Non-Memoryless Analog Network Coding in Two-Way Relay Channel

Shengli Zhang, Soung Chang Liew, QingFeng Zhou, Lu Lu, Hui Wang

TL;DR

This paper introduces a novel "non-memoryless" Analog Network Coding (ANC) scheme for two-way relay channels. By leveraging the inherent correlations within channel-coded packets at the relay, this approach significantly improves the accuracy of the superimposed signal estimate, leading to better overall system performance compared to existing memoryless ANC methods.

What Problem Does This Paper Address?

Two-way relay channels are crucial for efficient wireless communication, allowing two nodes to exchange information via a relay. Physical Layer Network Coding (PNC) enhances throughput by having the relay combine and retransmit signals. Analog Network Coding (ANC) is a type of PNC where the relay performs analog processing (like amplification) on the superimposed signals from the two end nodes. However, most existing ANC schemes treat each symbol independently (memoryless), failing to exploit the structure and redundancy present in channel-coded packets. This oversight limits their performance, especially in scenarios where noise can accumulate over multiple hops. The paper aims to bridge this gap by proposing a non-memoryless ANC scheme that utilizes channel decoding principles at the relay to improve the estimation of the superimposed signal.

Methods

The core innovation lies in the relay's decoding strategy. Instead of simply amplifying the received superimposed signal, the relay node employs a soft-input soft-output decoder. This decoder treats the combined received signal from the two end nodes as a "virtual code" and uses a belief propagation algorithm to estimate the joint probability distribution of the transmitted symbols from both nodes. This process explicitly accounts for the correlations introduced by the channel coding at the end nodes. The output of this decoding process is a refined Minimum Mean Square Error (MMSE) estimate of the superimposed signal, which is then broadcast back to the end nodes. The paper also analyzes the performance in terms of Signal-to-Noise Ratio (SNR) improvement and derives a lower bound for a simple repeat code.

Datasets

Not discussed in the paper. The evaluation is based on numerical simulations using randomly generated LDPC codes and channel coefficients.

Key Insights and Ideas

  • Exploiting Channel Coding Redundancy: The most significant insight is that by treating the superimposed signal as a "virtual code" and applying belief propagation, the relay can effectively leverage the correlations within channel-coded packets. This moves beyond simple symbol-by-symbol processing.
  • Non-Memoryless Advantage: The non-memoryless approach allows the relay to overcome the limitations of memoryless schemes, particularly in mitigating noise accumulation and improving the accuracy of the superimposed signal estimate.
  • LDPC Code Adaptability: The paper highlights that different types of LDPC codes (varying in complexity and degree distribution) perform optimally in different SNR regimes. Simpler codes (like repeat codes) are better at lower SNRs, while more complex codes excel at higher SNRs. This suggests a flexible approach to code selection based on channel conditions.
  • MMSE Estimation Refinement: The proposed scheme achieves a more accurate MMSE estimate of the superimposed signal compared to conventional memoryless ANC, which directly translates to improved end-to-end communication performance.

Key Results

The proposed non-memoryless ANC scheme demonstrates significant performance improvements:

  • SNR Improvement: The scheme offers an SNR improvement at the relay node that is lower bounded by 1/R (where R is the code rate), with a repeat code achieving a lower bound of -10log(R) dB.
  • MMSE Reduction: Numerical simulations show that the non-memoryless ANC scheme can reduce the Mean Squared Error (MSE) at the relay by 3dB to 6dB compared to memoryless ANC.
  • Code Selection Impact: The performance varies with LDPC code complexity. For instance, in simulations, a (1,2) LDPC code (repeat code) performed best at SNRs below 2.5 dB, while a (3,6) LDPC code was superior at higher SNRs.

Limitations and Future Work

  • Complexity: The proposed non-memoryless scheme is inherently more complex than memoryless ANC due to the channel decoding process at the relay.
  • Code Rate vs. Capacity: The paper acknowledges that for most channels, achievable end-to-end code rates are beyond the relay's capacity, posing a unique challenge for relay decoding.
  • Future Work: The authors suggest extending the scheme to other channel codes like Reed-Solomon (RA) codes and Turbo codes. They also mention exploring more sophisticated combinations of channel coefficients (a, b) for the broadcast signal and investigating the impact of different code complexities in various SNR regions. The paper also hints at future work on hybrid PNC schemes.

Summary updated

BibTeX
@inproceedings{zhang2011non,
  author = {Shengli Zhang and Soung Chang Liew and QingFeng Zhou and Lu Lu and Hui Wang},
  title = {{Non-Memoryless Analog Network Coding in Two-Way Relay Channel}},
  booktitle = {ICC},
  year = {2011},
  doi = {10.1109/icc.2011.5963216},
}
Figure