Estimation and detection from coded signals

1. Estimation and detection from coded signals Overview of research at UGent (37) and cooperation with NEWCOM partners Marc Moeneclaey, UGent - TELIN dept. Marc.Moeneclaey@telin.UGent.be

2. Outline • Coding + linear modulation, AWGN channel • MAP detection* carrier phase and timing known* carrier phase and timing unknown • ML synchronization* Cramer-Rao bound, modified Cramer-Rao bound* expectation-maximization (EM) algorithm • Numerical results • Related research (channel estimation, OFDM, MIMO,CDMA, multipath fading, time-varying parameters, ...) • Cooperation with NEWCOM partners

3. Coding + linear modulation, AWGN channel codedsymbols pulse h(t) infobits r(t) encoding,interleaving,mapping AWGNchannel linear modulation b a received signal q : carrier phaset : time delay h(t) : square-root Nyquist pulse b a : turbo code, LDPC code, BICM, ....

4. MAP detection, (q, t) known (1/2) MAP detection (on individual infobits) achieves minimum BER r : vector representation of r(t) Assuming (q, t) = (q0, t0) is known, APP is given by (many terms !) Summation over information sequences can be done efficiently (message-passing in factor graph of encoder+interleaver+mapper)

5. MAP detection, (q, t) known (2/2) Receiver structure zm(t0) conventionalMAP detector x r(t) h*(-t) mT+t0 exp(-jqo)

6. MAP detection, (q, t) unknown When (q, t) is unknown, APP is given by  conventional MAP detectorhard to compute, because of integration over (q, t) we look for a (suboptimum) alternative

7. Synchronization : estimate of (q, t) Strategy : use conventional MAP detector, but provide estimate (instead of correct value) of (q, t) no integration over (q, t) conventionalMAP detector x r(t) h*(-t) Synchronizer

8. ML synchronization ML estimation of (q, t) : no integrations required,but sum contains many terms that are function of (q, t) ML estimate hard to compute (2-D search)

9. Cramer-Rao lower bound (1/2) Motivationfor ML estimation : for long observation intervals, mean-square estimation error (MSEE) converges to Cramer-Rao lower bound (CRB) on MSEE When the data symbols are not a priori known to the receiver, computation of the CRB is hard, especially for coded transmission. A simpler but less tight bound is the modified CRB (MCRB), which basically assumes that the data symbols are known to the receiver

10. Cramer-Rao lower bound (2/2) QPSK BPSK (phase estimation) Large SNR : CRB  MCRB Small SNR : CRBuncoded > CRBcoded > MCRB  use code properties during estimation

11. Expectation-maximization algorithm (1/2) Direct application of ML estimation is complicated. ML estimate can be obtained iteratively by means of EM algorithm soft symbol decision : requires APPs of symbols (by-product of MAP detector)

12. Compute Expectation-maximization algorithm (2/2) conventionalMAP detector x r(t) h*(-t) soft decisions

13. Convergence behavior of EM algorithm phase estimation error at i-th iteration : negativezero-crossings : stable positivezero-crossings : unstable

14. Convergence behavior of EM algorithm phase estimation error at i-th iteration :  initialization is critical acquisition range

15. Initialization of EM algorithm (1/4) • DA initialization (pilot symbols) : - consumes power and bandwidth resources- does not exploit unknown data symbols (many pilot symbols needed) • NDA initialization (O&M for timing, V&V for phase)- gives rise to phase and timing ambiguities

16. Initialization of EM algorithm (2/4) Proposed solution - Different initializations, e.g., - Parallel EM algorithms with different initializations, each running for only one (or a few) iterations - Continue with the EM algorithm that gave the largest maximum in the maximization step after one (or a few) iterations

17. Initialization of EM algorithm (3/4) Example 1 : phase estimation, QPSK (perfect ambiguity resolution) other equil. points

18. Initialization of EM algorithm (4/4) Example 2 : phase estimation and ambiguity resolution, QPSK other equil. points

19. Numerical results : exploiting code properties partial (or no) exploitation of code properties  degradation of MSE does not exploit code properties uses infobit APPs only, assumes parity bits are uncoded uses infobit APPs and parity bit APPs

20. Numerical results : phase estimation (1/2)

21. Numerical results : phase estimation (2/2)

22. Numerical results : phase estimation + ambiguity resolution

23. Numerical results : timing estimation (1/2)

24. Numerical results : timing estimation (1/2)

25. Numerical results : timing estimation + frame synchronization

26. Related research at UGent Extension of phase and timing estimation from linear modulationtransmitted over AWGN channel to the following : • Carrier frequency estimation and channel estimation • Multiuser CDMA, OFDM, UWB • Multipath fading channel, MIMO channel • Multidimensional mapping + optimization • Estimation of time-vaying parameters by means of iterative feedback algorithm • etc.

27. Cooperation within NEWCOM (1/3) joint papers on EM-based parameter estimation • EURASIP JWCN paper accepted UGent + UCL(Vandendorpe) + UoP(Luise) • ISSSTA’04 UCL(Vandendorpe) + UGent • IEEE Trans.Comm. paper submittedUCL(Vandendorpe) + UGent • Lecture notes in Computer Science, 2004ETH(Loeliger)+UGent

28. Cooperation within NEWCOM (2/3) joint papers on estimation of time-varying parameters • EUSIPCO 2005 paper submitted ISIK(Panayirci) + UGent • Globecom 2005 paper submitted ISIK(Panayirci) + UGent

29. Cooperation within NEWCOM (3/3) UGent would like to • Continue existing cooperations • Enter new cooperations Broad topic : “estimation and detection from coded signals” Type of cooperation : • Joint publications • Joint research projects • etc.

30. Joint publications (1/2) 1) C. Herzet, H. Wymeersch, L. Vandendorpe and M. Moeneclaey , "On Maximum-Likelihood Timing Estimation", Submitted to IEEE Transactions on Communications.Joint UCL(038) - UGent (037) contribution 2) N. Noels, V. Lottici, A. Dejonghe, H. Steendam, M. Moeneclaey, M. Luise , L. Vandendorpe, "A Theoretical Framework for Soft Information Based Synchronization in Iterative (Turbo) Receivers”, accepted by EURASIP Journal on Wireless Communications and NetworkingJoint UGent(37)-UCL (38)- UoP (13) contribution 3) Ramon V., Herzet C., Vandendorpe L. and Moeneclaey M. , "EM algorithm-based estimation of amplitude, carrier phase and noise variance in multiuser turbo receivers", Proc. Eighth International Symposium on Spread Spectrum Techniques and Applications, 2004, ISSSTA04, Sydney, Australia, Aug 30 - Sep 02 2004, pp. 550 - 554Joint UCL (038)-UGent (037) contribution

31. Joint publications (2/2) 4) J. Dauwels, H. Wymeersch, H.-A. Loeliger and M. Moeneclaey "Phase Estimation and Phase Ambiguity Resolution by Message Passing", Telecommunications and Networking, Lecture Notes in Computer Science, vol. 3124, pp. 150-155, Springer-Verlag, Berlin, 2004.Joint ETH(26)- UGENT(37) contribution 5) E. Panayirci, H. A. Cirpan and M.Moeneclaey, ”A Sequential Monte Carlo Method for Blind Phase Noise Estimation and Data Detection”, submitted to EUSIPCO 2005Joint ISIK(07)-UGENT(37) contribution 6) E. Panayirci, H. A. Cirpan, M. Moeneclaey and N. Noels "Blind Data Detection in the presence of PLL phase noise by sequential Monte Carlo method”, submitted to Globecom’05 Joint ISIK(07)-UGENT(37) contribution