A Deep Equilibrium MIMO Detector With Adaptive Depth

In the massive multiple-input multiple-output (MIMO) system, the model-driven detection network that unfolds the traditional iterative algorithm can achieve sub-optimal performance with relatively low complexity. These detectors usually have a fixed number of layers for different channel environments. However, the difficulty of the detection problem varies across channel scenarios. To improve the detection efficiency of the network, the number of layers of the model-driven detection network should change adaptively as the channel scenario changes. To improve the adaptability of the number of layers of the deep-learning based iterative soft thresholding algorithm (DISTA) to the channel environment, this paper proposes a deep MIMO detector named DE-ISTA with adaptive depth based on the deep equilibrium model, which reconstructs the signal recovery iteration into a fixed-point iteration framework. Experiments show that DE-ISTA is able to achieve comparable performance to DISTA with fewer parameters and iterations on average.