Home

DSP-Based Implementation of a Digital Radio Demodulator on the

1. gt gt x Figure 3 12 DAB convolutional encoder 87 3 7 Errors Detection and Correction 11 040 4 1 0 4 20 aj_3 0 6 12 Qi 0 1 D Qi 4 P Qi 6 T3 048 ai 2 O Aj_3 0 5 D 0 6 In the corresponding trellis and therefore the Viterbi decoder four paths emerge from each node At the higher levels there are 64 nodes in each level and four paths enter each node one path amongst two must be chosen at each of these nodes Each branch represents a sequence of four transmitted bits The coding gain of a convolutional code depends on the code rate the constraint length and uniqueness properties endowed by the generator polynomials their choice is not obvious and may be the result of extensive computer searches as may be the puncturing codes Assessment of the number of errors can be corrected is quite complicated because of the unbounded length of the code word Nevertheless one useful parameter in the context of Viterbi decoding is the free distance This is the minimum Hamming distance between any two code words as the length and the number of code words considered approaches infinity On the basis that truncating the length of the Viterbi decoder from infinity to four or five times the constraint length incurs little penalty the free distance is a first order guide to the number of errors that can always be corrected in an input seq
2. Y A nor Y Y X Y Accumulator A Accumulator B x0 x1 yo y1 a0 al L bO b1 24 bit 24 bit 56 bit 56 bit y P 56 56 56 56 24 24 24 24 m gt 24 gt 56 56 56 5656 Y Y Y Y Y Y Y Y Connection Network 56 Pre Add Sub XMUL Multiplier 24 X24 YMUL Multiplier 24 X 24 48 gt 56 M Logic Shift Left Data path RSS units and Xbus and Ybus move buses Figure 2 2 step function 52 2 3 The Data Path ways signed unsigned complex SIMD multiplication with a pre add subtract unit before the XMUL multiplier e Two ALU units XALU and YALU The XALU executes single arithmetic and XMAC operations and the YALU executes parallel arithmetic and YMAC operations 2 3 1 Multipliers The CoolFlux BSP contains 2 multiply units XMUL and YMUL The both have two inputs of 24 bits They read their operands from the X and Y register files The XMUL can be used for signed signed unsigned signed unsigned unsigned complex and SIMD multiplications The YMUL supports signed signed complex and SIMD multiplications When doing a complex multiplication the inputs are considered to contain an imaginary part and a real part both 12 bits A complex multiplication does the following actions output imag inp
3. trame freme lt E 8 audio frame Am rm Em audio EE al ELS ERI pg ey frame n 1 Figure 3 35 Frame structure of MPEG Audio Layer II and corresponding DAB audio frame Each audio frame starts with a header consisting of a syncword and audio system related information A Cyclic Redundancy Check CRC following the header protects a part of the header information the bit allocation and the ScFSI fields After the CRC follows bit allocation ScFSI and Scale Factors The sub band samples which will be used by the decoder to reconstruct the PCM audio signal are the last audio data part in the MPEG Audio Layer II frame before the ancillary data field This ancillary data field which is of variable length is located at the end of the MPEG Audio Layer II frame An adaptation of the MPEG Audio Layer II frame to the DAB audio frame is performed in order to introduce e specific DAB Scale Factor Error Check ScF CRC e a fixed and a variable field of Programme Associated Data F PAD and X PAD The first four bytes of the DAB audio frame contain the MPEG Audio header This header carries information for the audio decoder In the DAB system some of this information is currently defined as static information This is e Syncword set to external synchronization of the audio decoder e Layer set to Layer II layer Layer IT 113 3 11 Audio Coding e protection bit set to CRC protection on 3 11 2 CRC check for
4. o o 34 1 7 1 How to characterize digital FIR filters 34 1 8 Error Measures si a Te e a a 36 Contents 1 81 Bit ErrorRate ee arean aaa ii ROS A S ess 36 1 82 Modulation Error Rate 37 1 9 Errors Detection and Correction o e e 38 1 9 1 Viterbi Decoder micras a 38 1 9 2 D zpunctuting A e A A ta 46 1 10 Digital Signals Processors DSP o o ooo o 47 CoolFlux BSP Architecture 49 Ql OVeEVIeW A ah A Rp eue td ed oi ad 49 2 2 Registers s a Ree ge I 9 RT AAA ek EAR WEAR ES 50 2 3 Ehe Data Pathi wia een utes om me a tds dp do dpi deve Bolten at el va 50 2 9 1 Multipliers 15 0 da a ee ia amp 53 2 3 2 Arithmetic and Logic Units XALU and YALU 54 2 4 Move Buses and RSS Units 2 0 0 000 200 0000 000 55 2 4 1 Xbus and Ybus Move Buses lees 55 2 4 2 Moving to an Accumulator eee 56 2 43 Moving from an Accumulator ee 57 2 4 4 Round Saturate and Select Units RSS llle 58 2 5 Memories and I O suse bue m e ted s 60 2 5 1 Individual Data Memory X Y llle 60 2 5 2 Dual Precision Memory XY lees 60 2 05 93 1 0 Memory Space ew ARR REA E g 60 2 6 X and Y Addressing Units raonar ra aea paa e e AA A e a 61 2 6 1 Indexed Stack Pointer Addressing ooo a 62 2 6 2 Indexed Pointer Addressing less 62 2 Program Control Unit mii iia ege Da p HR ns 62
5. Note that a bit in this case is not represented by 1 and 0 as the reader might have expected Instead the two values 1 and 1 are used 80 3 5 Quantization In picture 3 6 we show a noiseless constellation diagram resulting from the differential demodulator This output is the input to the quantization module Scatter plot for the 3 FIC symbols T T T 15 o Wee 1 i i 1 5 1 0 5 0 0 5 1 1 5 In Phase Figure 3 6 Constellation diagram without noise As a curiosity the bits in the above constellation diagram figure 3 6 belong to a real audio signal The first quadrant corresponds to the bits 1 1 the second to 1 1 the third to 1 1 and the fourth to 1 1 This signal is absented from noise This was only shown with academic purposes In MATLAB we can add white Gaussian Noise to simulate a more realistic channel See that the more noise we add the more and more away are the points from each other on the constellation diagram 81 3 5 Quantization 15 Scatter plot for the 3 FIC symbols 15 Scatter plot for the 3 FIC symbols T T r T T m mm ie AS 1 05 0 o5 1 15 In Phase In Phase Scatter plot for the 3 FIC symbols Scatter plot for the 3 FIC symbols 15 1 5 AS In Phase In Phase Figure 3 7 Constellation diagram adding noise We see that it is possible for a noisy signal
6. eee B 8 Energy Dispersal Unscrambling Contents B 9 Canal de Sincronizaci n 179 B 9 1 S mbolo Nulo 4 2 eo e GO a s 180 B 9 2 TEPR symbol sa eeg sa A 180 B 10 Implementaci n hardware del demodulador DAB 181 B10 DAB datapath 200 ewe el as e Bee ES 184 B 10 2 Memorias locales e e 184 B 10 3 Interfaz de depuraci n ee 184 B 10 4 Interfaz de interrupciones 0 0000000007 185 B 10 5 Buffers Compartidos 2 0 0 0 0 e 185 B 11 Descripci n de los m dulos BSP o e e o 186 B 11 1 BS Ps Sincronizaci n y buffering de muestras 186 B 11 2 BSP FFT Deinterleaving Demapping 187 B 11 3 BSP FIC and CIF Decoding Audio decoding 188 B 12 Conclusion s e oro e even e dnd ge xd ORE RA e MENSES 188 B 12 1 Punto de partida e c 2 o9 ode momo og es ovo m m ex 189 B 12 2 Objetivos propuestos y cumplidos css 190 B 12 3 Aspectos f bL rOS x ahaha eee RR Reus hen eben d e 190 B 12 4 Dificultades encontradas lee 191 vi Contents El presente documento es la memoria del proyecto de fin de carrera realizado por lvaro Mart nez L pez y Elena P rez Toril Gracia en la compa a NXP Semiconductors en Lovaina B l gica para la asignatura Sistemas Inform ticos de la titulacifon de Ingenier a en Inform
7. gt b ooenui v ohu g dS XNIJI009 Buipo2ep dW 49A893 19 U19p oui Buipo2eg Jld IHA po x iu Sel 440009 000009 iur 19010 wr xne b dSg xni3ioo i i 1eddewueg 41ojejnpouiep ASADO JeA gojiojurog baly 133 ERG 7 442008 000008 33400v 00000v 442008 000008 ev0z z z sayng nou eia las x90 q t xo WaWwx 340009 000009 0 dSg XNIJ1009 P oq uonoa402 cuanbaly 95v uoneziuo1u2u S 8 34400v 00000v geoz z 1 one NOM BIE los 9010 i eo oux 340009 000009 esey zo lt gt 15e 4 008 000008 bz01 z o anng mon Vea les 490 Q j Xni DVLP xa owa 420 0000 gt Gey Figure 4 10 DAB Architecture 130 4 1 An overview to CoolFlux BSP Around the three BSP instances several modules ensure the data flow between the different BSP s and from input to output The I Q data provided by the zero IF DAB front end has to be stored into buffer 0 A couple of alternating buffers are present between each pair of cores There is a common JTAG module to access the three CoolFlux BSP instances Each BSP has its own local memory and an address decoder for X memory see image 4 11 Each BSP has also an interrupt controller When all data has been processed the last core writes the output data often audio to customer specific circuitry by using the external memory X as the output interface Complex data flow done in hardware is bi directional
8. d PAIS SEES modulation 2kHz 1kHz 200MHz 1kHz 2kHz PNE signal 200 001 MHz integrator x fe a modulation signal x C A modulation signal 199 999 MHz Figure 1 16 generation of an OFDM signal are allowed to overlap each integrator will be presented with interference or crosstalk contributions as well as the wanted demodulated signal However if the radio frequencies could be chosen so that over the period of integration the symbol duration the integrals of all the interference signals amounted to zero then mutual interference would be cancelled This condition is known as orthogonality and is achieved when the carrier frequencies and the local oscillator frequencies are located on a regular comb where the frequency interval is equal to the reciprocal of the symbol duration With reference to Fig 1 17 take for example the modulated carrier at 200MHz with neighbors at 1 kHz 2 kHz etc either side The modulation state of each carrier is held constant over each symbol so each carrier is temporarily a sinusoidal wave with a particular phase and or amplitude representing the modulation When the incoming signal is acted upon by the mixer with the 200 MHz local oscillator the 200 MHz wave will produce a DC output signal and contributions from the neighbors will produce 1 kHz 2 kHz etc AC components the 400 MHz products will be neglected When the comp
9. Values 0 30 and 31 shall be reserved for future use of the FIG data field except for 31 11111 when used with FIG type 7 111 which is used for the end marker e FIG data field Generally FIGs may be arranged in any order except where special operational require ments dictate otherwise FIGs shall not be split between FIBs An example of FIG type is shown below fig 3 31 Type 0 field for extension 1 3 or 4 bytes cH E v k t 6 bits 10 bits 1 bit 7 or 15 bits b45 Dro bg bo bz orbis SubChid Start Short Long Size and protection Address form 1 bit 6 bits be bs bo Short form Table Table switch index 3bits 2 bits 10 bits b14 bio b11 bio bg bo Long form Option Protection Sub channel level size Figure 3 31 Structure of the subchannel organization field This image shows the structure of a FIB type 0 and FIG type 1 field it is used for the subchannel organization Each subchannel is defined by its own ID As an example of the fields contained into a FIG data field we describe e SubChlId Sub channel Identifier this 6 bit field coded as an unsigned binary number shall identify a sub channel e Start Address this 10 bit field coded as an unsigned binary number in the range 0 to 863 shall address the first Capacity Unit CU of the sub channel e Short Long form this 1 bit flag shall indicate whether the short or the long form of the size and protection field is used 0 shor
10. The next sections further detail the interconnection of the modules 4 1 6 Local memories Each CoolFlux BSP instance has three local memories P X and Y The memory sizes used in the design are described in the next table Instance Memory Size CoolFlux BSP instance 0 P Memory 2 Kwords 0x000000 0x0007FF X Memory 8 Kwords 0x000000 0x001FFF Y Memory 6 Kwords 0x000000 0x0017FF CoolFlux BSP instance 1 P Memory 1 Kwords 0x000000 0x0003FF X Memory 5 Kwords 0x000000 0x0013FF Y Memory 5 Kwords 0x000000 0x00013FF CoolFlux BSP instance 2 P Memory 14 Kwords 0x000000 0x0037FF X Memory 32 Kwords 0x000000 0x007FFF Y Memory 32 Kwords 0x000000 0x007FFF Figure 4 11 CoolFlux DAB Local memories 4 1 7 Debug interface Each instance of the CoolFlux BSP has a debugging unit There is however only one JTAG interface that can communicate with any of these three instances To do the multi core debugging a multi core debugger was necessary which was provided by NXP Semiconduc tors 131 4 1 An overview to CoolFlux BSP Each CoolFlux BSP instance has four hardware breakpoints It is possible to stop all cores when any of the instances has hit a hardware breakpoint In the multi core debugger it is possible to export a breakpoint from one core to another which allows us to debug the BSP cores together in a relatively simple way When selecting Breakpoint for the Bre
11. To compute a just multiply both sides of 4 10 by cos t and integrate From orthog onality the only factor that is not 0 is a1cos t cos t hence 27 27 2T F f cos t dt f t cos t dt ji a1cos t dt vai 0 0 0 hence Following the same argument we arrive to the general formula to compute the a s ak i F t cos kt dt T Jo Following an analogous argument the coefficients b s can be computed as 1 27 bk f t sin kt dt T Jo 153 A 4 Fourier Transform in Periodic Complex Functions A 4 Fourier Transform in Periodic Complex Functions So far we have discussed how a real function can be approximated by the Fourier serie 4 1 However we can extend the same argument to complex valued functions This might seem a mathematical nicety because after all who cares about complex numbers Well it is not the objective of this course to convince the reader of the usefulness of complex numbers Many physical facts are best modeled with complex numbers In the field signal modulation a QPSK signal carries two bits in each baud The first and second bits are coded in the real and imaginary parts of a complex number respectively Then the complex number is modulated into the carrier signal and launched trough the air In signal processing the Fourier Transform is surely the most performed operation If we had not the Fourier Transform for complex signals we could not process the information containe
12. zo lt gt 15e 4 008 000008 bz01 z o anng mon Vea les 490 Q j Xni DVLP xa owa 420 0000 gt Gey Figure B 6 DAB Architecture 183 B 10 Implementaci n hardware del demodulador DAB B 10 1 DAB datapath Alrededor de las tres instancias BSP existen varios m dulos que aseguran el flujo de datos desde la entrada a la salida y entre los diferentes BSP s Los datos IQ de entrada deben ser almacenados en el buffer 0 Entre cada par de n cleos se encuentra un par de buffers alternos Cada BSP tiene su propia memoria local y una direcci n en memoria X Adem s cada uno tiene su controlador de interrupciones Cuando todos los datos han sido procesados el ltimo n cleo escribe los datos de salida en un buffer de intercambio El flujo de datos real izado en hardware es unidireccional En las siguientes secciones se explicar la interconexi n de los m dulos B 10 2 Memorias locales Cada instancia BSP tiene tres memorias locales X Y y P Los tama os utilizados en el dise o se muestran en la siguiente tabla B 7 Instance Memory Size CoolFlux BSP instance 0 P Memory 2 Kwords 0x000000 0x0007FF X Memory 8 Kwords 0x000000 0x001FFF Y Memory 6 Kwords 0x000000 0x0017FF CoolFlux BSP instance 1 P Memory 1 Kwords 0x000000 0x0003FF X Memory 5 Kwords 0x000000 0x0013FF Y Memory 5 Kwords 0x000000 0x00013FF CoolFlux BSP instance 2 P Memory 14 Kwords 0x000000 0x0037FF X Memory
13. 3 9 1 Null Symbol Detection All the carriers are attenuated by at least 20 dB for the 1 297 ms duration of the null symbol in Mode 1 This discontinuity can be detected from the envelope of the incoming RF signal In order to detect the null symbol we must set up a window of size 512 words for which we calculate the average power of the incoming signal The idea is that the average power must fall below a given threshold in order to infer the null symbol is coming in When the average power falls below a given threshold we can be certain the null symbol is coming in Let s a bi be some sample The power of s is computed according to this formula P s a bi a bi a b In picture 3 16 we show the average value of a real DAB signal mode I You might note the abrupt falls of the power during the detection of the null symbol As expected the null symbol is presented each 96 ms for a DAB signal in mode I 91 3 9 Synchronization channel Output of Null detector 8 55 188 55 59 E Figure 3 16 Null symbol detection It might be the case that the average power falls incidentally below the threshold for a very short time due for instance to some obstacle in the path This event might be misinterpreted as the presence of the null symbol To avoid the possibility of such a false positive we force the average power to pass through a low pass filter In this way spurious
14. El L I qr eus gusu gouw jur szr qas BRE S gga wawx ETT m o zeba wawx DS zioyeuiuus Zgxp wa 0 no Opoyeurune y Lex ure ux mo mje m 1888 wawx wi szi Jes znova ogg usu E no K4 1 gga wax adoos boyd 3 E gioyeujuns E Lojeursa L 3 E 10je uiua fejdsig un sp ITY as nd asal Ou 4 du Jojgjeuar as ssp yasas uais g Ladoos m Viejdsig iy B c Figure 4 4 Simulink structure for DAB 122 4 1 An overview to CoolFlux BSP In the picture the main module in blue is the macro definition generated from the VHDL code for 1 BSP which once the code has been synthesized is considered a black box system and cannot be modified within Matlab alone The first very modest goal was to synchronize our demodulator with the broadcaster not to listen at the actual sound and so one BSP was enough for this task Below we will present the complete datapath with the three BSPs It is not in the scope of this project the design of the hardware outside the FPGA so we shall limit ourselves to comment out these components from a bird sight The oscilloscope is an essential device in order to debug the system in real time On the oscilloscope we can study the shape of the desired signal in the time domain The oscilloscope used for this project has two channels What we want to show at the output channels in the oscilloscope must be previously selected through the sour
15. Once all the algorithms have been introduced we describe their implementation on a FPGA in order to perform real time processing of the signal Although one CoolFlux BSP is capable of running the whole demodulator in real time on its own the FPGA is constrained to 50 MHz against 300 MHz of an actual BSP Thus the demodulation chain had to be pipelined into three BSP cores all of which were mapped on one FPGA Resumen Digital Audio Broadcasting DAB es un est ndar de emisi n de radio digital desarrol lado por EUREKA como un proyecto de investigaci n para la Uni n Europea Eureka 147 Este sistema pronto reemplazar a la mayor a de los m todos existentes de transmisi n de radio en muchas partes del mundo El principal motivo para el desarrollo del sistema DAB era conseguir un sistema para la recepci n m vil sin alteraciones de la se al Los principales problemas de la recepci n m vil provienen de la propagaci n mutipath la onda electromagn tica est sujeta a fen menos de reflexi n dispersi n y difracci n y llega al receptor como una superposici n incoherente de se ales lo que en sistemas cl sicos de radio tales como FM o AM se traduce en una recepci n pobre o incluso en la p rdida total de los datos El presente documento mostrar el desarrollo del software de un demodulador de DAB y su posterior implementaci n hardware en la arquitectura CoolFlux BSP el sistema para aplicaciones de bajo consumo desarrollado p
16. is sometimes used for very large arrays of samples In the simplest form of FFT the frequency domain samples appearing in its output array cover the same frequency range as the parent DFT but their arrangement is rather different It was noted earlier that the useful samples output by a DFT cover the range f 2 and it was implied that they are symmetrically disposed about 0 Hz However it was also noted that the spectrum is repeated centered about harmonics of fs so the negative frequency samples re appear between f 2 and f the sample at exactly f is a replica of the sample at 0 Hz By convention it is the range 0 Hz to one sample below f which is represented in the output array of the FFT The inverse FFT can be derived from the inverse DFT in a similar way and these comments apply equally to its input array of frequency domain samples ll 1 6 6 Applications to DAB The DAB system uses 1536 carriers in Mode 1 This requires a 2048 sample inverse FFT in the transmitter and a 2048 sample FFT in the receiver The way that an FFT is implemented in hardware depends on the required balance of 27 1 6 Fast Fourier Transform FFT speed versus hardware complexity Clearly parallel processing should yield the greatest speed whilst using several processes consecutively should reduce the amount of arithmetic hardware although it may increase the requirement for temporary storage For DAB there are options which a
17. I should say that this work is only an intuitive introduction to the Fourier Transform in one dimension The Fourier Analysis is in itself a vast discipline in which a lot of research is done even in present days I do not pretend to cover everything with this work For example we have explained functions R gt R and IR gt C However in the field of DIgital Signal Processing it is very usual to work with functions C C The intuition behind these functions is the same studying the variability of the dependent variable with respect the independent variable but explaining them requires knowledge in complex variable and it might not be so intuitive Fourier Analysis can even be extended to multidimensional domains C gt C This is useful for example for dealing with image processing from which I borrow some practical examples for the next chapter 162 A 7 Spectral Analysis A T Spectral Analysis We have talked about the time domain and the frequency domain of a function periodic or aperiodic however we have not represented graphically a function in the frequency domain Whereas in the x axis in the time domain we plot the variable t in the frequency domain we shall plot the frequency f We will make a distinction here from the rest of this text So far we have dealt with orthogonal functions in the form f t sin kt and f t cos kt with k Z That representation ignores physical units like seconds and Hertz
18. In the area of Leuven where this project was developed two radio channels could 124 4 1 An overview to CoolFlux BSP be selected which corresponded with both the flemish and the french emission It turned out that both signals had more or less the same quality 4 1 4 2 Tektronix TDS210 Tektronix TDS210 is a two channel digital real time oscilloscope used for testing Offering an unbeatable combination of performance reliability and versatility the TDS200 Series offers breakthrough digital and real time advantages By sampling at 10 and 16 times their bandwidths on all channels the TDS200 Series oscilloscopes provide accurate real time acquisition up to their full bandwidth Digital storage technology supports features including automatic measurements peak detect storage of two reference waveforms and five instrument setups and autoset Peak detect and high sample rates minimize aliasing while capturing waveform details useful for the user Figure 4 6 Tektronix TDS210 two channel digital real time oscilloscope The user interface is similar to that of an analog oscilloscope but with improvements that reduce learning time and increase efficiency Knobs and buttons are grouped by function and provide direct access to controls and each vertical channel has its own dedicated scale and position knobs Readouts or menus are displayed on screen at all times allowing users to more quickly and accurately determine instrument set
19. it follows 11 1 3 QPSK Quadrature Phase Shift Keying that we can send both signals through a single channel that is a single frequency without having them interfering with each other The following picture shows some random bits mapped in the I and Q components of a QPSK signal Both the I and Q components are added together to form the actual QPSK signal 1 0 0 1 Wut wr X XU er SS Wr d 1 0 1 0 H Time 0 T ar 37 4T 11 00 01 10 Data Figure 1 11 I and Q components mapped into QPSK signal The next picture shows a conceptual transmitter structure for QPSK The binary data stream is split into the in phase and quadrature phase components These are then sepa rately modulated onto two orthogonal basis functions In this implementation two sinu soids are used Afterwards the two signals are superimposed and the resulting signal is the QPSK signal Note the use of polar non return to zero encoding These encoders can be placed before for binary data source but have been placed after to illustrate the conceptual difference between digital and analog signals involved with digital modulation 2 Digital Signals Analog Signals t cos 2x f VE v Es f 10014 V O Y NRZ Encoder Binary Bitstream _ A Demultiplexor 11000110 OPSK signal Y gt NRZ Encoder x 101 fE iota x LL f sin 2s fet Figure 1 12 QPSK modulator 12 1 4 FDM Freque
20. the array of modulation signals which are to be applied to the carriers during a single symbol provide a specification of the spectrum of the required composite signal for that symbol The modulation of the carriers is intended to remain static during each symbol so each modulation signal contains one sampled value per symbol The array of modulation signals can then be thought of as a series of samples which make up a function of frequency The composite OFDM signal will be produced by the iDFT as a series of samples which follow one another in time so it can be thought of as a function of time The array of oscillator signals can be thought of as a function of both frequency and time the array of different frequencies forms a series with respect to frequency as with the modulating signals and if each oscillator provides a sine wave a sampled version of this forms a series with respect to time With this nomenclature the iDFT and the contents of figure Fig 1 15 can be expressed highestfrequency f t 5 function of frequency x oscillator signal lowest frequency To produce the first sample of the function of time each of the modulation samples is multiplied by the first time sample of its corresponding oscillator signal and all of the products are added together For the second sample the second time samples of the os cillator signals are used and so on In this way any number of samples of the function
21. tica en la Universidad Complutense de Madrid Los autores de este proyecto autorizan a la Uni versidad Complutense de Madrid a difundir esta memoria con fines acad micos no comerciales y mencionando expresamente a sus autores lvaro Mart nez L pez Elena P rez Toril Gracia Universidad Complutense de Madrid June 2010 Acknowledgements The realization of this project has been possible thanks to the contribution of many people In the first place I want to thank Luis Pi uel Moreno the director of this project for giving me the wonderful opportunity of going to Belgium for working on this project I want to thank Pancho my mentor at NXP for teaching me everything I know about signal modulation embedded systems and digital signal processors Thanks a lot for those marvelous 6 months in Leuven I also want to mention my gratitude towards Ludo Marleen and Sandrine from NXP for their help and support specially during the month of august Many thanks to Elena my project colleague and friend for her companionship For me this is not only a final year project but a symbol that represents all my trajec tory during the years I spent as an undergraduate student at university As such I want to mention all that people that has contributed to this venture First of all thank to my family specially my mother sister and grandmother for their love dedication and support while raising me Many thanks to my friends Emilio Pedroche Cave
22. 1 An overview to CoolFlux BSP datao out data outt RAM RAM Block 0 Block 1 1024 x 24bit 1024 x 24bit addro in datao in weon addri in data1 in wet in Block sel gt 5 4 4 addr ino data ino wed data outo adir int data int wet data out To From Memory Domain A To From Memory Domain B Figure 4 13 Shared buffer architecture Each shared buffer connects two hardware modules in the design The first shared buffer is between customer specific hardware and CoolFlux BSP instance 0 The second shared buffer is between the CoolFlux BSP instance 0 and CoolFlux BSP instance 1 The third shared buffer is between the BSP and BSP Each shared buffer contains 2 synchronous physically independent memories The memories of shared buffer 0 are of 1024 word of 24 bits The memories of shared buffer 1 and 2 are of 2048 words of 24bits While one memory is being accessed by one hardware module in block 0 memory domain the second memory can be accessed by the second hardware module in block 1 memory domain At some point in time this order can be changed At that moment the second memory is accessed by the first hardware module while the first memory can be accessed by the second hardware module This change is controlled by the signal block sel When this signal changes its value the connections between two memories and the two interfaces of the shared buffer are swapped so that there wi
23. 2 7 1 Hardware Loop Stack oaaae 63 2 8 Hardware Interface evita aa LUE ad teta 63 28 gt O MEDIOS ct A AA A 64 2 8 2 PMA uet dh oh Ru ARN ee ep p Ge ee 64 Contents 2 9 Multi Cycle Instructions and Delay Slots 64 2 9 1 Multi cycle Instructions 0 2 000004 64 2 9 2 Delay Slots sec ig lp ee ee ae eee a ieee A ES 65 2 10 Hardware Loops s mne a AAA es 66 2 11 Debug Mode a aacc e EAM MASS SO ae 67 2 12 Interr pts ok a A ee Pe a ot 67 2 13 BSP C Complle mA a Bd de x 67 2 13 1 Optimizations ses amp 5 ierst 5 248 2 SER GARE S BAIA led 70 DAB Implementation on the CoolFlux BSP 71 3 1 Automatic Gain Control e 74 A AA E A 76 3 3 Differential Demodulation e o 77 3 4 Frecuency deinterleaving 0 20000 00 0000000 78 3 5 J Quantization DA re Pes c e a dde el oe 80 3 0 1 Hard Decision ae a oe eH 80 3 5 2 Soft Decision aswel 82 3 6 Time De nterleaving e 83 3 7 Errors Detection and Correction o e 87 3 8 Energy Dispersal Unscrambling e 89 3 9 Synchronization channel e 90 3 9 1 Null Symbol Detection 22e 91 3 92 TERRY etc a a utes 93 3 9 3 Coarse Frequency Correction a 96 3 9 4 Fine Frequency Correction ees 98 3 9 5 Fine time synchronization 2A 100 3 10 Transport mechanisms 0 00 ee eee ee eee 102 3 10 1 Fast Info
24. 32 Kwords 0x000000 0x007FFF Y Memory 32 Kwords 0x000000 0x007FFF Figure B 7 Memorias CoolFlux DAB B 10 3 Interfaz de depuraci n Cada uno de los n cleos BSP tiene una unidad de depuraci n y una interfaz JTAG que permite comunicarse con cada uno de ellos Para hacer esta depuraci n multi n cleo fue necesario un depurador multi n cleo proporcionado por NXP Cada Coolflux BSP tiene cuatro posibles breakpoints siendo posible detener todos los n cleos cuando alguno de ellos alcanza este punto En el depurador multi n cleo es posible exportar un breakpoint de uno a otro lo que permite depurar los n cleos de una forma relativamente simple 184 B 10 Implementaci n hardware del demodulador DAB B 10 4 Interfaz de interrupciones Todas las interrupciones van directamente al controlador de interrupciones y ste es el encargado de manejarlas en orden de prioridad Cada BSP puede generar una interrupci n al BSP contiguo escribiendo en una determi nada direcci n Input Output La interrupci n se genera independientemente del valor que se haya escrito B 10 5 Buffers Compartidos Para la comunicaci n entre los BSP se utilizan buffers intermedios entre ellos En total existen tres buffers compartidos en la ruta de datos El dise o de cada buffer intermedio se muestra en la figura B 8 datad_out data outi RAM RAM Block O Block 1 1024 x 24bit 1024 x 24bit 4 ad
25. A 7 noisy signal restored In the last picture we see the original signal restored A 8 2 Image Compression So far we have dealt with one dimensional signals in which the independent variable was the time An image is just a two dimensional signal where the independent variables are the coordinates measured in pixels from the top right corner of the picture The dependent variable is a three dimensional vector indicating the color components of the pixel Each 168 A 8 Applications of the Fourier Transform fY zeros size eY A B size eY for i 1 to B do if abs eY i n lt 1 then fY i 0 else fY i eY 1 end if end for ifY if ft fY cy ifY subplot 2 1 1 plot t cy grid on axis 0 et 10 10 xlabel Time ls ylabel Amplitude Y f ft cy n size cy 2 2 amp spec abs Y n subplot 2 1 2 freq 0 79 2 x n dt plot freq amp_ spec 1 80 grid on zlabel Frequency H z ylabel Amplitude 169 A 8 Applications of the Fourier Transform pixel is colored by mixing different quantities of Red Green and Blue RGB Model Every thing explained so far for one dimensional signals applies here too Think again of sound The sound received in your ear is a variation of the air pressure with respect to time An image is a variation of intensity of light in a two dimensional plane Therefore we might extrapolate Fourier Analysis to
26. Figure A 2 step function filt sin t sin 3t folt sin t T sin 3t sin 5t zsin 7t t L sin 9t 1 1 1 1 fa t sin t 3 on 3t Sin 5t zsin 7t gin 90 1 l l l 73 sin 13t 15 5in 1t 15070 39 5in 19t L li 57 sin 21t 9g Sini23t We can think of the last three signals as composed of different subsignals each with a constant frequency amplitude and phase although in this example all have the same phase For example if t represents seconds the signal f2 t is composed of 5 subsignals 111 13257 with the frequencies inHz 3THz an Hz InHz and 2nHz and with amplitudes 1 and rt Notice that the more sines i e subsignals we add to the summation the more accurate is the approximation to the signal illustrated in figure A 2 In general the last intuition can be formalized as follows A function f t sufficiently well behaved can be expressed as F f ao a1cos t ay sin t azcos 2t basin 2t A 1 147 A 1 Motivation 2 n n 4 0 1 2 3 4 5 6 rads Figure A 3 sine function Mathematicians are not very happy with approximations so they will approximate the first signal by the infinite serie F f ao Y an cosnt Y bn sin nt A 2 n 1 n 1 In general it can be formally proved that the serie F converges to the original signal f The proof will not be provided here even Fourier was not able to prove it In any practical
27. Instruction de 1 por lo que cada BSP mapeado en la FPGA no puede procesar m s de 50 MIPS La cantidad total de MIPS requeridos para la demodulaci n result ser de 80 despu s de varias optimizaciones En teor a se podfia de modular la se al con tan s lo 2 BSPs Sin embargo la adici n de m s n cleos aumenta el n mero de MIPS debido a las transferencias de datos que se llevan a cabo entre ellos Por ello la implementaci n del demodulador se ha realizado con 3 BSPs De ahora en adelante nos referiremos a ellos como BSP BSP y BSP En la imagen siguiente B 6 observamos las conexiones entre los distintos BSPs La separaci n de la funcionalidad en 3 DSP no fue una simple divisi n del c digo en tres m dulos transmitiendo los datos de uno a otro Los requerimientos de memoria y MIPS fueron factores a tener en cuenta 182 Implementaci n hardware del demodulador DAB B 10 gt b ooenui v ohu g dS XNIJI009 MzE A Buipo2ep dW 49A893 19 U19p oui Buipo2eg Jld IHA 344008 000008 ev0z z z sayng nou eia po x iu sel 440009 000009 iur x2oiq Jul xne b dSg xni3ioo H t A seddeweg 41ojejnpouiep ASADO JeA gojiojurog baly 133 ERG 7 33400v 00000v 442008 000008 las 49010 t xo WaWwx 340009 000009 0 dSg XNIJ1009 P oq uonoa402 cuanbaly 95v uoneziuo1u2u S 8 34400v 00000v geoz z 1 one NOM BIE los 9010 i eo oux 340009 000009 esey
28. OFDM during 75 consecutive symbols preceded by the null and Phase Reference symbols By dividing the bit stream into 75 blocks of 3072 consecutive bits and associating pairs of bits within a block 1536 bit pairs can be made available during each symbol to be mapped onto the carriers aE LH 96 ms TM 1 or 24 ms TM ll or 48 ms TM IV Null TFPR FIC 1 MSC 72 FIC 2 FIC 3 MSC 1 MSC2 MSC3 MSC 71 i Figure 3 5 Transmission frame structure If the mapping was a simple one to one relationship e g the first two incoming bits T8 3 4 Frecuency de interleaving were associated as a bit pair and this always modulated the lowest frequency carrier etc static selective fading or relatively narrow band interference could impair one or more of the sub channels selectively they could suffer most of the errors Even in the case of mobile reception long strings of bits would be subject to relatively correlated fading events and this could give rise to bursts of errors As noted before this is the least favorable condition for successful error correction by a Viterbi decoder Instead the mapping is based on a static pseudo random series and when the bit pairs are recovered in the receiver fading and interference events which are correlated amongst groups of adjacent carriers are dispersed within the bit stream i e they are dispersed in time Subsequent bi
29. The numbers output by the FF T represent the phasor sums of these components so the array of numbers contains many shifted versions of the CAZAC sequence superimposed with different amplitudes This is indicated in Fig 3 24 by the use of smaller typefaces 99 3 9 Synchronization channel When the correlation is performed these different amplitudes are preserved in the nu merous values of correlation which are found at different discrete shifts as illustrated in Fig 3 25 A magnitude of Correlation frequency error 16 P 0 HAs cce 1 shift Figure 3 25 magnitude of correlation vs shift with a small frequency error The variation of the magnitude of the correlation at each discrete shift with increasing frequency error follows a sinx x law as illustrated by the dashed line in Fig 3 25 the square is introduced by taking the magnitude In this case the variable x is the sum of the magnitude of the error and the amount of shift multiplied by 7 The amount of shift is an integer so for a given error sinx has the same value for all shifts Therefore the magnitude at each shift is proportional to 1 2 and individual values are related to one another by a quadratic equation i e of the form a bx cx 0 The magnitude of the error can then be calculated from the magnitude of the correlation at different shifts and the sign of the error can be established with an additional measurement The middle three val
