We have multiple arrays of microphones, each array well seperated from others on the ground. Our task is to track an aircraft that happens to pass by. But the terrain is such that there is a lot of reverberation. Now:
1) What prefiltering technique is best to remove reverberation effects.
2) Which weightings during GCC(Generalised Cross Correlation) are best against reverberation, while at the same time without increasing error by much.
Thank you very much.
Lay Jain,
Massachusetts Institute of Technology
If all you want to do is detect, then reverb is not a problem. It also tells you when an aircraft is present or not. You could detect entirely on reverb for instance. So it seems that you also want to localize and track the aircraft.
The best method will depend on:
Perhaps the best generic method to localize and track amid reverb is Matched Field Processing (MFP) used in sonar.
First you need a computationally fast, high-fidelity model of aircraft sound generation and propagation in your environment, to model the sound from the aircraft to your arrays, including the terrain reverberation.
Then MFP is a global optimization problem: search aircraft type and behaviour space to find the type and behaviour that best matches the signal you observe across all of your arrays. You can also throw in some unknown environmental parameters to improve matters.
The objective function that measures the quality of the fit is very important. Some functions support localization and tracking, some ruin it. You need a coherent objective function to capitalize on reverb, using it to for some information about aircraft location.
MFP is very computationally intensive; often prohibitively so in practice, especially if you need near real time tracking. Forget about real time.
Of course in passive contact localization and tracking amid reverberation there is no guarantee that the sound field, or more particularly, the sound field as sampled by your arrays, holds enough information to answer the questions you ask of it, such as “where is the aircraft at the moment you hear it?”. The information is often ambiguous. The aircraft could be in various places. Observing longer adds more information, generally improving the localization. Observing longer is often OK when looking for distant submarines because you’ll have lots of time to react. But it may not be OK when looking for aircraft that are already upon you.
Maybe "reverberation" might be considered simply as "multipath" for aircraft noise in outdoor environments, and in that case you might solve it explicitly e.g using adaptative beamforming algorithms. Acoustic tracking of low flying helos is a conventional problem solved satisfactorily from many years in open field, I know some systems based on likelihood maximisation, it take a fraction of a second to get iteratively the most likely position, then your tracking algorithm will have to reject aberrant dots. A key issue is now far your aircraft noise and course is stationary or not, as rightly commented by Ronald! Of course if you deal with supersonic low flying jets, you will find the Mach wave direction, not directly the plane heading... [An issue solved in acoustic sniper detection systems by tabulating the speed (hence the Mach cone angle) for a library of ammunitions and associated signatures]
Spectral Subtraction combined with Beamforming should help. Check this out! https://www.researchgate.net/profile/Deepak_Gala2/publication/307930113/inline/jsViewer/5984aeb90f7e9b6c852f4d67
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing