Work

Hybrid GRU–LSTM–CNN deep learning model for cooperative spectrum sensing in cognitive radio networks

Year: 2026

Type: article

Abstract: This paper presents a reliability-aware cooperative spectrum sensing framework for Cognitive Radio (CR) networks based on a hybrid GRU–LSTM–CNN deep learning architecture. The proposed framework integrates Convolutional Neural Networks (CNNs) for spectro-temporal feature extraction, Long Short-Term Memory (LSTM) networks for long-term temporal dependency learning, and Gated Recurrent Units (GRUs) for efficient short-term sequential modeling. Each Secondary User (SU) generates calibrated soft sensing reports consisting of Primary User (PU) detection probability and predictive uncertainty obtained through Monte Carlo dropout. To improve sensing reliability, an adaptive fusion mechanism assigns reliability weights according to signal quality, confidence level, and uncertainty information. Furthermore, a Straight-Through Estimator (STE)-based quantized reporting strategy is introduced to reduce communication overhead, while an outlier-robust fusion scheme mitigates Spectrum Sensing Data Falsification (SSDF) attacks. Extensive simulations under Rician fading channels with 15% and 25% channel estimation errors, together with validation on the RadioML RML2016.10a dataset, demonstrate the effectiveness of the proposed framework. The proposed method achieves near-perfect detection performance (Pd ≈ 1) at SNR values as low as −4 dB, attains an AUC of 0.87 on real-world RadioML data, maintains stable false-alarm performance even when 30% of SUs are malicious, and reduces sensing-report communication overhead by 89–97% compared with conventional full-precision soft reporting. These results confirm that the proposed framework provides a robust, communication-efficient, and scalable solution for next-generation CR, IoT, and beyond-5G/6G wireless communication systems.

Source: Results in Control and Optimization

DOI: 10.1016/j.rico.2026.100823

Authors: Arun Kumar, Mehedi Masud, Nishant Gaur, Aziz Nanthaamornphong

Institutions: Sikkim Manipal University, Taif University, Sohar University, JECRC University, Prince of Songkla University

Language: English

Topic: Cognitive Radio Networks and Spectrum Sensing

Subfield: Computer Networks and Communications

Field: Computer Science

Domain: Physical Sciences

Open Access status: gold

Estimated APC paid: $1,470

Funder: Taif University