Deep Unfolding-based Distributed MIMO Detection
Masaya Kumagai, Ayano Nakai-Kasai, Tadashi Wadayama · 2023
This paper proposes a distributed multiple-input multiple-output signal detection algorithm based on DU (deep unfolding), which is known as the optimizing method of parameters involved in an iterative algorithm by using commonly used deep learning techniques such as backpropagation and stochastic gradient descent. In the distributed system, multiple distributed SPUs (sub processing units) having multiple antennas execute an iterative signal processing process to detect transmitted signals in cooperation with a CPU (central processing unit). In applications where SPUs are located distributedly, SPUs are placed under heterogeneous SNR (signal-to-noise ratio) environments depending on geographical location and distances from transmitters. We introduce weighting parameters corresponding to each SPU into the aggregation process of the proposed method to incorporate the heterogeneous SNR environments into the detection algorithm. In addtion, we apply DU to optimize the parameters and obtain improvement of the detection perfomance. Results of numerical experiments indicate that the proposed method with the parameters learned by DU improve detection performance by appropriately tuning the weighting parameters according to the heterogeneous SNR environments.