Blind Deinterleaving of Signals in Time Series with Self-attention Based Soft Min-cost Flow Learning

TL;DR

This paper introduces a novel end-to-end learning framework that uses self-attention neural networks to deinterleave and cluster signals in time series data, specifically radar signals, by approximating min-cost flow optimization.

Contribution

It formulates signal deinterleaving as a bi-level optimization problem and proposes a trainable self-attention based approach to learn costs for clustering in time series.

Findings

Effective clustering of radar signals demonstrated

Outperforms existing methods in challenging scenarios

Efficient and scalable solution for signal deinterleaving

Abstract

We propose an end-to-end learning approach to address deinterleaving of patterns in time series, in particular, radar signals. We link signal clustering problem to min-cost flow as an equivalent problem once the proper costs exist. We formulate a bi-level optimization problem involving min-cost flow as a sub-problem to learn such costs from the supervised training data. We then approximate the lower level optimization problem by self-attention based neural networks and provide a trainable framework that clusters the patterns in the input as the distinct flows. We evaluate our method with extensive experiments on a large dataset with several challenging scenarios to show the efficiency.