Analysis of an Open Channel Identifier using Stochastic Gradient Descent and GNU Radio

Abstract

In this paper, we address the problem of radio spectrum crowding by using a stochastic gradient descent neural network algorithm on simulated cognitive radio data to identify open and closed channels within a specified RF range. We used GNU Radio 3.8 flowgraphs to simulate cognitive radio data for standard U.S. Wi-Fi channels, and to design both the neural network and classical power estimation algorithms. Our methods include the possibility for leveraged use in many spectrum sensing applications such as channel detection, modulation classification, and radio fingerprinting. We provide analytical insight into the performance of this neural network that goes beyond that of previous work in this immediate field. These analyses will show the stochastic gradient descent algorithm achieves an advantageous accuracy over the traditional channel occupation algorithm.