Title:On Purely Data-Driven Massive MIMO Detectors

Abstract:To enhance the performance of massive multi-input multi-output (MIMO) detection using deep learning, prior research primarily adopts a model-driven methodology, integrating deep neural networks (DNNs) with traditional iterative detectors. Despite these efforts, achieving a purely data-driven detector has remained elusive, primarily due to the inherent complexities arising from the problem's high dimensionality. This paper introduces ChannelNet, a simple yet effective purely data-driven massive MIMO detector. ChannelNet embeds the channel matrix into the network as linear layers rather than viewing it as input, enabling scalability to massive MIMO scenarios. ChannelNet is computationally efficient and has a computational complexity of $\mathcal{O}(N_t N_r)$, where $N_t$ and $N_r$ represent the numbers of transmit and receive antennas, respectively. Despite the low computation complexity, ChannelNet demonstrates robust empirical performance, matching or surpassing state-of-the-art detectors in various scenarios. In addition, theoretical insights establish ChannelNet as a universal approximator in probability for any continuous permutation-equivariant functions. ChannelNet demonstrates that designing deep learning based massive MIMO detectors can be purely data-driven and free from the constraints posed by the conventional iterative frameworks as well as the channel and noise distribution models.