A Novel Hybrid Approach to Wireless Channel Estimation Using Vision Transformers (ViTs)

Abstract

Accurate channel estimation is vital for mitigating signal distortions in wireless communication systems, particularly in highly time-dispersive environments where multipath propagation induces long-range temporal dependencies that significantly degrade system performance. Conventional approaches, including Least Squares and Maximum Likelihood estimators, as well as modern Convolutional Neural Networks (CNN)-based methods, struggle to capture these global correlations because of their inherently local receptive fields, limiting estimation accuracy under severe channel dispersion. To mitigate this limitation, this work proposes a hybrid framework is introduced where that integrates Compressive Sensing (CS) with Vision Transformers (ViTs) for robust channel estimation. The self-attention mechanism inherent to ViTs enables effective modelling of long-range dependencies across the channel response, while CS exploits the inherent sparsity of wireless channels to reduce sampling and computational overhead. Extensive simulation results indicates that the proposed CS+ViT approach achieves a Root Mean Square Error (RMSE) of 0.08 and a Bit Error Rate (BER) of 1.12 × 10⁻⁶, substantially outperforming CS+CNN and conventional estimation techniques. These results confirm the superior capability of the proposed approach in handling complex, time-dispersive multipath channels, offering a promising direction for next generation wireless systems.