Adaptive communications techniques for the underwater acoustic channel

Transcription

1 Adaptive communications techniques for the underwater acoustic channel James A. Ritcey Department of Electrical Engineering, Box University of Washington, Seattle, WA Tel: (206) , Fax: (206)

2 Agenda Project overview Modulation Alternatives Bit-interleaved Coded Modulation (BICM)

3 Contract Overview 3 year project Start Date April 2007 Summer salary Ritcey One PDRA II Currently Mr. Chantri Polprasert (PhD Candidate in Electrical Engineering)

4 4 Tasks in SOW Task 1 Underwater Acoustic Channel Modeling. Task 2 Channel Adaptation and Multichannel Combining. Task 3 Coded Modulation. Task 4 Algorithm validation.

5 Schedule and Effort: Start in April 2007 Year Task 1 Task 2 Task 3 Task % 25% 0% 0% 2 25% 25% 25% 25% 3 0% 25% 25% 25%

6 Dissemination Conference and journal publications Student project and theses ONR reports and presentations Software and data analysis

7 Project Plans Channel Modeling Modulation and Adaptation Coded Modulation using BICM Algorithm Validation

8 Work to Date starting April 2007 Develop simulation programs of different transmission systems over multipath fading channels in Matlab: Uncoded OFDM ZP-OFDM CP-OFDM Uncoded SC-FDE ZP-SC-FDE CP-SC-FDE Linear equalizer (MMSE) Non-linear equalizer (DFE) Coding: BICM and BICM-ID BICM-ID over SC-FDE and OFDM

9 Modulation Options Modulation Types Cyclic Prefix Orthogonal Frequency-Division Multiplexing (CP-OFDM) Zero Padding Orthogonal Frequency-Division Multiplexing (ZP-OFDM) Cyclic Prefix Single-Carrier Frequency Domain Equalization (CP-SC-FDE) Zero Padding Single-Carrier Frequency Domain Equalization (ZP-SC-FDE)

10 System Parameters - BPSK R Bit rate (bits/s) N No. of subcarriers, blocksize P No. of CP or ZP samples T = (N+P)/R OFDM Symbol duration (s) Δf = R/N Subcarrier spacing (Hz) B = NΔf Total Bandwidth (Hz)

11 OFDM Included in DAB/DVB standard in Europe and the DSL modem in the US Used in fixed broadband wireless systems Combats multi-path fading by transmitting orthogonal symbols in parallel using narrow-band sub-channels Two variants are considered based on the sequence inserted at the transmitter to avoid Inter-block Interference (IBI): CP-OFDM ZP-OFDM

12 CP & ZP OFDM CP: A copy of the last part of the symbol prepended to the transmitted symbol ZP: A sequence of zero symbols appended after the transmitted symbol

13 OFDM Variations CP-OFDM CP is inserted at the beginning of each transmission block No equalizer required, but Susceptible to fading on each subcarrier Peak-to-Average Power Ratio (PAPR) ZP-OFDM ZP is appended after the transmitted symbols Equalizer needed at the receiver Overlap-Add avoids deep fades Increased receiver complexity over CP-OFDM

14 CP OFDM CP OFDM Transmission block diagram

15 CP OFDM { } = 0, L 1, s [ s, s ], N: a number of DFT points n n N n 1 The n OFDM symbol: S = s e, k = 0KN 1 N 1 th j2 πik / N kn, in, N i= 0 { } th The n block: S = [ S, LS, S, LS ], P: length of CP { h } = h0, Lh 1, ' n N P, n N 1, n 0, n N 1, n [, ]: Quasi-static channel impulse response n n L n Receiver: assume that P L 1 L 1 y = hs + n, k= 0KN+ P 1 kn, l k ln, k l= 0 zkn, = ykn,, k = PKN + P 1 ({ }) ( ) R = H s + V, m= 0KN 1, H = DFT h, V = DFT {} n mn, mn, mn, m mn, N n k N s% = W R. Assuming perfect CSI, W = H 1 mn, mn, mn, mn, mn, ({ }): N-size DFT of { } DTF r r N n n

16 ZP OFDM

17 ZP OFDM { } = 0, L 1, s [ s, s ], N: a number of DFT points n n N n 1 The n OFDM symbol: S = s e, k = 0KN 1 N 1 th j2 πik / N kn, in, N i= 0 { } th The n block: S = [ S, LS,0, L,0 ], P: length of ZP { } = 0, L 1, ' n 0, n N 1, n 1 P hn [ h n, hl n]: Quasi-static channel impulse response Receiver: assume that P L L 1 y = hs + n, k= 0K N+ P 1 kn, l k ln, k l= 0 z kn, ykn, + yk+ Nn,, k = 0KP 1 = ykn,, k = PK N 1 s% = W R. Assuming perfect CSI, W = H 1 mn, mn, mn, mn, mn, ({ }) ( ) R = H s + V, m= 0KN 1, H = DFT h, V = DFT {} n mn, mn, mn, k mn, N n k N ({ }): N-size DFT of { } DTF r r N n n

