Enhancing TDE-based Drone DoA Estimation with Genetic Algorithms and Zero Cyclic Sum

Abstract

This paper discusses a way to enhance an acousticbased approach to obtaining the direction of arrival (DoA) of a drones ego noise using a microphone array. We focus on obtaining better time delay estimations (TDE) from a set of possible candidates. Recently, a large number of works have been put forward to detect and classify drones with different techniques. However, more investigation is required to tackle the drone DoA estimation problem using the time difference of arrival between pairs of microphones for the case of strongly corrupted audio signals, possibly by noise and multipath. The main problem in a complex acoustic environment is accurately estimating the time difference of arrival. With a traditional approach, this task becomes nearly impossible without the line of sight assumption, that is, whenever the highest cross-correlation peak between signals does not correspond to the delay between them. This paper uses genetic algorithms to search for the correct delays between pairs of microphones among a set of possible delays (primary and secondary delays). We define a fitness function based on the concept of zero cyclic sum of closed loops, i.e., when forming a closed loop, the sum of all theoretical delays should equal zero. A drawback of closed loops is that incorrect delays may result in a zero-sum; we thus created a fitness function that considers all possible closed loops of a given array. We exploited different approaches to estimate the direction of arrival using the combination of genetic algorithms and zero cyclic sum. In our experiments, the method successfully found all correct delays in simulations, providing strong evidence of its effectiveness when a correct delay exists among multiple possible delays. Furthermore, in experimental trials, it significantly enhanced the number of correct delays detected, further validating its utility and potential in practical scenarios.