Iterative Demodulation and Decoding of DPSK Modulated Turbo Codes over Rayleigh Fading Channels

1
Iterative Demodulation and Decoding of DPSK
Modulated Turbo Codes over Rayleigh Fading
Channels

• Bin Zhao, Ph.D. student
• Matthew Valenti, Assistant Professor
• Dept. of Comp. Sci. Elect. Eng.
• West Virginia University

2
Introduction

• With coherent detection, turbo codes have
  remarkable performance in AWGN and fading
  channels.
• However with non-coherent or partially coherent
  detection, severe penalty in energy efficiency
  occurs Hall Wilson
• 3.5 dB loss in energy efficiency with
  differentially detected DPSK.
• 67dB loss in energy efficiency with
  non-coherently detected FSK.
• Possible solutions to achieve near coherent
  detection performance under unknown fading
  channels
• Pilot symbol assisted modulation Valenti
  Woerner
• Trellis based modulation with per survivor
  processing (PSP)
• Turbo DPSK Hoeher Lodge
• Only convolutional codes considered to date for
  turbo DPSK.
• We extend the concept of turbo DPSK by cascading
  turbo outer codes with an accumulator inner code.

3
Extended Turbo DPSK Structure

• Code polynomials (1,13/15)
• UMTS interleaver
• for turbo code.
• -- 640 data bits.
• S-random channel interleaver.
• Soft-output, trellis- based APP demodulator for
  DPSK.
• Iterative decoding and demodulation.

4
A Simple Analytical Tool for Extended Turbo DPSK

• Similar to the tunnel theory analysis.
• S. Ten Brink, 1999.
• Suppose Turbo decoder and APP demodulator ideally
  transform input Es/No into output Es/No.
• APP demodulator DPSK ? BPSK
• Turbo code decoder Turbo Code ? BPSK
• Convergence box shows minimum SNR required for
  converge.
• corresponds to the threshold SNR in the tunnel
  theory.
• convergence box location

code rate 1/2
rate Es/No Eb/No
½ 0.5 dB 3.5 dB
? -1.3 dB 3.5 dB
5
Extended Turbo DPSK in AWGN Channel

• Rate 1/3 code.
• 3 dB energy gap between turbo codes with
  differentially detected DPSK and coherent BPSK.
• 2.5 dB energy gap between extended turbo DPSK and
  coherent turbo codes
• 0.5 dB processing gain from iterative decoding
  and APP demodulation.

2.5 dB
3 dB
6
Extended Turbo DPSK in AWGN Channel

• Rate ½ code.
• 2.5 dB energy gap between turbo codes with
  differentially detected DPSK and coherent BPSK.
• 1.75 dB energy gap between extended turbo DPSK
  and coherent turbo codes.
• 0.75 dB processing gain with iterative decoding
  and APP demodulation.
• Why does the gap close?
• Rate ½ and ? extended turbo DPSK perform
  essentially the same at 3.5 dB, which matches our
  prediction.
• However, the coherently detected turbo code
  performs worse at rate ½ than at rate ?.

1.75 dB
2.5 dB
7
Extended Turbo DPSK in Fading Channel

• Rate 1/3 code in fully interleaved Rayleigh
  flat-fading channel.
• Energy gap between turbo codes with DPSK and
  coherent BPSK expands to 4.25 dB.
• Energy gap between extended turbo DPSK and turbo
  codes expands to 3.75 dB.
• 0.5 dB processing gain with iterative decoding
  and APP demodulation.
• Why is the gap so large?
• The energy gap between BPSK and DPSK is much
  larger in fading than it is in AWGN.

3.75 dB
4.25 dB
8
Extended Turbo DPSK in Fading Channel

• Rate ½ code in fully interleaved Rayleigh
  flat-fading channel.
• 3.5 dB energy gap between turbo codes with DPSK
  and coherent BPSK.
• 3 dB energy gap between extended turbo DPSK and
  turbo codes.
• 0.5dB processing gain with iterative decoding
  and APP demodulation.
• Unlike AWGN case, extended turbo DPSK requires 1
  dB more SNR in fading at rate ? than at rate ½.

3 dB
3.5 dB
9
Conclusions and Future Work

• Extended turbo DPSK performs considerably worse
  than turbo codes with BPSK modulation and
  coherent detection (both coherently detected).
• The energy inefficiency of extended turbo DPSK
  results from a large energy gap between BPSK and
  DPSK modulation at low Es/No where turbo codes
  typically operate.
• At higher code rates, the energy gap deceases
  significantly in both AWGN and fading channels.
• An analytical tool is developed to predict the
  performance of extended turbo DPSK.
• In particular, the minimum SNR required for the
  iterative decoding structure to converge can be
  reliably predicted.

• Higher rate (higher than ½ ) extended turbo DPSK
  will be studied to better confirm our
  conclusions.
• Per-survivor based DPSK demodulation will be
  applied to estimate the channel information for
  turbo codes under unknown fading channels.
• The analytical tool we developed will be further
  studied to predict the performance of more
  generalized code concatenation schemes.
• e.g. serial concatenated convolutional codes.
• Further results to appear at the 2001 Asilomar
  Conference on Signals, Systems, and Computers
  (invited paper).