Online Learning of Modular Bayesian Deep Receivers: Single-Step Adaptation with Streaming Data

TL;DR

The paper tackles the challenge of deploying DNN-based wireless receivers in highly dynamic channels where conventional online SGD learning incurs high latency. It reframes online training as Bayesian tracking in a dynamic parameter space and leverages modular deep receivers (notably DeepSIC) to enable localized, single-step updates with streaming pilot data. By developing cm-ekf and Bayesian online natural gradient methods, and by organizing updates module-wise with pipelining, the approach achieves fast adaptation, reduced latency, and improved robustness across realistic channel models (Sionna, COST2100, QuaDRiGa). Experimental results show near-optimal BER and strong tracking performance with substantially lower update overhead compared to end-to-end SGD baselines, supporting practical deployment for edge AI-enabled receivers in time-varying wireless environments.

Abstract

Deep neural network (DNN)-based receivers offer a powerful alternative to classical model-based designs for wireless communication, especially in complex and nonlinear propagation environments. However, their adoption is challenged by the rapid variability of wireless channels, which makes pre-trained static DNN-based receivers ineffective, and by the latency and computational burden of online stochastic gradient descent (SGD)-based learning. In this work, we propose an online learning framework that enables rapid low-complexity adaptation of DNN-based receivers. Our approach is based on two main tenets. First, we cast online learning as Bayesian tracking in parameter space, enabling a single-step adaptation, which deviates from multi-epoch SGD . Second, we focus on modular DNN architectures that enable parallel, online, and localized variational Bayesian updates. Simulations with practical communication channels demonstrate that our proposed online learning framework can maintain a low error rate with markedly reduced update latency and increased robustness to channel dynamics as compared to traditional gradient descent based method.

Figures (10)

• Figure 1: Illustration of proposed framework for single-step online adaptation of modular Bayesian deep receivers
• Figure 2: Transmission scheme with synchronization and periodic pilot-data blocks.
• Figure 3: DeepSIC model architecture illustration.
• Figure 4: Pipelined DeepSIC model architecture illustration.
• Figure 5: Symbol error rates, linear rotation channel.