Off-Grid Channel Estimation Based on Model-Driven Deep Learning With Adaptive Depth for Millimeter Wave Massive MIMO

In millimeter-wave (mmWave) communication systems, conventional channel estimation methods suffer from limited angle quantization resolution, leading to off-grid problems that degrade estimation accuracy. To address this issue, we propose an adaptive-depth model-driven deep unfolding neural network for channel estimation. In this framework, the parallel interference cancellation assisted multi-grid matching pursuit (PIC-MGMP) algorithm is unfolded into a deep neural network, with an adaptive stopping mechanism incorporated to enable depth flexibility. Each neural network layer progressively optimizes the angle quantization matrices through training, thereby effectively mitigating off-grid effects. Simulation results demonstrate significant improvement in estimation accuracy compared to conventional channel estimation approaches.