GPU acceleration of Threat Map computation and application to selection of sonar field controls

Improving Automated Search for Underwater Threats Using Multistatic Sensor Fields by Incorporating Unconfirmed Track Information

“…The Incze and Dasinger method is computationally expensive because of the large number of virtual targets. However, Simakov and Fletcher [7] implement this model on the GPU, allowing for tens of thousands of virtual targets to be used. We use this approach to calculate the threat map in this paper, and Figure 2 shows an example of how this threat map evolves as sonobuoys ping.…”

“…After each sonar transmission we first apply a standard threat map update [7], which accounts for ping-induced reduction of weights of virtual targets. In the simulations discussed in this paper, γ k used in (1) had the following values: γ0 fixed and…”

“…A threat map is used to model the spatial PDF of an existing, but undetected target [1][2][3][4][5][6][7]. This spatial PDF is updated as sonobuoys in the field transmit waveforms and process returns, lowering the probability that an undetected target could exist in areas near the transmitting sonobuoy according to the probability of detection.…”

Non‐myopic sensor scheduling for multistatic sonobuoy fields

“…Greedy scheduling algorithms define a reward for all possible actions a single step ahead and pick the action that maximises this reward. Significant work has since been undertaken to define new metrics for sensor scheduling using greedy algorithms [1,[6][7][8][9][10][11]. Greedy algorithms have the advantage of being computationally efficient, but can have issues effectively prioritising resources in complex multi-target scenarios, especially, as in this case, when conserving the battery power of the sonobuoys is important.…”