REINFORCEMENT LEARNING-BASED 5G NETWORK OPTIMIZATION USING BER SIMULATION

5G Wireless Communication Based on Reinforcement Learning Essay 5G wireless communication technologies are predicted to be able to maintain dependable communication while providing the highest possible data rates and minimum delays over time-varying channels. The paper explores the feasibility of employing adaptive modulation as a Markov decision process. The paper trains a Deep Q Network (DQN) controller to make decisions on the most appropriate modulation, out of QPSK, 16-QAM, and 64-QAM options, based on the estimated Signal-to-Noise Ratio (SNR). The communication system has a 20 MHz bandwidth, subject to Rayleigh fading and additive white Gaussian noise. The transmitter modulates binary information data with selected encoding methods. The receiver demodulates and decodes the input data. The BER, latency, throughput, and spectral efficiency are evaluated for the feedback process to the DQN agent. The reward function in the paper's experiments prioritizes lower BER, lower latency, and higher throughput for the selected action.The results provided by the authors demonstrate that the BER for all three modulation schemes decreases with increasing SNR as expected. The results also confirm that higher-order modulations have higher latency than lower-order ones on average. Most importantly, the experiments have shown that the DQN agent learned to switch between different modulations depending on the conditions and follow the optimal policy of switching from QPSK to 16-QAM and eventually to 64-QAM with rising SNR. The study therefore proves the feasibility of adopting a learning-based approach to modulation control and serves as a valuable basis for future research into the 5G physical layer.