Enhancing model based acoustic localisation using quantum annealing

In model based acoustic localisation (MBAL) the locations of underwater objects are estimated by comparing sensor measurements with model predictions. To obtain high quality predictions, computationally demanding propagation models need to be run for a large set of environmental parameters. The computational resources that are available onboard are typically not sufficient to perform accurate MBAL estimations on real-time data. In this work, we develop a quantum algorithm that uses quantum annealing to enhance underwater acoustic localisation. A restricted Boltzmann machine (RBM) is trained to predict the probability distribution of the candidate location of an underwater target. Advantage of this approach is that part of the computation can be moved to offline-training. Moreover, once the model is trained, the probability distribution can be sampled more efficiently using a quantum annealer. Potentially, this could enable real-time accurate target estimations to be made onboard. The RBM is applied to a simplified multi-sensor horizontal localisation problem where we assume a constant and linear acoustic propagation. Using simulated annealing we show that the RBM is able to learn probability distributions that resemble target locations. First results show that the training and sampling can be done using quantum annealing hardware by D-Wave Systems for a limited size example. Our contribution is the first work that explores how quantum algorithms can be applied for more efficient information processing for acoustic underwater localisation