Sensor Staggering in Multi-Sensor Target Tracking Systems Ruixin Niu, Pramod Varshney, Kishan Mehrotra and Chilukuri Mohan Department of Electrical Engineering and Computer Science 121 Link Hall Syracuse University Syracuse, NY 13244 USA Email: varshney@ecs.syr.edu Abstract — For a multi-sensor target tracking sys- tem in a cluttered environment, the effects of tem- porally staggered sensors on system performance are investigated and compared with those of synchronous sensors. Aprobabilisticdataassociationfilter(PDAF) is used to track the target. Measurements from lo- cal sensors are fused in a centralized manner for the system with synchronous sensors. The system perfor- mances are compared both in terms of track life and in-track percentage. For a wide variety of scenarios, simulation results show the superiority of temporally staggered sensors over synchronous sensors. I. Introduction Multi-sensor target tracking systems have generated in- tensive interest because of their enhanced estimation per- formance, surveillance coverage and robustness. Different kinds of multi-sensor tracking system architectures, are introduced and compared in [2] and [8]. With regard to the temporal issues of the system, there exists the so called “out of sequence measurement” (OOSM) problem. In this scenario, measurements from different sensors are collected at the same time, mean- ing they have the same time stamp. However, due to different transmission delays in the sensor network, it is possible that a measurement with time stamp τ arrives af- ter the target state estimation has been updated to time t>τ . Many authors have investigated this problem and proposed methods to deal with the OOSM[3, 4, 6, 7]. In an OOSM problem, the sensors are still assumed to operate synchronously and only the measurements arrive asynchronously. Much less attention is paid to the sys- tems with asynchronous sensors [5, 10], because the esti- mation accuracy at each sensor’s sampling time is worse than the estimation accuracy at the sampling time of the system with synchronous sensors. In spite of this dis- advantage, however, a system with temporally staggered sensors has much lower maximum prediction errorbecause it is more frequently updated with new measurements. In our previous work [9], the effects of asynchronous sensors, or temporally staggered sensors were investi- gated and compared with the traditional system with syn- chronous sensors under the assumption that neither false alarms nor missed detections exist. To make a fair com- parison, a new metric of performance, the average esti- mation error variance (AEV), was used. Through both analytical and numerical methods, we obtained extensive results on how to optimally stagger sensors over time for different scenarios. In this paper, we extend our research to more re- alistic cases, where there always exists the problem of measurement-origin uncertainty caused by false alarms, clutters and missed detections. In such cases, we sim- ply do not know which measurement, if any, is from the target. To deal with this problem, many algorithms have been developed, such as the nearest neighbor standard fil- ter (NNSF), the strongest neighbor standard filter (SNSF) (if the signal intensity information is available), and the probabilistic data association filter (PDAF) [2]. At each time step, they all need prior information, i.e., the pre- diction from the last time step, to process the new mea- surements. If the prediction is not accurate enough, the system will lose track of the target quickly. Thus, it is imperative that the maximum value of the prediction er- ror be kept to a minimum. In this paper, we propose to use sensor measurement staggering as a means to keep the maximum prediction error under control. This will help maintain tracks over a longer period. In Section II, we introduce the measurement collection model and the target dynamic model. In Section III, the realistic environment with false alarms and missed detec- tions is discussed. In addition, we describe briefly the probabilistic data associate filter (PDAF) which is chosen to track in a cluttered environment. A criterion to judge whether a target is in track, is proposed. In Section IV, for two sensors with the same performance, we compare the performance of the system with uniformally staggered sensors to that with synchronous sensors. In Section V, the case where two sensors have different performances is studied, and the best sensor staggering schemes are obtained for various system parameters via simulation. Some concluding remarks are made in Section VI.