I'm working on detecting signals buried under the sensor noise using a multiple sensor system and using some technique (correlation?) to take what is common to all the sensors and cancel out the random, uncorrelated part of the sensors. The signal I want to recover is some 30 times smaller than the rms noise. The smaller I could recover the better, though. can anyone point to any technique for this case?


Your ability to achieve coherent gain is going to depend on a number of factors and your upper bound on gain is 10log10(N)dB, where N is the number of sensors.

The essential idea is that you need to align your signals in time and add them resulting in a superposition. The signals need to be nearly identical at each sensor output.

In beamforming, there is a notion of signal source located at some distant point and plane waves are incident on the sensors. One can infer the anticipated time delays from geometry. One can also assume other wavefront geometries such as spherical wave incidence.

Effects such as polarization and sensor orientation complicate the similarity of the received signal at each sensor.

Some signals have multiple propagation paths and these paths aren’t all going to be free space paths. Seismic signals have pressure and sheer wave propagation.

The assumption that noise is independent at each sensor isn’t a given for every application.

These are all general considerations and you need to offer more detail on your problem if you want more help. The 10log10(N) upper bound should apply in any case.

  • $\begingroup$ Thanks Stanley, N above is the number of sensors, I guess. It is for a seismic application. Sensors will be very close so I assume vibration will be the same for all of them. Measures will also be simultaneous. $\endgroup$ – Joan Oct 28 '17 at 7:21
  • $\begingroup$ N is the number of sensors. $\endgroup$ – user28715 Oct 28 '17 at 14:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.