An Efficient Layer 2 Mesh Communications Protocol for

An Efficient Layer 2 Mesh Communications Protocol for Space Sensor Networks Loren Clare, Jay Gao, Esther Jennings, and Clayton Okino Jet Propulsion Laboratory, California Institute of Technology Presented at Space Internet Workshop Hanover, Maryland 8 -10 June 2004

Outline • • The need for multi-spacecraft sensing Distributed spacecraft mission types Why network? Networking solution approach, described through an example • Extension to Demand-Driven traffic scheduling • Conclusions 2

Multi-Spacecraft Sensing Missions Many phenomena can only be measured using multipoint sensing: – multiple sensors that are – spread over a spatial regime of interest and – simultaneously measure the target phenomena The need for multipoint (multi-spacecraft) sensing has long been recognized – Space Science Board of the NAS in 1974 for large-scale “geospace” phenomena (“space weather”) • Interplanetary Monitoring Platform (IMP-7 and IMP-8) s/c launched in early 70 s • International Sun-Earth Explorer (ISEE); 3 spacecraft; late 70 s – “able to break the space-time ambiguity inevitably associated with measurements by a single spacecraft on thin boundaries which may be in motion, such as the bow shock and the magnetopause. ” • Dynamics Explorer (DE); 2 spacecraft; launched 1981 • Many subsequent missions (GEOTAIL, WIND, INTERBALL, SOHO, POLAR, Cluster, …) – Space Studies Board (NRC) decadal strategy August 2002: 7 of 9 recommended moderate-class programs are multi-spacecraft – 2003 SSE Strategy: “Constellation technology must be developed to permit collecting data efficiently and simultaneously at dispersed locations” – “Sensor Web” concept is critical component of Earth Science strategic plan 3

Multipoint Sensing Classes Multipoint sensing applications fall into 3 classes: Pixellation/Voxellation of space Beamformation Tomography/Rendering Each class has associated data collection and processing needs for combining the multiple sensor signals => different traffic models 4

Additional Reasons for Distributed Sensing • Coverage of large (possibly sculpted) area via union of many spatially dispersed sensors • Incremental sizing (evolution/extension, replenishment) • In situ sensing: mitigates sensor range limitations and overcome ambient environmental noise • Speed through parallel actions • Fault tolerance • Mix multiple sensor modalities at appropriate densities 5

Why Use a Communications Network? Why not just store data and dump at perigee? Incorporating intersatellite links and networking enables: • Access to any/all spacecraft in the multi-spacecraft mission is continuously provided via single ground contact with any spacecraft – Increases ground operations efficiency – Enables automated operation of the whole “act as a single mission spacecraft for coordinated observations” • Real-time coordinated observations and processing – Alert/cue ground-based assets (e. g. , gamma ray bursts) • E. g. , on March 29, 2003 the High-Energy Transient Explorer (HETE) detected a gamma burst and cued the European Southern Observatory's Very Large Telescope, which confirmed a correlated supernova explosion (http: //www. gsfc. nasa. gov/topstory/2003/0618 rosettaburst. html); Gamma Ray Burst Coordinate Distribution Network: 10 -20 second latency – Event-based interactions among distributed sensor spacecraft • cueing, data aggregation (compression), fusion (improves resource use) • Autonomous cooperative processes among distributed spacecraft – precision navigation; constellation control and reconfiguration – network time synchronization for precise time-stamping of sensor data 6

What If No Crosslinks? Suppose there are no crosslinks. Data is stored onboard and each s/c dumps its data to Earth when it is near perigee. Data delivery latency is therefore approximately equal to the orbital period of the spacecraft. For example, for the Mag. Con mission, worst case is Note that storage requirements are substantial, in addition to age of data. 7

Uniqueness of Space-Based Sensor Networks Differences from conventional networks: • Nodes are moving, although deterministically – Unlike typical sensor networks, topology is dynamic – Unlike ad hoc networks, motion (and topology) is predictable – Unlike typical sensor networks, have natural load-balancing • Long ranges between adjacent nodes – Must use directional transmit and receive antennas • Largely ignored in literature, although some recent interest (e. g. for FCS); no known sensor network results – Multihop needed for ground operations efficiency and communications & energy efficiency 8

Assumptions • Sensor network, with – traffic originating at satellite nodes and destined to multiple ground stations on Earth, and – traffic originating at Earth stations and destined to satellites • Supports half-duplex or full-duplex operation • Directional antennas are used, so that “hidden terminal” interference does not arise • Network is synchronized 9

