Performing Hybrid Spectrum Sensing with an Adaptive and Attentive Multi-stacked Deep Learning Network in a Cognitive Radio Network

Cognitive Radio Network (CRN) includes Secondary Users (SUs) and Primary Users (PUs)
to perform better communication. The SUs present in the CRN observe the spectrum band to obtain
the white space opportunistically. Employing the white spaces supports enriches the effectiveness of
the spectrum. Due to the promising learning capacity of Deep Learning (DL) and Machine Learning
(ML) models, various experiments in the previous years have utilised the deep or shallow multi-layer
perceptron mechanism. However, these mechanisms do not apply to the time series data because of the
memory element’s absence. One of the primary issues in spectrum sensing is to model the test statistic.
Conventional mechanisms normally employ the model-aided attributes as a test statistic, including
eigenvalues and energies. However, these attributes cannot be precisely characterised in the real world.
Hence, a DL-assisted hybrid spectrum sensing technique in the CRN is implemented. At first, the data
are gathered from appropriate databases. Further, an Adaptive and Attentive Multi-stacked Network
(AAMNet) is developed for the hybrid spectrum sensing process. The AAMNet is developed by com￾bining three different deep networks such as Convolutional Neural Network (CNN), Long Short-Term
Memory (LSTM), and autoencoder. The spectrum sensing process by the proposed AAMNet is enhanced
further using the Random parameter Improved Duck Swarm Algorithm (RIDSA) for parameter opti￾misation. The availability of spectrum is identified for better spectrum utilisation with the help of the
developed hybrid spectrum sensing process. Throughout, the analysis of the proposed method is checked
by evaluating the resultant outcomes with various heuristic approaches and deep learning methods