Wireless Channel Estimation based on Transformer and Super-Resolution

This paper presents SLRTNet, a deep learning model designed for channel estimation in OFDM systems. SLRTNet employs a super-resolution network in combination with a transformer architecture, which features an improved encoder-decoder structure. To extract essential input information, a transformer encoder with a multi-head attention mechanism and a Feedforward Net that utilizes LRR Blocks are used. A multi-level residual network is implemented as a transformer decoder to improve the model's generalization ability. The simulation results show that SLRTNet outperforms other deep learning-based neural network schemes. Compared with the conventional LS algorithm, SLRTNet exhibits a performance gain of 0.7 to 2 dB in the low signal-to-noise ratio (SNR) regime and 5.5 to 8 dB in the high SNR regime. This indicates that SLRTNet has good performance in channel estimation for OFDM systems and improves the accuracy of channel estimation effectively.