1. VIRGINIA POLYTECHNIC INSTITUTE & STATE UNIVERSITY MOBILE & PORTABLE RADIO RESEARCH GROUP MPRG Turbo Codes and Iterative Processing IEEE New Zealand Wireless Communications Symposium November 1998 Matthew Valenti Mobile and Portable Radio Research Group Bradley Department of Electrical and Computer Engineering Virginia Tech Blacksburg, Virginia

2. Outline n Introduction u Forward error correction (FEC) u Channel capacity n Turbo codes u Encoding u Decoding u Performance analysis u Applications n Other applications of iterative processing u Joint equalization/FEC u Joint multiuser detection/FEC

3. Error Correction Coding n Channel coding adds structured redundancy to a transmission. u The input message m is composed of K symbols. u The output code word x is composed of N symbols. u Since N > K there is redundancy in the output. u The code rate is r = K/N. n Coding can be used to: u Detect errors: ARQ u Correct errors: FEC Channel Encoder

4. Power Efficiency of Existing Standards

5. Turbo Codes n Backgound u Turbo codes were proposed by Berrou and Glavieux in the 1993 International Conference in Communications. u Performance within 0.5 dB of the channel capacity limit for BPSK was demonstrated. n Features of turbo codes u Parallel concatenated coding u Recursive convolutional encoders u Pseudo-random interleaving u Iterative decoding

6. Motivation: Performance of Turbo Codes. n Comparison: u Rate 1/2 Codes. u K=5 turbo code. u K=14 convolutional code. n Plot is from: L. Perez, “Turbo Codes”, chapter 8 of Trellis Coding by C. Schlegel. IEEE Press, 1997. Gain of almost 2 dB! Theoretical Limit!

7. Concatenated Coding n A single error correction code does not always provide enough error protection with reasonable complexity. n Solution: Concatenate two (or more) codes u This creates a much more powerful code. n Serial Concatenation (Forney, 1966) Outer Encoder Block Interleaver Inner Encoder Outer Decoder De- interleaver Inner Decoder Channel

8. Parallel Concatenated Codes n Instead of concatenating in serial, codes can also be concatenated in parallel. n The original turbo code is a parallel concatenation of two recursive systematic convolutional (RSC) codes. u systematic: one of the outputs is the input. Encoder #1 Encoder #2 Interleaver MUX Input Parity Output Systematic Output

9. Pseudo-random Interleaving n The coding dilemma: u Shannon showed that large block-length random codes achieve channel capacity. u However, codes must have structure that permits decoding with reasonable complexity. u Codes with structure don’t perform as well as random codes. u “Almost all codes are good, except those that we can think of.” n Solution: u Make the code appear random, while maintaining enough structure to permit decoding. u This is the purpose of the pseudo-random interleaver. u Turbo codes possess random-like properties. u However, since the interleaving pattern is known, decoding is possible.

10. Recursive Systematic Convolutional Encoding n An RSC encoder can be constructed from a standard convolutional encoder by feeding back one of the outputs. n An RSC encoder has an infinite impulse response. n An arbitrary input will cause a “good” (high weight) output with high probability. n Some inputs will cause “bad” (low weight) outputs. DD Constraint Length K= 3 DD

11. Why Interleaving and Recursive Encoding? n In a coded systems: u Performance is dominated by low weight code words. n A “good” code: u will produce low weight outputs with very low probability. n An RSC code: u Produces low weight outputs with fairly low probability. u However, some inputs still cause low weight outputs. n Because of the interleaver: u The probability that both encoders have inputs that cause low weight outputs is very low. u Therefore the parallel concatenation of both encoders will produce a “good” code.

