Signal Processing - September 2017 - 128

Table 1. The teams that participated in the 2014 and 2015 Microsoft Indoor Localization Competitions. Teams in each category
are listed in order of the localization accuracy they achieved (highest to lowest) that year. (continued )

Infrastructure free

Team

Team's Affiliation

Country

Technical Approach

Global
Rank

Maróti et al. [72]

University of Szeged

Hungary

Radio interferometry

Lin et al. [73]

National Chiao-Tung University

Taiwan

Mobile device encounterings

Nikodem et al. [53]

Wroclaw University of Science and
Technology

Poland

2.4-GHz ToF

Deora and Krishnamachari [74]

University of Southern California

United States

Zigbee beacons

Kuo et al. [47]

University of Michigan

United States

Visible light

Mirshekari et al. [75]

Carnegie Mellon Unversity

United States

Structural vibration

Sánchez et al. [45]

European Commission, Joint Research
Center

Italy

Lidar

SPIRIT Navigation [76]

SPIRIT Navigation

Russia

Wi-Fi + magnetic + IMU
fingerprinting

Guimarães et al. [39]

Fraunhofer Research Center

Portugal

Wi-Fi + magnetic + IMU
fingerprinting

Zou et al. [60]

Nanyang Technological University/
University of California, Berkeley

Singapore/United
States

Wi-Fi + IMU fingerprinting

Wu et al. [77]

University of Windsor

Canada

Wi-Fi + IMU fingerprinting

Herrera et al. [78]

Navix

Mexico

Wi-Fi + IMU fingerprinting

Ghose et al. [64]

Tata

India

Wi-Fi + IMU fingerprinting

localization, fingerprints need to be carefully engineered so that
even nearby locations have sufficiently different fingerprints. To do
so, researchers have exploited more physical layer signal quality
indicators that go beyond RSSI [i.e., signal-to-noise ratio (SNR),
frequency offset, and more] and have also expanded fingerprints
to include widely available wireless signals that go beyond WiFi, such as FM and TV signals [11]-[18] (Figure 1).
The major challenges in implementing wireless fingerprinting are the overhead of the manual process for building
the fingerprint database and the inherent noise in the wireless
signals that can affect localization accuracy. Wireless signals
change over time, and they are affected by the number of people in the indoor environment as well as by the placement of
big objects, such as furniture. In addition, access points disappear or appear continuously, affecting the stability of wireless fingerprints.
An alternative to wireless fingerprinting is pedestrian dead
reckoning (PDR) [39]. PDR leverages the on-board sensors of
mobile devices, such as accelerometers, gyro, and compass,
to count the steps and turns of the person holding the device.
Assuming the person entered the indoor environment at a
known point (i.e., entrance to a shopping mall), the number of
steps and turns the person took can be used to estimate his or
her location in the space at any given time. If the map of the
indoor environment is available, PDR-based techniques can
achieve higher accuracy by constraining human movement
based on the map.
The major challenge with PDR is that sensor data tend
to drift over time. Even though initially very accurate, as
128

the person continuously moves, the noise in the inertial
sensors accumulates over time, impacting the overall localization accuracy. More recently, PDR techniques have been
used in a complementary way to wireless fingerprinting.
The inertial sensor data is used to filter out noise in the wireless fingerprints and vice versa, leading to more reliable
indoor localization.

Hardware-based approaches to indoor location
Ranging primitives
Hardware-based approaches rely on ranging, the process of
estimating the distance between two devices (i.e., between
the device that needs to be localized and one of the custom
hardware devices that have been preinstalled in known locations). Figure 2 shows the three fundamental techniques
used for estimating this distance. In all cases, distance estimation is achieved by accurately timestamping the transmission and reception of wireless signals exchanged
between the participating devices. To accurately timestamp
these signals, all participating devices need to be tightly
time synchronized.
The most common way of estimating the distance between
two devices [Tx and receivers (Rx) in Figure 2] is for these
devices to measure the ToF of a single wireless transmission.
By timestamping the wireless signal at the time of tranmission
(t1 in Figure 2) and at the time of reception (t2 in Figure 2), one
can measure the ToF, the time it takes the wireless signal to
travel from one device to the other. If the speed at which the

IEEE SIGNAL PROCESSING MAGAZINE

September 2017

Table of Contents for the Digital Edition of Signal Processing - September 2017