Channel State Prediction Scheme Based on Deep Learning

The rapid growth of wireless communication and data transmission needs has led to an increasing demand for more efficient and reliable transmission methods in communication systems. In this context, link adaptive coding techniques have been gradually developed. However, in complex and changing channel environments, the performance of conventional fixed coding and modulation schemes is limited when the channel changes frequently. To cope with these problems, communication systems require coding methods that can dynamically respond to the channel state. Channel estimation becomes a key technique in link adaptive coding systems, which can monitor and feedback channel changes, such as signal fading, interference, and multipath effects, in real time, and provide accurate channel state information for adaptive coding, so as to support the system to flexibly adjust the coding and modulation schemes according to different channel conditions. The article explores the limitations of traditional channel estimation techniques in rapidly changing channels and proposes a novel fast channel prediction model that combines convolutional neural network (CNN) and Transformer. The model extracts the spatial features of the signal through CNN and captures the global features using Transformer to achieve highly accurate channel prediction. Simulation results show that compared with the CNN-LSTM model, the prediction accuracy of the CNN+Transformer model is improved by 88.7%, and the Normalised Mean Square Error (NMSE) value of the multipath delay prediction is reduced to 0.0096, which is an excellent performance.