18 SC-FDE Single Carrier alternative to OFDM 1,2 Similar performance to OFDM with same computational complexity 2 variants ZP-SC-FDE CP-SC-FDE Frequency Domain Equalizer Linear: Zero-forcing (ZF), Minimum Mean Square Error(MMSE) Non-Linear: Decision feedback (DFE) Frequency domain feedforward filter Frequency or Time domain feedback filter 2,3 1-IEEE Std TM Falconer et al., Falconer 2002

19 SC- Frequency Domain Equalizers MMSE Given information block size M, ZP-SC-FDE outperforms CP- SC-FDE 4 DFE ZP-SC-FDE eliminates cyclic intersymbol interference and outperforms CP-SC-FDE 5 4-Ohno Chung and Hwang 2005

20 OFDM & SC-FDE Comparison OFDM Modulation Pros Cons Combats ISI using parallel narrowband transmission. Flat fading; channel coding is required High PAPR SC-FDE Yields multi-path diversity gain for uncoded transmission Low PAPR Resistance to frequency offset Susceptible to frequency offset (ICI) High computational complexity when calculating DFE coefficients PAPR: Peak-to-average power ratio

21 OFDM & SC-FDE OFDM SC-FDE Increased computational complexity at the receiver

22 CP SC-FDE

23 CP SC-FDE { s } = s0, Ls 1, [, ], N is the information blocksize n n N n { } th The n block: s = [ s, Ls, s, Ls ], P: length of CP { h } = h0, Lh 1, ' n N P, n N 1, n 0, n N 1, n [, ]: Quasi-static channel impulse response n n L n Receiver: assume that L 1 mn, l m ln, k l= 0 P L 1 y = hs + n, m= 0KN + P 1 zmn, = ymn,, m= PKN + P 1 {{ }} { } { } R = H S + V ; H = DFT h, V = DFT { n }, S = DFT { s } kn, kn, kn, k kn, N n k N k kn, N n π s% = W R exp( j mk), m= 0KN 1 N N mn, kn, kn, N k = 0 Wkn,, k = 0K N 1: Feedforward filter coefficients {{ }}: N-size DFT of { } DTF r r N n n

24 ZP SC-FDE

25 ZP SC-FDE { s } = s0, Ls 1, [, ], N is the information blocksize n n N n { } th The n block: s = [ s, Ls 0, L0 ], P: length of ZP { h } = h0, Lh 1, ' n 0, n N 1, n 1, n P, n [, ]: Quasi-static channel impulse response n n L n Receiver: assume that L 1 mn, l m ln, k l= 0 P L 1 y = hs + n, m= 0KN + P 1 {{ }} { } { } R = H S + V ; H = DFT h, V = DFT { n }, S = DFT { s } kn, kn, kn, k kn, N n k N k kn, N n π c = W R exp( j mk), m= 0KN + P 1 N N mn, kn, kn, N k = 0 Wkn,, k = 0K N 1: Feedforward filter coefficients s% mn, = cmn,, m= 0KN 1 ({ }): N-size DFT of { } DTF r r N n n

26 Recent Tasks Software development in MATLAB Performance Comparison uncoded Known channels

27 Performance comparison between ZP-SC SC-FDE and CP-SC SC-FDE using MMSE over 8-tap 8 Rayleigh fading* 10-2 ZP 32 ZP 64 ZP 128 CP 32 CP 64 CP 128 BER EbNodB ZP-SC outperforms CP-SC at increased complexity ZP-SC-FDE improves as the blocksize decreases CP-SC-FDE improves as the blocksize increases *Ohno 2006

28 Uncoded OFDM & ZP-SC SC-FDE Comparison BER OFDM LE-MMSE DFE 10 taps DFE 20 taps DFE 30 taps EbNo(dB) Performance comparison between OFDM & SC-DFE over 30-tap exponential decay with 1024-point FFT using QPSK modulation

29 Impact of imperfect feedback taps in DFE Correct symbols FB Decision-directed FB 10-2 BER EbNo(dB) Performance comparison between SC-FDE-DFE over 4- taps fixed channel with 256-point FFT using QPSK modulation

