A hybrid deep learning framework for cooperative resource management in 5G using spike-driven transformers and cycle-consistent adaptation

The rapid evolution of 5G wireless communication systems has significantly advanced wireless communication technologies. However, the increasing demand for high data throughput and stringent QoS requirements presents major challenges for conventional resource management approaches. Traditional optimization algorithms often fail to adapt efficiently to rapidly changing channel conditions, resulting in underutilized resources and increased computational complexity. While various resource allocation strategies have been proposed, many do not fully exploit the dynamic nature of wireless environments. To address these limitations, this work proposes a novel cooperative resource allocation framework that integrates an SDT with a CCAAN. This hybrid approach leverages the SDT’s capability to capture fine-grained spatiotemporal patterns in channel dynamics and the CCAAN’s strength in domain adaptation under non-stationary conditions. The inclusion of a BLSO ensures efficient hyperparameter tuning, balancing model accuracy with low computational overhead. The method enables adaptive resource allocation under fluctuating wireless conditions, leading to more precise and efficient resource distribution. Experimental results demonstrate that the proposed SDT-CCAAN framework achieves 99.8% accuracy, 32.5 dB SINR, and 98.7% sensitivity, while maintaining a low processing time of 12.5 ms. These results demonstrate improved performance compared with conventional methods in terms of efficiency, reliability, and adaptability. This work highlights the potential of combining spiking neural architectures with adversarial learning and nature-inspired optimization to meet the complex demands of next-generation wireless networks.