Online Optimization for Learning to Communicate over Time-Correlated Channels

Zheshun Wu, Junfan Li, Zenglin Xu, Sumei Sun, Jie Liu

Abstract

Machine learning techniques have garnered great interest in designing communication systems owing to their capacity in tackling channel uncertainty. To provide theoretical guarantees for learning-based communication systems, some recent works analyze generalization bounds for devised methods based on the assumption of Independently and Identically Distributed (I.I.D.) channels, a condition rarely met in practical scenarios. In this paper, we drop the I.I.D. channel assumption and study an online optimization problem of learning to communicate over time-correlated channels. To address this issue, we further focus on two specific tasks: optimizing channel decoders for time-correlated fading channels and selecting optimal codebooks for time-correlated additive noise channels. For utilizing temporal dependence of considered channels to better learn communication systems, we develop two online optimization algorithms based on the optimistic online mirror descent framework. Furthermore, we provide theoretical guarantees for proposed algorithms via deriving sub-linear regret bound on the expected error probability of learned systems. Extensive simulation experiments have been conducted to validate that our presented approaches can leverage the channel correlation to achieve a lower average symbol error rate compared to baseline methods, consistent with our theoretical findings.

1. Introduction

In the field of machine learning for communications, most theoretical "generalization bounds" rely on the assumption that wireless channels are Independently and Identically Distributed (I.I.D.). This assumes that the channel state at any given moment is a random draw completely independent of the previous state. In reality, wireless channels possess "memory" due to factors like user mobility or hardware imperfections, creating time-correlated channels. When this correlation exists, standard ML models can struggle to generalize, and traditional statistical proofs often fail. This paper moves beyond these unrealistic assumptions by applying Optimistic Online Mirror Descent (OMD) to develop algorithms that adapt to time-correlated channels in real-time.

2. Problem Formulation: Time-Correlated Channels

Unlike I.I.D. models, time-correlated channels exhibit temporal dependence where the current state is heavily influenced by previous states. The communication process is modeled in rounds, where each round corresponds to a coherence time during which the channel remains relatively constant but varies between rounds according to a underlying statistical process.

Illustration of time-correlated channel rounds where channel statistics vary across coherence times.

3. Proposed Framework: Optimistic Online Mirror Descent

The researchers frame the communication problem as an Online Optimization task. Instead of a one-time training phase, the system updates itself at every round based on a small training sequence. The core technique is Optimistic OMD, which treats channel correlation as a "hint." By using the gradient from the previous round as a predictor for the current one, the algorithm converges much faster than standard gradient descent.

This framework is applied to two primary tasks:

  1. Learning a Channel Decoder: For fading channels, a hinge-type surrogate loss function approximates error probability, optimized using a Euclidean regularizer.
  2. Codebook Selection: For additive noise channels, the problem is treated as a Multi-Armed Bandit (MAB) task, where the receiver selects the best modulation scheme from a super-codebook using a Log-Barrier regularizer.
The two communication tasks: (a) learning a decoder and (b) learning a codebook via training sequences.

4. Theoretical Guarantees and Surrogate Losses

Minimizing the Symbol Error Rate (SER) directly is difficult because it is a discrete, non-convex metric. The authors derived a convex "hinge-type" surrogate loss that upper-bounds the error probability, making it mathematically possible to provide rigorous performance guarantees. They proved sub-linear regret bounds, meaning that over time, the average performance of the online algorithm becomes as effective as if the optimal fixed strategy had been known from the start.

5. Experimental Evaluation

Using synthetic channel models like first-order Markov processes and complex noise distributions (Gaussian and Laplace Mixture Distributions), the study demonstrated several key findings:

  • Superior Accuracy: The proposed OMD method achieved a significantly lower Average SER compared to standard baselines like Least Squares (LS) estimation and Online Gradient Descent (OGD).
  • Faster Convergence: The "Optimistic" approach reached low error rates in fewer communication rounds than non-optimistic methods.
  • Correlation Sensitivity: Experiments confirmed that higher channel correlation leads to better performance, validating that the model successfully exploits temporal dependence.

6. Conclusion

By dropping the I.I.D. assumption, this work provides a more realistic and theoretically sound approach to learning-based communications. While challenges remain regarding computational overhead for low-power devices and feedback reliability in poor conditions, the use of Optimistic OMD offers a powerful path forward for adaptive, real-time communication systems.

Summary updated

Abstract

Machine learning techniques have garnered great interest in designing communication systems owing to their capacity in tacking with channel uncertainty. To provide theoretical guarantees for learning-based communication systems, some recent works analyze generalization bounds for devised methods based on the assumption of Independently and Identically Distributed (I.I.D.) channels, a condition rarely met in practical scenarios. In this paper, we drop the I.I.D. channel assumption and study an online optimization problem of learning to communicate over time-correlated channels. To address this issue, we further focus on two specific tasks: optimizing channel decoders for time-correlated fading channels and selecting optimal codebooks for time-correlated additive noise channels. For utilizing temporal dependence of considered channels to better learn communication systems, we develop two online optimization algorithms based on the optimistic online mirror descent framework. Furthermore, we provide theoretical guarantees for proposed algorithms via deriving sub-linear regret bound on the expected error probability of learned systems. Extensive simulation experiments have been conducted to validate that our presented approaches can leverage the channel correlation to achieve a lower average symbol error rate compared to baseline methods, consistent with our theoretical findings.
BibTeX
@article{wu2026online,
  author = {Zheshun Wu and Junfan Li and Zenglin Xu and Sumei Sun and Jie Liu},
  title = {{Online Optimization for Learning to Communicate over Time-Correlated Channels}},
  journal = {IEEE Trans. Cogn. Commun. Netw.},
  year = {2026},
  doi = {10.1109/tccn.2025.3595551},
}
Figure