Deep-Learning Channel Estimation for OFDM Systems

In OFDM (the modulation of Wi-Fi, LTE and 5G) the receiver must know the radio channel on every subcarrier before it can recover the data, but it only measures the channel on a few pilot tones. This project estimates the channel of 64-subcarrier OFDM symbols from 8 comb pilots and compares the classical estimators - least squares (LS) with linear, cubic-spline or DFT interpolation and linear minimum mean-square error (LMMSE) - with a fully connected deep neural network (DNN) and a 1-D convolutional denoiser. It measures the mean-square error against the true channel and the 16-QAM bit error rate from 0 to 30 dB SNR, on simulated 3GPP multipath channels and on real ray-traced DeepMIMO campus channels. A web application estimates uploaded GNU Radio captures, GNU Octave re-checks the computation, and a GNU Radio flowgraph and a complete MATLAB version are included.

Category: Electronics and Communication Engineering (Wireless Communication, 5G / LTE Physical Layer, OFDM, Signal Processing, Deep Learning, GNU Octave, GNU Radio, MATLAB, Python)

Electronics & Communication Engineering Project E02 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.