The Adaptive Modulation Coding Technology Research Based On Reinforcement Learning

In modern wireless communication systems,transmission systems with high data rate and spectrum utilization have always been proposed,but wireless channel environments are time-varying and fading.High-data-rate signals can easily generate the inter symbol interference(ISI)in frequency-selective fading channels Interference,thus affecting the overall transmission performance of the system.In order to improve the throughput of the system,adaptive modulation and coding(AMC)is designed in the transmission link.This technique dynamically adjusts the modulation and coding scheme(MCS)of the transmitted signal according to the channel state information(CSI)fed back by the receiving end at the transmission rate and transmission reliability to find a balance between,in order to improve system throughput.However,the conventional AMC technology selects the corresponding modulation and coding scheme(MCS)according to the threshold calculated by combining the modulation and coding scheme and the CSI fed back by the receiving end in advance.This method depends largely on the assumed channel model,in fact,the traditional AMC algorithm is often difficult to meet the requirements of the frame error rate(FER)because the channel does not completely follow the ideal distribution,the additive noise is not all Gaussian and the amplifier has nonlinear effects.Therefore,we propose an adaptive modulation and coding algorithm based on reinforcement learning,which does not rely on a perfect mathematical model and can autonomously learn in the wireless channel,and then determine the correspondence of the signal to noise ratio and MCS according to the actual BER performance of the system.At the same time,deep learning is introduced on the basis of reinforcement learning,and an MCS selection algorithm based on depth Q network(DQN)is proposed.Instead of using low-dimensional channel quality indicators(such as signal-to-noise ratio),the model directly uses the estimated channel values and then learns more accurate CSI through the neural network,which helps to select a more accurate modulation and coding scheme.The results of the link simulation show that AMC based on reinforcement learning have higher system throughput than the traditional look-up table AMC in the presence of non-linear distortion in the amplifier,receiver SNR estimation error and strong selective fading in the channel,and AMC based on reinforcement learning is more generalized than the traditional look-up table AMC technology.