30. accumulator can be the source of a move operation in 2 ways as a full accumulator a0 al b0 b1 through one of its sub registers ao0 aol ah0 ah1 al0 all bo0 bol bhO bh1 bl0 bl1 In the case a sub register is the source of a move operation the data width is either 8 or 24 So the conversion to 24 bits is straightforward sign extension with 16 bits or 57 2 4 Move Buses and RSS Units no conversion The selection of the correct sub register with eventual sign extension is performed by the select part of an RSS unit A full accumulator can be the source of a move operation in 4 ways a 24 bit scalar move a 48 bit scalar move e a 24 bit packed move e a 48 bit packed move In 24 bit scalar move assignment the 56 bit value is converted to 24 bits through the round and saturate part of the RSS unit and transferred to the destination register via one of the two move buses In a 48 bit scalar move l assignment the Xbus and Ybus move buses are combined to move a 48 bit value in a single processor cycle The 56 bit source value is converted to 48 bits through the saturation logic of the RSS unit A 48 bit scalar move operation is only possible between two accumulators or from one of the accumulators to the XY memory 6 In a 24 bit packed move c assignment the 56 bit value is converted to 24 bits through the round and saturate part of the RSS unit and transferred to the destination
31. audio side information A CRC check word for detecting errors within the significant side information of a DAB audio frame has been inserted in the bit stream just after the DAB audio frame header The error detection method used is CRC 16 whose generator polynomial is Gila x r z 1 The bits included into the CRC check are e 16 bits of DAB_audio_frame_header starting with bit_rate_index and ending with emphasis e a number of bits of audio_data starting with the first bit These bits include bit allocation information and ScFSI The initial state of the shift register for the calculation of the CRC is 1111 1111 1111 1111 If the final output of the shift register and the CRC check word in the DAB audio frame are not identical a transmission error has occurred in the protected field of the audio bit stream H 114 Audio Coding 3 11 WO SMQN 95 gt lt OF AjPuonpuoo CANE 6 0 t r St 8 92 91 Spueq qnsispueq qnsspueq qnsspueq qns paues es 185 OBS a S L sadwes soddwes jew pres naj ar puomouks se fo ee ouo ra A Y OHO sajdwes pueq qns 39S wes oipne gya 1xeu ay 0 siajas e Lm a u L u epow 091975 sajdwes ndu ojpne WOd ZSL 40 pIJeA uuea1 s yq Peapod jo einjonujs eure Figure 3 36 Frame structure of MPEG Audio Layer II 115 Chapter 4 Hardware implementation of the DAB demodula
32. be built up and in many cases this is found to approximate to an exponential curve see figure 1 9 The distribution is characterized by a single parameter known as the delay spread which can be interpreted as either the mean or the standard deviation in the case of an exponential distribution The next equation expresses the sta tistical distribution of the echo power as a function of the delay The parameter T is the delay spread The incremental effects of greater delay or greater echo power are similar they both increase the potential for inter symbol interference so the delay spread interpreted as the mean is a guide to the interference potential of a given type of environment The average degree of data corruption is proportional to the ratio of the symbol duration to the delay spread For the case of outdoor reception from a single terrestrial transmitter the delay spread can range from less than 0 5s to 5s or more a typical median value for 50 of locations is around 1 us Echoes with delays outside this range can be encountered but the percentage of locations for which they contribute significantly to the delay spread is small less than 1 Direct reception from a satellite yields values of 0 5 us or less Thus it was a pre requisite for the DAB system that the symbol duration should be much greater than the values of delay spread encountered in common broadcasting environments In other words the symbol duration s
33. changes in the signal power will be attenuated An example of a real filtered signal is represented in picture 3 17 with pink color The red line is an indicator that changes from 0 05 to 0 when the filtered average power of the signal falls below the decision threshold or equivalently when the null symbol is detected Note that the filtered signal is much softer than the unfiltered one D 10 20 30 40 50 50 70 Figure 3 17 Low pass filtered signal So up until now we have performed the coarse time synchronization and so we are almost at the beginning of one frame The TFPR symbol starts approximately within a 92 3 9 Synchronization channel fixed offset measure in number of samples from the point where the average filtered power falls below the aforementioned threshold Such approximation is not permissible in order to start demodulating the actual audio signal In fact the TFPR serves as a symbol to do the fine time synchronization the coarse and fine frequency synchronizations 3 9 2 TFPR symbol During the TFPR symbol all carriers are transmitted at normal power with the normal symbol duration and guard interval these are necessary requirements for its use as the phase reference for all carriers The carriers are modulated in a fixed pattern and this pattern is reproduced when the incoming signal is analyzed by the FFT in the receiver However the position of the pattern within the array of numbers outp
34. correction decoder where the redundant data and the soft decision detail are used to reconstruct the un coded bit stream symbol by symbol as accurately as possible The reconstructed bit stream is de multiplexed and the audio program data are con verted back to analogue signals which are reproduced by a loudspeaker 72 Buipo a I Duipooeg 319 ruens eubis ojpny y38 S 1epoou3 H BUONNJOAUO BunquieJosun Buipooeg DW jesuedsig IQJ8lIA Buumoundeg eig 2 K61eu3 eea SN 2480 2j aly Ke uoneziuoiuoug NZD m 6umeaaywieg baly JOJE NPOWap enuas Figure 3 1 A simplified DAB transmission system 73 3 1 Automatic Gain Control 3 1 Automatic Gain Control Automatic gain control AGC is an adaptive system found in many electronic devices The average output signal level is fed back to adjust the gain to an appropriate level for a range of input signal levels For example without AGC the sound emitted from an AM receiver would vary to an extreme extent from a weak to a strong signal the AGC effectively reduces the volume if the signal is strong and raises it when it is weaker Because of mobile operation the average signal amplitude constantly varies by about 20 dB Hence for COFDM receivers careful AGC design is a vital feature A special scheme called a null symbol a break in the transmitted signal of 1 ms in DAB transmission mode I for example is used for synch
35. data were held static for two consecutive symbols the waves would be continuous over 2 ms In the DAB receiver in order to accomplish a single FFT operation it would not matter 33 1 7 Finite Impulse Response Filters FIR at what instant the time window began as long as the input waves were continuous over 1 ms A time displacement would only change the apparent phase of each of the waves which would alter the absolute values of the data output by the FFT but since the phase modulation is coded differentially a slow change would not corrupt the decoded data Thus this example of an oversized guard interval would permit time agility in the receiver and the simultaneous reception of delayed signals as long as the limit of temporal coherence was not exceeded i e the maximum vehicle speed would be limited further in this case Of course there is no need to repeat the inverse FFT operation when the output samples can simply be stored and read out to the DACS directly after the active symbol Also as noted in the main text the DAB system actually uses the more economical compromise of a 246 us guard interval so only about one quarter of the output samples need to be stored According to available information the samples making up the guard interval appear to be applied to the DACs before each active symbol so they are repetitions of the last quarter of those produced during the active symbol This alternative arrangement does no
36. described here as it is defined in the fre quency domain that is at the input to the inverse FFT in the transmitter or the output of the main FFT in the receiver and not in terms of the time domain signal that is transmit 101 3 10 Transport mechanisms ted The time domain signal is related to this by the Fourier transform Now the PSD of a signal in one domain is related by the Fourier transform to the autocorrelation function of the transformed signal in the other domain In this case the autocorrelation function of the DAB signal described in the frequency domain is that of the CAZAC sequences which could be described as an impulse The Fourier transform of an impulse is a constant value viz for a true impulse in the time domain the frequency spectrum is white so it follows that the PSD of the transmitted DAB signal described in the time domain is constant If the Phase Reference symbol was repeated the PSD would be constant over the duration of any number of repeated symbols Another benefit of the constant amplitude nature of the Phase Reference symbol block is that the process of division by the stored array in the receiver only has the effect of subtracting phases The same effect can be achieved by multiplying the array output by the main FFT by a stored array representing the conjugate of the PhaseReference symbol block and multiplication is an easier process to carry out digitally Generally this would i
37. easily be removed A spectral component at the Nyquist limit must by definition contain an unwanted alias If the Nyquist criterion is just satisfied in the time domain sampling then the resulting sampled spectrum cannot contain useful information at frequencies above half the time sampling frequency If the time sampling frequency is f and the time window duration is T then the number of samples processed N T f The interval between the frequency domain samples is 1 T and the useful range lies between f 2 so the number of useful samples is f 2 1 T N 2 that is N 2 samples at positive frequencies and N 2 at negative frequencies Thus the total number of useful samples in the result is equal to the number of samples input With N time domain and N frequency domain samples a total of N coefficients are needed in the summation The frequency domain sampling of the result has a similar although reciprocal effect in the time domain that is the result of the DFT applies to the time windowed input signal as if it were periodic with a period equal to the time window duration If the input signal really is periodic i e it is composed of the same waveform during consecutive and contiguous time windows the results of consecutive DFT calculations will be the same If it is not subsequent DFT calculations will yield different results 1 6 4 Computation of the DFT If a computer program was set up to impl
38. encoded bit 43 1 9 Errors Detection and Correction 4 3 o o o 44 o 03 2 gt e o decision boundaries o o o o e 9 o e o o ee eee e e o o o S eee be oo o Ti 4 Figure 1 25 Decision boundaries pairs to be 01 The Viterbi decoder would then have to decide between the two possible encoded bit pairs 00 or 11 This decision depends on how far away the received bit pairs are from the two possible transmitted bit pairs Figure 1 26 The metric used to measure this is the Euclidean distance It is also referred to as the local distance calculation local distance n i y S n G n j encoded bits associated with a given input j where n denotes the time instance and 2 denotes the path calculated for The above equation can be further simplified by observing the expansion local distance n i 2 S7 n 28 n G n GZO and noting that S7 n and G5 n are constants in a given symbol period These terms can be ignored since we are concerned with finding the minimum local distance path The following measure will suffice for determining the local distance Note that the 2 is taken out which implies local distance n i Y 85 n G n j that instead of finding the path with the minimum local distance we would look for the path with the maximum local distance n i Going back to Figure 1 24 the branch metrics ca
39. for results of XMUL YALU for results of YMUL The multipliers are always accessed via MAC operations multiply accumulate which take 2 processor cycles to complete In the first cycle the multiplication is performed on XMUL or YMUL In the second cycle the result is accumulated through the XALU or YALU units Each processor cycle a new MAC operation can be started for which the multiplication runs in parallel with the accumulation of the previous MAC operation This means that n consecutive MAC operations only take n 1 processor cycles to complete ll 2 3 2 Arithmetic and Logic Units XALU and YALU The XALU and the YALU always operate on 56 bit operands and generate 56 bit results They take their inputs from the A B X and Y register files or from the temporary registers at the outputs of the multipliers All these inputs are converted to 56 bits The result of the XALU can be stored in the A B X and Y registers the result of the YALU can be stored in a B register The result is converted to 24 bits when it is written to an X or Y register The possible operations in scalar mode for XALU are e Add subtract add and divide by 2 subtract and divide by 2 reduced add reduced subtract Logic and logic or logic xor Divide step e Min max e Negate exponent absolute value sign extend e Logic and arithmetic shift reduced shift 54 2 4 Move Buses and RSS Units e Compare operations e Load immediate value A
40. frequency response is revealed by holding the definition of the Phase Refer ence symbol block in the receiver as an array of complex numbers in ROM and by dividing the main FFT outputs by their counterparts in this stored array The resulting array is then applied to an inverse FFT which outputs a further array representing the impulse response The magnitudes of elements in this array are then calculated and can be used to derive the timing reference for the following transmission frame This involves searching for the peaks and applying an algorithm for the chosen timing strategy in order to derive a number representing the amount by which the timing of the main FFT processing window should be advanced or retarded For accuracy in this method the transmitted signal must have a constant Power Spectral Density PSD at different frequencies within the bandwidth of the DAB ensemble i e its spectrum must be white for the duration of the Phase Reference symbol With this provision the same gain can be applied at all frequencies to avoid distorting the influence of channel noise To a first order and certainly over the period of only one symbol the PSD is effectively constant because the carriers are all transmitted with the same amplitudes This might not be the case if the spectrum of the DAB signal were considered over a period of many symbols because it would also depend on the carrier phases The Phase Reference symbol block has been
41. han sido necesarios considerables pasos de optimizaci n Una vez finalizados todos los procesos de optimizaci n y posibles mejoras del proyecto el paso siguiente ser a la implementaci n en un chip dedicado con el dise o del demodulador 190 B 12 Conclusiones para su posterior comercializaci n Actualmente son varias las empresas interesadas en su compra para la instalaci n en equipos port tiles tales como reproductores MP3 tel fonos m viles y radios para autom viles B 12 4 Dificultades encontradas Una dificultad a adida ha sido el hecho de lidiar con un sistema en tiempo real con las complicaciones que ello conlleva ya que la se al tiene que ser procesada dentro de unas restricciones de tiempo muy exigentes Durante la especificaci n se calculaba que para demodular la se al en tiempo real ser a posible utilizar un s lo DSP Durante el prototipado del mismo en una FPGA algunas restricciones del hardware principalmente la frecuencia de reloj de sta nos obligaron a implementarlo con tres m dulos DSP Con respecto al procesamiento de se al digital es importante recalcar la importancia de los c digos de correcci n de errores ya que sin ellos no ser a posibe la recepci n de una se al limpia y sin perturbaciones y tal vez ni siquiera audio ya que algunos frames contienen informaci n importante de cabeceras Estos m todos de correcci n han supuesto gran parte de la implementaci n del sistema Por razo
42. image processing It happens that the range of frequencies that the human brain can process is limited Psychoacoustics is the study of subjective human perception of sounds They same holds for colors We cannot perceive very high changes in color in a small space Recall that more frequencies implies more Fourier Coefficients to represent the information If we cannot perceive some frequencies why should we waste space to store them As we are about to see by applying the FFT to an image removing the unperceived frequencies and then going back to the spatial domain we can considerably reduce the size of a picture The next MATLAB script traverses all colors and for each color all rows In the first block of loops each row is transformed using the DCT then high order samples are discarded and the low order samples are reverse transformed using IDCT The same is done for columns which results in a higher compression ratio 170 A 8 Applications of the Fourier Transform clear all foto imread f22 jpg ancho size foto 2 mitadSuperior floor ancho 2 cuartoSuperior floor ancho 4 octavoSuperiorh floor ancho 8 fotoComprimida2 fotoComprimidaA fotoComprimida8 for k 1 3 do for i 1 size foto 1 do filaDCT dct double f oto i k fotoComprimida2 i k idct filaDCT 1 mitadSuperior ancho fotoComprimidaA i k idct filaDCT 1 cuartoSuperior ancho fotoComprimi
43. instruction irl i2 k k 1 k 2 decode instruction decode instruction decode instruction decode instruction decode instruction i it i 2 k k 1 execute execute execute execute execute k instruction 1 1 instruction i instruction i 1 instruction 1 2 Some instructions of the CoolFlux BSP have two variants the first is multi cycle no delay slots the other has a number of delay slots i e the number of cycles Ai 1 The option with the delay slots takes as many cycles to execute as the first variant but instead of stalling the instructions in the delay slots are executed l l 2 10 Hardware Loops The CoolFlux BSP supports up to 4 nested hardware loops When a loop instruction is executed the start address the end address and the number of times the loop must be executed the loopcount value are pushed onto the hardware loop stack This loop stack is four positions deep When it is full and the program tries to initiate another loop the instruction is ignored When the address of the instruction that has been fetched last equals the end address of the loop at the top of the stack the loopcount value of this loop is decremented When this new loopcount value equals 0 the loop is popped from the stack When it is not equal to zero the new loopcount value is saved and the program counter is set to the start address of the loop and the instruction at the start address is the next instruction being fetched Note that the loopcount value at the end
44. integrals infinite sums and so on An integral is a summation of infinite rectangles of infinitesimal base and infinitesimal area Thanks to the Fundamental Theorem of Calculus we find that there is a tight connection between derivatives and integrals In most cases we can analyti cally calculate the integral of a function by knowing that integral and derivation are inverse operations However the continuum does not exist in the real world Quantum mechanics the paradigm that best describes the known Universe in its most microscopic scale teaches us that the magnitudes we measure are discrete rather than continuos Moreover computers handle discrete numbers in discrete steps of time Any integral is substituted by a finite sum in order to be computed by a machine This translates the Fourier Transform into the Discrete Fourier Transform DFT The integral in the Fourier Transform becomes the summation N X k M z ne n20 N 1 A 23 n 0 mn whereas the inverse DFT is N 1 1 ink a n So X ke A 24 k 0 x is the vector of samples x ro 24 2 1 This vector is obtained by discretizing the continuos signal f in equally spaced points The process of sampling is usually done by a device called Analog to Digital Conversor ACD A note about sampling There is a theorem due to Nyquist and Shannon that says that the sampling frequency must be at least the double of the maximum frequency present in the continuos perio
45. is they must run in real time As we shall see more deeply below in DAB the broadcaster transmits an audio frame each 96ms therefore our DSP must be capable of processing the signal that is extract the audio in less than this time Otherwise part of the audio will be missed For this reason a DSP has specialized hardware not available in a general purpose computer to perform such common tasks All these constraints is what makes Digital Signal Processing a very motivating and interesting field of study Figure 1 27 illustrates the abstract architecture used for any dig ital signal applicationl l Analog Digital Signal Analog ABO Processing DAE Signal Figure 1 27 Electromagnetic signal to audio 48 Chapter 2 CoolFlux BSP Architecture 2 1 Overview The CoolFlux BSP is a general purpose high precision ultra low power low cost dual Harvard dual MAC BSP that has extensive support for fractional arithmetic as well as highly efficient support for the data types and operations encountered within ANSI C In addition CoolFlux BSP supports complex arithmetic sub word parallelism and larger addressing space The CoolFlux BSP is capable of sustaining two MACs two memory operations and two pointer updates per cycle making it highly cycle efficient for computationally intensive applications Many of the arithmetic operations Ai such as the multiplication additions division etc can be done in scalar mode or in packed
46. is known as the channel and as we shall see a great effort is devoted in DAB to transmit a clean signal through it figure 1 1 Figure 1 1 Elements in radio communications 1 1 Signal Modulation The main concept behind radio communications is that of modulation An electromagnetic wave has three main attributes that can be varied in time to convey information amplitude A frequency w and phase 4 figure 1 2 1 1 Signal Modulation H time I second 1 wavelength A amplitude amplitude wavelength A Figure 1 2 Wave attributes Figure 1 3 shows three waves with the same frequency and amplitude but different phases Figure 1 3 Three waves with different phase In radio communications two waves are used to transport the message the base signal and the carrier signal We can multiply the former by the later to form the modulated signal The device used to carry out this process is known as a modulator In the same way a demodulator is a device that takes as its input the modulated signal to extract the base signal hence recovering the information Classically information was sent in an analog fashion by varying the amplitude fre quency or phase within a continuous range As an example let s mention the well known FM and AM radio systems picture 1 4 in which the amplitude and frequency are changed over time respectively Pl By contrast digital information is ma
47. mode Packed mode can either be a complex operation or an SIMD operation This makes it possible for instance to do four 12 bit multiplications in parallel The CoolFlux BSP architecture is a load store architecture that is e All memory operands have to be explicitly moved to the registers e All execution units take their input values from registers and write their results back into a register e All results produced in the registers have to be explicitly moved to memory 49 2 2 Registers Because of its Instruction Set Architecture design the CoolFlux BSP is also highly efficient at supporting control code as well making it very well balanced for complete embedded applications in which 20 80 80 of the time is spent for 20 of instructions code mix is typical When designing the CoolFlux BSP at the level of its micro architecture much attention has been paid to efficiently support the operations needed by ANSI C making the CoolFlux BSP an excellent compiler target whilst maintaining very low power consumption l 8l 2 2 Registers The CoolFlux BSP is a load store architecture making registers important All execution units take their inputs from registers and write their results back to registers Memory accesses are considered to be operations A store operation takes the data from a register and writes it to memory while a load operation reads a value from memory and stores it in a register The registers of the CoolF
48. obtained when the outputs of all shift register stages are set to value 1 The first 16 bits of the PRBS are given in table 3 15 89 3 9 Synchronization channel Figure 3 15 PBRS indexing table 3 9 Synchronization channel Many of the processes in the DAB receiver require accurate synchronization with aspects of the frequency and timing of the incoming signal and in mobile reception conditions all of these aspects can be modified by the Doppler shift The following aspects of synchronization need to be considered e The rate at which FFT computations are carried out must be equal to the reciprocal of the total symbol duration of the incoming signal to avoid inter carrier crosstalk e The channel decoding and de multiplexing process must be clocked at the same rate as data appear in the incoming signal so the correct data are processed and selected e The frequencies of components of the baseband signal input to the FFT must be such that the carrier frequencies lie precisely on the correct teeth of the comb with frequency separation equal to the reciprocal of the active symbol duration to avoid corruption of the differential coding and to avoid inter carrier crosstalk e The FFT processing window must be timed to start at the point in each total symbol which makes the greatest constructive use of receivable multipath signals Synchronization of a DAB receiver is performed in several steps e Coarse time or
49. of time can be produced which represent the composite OFDM signal for one symbol In practice this number is minimized in order to constrain the demand for processing and the fundamental limit is set by the so called Nyquist criterion the time sampling rate must be 17 1 5 Orthogonal Frequency Division Multiplexing OFDM at least twice the frequency of the highest frequency component represented in the function of time to avoid aliasing distortion The period of time over which the whole calculation is performed can be called the processing time window and it is a requirement for correct operation of the iDFT that the duration of this window and the interval between the oscillator frequencies should have a reciprocal relationship In this case the required time window i e symbol duration is 1 ms so the oscillator frequencies are separated by 1 kHz The required number of carriers 1536 defines the highest frequency represented in the function of time so the minimum time sampling rate is then defined The iDFT process is repeated for subsequent symbols in each case with a new set of values in the array of modulation signals It is relatively straightforward to implement this transform in a computer program or by means of digital hardware of some other form where all of the sample values are represented by digital numbers Furthermore the availability of fast ADCs and DACs means that nowadays it is entirely practi
50. products are added together To recover the second modulation signal the second frequency samples of the oscillator signals are used and so on As before the minimum time sampling rate is set by the Nyquist criterion and the processing window duration 1 ms must be equal to the reciprocal of the carrier frequency separation 1 kHz Thus a fixed number of samples of the function of frequency can be produced which represent the modulation signals recovered from the individual carriers The process is then repeated for subsequent symbols to yield the subsequent sets of modulation signals It is within the scope of modern VLSI technology to perform this decomposition process within one integrated circuit and of course in such a fully digital system as DAB it is unnecessary to provide additional ADCs and DACs purely for these tasks In practice an algorithm ie a means for performing a computation which yields the same result is used to perform the DFT in the receiver and this is known as the Fast Fourier Transform or FFT An inverse FFT is used to generate the OFDM signal in the transmitter The advantage of the FF T is increased processing speed for a given level of complexity n FFT operates with complex numbers in digital form which represent the amplitudes and phases of its sampled input and output signals the multiplications to which the last two sections have referred are actually complex multiplications 21 1 5
51. shown that 29 1 6 Fast Fourier Transform FFT if a sine wave is expressed as a purely imaginary function i e it is presented to the Q port and zero is presented to the I port the resulting transform is purely real It follows that for an inverse FFT to output a single sine or cosine wave it must be presented with frequency domain data for the negative frequency component as well as for its positive frequency counterpart and the relationship between these data and their real imaginary status must follow certain rules 6l 1 6 9 Linearity of the transforms At first sight this would appear to imply that an N sample FFT could only provide N 2 useful frequency domain samples so the 1536 carriers used by the DAB system would require the use of a 4096 sample FFT and inverse However there is a simple way to halve this requirement by exploiting a useful property of the FFT and its forebears that of linearity If two time domain waveforms are added and the sum is transformed the result is the sum of the two corresponding spectra By reversing the sign of one of the input waveforms the result is the difference between the two spectra This property also applies in reverse to the inverse transforms Therefore if a real cosine wave and an imaginary sine wave of equal amplitudes and the same frequency f are complex added and applied to an FFT the result is one positive impulse function at minus f at the real output port multi
52. string of bits and f be a function defined as f s reverse s 83 3 6 Time De Interleaving out qo briso Brigo brizo rizo Dreimo Broo Droo b s 0 Dr 7 0 Dr 6 0 b so Diao bio boo bao Dro qi bis Or 6 1 bs ba Or 3 1 Or 2 1 b b q2 b 12 Or 102 bio bes Dr 72 Dr 6 2 bs ba Dr 32 Dr22 b bi q3 bass Dr 2 3 bas bes qa basa Or 12 4 bria br 104 Dr 9 4 Or 8 4 boa basa Dr 5 4 Dr 44 baa boa baa ba qs b ss Dr 4 5 bias bias Or 1 5 Or Qs bos r 8 6 bizs bis Dr s 6 Dr 4 6 bis bias Dr 1 6 Dr q7 bi 01 7 As biis Dr 13 8 biis bouis Dr 10 8 Dr 9 8 b ss bos Dr 6 8 Dr 5 g bias bass bos biis bis qo bio Dr 5 9 bas bis Dr 2 9 Dr 1 9 by quo D 10 10 Dr 9 10 Drsio benio Dr 6 10 Dy 510 D 4 10 bio Dr 2 10 Dr 1 10 Dro Qu bau ua bu qu buo Danae Beto12 ron Desa Derio Deo Dres12 raro beso Dio bus bus qis bas ras bes baa bus qiu Desa bona Droa Desi Dada Besta Deana Dera bua q s bus Figure 3 8 Time De Interleaving data structure where reverse s is the original string put in reverse order So for example reverse 1010 0101 The bit b are delayed by reverse i frames For example in mode I the frame du ration is 96ms Thus bit 0 will be transmitted without delay bit no 1 will be transmitted with a delay of 96 ms bit no 2 will be transmitted with a delay of 2 96 ms and so on until bit no 15 will be transmitted with a delay of 15 24 ms A buffer is required to store the bits that will be transm
53. that it may be transmitted via radio Modulation results in shifting the signal up to much higher frequencies radio frequencies or RF than it originally spanned A key consequence of the usual double sideband amplitude modulation AM is that usually the range of frequencies the signal spans its spectral bandwidth is doubled Thus the RF bandwidth of a signal measured from the lowest frequency as opposed to 0 Hz is usually twice its baseband bandwidth Steps may be taken to reduce this effect such as single sideband modulation the highest frequency of such signals greatly exceeds the baseband bandwidth 241 Power frequency Signal at baseband Signal at RF radio frequency Figure 4 3 Comparison of the equivalent version of a signal and its AM modulated RF version showing the typical doubling of the occupied bandwidth 4 1 2 DAB as a Real Time Application A system is said to be real time if the total correctness of an operation depends not only upon its logical correctness but also upon the time in which it is performed This was one of the main problems of the DAB implementation on the FPGA If the demodulator were not fast enough for processing all the OFDM symbols within the duration of one frame some data might be lost Recall that the duration of one frame for DAB mode I is 96 ms Due to this time constraint some problems arose once we tested the code on the FPGA 4 1 3 DAB hardware definition Once all th
54. that the values of some bits are flipped those close to the quadrant boundaries thus corrupting the information We shall see that even in this case we can recover the original signal with error correction codes 3 5 2 Soft Decision For a transmission channel which introduces only Gaussian noise such as a satellite down link it is found that a 3 bit representation is optimum giving 2 dB improvement over hard decision In this case an infinite number of bits would only increase the improvement to about 2 25 dB When the channel also introduces fading which cannot be removed by conventional AGC there can be value in increasing the resolution to 4 bits A multi bit representation of the demodulator output signal contains an indication of 82 3 6 Time De Interleaving the confidence with which each bit was demodulated For example with a three bit word given 000 it can be inferred with great confidence that a 0 was transmitted but 010 suggests that it was probably a 0 etc The two less significant bits of each word can be considered as a weighting factor and this can be applied in decisions made within the Viterbi decoder A neutral value half way between 000 and 111 would indicate no confidence at all or zero weighting so no decisions should be made on the basis of this word ll 3 6 Time De Interleaving We have already seen the functionality of the frequency interleaving For mor
55. where 1 1 Co 3 Cn 3 an ibn y Cn 3 an ibn So far we talked about orthogonality of vectors in IR and we a little effort we extended 155 A 4 Fourier Transform in Periodic Complex Functions the definition to functions from IR gt R We recapitulate here the definition of orthogonal ity of vectors in a vectorial space by defining it in its most abstract form Let V a vectorial space with inner product denoted by lt gt V gt V Two vectors v v V are orthogonal if lt v v gt 0 The inner product in C is defined as follows Let 2 2 C then lt 2 2 gt zz where z is the complex conjugate of 2 With these definitions at hand we are now ready to define orthogonality of functions C C Let fxg C C C C f g are orthogonal in J if f t g t dt 0 A 15 tel From this it follows that the set of functions e n z is orthogonal in the interval 0 27 Assume n Z m 2n ER 2n j git gimt dt p eiMte imt dt 2 0 0 cos nt isin nt cos mt isin mt dt 0 0 On the other hand if n m Ja cos nt isin nt cos mt sin mt dt 2x 40 0 The coefficients c can be calculated as we did in the case of real valued functions Recall that f t 5 Cne 2T oo lt fee M cnet e dt 0 n 0o oo 27 ye on emm t q 0 n 00 156 A 4 Fourier Transform in Periodic Complex Functions 5 en f Er m t isi
56. 0101 Additional code need to be added to reverse the ordering of the decoded output to exactly reconstruct the sample input H 1 9 2 De puncturing The addition of redundancy to the signal increments the data bandwidth in a 400 To palliate this effect some bits of the outgoing signal are removed Puncturing is the process 46 1 10 Digital Signals Processors DSP of removing some of the parity bits after encoding with an error correction code This has the same effect as encoding with an error correction code with a higher rate or less redundancy However with puncturing the same decoder can be used regardless of how many bits have been punctured thus puncturing considerably increases the flexibility of the system without significantly increasing its complexity In DAB a pre defined pattern of puncturing is used in the encoder The inverse operation known as depuncturing is implemented by the decoder 01 If the code word which results from puncturing is identical to that produced by a ded icated non punctured encoder of the same rate there is no loss of error correction perfor mance However if a large range of different rates are required in practice puncturing can cause a slight loss of performance relative to a non punctured code of the same rate this depends on the choice of generator polynomials and which bits are punctured 1 10 Digital Signals Processors DSP Traditionally during the analog era the common wa
57. 27 DAB Transmission mode independent description Each transmission frame is divided into a sequence of OFDM symbols each symbol consisting of a number of carriers The Fast Information Block FIB and the Common Interleaved Frame CIF are intro duced in order to provide transmission mode independent data transport packages associ ated with the FIC and MSC respectively The data carried in the MSC is divided at source into regular 24 ms bursts correspond ing to the subchannel data capacity of each CIF The CIF contains 55 296 bits divided at 864 capacity units 64 bits each Fast Information Block FIB is a data burst of 256 bits The sequence of FIBs is carried by the Fast Information Channel FIC The structure of the FIB is common to all transmission modes The synchronization channel symbols comprise the null symbol and the phase reference symbol The null symbols are also used to allow a limited number of OFDM carriers to convey the Transmitter Identification Information TII Both the organization and length of a transmission frame depend on the transmission mode The Fast Information Block FIB and the Common Interleaved Frame CIF are introduced in order to provide transmission mode independent data transport packages associated with the FIC and MSC respectively 104 3 10 Transport mechanisms Table 3 28 gives the transmission frame duration and the number of FIBs and CIFs which are associated with each transmiss