30 Coded Modulation Bit Interleaved Coded Modulation (BICM) Block Diagram Labeling Issues Analytical Performance Evaluation Numerical Results with Iterative Decoding

31 BICM and BICM-ID ID Review Bit-interleaved coded modulation (BICM) Large diversity order through bit-wise interleaving First introduced by Zevahi, 1992 Thorough ana BICM with iterative decoding (BICM-ID) Constellation labeling design 8-PSK: Li and Ritcey, QAM: Chindapol and Ritcey, 1999 Imperfect CSI over Rayleigh fading: Huang and Ritcey 2003 Space Time Block Codes: Huang and Ritcey 2005

32 BICM-ID ID Block Diagram Encoder S/P Modulator Channel Decoder S/P Demodulator : bit-wise interleaver -1 : bit-wise deinterleaver Labeling map μ 2 m -ary constellation χ Log-likelihood bit metric λ

33 Error-Free Feedback Bound (EFF bound) Motivation Fast numerical evaluation Accurate BER floor calculation BICM union bound d min : the minimum Hamming distance of the convolutional encoder W I (d): the total input weight of error events at d f(d,μ,χ): the pair wise error probability (PEP) k c /n c : the code rate

34 EFF Bound PEP: the Laplace transform of the p.d.f. of the metric difference The metric difference Conditional Gaussian random variable with mean and variance :the constellation point having the same binary bits as those of x except the ith bit position : the subset of χ whose label has binary value b at the ith bit position

35 Simulation parameters Convolutional encoder Rate: ½, 1/3, 2/3, Memory: 2, 3 Modulation: 8-PSK, 16-QAM Information blocksize: 5000 Simulate 10 7 information bits Mapping 8-PSK: Gray, Set partitioning (SP), Semi-SP (SSP) 16-QAM: Gray, Msp Channel: AWGN, Rayleigh Perfect CSI at the receiver No. of iteration: 8

36 Tightness of the EFF bound Pass Pass 8 BER Simulation EFF bound EbNo(dB) Performance of 16QAM BICM-ID with MSP labeling over Rayleigh fading. A four-state rate 1/2 convolutional encoder.

37 Modified Set Partitioning (MSP) Labeling scheme for 16QAM a) Decision region of each bit before iterative decoding* b) Decision region of each bit after iterative decoding* *Chindapol and Ritcey, 1999

38 Impact of labeling -Rayleigh Pass Pass 1 BER 10-3 Pass Pass1 Gray Pass8 Gray Pass 8 Pass1 MSP Pass8 MSP EbNo(dB) Performance of 16QAM BICM-ID with Gray and MSP labeling over Rayleigh using a rate-1/2 four-state convolutional encoder

39 Impact of labeling - AWGN Pass 1 BER 10-2 Pass 1 Pass Pass Gray Mapping MSP Mapping EbNo(dB) Performance of 16QAM BICM-ID with Gray and MSP labeling over AWGN using a rate-1/2 eight-state convolutional encoder

40 8PSK and 16QAM - Rayleigh Pass 1 Pass BER Pass Pass PSK 16QAM Eff bound EbNo(dB) Performance comparison of 8-PSK and 16-QAM BICM-ID over Rayleigh fading channels.

41 8PSK and 16QAM - AWGN Pass 1 Pass BER 10-3 Pass Pass 6 8PSK 16QAM EbNo(dB) Performance comparison of 8-PSK and 16-QAM BICM-ID over AWGN channels.

42 Impact of code memory Pass Pass 8 4-state 8-state Eff bound Impact of code memory on the performance of 16-QAM BICM-ID with MSP labeling and a rate ½ convolutional code over Rayleigh fading channels

43 Impact of code rate 10 0 Pass Pass Pass 8 BER Pass 8 Rate2/3 Rate1/3 Eff bound EbNo(dB) Impact of code rate on the performance of 8-PSK BICM-ID with SSP labeling and a rate ½ convolutional code over Rayleigh fading channel

44 Impact of information block size Eff bound 10-3 BER EbNo(dB) Impact of information blocksize on the performance of 16-QAM BICM-ID with MSP labeling and a rate ½ convolutional code over Rayleigh fading channel

45 Application to UWA Coherent Signaling Channel Estimation Iterative channel estimation decoding Integration with OFDM and SC-FDE Application to UWA realistic channels

46 Upcoming Work BICM/BICM-ID over OFDM/SC-FDE Perfect CSI Use iterative decoding to combat multi-path fading Impact of labeling, code rate over the BER performance Its performance over different types of equalizer e.g. DFE, MMSE Adaptive modulation and equalization Imperfect CSI Use iterative decoding to combat imperfect estimate of the fading Array Combining Joint estimation and decoding