A system for automatic classification of aircraft flyovers using acoustic data

Abstract

An overview of a system for the automatic classification of aircraft from flyover data is presented. The system is passive, utilising acoustic sensors to measure both broadband and narrowband energy. Aspects of the system architecture, sensor design and signal processing are covered. The processing is divided into three streams: broadband, narrowband and cepstrum. Each processing stream is capable of extracting flight parameter estimates from the acoustic data. Broadband estimation is based on the time-delay cross correlation of signals from multiple sensors. Narrowband estimation makes use of the spectrogram of the data to extract frequency lines produced by the aircraft and subject to the acoustical Doppler effect. Cepstrum processing tracks the primary rahmonic in the cepstrogram due to multipath interference. A novel hidden Markov model tracking technique is applied to form tracks on the noisy spectrogram and cepstrogram data. Examples of real data processing and flight parameter estimates for classification are given.