58. AYER il BIT STREA LEFT AUDIO CHANNEL RIGHT AUDIO CHANNEL Y DECODING OF E BEEN n DECODING OF BIT ALLOCATION BIT ALLOCATION DETERMINATION OF COMBINED sus SANDS DECODING OF DECODING OF SCFSI SCFSI DECODING OF BIT ALLOCATION OF coveweDsuesawos DECODING OF DECODING OF SCALE FACTORS SCALE FACTORS REGUANTIZATION OF SUS BAND SAMPLES OF REQUANTIZATION OF COMSRED SUBBANDS REQUANTIZATION OF SUB BAND SAMPLES SUB BAND SAMPLES A SUB BAND SYNTHESIS SUB BAND SYNTHESIS FILTERING FILTERING OUTPUT OF OUTPUT OF PCM AUDIO SAMPLES PCM AUDIO SAMPLES Figure 3 34 General MPEG Audio Layer II stereo encoder flow chart 3 11 1 Formatting of the audio bit stream The frame format of the audio encoder shall take the bit allocation ScFSI Scale Factor selection information ScF and the quantized sub band samples together with header in formation and a few code words used for error detection to format the MPEG Audio Layer II bit stream It shall further divide this bit stream into audio frames each corresponding to 1152 PCM audio samples which is equivalent to a duration of 24 ms in the case of 48 kHz sampling frequency and 48 ms in the case of 24 kHz sampling frequency The principal 112 3 11 Audio Coding structure of such an MPEG Audio Layer II frame with its correspondence to the DAB audio frame can be seen in figure 3 35 am MPEG Audio Layer Il frame
59. BSP pause the real time execution you might lose the synchronization with the transmitter In this case after watching at the desired state elements the program might be reset which means that the data structures of the demodulator must be properly initialized and another coarse synchronization must be performed 4 1 4 Components Description What follows is a technical description of the hardware components required to demodulate the radio signal 4 1 4 1 Maxim 2172 v1 0 25 The MAX2172 direct conversion to low IF tuner for Digital Audio Broadcast is designed for digital audio broadcast and terrestrial digital multimedia broadcast T DMB applications covering an input frequency range of 168MHz to 240MHz VHF IIT 1452MHz to 1492MHz L band and also 87MHz to 108MHz FM The MAX2172 achieves a high level of com ponent integration allowing low power and tuner on board designs The direct conversion to low IF architecture eliminates the need for an IF SAW filter while providing a balanced 2 048MHz center frequency baseband output to the demodulator Figure 4 5 MAX2172 Tuner A sigma delta fractional N synthesizer is incorporated to optimize both close in and wideband phase noise performances for OFDM applications where sensitivity to both 1kHz phase noise and wideband phase noise related to strong adjacent can be a problem P1 This component allowed us to select the different channels of the digital radio broad casters
60. Cycle n 3 Cycle n 4 fetch instruction fetch instruction fetch instruction fetch instruction fetch instruction 1 1 i2 k k 1 k 2 decode instruction decode instruction decode instruction decode instruction i itl k k 1 execute 1 1 execute i execute k The processor has to wait for the correct instruction to be fetched and decoded it is idle for a few cycles it stalls In the figure above it is as if instruction i takes three cycles to execute Because of pre decoding in the fetch stage not all flow control instructions take 3 cycles to execute because the fetching of the correct instruction k can be started a cycle earlier Also when a branch is not taken no jump the pre fetching scheme will work which means that the number of cycles for a conditional branch is 1 if not taken when the condition code evaluates to false and 3 if the branch is made when the condition code evaluates to true l 2 9 2 Delay Slots There is an alternative to stalling the processor in case of flow control instructions This is called executing delay slots What happens is that the instructions that have already been fetched are not flushed but decoded and executed In cases that useful instructions can 65 2 10 Hardware Loops be inserted into these delay slots there is a benefit The following diagram shows what happens Cycle n Cycle n 1 Cycle n 2 Cycle n 3 Cycle n 4 fetch instruction fetch instruction fetch instruction fetch instruction fetch
61. Fourier Transform from discretized input signal This function can be called by typing fft inputdata Recall that the Fourier Transform returns a list of complex numbers Therefore we cannot plot the amplitude versus frequency directly in a two dimensional plane In order to obtain the amplitude within each frequency present in the signal we must take the module of the complex numbers in the output list So by calling abs fft input data we obtain the last graphs I exhibit here the code in MATLAB to generate both the time and frequency domain functions clear all N 100 T 3 4 P 10000000 t 0 N 1 N t txT f 3xsin 2x pi 101 p abs f F0 f N 2 p p 1 N 2 freq 0 N 2 1 T subplot 2 1 1 plot f t axis 5 5 0 3 4 subplot 2 1 2 plot freq p axis 0 20 0 4 165 A 8 Applications of the Fourier Transform A 8 Applications of the Fourier Transform A 8 1 Noise Filter As a first example we will apply the Fourier Transform to one dimensional waves like sound I will show how to restore signal affected by gaussian white noise Let s plot a wave together with its spectra dE 0 2 4 6 8 10 12 14 16 18 20 Frequency Hz Figure A 5 Noiseless signal example and its spectrum The following code generates the last function 166 A 8 Applications of the Fourier Transform dt 1 100 et 4 t 0 dt et y 3 x sin 1 x 2 pi x
62. If the average power is greater than the power reference the gain factor 74 3 1 Automatic Gain Control is decreased by a small amount On the other hand if it is lower then the gain factor is increased Note the word slightly in the last sentence This is because in order to avoid the gain factor to be oscillating over and over again we define a range of values for which no change of the gain factor is performed Typically this range is around 0 9 x Power Reference 1 1 x Power Reference With all this at hand we can write the pseudo code for the AGC while true do AvgPwr 0 gain 1 pwrRef 12dB tav 1 25 for i 0 to WindowSize do current IQ Sample current IQ Sample x gain AvgPwr tav x IQpwr 1 tav x AvgPwr if AvgPwr lt 0 9 x PwrRef then gain 1 1 x gain end if if AvgPwr gt 1 1 x PwrRef then gain 0 9 gain end if end for end while As an example of the workings of the last algorithm in picture 3 2 we show an example of a real DAB signal received at NXP Offices in Leuven It turned out that the power of the incoming signal was not very good to be subsequently processed Nevertheless as can be seen the AGC performs well its task and amplifies the signal to a amplitude such that it can be processed with all the available resolution in the CoolFlux architecture 24 bits For a better understanding of the AGC block we show in picture 3 3 the average signal power of the las
63. Orthogonal Frequency Division Multiplexing OFDM A principal difference from the DFT is that the number of time samples must be equal to the number of frequency samples This means that if all of the available samples are used then the time sampling rate can only just satisfy the Nyquist criterion In most practical implementations of an FFT the number of time or frequency samples is made equal to 2 raised to some power so a 2048 sample FFT is used to process the 1536 carrier DAB signal In that case some headroom is provided against aliasing When an FFT processor is presented with a digital representation of a time domain signal i e of a voltage varying with time it has the effect of analyzing the spectrum of the signal and it outputs numerous baseband signals each corresponding to a particular range of input frequencies Each baseband signal represents the amplitude and phase of whatever component of the signal is present in that particular frequency range Thus the function of the FFT can also be visualized as that of a bank of band pass filters followed by frequency down converters The effective filter bandwidths are contiguous and are each equal to the reciprocal of the duration of the processing time window The centre frequencies of the pass bands are integer multiples of this reciprocal The frequency response of each filter has a sin f f shape where f represents the relative frequency with appropriate scaling By making t
64. Proyecto de Sistemas Inform ticos Curso 2009 2010 DSP BASED IMPLEMENTATION OF A DIGITAL RADIO DEMODULATOR ON THE ULTRA LOW POWER PROCESSOR COOLFLUX BSP Authors Elena P rez Toril Gracia Alvaro Mart nez L pez Project Supervisor Luis Pi uel Moreno Departamento de Arquitectura de Computadores y Autom tica Facultad de Inform tica Universidad Complutense de Madrid Contents 1 Preliminary concepts in DAB 2 11 Signal Modulation o e e a i iaceo cemana gut hikkaa a aa 2 1 2 Multipath Propagation e de rs b a e 5 1 3 QPSK Quadrature Phase Shift Keying llle 10 1 4 FDM Frequency Division Multiplexing 13 1 5 Orthogonal Frequency Division Multiplexing OFDM 14 1 6 Fast Fourier Transform FFT lee 23 1 60 14 Overview 26sec Rue AA ee A ee T ee y E eee de Redes 23 1 6 2 Digital Implementation o e e 24 1 6 3 The Discrete Fourier TransforM o 24 1 6 4 Computation of the DFT 0 25 1 6 5 The Fast Fourier Transform lee 26 1 6 6 Applications to DAB ice ae ee Se ba ae ae ee ES 27 1 6 7 Complex Numbers ux eR ee a ee Ae eee 28 1 6 8 Negative Frequencies 2 o 0 00000 0000000008 29 1 6 9 Linearity of the transforms 000 30 1 6 10 Combination oflandQ 0 2 0 0 000 31 1 6 11 Addition of the Guard Interval o o 33 1 7 Finite Impulse Response Filters FIR
65. S units which can operate in different modes The modes for the rounding are e truncation simple biased truncation sign magnitude rounding simple biased balanced bias rounding The modes for the saturation are e saturation e wrapping The rounding and saturation is done on the entire word or on the two sub words de pending on the type of move scalar or packed The way the rounding and saturation works however is exactly the samel l The mode of operation is equal for both RSS units and is controlled by setting and clearing the sr b sr r and sr s bits in the sr status register The assignment operator determines which conversion 24 or 48 bit scalar or packed must be executed e 24 bit scalar move e 48 bit scalar move e c 24 bit packed move cl 48 bit packed move 59 2 5 Memories and I O 2 5 Memories and I O The CoolFlux BSP is a dual Harvard architecture and has separate X and Y data memories Each memory has 16 Mword of 24 bits In a single processor cycle each memory can be accessed separately so that up to 2 data items can be read or written in a single cycle There is a special mode in which the X and Y memories are combined logically into a single XY memory that stores and loads 48 bit data items Next to the data memories the I O of the CoolFlux BSP is memory mapped Also the I O space has 16 Mword of 24 bits From a software point of view I O behaves as a memory to which dat
66. SPs CoolFlux BSP does not offer a separate compressed 16 bit instruction set Wide instruction words often lead to high program memory use which in turn can increase chip size and power consumption The BSP has several features to keep code size low multiple operations can be encoded in a single instruction to reduce code size on tasks that are parallelizable and for sequential code the core supports an instruction mode that essentially allows two 16 bit instructions to be packed into a 32 bit instruction As shown in figure 4 1 the CoolFlux BSP includes two 24x24 bit multiply accumulate MAC units two 24 56 bit ALUs and one 24 bit ALU two 24 bit data memories and a 32 bit program memory and two address generation units The key difference is that the new core supports three modes of operation rather than just one scalar used in the CoolFlux DSP core SIMD and complex The mode is determined based on the class of instruction used and a status bit Program Memory X Data Memory Y Data Memory a a a A a 32 24 24 24 24 24 a a a a Y v v Interrupts JTAG DMA yo Figure 4 1 CoolFlux BSP architecture 117 4 1 An overview to CoolFlux BSP In SIMD mode each MAC unit or ALU is split such that it executes two 12 bit operations the 56 bit ALUs can also perform dual 28 bit operations In complex mode the processor treats input data as a complex number with real and imaginary components which can be 12 or 24 bi
67. User Manual IP Programmer for CoolFlux BSP November 2008 35 Wikipedia JTAG 26 June 2010 http en wikipedia org wiki JTAG 194 Bibliography 36 87 38 39 40 41 42 43 44 45 Wikipedia Automatic gain control http en wikipedia org wiki Automatic gain control Wolfgang Hoeg et l Digital Audio Broadcasting Principles and Applications of Digital Radio Second Edition Wiley Press 2003 Zonst Anders E Understanding the FFT a tutorial on the algorithm and software for laymen students technicians and working engineers Citrus Press 2000 Zonst Anders E Understanding FFT applications a tutorial for students and working engineers Citrus Press 2004 Villanueva Diez Ignacio Apuntes de Ampliacion de Calculo Universidad Complutense de Madrid Rafael C Gonz lez Richard E Woods Digital image processing Prentice Hall 2007 Richard Burden J Douglas Faires An lisis num rico International Thomson Editores 1998 Luis V zquez Mart nez et l M todos num ricos para la F sica y la Ingenier a McGraw Hill 2009 Stein E M and G Weiss Introduction to Fourier Analysis on Euclidean Spaces Princeton University Press Smith Steven W Chapter 8 The Discrete Fourier Transform The Scientist and En gineer s Guide to Digital Signal Processing Second ed San Diego Calif California Technical Publishing 1999 195
68. a can be written and from which data can be read ll 2 5 1 Individual Data Memory X Y The CoolFlux BSP has 2 separate 16 Mword 24 bit X and Y data memories Typically accesses to both memories are used in parallel for intense BSP code so as to provide sufficient memory bandwidth for the computations being performed l l 2 5 2 Dual Precision Memory XY There are dual precision load and store instructions which can move a single 48 bit data item long to and from the memory sub system For these instructions both X and Y memories and both Xbus and Ybus move buses are combined used in parallel In the memory the corresponding locations in both X and Y memory need to be reserved since in this mode both memories get the same address This memory is referenced as XY memory It uses the X addressing unit for addressing and has the same addressing possibilities as the X memory 1181 2 5 3 I O Memory Space The I O for the CoolFlux BSP is memory mapped I O space is 16 Mwords 24 bit wide The I O memory space can be addressed as the Y memory space using the Y addressing unit to generate its addresses 5l 60 2 6 X and Y Addressing Units 2 6 X and Y Addressing Units The X and Y addressing units generate the addresses for the memory accesses The following addressing modes are supported e direct addressing e indirect addressing with post modification e indexed stack pointer addressing Indirect addressing is suppor
69. act in the industry of communications The development and improvement of techniques of Very Large Scale Integration VLSI has stimulated the development of more powerful smaller faster and cheaper computers for specific applications This technology has made it possible to implement highly sophisticated digital systems capable of performing digital signal processing that previously were done by analogical and more expensive means The high pressure from the consumers of more and more multimedia applications in their mobile devices has given rise to a growing interest for embedded specialized architectures of low power consumption This means that achieving the maximum efficiency with the lowest power consumption and of course at the lowest cost has become one of the biggest challenges of the companies in the sector of DSP 141 5 0 4 Start point For the development of this project it has been essential a big effort from the members of the team in order to understand the concepts that it englobes Previously to the development of this project it was necessary a deep study of both digital signal processing not covered in the studies in Computer Engineering and its applications to DAB and the CoolFlux BSP architecture Nevertheless thanks to the help of experts in the field we have been able to achieve the initial expectations In the following sections we explore the results obtained during the implementation and we conclude with the m
70. adena de bits resultante se hace pasar por los m dulos del depuncturing viterbi y dispersi n de energ a Al finalizar este proceso se debe extraer la informaci n del FIC n mero de subcanales direcciones de comienzo y final de stos etc para poder proceder con la demodulaci n de un subcanal el audio elegido por el usuario El usuario entonces selecciona el canal que desea escuchar y se procede a procesar el MSC ste contiene el audio de la emisora seleccionada La demodulaci n del MSC difiere ligeramente de la del FIC en que ahora adem s del depuncturing viterbi y dispersi n de energ a tenemos que realizar un deinterleaving temporal de los bits Este proceso es costoso en t rminos de memoria y tiempo Finalmente introducimos la cadena de bits resultante al decodificador de MP2 Este ltimo almacenara en un buffer de salida del sistema los bytes de audio que deber reproducir el cliente del CoolFlux B 12 Conclusiones El mercado de los receptores ha estado hasta el momento dominado por receptores est ticos o destinados a su integraci n como accesorios de autom viles Los receptores port tiles son los que presentan un mayor problema de dise o debido a la limitaci n que han de presentar tanto de tama o como de consumo el ctrico Una de las motivaciones que llevaron a cabo la planificaci n de este proyecto por parte de la compa a fue la ausencia de receptores port tiles de bajo consumo actualmente en el mercado La pro
71. akOut signals and BIO for the BreakIn signals all instances of the CoolFlux BSP are interconnected and they all stop when one of the instances has hit a hardware breakpoint 4 1 8 Interrupt interface Each CoolFlux BSP instance has an interrupt controller All interrupts go directly to the interrupt controller and it ensures that all interrupts are handled correctly and follow the priority scheme The priority is taken when more then one interrupt is waiting to be handled Once an interrupt is ongoing it will not be interrupted by a higher priority interrupt unless explicitly written in the software The interrupt controller detects a rising edge on the interrupt lines and the duration of the interrupt cab be as small as one clock cycle Each BSP can generate an interrupt in its neighboring BSPs by writing to a certain I O location The interrupt is generated independently of what value was written The table 4 12 indicates the mapping Interrupt Line From BSPO From BSP1 From BSP2 To BSPO NA 0x20 NA To BSP1 0x21 NA 0x20 To BSP2 NA 0x21 NA Figure 4 12 CoolFlux DAB Interrupt generation IO addresses 4 1 9 Shared Buffers We need some mechanism for communicating the input customer specific circuitry with the first BSP and the BSPs between each other To deal with this problem we were provided with a VHDL instance consisting of a couple of alternating shared buffers as shown in picture 4 13 132 4
72. all software devel opment tools including C compiler instruction set simulator and debugger can be created instantaneously from the IP Designer retargetable tool suite This ASIP specific SDK can then be distributed to the community of system integrators who want to implement their application software on the newly designed ASIP core The SDK is identical to the one in IP Designer except that the processor model is read from a shared library or DLL rather than from nML source code Two applications included within the tool suite provided by Target Compiler Technolo gies have been extensively used during the development of the DAB demodulator e Chess The C compiler It directly produces an executable in machine code Directives added to the C code allow optimization e Checkers The simulator which simulates the machine code and produces diagnostics profiling info 69 2 13 BSP C Compiler A library of building blocks for complex arithmetic SIMD acceleration FFT and modem functions is available To allow bit true simulation on the host machine of the C code intended for the CoolFlux BSP a library has been developed which supports the CoolFlux BSP types operations and intrinsic functions that is functions that have a direct efficient hardware implementation When this library is included the C code for the CoolFlux BSP can be compiled with a standard C compiler under Windows or linux This allows you to simula
73. application we must choose a finite value of n The question we pose now is given a signal that represents a physical quantity for example a fragment of sound how do we get the amplitude and phase of each frequency present in the signal This is an important question since the energy of a wave is a magnitude that depends both on the frequency and the amplitude of the wave So the energy carried by a signal is equal to the sum of the energy of each subsignal this result is derived from the Parseval Theorem found in advanced texts in this topic Therefore we could obtain how much energy is carried in the signal due to the presence of a given frequency We wish an algorithm that given a signal and a frequency returns the amplitude of that frequency in the signal If the frequency is not present in the signal the amplitude should be 0 148 A 2 Approximation by the minimum mean square error A 2 Approximation by the minimum mean square error As we said our intention is to approximate a function by means of other function which is more useful to the problem at hand But let s be more formalists about what approximation means in the context of Fourier Analysis When approximating a function we want to reduce the mean square error between the original and approximated function Intuitively we want the graphs of both function to be as close as possible For the sake of simplicity be
74. ardware s lo en el primero de ellos Obs rvese que la memoria total de la FPGA era pr xima a 45KB de modo que el espacio utilizado tuvo que ser reducido ya que todo la memoria ten a que ser dividida y compartida por los tres m dulos B 11 Descripci n de los m dulos BSP B 11 1 BSP Sincronizaci n y buffering de muestras Este es el m dulo responsable de la tarea m s compleja en t rminos de procesamiento de se ales la sincronizaci n de la se al de entrada La entrada de este BSP proviene de la se al de radio y los datos resultantes son envi ados al siguiente BSP Lo primero de todo es inicializar la estructura del demodulador y despu s se habilitar las interrupciones Si se ejecuta una interrupci n antes de inicializar apropiadamente los par metros del sistema pueden surgir datos incorrectos Al comienzo de las pruebas del sistema en tiempo real surgieron algunos problemas debido a que nuevos s mbolos de entrada sobreescrib an a los anteriores sin que stos hayan sido procesados Para evitar esto la primera soluci n propuesta fue incrementar el tama o del buffer de entrada a 16K Words pero debido a las restricciones de memoria de la FPGA ste era un tama o intolerable Una soluci n mas eficiente fue crear una m quina de estados De este modo en lugar de trabajar con un buffer auxiliar se utiliza el propio buffer de entrada al sistema para reescribir los datos de salida del AGC y como entrada al algoritmo que impl
75. ariable Fourier Analysis will allow us to think of a signal as a magnitude think of it as information varying in frequency the so called frequency domain This may sound quite bizarre right now but my intention is to change this idea in the chapters that follow A natural question is why we do ever want to deal with a signal in the frequency domain Let s say that signals might interact with other systems in the Universe in one way or another depending upon the frequencies i e subsignals contained within the signal The way a signal interacts with any other system is known as the frequency response of the system For example the sound emitted by two different instruments can be distinguished in part because they have different frequencies and as a consequence of that both signals interact differently with your ear and brain If we are to code a sound in a computer we will obtain a lot of advantage by filtering those frequencies in the record that the brain cannot perceive presumably reducing the size of the file After all why do we want to store information that cannot be perceived Take the following step function defined as 1 si lt T 1 SILT It might not seem clear that it is composed of sinusoids but hold on Let s plot the sine function f t sin t Note that the frequency f is jr Hz The following three pictures show in turn the representations of the signals 146 A 1 Motivation
76. as a reasonably reliable estimate for the ideal 37 1 9 Errors Detection and Correction transmitted signal in MER calculation It is defined in dB as MER dB 10x logy Primal l Perroi where Pignal is the power of ideal transmitted signal and Perror is the RMS power of the error vector This value was around 29dB and it is the typical MER figure quoted for digital radio systems such as DAB 1 9 Errors Detection and Correction 1 9 1 Viterbi Decoder In telecommunication a convolutional code is a type of error correcting code in which each m bit information symbol each m bit string to be encoded is transformed into an n bit symbol where m n is the code rate n gt m and the transformation is a function of the last k information symbols where k is the constraint length of the code Convolutional codes are used extensively in numerous applications in order to achieve reliable data transfer including digital video radio mobile communication and satellite communication To convolutionally encode data start with k memory registers each holding 1 input bit Unless otherwise specified all memory registers start with a value of 0 The encoder has n modulo 2 adders and n generator polynomials one for each adder see figure 1 21 below An input bit m4 is fed into the leftmost register Using the generator polynomials and the existing values in the remaining registers the encoder outputs n bits Now bit shift all regis
77. ative frequencies from 1023 kHz to 1 kHz Of these active data are applied to the 1536 surrounding 0 Hz that is those representing 1 kHz to 768 kHz and 1 kHz to 768 kHz and static zeros are applied to the remainder The inverse FFT then produces 2048 time domain samples which are output to the DACs within 1 ms Following up conversion by the quadrature modulation system carriers appear at frequencies from fo 768 kHz to fo 768 kHz The input data are changed from symbol to symbol giving the effect of QPSK modulated carriers Of course discontinuities occur when the signal is re configured abruptly at the symbol boundaries and this changes the fine detail of the spectrum If the modulation data 32 1 6 Fast Fourier Transform FFT are random then over many symbols the power spectrum of each modulated carrier takes on a sin f f 2 distribution as described in the main text In this approach the sample which represents 0 Hz in the baseband signals is not used If data were input to this sample the corresponding I and Q output signals would be static DC voltages and if the mixers in the quadrature modulation system could handle such signals the output signal would be a wave at the local oscillator frequency However the accuracy with which the phase of this wave could be controlled could be compromised by drifts in the DACs and the mixers themselves so this zero carrier is not used to carry modulation The mirr
78. bstracto la entrada del demodulador DAB consiste en una se al modulada OFDM DQPSK representada num ricamente por un conjunto de n meros complejos Generalmente la salida resultante del proceso de demodulaci n es el sonido real aunque tambi n es posible obtener otros tipos de datos como informaci n textual sobre predicciones del tiempo el tr fico etc La demodulaci n DAB est dividida en una cadena de m dulos cada uno de los cuales tiene una funcionalidad diferente El objetivo de este cap tulo es explicar la funcionalidad de cada uno de los m dulos junto a su implementaci n en la arquitectura CoolFlux BSP La siguiente figura ilustra la concatenaci n de los diferentes m dulos 173 B 2 Control Autom tico de Ganancia coo an MC OA CO Time Deinterleaving Fic Data Differential Freq demodulator Deinterleaving o E Depuncturi VT Convolutional Energy Fic Data Dispersal FIC Decoding Unscrambling Encoder BER Mp2 gt Fic Info Audio Signal CIF Decoding Figure B 1 Cadena de demodulaci n Mac Data B 2 Control Autom tico de Ganancia Este sistema conocido en ingl s por las siglas AGC Automatic Gain Control lo podemos encontrar en muchos dispositivos electr nicos Su funci n es ajustar el nivel de potencia de la se al de entrada para aprovechar la m xima resoluci n que ofrece la arquitectura 24 bits de precisi n simple y 48 bits de precisi n doble Es
79. cable to carry out symbol by symbol processing in real time With modern VLSI technology the complexity i e the number of carriers is of relatively minor importance The recovery of the modulation signals from an OFDM signal i e decomposition of the OFDM signal is rather less straightforward but essentially this follows from the generation process by interchanging time and frequency It was mentioned earlier that integrating over each symbol is an efficient way to determine the modulation state of a carrier and an extension of this principle provides a useful starting point To simplify the explanation the receiver should initially be visualized as containing a large bank of local oscillators mixers i e multipliers and integrators although as before such an arrangement is not used in practice Each oscillator mixer combination acts as a demodulator in the manner of a direct conversion radio receiver The incoming signal is fed equally to all of the demodulators and each of these is followed by an integrator Each integrator operates over a limited period of time before yielding a result This is illustrated in Fig 1 16 Each modulated carrier is demodulated by the mixer which is fed with a local oscillator signal of the corresponding frequency but since the spectra of adjacent modulated carriers 18 1 5 Orthogonal Frequency Division Multiplexing OFDM spectrum of incoming composite signal i 1 X l X
80. ces constant5 and constant6 shown on the last scheme A pulse generator is necessary as an input for the Analog Digital Converter as we can see at the top left of the image This pulse generator marks the sampling rate of the input signal There is also a JTAG module for testing purposes later on we shall explain how it works IQ demod schema corresponds with the AGC algorithm and Null Detection The fol lowing module IQ input takes the IC signal coming from the AGC and writes the corre sponding samples into the specified address addr in These samples will be the input to the BSP Output regs contains a set of registers where the values at the output of the BSP are allocated They are connected to the scopes or else as an input to the BSP returning the address of the next available word As can be seen in the picture some scopes were added within the data path in order to facilitate the debugging process These scopes can generate a graph with the required signals Although they turned out to be quite helpful these are not enough for completely debugging the program The debugger provided by Target is essential in order to see the state of the DSPs essentially registers and memories The debugger itself allows us to pause the execution at any time and check what is inside of each word of memory in such moment or even dumping the content of the memories to a graph Of course whenever you 123 4 1 An overview to CoolFlux
81. code In our case we just rewrite the number of the symbol in the 1536th position into the buffer Following the same methodology an interruption is signaled to BSP to make it start running once this BSP is done Throughout the implementation of this module we had to face up some problematic issues concerning the processing time Gradually we added to the chain the modules il lustrated in figure taking into account the number of million of instructions per second MIPS each one contributed to the total time available for this core Thanks to the com piler we could further optimize the FF T and the quantization algorithm In this way we could fit the FFT the frequency deinterleaver the differential demodulator and the quanti 139 4 2 Description of the BSP instances zation to this BSP thus liberating the last BSP which was already quite overloaded from performing even more computations 4 2 3 BSP FIC and MSC Decoding Audio decoding This is the last core of the three and so at the end it must output the raw audio for being reproduced by customer specific circuitry This first thing this processor does is to buffer all the OFDM symbols that conform the FIC When the whole FIC data has been buffered it is subjected to further processing in turn the de puncturing viterbi and unscrambling modules Once the FIC has been demodulated we extract the actual information from it This information is stored in memory in an str
82. cy domain spectrum the equivalent time domain signal is produced which can be up converted to the final frequency and transmitted The changes of modulation states from symbol to symbol are effected by changing the numbers input to the array In the receiver from the incoming time domain 23 1 6 Fast Fourier Transform FFT signal the spectrum is re constructed as an array of numbers representing the individual modulated carriers from which their modulation states can be determined The link between the time and frequency domains is process of transformation Bearing in mind the number of carriers 1536 in Mode 1 it would be out of the question to perform this process in a domestic receiver using analogue circuitry e g banks of oscillators and filters The solution is to implement the transformation digitally and there are several possible approaches of which one of the most rapid is the FF T algorithm The treatment in the section 1 5 in practical terms show how the discrete Fourier transform DFT can be developed from a block diagram of the OFDM decomposition process I6 1 6 2 Digital Implementation It is important to impose time limits in order to limit the extent of the summation In practice the number of input samples which are available to be transformed may already be defined e g by the symbol duration in the case of DAB and this automatically imposes time limits It is convenient to think of this as the appl
83. d within the QPSK signal There is a tight and mysterious relation between sinusoids the number e 2 7183 and the imaginary number 7 To understand this let s introduce another serie reminiscent to that of Taylor for approximating a real function This is called the Maclauring s Serie and says that a function f t can be approximated as follows f t f 0 FO al oe a P d A 13 We know from our childhood that the derivative of e is equal to itself We can write e as a Maclauring s serie We know also that the derivative of sin t and cos t are almost equal to themselves The derivative of sin t is cos t whose derivative is in turn sin t and so on Next we expand the Maclauring s series for both functions 1 1 sin t 4 25 354 6 120 1 1 cos t 1 a 3i 154 A 4 Fourier Transform in Periodic Complex Functions Finally we unfold the function e We are now in position to see that cosines sines e 7 and i are all together related in the following well known equality e cos t isin t It follows that eit e it t cos t 5 et 3 e t es sin t 3i Our original Fourier serie 4 1 becomes now oo oo f t ao an cosnt b sinnt n 1 n 1 3 oo eint Lg int oo eint eint ao gt a T 1 E 2 2i n 1 n 1 Since 4 oo f t co J je tee n 1 oo oo o0 Co 1 c et 1 c ent 1 Cy e A 14 n 1 n n 00
84. da8 i k idct filaDCT 1 octavoSuperiorh ancho end for end for Now do it for columns altura size foto 1 mitadSuperior floor ancho 2 cuartoSuperior floor ancho 4 octavoSuperior floor ancho 8 fotoComprimida2 full fotoComprimida4 full FotoComprimida8 full for for k 1 3 do for i 1 size foto 2 do colsDCT2 dct double fotoComprimida2 i k colsDCT4 dct double f otoComprimidaA i colsDCT8 dct double fotoComprimida8 i fotoComprimida 2 full i k idet filaDCT 1 mitadSuperior altura fotoComprimida4 full i k idct filaDCT 1 cuartoSuperior altura fotoComprimida8 full 1 k idet filaDCT 1 octavoSuperiorh altura end for end for k k k 1 1 4 171 A 8 Applications of the Fourier Transform Finally we save the compressed pictures imvrite uint8 fotoComprimida2full Desktop pict2x jpg imwrite uint8 fotoComprimida4 full Desktop pictAx jpg imwrite uint8 fotoComprimida8 full Desktop pict8x jpg Whereas the original image occupied 253 KBytes we have reduced the size to 167 KBytes and 57 KBytes respectively 172 Appendix B Resumen en espanol del proyecto B 1 Implementaci n software de DAB sobre la arquitec tura CoolFlux BSP En este cap tulo explicaremos el proceso de demodulaci n de DAB junto a su imple mentaci n en CoolFlux BSP Desde un punto de vista a
85. dditionally some of these operations can be made conditional The YALU supports add subtract add and divide by 2 subtract and divide by 2 e In packed operation the XALU can perform the following operations e Add subtract add and divide by 2 subtract and divide by 2 e Divide step Absolute value e Compare operation e Logic and arithmetic shift e Conjugate operation e Packing operation e Unpacking operation e Conditional negate operation Note that for all these operations the result is the same regardless of whether the operation SIMD or a complex operation 18 2 4 Move Buses and RSS Units 2 4 1 Xbus and Ybus Move Buses Moves between the registers and also between the memories and the registers are essential because of the load store architecture of the CoolFlux BSP Two central buses are provided to move the data in parallel with the operations in the data path These buses provide a connection between all registers and between all registers and the memories and support the move of data e from a source register to a destination register 55 2 4 Move Buses and RSS Units e from a source register to a memory store e from the memories to a destination register load The buses support single and parallel moves In a single move operation a single 24 bit data item is moved in a single processor cycle With a parallel move operation two 24 bit data items are moved in a single processor cycle b
