Large Array Beamforming using Graph Signal Processing
Abstract
There are many modern systems which advantageously combine signals from many sensors to improve resolution and noise suppression. Principles and algorithms for combining these signals, for example for passive acoustic or passive radio-frequency systems, should, in principle, offer improved performance as the number of sensors increase. Typically, these systems need to track and enhance signals in real time and with low compute latency, which limits this number of sensors and hence overall performance. As systems with thousands or more sensors become available, how can the models of locality continue to be advanced, perhaps dramatically, to allow performance to improve? An example of a multi-sensor system is a minimum variance distortionless response (MVDR) beamformer. Such a system requires an estimate of full covariance, that is the covariance of all possible pairs of sensors, resulting in an ???? × ???? matrix of parameters, where N is the number of sensors. This number of parameters results in at least an ????(????2) growth in computation as the size of the sensor array grows. (Even faster growth when unconstrained matrix inversion ids needed.) Many modern efforts for large-sized arrays reduces this intractable number of parameters by, for example, using diagonal covariance, truncation to leave only large eigenvalues, and/or sparseness assumptions where only the largest magnitude covariance values are used. Are there other formal method, not yet investigated for beamforming and direction-ofarrival estimation, which can greatly broaden the ability to improve local computation, reduce delays, and/or improve accuracy for direction of arrival estimation and beamforming for processing gain from large and typically irregular arrays of sonar and radio-frequency sensors? Our likely new answers to this question come from the emerging field of signal processing on graphs.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 20, 2018
- Source ID
- N000141812109
Entities
People
- Les Atlas
Organizations
- Office of Naval Research
- United States Navy
- University of Washington