12. Iterative Decoding n There is one decoder for each elementary encoder. n Each decoder estimates the a posteriori probability (APP) of each data bit. n The APP’s are used as a priori information by the other decoder. n Decoding continues for a set number of iterations. u Performance generally improves from iteration to iteration, but follows a law of diminishing returns. Decoder #1 Decoder #2 DeMUX Interleaver Deinterleaver systematic data parity data APP hard bit decisions

13. The Turbo-Principle n Turbo codes get their name because the decoder uses feedback, like a turbo engine.

14. Performance as a Function of Number of Iterations n K=5 n r=1/2 n L=65,536

15. S0S0 S3S3 S2S2 S1S1 0/00 1/11 0/01 1/10 0/01 0/00 1/11 i = 0i = 6i = 3i = 2i = 1i = 4i = 5 The log-MAP algorithm The log-MAP algorithm: Performs arithmetic in the log domain Multiplies become additions Additions use the Jacobian Logarithm:

16. Performance Factors and Tradeoffs n Complexity vs. performance u Decoding algorithm. u Number of iterations. u Encoder constraint length n Latency vs. performance u Frame size. n Spectral efficiency vs. performance u Overall code rate n Other factors u Interleaver design. u Puncture pattern. u Trellis termination.

17. Performance Bounds for Linear Block Codes n Union bound for soft-decision decoding: n For convolutional and turbo codes this becomes: n The free-distance asymptote is the first term of the sum: n For convolutional codes N is unbounded and:

18. Free-distance Asymptotes n For convolutional code: u d free = 18 u W d o = 187 n For turbo code u d free = 6 u N free = 3 u w free = 2

19. Application: Turbo Codes for Wireless Multimedia n Multimedia systems require varying quality of service. u QoS u Latency F Low latency for voice, teleconferencing u Bit/frame error rate (BER, FER) F Low BER for data transmission. n The tradeoffs inherent in turbo codes match with the tradeoffs required by multimedia systems. u Data: use large frame sizes F Low BER, but long latency u Voice: use small frame sizes F Short latency, but higher BER

20. Influence of Interleaver Size n Constraint Length 5. n Rate r = 1/2. n Log-MAP decoding. n 18 iterations. n AWGN Channel. Voice Video Conferencing Replayed Video Data

21. Application: Turbo Codes for Fading Channels n The turbo decoding algorithm requires accurate estimates of channel parameters: u Branch metric: u Average signal-to-noise ratio (SNR). u Fading amplitude. u Phase. n Because turbo codes operate at low SNR, conventional methods for channel estimation often fail. u Therefore channel estimation and tracking is a critical issue with turbo codes.

22. Fading Channel Model n Antipodal modulation: n Gaussian Noise: n Complex Fading: u  is a constant. F  =0 for Rayleigh Fading F  >0 for Rician Fading u X and Y are Gaussian random processes with autocorrelation: Turbo Encoder Channel Interleaver BPSK Modulator Turbo Decoder De- interleaver BPSK Demod

23. Pilot Symbol Assisted Modulation n Pilot symbols: u Known values that are periodically inserted into the transmitted code stream. u Used to assist the operation of a channel estimator at the receiver. u Allow for coherent detection over channels that are unknown and time varying. segment #1 symbol #1 symbol #M p symbol #1 symbol #M p pilot symbol segment #2 symbol #1 symbol #M p symbol #1 symbol #M p pilot symbol pilot symbols added here

24. Pilot Symbol Assisted Turbo Decoding n Desired statistic: n Initial estimates are found using pilot symbols only. n Estimates for later iterations also use data decoded with high reliability. n “Decision directed” Turbo Encoder Insert Pilot Symbols Channel Interleaver Channel Interleaver Insert Pilot Symbols Compare to Threshold Filter Turbo Decoder Channel Deinterleaver Remove Pilot Symbols Delay

25. Performance of Pilot Symbol Assisted Decoding n Simulation parameters: u Rayleigh flat-fading. u r=1/2, K=3 u 1,024 bit random interleaver. u 8 iterations of log-MAP. u f d T s =.005 u M p = 16 n Estimation prior to decoding degrades performance by 2.5 dB. n Estimation during decoding only degrades performance by 1.5 dB. n Noncoherent reception degrades performance by 5 dB.

