ErrorCorrecting SequenceBased Localization for Wireless Networks A New

Error-Correcting Sequence-Based Localization for Wireless Networks: A New Paradigm Bhaskar Krishnamachari Autonomous Networks Research Group Dept. of EE-Systems USC Viterbi School of Engineering http: //ceng. usc. edu/~anrg bkrishna@usc. edu ARO Workshop, Seattle, June 1415, 2005 1

Overview • Location information is a fundamental building block for self-organized wireless ad-hoc and sensor networks. It is important for – – stamping sensor measurements target tracking topology formation routing and querying • Thus far, the primary focus in designing localization algorithms has been on functionality. • Critical challenges of fault-tolerance and security have been largely ignored. 2

Securing Localization • Localization algorithms can be made secure and robust in a number of complementary ways: – developing tamper-proof hardware – securing measurements through cryptographic algorithms – patches to existing algorithms to address identified vulnerabilities – developing a fundamentally new class of localization algorithms 3

Thesis • A new class of sequence-decoding localization algorithms, with the potential to automatically detect and correct errors introduced by the environment as well as malicious attackers, will be a key component of future tactical wireless networks. 4

Traditional Forward Error Correction encoder original message higher dimension codeword channel with error decoder corrupted packet “nearest” correct codeword received message • FEC is at the heart of modern high-performance wireless communication. • A major field of research for several decades • Latest FEC techniques (turbo codes, LDPC codes) can provide low 5 error communication within 0. 1 d. B of theoretical Shannon limit

Error Correcting Localization encoder ideal signals corrupted signals (RSS, TDOA, AOA, etc. ) noise/environmental errors malicious errors codeword decoder corrupted codeword “nearest” correct codeword decoded location 6

Ecolocation • A novel RF-only sequence-based error-correcting localization technique currently under development • Empirically shown to have superior performance compared to state of the art techniques • “Tip of the iceberg” Reference: Yedavalli, Krishnamachari, Srinivasan, Ravula, “Ecolocation: A sequence based technique for RF-only localization in wireless sensor networks, ” IPSN 2005. 7

Ecolocation • Basic idea: look at the sequence indicating relative ranking of RSSI measurements, not absolute values • Each sequence ideally corresponds to a unique location region • Provides a way to decode location with high accuracy, even given a possibly erroneous sequence. 8

The Basic Algorithm ¨ Unknown node sends a beacon. ¨ Nearby reference nodes measure RSSI and send to computation point. ¨ Sequence is determined and expressed as a set of ordering constraints. ¨ Most likely location is computed based on this measured sequence 9

Illustration • Sequence: ADBC DC B AD BC D C A AB AC DB 10

Motivation • Ordered sequence is inherently more robust to amplitude fading fluctuations than absolute signal strengths • Many corrupt sequences do not correspond to any valid locations - hence error is easily detected and can be corrected in most cases by mapping to nearest valid sequence. Specifically, the number of feasible codeword sequences is only O(n 4) out of n! possible (corrupt) sequences. 11

Location Determination ¨ Consider a grid of location points in the environment ¨ Determine ideal sequence for a given possible location of the unknown node ¨ Look at the measured sequence and compare with above to determine number of satisfied/violated constraints ¨ Identify location(s) that maximizes the number of satisfied constraints • Optimizations: Multiresolution search/Greedy approaches can significantly cut down on search time and computation 12

An alternative approach • Precompute regions in the location space corresponding to feasible error-free sequences (not all possible sequences are feasible) • Determine the feasible sequence that “best” matches received sequence and return the corresponding location • Can yield a much faster solution, can also be optimized through multi-resolution/greedy approaches 13

Order Constraints 1 2 A 5 F 3 4 B C D E 14

Constraint Violations 1 2 A 4 F 3 5 B C D E 15

Illustration NO ERRONEOUS CONSTRAINTS 16

Illustration 13. 9% ERRONEOUS CONSTRAINTS 17

Illustration 22. 2% ERRONEOUS CONSTRAINTS 18

Illustration 47. 2% ERRONEOUS CONSTRAINTS 19

Evaluation • Simulation Model: – RSS samples generated using log-normal shadowing model • Simulation Parameters: – RF Channel Characteristics • Path loss exponent (η) • Standard deviation of log-normal shadowing model (σ) – Node Deployment Parameters • • Number of reference nodes (α) Reference node density (β) Scanning resolution (γ) Random placement of nodes 20

RF-only State of the Art • Pattern Recognition (e. g. RADAR) • Centroids • Approximate Point in Triangle (APIT) • RSSI-based Maximum Likelihood Estimation (MLE) • RSSI-based Minimum Mean Squared Error Estimation (MMSE) • Proximity (nearest reference, an extreme special case of ECOLOCATION) 21

22

Experiments with Real Measurements • Outdoors: Parking Lot. – Eleven MICA 2 motes placed randomly in an unobstructed 144 sq. m area – Locations of all motes estimated and compared with true position • Indoors: 3 rd floor of EE building – Twelve MICA 2 motes placed randomly in an obstructed 120 sq. m area in an office building – unknown node placed at five locations for position estimation 23

Empirical Results (1) 24

Empirical Results (2) 25

Empirical Results (3) 26

Empirical Results (4) 27

Observations • Ecolocation is self-configuring - it does not require prior measurement of environment. It is robust and efficient in dense settings. • Can be easily extended to 3 D environments and to incorporate other available information (including antenna orientations, operational area constraints) • Most importantly, Ecolocation can also detect and mitigate induced errors from malicious nodes. (Each adversary can forge at most n-1 constraints out of n(n-1)/2 ) 28

Research Agenda • Intermediate term: develop Ecolocation – Full, optimized testbed implementation of Ecolocation taking into account resource constraints on energy, computation, and communication – Quantifying security using different adversarial models – Theoretical analysis of gains from error correction (is there an equivalent to coding gain in communications? ) 29

Research Agenda • Long term: Develop and analyze a wide range of sequence/codeword-based error correcting localization algorithms suitable for different contexts: – with other signal measurement modalities (angles, TDo. A-based ranges, etc. ) – under different density/mobility assumptions – for network localization (multiple unknown nodes) 30

Additional Thoughts • Enable multiple “competing” solutions • Develop a “standard suite” of benchmark problems for comparisons – realistic empirical traces or real common test-bed – different environmental operating conditions (density, mobility, resource constraints, indoor/outdoor, interference) – different modalities (pure RF, multimodal TDo. A) – different localization requirements (single/multiple unknown node, cooperating/non-cooperating nodes, different accuracy and precision requirements, etc. ) – different attack models and assumptions 31