An efficient spectrum sensing model using adaptive residual gated recurrent unit in cognitive radio system

One of the significant technologies is spectrum sensing that is employed to resolve the limited accessibility of the spectrum resource. Spectral signal validation acts as a significant part in forecasting and characterizing the occupancy conditions of dynamic frequency platforms. To improve the spectrum usage, the technology of Cognitive Radio (CR) is suggested. This research addresses the concerns in spectrum sensing in CR by developing a heuristic-based deep learning model. At first, the signal data collected from the network is subjected to feature extraction based on the developed Ensemble Autoencoder with Multihead Coordinate Attention Fusion Network (EA-MCAFNet). The developed EA-MCAFNet model includes rigorous characteristics to operate at noisy conditions, as it combines the ability of the Autoencoder and MCAFNet. The developed EA-MCAFNet identifies both the spatio-temporal features from the signal data, and these extracted features are sent to the developed Adaptive Residual Gated Recurrent Unit (ARGRU) for spectrum state detection. Here, the layers within GRU are set for parameter tuning based on the newly implemented Movement-updated Revolution Optimization (MRO). The suggested network is effective and has high reliability to operate in low Signal-to-Noise Ratio (SNR) conditions, and successfully achieves high spectrum state detection accuracy. Experiments follow a comparative study against the well-defined methods, and the results are achieved on standard data sources to ensure the efficacy of the approach.