FILTERING NOISY 802.11 TIME-OF-FLIGHT RANGING MEASUREMENTS FROM COMMODITIZED WIFI RADIOS

FILTERING NOISY 802.11 TIME-OF-FLIGHT RANGING MEASUREMENTS FROM COMMODITIZED WIFI RADIOS

 

ABSTRACT

Time-of-flight (ToF) echo techniques have beenrecently suggested for ranging mobile devices over WiFiradios.However, these techniques have yielded only moderate accuracyin indoor environments because WiFiToF measurements sufferfrom extensive device-related noise which makes it challengingtodifferentiate between direct path from non-direct path signalcomponents when estimating the ranges. Existing multipathmitigation techniques tend to fail at identifying the direct pathwhen the device-related Gaussian noise is in the same orderof magnitude, or larger than the multipath noise. In order toaddress this challenge, we propose a new method for filteringranging measurements that is better suited for the inherentlarge noise as found in WiFi radios. Our technique combinesstatistical learning and robust statistics in a single filter.Thefilter is lightweight in the sense that it does not require specializedhardware, the intervention of the user, or cumbersomeon-site manual calibration. This makes our method particularlysuitable for indoor localization in large-scale deployments usingexisting legacy WiFi infrastructures. We evaluate our techniquefor indoor mobile tracking scenarios in multipath environmentsand, through extensive evaluations across four different testbedscovering areas up to 1000 m, the filter is able to achieve a median2-D positioning error between 2 and 3.4 m.

PROPOSED SYSTEM:

We have used our filter to estimate the location ofWiFi mobile devices in four different experimental testbeds.Our results show that our technique is able to significantlyreduce the median ranging error over classical estimators. Ourmain contributions in this paper are the following:• We present a firmware-based ranging architecturefor round-trip time measurements running in theMAC processor of WiFi chipsets, and we quantify andcompare the amount of noise that comes from the WiFidevices and the noise from the WiFi radio signal propagationover the wireless channel.• We propose a filter based on statistical learning and robuststatistics to estimate the distance range from a series ofnoisyToF measurements. Our filter does not require anymanual calibration and manages to estimate the distanceto a remote WiFi device with median error between1.7 and 2.4 meters.• Finally, we evaluate our system for indoor mobile devicetracking. We demonstrate across four different experimentaltestbeds that mobile devices associated to theWiFi infrastructure can be tracked continuously using ourfilter with a median accuracy between 2.0 and 3.4 meters.

EXISTING SYSTEM:

The indoor localization literature is vast, includingtechniques using signal strength, the angle ofarrival| or combining WiFi signals with inertial sen-sors as found in smartphones. In this section, we focuson time-based localization techniques which are most relatedto ours in addition to some general NLOS mitigation basedsolutions.ToF Echo techniques based on packet exchanges inWiFi networks were first proposed in and refinedin. However, unlike our work, none of theseapproaches address the effect of non-Gaussian noise such asin multipath-rich environments. A direct comparison as presented in this work shows that the error with ourstatistical filter can be reduced in indoor environments bya factor of more than two compared to classical estimatorsthat do not compensate for the bias of the multipath.Galloet al. [30] introduced directional Yagi antennas toeliminate the effect of multipath and other noise sources fromWiFi echo techniques. They achieved a positioning accuracyof less than 5 min8 from 10 positions. In contrast, our filterworks with omnidirectional antennas in environments withmultipath.There have been many attempts to harness the time-of-flight of wireless signal in indoor propagation environmentsdespite multipath. Common proposals to combat the multipathproblem are for example the use ofultrawidebandsignalsor frequency diversity. However,these approaches require specialized hardware or softwaredefinedradios which increase costs and hinders localizationat larger scales. SAIL is a ToF system using WiFithathas been designed for localization in multipath environments.However, SAIL requires inputs from the inertial sensors inthe smartphone. SAIL achieved median error of ˜ 1 mand80-percentile error of ˜ 5 m, which is comparable to ourfilter, at the drawback of requesting the collaboration from themobile user through the installation of a dedicated applicationon smartphones. ToneTrack tries to overcome the problemof limited bandwidth, inherent to WiFi time-based localizations.It combines channels to form virtual larger bandwidthswithout increasing the radio’s sampling rate, taking benefit offrequency hopping, to increase the resolution ofToFprofiles.The work SpotFi uses APs equipped with 3 antennas andcommodityWiFi chipsets. It jointly estimates the Angle ofArrival (AoA) and ToF pairs of each path using the channelstate information, and estimates the likelihood that these pairscorrespond to the direct path between the AP and the target.In contrast to our work, SpotFi does not rely on ToFforranging. Other solutions try to identify and mitigate the NLOSeffects of WiFi RSS measurements. The proposed solutioncombines a machine learning technique to first extract typicalfeatures from the training data collected during extensiveindoor measurement campaigns and estimate the ranges usinga regression model, and an identification approach based onhypothesis testing. The approach still needs a training phasethat requires an offline classification or calibration. Still inRSS based ranging, proposes to use a GMM model tofilter corrupted range estimations caused by NLOS radio propagationby modeling distributions of LOS and NLOS sets ofestimates. In this work, we have demonstrated that the GMMmodel does not work well in WiFiToF and we have proposeda combination of statistical learning and robust statistics thatoutperforms the classical expectation-maximization algorithmused by the GMM model.