26. Other Applications of Turbo Decoding n The turbo-principle is more general than merely its application to the decoding of turbo codes. n The “Turbo Principle” can be described as: u “Never discard information prematurely that may be useful in making a decision until all decisions related to that information have been completed.” -Andrew Viterbi u “It is a capital mistake to theorize before you have all the evidence. It biases the judgement.” -Sir Arthur Conan Doyle n Can be used to improve the interface in systems that employ multiple trellis-based algorithms.

27. Applications of the Turbo Principle n Other applications of the turbo principle include: u Decoding serially concatenated codes. u Combined equalization and error correction decoding. u Combined multiuser detection and error correction decoding. u (Spatial) diversity combining for coded systems in the presence of MAI or ISI.

28. Serial Concatenated Codes n The turbo decoder can also be used to decode serially concatenated codes. u Typically two convolutional codes. Outer Convolutional Encoder Data n(t) AWGN Inner Decoder Outer Decoder Estimated Data Turbo Decoder interleaver deinterleaver interleaver Inner Convolutional Encoder APP

29. Performance of Serial Concatenated Turbo Code n Plot is from: S. Benedetto, et al “Serial Concatenation of Interleaved Codes: Performance Analysis, Design, and Iterative Decoding” Proc., Int. Symp. on Info. Theory, 1997. n Rate r=1/3. n Interleaver size L = 16,384. n K = 3 encoders. n Serial concatenated codes do not seem to have a bit error rate floor.

30. Turbo Equalization n The “inner code” of a serial concatenation could be an Intersymbol Interference (ISI) channel. u ISI channel can be interpreted as a rate 1 code defined over the field of real numbers. (Outer) Convolutional Encoder Data n(t) AWGN SISO Equalizer (Outer) SISO Decoder Estimated Data Turbo Equalizer interleaver deinterleaver interleaver ISI Channel APP

31. Performance of Turbo Equalizer n Plot is from: C. Douillard,et al “Iterative Correction of Intersymbol Interference: Turbo- Equaliztion”, European Transactions on Telecommuications, Sept./Oct. 1997. n M=5 independent multipaths. u Symbol spaced paths u Stationary channel. u Perfectly known channel. n (2,1,5) convolutional code.

32. Turbo Multiuser Detection n The “inner code” of a serial concatenation could be a multiple-access interference (MAI) channel. u MAI channel describes the interaction between K nonorthogonal users sharing the same channel. u MAI channel can be thought of as a time varying ISI channel. u MAI channel is a rate 1 code with time-varying coefficients over the field of real numbers. u The input to the MAI channel consists of the encoded and interleaved sequences of all K users in the system. n MAI channel can be: u CDMA: Code Division Multiple Access u TDMA: Time Division Multiple Access

33. System Diagram Convolutional Encoder #K n(t) AWGN SISO MUD Bank of K SISO Decoders Estimated Data Turbo MUD interleaver #K multiuser deinterleaver multiuser interleaver MAI Channel APP Convolutional Encoder #1 interleaver #1 Parallel to Serial “multiuser interleaver”

34. Simulation Results: MAI Channel w/ AWGN n From: u M. Moher, “An iterative algorithm for asynchronous coded multiuser detection,” IEEE Comm. Letters, Aug.1998. n Generic MA system u K=3 asynchronous users. u Identical pulse shapes. u Each user has its own interleaver. n Convolutionally coded. u Constraint length 3. u Code rate 1/2. n Iterative decoder.

35. Conclusion n Turbo code advantages: u Remarkable power efficiency in AWGN and flat-fading channels for moderately low BER. u Deign tradeoffs suitable for delivery of multimedia services. n Turbo code disadvantages: u Long latency. u Poor performance at very low BER. u Because turbo codes operate at very low SNR, channel estimation and tracking is a critical issue. n The principle of iterative or “turbo” processing can be applied to other problems. u Turbo-multiuser detection can improve performance of coded multiple-access systems.