Abstract

The rapid growth of immersive applications, massive Internet of Things (IoT) connectivity, intelligent transportation, and ultra-reliable services in 6G networks increases demand for radio spectrum. Conventional static allocation cannot adequately utilize spectrum that remains temporarily unused, while conventional sensing techniques experience performance degradation under low signal-to-noise ratio (SNR), noise uncertainty, fading, and rapidly changing channel conditions. Recent studies indicate that deep learning can improve spectrum sensing by learning discriminative signal characteristics directly from received observations. Existing approaches generally address spectrum sensing and spectrum allocation as separate tasks, which can limit their responsiveness to instantaneous spectrum occupancy and traffic conditions. This study proposes a novel Deep Adaptive Spectrum Sensing and Allocation Network (DASSAN) that integrates a CNN-BiLSTM feature-learning module with an attention mechanism for spectrum occupancy detection and a deep reinforcement learning allocator for dynamic channel selection. The CNN extracts local spectral features, while BiLSTM captures temporal channel occupancy patterns. The attention mechanism emphasizes informative features, and the reinforcement-learning agent jointly considers spectrum availability, interference, traffic demand, and channel quality to select allocation decisions. The integrated framework is designed to improve sensing reliability, spectrum utilization, throughput, and allocation adaptability in dynamic 6G environments. The simulated results show that DASSAN achieves 99.1% accuracy, 99.3% detection probability, 3.4% false alarm rate, 128.3 Mbps throughput, and 95.8% spectrum utilization at 20 dB SNR. At 0 dB, it achieves 93.1% accuracy, 95.1% detection probability, 6.4% false alarm rate, 71.9 Mbps throughput, and 83.9% spectrum utilization. These results demonstrate the effectiveness of integrating deep spectrum sensing with adaptive reinforcement-learning-based allocation for dynamic 6G environments.

Authors

D. Swamydoss¹, B. Gopinathan²
Adhiyamaan College of Engineering, India¹, Viswam Engineering College, India²

Keywords

Deep Learning, Spectrum Sensing, Dynamic Spectrum Allocation, 6G Networks, Cognitive Radio