The Intelligent Receiver Scheme With Joint Training for UWB

Abstract

Current schemes are inadequate for achieving low bit error rate (BER) communication under extreme interference and limited pilot samples. Therefore, we propose a receiver scheme based on a spiral multi-hybrid convolutional network (SMMCNet). Specifically, the SMMCNet framework enhances decoding capability at low signal-to-noise ratios (SNR) by leveraging the statistical characteristics of offline white noise. The Spiral Multi-scale Hybrid Convolutions (SMMCov) reduce feature channel dimensions in multi-scale convolutions, enabling a lightweight deep network. The dual-layer shared connection mode allows deep-level, small-channel convolutions to capture diverse depth, multi-channel, and multi-scale target signal features, enhancing SMMCNet’s feature learning capability with limited samples. In extreme multipath simulations, the receiver achieves a bit error rate two orders of magnitude lower than that of a traditional receiver, with significantly fewer parameters than other deep learning receivers.