1. Institute for Software Integrated Systems Vanderbilt University Node Density Independent Localization Presented by: Brano Kusy B.Kusy, M.Maroti, G.Balogh, P.Volgyesi, J.Sallai, A.Nadas and A.Ledeczi (Vanderbilt University, Nashville, TN) L.Meertens (Kestrel Institute, Palo Alto, CA)

2. Overview Introduction Radio Interferometric Ranging (RIPS) Improving RIPS Ambiguity of Interferometric Range Multipath Interleaving Ranging and Localization Scalability of RIPS in Time Maximum Number of Independent Interferometric Measurements Scheduling Evaluation Tracking Demo Conclusions

3. Introduction Ranging: determine distances between nodes Localization: find physical 3-D locations of nodes State of Art: Time of flight - acoustic ranging: sound – tens of meters range; 10cm accuracy; ultrasound – 5m range, 6cm accuracy; RF Time of flight: GPS – anywhere outdoors; accuracy is feet with DGPS UWB – <10m range; 6 inches accuracy; Radio signal strength - RSSI: MS RADAR – 35m range; 3m accuracy; Calamari – 20m range; 2m accuracy; Range Free: SPOTLIGHT – 170m range; ~20cm accuracy; Radio interference – RIPS: 170m range; 4cm accuracy; Contribution: improved the range of RIPS from 20m to 170m improved RIPS ranging error distribution addressed scalability constraints

4. Radio Interferometric Ranging Interference: superposition of two or more waves resulting in a new wave pattern Interferometry: cross-correlates a signal from a single source recorded by 2 observers, used in geodesy, astronomy, … Our novel technique - RIPS: no processing power to correlate high freq radio signals in sensornets, instead we utilize two transmitters to produce low frequency interference signal directly 1.Signal strength is not crucial: no dependence on orientation, power level, hardware deviations 2.Low freq envelope (of composite signal): inexpensive HW 3.High carrier freq: high accuracy

5. Radio Interferometric Ranging φ CD = ( d AD -d BD +d BC -d AC ) mod λ Senders (A, B) transmit simultaneously pure sinusoid waves high carrier freq (400 MHz) small freq difference (500 Hz) Receivers (C, D) measure radio interference sample RSSI (9 KHz) find beat frequency (500 Hz) measure phase offset of RSSI use 1 μs timesync to correlate phase offsets result: (d AD -d BD +d BC -d AC ) mod λ d XY : distance of X and Y λ: wave length of carrier freq

6. Overview Introduction Radio Interferometric Ranging (RIPS) Improving RIPS Ambiguity of Interferometric Range Multipath Interleaving Ranging and Localization Scalability of RIPS in Time Maximum Number of Independent Interferometric Measurements Scheduling Evaluation Tracking Demo Conclusions

7. Ambiguity of Interferometric Ranges φ CD = (d AD -d BD +d BC -d AC ) mod λ interferometric range, or q-range Interferometric range (q-range) for 4 fixed nodes A, B, C, D is a solution of the following equation: In general, infinitely many ranges solve this equation. We eliminate incorrect solutions by measuring phase offsets at multiple carrier frequencies. We then solve the following system of equations: φ 1 CD = (d AD -d BD +d BC -d AC ) mod λ 1 φ k CD = (d AD -d BD +d BC -d AC ) mod λ k...

8. Ambiguity of Interferometric Ranges How do we solve the system of equations and what are the ranging errors of our method? d φ1 φ2 φ3 Define the ranging problem as a frog jumping problem: we have k frogs jumping along a line phase offsets: the points where the frogs start jumping wavelengths: lengths of the frogs’ jumps interferometric range r: place where all frogs align

9. Ambiguity of Interferometric Ranges d φ1 φ2 φ3 λ 1 λ 2 λ 3 Error of type 1 Define the ranging problem as a frog jumping problem: we have k frogs jumping along a line phase offsets: the points where the frogs start jumping wavelengths: lengths of the frogs’ jumps interferometric range r: place where all frogs align Ideal world: discrepancy function has minimum at true q-range Our world: world of phase offset errors r phase-offset discrepancy: a mean square error of φ i +n i λ i from r Extra solutions exist small number of wavelengths to the left and right of r Error of type 1 is minimized if λ separation is maximized.

10. Ambiguity of Interferometric Ranges d Define the ranging problem as a frog jumping problem: we have k frogs jumping along a line phase offsets: the points where the frogs start jumping wavelengths: lengths of the frogs’ jumps interferometric range r: place where all frogs align phase-offset discrepancy: a mean square error of φ i +n i λ i from r

11. Ambiguity of Interferometric Ranges d Error of type 2: Define the ranging problem as a frog jumping problem: we have k frogs jumping along a line phase offsets: the points where the frogs start jumping wavelengths: lengths of the frogs’ jumps interferometric range r: place where all frogs align phase-offset discrepancy: a mean square error of φ i +n i λ i from r Error of type 2 is minimized if f separation is minimized.

12. Multipath – Ground Reflection RIPS ranging error distribution observations: 1.gets significantly worse for distances above 25m, if nodes on the ground (campus) 2.improves when lifting motes 1m above the ground (campus) 3.is good for distances above 100m, even if nodes are positioned on the ground (field – no trees, no buildings close by) (3) suggests that multipath is at play here! But why elevating the motes helps the problem? Answer: radio signal reflected from surfaces under small angles has significant amplitude and opposite phase than the original radio signal.

13. discrepancy function in multipath environment: Interleaving Ranging and Localization Recap: q-range r - solution of equation φ i +n i λ i =r phase-offset discrepancy function - alignment of points true q-range Our interleaved algorithm: 1.find all q-ranges 2.localize using these q-ranges 3.constrain q-range search with the intermediate locations Our observations: it rarely happens that all q-ranges are affected by multipath 30-40% of good (<30cm error) q- ranges give us good (<few m error) locations even if q-range is affected by multipath, discrepancy function has local minimum at true range

14. Evaluation: setup 1 UCB Richmond Field Station: 50 node setup, 9m neighbor distance, 108m max distance moderate multipath present (trees, buildings close by) GPS ground truth has few m error 68% of ranges has <1m error better ground truth needed manually localized subset of 30 nodes have ~5cm error 68% of measured ranges has <10cm error

15. Evaluation: setup 2 rural area close to Nashville: 16 node setup, deployed on the ground, 35m avg, 170m max neighbor distance covering 12000m 2 area (2 football fields) no multipath present, other than the ground reflections localization: 4cm avg, 12cm max error q-range error distribution 72% of ranges has <30cm error error distribution after 3 iterations 98% of ranges has <30cm error

16. Dirty Bomb Detection and Localization Demo RIPS was used in our tracking demo yesterday 12 XSM anchor motes used near real-time tracking:  2 sec location refresh rate,  3 sec delay, high accuracy:  better than 1m error long range:  often larger than 100m ranges used for localization  covered 80m x 90m area tracked 1 mote only theoretically, we can track arbitrarily many motes as only anchor motes need to transmit completely stealthy

17. Conclusions and Questions  Introduced improved version of novel ranging technique for wireless sensor networks  Large range, high accuracy, requires no extra HW, stealthy,...  Had successful tracking demo QUESTIONS ???