Adaptive modulation and coding using deep recurrent neural network

Adaptive Modulation and Coding (AMC) is a promising technique to increase the average spectral efficiency of communication links. This research proposes a novel AMC method based on a supervised deep learning approach to maximize the average spectral efficiency of OFDM wireless systems while the bit error rate (BER) remains under a predefined threshold. The proposed method consists of a one-dimensional convolutional network that performs feature extraction and a long short-term memory network that learns the behavior of the channel. Input features are the magnitudes and phases of the estimated channel frequency response in the pilot subcarriers and signal-to-noise ratio. Datasets of various fading channel responses were generated using WINNER II. The proposed method was compared with previous methods based on different criteria, including average spectral efficiency, BER, the accuracy of predictions, the average delay of each prediction, and model complexity. The simulation results confirmed the superiority of the proposed AMC method.