Blind Symbol Rate Estimation using Convolutional Neural Networks for PSK Modulated Signals
Jeewani Kodithuwakkuge, Vaibhav Sekhar, Tharaka A. Lamahewa · 2021
The problem of symbol rate estimation in Phase Shift Keying (PSK) modulated signals is addressed. We show that machine learning based techniques can be used to estimate the symbol rate blindly from the baseband received signal. A Convolutional Neural Network was trained to obtain a fast and coarse estimate of the symbol rate compared to classical symbol rate estimators. This is beneficial for real time applications such as situational awareness where a large bandwidth needs to be analysed quickly. The network was trained using an input feature vector derived based on cyclostationarity theory. This feature vector is used in blind signal detectors and symbol rate can be estimated by observing the peak value of the feature vector. We show that symbol rate estimation accuracy can be significantly improved by using deep learning techniques, specially at low signal to noise ratios compared to classical signal processing techniques. The proposed method does not require prior knowledge of the PSK modulation order or pulse shaping filter and robust against time, phase and small frequency offsets.