1 FM-BASED INDOOR LOCALIZATION 20130107 TsungYun 2 Outline • Introduction • Architecture • Experiment • Result • FM-based Indoor localization • Temporal Variations • Different Buildings • Fine-Grain Localization • Conclusion 3 Introduction • The major challenge for fingerprint-based approach is the design of robust and discriminative signatures • Existing approaches exhibit several limitations • This paper study the feasibility of leveraging FM broadcast radio signals for fingerprinting indoor environments 4 Introduction • WiFi - The most popular design • the high operating frequency makes it susceptible to human presence • Optimized by frequency hopping to improve network’s throughput (RSSI values change across WiFi channels) • WiFi RSSI values exhibit high variation over time • the area of coverage of a WiFi access point is significantly reduced due to the presence of walls and metallic objects, easily creating blind spots (i.e. basement, parking lots, corners in a building, etc.) 5 Introduction • FM broadcast radio • No need for extra deployment • Lower frequency • Stronger signal strength • Lower power consumption • Outdoor localization • Zip code level [10] • Tens of meters [8] 6 Introduction • FM-Based indoor localization • internal structure of the building can significantly affect the propagation of FM radio signals • achieve similar room-level accuracy in indoor environments when compared to WiFi signals • FM and WiFi signals are complementary • their localization errors are independent • Combine FM and WiFi 7 Architecture • Training stage • Fingerprint database • Site survey artificially • Crowd-sourced from freely services (e.g. Google) • Positioning stage (Testing) • Find the closest fingerprint (1-NN) • Use Euclidean and Manhattan distance 8 Architecture 9 Architecture • Augment the WiFi wireless fingerprint to include the RSSI information obtained by FM radio signals • Extract more detailed information at the physical layer for FM radio signals • SNR (signal to noise): 0~128 db • Multipath: 0~100 • Frequency offset: -10~10 10 Architecture 11 Experiment • Three different buildings • Office building • 3 different floors • Totally 119 small rooms (9 ft x 9 ft) • 434 WiFi APs • Shopping mall • 13 large rooms of varying size and shape • 379 WiFi APs • Residential apartment • 5 different rooms • 117 WiFi APs 12 Experiment 13 Experiment • Hardware • WiFi Link 5300 from Intel • SI-4735 FM radio receiver from Silicon Lab • Data collection (the official building) • 3 random point each rooms • collect 32 FM & M WiFi signals each location • • (RSSI, SNR, MULTIPATH, FREQOFF) (WiFi signal) • each fingerprint • 3 data set A1, A2, A3 14 Result – FM-based Indoor localization • Focus on RSSI value only • Use 2 dataset as database, the other as testing data (the office building) • Average accuracy across 3 combinations • FM and WiFi RSSI values achieve similarly high room-level accuracies (close to 90%) 15 Result – FM-based Indoor localization • The localization errors in terms of physical distance are lower in the case of WiFi 16 Result – FM-based Indoor localization • 3 squares correspond to the 3 floors profiled 17 Result – FM-based Indoor localization • Leverage additional information at the physical layer (SNR, MULTIPATH, FREQOFF) to generate more robust FM signatures 18 Result – FM-based Indoor localization • Combining all signal indicators into a single signature achieves higher accuracy than any individual signal indicator 19 Result – FM-based Indoor localization • distance matrix (c) appears to be significantly less noisy 20 Result – FM-based Indoor localization • Combining FM and Wi-Fi 21 Result – FM-based Indoor localization • FM localization errors are not correlated with the WiFi errors • Using more FM indicators removes many of the localization errors by FM RSSI 22 Result – FM-based Indoor localization 23 Result – FM-based Indoor localization • All the erroneously predicted rooms are on the same floor and nearby the true rooms 24 Result – FM-based Indoor localization • Sensitivity to number of FM stations • About 30 FM stations are required 25 Result – FM-based Indoor localization • Sensitivity to number of WiFi APs • About 50 WiFi APs are required 26 Result – FM-based Indoor localization • Combine WiFi & FM signals • 50 WiFi APs and 25 FM stations are required 27 Result – Temporal Variations • FM • Continuous Monitoring of FM Signals Over Ten Days 28 Result – Temporal Variations • Using ten days data as testing data • FM signals are stable 29 Result – Temporal Variations • WiFi • Collect four additional sets of fingerprints on the second floor on four different days 30 Result – Temporal Variations • Temporal variations lead to noticeable degradation of accuracy in WiFi case • FM signatures seem to be less susceptible • Adding more datasets into the database can lead to notable gains in the localization accuracy • A bigger fingerprint database can better cope with temporal variations 31 Result – Different Buildings • Shopping Mall • 5 data set on three days (Weekends & Wed.) 32 Result – Different Buildings • Shopping Mall - 5 data set on three days (Weekends & Wed.) • The ceilings are taller and the rooms are sparser and bigger => like outdoor environment • FM signatures perform slightly worse compared to the office building • WiFi signatures perform significantly better • more fingerprints in the database increases localization accuracy 33 Result – Different Buildings • Residential Building • 2 data sets on two days, different FM stations • localization accuracies are independent of the building type • FM based indoor localization approach is applicable to other geographic regions with different FM broadcast infrastructure 34 Result – Fine-Grain Localization • More data collection (2-nd floor of the official B.) • 100 locations along the hallway • Distance between two adjacent locations is one foot • 3 data sets in 3 different days • Leave one out evaluation • use one and only one location at a time from the dataset as the testing fingerprint • Use the other 99 signatures as database 35 Result – Fine-Grain Localization • Each location is identified as one of its two neighbors on the line in terms of FM • WiFi RSSI signatures exhibit larger errors 36 Result – Fine-Grain Localization • • FM RSSI signatures have the necessary spatial resolution For more accurate fingerprinting, even better than WiFi signature 37 Result – Fine-Grain Localization • Temporal Variation • FM still outperforms WiFi significantly • Device Variation • Data set 3 is collected by a different FM receiver • Localization error doesn’t increase significantly 38 Conclusion • Propose to exploit additional information at the physical layer to create more reliable fingerprinting of indoor spaces • Demonstrate that FM and WiFi signals are complementary in the sense that their localization errors are independent • Study in detail the effect of wireless signal temporal variation