86. de up of discrete symbols also called bauds chosen 1 1 Signal Modulation carrier signal Carrier Signal Output output Figure 1 4 Amplitude Modulation and Frequency Modulation from a finite alphabet If the alphabet has 2 symbols then each symbol carries N bits of information The equivalent digital modulation schemes to those presented so far are known as PSK FSK figure 1 5 and ASK for which in turn finite values of phase frequency and amplitude are employed A variation of the first one called DQPSK will be our main concern later since it is the standard used in DABBI Data Carrier Modulated Signal Figure 1 5 Frequency shift keying Strictly speaking even though the information is digitally modulated the actual signal that goes out the transmitter is analog since nature does not allow a physical magnitude to change from one state to another in a finite amount of time without passing through all intermediate values An Analog to Digital Convertor ADC is a device that takes an 1 2 Multipath Propagation analog signal as its input and converts it to a digital representation in a process known as sampling 10 11 12 13 Figure 1 6 Sampling of an analog signal 1 2 Multipath Propagation Most existing means for radio communication which can use simple omni directional receiv ing antennas are affected adversely by multipath propagation par
87. decir este m dulo reduce la potencia de la se al de entrada si sta es muy fuerte y la aumenta si es muy d bil Inicialmente la se al de entrada se multiplica por un factor de 1 La potencia media se calcula durante un periodo fijo de muestras IQ Despu s de esto la ganancia aumenta o disminuye dependiendo de si la potencia media de la ventana de muestras es mayor o menor que la potencia de referencia 12 dB en el caso del CoolFlux Se necesita un filtro paso baja para evitar que el ajuste de la ganancia sea demasiado sensible a cambios bruscos en la se al de entrada Como ejemplo de un cambio brusco en la potencia de la se al de entrada tenemos el S mbolo Nulo cuyo prop sito es la sincronizaci n 174 B 3 FFT temporal del receptor Dicho s mbolo se caracteriza por tener una baja potencia en relaci n al resto de s mbolos que conforman la se al de radio Sin este filtro dicho s mbolo ser a amplificado perdiendo la propiedad de ser un s mbolo de potencia nula haciendo imposible su detecci n Una implementaci n software de un filtro paso baja puede ser como sigue AvgPur Tav IQpwr 1 Tav AvgPwr donde Qpwr es la potencia de la muestra IQ actual AvgPwr es la potencia media recibida en la ventana actual Tav es el par metro con el valor que queremos filtrar La siguiente gr fica muestra c mo la se al de entrada es amplificada progresivamente hasta que finalmente se estabiliza N tese la presencia d
88. decision offers better per formance results since it provides a better estimate of the noise i e less quantization noise is introduced In most circumstances the noise is strong enough just to tip the signal over the decision boundary If hard decision is used a significant amount of quantization noise will be introduced The quantized soft decision values are used to calculate the parameters for the Viterbi decoder Up to this point no Viterbi decoding has been performed Viterbi decoding begins after a certain number of encoded symbols have been received This length again is usually 41 1 9 Errors Detection and Correction Sample input pattern 1011010100 Time 0 1 2 3 4 5 6 7 8 9 10 Input 1 0 1 1 0 1 0 1 0 0 00 e e e e e e 01 e e e e e 1 9 e 1 e e e e e e e e e Encoded output bit pairs 11 10 00 01 01 00 10 00 10 11 Soft decision ideal 4 4 43 33 3 4 3 4 33 43 33 43 4 4 Soft decision nois 34 3 8 3 6 2 7 2 925 2 7 3 6 2 1 2 5 2 6 1 3 2 5 2 8 27 1 4 3 2 1 3 1 4 sy Soft decision 5 noise quantized 3 4 43 33 3 4 2 3 3 1 33 3 1 32 3 4 Branch metrics 00 7 1 6 1 1 4 0 4 1 7 01 1 7 0 7 5 2 6 2 5 1 10 1 7 0 7 5 2 6 2 5 1 11 7 1 6 1 1 4 0 4 1 7 State metrics 100 _ 93 f 86 108 107 115 128 128 138 d 137 158 00 e S 0 S 0 42 A X 7 132 qaNo 142 l Na as aa 01 0 o M39 44 i0 i T 2 N N s Ki A S 0 1 sx 109 100 a 122 E 120 139 sx 130 s
89. dependent demodulation and makes the greatest use of the available processing power lfl 1 6 Fast Fourier Transform FFT Please for a better understanding of this section refer to the appendix in which a mathe matical view of the Fourier Transform is given 1 6 1 Overview In developing DAB to combat multipath propagation attention has been paid to the effects of radio propagation in both the time domain and the frequency domain As noted in the first chapter these two domains provide different viewpoints for the same effects Equally when generating or receiving a signal certain aspects of the processing can be carried out more easily in one or other of the domains This is particularly true in the case of DAB where the multiple carrier RF signal is more easily synthesized and analyzed in the frequency domain but the symbol by symbol modulation is more easily treated in the time domain Indeed if a DAB signal is displayed on an oscilloscope it appears similar to band limited white noise punctuated by the null symbols every 96 ms in Mode 1 which gives little clue to the existence of multiple discrete carriers Successive stages of a DAB transmitter or receiver operate on constituents of the DAB signal in both of these domains In the channel encoder the spectrum of the signal is constructed essentially as an array of numbers each representing the instantaneous ampli tude and phase of one of the QPSK carriers From this frequen
90. detected the effect of the overlapping different symbols is to cause corruption of the data that is inter symbol interference For example figure 1 8 shows a symbol generated in the transmitter first row Let s suppose the transmitter is sending the binary string 01000 As we pointed out before the actual signal that travels through the air is not like a perfect step function but slightly smoother The symbol is echoed three times and arrives at the receiver as shown in the second row The receiver shall not receive 01000 but 01110 instead However rather than making a snap decision at one point in time during each symbol as implied in the figure in some cases it can be more efficient for the 1 2 Multipath Propagation detector to integrate the demodulator output signal over the whole of each symbol with respect to some timing reference which may be the direct signal This makes use of all of the received signal power In that case as long as the delay is less than the symbol duration the echo will carry the same modulation as the direct signal for some portion of integration period Undoubtedly this will have some effect because the demodulator output signal represents the phase of whatever is input in this case the vector sum of the direct and delayed signals However a method is available to prevent this from upsetting the operation of the detector that is differential modulation which will be describ
91. dic signal in order to recover the continuos signal from the discretized signal If this criterion is not satisfied we cannot recover the information contained within the original signal f from the sampled one a phenomenon known as aliasing Note that the frequency domain of a sampled version cannot have an infinite range 160 A 6 DFT of frequencies If fact if we sample the signal according to the Nyquist s criterion to a rate double that the maximum frequency of the continuos signal we shall not encounter frequencies bigger than half the sampling rate in the frequency domain Next I show the continuos signal f t 3 sin 27t and three sampled version with less and less sampling rate Note that in the case of the last one the information of the original signal is almost unrecoverable Everything said in the case of the continuos Fourier Transform applies equally well to the Discrete Fourier Transform For example Let f g R R f g are orthogonal if X f 0g t 0 A 25 teT where T to t1 gt ty 1 It can be shown that f t sin mt and g t sin nt where t to t1 tu 1 are orthogonal if and only if m n The orthogonality test is the same as that of continuous functions Finally for discrete complex valued signals f g are orthogonal if 161 A 6 DFT Y f 0g t 0 A 26 teT where T to ti tua To conclude with the theory
92. do que su equivalente hard decision B 7 Time De Interleaving Este m todo de entrelazado de datos se implementa para mayor tolerancia al ruido En lugar de enviar los datos en el orden temporal en el que se generan en el emisor se introduce una retraso en cada bit conforme a unas reglas precisadas en el est ndar De este modo hacemos 178 B 8 Energy Dispersal Unscrambling que bits que est n pr ximos en la cadena de bits original est n m s dispersos cuando viajan a trav s del canal El mismo retraso se aplica siempre a los bits que ocupan la misma posici n en distintos frames El algoritmo de Viterbi como ser explicado m s adelante es capaz de recuperar errores en la cadena de bits recibida aunque la longitud de la cadena corrupta no debe superar un cierto limite para que el algoritmo sea capaz de restaurar la cadena original Si alguna fuente de ruido afecta al canal durante un periodo de tiempo en la transmisi n de un frame s lo se ver afectado un grupo de bits no relacionados entre s Retrasando algunos bits en el emisor y reorden ndolos en el receptor evitamos que en el canal haya segmentos da ados mayores que la duraci n cr tica M s adelante el algoritmo de Viterbi ser capaz de recuperar los bits perdidos El time de interleaving s lo se lleva a cabo para los bits del MSC ya que los bits del FIC necesitan ser procesados m s r pidamente B 8 Energy Dispersal Unscrambling Para asegurar una disper
93. dro n data in adori_in datat in wet E Block sel addr inQ data ino wed addr int data ini wet data outi To From Memory Domain A To From Memory Domain B Figure B 8 Arquitectura de los buffers intermedios Cada uno de los buffers conecta 2 m dulos hardware en el dise o Mientras una memoria est siendo accedida por un m dulo hardware en el dominio de la memoria A la segunda memoria puede ser accedida por el segundo m dulo en el dominio de la memoria B En alg n momento de la ejecuci n este orden cambia accediendo por el primer m dulo hardware a la segunda memoria y viceversa Este cambio se debe a la se al block sel Cuando sta se al cambia su valor las conexiones entre las dos memorias y las dos interfaces cambian por lo que nunca habr un conflicto entre las memorias El estado de los bloques de memoria es oculto al software 185 B 11 Descripci n de los m dulos BSP El primer BSP comienza recibiendo los datos IQ de la memoria de entrada Las palabras impares almacenar n datos imaginarios y las pares guardar n valores reales de la se al Esto da lugar a 512 valores complejos Una vez cada 512 muestras IQ tiene lugar una interrupci n que realiza el cambio entre los dos buffers Cuando el primer BSP ha terminado se env a una se al al siguiente BSP para que comience su ejecuci n Cada BSP comienza a ejecutar cuando recibe una interrupci n del anterior o bien una interrupci n h
94. e software development process has concluded it is time to dump the code into the actual hardware to test it in real time Up until now the correctness of the code has been proved by means of the Target Simulator l First of all the processor data path 120 4 1 An overview to CoolFlux BSP must synthesized on the FPGA The hardware description language chosen for this task has been VHDL Special care was taken in considering the memory requirements The quantity of memory of an embedded system is a critical aspect that must be taken in mind Before mapping the actual memory into the FPGA we worked out the memory requirements for the design taking into account the total memory of the FPGA and the worst case of the demodulation process which corresponds to that of the maximum subchannel size at the maximum bit rate Once the FPGA was correctly configured and the software loaded on the next step was to connect the FPGA with the external world and the debugger A detailed description can be found below MATLAB Simulink proved to be an essential tool for easily interconnecting the FPGA with the rest of the hardware oscilloscope tunner debugger etc In picture 4 4 we can see the global structure of the architecture during the first iteration of development process 121 4 1 An overview to CoolFlux BSP 90 dsqy doy qep Bloje uiua L tyuezsuog zauezuog du z3vq 13s B S E poeu Giran E xn yezuog zava o Y i
95. e tolerance to burst noise a time interleaving of output data is also performed in the transmitter Instead of sending the bits in the order they are generated in the original data stream we introduce some rule for delaying each bit to transmit In this way we make that close bits in the original non time de interleaved stream that are likely related to each other be more dispersed If some noise affects the channel during a period of time whilst transmitting a frame only a group of non related bits will be actually corrupted In addition the Viterbi decoder see below which was one of the best methods to recover errors in a bit stream at the time of the DAB specification was written is able to repair only a corrupted bit stream lasting less than a critical duration By delaying some bits at the transmitter and reordering them at the receiver we avoid the channel to damage segments of the stream data of longer length than the critical duration The bits that compose a frame are each delayed by different amounts of subsequent frames The same amount is always applied to bits that occupy the same position in different frames Time interleaving shall be applied to the output of the convolutional encoder for all sub channels of the Main Service Channel MSC It shall not be applied to the FIC because this must be processed faster than the MSC The rule is the following Let s form groups of 16 bits for the current frame bo 51 015 Let s be a
96. e two measures are commonly used in digital modulation systems 1 8 1 Bit Error Rate This algorithm was implemented for measuring the accuracy of Viterbi decoding Since the Viterbi algorithm has to recover the missing information during the transmission it might be useful to know whether or not this information is reliable In digital transmission the bit error rate or bit error ratio BER is the number of received bits that have been altered due to noise interference and distortion divided by the total number of transferred bits during the studied time interval BER is a unitless performance measure often expressed as a percentage number As an example assume this transmitted bit sequence 0110001011 and the following received bit sequence 0010101001 The BER is in this case 3 incorrect bits divided by 10 transferred bits resulting in a BER of 0 3 or 30 The bit error probability p is the expectation value of the BER The BER can be considered as an approximate estimate of the bit error probability This estimate is accurate for a long studied time interval and a high number of bit errors In a communication system the receiver side BER may be affected by transmission channel noise interference distortion bit synchronization problems attenuation wireless multipath fading etc The BER may be improved by choosing a strong signal strength unless this causes cross talk and more bit errors by choosing a slow and robust
97. ected which lie in the range 256 to 1792 omitting all others and 1024 the centre carrier again This yields 1536 different values and 1024 is subtracted from each value giving 513 14 329 197 The 79 3 5 Quantization position of each value in this series gives the bit pair index 0 1 2 1535 and the value gives the carrier index so the first bit pair is directed to the carrier indexed 513 and so on Consecutive bit pairs are separated by widely varying amounts ranging from about 40 carriers i e their information is transmitted using frequencies separated by at least 40 kHz up to more than 1400 i e 1 4 MHz separation By this process the resulting dispersal in time of errors in the received de interleaved bit stream is limited to within one symbol block but errors are further dispersed by the time interleaving which is applied outside the frequency interleaving 6l 3 5 Quantization Once the differential demodulation has been carried on we must convert the resulting complex numbers to actual bits There exists two approaches to this problem namely hard decision and soft decision 3 5 1 Hard Decision First we present the hard decision algorithm which might be simpler for i 0 to 1535 do r real input i i imag input i if r 20 then output i 1 else output i 1 end if if i gt 0 then output i 1536 1 else output i 1536 1 end if end for
98. ective frequencies are different Let f g be two signals defined as f t sin mt and g t sin nt Hence 151 A 3 Orthogonality cos mt nt cos mt nt f t g t sin mt sin nt Firstly assume m n Thus 27k 27k 1 2nt cos 2n zu k an Therefore according to the definition they are not orthogonal On the other hand assume now that m z n then 20k i sin nt sin mt dt k 5 cosi nyt eos m moat 1 sin m n t sin m n t 3 m n m n E 20k i cos nt cos mt dt k eot n t cos m n t dt _1 sin m n t sin m n t 2nk 5 m n m n PRIM We almost have all the ingredients for computing the amplitudes of all the components in the Fourier Serie Finally we show under what conditions both a sine and a cosine are orthogonal Without loss of generality let f t sin mt and g t cos nt 20k f cos nt sin mt dt k 2nk 5 sin m 01 sin m n dt zx E 4n cos m ar 2 m n m n NM oa 152 A 3 Orthogonality Here we are we are ready to obtain the coefficients a b For commodity I rewrite here the Fourier Serie F f ao a1cos t a4sin t a cos nt b sin nt A 10 As n approaches infinity F f will be equal to f t lim f t F f A 11 n oo Therefore to obtain ag we integrate both sides of 4 10 27 F f dt ay2r A 12 0 therefore 1 2T ag f t dt 2m 0
99. ectrum analyzer to inspect the same signal The oscilloscope allows inspection of the way that a signal voltage changes as time progresses almost irrespective of the rate at which it is changing whereas the spectrum analyzer allows inspection of the content of the signal at different frequencies 1 2 Multipath Propagation i e rates of change almost irrespective of time Often different aspects of a phenomenon being investigated can be visualized more clearly in one or other of the two domains For example the vestigial sideband in a conventional AM television signal can be inspected using a spectrum analyzer but an oscilloscope gives no obvious hint to its existence If the effects of a phenomenon are described mathematically in each of the domains the two descriptions are related by the Fourier transform For example the frequency response of a filter is related to its time impulse response by the Fourier transform One effect of multipath propagation is to generate inter symbol interference When a direct signal and delayed echoes arrive at the receiving antenna there will be occasions when their modulation represents data from different symbols previous as well as present This is illustrated in figure 1 8 for the case of a single delayed signal although there may be many in practice Figure 1 8 inter symbol interference Returning to figure 1 8 when the combined signal is demodulated and the modulation state is
100. ed later Entirely different symbols overlap for the remainder of the integration so the degree to which the result is corrupted depends on the magnitudes of the echo signals and on their delays When a mobile receiver is traveling in a dense urban environment which imposes one of the most difficult multipath conditions short delay echoes are often received in greater numbers than long delay ones owing to multiple reflections of a relatively direct signal from local buildings and terrain Very long delay echo signals tend to have smaller magnitudes because they have travelled greater distances Therefore the majority of the collective echo power arises from short delay echoes so modulation schemes which use long symbols or low symbol rates provide the greatest tolerance to this effect because the proportion of each symbol that is corrupted is minimized We could measure the collective echo power received with different delays in time An idealized result is illustrated in figure 1 9 where the transmitted i e direct signal is shown as a single impulse in practice more complicated signals are used in such measurements to overcome the infinite bandwidth requirement of true impulses but the results can be presented in a similar fashion Power Delay Figure 1 9 Delay spread 1 3 QPSK Quadrature Phase Shift Keying By taking a large number of measurements in a particular type of environment the sta tistical distribution can
101. el s mbolo nulo cerca del punto 3 en el eje de abscisas y c mo el AGC ignora su amplificaci n Figure B 2 AGC amplificando una se al real B 3 FFT La transformada r pida de Fourier es un algoritmo fundamental en cualquier sistema de procesamiento de se ales No es el objetivo de este resumen en espa ol explicar para qu se utiliza la FFT en la demodulaci n de una se al OFDM Cons ltese el documento en ingl s para una exposici n detallada de la aplicaci n de la FFT a DAB Para el desarrollo del 175 B 4 Demodulaci n diferencial demodulador de DAB se probaron dos implementaciones de la FFT a saber la radix 2 y la radix 4 Nuestro trabajo consisti en analizar el n mero de MIPS empleados por cada una de ellas y comparar el error de precisi n que se obtiene en dichas implementaciones en punto fijo cuya precisi n es menor con distintas implementaciones en coma flotante que existen en la comunidad de software libre Para permitir una computaci n eficiente de la FFT el tama o del buffer de entrada debe ser una potencia de 2 La se al de DAB modo 1 es transportada por medio de 1536 portadoras por lo que el n mero de puntos de la FFT que se debe utilizar es de 2048 Los restantes puntos no se utilizan B 4 Demodulaci n diferencial La se al de radio DAB utiliza un esquema de modulaci n conocido como DQPSK En una se al QPSK se utilizan distintas fases de la onda portadora para convenir lo
102. ement a 16 sample DFT it would take as its input 16 complex numbers representing consecutive samples of the time domain signal to be transformed For each sample in turn the program would perform the complex multi plication of that sample and the appropriate coefficient for the first output frequency the results would be added together and stored This would then be repeated for the remaining 15 output frequencies giving the result 16 stored complex numbers representing samples of the spectrum at different frequencies It is important to note that each output sample has contributions from every one of the input samples 25 1 6 Fast Fourier Transform FFT This would require 162 i e 256 multiplication operations and 16 additions Multipli cations are more time consuming operations for a computer and generally the relationship between the number of multiplications and the number of input or output samples is a square law This can lead to excessive computing time for large numbers of samples which is a fundamental shortcoming of the DFT when implemented on a computer However there is a significant amount of redundancy in this long hand computation and algorithms have been developed to exploit this Notwithstanding this in some cases multiplication speed is less of a problem than other processes such as memory access and specialized integrated circuits are available which are designed to implement DFTs It was noted earlie
103. ementa la detecci n del s mbolo nulo La funcionalidad de este DSP se divide principalmente en dos bucles uno de ellos es lanzado durante la interrupci n emitida por el conversor anal gico digital y ejecuta el AGC 186 B 11 Descripci n de los m dulos BSP y la m quina de estados que analiza la se al y prepara los datos para almacenarlos en el buffer intermedio El otro bucle se encarga de corregir los errores de frecuencia y de fase Cuando un BSP no tiene trabajo til que realizar se duerme llamada a la funci n sleep lo que permite un mayor ahorro de energ a que la soluci n cl sica de esperar en un bucle que ejecuta NOPs El BSP est dormido hasta que recibe una interrupci n Entonces realiza los c mputos correspondientes y vuelve al estado dormido Una vez el BSP recibe el s mbolo TFPR comienza a calcular el error de fase y frecuencia Una vez se recibe una interrupci n se procesan 512 muestras El motivo de ello es que utilizamos un buffer de entrada de 1024 palabras donde se sit an consecutivamente las partes reales y imaginarias de las muestras Por este motivo si fuera necesario procesar un s mbolo de m s de 512 muestras por ejemplo el Null Symbol de 2656 muestras ser an necesarias varias vueltas en el bucle de ese mismo estado 2656 512 5 18 o sea 6 vueltas Las posiciones restantes se analizan en el siguiente estado B 11 2 BSP FFT Deinterleaving Demapping Una vez que la
104. ery small subset of all possible convolution codes The usefulness of the this feature will be apparent later on when Viterbi decoding is described Figure 1 24 illustrates the entire process of encoding an input pattern to quantizing the received signal to decoding the quantized signal The sample input pattern used is 1011010100 For the purpose of this example this input pattern is sufficiently long In practice for proper error correction the input pattern need to be longer than approximately 10 times the number of delay units plus 1 i e length of input pattern 10 k 1 k 1 is often referred to as the constraint length For this example k 2 which implies 30 input symbols need to be received before decoding to improve the bit error rate The first trellis diagram in Figure 1 24 illustrates the state transitions associated with each input This trellis diagram is used here to illustrate how the decoder operates Use Figure 1 23 to verify the state transitions and the encoder outputs The encoder outputs do not have to be generated using the trellis diagram They can be generated using the system equations directly The encoded outputs are transmitted as signed antipodal analog signals i e 0 is trans mitted with a positive voltage and 1 is transmitted with a negative voltage They are received at the decoder and quantized with a 3 bit quantizer The quantized number is represented in 2 s complement giving it a range of 4 to 3 Soft
105. essor using 3 pipeline stages fetch decode and execute As a consequence all flow control instructions i e instructions that possibly change the pc program counter need to execute in multiple cycles because the pipeline needs to be flushed and re filled on a change in the pc program counter valuel l l 2 9 1 Multi cycle Instructions The following figures show the working of the pipeline stages for non flow control instruc tions the address of the next instruction is the address of the current instruction 1 and the pipeline works perfect 181 When instruction i is a flow control instruction ie a branch a call a return from subroutine a conditional branch the next instruction to be executed after this instruction 64 2 9 Multi Cycle Instructions and Delay Slots Cycle n Cycle n 1 Cycle n 2 Cycle n 3 Cycle n 4 fetch instruction fetch instruction fetch instruction fetch instruction fetch instruction 1 1 i 2 i 3 i 4 i 5 decode instruction decode instruction decode instruction decode instruction decode instruction i it i 2 1 3 144 execute 1 1 execute i execute i 1 execute 1 2 execute 1 3 is not instruction i 1 the pre fetch scheme fails the instruction that is already fetched and decoded is not the instruction that must be executed next Another instruction at address location k the target of the branch has to be fetched and decoded which takes additional cycles Cycle n Cycle n 1 Cycle n 2
106. fer served both as input output to from the AGC and in turn as input to the Null Symbol Detection algorithm The synchronization processor mainly consists of two loops The first is run when an interruption is received and computes the Automatic Gain Control with the help of the aforementioned state machine both of which amplify the signal and store back the data into the input shared buffer The other one is placed into the main call of the processor and applies the frequency and phase corrections When the BSP has nothing to do it is put into sleep mode called by the function sleep in the main loop When the TFPR symbol has been copied to BSP BS Ps starts with the calculation of the phase and frequency error and performs the corresponding correction Once the input buffer is full with IQ samples a gain factor is applied to the input complex data as we explained before Normally the gain factor is within the range 2 2 Each word of the input buffer stores either the in phase or quadrature components of the signal successively Thus for each interruption a total of 512 samples are processed Therefore 512 is the maximum number of samples each state has to deal with For this reason if it were necessary to process a complete OFDM symbol for example 2656 samples in the case of the null symbol several loops in the same state had to be performed in the case of the null symbol 2656 512 5 18 so 6 loops must be executed In t
107. fore seeing the mean quadratic error for sinusoids let s put it forward with polynomials Let f t a b R be some function we want to approximate Let P x ant 491 a x ag be an n degree polynomial The mean square error thereinafter called E is a function of the coefficients an a1 ao That is we have to find the values a that minimize E By the way E is defined as b EG Plan sa 0 POP de A 3 To minimize E for each a we take partial derivatives for each a 149 A 2 Approximation by the minimum mean square error DE f P as 40 0a 0 A 4 From last two equations it follows b n B E f P an 40 2 f xI f x dx Ya ada A 5 a k 0 a So we have to solve the following system of n 1 equations n b b Sa f Pii al f x dx j 0 1 n A 6 k 0 a a As an example we apply the last algorithm for approximating the function f t sin mt by the polynomial P t azt a t ay so we obtain az 4 12251 a 4 12251 ag 0 050465 The following graph represents the functions f and P in blue and red respectively Notice how close they are even for a polynomial of such low degree 08r s S oel x f 0 21 0 2 0 Figure A 4 sin zt approximated by a polynom 150 A 3 Orthogonality A 3 Orthogonality Recall from elemental geometry that two vectors are orthogonal if the dot product between them is 0 In a two dimensio
108. frame synchronization e Coarse frequency synchronization on carrier accuracy Fine frequency synchronization on sub carrier accuracy e Fine time synchronization There is a special OFDM symbol called Null symbol The Null symbol of the DAB transmission frame provides a simple and robust way for the coarse time synchronization 90 3 9 Synchronization channel which is also called frame synchronization The underlying idea is to use a symbol with reduced signal level which can be detected by very simple means In practice a short time power estimation is calculated which is then used as input to a matched filter This filter is simply a rectangular window with a duration according to the Null symbol length Finally a threshold detector indicates the beginning of a DAB frame Coarse and fine frequency synchronization can be performed using the by means of an OFDM symbol called the TFPR symbol Time Frequency Phase Reference in the frequency domain This step clearly requires a sufficiently exact coarse time synchronization Frequency offsets are calculated using the various CAZAC Constant Amplitude Zero Auto Correlation sequences inside the TFPR symbol see below These sequences provide a tracking frequency error range of about 32 carriers Fine time synchronization is performed by calculating the channel impulse response based on the actually received TFPR symbol and the specified TFPR symbol stored in the receiver
109. he end of the Null Symbol to the beginning of the TFPR Symbol which is called WinOffsetNullSymbol are discarded When a resynchronization is needed every parameters in the DAB global register are reinitialized and the processor starts executing in this state e NULL COPY After recognizing the beginning of the signal all BSPs are going to start working but no one has information about the progress of the previous one so we should implement some kind of control signal for verifying that everything is going correctly In this case BSP should know whether we are synchronized with BSP or not The way of doing that is by copying a special known symbol into the shared memory before writing the TFPR symbol indicating the beginning of a frame Immediately after that an interruption is sent to BS Pj which indicates it to start processing In the future the verification information might be managed also by interruptions e SKIP TO TFPR Once we have sent the signal to the following BSP we continue by skipping the samples between the Null Symbol and the TFPR Symbol 137 4 2 Description of the BSP instances This is the time for applying the frequency correction of the frequency error calculated in the previous loop In turn the synchronization calculations performed during the current frame will be applied in the next one This frequency correction is the same for every symbol in the ensemble SAVE TFPR FRQ COR Wi
110. he input x n in the state definition In certain systems definition of the states may be more complicated and may require special considerations Explanation of such systems such as the radix 4 trellis is outside the scope of this example There are two trellis diagrams shown in Figure 1 23 They differ only in the way the destination states are ordered The purpose for reordering is to obtain higher computational efficiency The reordered trellis is partitioned into groups of states and state transitions that are isolated from other such groups This allows realizable specialized hardware to be designed that is dedicated to compute parameters associated with the particular group The structure of the groups shown in Figure 1 23 is called the butterfly due to the physical resemblance It is also referred to as the radix 2 trellis structure since each group contains 2 source and 2 destination states Note again that the structure of the reordered trellis is fixed depending on how the states are defined Not all trellis diagrams can be reordered 40 1 9 Errors Detection and Correction into butterflies Fortunately a number of commonly used convolution coders have this nice property An important feature of this convolution code is that in each butterfly there are only two out of four possible distinct outputs 00 and 11 or 01 and 10 Notice that the hamming distance of each output pairs is 2 This feature is available to a v
111. he next picture 4 14 we can see how the SM works 135 4 2 Description of the BSP instances Resynchonization Symbol copied Symbols copied 75 Symbol copied Symbol copied Proces OFDM Samples skipped Figure 4 14 State machine for BS Po NULL DETECTION This is the initial state where the coarse time synchronization takes place or a resynchronization if some problem occurs As explained above this state detects a drop in the signal for recognizing the null symbol Since the input signal needs some time to be established the gain must be adjusted to deal with the average power of the incoming signal it is preferable to wait some frames before starting with the detection Thus our decision was to discard the first four null symbols Each time we are in this state the minimum size between the maximum number of samples 512 and the remaining samples if it were the case are analyzed as explained before The following pseudo code illustrates this idea 136 4 2 Description of the BSP instances if state NULL DETECTION then usedSamples min remainingSamples SamplesT oS kip remainingSamples usedSamples SamplesToSkip usedSamples buffer pointer usedSamples if SamplesToSkip 0 then Vars state Y SamplesToRead dab nextState state Y end if end if When the null symbol is found the state machine jumps to the following state and the set of samples remaining from t
112. he window 1 ms long each bandwidth becomes 1 kHz and the centre frequencies fall on a regular 1 kHz comb The frequency response of each filter exhibits nulls at 1 kHz 2 kHz etc and this accounts for the cancellation of inter carrier interference noted earlier Therefore it should be clear that the absolute frequencies of the carriers presented to the FFT processor and their frequency separation are intimately related to the symbol duration and that any divergence from this relationship will cause some loss of orthogonality i e crosstalk between carriers All of these features apply equally but in a reversed sense to the inverse FFT used in the transmitter The absolute carrier frequencies and their separation are automatically related to the reciprocal of the symbol duration Of coursejt is possible to specify the spectrum of the OFDM signal such that certain carriers are suppressed i e their amplitudes are set to zero and this is done for the remainder of the 2048 frequency samples beyond the 1536 that are used for the DAB signal The spectrum could also be configured such that every 22 1 6 Fast Fourier Transform FFT other carrier was suppressed so the relationship between the frequency separation and the reciprocal of the symbol duration need not be 1 1 it could be 2 1 or some other integer ratio However a 1 1 relationship provides the greatest possible packing density consistent with the facility for in
113. her of the same size so in the same row in picture 3 8 Thus only one pointer is needed to mark the first element of the first queue The first elements of the rest of the queues can be deduced from the pointer to the first element of the first queue since they are within a fixed offset from it For example suppose 48 bits are received in a interleaved fashion Thus the third row of the above table would be that on picture 3 10 Note that the first element of the second queue is the address of the first element of the first queue plus 4 With this scheme only 16 pointer updates are needed for each frame Another final improvement consists on packing several bits in a single word of 24 bits So far we used one word for each input bit now we will pack 4 input bits into a single word thus reducing the memory requirements by four The BSP architecture does support the data type simd12 which makes possible to pack two integers into one word However it does not implement the type simd6 which would be required for this application Nevertheless we emulated this data type in software by making use of bit operations like masks shifts etc Figure 3 11 The counterpart is that more cycles of overhead will be added to the running time As ever some compromise must be found between memory space and number of cycles and of course energy consumption The input size depends on the subchannel configuration and so does the performance During the demodu
