ConvLSTM based Spectrum Sensing Scheme for Cognitive Radio

Abstract

Spectrum sensing is the key component for Cognitive Radio as it helps in finding the spectrum holes present on the spectrum. Now a days, Deep learning models shows promising result in finding the spectrum holes specifically CNN and LSTM network

 A deep learning model with an architecture framework shows remarkable results in recent years. It is made up of numerous neural layers that represent data at different abstract levels has the ability to learn large signal data semantically at a high level, which can be a possible solution to all such problems. In this paper, the ConvLSTM network has been implemented with 64 filters and 128 cells per convolutional layer for spectrum sensing on a 2, 80,500 sample size dataset. The proposed model in the given scenario with Adam optimization shows the best performance for spectrum sensing when compared based on probability of detection (P_d), probability of false alarm (P_f), probability of miss detection (P_miss), and Signal-to-Noise ratio (SNR) with conventional techniques of spectrum sensing such as energy detection and eigen value detection.