Video Transmission over Wireless Channel

Transcription

1 Bologna, Video Transmission over Wireless Channel Raffaele Soloperto PhD DEIS, University of Bologna Tutor: O.Andrisano Co-Tutors: G.Pasolini and G.Liva (DLR, DE) DEIS, Università di Bologna IEIIT-BO CNR

2 PhD Outline Focus on DVB-T and T2 standards: Signal Processing Linear and Non Linear Predistorsion MultiRate MultiStage Filters Channel Coding LDPC, G-LDPC and DG-LDPC Codes Measurements and Tele-Measurements Instrumentation and programmable circuits

3 PhD Outline Focus on DVB-T and T2 standards: Signal Processing Linear and Non Linear Predistorsion MultiRate MultiStage Filters Channel Coding LDPC, G-LDPC and DG-LDPC Codes Measurements and Tele-Measurements Instrumentation and programmable circuits VIDEO OVER WIRELESS SIGNAL PROCESSING II CHANNEL CODING MEAS. AND TELE MEAS. I, II, III II, III

4 PhD Outline Signal Processing: MultiStage-MultiRate Filters Design of MultiStage and MultiRate Filters with minimum group delay

5 Scientific literature: < proj fixed I) Ronald E.Crochiere, Lawrence R.Rabiner multistage multirate filters with minimum number of taps II) Raffaele Soloperto, Gianni Pasolini multistage multirate filters with minimum group delay

6 MultiStage MultiRate Filter cos(2π f 0 t) cos(2π f 0 t) X LPF X ADC + DAC DEC LPF INT M L X LPF X -sen(2π f 0 t) -sen(2π f 0 t) LPF Down sampl. M MultiRate Decimation (min. number of taps) LPF 1 Down sampl. M 1 LPF I Down sampl. M I

7 Measurements (1/2) B = 8 MHz Fc = 36 MHz

8 Measurements (2/2) B = 1 MHz Fc = 36 MHz

9 Iterative decoding of DG-LDPC codes Visiting PhD Student at DLR, Munich DE (Nov 2009 July 2010) DEIS, Università di Bologna IEIIT-BO CNR

10 Outline Introduction LDPC codes Generalized LDPC codes Doubly Generalized LDPC codes Graph representation Decoding algorithm Efficient encoding: Quasi-Cyclic DG-LDPC codes Conclusions

11 Outline Introduction LDPC codes Generalized LDPC codes Doubly Generalized LDPC codes Graph representation Decoding algorithm Efficient encoding: Quasi-Cyclic DG-LDPC codes Conclusions

12 Why DG-LDPC codes? Long LDPC/Turbo codes approach Shannon limit In the moderate/short block length regime, however, they show a gap from theoretical bounds (~1dB): Generalized LDPC codes (Leintmeier 98, Chiani 06, Liva 06) Non-binary LDPC codes (Mackay, 98) Doubly Generalized LDPC codes (Chiani/Paolini/Fossorier 06)

13 Low-Density Parity-Check Codes LDPC codes: (sparse) bipartite graph describing the parity-check equations. = H = = = c c c c c c c c c c c c Parity-check equations: Parity-check matrix of a Hamming (7,4): Bipartite graph: Variable nodes Check nodes

14 LDPC codes: decoding on the bipartite graph ML decoding is unfeasible even for rather short block lengths Belief propagation (BP): iterative, message-passing decoding algorithm. Sparse graph: the correlation among messages is reduced. BP ML. Complexity: graph, nodes, iteration number, etc bit reliabilites Channel observations (noisy symbol samples) y 1 y 2 y 3 y n...

15 Generalized LDPC Single Parity-Check Low-density parity-check codes (large number of simple nodes)

16 Generalized LDPC H H H H Block code (ex. Hamming) Block turbo codes (BTC) and generalized low-density codes (GLDC) (trade-off between node count and complexity)

17 Generalized LDPC Block code (ex. Hamming) Block turbo codes (BTC) and generalized low-density codes (GLDC) (trade-off between node count and complexity) Super Check Node can be stronger than conventional CN Error floors are lowered Unbalancing of the edcoding complexity: all the complexity at the SCN!

18 DG-LDPC codes STATE OF THE ART Generalized and Doubly Generalized LDPC codes with random component codes for the binary erasure channel, E.Paolini, M.P.C. Fossorier, M.Chiani, IEEE Transaction on Information Theory, Vol. 56, No 4, April 2010 Doubly-Generalized LDPC codes: stability bound over the BEC, E.Paolini, M.P.C. Fossorier, M.Chiani, IEEE Transaction on Information Theory, Vol. 55, No. 3, March 2009 On the Growth Rate of the Weight Distribution of Irregular Doubly-Generalized LDPC Codes, M.F. Flanagan, E. Paolini, M. Chiani, and M.P.C. Fossorier, Forty-Sixth Annual Allerton Conference, Allerton House, UIUC, Illinois, USA, September 23-26, 2008 Generalized Stability Condition for Generalized and Doubly-Generalized LDPC Codes, E.Paolini, M.P.C. Fossorier, M.Chiani, ISIT2007, Nice, France, June24 June29, 2007 Getting closer to the limits in the short/moderate length Balancing the decoder complexity at CNs and VNs

19 DG-LDPC codes Motivation of our work: Simplify decoding algorithm (new stopping criterions) Efficient encoding (design of QC-DGLDPC codes)

