A nearest neighbor algorithm for fast matched-field processing with a vertical line array

Selection of replica fields that are most like the data, i.e., the nearest neighbors (NNs) to the data, offers a way of reducing the computational search space in matched-field processing, thereby making larger physical search spaces or a larger number of frequencies practical. To enable selection of NNs a vector basis for the search space is required. The authors use the large eigenvectors of the covariance matrix for uncorrelated sources spread over the search region. This is not only a suitable vector basis of the search space, but also results in a dimensional reduction from the full set of eigenvectors, with a further computational saving. The replica vectors for the search region are partitioned by finding their projection on this vector basis. One can then select for matching only those replicas with similar squared projections on the vector basis. This selection process carries a modest cost in computing overhead, provided that the code, the partitioning, and the replica selection parameters are optimized. The detection performance and false alarm probability for the Bartlett beamformer, with and without selection of the replicas, were estimated from simulations of noisy data received on a vertical line array at practical time-bandwidth products. An order of magnitude speedup was obtained