Abstract: For Massive MIMO systems, classical detection algorithms face challenges in terms of both performance and complexity. In this paper, a good compromise between performance and complexity is achieved by organically combining parallel interference cancellation (PIC) and deep neural network (DNN). By applying the idea of PIC, the MIMO system was equated to multiple parallel single-input multiple-output (SIMO) systems, and the DNN was applied to the SIMO system for signal detection. The designed DNN modeled the signal detection as a deep learning classification problem and achieved blind detection of the received signal without channel state information. The simulation results show that the proposed algorithm has a significant advantage over the classical detection algorithms in terms of BER performance, and its BER performance is close to that of the maximum likelihood detection algorithm when the number of receiving antennas is larger than the number of transmitting antennas, and has better robustness.