ICDCS 2013 Pilot: Device-free Indoor Localization Using Channel State Information Jiang Xiao, Kaishun Wu, Youwen Yi, Lu Wang, Lionel M. Ni Department of Computer Science and Engineering Hong Kong University of Science and Technology July. 10 th, 2013
Outline Introduction & Motivation Related Work Hypotheses Methodology Performance Evaluation Conclusion 2
Indoor LBSs Safety Precaution Area Isolation Pilferage Prevention Residence Hospital Warehouse Device-free techniques are in need! 3
Indoor Device-free Localization (DfL) High cost for large environment Ultrasonic Existing Approaches Environment constraints (e.g., dark, smoke) Sensor High cost for dense deployment Camera Infrared Line-of-sight Can t work well in large scale, complicate, typical indoor environment! 4
Motivation WLAN Advantages ü Low Cost ü Good Scalability WLAN-based Indoor Device-free Localization! 5
Challenges 1. When to trigger localization? à Motion detection is the prerequisite. 6
Challenges 2. How to realize localization? 7
Outline Introduction & Motivation Related Work Hypotheses and Measurements Methodology Performance Evaluation Conclusion 8
Existing WLAN-based DfL Most of researches rely on RSS measurement MLE Alg.[PerCom] Nuzzer[TMC] FIMD [ICPADS] 2007 2009 2010 2012 2008 2011 Challenges [MobiCom] Survey [IEEE Invited Paper] RASID[PerCom] Pilot [ICDCS] 2013 Time M. Seifeldin M. Youssef Analyzed the challenges for DfL systems Proposed to use radio signal strength (RSS) Developed Nuzzer & RASDID DfL systems Surveyed the recent researches on DfL J. Wilson N. Patwari 9
Existing RSS-based DfL TX RX RSS Static Dynamic AP l1 l2 Static l3 l4 Dynamic DP Youssef et al. Nuzzer System [PerCom 11] 10
In static environment RSS Limitation 15dB High Variability Receiver Node ID single value RSS 2.4GHz RF band Coarse measuremen t 11
We want to find a reliable location feature for WLAN-based DfL. 12
Opportunity Data in OFDM Transmitter + Channel OFDM Receiver Data out In 802.11 n OFDM system, the received signal over multiple subcarriers is CSI Channel gain Previously, CSI à amplitude phase [SIGCOMM 10, MobiCom 11] 13
Observations CSI Property 1: Frequency Diversity Receiver single value RSS multiple values CSIs 2.4GHz S/P FFT RF band Baseband CSI Property 2: Temporal Stability RSS (dbm) RSS: Variant CSI amplitude CSI: relatively stable Time Duration (s) Time Duration (s) 14
CSI RSS Temporal Stability Frequency Diversity We want to harness fine-grained CSI for device-free indoor localization. 15
Our Work Hypotheses Feasibility of CSI Methodology Design Pilot System Evaluation Accuracy (Motion Detection & Localization) 16
Outline Introduction & Motivation Related Work Hypotheses Methodology Performance Evaluation Conclusion 17
Hypothesis1 Hypotheses 1/2 Ø CSI can reveal the abnormal status caused by appearance of human. 18
Hypotheses 1/2 Experiment 1: Anomaly Detection by CSI Feature CSI feature can reveal normal/motion behavior 19
Hypothesis 2 Hypotheses 2/2 Ø CSI can be leveraged to distinguish human in different locations. 20
Hypotheses 2/2 Experiment 2: Location Distinction by CSI Feature 21
Outline Introduction & Motivation Related Work Hypotheses Methodology Performance Evaluation Conclusion 22
System Overview 3 Components 23
Pilot Architecture Process CSI Passive Radio Map Construction Collect CSI Channel Estimation OFDM Demodulator Anomaly Detection Fingerprint Mapping Data Fusion Passive Fingerprints DB (Normal/Abnormal) Position Estimation Pilot DP Pilot Server 24
