Abstract—Massive MIMO is a new configuration of MIMO technology that uses many antennas up to the order of hundreds to serve tens or hundreds of User Equipment (UE) at the same time and frequency. Massive MIMO technology is one of the spatial diversity techniques used to increase the Spectral Efficiency (SE) of current Fifth-Generation (5G) communication systems. Massive MIMO has a high complexity in signal processing because it serves a large amount of user traffic at the same time. Therefore, this paper proposes using the Deep Deterministic Policy Gradient (DDPG), a deep learning method that significantly improves both SE and runtime performance, making it highly effective for large-scale wireless communication systems. In the simulation, we are modeling a massive MIMO system with multiple Access Points (APs) and User Equipment (UEs). We are training the channel using the proposed DDPG model. Then, we analyze each end-user path’s Signal-to-Interference plus Noise Ratio (SINR) and compare it with conventional massive MIMO (without deep learning). In addition, the complexity of the proposed DDPG model in terms of runtime is analyzed and compared with the Convex Optimization Algorithm (CVX). The simulation results indicate that the performance of the massive MIMO system is improved with the proposed DDPG model. It achieves an optimal and higher spectral efficiency (SE) of 85% compared to not using the DDPG method. Additionally, it achieves an average Signalto- Interference-plus-Noise Ratio (SINR) of 19.54 dB, while the conventional method only provides an average SINR of 15.32 dB. Furthermore, the proposed DDPG model has a lower complexity with a runtime ratio of 1:8000 compared to the CVX algorithm for the same number of epochs.