Deep-Learning-Based Initial Access Method for Millimeter-Wave MIMO Systems

Abstract

This letter proposes a deep-learning-based initial access (IA) method for a millimeter-wave (mmW) multiple-input multiple-output (MIMO) system. By detecting the primary synchronization signal (PSS) in the IA block, UE can discover and synchronize with the BS. PSS detection can be modeled as a binary hypothesis testing problem with unkonwn channel, carrier frequency offset (CFO), and timing offset (TO). Unlike the conventional methods using energy as the detection statistic, the proposed method uses probability as the detection statistic. Specifically, we first preprocess the received signal, and then input the results into a pre-trained convolutional neural network (CNN). The CNN can output the probability that PSS is present. After comparing the maximum probability with the threshold, the UE can determine whether there is PSS. Once PSS is detected, the estimation of TO can be obtained at the same time. Simulation results demonstrate that, the proposed IA method can outperform the conventional energy-based IA method in both miss detection rate performance of PSS and normalized mean square error (NMSE) performance of TO.