Aider: Artificial Intelligent-Based Deep Receiver for Wireless Communication Systems

Deep learning (DL) has found extensive applications in wireless communication, propelling DL-based physical layer orthogonal frequency division multiplexing (OFDM) receivers to the forefront. However, existing research mostly focuses on training frequency-domain signals to achieve bit recovery, which involves additional pre-processing steps that add complexity to receiver design and loss of temporal information. In this letter, we propose a novel attentive deep convolutional network (ADCN) with a complexity of O(N2) design that uses a learned linear transform and exploitation of the cyclic prefix (CP), which can replace these pre-processing procedures. Additionally, we enhance ADCN with an attention mechanism that captures the time dependence between OFDM signals, employs causal convolution to reveal the long-term properties of the signals, and addresses the issue of temporal information loss. According to 3GPP-defined channel models, we demonstrate that ADCN is superior to traditional methods and performs well in channels with large delay spread.