![[2602.20361] Learning During Detection: Continual Learning for Neural OFDM Receivers via DMRS](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.20361] Learning During Detection: Continual Learning for Neural OFDM Receivers via DMRS
Summary
This paper presents a continual learning framework for neural OFDM receivers that allows for real-time adaptation to changing communication channels without the need for periodic retraining.
Why It Matters
As communication channels evolve rapidly, traditional methods require frequent retraining, which can disrupt service. This research offers a solution that enables uninterrupted operation and efficient adaptation to channel variations, making it highly relevant for improving wireless communication systems.
Key Takeaways
- Introduces a zero-overhead online learning framework for OFDM receivers.
- Utilizes existing demodulation reference signals (DMRS) for simultaneous signal demodulation and model adaptation.
- Presents two innovative receiver architectures that enhance operational efficiency.
- Demonstrates effective tracking of channel distribution variations without performance degradation.
- Offers insights into pilot design strategies that support joint demodulation and learning.
Computer Science > Information Theory arXiv:2602.20361 (cs) [Submitted on 23 Feb 2026] Title:Learning During Detection: Continual Learning for Neural OFDM Receivers via DMRS Authors:Mohanad Obeed, Ming Jian View a PDF of the paper titled Learning During Detection: Continual Learning for Neural OFDM Receivers via DMRS, by Mohanad Obeed and Ming Jian View PDF HTML (experimental) Abstract:Deep neural networks (DNNs) have been increasingly explored for receiver design because they can handle complex environments without relying on explicit channel models. Nevertheless, because communication channels change rapidly, their distributions can shift over time, often making periodic retraining necessary. This paper proposes a zero-overhead online and continual learning framework for orthogonal frequency-division multiplexing (OFDM) neural receivers that directly detect the soft bits of received signals. Unlike conventional fine-tuning methods that rely on dedicated training intervals or full resource grids, our approach leverages existing demodulation reference signals (DMRS) to simultaneously enable signal demodulation and model adaptation. We introduce three pilot designs: fully randomized, hybrid, and additional pilots that flexibly support joint demodulation and learning. To accommodate these pilot designs, we develop two receiver architectures: (i) a parallel design that separates inference and fine-tuning for uninterrupted operation, and (ii) a forward-pass reusing design that...
