The Cricket Location-Support System N.B. Priantha, A. Chakraborty, H. Balakrishnan Presented by Lewis Girod
Overview • Cricket “listeners” use acoustic time-offlight measurements to determine the nearest beacon • Objectives: – Granularity of “a few square feet” – Different regions distinguishable – Low cost, both H/W and installation
• In general, the system seems to work
Related Work • Niru’s RF beaconing • “RADAR” from Microsoft Research – centralized computation, not clear how well it works
• Active Bat from ORL • The Cricket work differs: – It does not attempt to calculate any kind of absolute position, but only the nearest beacon – Interest in demonstrating that nearest beacon is “good enough” for application requirements • Knowing what room you are in, which part of room
Design Goals • Privacy: listeners are passive • Decentralized administration: owner of space installs and configures beacons as needed • Network heterogeneity: Cricket not coupled to any networking technology • Cost: under $10 per unit • Granularity: spatial regions can be as small as a “few square feet”. This will really depend on beacon placement
Metrics & Terminology • Precision: how well can a listener detect a boundary (rate of correct detection) • Granularity: The smallest possible size for a detectable geographic region • Objective is near 100% precision with a granularity of a few square feet
System Architecture • Beacons placed near ceiling where they have better chances of LOS with listeners • Beacons delimit boundaries: – On either side of doorway to delimit two adjacent rooms – On either side of non-physical “region boundaries” – In corners and near walls (physical boundaries)
Operation of Ranging System • Time of flight measurement – From beginning of RF transmission to ultrasound (US) pulse. – RF transmission is as long as the maximum US propagation time; “matching” US pulse must arrive during RF transmission. Avoids need to code pulses.
• Beacons are not coordinated – Interference prevented using randomized delays between beaconing
Interference Problems • Mnemonics:
RF-A
– RF-A – US-A – US-RA
US-A
US-RA
(reflection of US-A)
– RF-I (interfering RF signal)
– US-I – US-RI
RF-I US-I
True TOF
Case 1: RF-A / US-RA • Standard NLOS problem: – the direct path is blocked, longer reflected path detected
• Cricket uses placement of beacons to avoid this. – Emitters are on ceiling, pointed downwards towards floor and “into” their region: towards the probable location of listeners in their region – Larger number of beacons also mitigates this problem: beacon pointing downwards will have LOS, and will be shorter path than a reflection from farther away.
Case 2: RF-A / US-I, RF-I / US-A RF-A
2a.
I Sends early I
US-A
2b.
RF-I
I US-I
A Sends later
• Results of RF interference – 2a. RF-A transmission drowns out earlier but farther-away RF-I; US-I detected earlier – 2b. RF-I transmission from beyond a solid partition drowns out RF-A, but US-A detected.
• Solutions: – Random inter-beacon delays reduce likelihood of persistent errors – If RF TX range is larger than US TX range, case 2a is less likely.. more likely the RF messages will collide and neither will be received.
Case 3: RF-A / US-RI • Stray reflected US pulses may cause incorrect readings. • Solution: – Limit the beacon density and rely on random inter-beacon delays to avoid collisions. – Currently, Cricket system is engineered to have at most 6 beacons within range of each other at any given point
Overcoming Interference • Statistical approaches for filtering out bad data – Majority: (strawman) • select the beacon heard most frequently regardless of distance
– MinMean: • For each beacon, calculate the avg. distance, then select beacon with minimum value. • Problem: multipath causes modal behavior, this algorithm ignores that information, and averages across modes
– MinMode: • For each beacon, calculate the mode of last n samples, then select beacon with minimum mode. • Robust to stray signals
Deployment issues Problem arising from directionality
(top view) Using directionality to enable “virtual boundaries”
• Deployment critical to making Cricket work – US emitters are highly directional, which can easily lead to problems with reflections – Instead, Cricket exploits directionality by arranging emitters at boundary of region, facing in downwards from ceiling. • Each region must mark the border with one of its own beacons, separated by at least 4 feet • This requirement is needed because the beacons are on the ceiling; so the relative distances that must be distinguished are too close unless the beacons are separated.
Performance Analysis 4 feet
6 feet
• Boundary detection – Two beacons on ceiling, 4 feet apart – Listener moves outwards from center of two beacons; distances measured at 0.5 foot intervals – Region membership is detected correctly after the listener is more than 1 foot into the region
Robustness to Interference • Static performance with 2 listeners and 5 beacons – Experimental configuration • One listener had two nearby pure-RF interferers • The other was nearby a boundary marked by 3 beacons
– Interference detected in less than 1% of samples. • Caveat: the transmission rate was not given – This behavior is probably a direct result of the randomized transmission schedule – What is the tradeoff between beacon density, beaconing rate (and therefore response to dynamics), and interference rate?
– Majority performed poorly; minMean and minMode performed perfectly, but indistinguishably • Very little interference for the statistics to deal with
Tests Involving Mobility • In this experiment, a listener is moved through the environment – Stop at the first time a new region is detected – Take a series of samples at that point – Continue moving
• Tests performed very near boundaries – Good test: ambiguity should be highest – Results show that if several samples are taken the correct answer is determined with high probability
Conclusions: • Cricket works quite well at what it does • Location support, not location tracking: eliminates privacy concerns • Caveats: – Incompatible with ad-hoc beacon placement – Does not solve problem of fine-granularity location.. at best within 2-4 foot radius • Tradeoff between granularity of location and interference from neighboring beacons (they had a max of 6 beacons in range) • May mean it works best indoors in confined spaces?
– Does not attempt to interpolate coordinate system between beacons, simply presents name of closest beacon.