Weak sense Lp error bounds for leader-node distributed particle filters

The leader node particle filter is a partially distributed approach to tracking in a sensor network, in which the node performing the particle filter computations (the leader node) changes over time. The primary advantage is that the position of the leader node can follow the target, improving the efficiency of data collection. When the leader node changes, the particle filter must be communicated to the new leader. Exchanging a complete representation can require many thousands of bits, so the filtering distribution is more coarsely approximated, either by transmitting only a subset of particles or by training a parametric model. The concern is that this approximation could lead to instability and eventually tracking failure. The major contribution of this paper is the development of bounds on the error of the leader node particle filter. In contrast to previous bounds, which grow exponentially over time, the bounds indicate convergent behaviour.