Machine Learning Frequency Hop Sequence Recovery

Problem:
Frequency hopping is widely used to protect radio communications from jamming; however, it requires precise synchronization of hop sequence and timing between sender and receiver. This makes such systems less adaptable to dynamic electronic warfare (EW) environments. Additionally, radios can lose synchronization even in non-jamming conditions, leading to communication failure.

Background:
Recent research demonstrates that machine learning approaches can recover frequency hopping sequences from observed signals. While these methods show promise, they have not yet been translated into practical, real-time communication systems. Successful recovery of hop sequences could also address timing desynchronization issues.

Objective:
This project aims to integrate a deep learning model with a radio transceiver to recover frequency hop sequences in real time. An extended objective is to detect electromagnetic interference (EMI) and dynamically adapt the hop sequence to avoid jammed frequency bands.

Approach:
The project will proceed along two parallel tracks. First, a machine learning model will be developed to recover frequency hopping sequences from simulated data. Signal data will be transformed using Short-Time Fourier Transform (STFT) and used as input for training and testing various deep learning architectures informed by prior literature.
Second, the validated model will be integrated with a radio transceiver capable of transmitting and receiving frequency hopping signals. Experimental testing will evaluate system performance under synchronization errors and interference conditions. Operation within amateur radio bands will require appropriate FCC licensing.