The goal of the project is to classify different modulations of analog and digital RF signals using a Deep Neural Network. Towards that, a MATLAB machine learning example of Convolutional Neural Network (CNN) was modified and trained using synthetic In phase and Quadrature (IQ) data of eleven different modulation schemes. Then actual over the air RF data collected using a pair of Software Defined Radio (SDR) dongles were tested against the trained network. Next, the network classification accuracy for over the air transmission was improved by retaining the network using a smaller sample of low Signal-to-Noise Ratio (SNR) employing the concept of transfer learning. Applied to the data collected from a separate hardware, the retrained network classified modulation of two commercial radio stations with high accuracies and demonstrated the applicability of the technique to real word signals.
Authors
First Name
Last Name
Ujan
Talukdar
File Count: 1
Leave a comment
Submission Details
Conference URC
Event Interdisciplinary Science and Engineering (ISE)