MaxMB: A Signal Detector for OFDM Receiver Based on Multi-Axis Attention and MBConv
Xuefeng Wang, Yang Lu, Ruichen Zhang, Gaofeng Pan, Bo Ai, Dusit Tao Niyato · IEEE Transactions on Vehicular Technology · 2025
This paper proposes MaxMB Net, a deep learning (DL) based signal detector, that integrates multi-axis attention and mobile inverted bottleneck convolution (MBConv) to enhance the feature extraction. Through supervised training, we implement MaxMB Net in an orthogonal frequency-division multiplexing (OFDM) receiver for direct signal detection from the pilot signals and the received signals, and the complexity of channel estimation is alleviated. Extensive experiments across typical 5G New Radio (NR) and Long Term Evolution (LTE) channel models demonstrate that MaxMB Net achieves bit-error rate performance comparable to the ideal linear minimum mean-square error (LMMSE) receiver. Besides, MaxMB Net outperforms state-of-the-art DL-based receivers across all signal-to-noise-ratio regions.