Hybrid Neural Network-Based Signal Detection for Multi-User MIMO With Hardware Imperfections

This work deals with the issue of communication signal detection in multi-user multi-input multi-output (MIMO) with hardware imperfections, where low-cost users suffer from power amplifier nonlinearity (PAN) and in-phase and quadrature imbalance (IQI). To deal with the multi-user interference and nonlinear distortions due to hardware imperfections, we propose a novel signal detector using a hybrid neural network (HNN), which combines a Kolmogorov-Arnold network (KAN) and a conventional linear NN layer. The training of the HNN involves the update of both linear NN weights and nonlinear activation functions. It turns out that the proposed HNN detector with a much smaller model size significantly outperforms conventional polynomial-based techniques and state-of-the-art pure NN or KAN techniques.