Outline Introduction & Motivation Related Work Hypotheses Methodology Passive Radio Map Construction Anomaly Detection Position Estimation Performance Evaluation Conclusion 25
1. Passive Radio Map Construction DP1 DP2 AP1 AP2 AP1 AP2 26
1. Passive Radio Map Construction To process the measurement data To generate the fingerprints database In normal state: -- aggregated correlation sum CSI of n packets In dynamic state: 27
1. Passive Radio Map Construction Sensing Zone & Dead Spot 28
Outline Introduction & Motivation Related Work Hypotheses and Measurements Methodology Passive Radio Map Construction Anomaly Detection Position Estimation Performance Evaluation Conclusion 29
Anomaly Detection Localization Trigger N Alarm? Y Localization 30
2. Anomaly Detection Main idea: to check the probability of each CSI feature to be in normal profile. To estimate the distribution of Kernel density estimator Correlation of (M+W-1) Packets 1 0.8 Bandwidth Anomaly Detection Threshold Epanechnikov Quadratic Kernel Function CDF CDF of CSI Feature à 0.6 0.4 0.2 β 0 0 0.2 0.4 0.6 0.8 1 CSI Feature C β 31
Anomaly Detection N Alarm? Y Localization Position Estimation 32
Outline Introduction & Motivation Related Work Hypotheses and Measurements Methodology Passive Radio Map Construction Anomaly Detection Position Estimation Performance Evaluation Conclusion 33
3. Position Estimation To compare CSIs against the abnormal To select the best match (MAP Alg.) Link l Unknown Location Fuse Multiple RF Links 34
Outline Introduction & Motivation Related Work Hypotheses Methodology Performance Evaluation Conclusion 35
Experimental Setup Commercial Hardware Intel WiFi Link 5300 802.11n Router 36
Testbeds 1/2 Lab in HKUST 77m 2 37
Testbeds 2/2 Lobby in HKUST 776m 2 38
Evaluation Metrics Receiver Operating Characteristic (ROC) curve: interprets the detection performance in the presence of false alarm. Condition: Motion Condition: Static Test outcome: Motion True Positive (TP) Rate False Positive (FP) Rate False Alarm Test outcome: Static False Negative (FN) Rate Ignore Miss detection 39
Detection Accuracy in Lab DR>90%, 30% Refer to [PerCom 12] Youssef et. al. RASID system 40
Detection Accuracy in Lobby DR > 90% 10% Refer to [PerCom 12] Youssef et. al. RASID system 41
Localization Accuracy in Lab Refer to [PerCom 11] Youssef et. al. Nuzzer system 42
Localization Accuracy in Lobby Refer to [PerCom 11] Youssef et. al. Nuzzer system 43
Localization Accuracy in Lab Refer to [PerCom 11] Youssef et. al. Nuzzer system 44
Localization Accuracy in Lobby Refer to [PerCom 11] Youssef et. al. Nuzzer system 45
Outline Introduction & Motivation Related Work Hypotheses Methodology Performance Evaluation Conclusion 46
Conclusion We explore the feasibility of using CSI for indoor DfL. We design the Pilot system, and implement it on commercial products. CSI is a better tool for DfL compared to RSS as shown in experiments. 47
Future Work To facilitate multiple objects localization. To quickly establish database. To optimize AP-DP deployment. 48
Thanks. Questions? jxiao@cse.ust.hk PhD @ HKUST
Backup Slides
51
1. Passive Radio Map Construction Sensing Zone & Dead Spot 52
Hypothesis 2 Hypotheses 2/2 Ø CSI can be leveraged to distinguish human in different locations. Direct light-of-sight (LOS) blocking: the entity is located exactly be tween the AP and DP such that blocks the direct LOS transmitting link; Indirect non-light-of-sight (NLOS) reflection: the position of entity lies beside the LOS link and influences the multipath propagation of RF signals. 53