114. hould have a minimum value of at least 50 us for terrestrial broadcasting l6 51 1 3 QPSK Quadrature Phase Shift Keying So far we pointed out that PSK uses a finite set of phases to represent the symbols that compose the message QPSK uses four points on the constellation diagram equispaced around a circle With four phases QPSK can encode two bits per symbol shown in the diagram with Gray coding to minimize the Bit Error Ratio BER 10 1 8 QPSK Quadrature Phase Shift Keying o i S AU Q Figure 1 10 Constellation diagram Let T be the duration of one symbol The mathematical expression for each symbol i 1 2 3 4 is 2E Si t y F eosQm fet Qi 1 7 i 1 2 3 4 where Es is the energy employed to send the signal The last equation also shows that the less the energy employed for the transmission the closer the symbols will be in the constellation diagram thus the chance of one symbol becoming another is higher The signal constellation consists of the signal space 4 points E Es 2 a The factors of 1 2 indicate that the total power is split equally between the two carriers A QPSK signal is actually a composition of two orthogonal basis functions gi t Von palt ERE where f is the carrier frequency The first and second bases are called the in phase I and quadrature Q components of the signal respectively From the orthogonality of the I and Q components
115. iated with each state is determined Traceback uses this information to derive an optimal path through the trellis Referring to the second trellis in Figure 1 24 notice that a state at any time instance has exactly one incoming path but the number of outgoing paths may vary This results in a unique path tracing backward What are state metrics and how are the optimal incoming paths determined To un derstand the metrics used in Viterbi decoding consider the received encoded bit pairs GO n G1 n Suppose 11 is transmitted we would expect 11 to be received with a soft decision pair of 4 4 However channel noise may corrupt the transmitted signal such that the soft decision pair may be 0 4 3 3 resulting in a quantized value of 0 3 see Figure 1 25 Without knowing anything about the encoded bit stream the detector would decide that 3 4 or 01 is transmitted which results in a bit error The Viterbi algorithm on the other hand exploits the structure of the convolution code and makes its decision based on previously received data i e this is kept track of using the states Referring back to Figure 1 23 observe that an encoded bit pair is associated with only 2 possible originating states r n 1 z n 2 For example if 11 is transmitted then the originating state must be either state 00 or state 01 see Figure 1 23 Suppose that the original state is known to be 00 then it would be impossible for the detected
116. ication of a time window processing is only carried out on those samples which appear when the window is open This action also imposes a limit on the range of frequency values which need to be considered in the summation which will be explained later As is often the case in sampled systems a compromise has to be made between accuracy and an acceptable amount of processing Generally the sampling frequency must be greater than twice that of the highest frequency component in the input signal and this so called Nyquist criterion is applicable in most cases of time domain sampling Frequency compo nents above half the sampling frequency are not represented accurately and may need to be removed from the input signal by filtering A frequency component at exactly half the sampling frequency can be considered to be at the Nyquist limit ll 1 6 3 The Discrete Fourier Transform The result of the DFT provides a series of samples of the spectrum for negative and positive frequencies The time domain sampling of the input signal gives rise to repetitions 24 1 6 Fast Fourier Transform FFT of this two sided spectrum at higher frequencies centered on multiples of the sampling frequency i e the spectrum is periodic in terms of frequency Tf the sampling frequency is sufficiently great these can be removed by filtering Insufficient sampling frequency gives rise to overlapping spectra so called aliasing which cannot
117. ice spectrum of composite signal ES 200 001 MHz 2kHz 1kHz 200MHz 1kHz 2kHz modulation signal gt i 4 e 200 MHz modulation signal 199 999 MHz adder modulation n gt signal Figure 1 15 generation of an OFDM signal Each oscillator provides one of the carriers and the modulation is applied by each multi plier All of the modulated carrier signals are then added together to make up the composite signal The addition or summation can be considered as taking effect in the frequency domain all of the frequency components produced by the multiplications are combined without affecting the time waveform of any component If the modulation signals are com posed of symbols such that changes only occur at the symbol boundaries then during each symbol each modulated carrier is temporarily a sinusoidal wave with a particular phase and or amplitude representing the modulation 16 1 5 Orthogonal Frequency Division Multiplexing OFDM However 1536 such oscillators could occupy a large room The key to compact digital implementation lies in the recognition that this process parallels the operation of a math ematical process known as the inverse Discrete Fourier Transform iDFT The iDFT is a method of calculating the waveform of a signal for which the spectrum is known The iDFT operates with time domain and frequency domain variables which must be expressed as series of discrete samples In this case
118. ied values and generally the result will be a complex number In that case up to a certain degree of corruption the correlation for zero shift will still have the greatest magnitude The ruggedness of such a sequence can be demonstrated by adding a 16 element sequence of random numbers from the set 1 j 1 j as shown in Fig 3 21 This corresponds to 0 dB signal to noise ratio but the correlation peak for zero shift can be clearly identified Essentially the added sequence would need to mimic the original CAZAC sequence 95 3 9 Synchronization channel CAZAC sequence j 1 j 1 11 11 1j 11 1 1 1 random Sequence noise 1 1 ji 1 1j Aa 1 rtp ji jj 1 conjugate sequence j 1 j 1 1 1 11 j 1 j 111 1 1 magnitude of 99 x correlation y 10 Aa He ML pcs 0 1 2 3 a a EN 4 5 6 7 8 9 10 11 12 13 14 15 16 shift Figure 3 21 correlation with added noise or some multiplied and or generated and the likelihood shifted version for a large false correlation peak to be of this occurring is reduced by lengthening the sequence e If several of the original CAZAC sequences are concatenated end to end and the corre lation is performed between any group of 16 consecutive elements and the 16 element conjugate sequence the result will be periodic correlation peaks If the amount of shift is limited
119. ift clock causes the register to shift its content by one stage towards the MSb stage b 1 while 109 3 10 Transport mechanisms loading the tapped stages with the result of the appropriate XOR operations The procedure is then repeated for each data bit Following the shift after applying the last bit LSb of the data block to the input the shift register contains the CRC word which is then read out Data and CRC word are transmitted with MSb first The CRC codes used in the DAB system shall be based on the following polynomials G z a 24 r a 1 Note that all CRCs has to be passed correctly to ensure a good reception of the data For mode I 3 CRCs for each FIB block have to be checked so in total 12 CRC are tested IH 3 10 2 Main Service Channel MSC The MSC is a time interleaved data channel used to carry the audio and data service components together with possible supporting and additional data service components The MSC is divided into a number of subchannels Each of the subchannels is individually convolutionally coded with equal or unequal error protection Each subchannel may carry one or more service components The MSC is made up of Common Interleaved Frames CIFs The CIF contains 55 296 bits The smallest addressable unit of the CIF is the Capacity Unit CU comprising 64 bits Therefore the CIF contains 864 CUs which shall be identified by the CU addresses 0 to 863 The MSC is divided
120. ignal is 1 537 MHz Clearly the greater the number of individually modulated carriers that can be packed into the given bandwidth the greater the potential data capacity of the signal but the upper limit is set by the requirement for independent demodulation without mutual interference The significant bandwidth occupied by each modulated carrier is determined by the chosen symbol rate and the modulation scheme and a simple way to avoid mutual interference would be to separate adjacent carriers by frequency guard bands However such a sim ple Frequency Division Multiplexing FDM approach would be wasteful of RF spectrum Without guard bands the spectra of adjacent modulated carriers are likely to overlap and the allowable degree of overlap depends on the method of demodulation so this establishes a maximum packing density or a minimum separation between the carrier frequencies In either case some form of spectrum analysis technique is needed in the receiver The DAB system uses a technique known as Orthogonal Frequency Division Multiplex ing OFDM which allows the greatest possible packing density consistent with the use of practicable mainly digital processing techniques This requires the minimum separation of the carrier frequencies to be equal to the reciprocal of the symbol duration 1 kHz so the spectra of adjacent modulated carriers will certainly overlap In that case the maxi mum number of modulated carriers would be 1537 but i
121. into sub channels Each sub channel shall occupy an integral number of consecutive CUs and is individually convolutionally encoded Each CU may only be used for one sub channel A service component is a part of a service which carries either audio or general data The data carried in the MSC shall be divided at source into regular 24 ms bursts corresponding to the sub channel data capacity of each CIF Each burst of data constitutes a logical frame Each logical frame is associated with a corresponding CIF Succeeding CIFs 110 3 11 Audio Coding are identified by the value of the CIF counter which is signaled in the MCI The logical frame count is a notional count which shall be defined as the value of the CIF counter corresponding to the first CIF which carries data from the logical frame H 3 11 Audio Coding MPEG 1 Audio Layer II MP2 is the audio compressing format defined for DAB It uses a lossy compression algorithm that was designed to greatly reduce the amount of data required to represent an audio recording and sound like a decent reproduction of the original uncompressed audio for most listeners For transmission mode I the sampling rate is set at 48 kHz The format is based on successive digital frames of 1152 sampling intervals with four possible formats e Mono format e Stereo format e Intensity encoded joint stereo format stereo irrelevance e Dual channel uncorrelated format MP2 splits the input audio sig
122. ion frame for the four transmission modes Transmission mode Duration of Number of FIBs per Number of CIFs per transmission frame transmission frame transmission frame 11 ms 12 124 J 9 Figure 3 28 General transport characteristics of the transmission frame In transmission mode I the 12 FIBs contributing to one transmission frame shall be divided into four groups which are each assigned to one of the CIFs contributing to the same transmission frame The information contained in the first three FIBs shall refer to the first CIF the infor mation contained in the fourth fifth and sixth FIB to the second CIF and so on All FIBs contributing to a transmission frame in transmission modes II and III shall be assigned to the one CIF associated with that transmission frame In transmission mode IV the six FIBs contributing to one transmission frame shall be divided into two groups which are each assigned to one of the CIFs contributing to the same transmission frame The information contained in the first three FIBs shall refer to the first CIF and the information contained in the fourth fifth and sixth FIB to the second CIF The general structure of the FIB is shown later for a case when the useful data does not occupy the whole of a FIB data field The FIB contains 256 bits and comprises an FIB data field and a CRC The following subclauses describe the formation of the FIC and MSC I 3 10 1 Fast Information Chan
123. is responsible for the most arduous task in terms of signal processing the syn chronization of the input signal The input to this BSP comes from the sampled radio signal and once is processes by the synchronization algorithms it is sent to the next BSP The first task of this BSP is to initialize all the data structures of the demodulator and in turn enable the interruptions If the interruptions were set up before initializing the aforementioned data structures an interruption could occur thus starting the demod ulation process before all the parameters of the demodulator have the right values In short interruptions start disabled and are enabled after the initialization of the DAB data structures During the very first tests of the system in real time we had to face up a problem the rate at which the demodulator processed the samples in the input buffer was slower than the sampling rate of the incoming signal therefore the stored symbols in the input buffer were overwritten by the new samples before being processed For avoiding this problem 134 4 2 Description of the BSP instances the first and obvious idea was merely to increase the memory but the size necessary to buffer every OFDM symbol without losing data was 16K words and this was inconceivable for the project A synchronous state machine was created with the aim of saving memory in the first BSP In this way instead of working with an ancillary buffer the input buf
124. it carry several radio programs at the same time by bringing together data representing a number of audio signals and multiplexing them before transmission However this gives rise to a dilemma if the DAB signal were to use a single carrier then in order to achieve a wide bandwidth the symbol rate would need to be high but this conflicts with the requirement for long symbols A solution is to use not one but a multiplicity of carriers each at a different radio frequency By modulating each carrier independently at a low symbol rate by a small fraction of the data to be transmitted individually the carriers will then be relatively resistant to multipath echoes because of their long symbols The requirement for mobile reception imposes an upper limit on the symbol duration The changing characteristics of the transmission channel can have adverse effects on what ever modulation system is used and generally the maximum symbol duration is related to 14 1 5 Orthogonal Frequency Division Multiplexing OFDM the required maximum vehicle speed Thus the maximum symbol duration chosen for the DAB system is 1 ms In isolation this allows good reception at vehicle speeds of at least 100 km hr On the other hand it is found that the resistance to multipath effects improves as the bandwidth is increased accommodating more extreme multipath conditions which could otherwise cause flat fading Hence the actual bandwidth chosen for the DAB s
125. ited that say a float point implementation To allow a fast computation of the FT the size of the input buffer must be a power of 2 In DAB mode I there are 1536 subcarriers so the number of points of the FFT is 2048 Note that 2048 1536 512 output samples are not actually used 3 3 Differential Demodulation The DAB system uses a different method whereby the QPSK modulation on each carrier is applied differentially that is 2 data bits are signaled by the change of phase of a carrier at each symbol boundary rather than the absolute phase Detection can be achieved in the receiver by storing each output from the FFT for one symbol and comparing in some way the new value with the previous value This avoids the need for an absolute phase reference which can simplify the receiver implementation and correct operation resumes quickly after a deep fade as soon as two consecutive symbols have been received successfully The disadvantage is impaired performance in the presence of noise and interference i e the previous received symbol could be in error Up to 3 dB greater S N is needed to match the performance of coherent detection in the absence of fading In the DAB system the chosen method of modulation is 7 4 offset D QPSK where the offset is 45 degrees i e 7 4 in radians The four modulation states are signaled by 45 degrees and 135 degrees changes of the carrier phase at the start of each new symbol and this a
126. its own front end noise is demodulated giving a burst of audio noise In most common environments reflected and diffracted signals usually have smaller magnitudes than a direct signal but a direct line of sight path might not always be available Most reflections occur by lossy mechanisms i e not large sheets of metal and reflected signals often have further to travel so they are subject to greater spreading losses On the other hand the phenomenon by which fading affects only a fixed range frequencies within the signal is called selective fading WX Reflected H Paths y EE S e Direct Path Figure 1 7 Multipath propagation in the environment Transmitter The receiver performance can be enhanced by using a directional antenna which is only appropriate to fixed reception or a diversity system using more than one antenna which may be applicable to a more expensive vehicular installation Indeed the audio quality available from fixed FM receivers with rooftop directional antennas is very good but the market for radio consumption has changed from what was anticipated at the start of BBC FM services The widespread use of portable and mobile receivers now demands a means for delivery which offers the highest audio quality in most environments within a service area without the need for a complicated antenna system or one that may need adjustment On this basis there is little that can be done to rescue an FM signal or a
127. itted later In fact this is the most critical module in terms of memory requirements for the DAB Note that when we start receiving bits we must wait 16 frames 1536 ms in order to fulfill the buffer The number of bits within a subchannel is a multiple of 16 so without loss of generality we will consider in a first sight a group of 16 bits Let r be the current frame being processed So b is the i th bit within the frame emitted t frames ago To delay the bits correctly we implemented a data structure consisting of 1 array of 16 circular queues FIFO each one of a size equal to reverse i Bits are inserted always at the end of the queue Picture 3 8 shows this data structure and how bits are inserted 84 3 6 Time De Interleaving Argument Type Description P See below The pointer value to be update offs int The increment or decrement Start See below The start address of the array Len int The length of the circular buffer Figure 3 9 Structure of the circular buffer The output is made out of all bits in order from the first position of each queue to the last one So according to the picture the output would be b 15 00 7 107 11 20 3 30r 13 40 5 50 9 60r 1 70 14 85 6 9 br 10 10br 2 11br 12 12br 4 13br 8 14br 15 Note that the first bit is the first bit of the frame emitted 15 frames ago the second bit is the second bit emitted 7 frames ago and so on The CoolFlux BSP has special functions to supp
128. ki Phase shift_keying 8 Wikipedia Time division multiplexing Last Updated 27 April 2010 http en wikipedia org wiki Time division_multiplexing 9 Wikipedia Frequency division multiplexing Last Updated 23 June 2010 http en wikipedia org wiki Frequency division_multiplexing 192 Bibliography 10 11 12 13 14 15 16 17 18 19 20 21 Wikipedia Puncturing Last Updated 17 December 2009 http en wikipedia org wiki Puncturing Jen Ming Wu Associate Professor Dept of E E Inst of Comm Eng National Tsing Hua Univ Hsinchu Taiwan Example Viterbi algorithm http my com nthu edu tw jmwu comb195 viterbi example pdf Wikipedia Digital Signal Processors Last Updated 8 June 2010 http en wikipedia org wiki Digital signal processor Wikipedia Finite impulse response Last Updated May 16 2010 http en wikipedia org wiki FiniteN impulseN response Wikipedia Bit Error Ratio Last Updated May 20 2010 http en wikipedia org wiki BitN errorN rate http www insidedsp com http www insidedsp com tabid 64 articleType ArticleView articleId 319 New Details Emerge on NXPs CoolFlux BSP Core aspx http www insidedsp com http www insidedsp com tabid 64 articleType ArticleView articleId 319 New Details Emerge on NXPs CoolFlux BSP Core aspx NXP Semiconductors CoolFlux BSP Interrupt Controller version 1 6 January 2010 NXP Semiconductors CoolFlu
129. l objetivo principal de nuestro proyecto era dise ar un demodulador completo de radio digital orientado especialmente a dispositivos m viles siguiendo el est ndar europeo Digital Audio Broadcasting Para ello utilizar amos el procesador CoolFlux BSP propiedad de la empresa en la que este proyecto se llev a cabo aprovechando su alta capacidad de procesamiento de bajo consumo El objetivo final de nuestro proyecto consistir a en conseguir una se al de audio libre de errores y un demodulador lo m s optimizado posible en cuanto a espacio de memoria y tiempo de ejecuci n Estos dos ltimos aspectos juegan un papel muy importante en el consumo de energ a del receptor Actualmente el sistema es capaz de demodular una se al de radio captada directamente del medio y procesarla hasta concluir toda la cadena de demodulaci n que nos llevar a la obtenci n del audio Las complicaciones surgidas durante el desarrollo del proyecto no nos han permitido implementar el modelo m s optimo deseado no obstante se han realizado mejoras notables en t rminos de memoria y ciclos de computaci n durante el transcurso del mismo B 12 3 Aspectos futuros Gracias a las ventajas que CoolFlux BSP nos ofrece ser a posible reducir de forma notable tanto las dimensiones utilizaci n de memoria del dise o actual como su consumo El objetivo principal del proyecto no englobaba reducir al m ximo estos t rminos pero incluso para una primera implementaci n
130. lation of a DAB signal mode IT with 50 frames and a subchannel size of 208 capacity units CU we obtained 2053293 cycles in total for the time de interleaver Since each frame in mode II takes 24ms we can easily compute the number of MIPS 2053293 1M 24mS 50 frames 1 71MIPS 86 3 7 Errors Detection and Correction According to the DAB standard the maximum size allowed for a subchannel is of 416 CU s for each frame giving a bit rate of 384kbit s Since each CU contains 64 bits the subchannel maximum size has a total of 26624 bits Each basic structure as the one shown in the picture above is made up of 1 2 16 136 words and since each word holds 4 bits 16 4 64 bits can be de interleaved with 136 words Hence the number of words needed to fully de interleave a subchannel of maximum size is 56576 55 25 KB 3 7 Errors Detection and Correction The convolutional coding used in the DAB system employs these principles but is consid erably more complicated than the examples given so far The constraint length is 7 so the encoder has 64 states Fundamentally it uses 4 generator polynomials and the generated bits are transmitted in a fixed sequence cycling through the 4 bits so the encoder operates at rate 1 4 This can provide very powerful error correction albeit with extreme consumption of the available bit rate gt o gt e gt gt i Y e gt Xi E EE X
131. lation of the CRC word A 16 bit Cyclic Redundancy Check word is calculated on the FIB data field The implemen tation of CRC codes for data and audio transmission allows the detection of transmission errors at the receiver side For this purpose CRC words shall be included in the transmitted data These CRC words shall be defined by the result of the procedure described here A CRC code is defined by a polynomial of degree n G x gu ix 71 gaz gix 1 n gt 1 gi 0 1 4 1 n 1 The CRC calculation may be performed by means of a shift register containing n register stages equivalent to the degree of the polynomial The stages are denoted by bo by_1 where bo corresponds to 1 b to x1 b2 to 22 64 1 to z4 4 The shift register is tapped by inserting XORs at the input of those stages where the corresponding coefficients g of Data Input the polynomial are 1 yo y LSb 50 bi bn2 Bni MSb Figure 3 33 General CRC block diagram At the beginning of each CRC word calculation all shift register stage contents shall be initialized to 1 The CRC word shall be complemented 1 s complement prior to transmission If the final output of the shift register and the CRC check word in the DAB audio frame are not identical a transmission error has occurred in the protected field of the audio bit stream After applying the first bit of the data block MSb first to the input the sh
132. level parallelism and pipelining and of the heterogeneous aspects of the ASIP architecture The C compiler produces machine code in the Elf Dwarf Debug With Arbitrary Record Format object file format The C compiler is complemented with an assem bler disassembler and linker Executable and Linkable Format e An instruction set simulator ISS that can execute the compiled machine code both in an instruction accurate and a cycle accurate way The ISS supports cosimulation and can automatically be encapsulated in SystemC ESL models The ISS comes with a graphical C source level debugger that can connect both to the ISS and to the hardware target for on chip debugging Profiling capabilities are provided as a basis for architectural or algorithmic refinement e A hardware generator that creates a synthetizable RTL implementation of the ASIP core in VHDL and Verilog for implementation in an ASIC or FPGA Optionally the generated RTL model can contain support for on chip debugging 68 2 13 BSP C Compiler e A test generator that produces a large number of processor specific assembly test programs with high fault coverage User Defined ASIP Synthesis RTL Generator Synthesizable RTL VHDUVerilog y 3rd party tools RTL RTL Simulator Synthesizer O SDK Generation O Architectural Optimization Hardware Generation O Verification Under Target s IP ProgrammerTM service ASIP specific versions of
133. liferaci n y popularidad de dispositivos m viles hace que el bajo consumo unido a la alta productividad se haya vuelto imprescindible 188 B 12 Conclusiones El imparable avance en el campo de la electr nica particularmente en las t cnicas de fabricaci n de circuitos integrados ha tenido un gran impacto en la industria de las co municaciones El r pido desarrollo de la tecnolog a de integraci n a gran escala VLSI Very Large Scale Integration de circuitos electr nicos ha estimulado el desarrollo de com putadores digitales m s potentes peque os r pidos y baratos cuyo hardware responde a prop sitos espec ficos Esta tecnolog a ha hecho posible construir sistemas digitales alta mente sofisticados capaces de realizar funciones y tareas del procesado de se al digital que normalmente eran demasiado dif ciles y o caras con circuiter a o sistemas de procesado de se ales anal gicas Dentro del procesamiento de se al digital en los ltimos a os se ha experimentado un direccionamiento de las tendencias del mercado tecnol gico hacia las telecomunicaciones m viles de altas prestaciones lo que ha provocado un creciente inter s por arquitecturas empotradas de bajo consumo La proliferaci n y popularidad de dispositivos port tiles hace que el bajo consumo unido a la alta productividad se haya vuelto imprescindible Adem s como consecuencia de la complejidad creciente de los protocolos de comunicaci n y del advenimien
134. ll never be a clash between the memories and the hardware modules Note that for the running program the addressing space always remains the same independently of the actual physical memory that is being read The memory block switching is hidden for the software This peripheral toggles the data buffers when the corresponding IO register is written The buffers are toggled independently of what value was written Each BSP can toggle any of its shared buffers by writing to a certain IO location 133 4 2 Description of the BSP instances The first BSP starts receiving the IQ data signal from the input buffer The even and odd words of the input buffer keep the in phase and quadrature components of the signal respectively The input buffer has space for a total of 512 IQ samples When the input buffer is full an interruption is triggered to swap the two input buffers When the first BSP has finished to perform its tasks it triggers a signal to activate the next BSP within the chain In sum each BSP starts working as soon as it receives a signal from the previous BSP or a hardware interruption only applicable for BSP Note that the total memory of the Board was just around 45KB so all this space had to be divided and shared between the modules 4 2 Description of the BSP instances The aim of this section is to explain in more detail the BSP cores 4 2 1 BSP Synchronization Time correction Save OFDM sym bols This BSP
135. lux BSP are organized in register files All registers in a register file have the same name but a different index Operations work on individual registers but in their syntax description register file names are used The reason is that all registers in a register file can be used equally in an operation Individual registers must be used as source and destination registers in real operations The registers have different data widths The architecture figures presented above show the data widths of the registers They also show how the register files are connected to the different execution unitsl 3l 2 3 The Data Path The data path consists of e Four register files A B X and Y connected to the arithmetic units e Two multiplier units XMUL and YMUL 24x24 bits which can be used in different 50 2 3 The Data Path pc 24 Program sr 24 Control reset isr 24 Tm habe Ir 24 Stack interrupts ilr 24 0 3 Xbus Ybus 24 24 X addressing unit Y addressing unit XAGU YAGU X Address Generation Unit Y Address Generation Unit Addressing units Xbus and Ybus move buses memories DMA and program control unit Figure 2 1 Addressing Units Xbus and Ybus move buses DMA and program control units 51 2 3 The Data Path Xbus Ybus
136. m inputs then there will be 2 state transitions The total number of state transitions at a point in time is the product of the number of state transitions and the number of states 2 The state diagram offers a complete description of the system However it shows only the instantaneous transitions It does not illustrate how the states change in time To include time in state transitions a trellis diagram is used Figure 1 23 Each node in the trellis diagram denotes a state at a point in time The branches connecting the nodes denote the state transitions Notice that the inputs are not labeled on the state transitions They are implied since the input x n is included in the destination states r n z n 1 Another thing to note is that the state transitions are fixed by the definition of the source x n 1 z n 2 and the destination states The state transitions are independent of the 39 1 9 Errors Detection and Correction x n 1 x n 2 G n G n x n 10 1 Figure 1 22 State Diagram Origional Trellis Reordered Trellis x n 1 x n 2 G0 Gl x n x n 1 x n 1 x n 2 G0 G1 x n x n 1 AA e X N 00 yA 00 00 00 00 cm n 01 11 01 01 00 10 00 10 10 01 10 10 10 01 01 01 E a Hu m e ul n 10 Tm Figure 1 23 Trellis diagram system equations This a consequence of including t
137. magnitudes 1 and 7 Therefore to compute the branch metric you only need to do 1 add 1 subtract and 2 negation Once the branch metrics are calculated the state metrics and the best incoming paths for each destination states can be determined The state metrics at time 0 are initialized to 0 except for state 00 which takes on the value of 100 This value is arbitrarily chosen and is large enough so that the other initial states cannot contribute to the best path This basically forces the traceback to converge on state 00 at time 0 The state metrics are updated in two steps First for each of the two incoming paths the corresponding branch metric is added to the state metric of the originating state The two sums are compared and the larger one is stored as the new state metric and the cor 45 1 9 Errors Detection and Correction responding path is stored as the best path Take state 10 at time instance 2 for example Figure 1 24 The two paths coming in to this state originates from states 00 and 01 figure 1 23 Looking at the second trellis we see that state 00 has a value of 93 and state 01 has a value of 1 The branch metric of the top path connecting state 00 at time 1 and state 10 at time 2 is 1 The branch metric of the bottom path between states 01 and 10 is 1 The top path gives a sum of 93 1 94 and the bottom path give a sum of 1 1 0 As a result 94 is stored as the state metric for state 10 at time 2 and the
138. modulation scheme or line coding scheme and by applying channel coding schemes such as redundant forward error correction codes 36 1 8 Error Measures The transmission BER is the number of detected bits that are incorrect before error correction divided by the total number of transferred bits including redundant error codes The main problem of this algorithm is that the signal has to be decoded before 4l The implementation of this error measure algorithm was carried out in the following 3 768 way lutional Figure 1 20 BER calculation 1 8 2 Modulation Error Rate The aim of this algorithm is to check if the received data is good enough for ensure a good reception The modulation error ratio or MER is a measure used to quantify the perfor mance of a digital radio transmitter or receiver in a communications system using digital modulation A signal sent by an ideal transmitter or received by a receiver would have all the constellation points precisely at the ideal locations however various imperfections in the implementation such as noise low image rejection ratio phase noise carrier suppres sion distortion etc or signal path cause the actual constellation points to deviate from the ideal locations Transmitter MER can be measured by specialized equipment which demodulates the received signal in a similar way to how a real radio demodulator does it Demodulated and detected signal can be used
139. n n n oO 27Cy so 1 27 g te dt 2m 0 By the way the function c Z gt C defined as c n Cn is called the spectrum of the function Equivalently 1 2n c n fte tdt Sr A 157 m t dt A 16 A 5 Integral Fourier Transform A 5 Integral Fourier Transform So far we have talked about periodic signals and how to obtain the frequency components that is the Fourier Coefficients We made the implicit assumption that the period was 27 First we redefine the Fourier Serie for a signal of an arbitrary period After that we will deduce the Integral Fourier Transform that will allow us to deal with aperiodic functions It is not at all difficult to see that the Fourier Serie of a signal f of period T is 1 9d 2mint 1 e 2mint f z M ene zb No ene ro A 17 hence en 2e 10 as A 18 Note that for convenience instead of taking the interval 0 T we took 7 5 This is legitimate and even though the Fourier coefficients will be different to those of the original serie the latter form still converges to f You might have studied the Reimann s integral in elemental calculus Let h a b R an integrable function We split up the interval a b in subintervals with the same length wo so a b a xo z1 Xn b and z 2 1 wo Then k b lim 5 h a nwo wo p h t dt A 19 wo 0 By taking n 7 and A nth T we transform the Fourier serie i
140. n a Riemann sum oo fia ee AE A 20 n 00 If T oo the Riemann sum tends to the Fourier transform given by F f f t e dt teR A 21 On the other hand the inverse Fourier transform is given by 158 A 5 Integral Fourier Transform so f F Oet de A 22 The intuition behind this is that we consider a non periodic serie as a periodic function with infinite period Up until now we have presented a lot of integrals and assume that they can be calculated using symbolic calculus We might well use numerical methods to evaluate the integrals For example the Fourier Coefficients ay a1 01 a4 bn can be approximated by calculating the integral by means of the trapezoidal rule To obtain the area under a curve divide that area in trapezoids calculate the area of each trapezoid and then sum them all together 0 01 02 03 04 05 06 07 08 09 1 tme An multitude of methods exist for approximating integrals But the aim of this work is not how to approximate integrals rather it is how to approximate a function by sinusoids How the Fourier Transform is applied in computers and DSPs depends upon the Discrete Fourier Transform and the Nyquist Theorem which will be presented in a moment 159 A 6 DFT A 6 DFT So far we have presented the essential ideas of Fourier Analysis for continuos functions However the ideas already introduced of the Fourier Transform implies calculus
141. n now be calculated using the above 44 1 9 Errors Detection and Correction Originating states 00 01 Originating states 10 11 G n Gin amp Q3 43 nu e e o 0 O o o o o o oa ee o o o e o 0o 0o o ge o o o e 9 o o o e o o o 9 g e e 6000 90 9 Gn IA 4 gt Gyn o o o o gt o o eee LE eee o o o TA e o e o Lx o o en o en ove ee O ees so oe e o oO 4 4 3 4 Decision boundaries Figure 1 26 Decision boundaries for different states local distance equation As an example the branch metrics for time 0 and 1 is calculated here The received quantized soft decision pair is 3 4 The branch metric corresponding to Go n Gi m 0 0 is equal to 3 1 4 1 7 For Go n Gl n 0 1 it is 3 1 4 1 1 For Go n Gq n 1 0 it is 3 1 4 1 1 And for Go n Gi n 1 1 it is 3 1 4 1 7 Notice that the G n s used in the local distance calculation are 1 and 1 rather than 3 and 4 respectively There are three reasons First using unities makes computation very simple The branch metrics can be calculated by adding and subtracting the soft decision values Second 3 and 4 can be scaled to approximately 1 and 1 respectively As mentioned in the derivation of local distance 1 constant scaling can be ignored in determining the maxima and the minima Finally note that there are only two unique metric