CONCLUSION

We have developed a new filter based on statistical learningand robust statistics to improve the ranging accuracy in thepresence of noisy ToF measurements. Our filter does notrequire additional information form the devices or the userbesides the ToF values which makes it applicable to a widerange of applications. We have shown how to apply our filterfor indoor localization and tracking of COTS WiFideviceswith legacy 802.11 Access Point infrastructures. In indoordeployments with multipath, our filter outperforms conventionalToF based range estimators by a factor of more than two.We have demonstrated the accuracy of the filter to estimatethe position in static and mobile settings. In static conditions,the filter achieved a median error of 2.0-3.1 meters. In mobilesettings, the error was only slightly higher with a median errorof 2.6 – 3.4 meters. We have shown that the performance ofour filter can be achieved with online model calibration,andhence does not require any cumbersome onsite pre-calibrationefforts. The system has also participated in the MicrosoftIndoor Localization Competition 2016, achieving an averageerror of 3.17 meters in a challenging and uncontrollableenvironment with 5 APs covering 500 m. This resulted inranking 5th out of ten teams in the final. Our system wasthe only solution presented at the competition that did notrequire any customized software in the mobile device, neitherinputs from inertial sensors in the mobile device. Since ToFisbecoming readily available in WiFi chipsets, we foresee thatour approach can be applied by the provider of positioningsystems in airports, malls, museums, homes for context-awarenetworking as well as data analytics that require positioningdata. The advent of new WiFi chipsets operating on widebandchannels (such as the 160 MHz clock of 802.11ac) willgreatly help to increase the accuracy and the integration withphase information will also be needed to improve the systemperformance.

REFERENCES

[1] P. Bahl and V. N. Padmanabhan, “RADAR: An in-buildingRF-based user location and tracking system,” in Proc. IEEE INFOCOM,Mar. 2000, pp. 775–784

.[2] M. Youssef and A. Agrawala, “The horus WLAN location determinationsystem,” in Proc. MobiSys, New York, NY, USA, 2005, pp. 205–218.

[3] H. Lim, L.-C. Kung, J. C. Hou, and H. Luo, “Zero-configuration, robustindoor localization: Theory and experimentation,” in Proc. 25th IEEEInt. Conf. Comput. Commun. (INFOCOM), Apr. 2006, pp. 1–12.

[4] A. Goswami, L. E. Ortiz, and S. R. Das, “Wigem: A learning-basedapproach for indoor localization,” in Proc. CoNEXT,NewYork,NY,USA, 2011, pp. 3:1–3:12.

[5] K. Chintalapudi, A. P. Iyer, and V. N. Padmanabhan, “Indoor localizationwithout the pain,” in Proc. ACM MobiCom, 2010, pp. 173–184. [Online].Available: http://doi.acm.org/10.1145/1859995.1860016

[6] J. Xiong and K. Jamieson, “ArrayTrack: A fine-grained indoor locationsystem,” in Proc. 10th USENIX Symp. Netw. Syst. Design Implement.(NSDI), Lombard, IL, USA, 2013, pp. 71–84.

[7] S. Sen, J. Lee, K.-H. Kim, and P. Congdon, “Avoiding multipath to reviveinbuildingWiFi localization,” in Proc. 11th Annu. Int. Conf. MobileSyst., Appl., Ser. (MobiSys), New York, NY, USA, 2013, pp. 249–262.

[8] J. Gjengset, J. Xiong, G. McPhillips, and K. Jamieson, “Enabling phasedarray signal processing on commodity WiFi access points,” in Proc.Mobicom, New York, NY, USA, 2014, pp. 153–164 .

[9] M. Kotaru, K. Joshi, D. Bharadia, and S. Katti, “SpotFi: Decimeter levellocalization using WiFi,” in Proc. ACM Conf. Special Interest GroupDataCommun. (SIGCOMM), New York, NY, USA, 2015, pp. 269–282.