Pulse Separation Using Time-Frequency Mask and Machine Learning

This paper intends to demonstrate the use of machine learning to deinterleave pulse trains in radar by implementing the cocktail party problem. The cocktail party problem is a phenomenon when a person is selectively listening to a specific conversation from a mix of conversations at a cocktail party. In our case, instead of a mix of conversations, it's a mix of pulse trains that a radar is trying to separate. The radar could use machine learning to separate the intercepted pulse train using a deep learning neural network. The pulse trains are converted to a time-frequency representation to create an ideal time-frequency mask and isolate the desired pulse trains. The mixed pulse train and the time-frequency mask is used to train the neural network to estimate the desire pulse trains from mixed pulse train. This is a process of using a time-frequency mask with a neural network to deinterleave pulse trains in radar.