142. n practice one carrier is not used Therefore the maximum number of modulated carriers in the DAB signal is 1536 See the appendix about the Fourier Transform for a mathematical explanation of orthogonality In this scheme the possibility arises that data conveyed by some of the individual carriers will not be received successfully because of selective fading but in this case the application of FEC is most efficient and necessary because the loss of a small number of carriers represents the loss of only a small fraction of the total data Thus for the same 15 1 5 Orthogonal Frequency Division Multiplexing OFDM degree of protection the amount of redundant data that needs to be transmitted is much smaller than would be required for scheme using a single carrier For the DAB system the abbreviation OFDM is prefixed with a C for coded to indicate the application of FEC giving COFDM With all that at hand we can deduce the total bit rate of a DAB ensemble 1536 subcarriers 2 bits per symbol DQPSK 1000 symbols sec 3 072 Mbits sec Generation of an OFDM signal is easily visualized because in principle it could be carried out by partly analogue means using a large bank of synthesized oscillators followed by modulators i e multipliers This principle is illustrated in picture 1 15 where three of many oscillators are shown although it should not be inferred that such a cumbersome arrangement is used in pract
143. n the channel P l In our case a power of 3V was used 4 1 4 5 Hewlett Packard 8596E Oscilloscope The HP Agilent 8596E is an easy to use microwave Spectrum Analyzers with 9 kHz to 12 8 GHz range that offers a wide range of performance features and optional capability to meet the measurement needs BH 126 4 1 An overview to CoolFlux BSP This instrument allows us to visualize the radio signal broadcasting for the area of Belgium 4 1 4 6 CoolFlux simulator CoolFlux debugger Chess compiler Development toolkit designed and provided by NXP Semiconductors with the architecture explained in the chapters above 4 1 4 7 Reception Antenna We employed a simple omnidirectional antenna An electromagnetic field from any direction induces an alternating current at the antenna s terminals During the simulations of the DAB we could appreciate the phenomenon of multipath propagation described earlier Even somebody moving around the antenna could cause the signal to be attenuated It could happen that due to hard signal distortions the receptor might get desynchronized with the emitter In this case a resynchronization might be required 4 1 4 8 JTAG Joint Test Action Group Although it was originally designed for testing printed circuit board assemblies today JTAG is also used for accessing sub blocks of integrated circuits making it an essential mecha nism for debugging embedded systems which may not support any othe
144. nal euclidean space a vector is an arrow from the origin to a given point on the plane and two vectors are orthogonal if the form 90 degrees between them Analytically if v 4 91 and v9 x3 y2 are two vectors in R then they are orthogonal if and only if 1 12 y1Y2 0 More general two vectors v 11 2 Ln v 2 25 27 R are orthogonal if X riz 0 A 7 i 1 The notion of vector in mathematics is actually much more general than the usual concept of vector in geometry In fact it can be found in any text of linear algebra that functions from and to N R C belong to a vectorial space and so are vectors too Let f g I C R R f g are orthogonal in the interval J if f t g t dt 0 A 8 tel Note the parallelism between the notion of orthogonality of vectors in a Euclidean space and that of two functions in a functional space From the paragraph above we can extend the concept of orthogonality to a set of functions fi fa fn This set orthogonal in I if f t fy Odt 0 1 4 j A 9 Recall we are looking for a procedure that given a signal composed of many frequencies it returns the amplitude of each frequency component i e The coefficients a b of the Fourier Serie Now we show that two sinusoids of the same phase are orthogonal in an interval multiple of 27 if and only if the frequency of one of them is an integer multiple of the frequency of the other and the resp
145. nal into 32 sub bands and if the audio in a sub band is deemed to be imperceptible for the human auditory system then that sub band is not transmitted Conversely MP2 shows a better behavior than MP3 in the time domain due to its lower frequency resolution which implies less codec time delay simpler editing and native ruggedness to the digital recording and digital transmission errors Moreover the MP2 sub band filter bank provides an inherent transient concealment feature due to the specific temporal masking effect of its mother filter T his unique char acteristic of the MPEG 1 Audio family codecs implies a very good sound quality on audio signals with rapid energy changes such as percussive sounds both on the MP2 and the MP3 codecs which use the same basic sub band filter bank The simplified block diagram of the audio decoder in the receiver shown in the following figure 3 34 accepts the DAB audio frame in the syntax defined later which is a conformant subset of the MPEG Audio Layer II 111 3 11 Audio Coding This allows the use of an MPEG Audio Layer II decoder The DAB audio frame shall be fed into the audio decoder which unpacks the data of the frame to recover the various elements of information The reconstruction block shall reconstruct the quantized sub band samples An inverse filter bank shall transform the sub band samples back to produce digital PCM audio signals at 48 kHz sampling frequency H INPUT MPEG AUDIO L
146. ncy Division Multiplexing On the other hand a receiver might be implemented as follows The matched filters can be replaced with correlators Each detection device uses a reference threshold value to determine whether a 1 or 0 is detected a 4 Matched Filter to e Y Decision Device Device PA OPSK signal TS Binary Bitstream 3 Sampling at Time Interval 7 Multiplexer E gt ai Mute 11000110 E Decision Dev ice eo i gt Matched Filter to o f Figure 1 13 QPSK demodulator Despite its apparent simplicity QPSK has a major drawback QPSK uses absolute phases e g 0 90 180 270 to determine whether the current symbol is 00 01 11 or 10 There can happen that due to noise in the channel the current and future QPSK symbols suffer a phase shift of say 90 thus rotating the constellation diagram Of course the conveyed digital information would be erroneous A better approach would be to differentially encode the phases of two consecutive QPSK symbols This scheme is called DQPSK and it is the standard in DAB The actual phase is not that of the current symbol but the difference between the last symbol and the current one In this case all the symbols that pass through the channel will be shifted in time thus in phase by the same offset Therefore operating with the difference of phases between two subsequence symbols will cancel out that offset 1 4 FDM Frequency Division Mul
147. nd cosine wave in the I channel is multiplied by the cosine local oscillator signal and the baseband negative sine wave in the Q channel is multiplied by the sine local oscillator signal Each multiplication produces sum and difference frequency cosine components i e double sideband suppressed carrier AM 31 1 6 Fast Fourier Transform FFT e d T I DAG es em ure XK A mj iFFT sin 2x fo t gt Mar gt XG MEANS ie output AD 3 signal NUI Y Q DAC gt E X Figure 1 18 the quadrature modulation system but in this case the difference frequency components cancel when the mixer outputs are added Thus the final signal contains a single cosine wave at the local oscillator frequency plus the baseband frequency In the channel encoder protocol a 2048 sample inverse FFT is indeed used to produce the 1536 carrier Mode 1 signal The computations are processed with a time window duration of 1 ms the symbol duration Therefore the frequency domain sampling has an interval of 1 kHz so adjacent input samples correspond to sinusoidal output waves separated by 1 kHz Assuming conventional ordering of the input array the first 1024 samples represent positive frequencies from 0 Hz to 1023 kHz in the baseband signals The 1024th sample represents 1024 kHz which is at the Nyquist limit and is probably unusable The remaining samples represent neg
148. nel FIC The FIC is made up of FIBs 105 3 10 Transport mechanisms 3 10 1 1 Fast Information Block FIB The general structure of the FIB is shown figure 3 29 for a case when the useful data does not occupy the whole of a FIB data field The FIB contains 256 bits and comprises an FIB data field and a CRC 256 bits 30 bytes 16 bits FIB data field useful data field FIG k FIG t End Paddi ieee ee OS ee FIG header 3 bis 5 bits FIG type Length FIG data field Figure 3 29 Structure of the FIB 3 10 1 2 Fast Information Group FIG The FIG shall comprise the FIG header and the FIG data field The following definitions apply e FIG header shall contain the FIG type and the length FIG type this 3 bit field shall indicate the type of data contained in the FIG data field The assignment of FIG types is given in the following table FIG type FIG type FIG application number 0 000 MCI and part of the SI Labels etc part of the SI Reserved Reserved 101 FIC Data Channel FIDC boa o J L2 00 L 3 on Reserved 4 10 L5 10 8 Conditional Access CA mm 7 t1 in house except for Length 31 Figure 3 30 List of FIGs Length this 5 bit field shall represent the length in bytes of the FIG data field and is expressed as an unsigned binary number MSb first in the range 1 29 106 3 10 Transport mechanisms
149. nes de eficiencia energ tica en el dise o del procesador ste no incluye una unidad para realizar c lculos en punto flotante Dicha limitaci n dificulta m s el proceso de desarrollo ya que ahora es el programador el que debe escalar los n meros para que estos no desborden y saturen la se al Adem s de todo esto es importante recalcar que cont bamos con m dulos gen ricos de software ya implementados por la compa a como el decodificador MP2 Sin duda esto supuso una ventaja para alcanzar m s r pidamente nuestros objetivos Sin embargo la adaptaci n e inclusi n de dichos m dulos a nuestros demodulador supuso una dificultad a adida 191 Bibliography 1 ETSI EN 300 401 v1 4 1 Radio Broadcasting Systems Digital Audio Broadcasting DAB to mobile portable and fired receivers European Broadcasting Union June 2006 2 Wikipedia Frequency Modulation Last Updated 7 June 2010 http en wikipedia org wiki Frequency modulation 3 Wikipedia Phase Shift Keying Last Updated 24 mar 2010 http en wikipedia org wiki Phase shift keying 4 Wikipedia Analog Digital Converter Last Updated 17 June 2010 http en wikipedia org wiki Analog to digital converter 5 Wikipedia Multipath propagation Last Updated 10 June 2010 http en wikipedia org wiki Multipath propagation 6 BBC RED White Paper June 2003 7 Wikipedia Quadrature Phase Shift Keying Last Updated 24 mar 2010 http en wikipedia org wi
150. nos en lo que se conoce como diagramas de constelaci n como muestra la siguiente figura hs Scatter plot for the 3 FIC symbols T t TTL ut 7 oe 4 d 6 o Net 2 ae ce P a CS E e WU uh i ex ack xr urs e o Vs ent AE et fox EEES gt x e In Phase Figure B 3 Diagrama de constelaci n B 6 1 Hard Decision Conceptualmente este m todo es el m s sencillo Sea a bi un n mero complejo Cada componente a b se asocia a uno de los valores del par binario 1 1 seg n a b sean positivos o negativos respectivamente A continuaci n presentamos el pseudo c digo 177 B 7 Time De Interleaving for i 0 to 1535 do r real input i i imag input if r gt 0 then output i 1 else output i 1 end if if i2 0 then output i 1536 1 else output i 1536 1 end if end for Todos los puntos que caen en el primer segundo tercer y cuarto cuadrante corresponden respectivamente a los dibits 1 1 1 1 1 1 1 1 B 6 2 Soft Decision Mientras un decodficador con hard decision opera con datos de un conjunto fijo de valores 1 6 1 las entradas al soft decision pueden tomar m s valores dentro de ese rango t picamente 16 valores diferentes Esta informaci n extra indica con mayor fiabilidad cada punto de entrada con los datos originales De esta manera este decodificador se comportar mejor en presencia de rui
151. ntroduce a scaling factor equal for all elements of the array but in this case the stored values all have unit amplitude so there is no scaling factor Described as an array the Phase Reference symbol block has 1536 elements in Mode 1 so a 2048 sample inverse FFT must be used The samples output by the inverse FFT represent the duration of the active symbol so the resolution of the impulse response is then about 0 5 us It is worth noting that absolute magnitudes are not important in this case indeed they depend on the action of AGC which is entirely separate from the processing of the Phase Reference symbol An example of an impulse response is illustrated in Fig 3 26161 3 10 Transport mechanisms The DAB system is designed to carry several digital audio signals together with data signals Audio and data signals are considered to be service components which can be grouped together to form services The DAB system transmission frame consists of three different channels 102 3 10 Transport mechanisms Impulse response for frame 11 6 0 500 1000 1500 2000 Figure 3 26 example of an impulse response for a multipath channel 1 Main Service Channel MSC Used for carrying audio and data service components The MSC is a time interleaved data channel divided into a number of sub channels which are individually convolu tionally coded with equal or unequal error protection Each sub channel may carry one or more ser
152. number this would be true for many sequences However if the sequence is shifted by one element i e the first number takes the place of the second one etc and the last number takes the place of the first one and the shifted sequence is multiplied by the conjugate of the original sequence the sum of the products is zero Furthermore the same zero result occurs for any amount of shift less than the length of the sequence in either direction This means that the sequence has Zero Autocorrelation and this property is relatively uncommon Correlation is the process of multiplying the element values and accumulating the result and the auto prefix indicates that the process is carried out on a sequence and the conjugate of the same sequence An example of a 16 element CAZAC sequence its conjugate and the autocorrelation with no shift are shown in Figure 3 18 note that the conjugate has the signs of the j s reversed CAZAC sequence j 1 j 1 11 11 j 1j 1111 1 X conjugate sequence j 1 j 1 11 1 1j 14 1 1 1 1 1 1494 94 97 467497 4 74174141414141414141 16 Figure 3 18 a 16 element CAZAC sequence its conjugate and the autocorrelation with no shift Performing the autocorrelation with a shift of 1 element or any other discrete amount up to 15 elements gives the zero result as shown in Fig 3 19 Note that the zero result arises because of cancellation so it is essential that all elements of the
153. ny other analogue signal in the presence of severe fading or interference Much of the damage is done in the receiving antenna and there is little scope for post processing to rectify the situation However in broadcasting and many other fields of communication a solution is being sought in the application of digital techniques implying a radical departure from the classical methods of broadcasting This introduces numerous problems for the broadcasters and the receiver manufacturers none of which is technically insurmountable nowadays but it offers the great advantage of post processing in the form of error correction 1 2 Multipath Propagation The effect of fading and interference on a digital system is to introduce errors into the received signal but improvements are possible by virtue of the numerical nature of the bit stream Error detection can be achieved in the receiver by sending a small amount of additional data derived from the original data such as checksums By sending further additional data it becomes possible to correct errors Such additional data are often referred to as redundant data but this does not imply that they are not needed only that they carry no new information The trivial case would be simply to transmit all of the original data twice with independent checksums so with the benefit of error detection a complete set of good data could be reconstructed However this would make relativel
154. o de cierto umbral de la potencia media de la se al de entrada En la siguiente figura se muestra el valor medio de una se al DAB real en modo I Obs rvense las abruptas ca das de potencia correspondientes a la llegada de un s mbolo nulo el cual es recibido una vez cada 96ms Output of Null detector T i L i i i 0 50 100 150 200 250 Figure B 4 Detecci n del s mbolo nulo Para evitar el caso de que se confunda por error una ca da accidental de la potencia con un s mbolo nulo se realiza un filtro paso baja con el fin de atenuar cambios esp reos B 9 2 TFPR symbol El s mbolo TFPR esta compuesto siempre por un patr n fijo de muestras Durante la transmisi n del s mbolo TFPR todas las portadoras son transmitidas a una potencia normal Dicho patr n se analiza en el receptor una vez la se al ha sido procesada por la FFT No obstante la posici n del patr n dentro del array de salida de la FFT depende de la frecuencia 180 B 10 Implementaci n hardware del demodulador DAB de la se al recibida y del oscilador RF del receptor El objetivo es localizar el patr n en una determinada posici n para que las portadoras puedan ser identificadas de forma un voca en el siguiente frame El mismo patr n es almacenado en el receptor De esta forma el error en frecuencia se puede determinar midiendo la desviaci n entre los patrones recibidos y almacenados El s mbolo TFPR tambi n facilita medida
155. oduce combinations of sine and cosine waves What is then needed is a method for combining the sampled waves appearing at the I and Q output ports to produce the required single output signal and this can be achieved using a quadrature modulation system S 1 6 10 Combination of I and Q The I and Q data are applied separately to a pair of DACs or one with time multiplexing to produce a pair of sampled baseband signals which are then applied to low pass filters to construct analogue signals i e to remove the artifacts of sampling Note that these filters are not used to implement matched filtering e g cosine roll off as is used in some other digital modulation systems that function is effectively carried out by the inverse FFT and the FFT in the receiver The filtered baseband signals are applied to a pair of mixers i e multipliers which are also fed with synchronous local oscillator signals having 90 degrees i e 7 2 phase difference that is cosine and sine waves The local oscillator frequency is equal to the desired centre frequency of the final signal The outputs of the two mixers are then added to give the final signal which after band pass filtering to remove harmonics and other imperfections can be transmitted This quadrature modulation system is illustrated in Fig 1 18 Taking again the example of a single positive frequency sample applied to the real input port of the inverse FFT with a positive value the baseba
156. of the loop is always 1 since the new loopcount value is only written if the loop has not ended The loop instruction has 2 initial delay slots which means that the actual start address of the loop the program counter value to which the loop jumps back at the end is the value of the program counter after 2 processor cycles 66 2 11 Debug Mode The loop end address is relative and thus calculated The loop end address is calculated in the same way as done for the relative branches Loop end address PC value do instruction 2 of fset Loop end address PC value do instruction 1 of fset The first one is taken when there are two single cycle instructions in the two delay slots while the second one is taken when there is a two cycle instruction in the two delay slots 8l 2 11 Debug Mode It is possible to enter debug mode using a software instruction Whenever the debug instruc tion is executed the programs stops and goes into debug mode This can be useful when debugging the design on an actual target for instance on an FPGA The debug instruction is a multi cycle instruction that take 3 cycles to execute There are no delay slots l 2 12 Interrupts The CoolFlux BSP supports 3 vectored hardware interrupts as well as 4 software interrupts On top a hardware reset is treated as an interrupt located at interrupt vector table location 0 When an hardware or software interrupt occurs the status register sr i
157. or NXP Semiconductors destinado a disposi tivos m viles de consumo energ tico limitado Para la implementaci n del sistema DAB se ha realizado un gran trabajo en cuanto a procesamiento de se ales y se han aplicado t cnicas de optimizaci n destinadas a conseguir el menor consumo de potencia posible con el m nimo almacenamiento de memoria y ciclos de computaci n Tambi n se han aplicado numerosos c digos de correcci n de errores a los datos para asegurar una alta calidad del audio emitido Anteriormente a este trabajo se realizaron estudios de viabilidad en los que se concluy que tan s lo ser a necesario un BSP para demodular la se al en tiempo real Sin embargo el prototipado del sistema en una FPGA cuya frecuencia de reloj es considerablemente m s limitada que un sistema real exig a la necesidad de utilizar 3 instancias del BSP conectadas en cascada Keywords DSP BSP CoolFlux Digital Radio DAB Demodulator FFT Digital Signal Error correction Real Time Processor Chapter 1 Preliminary concepts in DAB Before we introduce the actual implementation of the DAB demodulator on the CoolFlux BSP we present some theoretical concepts in radio communications The problem at hand is that of transmitting information in an efficient manner using the electromagnetic spectrum In this scenario we have the broadcaster that is the agent that transmits the information and the receiver The space between them
158. or image of this approach is used in the experimental DAB receivers where the incoming signal is converted to an IF and band pass filtered and then applied to a similar quadrature modulation system In this case the local oscillator frequency is the centre frequency of the IF band and of course ADCs are substituted for the DACs What has been described is the approach that has been taken so far in all successive generations of experimental DAB transmitter equipment However it is worth noting that the FFT can be applied to multi carrier signal generation and reception in other ways some of which do not require the quadrature modulation system Also it is possible to implement a quadrature modulation system in the digital domain rather than the analogue domain l l 1 6 11 Addition of the Guard Interval It was noted earlier that if the frequency domain data input to an inverse DFT remain static for consecutive time windows then the time domain result will be periodic over the consecutive windows and this applies equally to the inverse FFT In the DAB channel encoder the window duration is equal to the active symbol duration 1 ms in Mode 1 and the signals output by the transform are all sinusoidal waves at harmonics of 1 kHz so they are periodic over whole multiples of the active symbol duration Therefore consecutive inverse FFT operations with the same input data would yield continuous waves in the baseband signals For example if the
159. ore taps in a filter results in better stopband attenuation less of the part we want filtered out less rippling less variations in the passband and steeper rolloff a shorter transition between the passband and the stopband e Multiply Accumulate MAC In the context of FIR Filters a MAC is the op eration of multiplying a coefficient by the corresponding delayed data sample and accumulating the result There is usually one MAC per tap There are a couple different basic filter responses Each will have a unique frequency response based on its cut off frequency the number of taps used its roll off and amount of ripple The various attributes describing a filter may be seen in the following diagram FIR High Pass Filter FIR Low Pass Filter Magnitude 0 T i i i 0 i i i h uU 01 02 03 04 05 06 0 031 02 03 204 05 06 Normalized Frequency Normalized Frequency Figure 1 19 FIR filters While high pass filters remove low frequency contents low pass filters remove high fre quency values This kind of filters are used for synchronization and also for null symbol detection and automatic gain control As we have seen before these filters could be devel oped in software just updating properly the values for the filter coefficients H3 35 1 8 Error Measures 1 8 Error Measures Some error detection measures were implemented for testing how precise was the system such as BER and MER Thes
160. ort circular buffers As explained in the CoolFlux BSP Assembly Programmers Manual l the start address of such circular buffers must be aligned to a start address that is a multiple of a power of 2 The power of 2 must be equal to or higher than the length of the buffer To efficiently support circular buffers two special functions are available to update pointers TP cyclic_add TP p int offs TP start int len TP cyclic_add_enc TP p int offs TP start int len or The cyclic add function adds an offset offs to a pointer in a cyclic buffer When the pointer reaches the end of the buffer the pointer is automatically wrapped On the CoolFlux BSP the start address is deduced by the hardware from the address of the pointer p the size of the array and the constraint of correct alignment An obvious way to extend the structure in figure 3 8 solution to any multiple of 16 bits would be to replicate the last structure n times one below the other giving a total of 16n queues However 16n pointer update operations would be required each time a frame is received making it quite inefficient A more efficient way is to reserve memory for one queue of a given size consecutively 85 3 6 Time De Interleaving 1 queue 2 queue 3 queue bio bis b s b 5 bres b s5 bas b 13 be103 bio bs q3 b s5 Figure 3 10 deinterleaver for any multiple of 16 bits 24 bits b s5 bio Dira b Figure 3 11 Data Type isimd6 after the ot
161. os 212t Notice that only the amplitude has changed with respect to the last graph but the frequency remains the same ie the peak in the frequency domain is still in 2Hz Let s see what happens with a slightly more complex signal made up of 3 different frequencies for example v t 3 x sin 275t 2 cos 213t 8 cos 2710t We should see in the frequency domain three peaks at points 3 5 10 in the z axis TN 1 4 Nr 1 I l i Ji IM i l 1 NI M IM Ili n li n I i ll a T os EI PN FE o 08 1 15 2 25 3 2 4 6 v We have just analyzed some quite nice signals made up of at most 3 different frequencies but they do not contain any interesting amount of information Just to show a real example I have loaded into MATLAB an audio file from the Internet and plot it both in the time and frequency domain Even incredible it might seem the signal in the time domain is made up of sinusoids just as were the signals we examined so far See that most of the information is contained within the frequencies between 0 DC and 0 15 Hz 164 A 7 Spectral Analysis Now that we have seen how to pass from the time domain to the frequency domain we will see how to do that in MATLAB In MATLAB there exists a function that calculates the
162. osite signal output by this mixer is integrated over 1 ms all of the AC components will cancel because 19 1 5 Orthogonal Frequency Division Multiplexing OFDM they contain whole cycles but the DC signal will accumulate to produce an output signal representing the modulation state of the 200 MHz carrier alone spectrum of incoming signal PANA Wes pis 2kHz 2kHz 1kHz 200MHz 1kHz 2kHz 4 1kHz Ar f N integrator gt x gt OHz gt J 1 a LY one oN NS Ke modulation 24 Signal 1kHz i y of man 200 MHz y j A 2kHz 4 gt integration period symbol duration 1 ms Figure 1 17 demodulation without mutual interference A similar argument applies for each of the other carriers with input frequencies and local oscillator frequencies separated by the reciprocal of the symbol duration T he overall effect of this process is to analyze the spectrum of the incoming signal and to output numerous signals each representing the modulation of one of the carriers If analogue processing were relied upon the acceptable complexity of a domestic receiver would probably limit the maximum number of carriers to as few as 16 but once again the solution is to represent the process digitally It is probably not surprising that the decom position process has a direct counterpart in mathematics known as the forward DFT or simply the DFT The DFT is the digital counterpart of the well known Fourie
163. otives for which we think all the goals of the project have been accomplished 5 0 5 Proposed and accomplished goals The main objective of this project was to design a complete digital radio demodulator following the standard of the european project DAB for the architecture CoolFlux BSP property of NXP Semiconductors and perhaps if there was time simulating it on a FPGA For such ambitious task it was required a highly optimized code in terms of memory consumption and execution cycles The last two aspects play an important role in the actual power consumption Our initial goal was not only achieved but we were able to simulate the demodulator on a FPGA and reproduce audio in real time from a real flemish broadcaster signal 5 0 6 Future aspects Thanks to the advantages that CoolFlux BSP together with its compiler offer us in principle it could be possible to reduce even more the memory requirements and power consumption Once the software is tested enough and further optimized the next step would be to implement a dedicated chip for the commercialization At the present there are several companies interested in this product for its installation in portable devices such as MP3 players telephones and so on 142 5 0 7 Difficulties found within the development phase One of the main burdens has been that of developing a system that had to be able to process within very strict time constraints In fact this part took us aro
164. plied by the amplitude of either wave the positive frequency component is cancelled Since the two input waves are purely real and imaginary respectively the complex addition consists of no more than applying them simultaneously to the appropriate I and Q input ports If the sign of the imaginary sine wave is reversed the result is one positive real impulse function at plus f and the negative frequency component is cancelled Combinations of two sine waves or two cosine waves do not cause cancellation of the second impulse function An FFT would output a single impulse function as one sample amongst many having a non zero value By the reverse argument if a single positive frequency sample is applied to the real input port of an inverse FFT with a positive value the time domain result is a cosine wave at the I output port and a negative sine wave at the Q output port The amplitudes of these 30 1 6 Fast Fourier Transform FFT sampled time waveforms are equal and are proportional to the input sample value and their frequencies correspond to the position of the sample in the input array With a negative value input sample the result is a negative real cosine wave and a positive imaginary sine wave These and other permutations can be derived from the above table by reading right to left Thus it is possible to use the negative frequency samples of an inverse FFT indepen dently of their positive frequency counterparts to pr
165. pointer gives full stack support for the X memory The indexed stack pointer addressing mode implements indexed indirect addressing using the stack pointer as the base pointer meaning that the address is calculated using an offset with respect to the stack pointer In this operation the stack pointer is not automatically updated A full range of operations is provided for updating the stack pointer A number of operations allow moving all registers efficiently from and to the stackl l 2 6 2 Indexed Pointer Addressing The indexed pointer addressing mode implements indexed indirect addressing using the a pointer as the base pointer meaning that the address is calculated using an offset with respect to this pointer In this operation the pointer is not automatically updated A full range of operations is provided for updating the pointers l 2 7 Program Control Unit This unit controls the program counter the status register and the registers used for inter rupt and subroutine linking It controls the flow of the program and implements branches subroutine calls hardware loops and it handles interrupts It also keeps track of the status register It implements 4 registers that are directly accessible in the code The status register contains three flags zero negative and overflow which are set by the XALU There are a number of control bits that control the operation of the CoolFlux BSP Three bits control the action of both RSS unit
166. r AFC this has the effect of averaging noise interference and channel variations l l 3 9 4 Fine Frequency Correction Having achieved coarse AFC the residual frequency error will be less than the carrier frequency separation One effect of a small frequency error is to cause crosstalk between the FFT outputs so the correlation peak will be dispersed to some degree the magnitude of the correlation will rise and fall as the conjugate sequence is shifted and the true peak will lie between two discrete amounts of shift Take for example four adjacent carriers labelled D E F and G to facilitate this explanation which carry part of one of the CAZAC sequences elements labelled P Q R and S The FFT in the receiver operates like a bank of band pass filters followed by demodulators each filter has a sinf f frequency response f being the relative frequency which is broad but exhibits distinct nulls With no frequency error the nulls coincide with the frequencies of the adjacent carriers as illustrated in Fig 3 23 so there is no crosstalk Only one of the frequency responses is shown fully in order to preserve clarity transmitted CAZAC sequence P Q R S carriers in D E F G transmitted signal FFT response DINE demodulator outputs d e f g recovered LL CAZAC sequence P Q R S Figure 3 23 transmission and reception of four carriers with no frequency error Each filter demodulato
167. r BSP Assembly Programmer s Manual version 1 4 November 2008 NXP Semiconductors CoolFlux BSP C Programmer s Manual version 0 3 November 2008 Target The ASIP company http www retarget com resources pdfs IPDesignerlIPProgrammer pdf Maxim 2172 datasheet http www maxim ic com datasheet index mvp id 5883 193 Bibliography 22 Xilinx http www xilinx com support documentation index htm 23 Philips PE 1542 00 Operating manual 24 Wikipedia Baseband Last Updated Abr 2010 http en wikipedia org wiki Baseband 25 Tektronix Datasheets Digital Storage Oscilloscopes http www2 tek com cmswpt psdetails lotr ct PS amp cs psu amp ci 14408 amp lc EN 26 Target The ASIP company Bridge Linker User Manual November 2008 27 F van de Laar N Philips et al General purpose and application specific design of a DAB channel decoder Winter 1993 28 P Kabal and B Sayar Rounding and Scaling in Fixed Point FFT Implementation June 12 1985 29 Matteo Frigo and Steven G Johnson FFTW version 3 2 2 http www fftw org 30 Wikipedia NICAM 26 June 2009 http en wikipedia org wiki NICAM 31 Agilent Technilogies Product description ttp www ome agilent com agilent 32 NXP Semiconductors Software Way of Working CoolFlux DSP Software Development Junary 2008 33 NXP Semiconductors CoolFlux DSP Self explanatory Training Manual May 2009 34 Target The ASIP company Chess Compiler
168. r Transform which relates the time and frequency domains but the input and output functions of the DFT are series of discrete samples rather than continuous signals The DFT is related to the iDFT essentially by interchanging time and frequency The incoming OFDM signal can be sampled in time and the series of samples can be thought of as a function of time As before each of the modulation signals contains one sample per symbol so the array of modulation signals can be thought of as a series of samples which make up a function of frequency that is a description of the spectrum 20 1 5 Orthogonal Frequency Division Multiplexing OFDM of the incoming composite signal The array of oscillator signals can be thought of as a function of both time and frequency as before The action of integrating each output signal corresponds in discrete terms to the sum mation of numerous consecutive discrete samples in time so the decomposition process and the contents of 1 16 is modeled by the DFT which can be written as last time sample f frequency C function of time x oscillator signal first time sample To produce the first sample of the function of frequency that is the modulation signal from the first e g the lowest frequency carrier each sample of the incoming function of time is multiplied by the first frequency sample of the array of oscillator signals e g the lowest frequency one and all of the
169. r debug capable communications channel Embedded systems development relies on debuggers talking to chips with JTAG to perform operations like single stepping and breakpointing 35 A JTAG interface is a special four five pin interface added to a chip designed so that multiple chips on a board can have their JTAG lines daisy chained together if specific conditions are met and a test probe need only connect to a single JTAG port to have access to all chips on a circuit board The connector pins are e TDI Test Data In e TDO Test Data Out e TCK Test Clock 127 4 1 An overview to CoolFlux BSP e TMS Test Mode Select e TRST Test Reset optional Test reset signal is not shown in picture 4 8 TMS TMS TMS TK DEVICE 1 Y DEVICE 2 TK DEVICE 3 TDI TDO TDI TDI TDO Figure 4 8 JTAG interface Since only one data line is available the protocol is necessarily serial The clock input is at the TCK pin Configuration is performed by manipulating a state machine one bit at a time through a TMS pin One bit of data is transferred in and out per TCK clock pulse at the TDI and TDO pins respectively 13 The connection between different hardware components is shown in the schema 4 9 Antenna DAB Tuner DSP JTAG debugger Figure 4 9 Components connection 128 4 1 An overview to CoolFlux BSP 4 1 5 DAB demodulator datapath The implementation of the DAB was done with several CoolFlux BSP cores due
