Implementation of a Machine Learning based OFDM detector using GNU Radio

Software-Defined Radios (SDRs) are powerful and versatile tools that can implement various communication systems in the software domain. The user can redefine the system by modifying the code. GNU Radio is an open-source SDR that uses a block-based graphical interface with a Python backend framework. Machine Learning (ML) technologies are powerful tools that can bolster the development of the next generation of wireless communication systems. Orthogonal frequency-division multiplexing (OFDM) is a modulation scheme used in modern communication systems such as 4G/5G, WiMax, etc. This study uses the powerful ML ecosystem in Python to model conventional OFDM receivers using the GNU Radio ecosystem. The work trains and compares Deep Neural Networks as well as Linear Regression algorithms (using PHY-layer data) to model the receiver. The trained models improve upon the Bit Error Rate (BER) of conventional systems by a factor of 10
^-2
.