Automatic Modulation Classifier Based Spectrum Sensing in the context of Cognitive Radio

Abstract

Cognitive Radio (CR) is one of the technologies that promises to enhance the spectral efficiency, by having secondary users access channel(s) allocated to a primary user but are idle in a given area or time. Spectrum sensing is one of the key functionalities of the CR system that detects channels that are occupied by the primary users so that secondary users could avoid them in-order not to cause harmful interference. Many spectrum sensing methods have been proposed in the literature, each with its own merits and demerits. The main demerits for most of them is, either the need for prior knowledge of the primary user signal, or poor performance in the face of low Signal to Noise Ratio (SNR). The aim of this study was to develop a spectrum sensing method that has the ability to detect the presence of unknown signal in a channel, even under moderately low SNR using an Automatic Modulation Classifier (AMC) based on the Convolution Neural Network (CNN). The logic behind the use of the proposed classifier is the fact that today, every wireless communication system makes use of some form of modulation scheme. Therefore, a detection of a signal with a given modulation scheme would imply the presence of a primary user signal in that channel. If no signal with a known modulation scheme is detected, then it could be concluded that the channel is idle and could be used by secondary users. Using MATLAB, synthetic channel-impaired wave forms were generated for eight digital and three analog modulation schemes. Using the generated wave forms, a six-layer CNN was trained, validated and tested. The trained CNN was then used as an AMC, to classify input signals according to their modulation type. The proposed classifier was able to attain up to 100 % classification accuracy for some of the modulation types with an average accuracy of 95.3% for an SNR of 30dB. The obtained results showed that the proposed classifier was able to recognize modulation type of the input signals and hence identify the occupied and the unoccupied channels. Therefore, it could be used for spectrum sensing in the context of cognitive radio. Keywords: Automatic Modulation Classifier, Cognitive Radio, Convolution Neural Network, Signal to Noise Ratio, Spectrum Sensing.