Coding for Gaussian Two-Way Channels: Linear and Learning-Based Approaches

Junghoon Kim, Taejoon Kim, Anindya Bijoy Das, Seyyedali Hosseinalipour, David J. Love, Christopher G. BrintonIEEE Trans. Information Theory 2023 · DOI · OpenReview

TL;DR

This paper introduces novel linear and RNN-based coding strategies for Gaussian Two-Way Channels (GTWCs) to improve communication reliability. By jointly designing encoders and decoders, these methods effectively handle user cooperation and power constraints, outperforming conventional schemes in sum-error performance.

What Problem Does This Paper Address?

Modern communication systems, from WiFi to social media, rely heavily on two-way communication where users exchange information bidirectionally. The Gaussian Two-Way Channel (GTWC) is a fundamental model for this, but improving its communication reliability, especially under power constraints and with user cooperation, presents significant challenges. While channel capacity is well-understood, achieving robust error performance in finite block-length codes for GTWCs is an open problem. Existing work on feedback-enabled one-way channels has developed sophisticated linear and learning-based coding techniques, but directly generalizing these to the coupled, two-way nature of GTWCs is non-trivial due to the inherent coupling of user encoding processes and the need for effective decoding that captures this coupling.

Methods

The authors propose two main approaches:

  • Linear Coding: This approach focuses on tractable mathematical derivations. It formulates a signal model that explicitly captures the coupling of encoders. They derive an optimal linear decoding strategy and convert the sum-error minimization problem into a weighted sum-power minimization problem. This allows for new insights into user cooperation and the development of an efficient joint encoder-decoder design algorithm. The method is extended to medium/long block-lengths.
  • Learning-Based Coding (RNNs): Inspired by deep learning advancements, this approach uses recurrent neural networks (RNNs) to provide greater flexibility. They introduce a state propagation-based encoding using interactive RNNs to capture user cooperation and a power control layer for efficient power management. For decoding, bi-directional RNNs with an attention mechanism are employed to exploit correlations among received symbols. The entire architecture is trained end-to-end using an autoencoder framework.

Datasets

Not discussed in the paper. The paper focuses on theoretical development and simulation-based evaluation.

Key Insights and Ideas

  • User Cooperation is Key for Reliability: While cooperation doesn't increase GTWC capacity, it significantly improves communication reliability by allowing users to help each other, especially in asymmetric channel conditions.
  • Coupling is a Double-Edged Sword: The coupling of user encoding processes is a core challenge but also the source of potential gains. Effectively capturing this coupling in both encoding and decoding is crucial.
  • Linear vs. Learning Trade-offs: Linear coding offers low complexity and good performance at high SNRs, while RNN-based coding excels at low SNRs and in noisy, stochastic environments due to its ability to learn complex feature representations.
  • Helper Concept Extends to GTWCs: The idea of a user with a better channel acting as a "helper" to improve the other user's performance is validated and extended to the context of weighted sum-power minimization.
  • RNNs Offer Flexibility: The learning-based approach, particularly with bi-directional RNNs and attention, allows for more sophisticated handling of channel dynamics and user interactions compared to linear methods.

Key Results

  • Both proposed linear and RNN-based coding strategies significantly outperform conventional channel coding schemes (which do not utilize user cooperation) in terms of sum-error performance.
  • Linear coding performs best at high SNRs, while RNN-based coding excels at low SNRs.
  • In asymmetric channels, the user with a higher SNR sacrifices some of its error performance to help the user with a lower SNR, leading to improved overall sum-error performance.
  • The RNN-based coding demonstrates robustness to high channel noises.
  • The paper provides detailed analysis of power distribution, block-length gain, and the impact of varying rates and modulation orders.

Limitations and Future Work

  • Optimality Loss in Linear Coding: The iterative optimization for linear coding, while tractable, may incur a slight optimality loss compared to exhaustive search methods, though this loss is shown to be less than 3% for small N.
  • Convergence of Iterative Optimization: The theoretical convergence of the iterative optimization process for linear coding is challenging to prove and is demonstrated empirically.
  • Complexity of RNNs: While offering flexibility, the RNN-based coding architecture has higher encoding/decoding complexity.
  • Future Research: The authors suggest exploring concatenated coding with their schemes and developing error correction codes based on their learning architecture. Investigating lower bounds on sum-BLER or upper bounds on coding rates for cooperative GTWC systems in the finite block-length regime is also identified as an interesting avenue.

Summary updated

Abstract

Although user cooperation cannot improve the capacity of Gaussian two-way channels (GTWCs) with independent noises, it can improve communication reliability. In this work, we aim to enhance and balance the communication reliability in GTWCs by minimizing the sum of error probabilities via joint design of encoders and decoders at the users. We first formulate general encoding/decoding functions, where the user cooperation is captured by the coupling of user encoding processes. The coupling effect renders the encoder/decoder design non-trivial, requiring effective decoding to capture this effect, as well as efficient power management at the encoders within power constraints. To address these challenges, we propose two different two-way coding strategies: linear coding and learning-based coding. For linear coding, we propose optimal linear decoding and discuss new insights on encoding regarding user cooperation to balance reliability. We then propose an efficient algorithm for joint encoder/decoder design. For learning-based coding, we introduce a novel recurrent neural network (RNN)-based coding architecture, where we propose interactive RNNs and a power control layer for encoding, and we incorporate bi-directional RNNs with an attention mechanism for decoding. Through simulations, we show that our two-way coding methodologies outperform conventional channel coding schemes (that do not utilize user cooperation) significantly in sum-error performance. We also demonstrate that our linear coding excels at high signal-to-noise ratios (SNRs), while our RNN-based coding performs best at low SNRs. We further investigate our two-way coding strategies in terms of power distribution, two-way coding benefit, different coding rates, and block-length gain.
BibTeX
@article{kim2023coding,
  author = {Junghoon Kim and Taejoon Kim and Anindya Bijoy Das and Seyyedali Hosseinalipour and David J. Love and Christopher G. Brinton},
  title = {{Coding for Gaussian Two-Way Channels: Linear and Learning-Based Approaches}},
  journal = {IEEE Trans. Information Theory},
  year = {2023},
  doi = {10.1109/tit.2025.3566324},
}
Figure