Underwater acoustic localization by probabilistic fingerprinting in eigenspace

In this paper, the underwater acoustic localization is given by probabilistic fingerprinting in eigenspace. The eigenspace of this study means the projection of PCA (principal components analyses). The goal is to predict the receiver location through wireless acoustic communication signals in underwater environments. It should be emphasized that our underwater localization is performed from wireless acoustic communication signals, but not from commercial localization systems. In other words, the hardware can be utilized for both communication and localization simultaneously in our experiments. Our underwater localization scheme is based on the fingerprinting of wireless acoustic communication signals in eigenspace of PCA (principal components analyses). It is based on fingerprinting and contains two stages, i.e., the off-line (i.e., training) and on-line (i.e., predicting) stages. In the off-line stage, there are some reference locations. At each reference location, acoustic communication signals at different frequencies are collected and sampled at discrete time points to constitute an acoustic-signal map. In the on-line (predicting) stage, acoustic communication signals at the unknown location are collected to constitute a signal vector. The problem becomes to predict the coordinate of the unknown location by comparing the signal vector with existing acoustic-signal maps. To reduce the complexity of acoustic-signal maps and overcome the severe fluctuation of measured data, all received signals are projected onto the eigenspace of PCA. Each component of the feature vector in eigenspace is assumed to be random Gaussian distribution. In addition, the components of the feature vector are assumed to be independent. The final probability that the signal vector occurred at an arbitrary reference location becomes the product of different Gaussian distribution functions. Such a probability is viewed as the weight for such a reference location. The unknown location can be a- pproximated by the weighted summation of different reference locations.