Deep learning-based performance optimisation of non-orthogonal frequency-guided transmission receivers

In the transition from 2G to 5G and the exploration of 6G, wireless communication systems face increasing challenges due to surging data traffic and diverse application demands. Traditional methods like FDMA and TDMA struggle to meet the needs of high-density scenarios due to limited spectrum utilization. To overcome these limitations, non-orthogonal frequency-guided transmission techniques offer a promising alternative. This paper proposes a deep learning-based receiver architecture optimized for processing non-orthogonal signals. Through extensive simulations, we evaluate the model's performance across various parameters, including the number of channels, batch size, and the integration of a global attention module. Our results show that fine-tuning these paraasmeters leads to a data recovery accuracy of 0.88, highlighting the effectiveness of the proposed approach.