Multi-Band Spectrum Detection for 6G Signals Using Hybrid LSTM Model

Spectrum sensing for 6^th Generation (6G) networks requires highly adaptive techniques to efficiently detect spectrum occupancy in dynamic and low-SNR environments. Traditional methods, such as energy detection and cyclostationary feature detection, often fail due to fixed detection thresholds and dependence on prior signal assumptions. To address these limitations, we employ Wavelet Packet Enhanced Hybrid Transform (WPEHT) where Wavelet packet decomposition extracts multi-resolution time-frequency features, preserving fine-grained spectral information across varying SNR conditions. These extracted features are processed using a hybrid Transformer network, where self-attention mechanisms refine initial feature representations and within spectral data. This model is trained on an enhanced dataset comprising 2G, 3G, 4G, 5G, Global System for mobile (GSM), and Long term evolution (LTE) signals, ensuring strong generalization to future 6G spectral environments. Compared to conventional methods such as Support Vector Machines (SVM) and Convolutional Neural Networks (CNNs), the proposed Bidirectional Long Short Term Memory (Bi-LSTM) model demonstrates superior classification accuracy, noise resilience, and real-time inference efficiency. Experimental results confirm that this approach enables high-precision spectrum occupancy detection, making it a promising solution for adaptive and intelligent 6G cognitive radio networks.