RC-Struct: A Structure-Based Neural Network Approach for MIMO-OFDM Detection

In this paper, we introduce a structure-based neural network architecture, namely RC-Struct, for MIMO-OFDM symbol detection. The RC-Struct exploits the temporal structure of signals through reservoir computing (RC). A binary classifier leverages repetitive constellation in system to perform multi-class incorporation RC allows be learned purely online fashion with extremely limited pilot symbols each OFDM subframe. enables efficient utilization precious training and an easy extension high-order modulations without substantial increase complexity. Experiments show that introduced outperforms both conventional model-based detection approaches state-of-the-art learning-based strategies terms bit error rate (BER). advantages over existing methods become more significant when rank link adaptation are adopted. sheds light on combining communication domain knowledge receive processing 5G/5G-Advanced Beyond.