Sensor array post-filter for tracking spatial distributions of signals and noise

A method for processing a beamformer output to attenuate unwanted signals and noise, comprising steps for:

• applying a beamformer to frequency domain transform coefficients derived from each channel of a sensor array to produce a beamformer signal corresponding to a target channel;
• computing inter-sensor phase differences representing instantaneous direction of arrival (IDOA) vectors for each of a plurality of non-repeating sensor pairs from the frequency domain transform coefficients;
• using a set of pre-computed IDOA prior distributions to estimate a probabilistic direction from which each IDOA vector originated;
• constructing a probabilistic model from the IDOA vectors, the probabilistic directions, and the beamformer signal;
• estimating signal and noise variances from the probabilistic model;
• constructing an adaptive post-filter from the estimated signal and noise variances and the beamformer output signal;
• and
• applying the adaptive post-filter to the beamformer signal to generate a frequency domain output signal exhibiting attenuation of signals originating from non-target directions and attenuation of noise from all directions.

A system for modeling a working space of sensor array, comprising using a computing device to:

• divide a spatial region corresponding to a working space of the sensor array into a plurality of regions;
• generating a plurality of training data elements for each of a plurality of spatial points within each region, each training data element representing a signal contaminated by some amount of correlated and non-correlated noise;
• converting the training data elements to instantaneous direction of arrival (IDOA) vectors corresponding to inter-sensor phase differences for each of a plurality of non-repeating sensor pairs;
• computing means and covariances of the training IDOA vectors and setting a prior probability of each location or region, to generate the IDOA prior distributions;
• and
• using the IDOA prior distributions to model the working space of the sensor array, said IDOA prior distributions being stored on a computer readable medium for use in processing signals captured by the sensor array.