1. Implementing Interval Algebra to Schedule Mechanically Scanned Multistatic Radars Richard W Focke (CSIR & UCT) Leon O Wabeke (CSIR) J Pieter de Villiers (CSIR & UP) Michael R Inggs (UCT) 14 th International Conference on Information Fusion Hyatt Regency Chicago, Chicago, Illinois, USA 5 th to the 8 th July 2011

2. System Context: Border Control & Surveillance Why? - Terrorism, piracy & poaching (small targets) - Trafficking of illegal substances / people How? - Limited reaction forces (coast guard, border police, military) - Many multisensor platforms (radar, cameras, EWR) - Need to co-ordinate the actions (sensors & forces) Advantages - Force multiplication for greater border protection - One complete picture for the commander © CSIR 2011 www.csir.co.za

3. System Context: Border Control & Surveillance © CSIR 2011 www.csir.co.za

4. Background: Resource Scheduling © CSIR 2011 www.csir.co.za

5. Problem Statement Scheduling of mechanically scanned multistatic radars - Each radar has a fixed timeline of tracking tasks - Assumed detection happens between updating tracks - Mechanical inertia places constraints on scheduling - Need to decide which measurements should coincide Advantages of measurements coinciding: - Radar cross section diversity: better detection & radar imaging - Radar clutter diversity: better signal to noise ratio - Measurement fusion gain: monostatic & multistatic measurements - Track fusion gain: more measurements & covariance ellipse overlap © CSIR 2011 www.csir.co.za

6. Experiment Setup: Simulations © CSIR 2011 www.csir.co.za

7. Experiment Setup: Simulations © CSIR 2011 www.csir.co.za

8. Experiment Setup: Simulations © CSIR 2011 www.csir.co.za

9. Experiment Setup: Simulations © CSIR 2011 www.csir.co.za

10. Experiment Setup: Simulations © CSIR 2011 www.csir.co.za

11. Experiment Setup: Simulations © CSIR 2011 www.csir.co.za

12. Prioritizing Tasks © CSIR 2011 www.csir.co.za

13. Heuristic Sensor Scheduling Algorithm Generation of dual visited target sequence: - Iterate through target list of 1 st radar, select a target of interest - Find target of interest in target list of 2 nd radar - Target added to prefix, repeat using remainder of sequences - Stop when no more targets are common to subsequences - Enumerates all possible sequences Selection of final sequence: - Select from highest priority targets (randomly if more than one) - Discount all sequences that do not contain this target - Final scheduled sequence is last remaining sequence © CSIR 2011 www.csir.co.za

14. Allen’s Interval Algebra (IA) [1] Two algebras defined: - Temporal constraints - Duration constraints Operators: - 13 temporal operators - 3 duration operators Algorithms to maintain consistency: - Constraint propagation (local) - Path consistency (global) Extensions available (fuzzy, probabilistic) © CSIR 2011 www.csir.co.za [1] J. F. Allen, “Maintaining knowledge about temporal intervals,” Communications of the ACM, vol. 26, no. 11, pp. 832-843, 1983.

15. Temporal IA © CSIR 2011 www.csir.co.za

16. Duration IA Make up of duration algebra: - No definitive algorithms given in his paper - Operators: shorter, longer and equals. - Shorter / longer has a multiplication factor - Interaction between temporal and duration algebras Potentially enables: - Dwell time requirements (detection & tracking) - Fixed duration scans (deterministic) © CSIR 2011 www.csir.co.za

17. IA Constraint Propagation

18. IA Path Consistency © CSIR 2011 www.csir.co.za

19. IA Sensor Scheduling Algorithm IA setup per radar: - One interval defined per tracking task - Each interval related to other intervals no ambiguity use the before or after operator ambiguous use all operators Determining dual visitations: - Select from highest priority tasks (randomly if more than one) - Equals operator set for that target interval on all radars - Network remains consistent schedule the visitation © CSIR 2011 www.csir.co.za

20. Azimuth Ambiguities in Target Sequences © CSIR 2011 www.csir.co.za

21. Algorithm Differences Interval Algebra scheduling algorithm - Performs local optima search - Handles azimuth ambiguities - No changes or extra processing to handle ambiguities Heuristic scheduling algorithm - Performs exhaustive global optima search - Does not cater for ambiguities (selects a random order) - Requires changes and much more procession to handle ambiguities © CSIR 2011 www.csir.co.za

22. Results © CSIR 2011 www.csir.co.za

23. Conclusions Catering for ambiguities more important than search type - IA: Local optima search with ambiguities - Heuristic: Global search without ambiguities IA is an elegant solution for handling time constraints - Allows for imprecise knowledge to be captured © CSIR 2011 www.csir.co.za

24. Future Work Computational complexity comparison - Change heuristic to perform local search (all enumerations) Performance comparison - Compare to Greedy Randomised Adaptive Search Procedure (GRASP) Investigate parallelization of IA scheduling algorithm More advanced sensor scheduling problems - Schedule turn-around events for each radar independently - Tasks requiring radar dwells (scan, track & image) Generalize IA to task organizing algorithm - Investigate applicability of IA to AESA radar scheduling © CSIR 2011 www.csir.co.za

25. Thank you Questions?