Technical Approach 0. Obtain potential topology G 1. Grow branches rooted at satellites that are 1 -hop away from any ground station 2. Compute the total load of a subtree rooted at each node 3. Load-balancing among different branches 4. Attach branches to ground stations (min. schedule) 5. Load-balancing among ground stations Cannot balance to improve schedule 6. Generate schedule from tree using Florens -Mc. Eliece algorithm 10

Derive Node Locations Example 16 -satellite, 3 -ground stations configuration 11

Grow Branches Lbranch(1) = 1 Lbranch(2) = 13 Lbranch(3) = 2 1 13 2 4 15 16 10 14 3 1 1 1 9 4 4 2 12 1 10 1 6 1 2 1 7 5 1 1 13 11 1 8 12

Load-Balancing Among Branches Lbranch(1) = 1 Lbranch(2) = 13 Lbranch(3) = 2 1 13 2 4 15 16 10 14 3 1 1 1 9 4 4 2 12 1 10 1 6 1 2 1 7 5 1 1 13 11 1 8 13

Load-Balancing Among Branches (cont) Lbranch(1) = 1 Lbranch(3) = 6 Lbranch(2) = 9 1 9 6 4 15 16 4 2 1 1 1 9 1 10 6 2 3 14 1 13 5 5 1 4 11 12 1 6 1 7 1 8 14

Attach to Ground Stations Canberra Goldstone 1 8 7 16 15 4 7 6 3 4 1 1 1 9 5 1 2 1 10 No improvements can be mad by load balancing among the ground stations (step 5) 6 13 12 1 6 1 1 2 7 8 14 1 11 15

Generate Schedule for Tree An algorithm for deriving an optimal (shortest-length) schedule for each tree rooted at a ground station with half-duplex directional links has been developed: Cedric Florens and Robert Mc. Eliece, “Scheduling algorithms for wireless ad-hoc sensor networks, ” Proceedings of IEEE GLOBECOM 2002, Dec. 1 -5, 2002 This algorithm holds for general traffic load distribution We apply this algorithm to each tree to obtain the final schedule 16

Example Schedule Table Schedule for 16 -satellite example: → 15 time slots to deliver all 16 packets 17

Mitigation of Propagation Delays Directionality of path flows permits schedule to be adjusted to remove effects of propagation delays • Operation: – Pull data from all satellites to Earth – Push Earth commands/data to satellites • Propagation losses only occur in transitions between these two operational modes • Can be applied to either Half-Duplex or Full-Duplex systems 18

Propagation Delays (Half Duplex) One Cycle of Schedule C 4 15 3 14 2 11 1 10 9 Canberra 19

Propagation Delays (Full Duplex) One Cycle of Schedule C 4 15 3 14 2 11 1 10 9 Canberra 20

Simulation A simulation was developed for performance characterization Simulation execution: • General topologies derived from random spatial distribution and internode range constraints • Traffic load generated from statistical model • Tree optimization algorithm executed • Link activation/routing schedule derived • Measure statistics on schedule length and throughput performance Example Topology 21

Simulation Results Performance Improvement using Optimized Tree Algorithm 1 ground station 2 ground stations 4 ground stations 6 ground stations 8 ground stations Schedule length using optimized tree algorithm 100. 73. 38 49. 76 40. 68 33. 17 Schedule length without optimized tree algorithm 159. 16 113. 52 77. 34 59. 73 47. 92 Percent length increase 59. 2% 54. 7% 55. 4% 46. 8% 44. 5% Schedule Length versus Number of Ground Stations 22

Simulation Results (continued) Schedule Length versus Number of Ground Stations Performance Improvement using Optimized Tree Algorithm Schedule Length versus Network Size 23

Simulation Results (continued) Schedule Length versus Number of Ground Stations Schedule Length Distribution (20 nodes) 80 70 60 2 GS 50 4 GS 40 6 GS 30 8 GS 20 10 0 0 50 100 150 200 250 Time Slots 24

Summary • Space-based sensor networks are emerging in order to enable new science requiring multipoint measurement • Interspacecraft communications (networking) will enable – Continuous access to any/all spacecraft in the multi-spacecraft mission via single ground contact with any spacecraft, thereby increasing ground operations efficiency and enabling automated operation of the whole – Real-time coordinated observations are made possible, such as alerting/cueing ground-based assets – Autonomous operations/processing among distributed spacecraft including precision navigation and formation control and reconfiguration • Presented a layer 2 mesh link activation/routing algorithm that maximizes throughput and minimizes latency 25