170. r or more convenient ones In this way properties of the former become easier to analyze by studying the properties of the latter functions This is even more important in Physics and Engineering where the ideal of Platonism hold by the most pure mathematicians is replaced by applicability Maybe the first example we learn in our lives about reduction of an arbitrary complex function to a simpler one is that discovered by Taylor in which a function is approximated around a point by a polynomial with all the accuracy we want 4 1st Four Terms of the Taylor Series Expansion of f z 5x x Figure A 1 A 4 degree polynomial approximated around point 0 5 by lower degree ones 145 A 1 Motivation But what does this have to do with the first paragraph Why did I mention the principle of the whole as a sum of the parts Well as we shall see a function can be thought of as composed of atomic subsignals which are simply sines and cosines each one with a different frequency and with a different amplitude just as matter is composed of atoms This is known as Fourier Analysis I do not intend to present the mathematical formality behind Fourier Analysis rather I will try to introduce it in physical terms with applications to Physics and Engineering We shall think of a function as a physical signal varying in time The classic notation of f a is substituted by f t and so on Rather than emphasizing time as the independent v
171. r outputs a complex number labelled d e f and g which carries 98 3 9 Synchronization channel information about the phase of one carrier and the received CAZAC sequence can be recovered from these In the presence of a frequency error the number output by each filter contains contri butions from all of the carriers that is crosstalk The strongest component represents the closest carrier and the relative amplitudes of the other components depend on the relative frequencies of the carriers they represent with a sinf f variation Considering the filter demodulator which outputs the number e in the absence of an error in the presence of an error its output number will contain a component e the prime indicates some difference from e principally a smaller amplitude f with an amplitude even smaller than f d and many other components with much smaller amplitudes The next higher frequency filter demodulator will output a number containing f g e etc and so on for all filter demodulators This is illustrated in Fig 3 24 carriers in transmitted signal FFT response KM Aes E frequency error demodulator outputs d e f g c n d t e n f n recovered g Cea mcos pem Tm IE CAZAC sequence o P Q R S T superimposed sequence P Q R S T U superimposed sequence N O P Q R S Figure 3 24 transmission and reception of four carriers with frequency error
172. r that the inverse Fourier transform uses an integral which is very similar to that of the forward Fourier transform so it follows that the inverse DFT uses a summation which is very similar to that of the forward DFT Also if the input frequency domain array remains static for consecutive inverse DFT calculations the time domain result will be periodic over consecutive time windows If the window duration is equal to or an integer multiple of the period of this result then the result will be a sampled waveform free from discontinuities i e glitches For example with a 1 ms window sinu soidal waves at 1 kHz and harmonics within the Nyquist limit can be portrayed without discontinuities l 1 6 5 The Fast Fourier Transform The FFT is a particularly efficient algorithm for implementing the DFT It increases the speed of processing by cutting down the number of multiplications from n to n 2 logs n where n is the number of input or output samples in cases where n is a power of 2 Thus representing a 1536 carrier DAB signal by means of 2048 samples an FFT would require 11264 multiplications whereas a DFT would require more than 4 million The number loga n has a value of 11 for DAB and can be called the index of the FFT i e 2 2048 The FFT can be derived from the DFT by expressing the summation using matrix arithmetic All of the computations relating the values of the output samples to the input samples can be expres
173. radio signals are usually thought of as using only positive frequencies but the mathematically rigorous definitions of their spectra should include components at negative as well as positive frequencies All of the transforms being discussed require these full definitions For example the spectrum of a cosine wave with frequency f contains two positive impulse functions ie lines in the spectrum at plus and minus f each multiplied by half the amplitude of the wave Of course by trigonometry cos x cos x so this is no different from the simplified view of a single positive impulse function at plus f having both halves of the amplitude In this simplified view the negative frequencies are effectively folded about 0 Hz over into the positive frequency range The spectrum is slightly more complicated for a sine wave of frequency f because sin x sin x the two impulse functions at f are each multiplied by half the amplitude but with opposite signs the positive frequency one having negative sign When such spectra are calculated using the FFT or DFT where the input waveform is expressed as a purely real function of time i e it is presented to the I port and zero is presented to the Q port the transform of the cosine wave is purely real whilst that of the sine wave is purely imaginary This might be expected in view of the orthogonal relationships of cosine and sine waves or real and imaginary numbers It can be
174. re more economical of hardware than using 2048 arithmetic devices in parallel and more economical of processing speed than using one arithmetic device for all computations In the DAB channel encoder the array of complex numbers representing the spectrum of the signal during each 1 ms symbol is applied to the inverse FFT which produces samples of the time domain signal for that symbol These can be converted to analogue form up converted and transmitted In this case full parallel processing is not necessary because the time domain samples need to be output consecutively and not simultaneously although all of the frequency domain samples must be available for each computation In the DAB receiver the frequency domain spectrum is derived from the incoming time domain signal symbol by symbol using the forward FFT However this signal appears via an ADC as a series of consecutive samples so a mirror image of this approach can be used This will produce the spectrum for each symbol when all of the samples have been received but computations can start when only a small number of time domain samples are available two for example It might seem wasteful to have to use 2048 samples to represent the 1536 carriers appar ently wasting 512 but these can be put to good use by purposely setting their amplitudes to zero This can be used in the encoder and the receiver to simulate a band pass filter with an amplitude frequency response much steeper a
175. register via one of the two move buses In this packed move both 28 bit sub words will go through the round and saturate part of the RSS unit and become two 12 bit sub words In a 48 bit packed move cl assignment the Xbus and Ybus move buses are combined to move a 48 bit value in a single processor cycle The 56 bit source value is converted to 48 bits through the saturation logic of the RSS unit In this packed move both 28 bit sub words will go through the saturation logic of the RSS unit and become two 24 bit sub words A 48 bit packed move operation is only possible between two accumulators or from one of the accumulators to the XY memory l l 2 4 4 Round Saturate and Select Units RSS The CoolFlux BSP has two round saturate and select units RSS one for the A accumula tors and one for the B accumulators They convert between values held in 56 bit precision 58 2 4 Move Buses and RSS Units in the accumulators and the precision of a particular move operation 24 bit or 48 bit and are used when a 56 bit accumulator is the source of a move operation scalar or packed as described above In case a sub register is used as source the RSS unit will select the correct part overflow high low of the accumulator and no actual conversion is done In case the full accumulator is the source its value needs to be converted from 56 bits to 24 or 48 bits which requires saturation and rounding Both are handled by the RS
176. rmation Channel FIC o o oo oo 105 3 10 1 1 Fast Information Block FIB 106 3 10 1 2 Fast Information Group FIG 106 iii Contents 3 10 1 3 Calculation of the CRC word 3 10 2 Main Service Channel MSC llle Soll Audio Coding ri 2g ie x TE SUE ddr Y Y duet 3 11 1 Formatting of the audio bit stream o a 3 11 2 CRC check for audio side information 4 Hardware implementation of the DAB demodulator 41 An overview to CoolFlux BSP 4 2 4 1 1 4 1 2 4 1 3 4 1 4 4 1 5 4 1 6 4 1 7 4 1 8 4 1 9 Baseband Signal Processor BSP o llle DAB as a Real Time Application o DAB hardware definition eee Components Description 22e 41 4 1 Maxim 2172 v1 0 5 55 n e wmm m 41 4 2 Tektronix TDS210 4 1 3 FPGA Board Virtex XCAVLX100 4 1 4 4 PE 1542 DC Power Supply Philips 4 1 4 5 Hewlett Packard 8596E Oscilloscope 4 1 4 6 CoolFlux simulator CoolFlux debugger Chess compiler 4 1 4 7 Reception Antenna een 4 1 4 8 JTAG Joint Test Action Group DAB demodulator datapath Local Memories ee cesera Re ES eA Y s Debug interface assente Sont Tee mter ene ean LOUP ad utet Interrupt interface ee ee Shared Buffers see a UTER CREE a a a Descrip
177. ro Blasco Agus and Saul Among other things I will never forget those marvelous days during our summer holidays in La Marina I also want to thank you Andr s for your wisdom and special understanding towards me In Memoriam my father A M L Abstract The new digital radio system Digital Audio Broadcasting DAB is a digital radio technology for broadcasting radio stations which will likely replace most existing methods for radio broadcasting in many parts of the world Mobile reception without disturbance was the basic requirement for the development of the DAB system The special problems of mobile reception are caused by multipath propagation the electromagnetic wave will be scattered diffracted reflected and reaches the antenna in various ways as an incoherent superposition of many signals with different travel times which in classical radio systems like AM or FM meant a deficit in the quality of sound or even complete cancellation This document will present the software development of a DAB demodulator for the CoolFlux BSP and its subsequent implementation on the ultra low power architecture CoolFlux BSP developed by NXP Semiconductors For the implementation of the DAB system a hard work in signal processing and code optimization techniques has been carried out aimed to get the smallest power and memory storage Important correction codes and measures have also been applied to the data for ensuring a good quality reception
178. ronization l In the case of the DAB input signal this must be scaled up or down to make use of the 24 bits of resolution available within the architecture The power reference for the CoolFlux is around 12dB Initially the incoming signal is multiplied by a gain factor set to 1 no amplification is performed The average power is computed during a fixed window of IQ samples After that the gain is turned up or down depending on whether the average power of the window is greater or lower respectively than the power reference A low pass filter is required in order to avoid the average power to go up or down too fast due to spurious abrupt i e high frequency changes in the power of the incoming signal As an example of a possible sharp change in the signal power is the reception of the null symbol Without the low pass filter the null symbol would be amplified up to 12dB thus making its detection impossible A low pass filter can be implemented in software as follows AvgPwr Tav x IQpwr 1 Tav x AvgPwr IQpuwr is the power of the current IQ sample Avg Pwr is the average power of all samples received so far within the current window T av is a parameter of the filter related to the frequency you want to filter Notice that If you increase Tav you force the power of the current symbol to contribute stronger to the average power Once the average power of a complete window has been calculated it is time to adjust the gain factor
179. rted to 56 bits by sign extending it with 8 bits A 48 bit 56 2 4 Move Buses and RSS Units move operation is only possible between two accumulators or from the XY memory to one of the accumulators In a 24 bit packed move c assignment the 24 bit value is split into two 12 bit sub words When converting to 56 bits each 12 bit sub word is converted to a 28 bit accumu lator sub word This is done by sign extending the upper 4 bits and padding with 12 zeros at the LSB side The 56 bit accumulator is then made by concatenating both sub words accumulator 55 52 accumulator 51 40 input 23 12 accumulator 39 28 accumulator 27 24 input 11 amp input 11 amp input 11 amp input 11 accumulator 23 12 input 11 0 accumulator 11 0 In a 48 bit packed move cl assignment the Xbus and Ybus move buses are com bined to move a 48 bit value This 48 bit value is converted to 56 bits by sign extending each of the 24 bit sub words with 4 bits A 48 bit move operation is only possible between two accumulators or from the XY memory to one of the accumulators l accumulator 55 52 input 47 amp input 47 amp input 47 amp input 47 accumulator 51 28 input 47 24 accumulator 27 24 input 23 amp input 23 amp input 23 amp input 23 accumulator 23 0 input 23 0 2 4 3 Moving from an Accumulator The
180. s We will normalize the functions by multiplying the arguments in the sinusoids by 27 f where f is the frequency of the sinusoid In this way if we convey that time is measured in seconds then the units of f are in Hz So for example the voltage in AC between two points can be given by V t 3sin 274t In this case the voltage will oscillate 4 times per second with a maximum peak value of 3 Volts Everything explained in earlier chapters applies equally well to the normalized representation If we have the function g t sin 3t we can normalize it just by calculating f 3 and substituting f into g t sin 2m ft The following pictures show on the left the function v t 3sin 272t In this case f 2Hz As expected we see a peak of amplitude 3 at the point 2 in the x axis in the frequency domain You might wonder why we do not obtain a perfect peak but there appears some residual frequencies around the point 2 in the z axis Presumably this is because we are dealing with discrete signals made up of little steps These steps seem to introduce such undesirable frequencies The frequency domain graph does not represent whether the frequency was due to a sine or a cosine component in the time domain This is because both sine and cosine do 163 A 7 Spectral Analysis not differ in frequency they do in phase In order to see this the following pictures shows the function v t 3sin 272t 2c
181. s sr b sr r sr s The sr ie bit is the 62 2 8 Hardware Interface Name Long name Width Description sr Status Register 24 bits Contains the machine control and status word Ir Link Register 24 bits Contains the return address after a call instruction ilr Interrupt Link Register 24 bits Contains the return address after an interrupt Isr Interrupt Status Register 24 bits Itis used to save the value of the status register sr during an interrupt interrupt enable disable bit Two bits are used for the packed division operation sr div0 sr div1 Four bits keep track whether an overflow could occur in a packed move There are two bits for the imaginary part or the first word and two bits for the real part or the second word sr ov i sr ov r sr ov i2 sr ov r2 A last bit differentiates between an SIMD operation and a complex operation sr simd When this bit is set and a packed operation is executed it will always be an SIMD operation Note that the only difference between an SIMD operation and a complex operation is in the multiply operation All other arithmetic operations and all moves are identical ll 2 7 1 Hardware Loop Stack Hardware loops are implemented through a hardware loop stack The stack is 4 levels deep so that up to 4 nested loops are supported by hardware When a loop is initiated the start and end address of the loop and the number of times the loop must be executed is pushed onto the loop s
182. s bits de informaci n La modulaci n DQPSK es esencialmente igual excepto que no se utilizan fases absolutas para convenir la informaci n sino que sta es codificada en la diferencia de fases entre el s mbolo actual y el previo B 5 Frecuency de interleaving Adem s del s mbolo de referencia de fase un marco completo de transmisi n o frame se compone de 75 s mbolos OFDM En DAB se utilizan 1536 portadoras en paralelo cada una de las cuales transporta 2 bits de informaci n en cada periodo Una estrategia naive para la transmisi n de la se al ser a dividir la cadena de bits en 75 bloques de 3072 bits y asociar el primer par de bits con la primera portadora el segundo par de bits con la segunda portadora y as sucesivamente Sin embargo dicha estrategia es propensa a errores de tipo burst El frequency de interleaver reordena la asignaci n de cada par de bits a cada portadora de una forma pseudoaleatoria convenida en el est ndar de DAB 176 B 6 Cuantizaci n B 6 Cuantizaci n Una vez que se ha llevado a cabo la demodulaci n diferencial el n mero complejo resultante de la resta tiene que ser convertido a bits reales La cuantizaci n es el proceso mediante el cual se reconstruye el dibit codificado en cada s mbolo IQ y enviado en la transmisi n Para resolver este problema se implementaron dos soluciones a saber la hard decision y la soft decision La entrada al cuantizador se puede representar en los ejes cartesia
183. s de la respuesta al impulso del canal de transmisi n tiene la longitud de un s mbolo normal y toda su capacidad es utilizada para funciones de sincronizaci n B 10 Implementaci n hardware del demodulador DAB La implementaci n de nuestra aplicaci n se ha realizado en la arquitectura CoolFlux BSP versi n 2 0 un procesador de se al baseband de ultra bajo consumo especialmente dise ado para aplicaciones de audio por la compa ia NXP Semiconductors donde se realiz este proyecto Una de las grandes dificultades encontradas en este proyecto fue la de poder realizar la demodulaci n de la se al al mismo tiempo que sta es transmitida A este tipo de sistemas se les conoce por sistemas en tiempo real Si el demodulador no fuera lo suficientemente r pido para procesar todos los s mbolos OFDM dentro de un frame de duraci n muchos datos podr an llegar a perderse 181 B 10 Implementaci n hardware del demodulador DAB DSP JTAG debugger Figure B 5 Conexi n de los componentes El primer paso para las pruebas fue generar el c digo VHDL para la FPGA con la es pecificaci n de la ruta de datos y crear la estructura del sistema y los dispositivos perif ricos osciloscopio sintonizador depurador etc en MATLAB Simulink Mientras un BSP es capaz de funcionar a 300MHz en una implementaci n real la FPGA utilizada Virtex XC4VLX100 s lo es capaz de funcionar a 50MHz CoolFlux BSP tiene un CPI Cycles per
184. s processed as follows 11 e Audio program signals are digitized and multiplexed together with ancillary data to produce an un coded bit stream e The bit stream is then encoded for forward error protection by adding redundant bits with appropriate calculated values 71 During each consecutive 1 ms symbol the coded bits are divided into 1536 pairs and each pair is differentially encoded with respect to its counterpart for the previous symbol The 1536 differentially encoded bit pairs are presented to an inverse FFT where each is used to define the phase of a QPSK carrier collectively they specify the spectrum of a 1536 carrier signal The inverse FFT synthesizes a time domain signal which has the specified spectrum and this signal is converted to analogue form frequency converted then transmitted This is the OFDM generation process and it is repeated symbol by symbol In the receiver the incoming OFDM signal is frequency converted to lower frequencies appropriate to the hardware digitized and applied to an FFT Here the spectrum is analyzed symbol by symbol and the phases of the 1536 carriers are determined This corresponds to OFDM decomposition The high resolution digital complex number which represents the phase of each carrier is divided by the value for the previous symbol in order to detect the differential QPSK The resulting 1536 differentially decoded numbers are passed to an error
185. s saved into the isr register the interrupts are disabled the return address of the interrupt is stored in the interrupt link ilr register and then the processor jumps to an address in the interrupt vector table This address is dependent of the source of the interrupt 2 13 BSP C Compiler To ease the development of applications on this processor a very efficient optimizing C compiler is available which is based on a retargetable tool called IP Designer TM patented by the company Target Compiler Technologies IP DesignerTM is a retargetable tool suite for ASIP Application Specific Processors Design application specific processors design The designer can easily describe each ASIP 67 2 13 BSP C Compiler Interrupt Vector Table Address Interrupt Source Remark 0 RESET Hardware reset 1 INTO Hardware interrupt 2 INT1 Hardware interrupt 3 INT2 Hardware interrupt 4 SWA Software interrupt 5 SW5 Software interrupt 6 SW6 Software interrupt 7 SW7 Software interrupt core in nML the world s first commercially available high level processor description lan guage that offers an unprecedented architectural breadth From nML an efficient SDK is generated instantaneously containing e An optimizing c compiler that maps algorithm descriptions into highly optimized ma chine code for the target ASIP The unique and patented graph based modeling and optimization technology in the C compiler enables full exploitation of instruction
186. se al ha sido sincronizada en el primer BSP el segundo BSP comienza con la demodulaci n de sta Se utiliza el centinela es decir un valor que no se encuentra entre los posibles de la emisi n de la se al 0000FFFF FFFF0000 como entrada a BSP para indicarle que empiece a trabajar Este m dulo consiste en una cadena de algoritmos como se explica en la figura Freq Differential eem F di Eom FFT Deinterleaving demodulator Quantization TO B8E2 Figure B 9 BSP modules Para asegurarse de que el ltimo BSP BSP recibe los datos en el orden correcto es necesario escribir en el buffer compartido el n mero del s mbolo enviado Es importante resaltar que tras la cuantizaci n solamente usamos 1536 palabras de las 2048 disponibles en el buffer intermedio Utilizamos este espacio libre para escribir este n mero En este m dulo es muy importante la reutilizaci n de la memoria ya que sin ello hubiera sido imposible la implementaci n 187 B 12 Conclusiones Despu s de esto se enviar la interrupci n correspondiente al BS P5 B 11 3 BSP FIC and CIF Decoding Audio decoding Este es el BSP del final del pipeline y como tal tiene que ser capaz de obtener los bytes del audio que se reproducir con circuiter a espec fica del cliente del sistema CoolFlux BSP DAB Para decodificar la informaci n del FIC es necesario almacenar en un buffer todos los s mbolos que lo conforman Una vez stos han sido recibidos la c
187. sed in a two dimensional matrix Individual elements of this matrix can be broken down into consecutive stages of simpler arithmetic that is they can be 26 1 6 Fast Fourier Transform FFT factorized just as x 31 2 can be factorized into x 1 x 2 The matrix as a whole can be factorized into a number of matrices containing simpler expressions and when this process is taken as far as possible the number of factored matrices is equal to the index This factorization process sometimes referred to as decimation introduces several simplifications some expressions always return zeros or ones so they need not be calculated and some others have counterparts in the same matrix which yield the same result but with the opposite sign The overall benefit is the reduction in the number of multiplications required There are different approaches to decimation which yield the same overall result but with greater internal complexity towards either the time domain or frequency domain end of the chain of matrices these are known as decimation in time and decimation in frequency respectively FFT is really a generic name for this type of algorithm and there are many variants the main differences being in the paths taken through the factored matrices The FFT algorithm is commonly based on a radix of 2 that is the numbers of input and output samples are equal to 2 raised to some power A larger radix e g 4 8 etc
188. sequence be represented in the autocorrelation calculation if the sequence was truncated for some reason the ZAC property would be lost The result for any discrete amount of shift can be presented as a histogram as shown in Fig 3 20 A shift of 16 elements or any multiple restores the original sequence so the autocorrelation peak is said to be periodic Some useful properties of a CAZAC sequence are as follows 94 3 9 Synchronization channel 1 j 1 j 1 431 11 j 1j 1111 1 X j 1j 1 11 11 j 1 j 1111 1 j j j j 1 1 1 1 j j j j 1 1 1 1 0 Figure 3 19 The autocorrelation with shift of one A autocorrelation 16 gt a Als e wc A o 012 3 4 5 6 7 8 9 10 11 12 13 14 15 16 shift Figure 3 20 Autocorrelation VS Shift e If all elements of the original sequence are multiplied by some constant factor which may be complex the magnitude of the result becomes multiplied by the same factor and the zero autocorrelation property is preserved e If the original sequence is corrupted by the addition of one or more shifted versions of the same sequence individual correlation peaks are found for each of the sequences at appropriate shifts Last property applies so if the different sequences are multiplied by constants the complex values of correlation are multiplied in the same way e If the original sequence is corrupted by noise or interference some or all elements will have modif
189. several own analysis their developers believe that it is particularly because the core can use larger intermediate results to maintain precision And when it is not users can switch to the slower 24 bit mode How well customers are able to make use of the 118 4 1 An overview to CoolFlux BSP cores maximum throughput using 12 bit data will have a big effect on the performance they are able to squeeze out of the corel 6l Power Consumption 31 y W MHz Q 0 8 V 65 nm CMOS 70 y W MHz 1 2 V 65 nm CMOS 40 u W MHz 0 8 V 90 nm CMOS 90 y W MHz 1 2 V 90 nm CMOS 53 y W MHz Q 0 8 V 130 nm CMOS 120 y W MHz Q 1 2 V 130 nm CMOS 390 u W MHz 1 8 V in 180 nm CMOS Performance Worst Case Commercial Conditions standard VT 150 MHz 1 8 V 180 nm CMOS 190 MHz Q 1 2 V 130 nm CMOS 245 MHz 1 2 V 90 nm CMOS 300 MHz Q 1 2 V 65 nm CMOS gt 5000 MOPS peak 4 1 1 Baseband Signal Processor BSP In telecommunications and signal processing baseband is an adjective that describes signals and systems whose range of frequencies is measured from zero to a maximum bandwidth or highest signal frequency it is sometimes used as a noun for a band of frequencies starting at zero X f B B f Figure 4 2 Spectrum of a baseband signal amplitude as a function of frequency A signal at baseband is often used to modulate a higher frequency carrier wave in 119 4 1 An overview to CoolFlux BSP order
190. si n apropiada de la energ a de la se al transmitida y evitar de este modo que patrones sistem ticos den como resultado una regularidad no deseada en la se al trasmitida del m dulo los bits de entrada son transformados mediante una suma modulo 2 cuyos par metros son los propios bits de entrada y una secuencia binaria pseudo aleatoria llamada PRBS Pseudo Random Binary Sequence Por consiguiente es necesario hacer alguna operaci n para reconstruir la se al original Esta operaci n es la misma que se realiza en el transmisor sumar m dulo 2 los bits aleatorizados y la secuencia PRBS La adici n modulo 2 equivale a una XOR a nivel de bit B 9 Canal de Sincronizaci n Debido a condiciones adversas del canal es posible que la frecuencia y fase de los distintas portadoras se vean perturbadas Dichas perturbaciones provocan que el receptor no de 179 B 9 Canal de Sincronizaci n module correctamente la se al Para evitar este tipo de errores se introducen dos s mbolos especiales de sincronizaci n en el frame de DAB B 9 1 S mbolo Nulo En el frame de radio existe un s mbolo especial llamado Null symbol con el que se realiza la sincronizaci n del frame Su detecci n se realiza gracias a que su potencia es cero Para detectar este s mbolo se utiliza una ventana de decisi n de 512 palabras para la que se calcula la potencia media de la se al entrante El s mbolo nulo se detecta cuando hay una ca da por debaj
191. t 5 sin 9 2 x pi t 2 x sin 4 x 2 pi x t subplot 2 1 1 plot t y grid on axis 0 et 10 10 xlabel Timels ylabel Amplitude Y fft y n size y 2 2 amp spec abs Y m subplot 2 1 2 freq 0 79 2 n dt plot freq amp spec l 80 grid on xlabel Frequency H z ylabel Amplitude We can simulate noise interference by calling the function randn Time s 6 T T T T T T T T T L ad al MN 2L i i d 4 0 AAN m p cT m 0 2 4 6 8 10 12 14 16 18 20 Frequency Hz Figure A 6 Noisy signal and its spectrum See that in addition to the three peaks of the original signal there appears residual frequencies with low amplitudes that were not present earlier Therefore if we are to filter the noise we must reduce to 0 such low amplitudes in the frequency domain Look carefully at the next code 167 A 8 Applications of the Fourier Transform noise randn 1 size y 2 ey y noise eY f ft ey n size ey 2 2 amp spec abs eY n figure subplot 2 1 1 plot t ey gridon axis 0 et 10 10 xlabel Timels ylabel Amplitude subplot 2 1 2 freq 0 79 2 n dt plot freq amp_ spec 1 80 grid on slabel Frequency H z ylabel Amplitude Time s i i i 1 i 1 i i 0 2 4 6 8 10 12 14 16 18 20 Frequency Hz Figure
192. t form 1 long form 107 3 10 Transport mechanisms e Sub channel size this 10 bit field coded as an unsigned binary number in the range 1 to 864 shall define the number of Capacity Units occupied by the sub channel The remaining fields are out of the aim of this documentation DAB ENSEMBLE ONE ALPHA 1 Ensemble Services Service components Audio Fast Information Channel Main Service Channel Figure 3 32 An example of the DAB service structure The organization of the sub channels services and service components in an ensemble is managed by the MCI The MCI serves five principal functions a To define the organization of the subchannels in terms of their position and size in the CIF and their error protection b To list the services available in the ensemble c To establish the links between service and service components d To establish the links between service components and subchannels e To signal a multiplex re configuration As FIC symbols come in they are processed In each FIC data different FIG types are contained usually this types are 0 1 and 7 Once all the symbols into a frame has been processed one structure for recover the information is created All services are stored which their respective service component For each service component just one subchannel belongs identified by its subchannel ID H 108 3 10 Transport mechanisms 3 10 1 3 Calcu
193. t imply additional storage merely a re ordering of the inverse FFT either method is equally valid l6 1 7 Finite Impulse Response Filters FIR These filters deserve attention since they are used several times during the implementation of the demodulator In signal processing there are many instances in which an input signal to a system contains extra unnecessary content or additional noise which can degrade the quality of the desired portion In such cases we may remove or filter out the useless samples 1 7 1 How to characterize digital FIR filters There are a few terms used to describe the behavior and performance of FIR filter including the following e Filter Coefficients The set of constants also called tap weights used to multiply against delayed sample values For an FIR filter the filter coefficients are by defini tion the impulse response of the filter 34 1 7 Finite Impulse Response Filters FIR e Impulse Response A filter s time domain output sequence when the input is an impulse An impulse is a single unity valued sample followed and preceded by zero valued samples For an FIR filter the impulse response of a FIR filter is the set of filter coefficients e Tap The number of FIR taps typically N tells us a couple things about the filter Most importantly it tells us the amount of memory needed the number of calculations required and the amount of filtering that it can do Basically the m
194. t pairs are then affected by events which are uncorrelated in the frequency domain and the dispersal of clusters of bit errors improves the performance of the Viterbi decoder This process is known as frequency interleaving In practice the bit pairs are formed by partitioning the incoming bit stream into blocks of 3072 consecutive bits and associating the 1st bit with the 1537th bit the 2nd bit with the 1538th bit and so on This adds a further element of dispersal The data carried in the Phase Reference symbol are not subjected to frequency inter leaving but are available as a further 1536 bit pairs mapped to particular carriers Each of the 1536 bit pairs to be transmitted is given an index 0 to 1535 and each of the 1536 carriers is given an index 768 to 768 omitting 0 which corresponds to the un modulated centre carrier The series then relates the bit pair index to the carrier index and is constructed as follows First an intermediate series is constructed starting with 0 as the value of the first term The value of the next term is calculated by multiplying the value of the previous term by 13 and adding 511 with the proviso that if the result is greater than 2047 then 2048 is repeatedly subtracted from it until the result is less than 2048 but positive i e the result is taken modulo 2048 This is repeated to yield 2048 different values 0 511 1010 1353 1221 Then from the intermediate series only those values are sel
195. t signal in blue in superposition with the gain factor in red Presumably the power of the input signal is increased maybe due to multipath propagation at around point 12 in the x axis Consequently the gain factor is immediately decreased to maintain the power of input signal at safe levels As an observation an excess in power does not 75 3 2 FFT Figure 3 2 AGC amplifying a real DAB signal NXP Offices Leuven Untere and tered I power afer AGC Figure 3 3 Average signal power in superposition with the gain factor UM M biet WIN NW li j LU make the DSP to blow up but it might cause some operations in the demodulation chain to overflow thus corrupting the information 3 2 FFT The FFT and its relation with DAB was explained in chapter 1 Recall that it is used to extract each of the 1536 subcarriers of the input signal Two implementations of the FFT were tested namely the radix 2 and radix 4 These FFTs were already implemented by the team of NXP so we did not have to rewrite it Our 76 3 3 Differential Demodulation work was to analyze both the number of MIPS consumed by each FFT and the error of the output vector Each input sample to the FFT is a complex number with type complex12 Thus both the real part and the imaginary part are represented with 12 bits Note that we are in a fixed point architecture so the precision of the FFT is more lim
196. t the band edges than can be achieved using an analogue filter 6l 1 6 7 Complex Numbers The Fourier transform operates with complex numbers in both domains and this applies to the DFT and FFT derived from it and their inverses Whilst it is usually necessary to consider components of a spectrum as complex having amplitudes and phases the waveform of a radio signal is purely real simply the variation of a voltage with the passage of time This is not to say that such a waveform could not be specified using complex quantities 28 1 6 Fast Fourier Transform FFT only that division of its specification into real and imaginary parts would require that they be combined in some way before the final waveform could be generated Essentially the input and output arrays of the FFT can be divided into real and imag inary parts When complex numbers are represented each real sample has an imaginary sample associated with it Conventionally in this field of engineering the real and imagi nary parts of the time domain array i e the input array of an FFT or the output array of an inverse FFT are referred to as the I and Q ports for In phase and Quadrature respectively l9 1 6 8 Negative Frequencies It was noted earlier that up to the Nyquist limit the DFT and the FFT produce as many output samples as are input but in each case half of the frequency domain samples represent negative frequencies Real
197. tack When a loop finishes the loop information is popped from the stack The stack consists ofl18l e Four 24 bit start addresses e Four 24 bit end addresses e Four 24 bit loop counts 1 16777215 2 8 Hardware Interface The CoolFlux BSP has extensive I O facilities These include a 16 Mword I O memory for mapping peripherals three hardware interrupts a data DMA interface to access the data 63 2 9 Multi Cycle Instructions and Delay Slots memories and a program DMA interface to access the program memory The CoolFlux BSP Interface Manual See bibliography outlines the hardware interfaces of the CoolFlux BSP in more detail including signals and timing l l 2 8 1 I O Memory There is a 16 Mword 24 bit I O memory that can be addressed in the same way as the Y memory Hl 2 82 DMA The CoolFlux BSP has an arbitrated memory access scheme to all memories which is supported by a fully hand shaken DMA access mechanism The data DMA port has a unified address space that can map onto X memory or Y memory The program DMA port has access to the P memory Access to the X memory Y memory and P memory is possible when the CoolFlux BSP is running a program The CoolFlux BSP can support program and data caches through the usage of core hold signals Refer to the CoolFlux BSP Interface Reference Manual for more details of this interface Ul 2 9 Mlulti Cycle Instructions and Delay Slots The CoolFlux BSP is a pipelined proc
