SCS-DSSS: a compressive sensing and deep semantic segmentation framework for cooperative spectrum sensing in 5G cognitive radio networks

The rapid proliferation of wireless devices and the advent of advanced 5G new radio (NR) standards have intensified the need for efficient spectrum utilisation. Cognitive radio (CR) with dynamic spectrum access (DSA) offers a solution, yet existing wideband sensing methods face a trade-off between accuracy and data overhead. This work proposes SCS-DSSS, an end-to-end framework integrating spectrogram compressive sensing (SCS) with DeepLabv3+ semantic segmentation using a ResNet-50 backbone. This framework performs joint spectral occupancy detection with signal-type classification (5G-NR/LTE) directly from compressively sampled spectrograms, reducing sensing load without sacrificing accuracy. Unlike traditional approaches, SCS-DSSS eliminates the need for handcrafted features or prior signal knowledge. Experimental results show a sub-band detection accuracy of 96.85%, with notably low false negative rates, even under 25% compression. The framework demonstrates strong robustness in low-SNR settings and is well-suited for cooperative CRNs in edge-deployed IoT and future 6G systems, enabling scalable and intelligent DSA.