A Hybrid Deep Learning Model for Efficient Spectrum Sensing in Cognitive Radio

Abstract

The massive developments of gaming devices as well as mobile apps have increased the demand of bandwidth. In wireless communication, cognitive radio (CR) technique has shown a significant amount of improvement in optimal utilization of spectrum band.  Spectrum sensing is a very important strategy through which new scope for spectrum sharing can be detected. In CR, spectrum sensing is crucial. To estimate spectrum sensing performance, two metrics are mostly considered, i.e., probability of detection (PD) and probability of false alarm (PFA).  Traditional sensing approaches often face problems due to PD and PFA. These two constraints can affect spectrum utilization. Spectrum sensing is a type of binary classification problem and researches have shown that neural network has achieved high accuracy in this aspect. This research work focuses on the utilization of deep neural network (DNN) for accurately sensing unoccupied spectrum band. The benefits of convolutional neural networks (CNN) and recurrent neural networks (RNN) are combined in our proposed hybrid model i.e. ResNet-LSTM. In our study, RadioML2016.10b dataset is used for the experiments. The results showed that proposed model found to be efficient when compared to the existing techniques such as CNN, ResNet, LeNet, LSTM, CLDNN. Further, while compared with earlier models like CNN-LSTM, DetectNet and DLSenseNet, the proposed hybrid model “ResNet-LSTM” has shown better spectrum sensing performance. Proposed ResNet-LSTM framework achieved 96.97% prediction accuracy with 96.52% precision and 96.83% recall. The prediction time reduced by 0.14 msec than CNN-LSTM model.