Edinburgh Research Explorer

Joint blind enhancement and passive source localisation of acoustic signals

Project: Funded ProjectResearch

Total award£119,225.00
Funding organisationEPSRC
Funder project referenceEP/H01699/1
Period1/09/0931/08/10

Description

This proposal addressed joint tracking and detection of signals received from a target which admits an acoustic signature. Processing of real-world analogue signals measured using a variety of sensors, such as microphones, and which propagate in multipath or reverberant environments is fundamental to a variety of applications. Within civilian and domestic settings it is important for teleconferencing and hands-free audio enhancement. Within the homeland security it is crucial in a wide variety of fields such as forensics and surveillance. Within the defence sector it is necessary for outdoor sniper detection and localisation.

Any signal radiated in a confined space exhibits multipath propagation due to reflections off surrounding obstacles. However, many existing signal processing technologies fail to explicitly model the multipath response. Consequently, multipath causes significant problems in signal enhancement and separation, signal detection, high-resolution source localisation, and joint detection, classification, and localisation technologies.

Blind multipath equalisation can be improved with accurate modelling of the acoustic distortion which, in turn, depends on knowledge of the target position, thereby requiring target tracking. However, many passive target tracking methods - those which do not actively emit a signature signal, such as active radar and sonar - suffer from the presence of multipath leading to substantial errors in tracking. Target tracking can thus be improved by modelling the effect of, or even equalising, the multipath, thereby allowing identification of the true source from signal reflections. Target tracking and blind multipath equalisation should therefore be solved jointly rather than separately.

This 12-month research programme addressed the detrimental effect of multipath mitigation by developing algorithms for joint blind detection and passive source localisation of acoustic sources. While initially the project focused on enhancing and localising speech sources in an indoor multipath environment for homeland security and domestic applications, the needs of the UK MOD dictated a stronger emphasis on sniper localisation.

The outcomes of this project included:

1) a thorough review of existing sniper localisation research, including an assessment of the accuracy of each technique;

2) an analysis of bullet trajectory and shock-wave geometries, the development of physical models for the measured shock-wave signature waveforms, known as the Whitman N-wave, and the development of a physical model relating the observed signals to the bullet trajectory and N-wave parameters;

3) the development of a sequential Monte Carlo detection algorithm for sniper-fire, and a particle filter based tracker for estimating bullet trajectory, assuming no reverberation.

Our approach in this project differs to existing published work in several important ways:

1) the proposed approach explicitly uses the target-sensor geometry to model the expected observations at the sensor rather than using time-difference of arrival (TDOA) and multi-lateration as in the existing literature; as a result, since the geometry is explicitly modelled, improved accuracy is expected; moreover, our approach provides a more unified and robust modelling technique;

2) the Rao-Blackwellised particle-filter algorithm marginalises parameters thereby reducing the state-space and again improves accuracy, as well as facilitating faster computation time;

3) our approach implicitly incorporates a detection algorithm, as well as localisation, and therefore this joint detection and localisation approach has several advantages over a hybrid approach which detects before localising.

As a result of this 12-month work programme, a conference and journal paper are in preparation entitled "Sniper detection and localisation by joint signal enhancement and localisation" by Christine Evers and James R. Hopgood.

Key findings

The outcomes of this project included:

1) a thorough review of existing sniper localisation research, including an assessment of the accuracy of each technique;

2) an analysis of bullet trajectory and shock-wave geometries, the development of physical models for the measured shock-wave signature waveforms, known as the Whitman N-wave, and the development of a physical model relating the observed signals to the bullet trajectory and N-wave parameters;

3) the development of a sequential Monte Carlo detection algorithm for sniper-fire, and a particle filter based tracker for estimating bullet trajectory, assuming no reverberation.

Plain English Description

This proposal addressed tracking and detection of signals received from a target which admits an acoustic signature. Processing of real-world analogue signals measured using a variety of sensors, such as microphones, and which propagate in multipath or reverberant environments is fundamental to a variety of applications. Within civilian and domestic settings it is important for teleconferencing and hands-free audio enhancement. Within the homeland security it is crucial in a wide variety of fields such as forensics and surveillance. Within the defence sector it is necessary for outdoor sniper detection and localisation.