Design and implementation of online learning assisted intelligent receiver

To address the issue of reliable communication under complicated scenarios, an online learning-assisted intelligent OFDM receiver was proposed. The variations of the channel environment could be precepted by the receiver, and the optimal parameters of the receiver under the current scenario were obtained by collecting data and training online. In the channel estimation module of the OFDM system, a performance comparator based on the mean square error of noisy channel samples was designed as the indicator of channel environment variations. To accelerate the online training progress, a lightweight neural network structure was applied. The proposed method was further implemented and verified based on universal software radio peripherals. The numerical simulation and over-the-air experimental results demonstrate that the proposed receiver can perceive and adapt to new environments effectively, and outperforms existing machine learning methods in terms of receiving performance and convergence rate with a limited number of pilots. © 2024 Editorial Board of Journal on Communications. All rights reserved.