20 Outline Introduction LDPC codes Generalized LDPC codes Doubly Generalized LDPC codes Graph representation Decoding algorithm Efficient encoding: Quasi-Cyclic DG-LDPC codes Conclusions

21 DG-LDPC Graph Noisy codeword The Tanner graph of DG- LDPC codes can be obteined from that of LDPC codes with original VNs and CNs replaced by Super Variable Nodes and Super Check Nodes respectively. Legend:

22 DG-LDPC Graph SVNs and SCNs are subcodes with lengths equal to the degrees of their corresponding super nodes! K v1 noisy codeword bits SVNs degree SUPER VARIABLE NODE (d vx, k vx ): subcode for the SVN d vx : block length of the subcode k vx : dimension of the subcode

23 DG-LDPC Graph SVNs and SCNs are subcodes with lengths equal to the degrees of their corresponding super nodes! SCNs degree (d cy, k cy ): subcode for the SCN d cy : block length of the subcode k cy : dimension of the subcode SUPER CHECK NODE

24 DG-LDPC. Decoding algorithm Iterative decoding based on belief propagation (BP) A Priori information at the i-th iteration Extrinsic information at the i-th iteration

25 Decoding Algorithm ITERATIONS SCN SUPER CHECK NODE ELABORATION SVN SUPER VARIABLE NODE ELABORATION APP LLR TRANSMITTED BIT HARD DECISION SYNDROME CHECK

26 EXIT Chart

27 Stopping Criterions STEP STOP: evaluation of parity-check equations every N iterations. THRESHOLD STOP: evaluation of parity-check equations after a fixed number of iterations (threshold) data rate [Kbps] Example: DG-LDPC code 500 SVNs (SPC(7,6)) 500 SCNs (Hamm(7,4)) R = 0.5 Transmitted codewords = 1000 Eb/N0 = 1.8 db (CER = 10-4 ) State of the art Proposed solutions Simulation on standard PC (Pentium IV, 3.00GHz, 3.00GB RAM)

28 Iterative decoding simulator In principle, we can simulate all possible iteratively decodable codes

29 Iterative decoding simulator In principle, we can simulate all possible iteratively decodable codes Single parity-check Low-density parity-check codes (large number of simple nodes) C C Convolutional code H H H H Block code (ex. Hamming) π Convolutional turbo codes (small number of complex nodes) Block turbo codes (BTC) and generalized low-density codes (GLDC) (trade-off between node count and complexity)

30 RESULTS DEIS, Università di Bologna IEIIT-BO CNR

31

32

33 Outline Introduction LDPC codes Generalized LDPC codes Doubly Generalized LDPC codes Graph representation Decoding algorithm Efficient encoding: Quasi-Cyclic DG-LDPC codes Conclusions

34 Protographs Type A Type B Protograph: small graph with N variable nodes and M check nodes Each check/variable node in a protograph identifies a check/variable node type Type c Type d Type e Multiple, parallel edges are allowed.

35 Protographs Derived graph: obtained by q replicas on the protograph. Cyclic edge permutations: the DG-LDPC code is quasicyclic (QC). Shift-register-based encoder.

36 Protographs Proposition. A sufficient condition for having a QC DG-LDPC is that the protograph is expanded by means of circulant permutation matrices Proof (sketch): Expand rows/columns of Γ according to the rule Permute rows/columns of according to a defined algorithm (not shown here) H DGLDPC is block circulant H DGLDPC THE CODE IS QUASI-CYCLIC

37 Protographs example (1/3) 7 SCNs Hamm(15,11) Q = 30 d c1 = 15 # SVNs = 450 # SCNs = 210 d v1 = 7 # bits = 1800 R = K v1 = 4 15 SVNs Hamm(7,4)

38 Protographs example (2/3) Adjaceny Matrix Rows/columns expansion Rows/columns permuation H QC-DGLDPC

39 Protographs example (3/3) Performance of a (1800, 960) QC-DGLDPC code and DG-LDPC code in terms of CER on the AWGN channel Q = 30 # SVNs = 450 # SCNs = 210 R = There is NO degradation of performance between a QC code and unstructured code!

40 Outline Introduction LDPC codes Generalized LDPC codes Doubly Generalized LDPC codes Graph representation Decoding algorithm Efficient encoding: Quasi-Cyclic DG-LDPC codes Conclusions

41 Conclusions Signal Processing: design of MultiRate MultiStages Filters with minimum group delay Channel Coding: General Purpose decoder machine for iterative codes (LDPC, G-LDPC, DG-LDPC..) efficient encoding: Quasi Cyclic codes Mesearements and TeleMeasurements: remotization of instrumentation and programmable circuits remotization of the connections by means of a programmable matrix

42 Publication

43 DELIVERABLES DVB2006 Sistema di filtraggio, Cancellatore d eco, Studio delle prestazioni del sistema e Pianificazione di rete. Sistema di filtraggio multistadio, Precorrezione non lieare, Cancellatore d eco. Sistema di filtraggio multistadio, Precorrezione non lieare, Precorrezione lineare, Cancellatore d eco. Sistema di filtraggio multistadio, Precorrezione non lieare, Precorrezione lineare, Cancellatore d eco, Modulatore. Attività di misura per la caratterizzazione del Cancellatore d eco a finestra fissa, con tecnologia ad impulsi e LMS. Deliverable Finale DVB2006 (Polo tecnologico TV digitale) : - DVB-T Echo Canceller (design & implementation); - DVB-T Echo Canceller (measurements).

44 THANK YOU!