198. te and func tionally debug the C code very fast before it is compiled and optimized for the CoolFlux BSP using the Chess compiler The compilation using your standard C compiler is called native compilation 119 2 13 1 Optimizations The directives for the C compiler allow code optimizations and steering whilst remain ing completely within the unified C language environment You are advised to write the majority of the code for the CoolFlux BSP in ANSI C resorting to assembly language programming only when absolutely necessary 119 70 Chapter 3 DAB Implementation on the CoolFlux BSP In this chapter we explain the demodulation process of DAB together with its actual im plementation under the CoolFlux BSP The modulation techniques explained in chapter 2 should be fairly enough to understand this chapter From an abstract point of view the input to the DAB demodulator consists of an OFDM DQPSK modulated signal represented numerically by a set of complex numbers The output is the actual sound other kind of data might be also present in the signal like textual information The DAB demodulation is divided up in a chain of modules each one performing a transformation on its data input thus providing partial results to the following module The aim of this chapter is to explain the functionality of each module together with their implementation and performance under the CoolFlux BSP A simplified DAB transmission chain i
199. ted by means of 2 pointer register files XPTR YPTR These are located in the addressing units The XPTR register file contains 8 pointers xptr0 xptr7 the YPTR register file contains 4 pointers yptrO yptr3 When the memories are accessed indirectly via one of these pointers the pointer value can be updated post modified in parallel with the access The calculations for this update are executed on the XAGU and YAGU the address generation units The post modification operation is specified as part of the memory access These operations use the registers in the other register files XSTEP YSTEP XMOD and YMOD The possible post modification operations are e nop e increment e indexed stack pointer addressing e decrement e increment by 2 e decrement by 2 e add the step register e subtract the step register e and a special mode which is used for cyclic buffer addressing and bit reverse address ing 61 2 7 Program Control Unit Besides the update operations that are part of the indirect addressing syntax there are a number of specific pointer operations that are executed on the XAGU and YAGU The operations that execute on the XAGU and YAGU are executed in parallel with the move operations on the Xbus and Ybus move buses the accesses to memory and the operations on the data path XMUL YMUL XALU and YALU HSI 2 6 1 Indexed Stack Pointer Addressing The X addressing unit contains a stack pointer sp This stack
200. ter values to the right m moves to mo mo moves to m 1 and wait for the next input bit If there are no remaining input bits the encoder continues output until all registers have returned to the zero state The figure 1 21 below is an example of a convolutional encoder x n is the input and Go n and G n are the encoded outputs The rate of the convolution coder is defined as the number of input bits to output bits This system has 1 input and 2 outputs thereby resulting in a coding rate of 1 2 The number of states of a convolution coder is determined 38 1 9 Errors Detection and Correction by its number of delay units There are 2 delay units in this system and therefore there are 2 4 states For a system with k delay units there are 2 states lt P O as w Lm D x n 2 x n 1 E an Figure 1 21 Convolutional Decoder The system block diagram can be expressed with the following equations Gi n x n a n 2 The state diagram of this system is depicted in Figure 1 22 The states are defined as x n 1 z n 2 pairs and the state transitions are defined as Go n Gi1 n z n The states are indicative of system memory The state transitions gives you the path and the outputs associated with a given input Since a state transition is associated with each possible input the number of state transitions depends on the number of possible inputs If there are
201. th the new value of the frequency correction a phase correction is applied to the TFPR Symbol This symbol is also copied to the shared memory as will be done with the rest of the symbols The TFPR symbol is written twice in different buffers One will be passed to the next core BS P and will be used as the previous symbol for the differential decoder see last chapter The other one is necessary because of its role as a synchronization symbol and is used in the main loop It is not possible to reuse this space of memory because they have different life cycles and when the TFPR symbol is used by BS P BS Py has already overwritten it Note that since the length of a symbol is bigger than 512 samples this state is for sure executed more than once SKIP GI This is the state responsible of skipping the Guard Interval samples For these samples a different algorithm of Frequency correction is applied which is called Fast Frequency Correction and differs from the ordinary frequency correction algorithm in that it just adjusts the phase without using the Sine and Cosine Table This algorithms allowed us to reduce the memory requirements and the processing time PROCESS OFDM Symbols are copied into the shared buffer and then a signal is sent to B SP The frequency correction is also applied here In this BSP every subchannel is saved as at this time we still do not know which ones are going to be used when the user selects a s
202. the carrier frequency separation the reproduced sequence will be shifted with respect to those outputs This can be detected by shifting the stored conjugate sequence or changing the local oscillator frequency and searching for the correlation peak Coarse AFC can be provided in this way with a capture range of at least 8 carrier separations i e 8 kHz in Mode 1 The Phase Reference symbol block is not subjected to frequency interleaving because this would require additional processing in the receiver and it would confer no advantage in conditions of selective fading a CAZAC sequence is equally sensitive to corruption of adjacent or non adjacent elements Also time interleaving cannot be used because the AFC needs to be updated from transmission frame to frame in order to optimize the receiver performance in conditions of changing radio frequency i e oscillator drift or changing Doppler shift and the attendant time delay could not be tolerated Consequently the ruggedness of this system is dictated by the length of the CAZAC sequences alone However very long sequences i e across many carriers could be corrupted by variations in the phase frequency response of the transmission channel The use of a large number of shorter sequences is relatively beneficial because the complex results of individual 97 3 9 Synchronization channel correlations can be added together as vectors and the magnitude of the resultant used fo
203. ticularly in a changing environment such as when the receiver is located in a moving vehicle The effect on broad cast FM radio is well known where in built up areas even if there is sufficient mean field strength mobile reception can be severely impaired by bursts of noise and audio distortion and sometimes a fluttering effect on the audio signal In addition to a direct signal from the transmitter the receiver is often presented with signals reflected and diffracted by buildings and the terrain These can combine construc tively or destructively in the receiving antenna as the relative lengths of the propagation paths change or as the wavelength changes Indeed the sensitivity to the wavelength or frequency is the main reason for the audio distortion with FM signals Constructive addition can give up to 6 dB enhancement for two signals of equal mag nitude but subtraction can cause complete cancellation This phenomenon is known as fading and multipath propagation is one of two mechanisms by which it can be caused the other is blocking of the propagation path by obstacles In the latter case the prob lem can often be overcome simply by increasing transmitter powers but this is not always true for multipath fading destructive combination of two signals of equal magnitude causes complete cancellation regardless of the transmitter power When an FM receiver is pre 1 2 Multipath Propagation sented with insufficient signal power
204. tings The display responds quickly to control adjustments and has a fast update rate 9l With this instrument we could carry out the testing checking in real time the values from the input signal registers and variables 125 4 1 An overview to CoolFlux BSP 4 1 4 3 FPGA Board Virtex XCAVLX100 Simulation on real time of the DAB demodulator was done in an FPGA Board Nallatech by loading the data through the CoolFlux Debugger version 1 1 0 Nallatech has its own API FUSE to manage the configuration control and communica tion with the rest of the system With it comes Dimetalk that for some functions transform restricted type of C code DIME C into VHDL and can generate the communication net work between the various algorithm blocks memory blocks I O interfaces l Figure 4 7 FPGA Board Virtex XC4VLX100 4 1 4 4 PE 1542 DC Power Supply Philips The PE 1542 is a stabilized D C power supply designed for supplying and testing electrical and electronic circuits It produces three precise output voltages which are galvanically separated All outputs are protected against short circuit Regulation of each output is achieved by transistor series regulator stages The transistor series regulator gives contin uously variable adjustments of the output voltages and currents with good accuracy and stability with minimum ripple By means of control each output voltage can be continu ously varied from OV to 20V or 7V depending upo
205. tion of the BSP instances ss 4 2 1 4 2 2 4 2 3 BSP Synchronization Time correction Save OFDM symbols BSP FFT Deinterleaving Demapping BSP FIC and MSC Decoding Audio decoding Contents 5 Conclusions 5 0 4 JS tart pOint ceci os ende AR Seer I P Y Run 5 0 5 Proposed and accomplished goals 50 0 B t ure aspects traca al cete eum Pm vam ee 5 0 7 Difficulties found within the development phase Appendices A The Fourier Transform AT Motivations ue sash AER ede A cies A 2 Approximation by the minimum mean square error A3 Orthogonality ya e rU VR ee eee iude A 4 Fourier Transform in Periodic Complex Functions A 5 Integral Fourier Transform ee ALO DET xxx soe a eR dr Ado Rd sues AT Spectral Analysis 5 o 4 4b kot b dodo ov ae ber ee y ox A 8 Applications of the Fourier Transform AsO ls Noise Filtern aei Ron RP Reus heh beu A 8 2 Image Compression 2A B Resumen en espanol del proyecto B 1 Implementaci n software de DAB sobre la arquitectura CoolFlux BSP B 2 Control Autom tico de Ganancia ooa B3 ERI 5t nuu aeneis pee Rare Ie daban a B 4 Demodulaci n diferencial len B 5 Frecuency deinterleaving eA B 6 Cuantizacl n duc ERROREM REI IAM VY B 6 1 Hard Decisi n 69S AE B 6 2 Soft Decision 4 iu acm a a a B 7 Time De Interleaving
206. tiplexing The process by which multiple broadcaster share the radio spectrum is called multiplexing If two or more broadcasters want to emit at the same frequency there has to be an agreement between both to send their signals at different times Otherwise their signal would interfere with each other destroying the information This turning on the use of the channel is what is called in telecommunications TDM Time Division Multiplexing l 13 1 5 Orthogonal Frequency Division Multiplexing OFDM A more common way of sharing the spectrum is to allow each transmitter to emit at a different frequency We avoid interference among different frequency carriers by introducing an additional guard space between them This process is called FDM Frequency Division Multiplexing As an example the next picture illustrates the spectra of a signal composed of 8 subcarriers l 8 Subcarriers Guard Guard Bend Guard Band Band lt Frequency Spectrum Figure 1 14 Frequency Spectrum 1 5 Orthogonal Frequency Division Multiplexing OFDM The bandwidth of a digitally modulated signal is proportional to the symbol rate The exact relationship depends on the modulation scheme and factors such as filtering but generally for a given modulation scheme doubling the symbol rate doubles the bandwidth The simple way to increase the bandwidth of the DAB signal without compromising its spectral efficiency would be to make
207. to de la era multimedia se ha producido un espectacular aumento de los requerimientos computacionales para abordar funcionalidades cada d a m s demandadas Por otro lado cabe comentar que el ciclo de vida de los dispositivos m viles y multimedia se est reduciendo considerablemente debido a la presi n del mercado lo que provoca a su vez una reducci n del tiempo de desarrollo de los procesadores Esto significa que lograr la m xima eficiencia computacional con el m nimo consumo de energ a y en el menor tiempo posible se ha convertido en la meta de varias compa as de desarrollo de DSPs B 12 1 Punto de partida El desarrollo de este proyecto ha supuesto un gran esfuerzo por parte de los integrantes del grupo debido al amplio rango de conceptos que ste engloba Previamente a su imple mentaci n fue necesario un estudio a fondo tanto de procesamiento de se ales digitales no cubiertos en el plan de estudios de la Ingenier a en Inform tica como del est ndar DAB y de la arquitectura CoolFlux BSP No obstante gracias a la ayuda de profesionales dentro 189 B 12 Conclusiones de este mbito hemos realizado un desarrollo del mismo que cumple ampliamente con las expectativas iniciales En los siguientes apartados exploraremos los resultados obtenidos en la implementaci n y concluiremos con los motivos por los cuales pensamos que los objetivos del proyecto han sido cubiertos B 12 2 Objetivos propuestos y cumplidos E
208. to less than 4 16 elements then a single central correlation peak can be produced with reduced correlation either side Fig 3 22 shows the sequence ex tended by 8 elements at each end in this case the conjugate sequence can be shifted by up to 8 elements in either direction 3 9 3 Coarse Frequency Correction These properties are used to provide AFC in the DAB receiver A large number of such 32 element extended CAZAC sequences are transmitted simultaneously in the Phase Reference symbol block Each element value is conveyed by the phase modulation one carrier and each sequence is contained in the modulation of 32 consecutive carriers working from the low frequency end of the ensemble for example 96 3 9 Synchronization channel Saja 1111 j 1 j1 11 11 pla pada 4 14 4 5 4911 11 11 x j 1j1 11 11 j 1 11111 d 4 20 magnitude of correlation 10 with added noise Fe er ere tte tia 8 7 6 5 4 3 2 1 0 12345678 shift Figure 3 22 Correlation with an extended sequence In the receiver the sequences are reproduced in the array of complex numbers output by the FFT and the conjugate sequence is held in ROM so the correlation can be performed as previously described If the time domain signal input to the FFT has no frequency error then the reproduced sequence will be aligned in some particular way with the FFT outputs but if a large frequency error exists i e one or more times
209. to the frequency constraints imposed by the FPGA Board Whereas one CoolFlux BSP is able to run at 290 MHz in a real implementation our FPGA runs at a maximum speed of about 50 MHz The CoolFlux BSP has a CPI Cycles per Instruction of 1 Therefore each BSP mapped on the FPGA might not process more than 50 MIPS The total amount of MIPS required for demodulating a DAB signal after several op timizations turned out to be around 80 future improvements are expected to reduce this number It might be expected that with two cores it would be possible to demodulate the signal in real time However the fact of having more than one core increases the number of MIPS due to the overhead of transferring the data amongst each other Thus the decision was to split up the demodulation chain into three DSPs From now on we shall refer them as BSP BSP BSP It might not seem obvious that the separation of the functionality into 3 BSPs was not just a matter of dividing the code into serial connected modules which pass the data from one to another Memory requirements is another important issue to take care of In a system on chip SoC the memory occupies essentially all the area of the chip in comparison with the processing elements Thus the division was also made attending to the memory size available for set of processing modules l The final division of the processing modules is shown in the next picture 129 4 1 An overview to CoolFlux BSP
210. top path is stored as the best path in the transition buffer The transition buffer stores the best incoming path for each state For a radix 2 trellis only 1 bit is needed to indicate the chosen path A value of 0 indicates that the top incoming path of the given state is chosen as the best path whereas a value of 1 indicates that the bottom path is chosen The transition bits for our example is illustrated in Figure 1 24 in the third trellis Traceback begins after completing the metric update of the last symbol in the frame For frame by frame Viterbi decoding all that is needed by the traceback algorithm are the state transitions The starting state is state 00 For sliding window decoding the starting state is the state with the largest state metric In our example the starting state for both types of decoding is 00 since it has the largest state metric 158 Looking at the state transition for state 00 we see that the bottom path is optimal state transition 1 implies bottom path Two things now happen First the transition bit 1 is sent to the decoder output Second the originating state state 01 of this optimal path is determined This process is repeated until time 2 which corresponds to the first input bit transmitted The decoded output is shown in Figure 1 24 in the last row Notice that the order with which the decoded output is generated is reversed i e decoded output 10101101 whereas the sample input pattern is 1011
211. tor 4 1 An overview to CoolFlux BSP The implementation of our application was done on the CoolFlux BSP Baseband Signal Processor version 1 1 0 an embedded ultra low power C programmable Baseband Signal Processor core which targets low power communications baseband processing The core is based on the similarly named CoolFlux DSP which was designed for use in low power audio applications and was introduced in 2004 This core relative to the older one version has been enhanced to increase its performance in baseband processing while retaining a small footprint and low power The CoolFlux BSP core will run at 290 MHz power consumption for the core only is about 20 mW at 1 2 volts and 290 MHz in a 65 nm process The core is designed to be used as a co processor or in standalone mode and can also be used as part of a multi core system The CoolFlux BSP s 24 bit width is a holdover from the CoolFlux DSP audio oriented processors often use 24 bit data paths as a compromise as 24 bit data eases implementation of high fidelity audio algorithms relative to a 16 bit machine and reduces area and power 116 4 1 An overview to CoolFlux BSP relative to a 32 bit design As described below the CoolFlux BSP can split its 24 bit data path into two 12 bit paths to increase computational throughput The CoolFlux BSP issues a single 32 bit instruction word per cycle and enables an orthogonal instruction set and efficient C compiler Unlike many D
212. ts wide The core can perform for example 12 bit complex multiplication i e four real multiplications and two additions as shown below in two cycles with single cycle throughput A Bi Ai Br Ar Br Ai Bi The core also explicitly supports complex addition and subtraction and can execute 24 bit complex calculations with lower throughput The BSP supports a range of specialized instructions for SIMD arithmetic complex arithmetic FFTs Viterbi processing and the CORDIC algorithm The new SIMD and complex math capabilities enable the core to calculate two taps per cycle for a 12 bit complex FIR filter for example or execute a 12 bit with 28 bit intermediate results radix 4 256 point complex FFT in 2480 cycles By way of comparison the CoolFlux DSP core requires 8930 cycles for a 24 bit with 56 bit intermediate results radix 2 FFT Overall the SIMD and complex modes provide a significant speedup across a range of algorithms but because they are 12 bit operations the speedup comes at the cost of precision and dynamic range Like with the CoolFlux DSP CoolFlux BSP has to be programmed in C It has added C language extensions to support complex data types and SIMD data types along with intrinsics for complex and SIMD functions The BSP has some unusual features particularly its complex 12 bit computational capabilities Perhaps a key question is whether 12 bits is enough for many baseband opera tions Based on
213. ubchannel 138 4 2 Description of the BSP instances 4 2 2 BSP FFT Deinterleaving Demapping Once the signal is synchronized the second BSP just have to start demodulating it when required that is when a certain value is received as entry in our case 0000FFFF FFFF0000 This entry is the one written in NULL COPY state Since this value can never exist in digital signal the recognition of this pattern will indicate the beginning of the new frame and can not be mistaken for a radio signal This BSP is a module consisting of a chain of algorithms as explained in the figure 4 15 5 Freq Differential ird FROM BST pe EET Deinterleaving demodulator Quantization 10 8522 Figure 4 15 BSP modules The FFT requires all the input data ready by the time it starts running so the 2048 words have to be written in the input buffer before the activating interruption is received Note that the FFT is computed for the TFPR symbol again The last BSP needs to get all the ensemble but the Reference Symbol To make sure the others are sent in the appropriate order it is necessary to write into the shared buffer between the second and the third BSP which symbol has been processed After performing the quantization of the input stream we obtain 1536 words out of 2048 so we can use the remaining space as extra space for any other task recall that in a SoC any available block of memory is quite welcome even at the cost of complicating the
214. ucture referred to as FicInfo Recall from last chapter that the FIC gives us information about the subchannel configuration and services This information includes the subchannel sizes start and end addresses and so on Hereinafter the user might select one subchannel and the demodulation of the MSC begins We only perform the demodulation of the OFDM symbols that form the subchannel ignoring the rest of symbols In other words the symbols before and after the selected subchannel are skipped The demodulation of the MSC is slightly different from that of the FIC since now we must carry out a time deinterleaving of the bits As pointed out in the last chapter this module is critical in terms of memory Once the bits are time reordered we perform the depuncturing viterbi and energy dispersal as we did with the FIC Finally the output of the energy dispersal serves as input to the MP2 decoder which in turn produces the raw audio stream 140 Chapter 5 Conclusions The market of radio receptors has been up until now dominated either by static devices of considerable dimensions or as embedded devices in vehicles Mobile receptors might be the most complex in terms of design due to the limitations both in size and power consumption The lack of low cost mobile demodulators was one of the biggest motivations for scheduling this project in the company Advances in the field of electronics concretely in integrated circuits has had a huge imp
215. uence the length of the decoder The coding used in the DAB system is based on an un punctured code for which the free distance is 10 so this indicates a capacity for always correcting up to 4 errors in the decoder trellis simultaneously The combination of a convolutional encoder and a Viterbi decoder works well in the presence of a continuous stream of randomly placed errors which may be caused by noise or inter symbol interference For the non coded case the BER of the recovered bit stream would increase with reducing S N of the input signal as illustrated in Fig 3 1316 88 3 8 Energy Dispersal Unscrambling BER 1 107 2 19 uncoded rate 0 5 theoretical 3 coded 10 measured 10 5 i coding gain 10 minus implementation margin 10 3 4 5 6 7 8 9 10 11 12 13 14 S N dB Figure 3 13 Relationship between BER and S N ratio 3 8 Energy Dispersal Unscrambling In order to ensure appropriate energy dispersal in the transmitted signal the individual inputs of the energy dispersal scramblers shall be scrambled by a modulo 2 addition with a pseudo random binary sequence PRBS prior to convolutional encoding The PRBS shall be defined as the output of the feedback shift register of figure 3 14 It shall use a polynomial of degree 9 defined by P X X X 41 TAE HHHH PRBS Figure 3 14 PBRS generation The initialization word shall be applied in such a way that the first bit of the PRBS is
216. ues offer the greatest signal to noise ratio and therefore the greatest accuracy Some straightforward arithmetic involving these three values provides a number representing the magnitude and sign of the frequency error 8l 3 9 5 Fine time synchronization Having achieved fine frequency control the Phase Reference symbol block can then be used to achieve fine time synchronization of the FFT processing window in the receiver In order to make the greatest constructive use of multipath signals it is necessary to measure the impulse response of the transmission channel and to time the beginning of the processing 100 3 9 Synchronization channel window on the basis of a calculation which weights the relative importance of different multipath signals The impulse response can be measured as follows The array of complex numbers output by the FFT partly illustrated in Fig 3 23 represents the amplitudes and phases of the carriers transmitted in the Phase Reference symbol block multiplied by the complex amplitude and phase frequency response of the channel albeit sampled at the carrier frequencies The time impulse response is related to the channel frequency response by the inverse Fourier transform so a sampled version of the impulse response is related to this sampled frequency response by the inverse DFT and this can be performed using an inverse FFT quite separate from the main FFT used for OFDM decomposition The channel
217. und the fifty percent of our working time Whereas one CoolFlux BSP is able to run at 290 MHz in a real implementation our FPGA runs at a maximum speed of about 50 MHz Therefore the implementation of the DAB was done with several CoolFlux BSP cores due to the frequency constraints imposed by the FPGA Board Another important aspect to highlight is the necessity to reproduce a clean signal free of errors The algorithms written for this task such as the Viterbi convolutional decoder synchronization correction and CRC codes were another big difficulty For the sake of energy efficiency the processor does not include a floating point unit As programmers we were not used to write algorithms in fixed point representation Fixed point programming is much harder than the float point counterpart since now the program mer is responsible of manually scaling the intermediate variables in order to not saturate the ongoing signal In addition it is important to mention that we were provided with generic algorithms already implemented by the company such as the the MP2 decoder For sure this was an advantage in order to achieve our goals faster However the main difficulty in this case was to work around with code that was not written by us 143 Appendices 144 Appendix A The Fourier Transform A 1 Motivation In the field of functional analysis in mathematics it has been fundamental to approximate a complex function by means of simple
218. ut a real input b imag r input a imag input_b_ real output real input a real xinput_b_ real input a imag input b imag It is also possible to get the following different intermediate results output imag r input a real input_b_imag output real r input a real input_b_ real output imag input a imag input b real output real input a imag xinput b mag When doing an SIMD multiplication the inputs are considered to contain two subparts of 12 bits each A SIMD multiplication does the following actions output SIM DO input a SIMDO0xinput b SIM DO output SIMD1 input a SIMDl xinput b SIMDI The XMUL unit is preceded by a pre adder subtraction unit This unit can be used in 3 modes pass add and subtract For simple multiplications it is used in pass mode where it just transfers one of its inputs to the output without doing any operation this output is then multiplied in the XMUL unit to another XY register In case it is used in add or subtract mode it adds subtracts two XY registers and this output is then multiplied 53 2 3 The Data Path with a third XY register This pre adder subtraction unit is not used in complex or SIMD mode In all cases the result of a multiplication is a full 48 bit number Before being used in the ALUs it is converted to a 56 bit number This 56 bit value is stored in a temporary register and in the next processor cycle used as an operand for accumulation through one of the ALU units XALU
219. ut by the FFT depends on the frequencies of the received signal and the RF local oscillator in the receiver The objective is to locate the pattern precisely at a particular position so the data conveyed by the carriers during the following frame can be identified uniquely The same pattern is stored in the receiver so the frequency error can be determined by measuring the misalignment between the stored and received patterns The TFPR symbol also facilitates measurement of the impulse response of the transmis sion channel and fine time synchronization can be achieved using this this will be discussed later The TFPR symbol block has the normal capacity of 1536 bit pairs in Mode 1 and the whole of this capacity is used for synchronization functions The data are self contained within the symbol block and can be decoded without reference to an earlier symbol The chosen fixed pattern has properties which assist its recognition in the presence of noise and interference it is based on a CAZAC Constant Amplitude Zero Autocorrelation sequence A CAZAC sequence is a sequence of complex numbers which may be called elements which has the following characteristics e Each element has the same magnitude for example its value can be 1 j 1 or j This accounts for Constant Amplitude 93 3 9 Synchronization channel e If the sequence is multiplied element by element by its complex conjugate the sum of the products is a large
220. vice components The organization of the sub channels and service components is called the multiplex configuration 2 Fast Information Channel FIC The FIC is a non time interleaved data channel with fixed equal error correction FIC carries information about the organization of the MSC subchannels such as informa tion on the multiplex structure and when necessary its reconfiguration Optionally FIC may include service information conditional access management information and data service 3 Synchronization channel The Synchronization Channel provides a phase reference and is used internally for de modulator functions such as transmission frame synchronization automatic frequency control transmitter identification and channel state estimation Each channel sup plies data from different sources and these data are provided to form a transmission frame The Synchronization channel the Fast Information Channel and the Main Service Chan 103 3 10 Transport mechanisms nel form a transmission frame see figure below 3 27 The MSC occupies the major part of the transmission frame Transmission frame lt gt non time interleaved time interleaved lt gt lt gt Fast Information Synchronization Channel FIC Channel Main Service Channel MSC FIB CIF C cr FIB Fast Information Blocks Common Interleaved Frame s FIBs CIFs Figure 3
221. voids 180 degrees phase changes In practice this requires only a little further manipulation of the complex numbers In absolute terms there are eight possible phase states four of which are available in any one symbol and the phase reference from the previous symbol rotates by multiples of 45 degrees from symbol to symbol These features are illustrated in Fig 3 4 where the phase of one carrier is shown during three consecutive symbols the actual phases shown are examples of many possible combinations 77 3 4 Frecuency de interleaving 1 1 0 1 1 1 e 1 0 0 0 0 1 1 0 0 0 symbol n 1 symbol n symbol n 1 Key possible phase bit pair 1 0 bit pair 1 1 previous phase present phase Figure 3 4 7 4 offset QPSK Recall that the input signal is DQPSK modulated Therefore the coded data is the difference of phases between the current OFDM symbol 1536 carriers and the previous one both of which are held in two buffers in XMEM Let z i z i be the i th sample in the input buffer and the previous buffer respectively It can be shown that the phase of complex number r where is the difference of phases between z i and i 3 4 Frecuency de interleaving Apart from the Phase Reference symbol block 75 symbol blocks of data or 230400 bits make up each transmission frame In principle these can be formed into a serial bit stream to be applied to the 1536 carriers by D QPSK and
222. x 14 140 O e ee e e O 6 e State transitions 0 0 1 0 1 1 0 1 0 1 00 e S 0 S 0 Y x 5 Y N 1 1 0 lox 0 1 01 e Poe 1 10 e 0 0 1 11 v e Decoder output 1 0 1 1 0 1 0 1 Figure 1 24 Trellis diagram states 42 1 9 Errors Detection and Correction longer than 10 k 1 and is application dependent In this example 20 encoded symbols are received before being decoded In some applications Viterbi decoding is operated on a frame of received symbols and is independent of the neighboring frames Is is also possible to perform Viterbi decoding on a sliding window in which a block of decoded bits at the beginning of the window is error free and the windows will have to overlap With frame by frame decoding a number of trailing zeros equalling to the number of states is added This forces the last state of the frame to be zero providing a starting point for traceback With sliding window decoding the starting point for traceback is the state with the optimal accumulated metric This starting state may be erroneous However the traceback path will converge to the correct states before reaching the block of error free bits Viterbi decoding can be broken down into two major operations metric update and traceback In metric update two things are done in every symbol interval 1 symbol 1 input bit 2 encoded bits the accumulated state metric is calculated for each state and the optimal incoming path assoc
223. y inefficient use of the available bit rate as there still remains a probability that some of the same bits could be in error in both of the received versions Many more complicated and ingenious methods have been developed for such Forward Error Correction FEC forward implies that action is taken at the transmitter which make more efficient use of a given amount of redundancy Powerful FEC by whatever method requires the transmission of substantial amounts of redundant data which increases the demand for radio frequency spectrum However the Eureka DAB system achieves greatly improved ruggedness without sacrificing efficiency in its use of radio spectrum by applying in addition what could be called advanced digital techniques To explain how the DAB system works we should first look more closely at the effects of multipath propagation in both the time and frequency domains Incidentally the word coding is used widely in the context of error correction and digital communications generally Sometimes the intended meaning is the whole principle of encoding and decoding data and sometimes it is used instead of encoding The former meaning will be applied here and in order to avoid ambiguity the terms encoding and decoding will be used where they are meant The time and frequency domains provide different viewpoints for the same effects a principle well known to anyone who has used an oscilloscope and a sp
224. y to design a radio receptor was to design a very specific circuit with analog electronic components like operational amplifiers resistances capacitances inductances and so on Everything was done in hardware Due to advances in both computer and semiconductor technology today we can replace that old fashioned complicated circuitry by an embedded computer which is much cheaper and more reliable The aim of these computers is to achieve the required performance while maintaining costs at a minimum price Often as in the case of this project the energy consumption and heat dissipation must be taken into account very carefully since the final demodulator will work with batteries and typically no fan is available on a portable device In the context of telecommunications these embedded processor are used to operate on digital signals hence the acronym of DSP Digital Signal Processors Typically algorithms for signal processing share a lot of features among them and a specific algorithm will be used massively during the demodulation process For example for sure any algorithm for audio or video processing will perform the Fast Fourier Transform FFT on the input sig nal Another common algorithm is the Viterbi decoder for the detection and correction of 47 1 10 Digital Signals Processors DSP errors Another typical requirement for a DSP is that it cannot spend more than a certain amount of time to manipulate the input signal that
225. y using the two buses at the same time In each move the source value is converted from its data width to a 24 bit value This 24 bit value is transferred to the destination register via the move buses where it is converted to the destination data width The way the conversions are done depends whether the move is a scalar or packed move The 24 bit buses are wide enough to make all 8 bit 15 bit 16 bit and 24 bit register and memory moves straightforward but transferring from and to a 56 bit accumulator requires special attention l l 2 4 2 Moving to an Accumulator The accumulator can be the destination of a move operation in 2 ways The full accumulator is the destination a0 al b0 b1 An accumulator sub register is the destination ao0 aol ah0 ah1 al0 all bo0 bol bhO bh1 bl0 bl1 In case a sub register is the target its data width is either 8 or 24 So the conversion from 24 bits to the target data width is straightforward the 8 LSBs are used or no conversion is needed The full accumulator can be the destination in 4 ways e a 24 bit scalar move e a 48 bit scalar move a 24 bit packed move e a 48 bit packed move In a 24 bit scalar move assignment the 24 bit value is converted to 56 bits by sign extending the upper 8 bits and padding with 24 zeros at the LSB side In a 48 bit scalar move 1 assignment the Xbus and Ybus move buses are combined to move a 48 bit value This 48 bit value is conve

DSP-Based Implementation of a Digital Radio Demodulator on the

Contents

Download Pdf Manuals

Related Search

Related Contents