Signal Processing - September 2017 - 131

Competition evaluation process

Wi-Fi Access Points

FM Radio Towers
...
...

SSID1
101.5

89.5

Wi-Fi Rx
...

...

Multipath

SNR

SFMi

RSSI

Frequency
Offset

Multipath

...
SNR

SSID3
SS

80.9

FM Rx

SFM1 SFM2 SFM3

SSID2

Frequency
Offset

93.6

RSSI

in indoor localization applications lies in
their ability to very accurately timestamp
the wireless transmissions. Conventional RF
signals reflect in indoor environment structures, resulting in distortions to the direct
path signal, which in turn makes accurate
timestamping challenging. In the case of
UWB (Figure 2), the narrower, but higher
bandwidth, pulses can still be reflected by
the indoor environment structures. However,
because of the pulses being very narrow, it
is now easier to differentiate the direct path
pulse from the reflected ones. As a result,
UWB signals can measure distances with
accuracies as low as a few centimeters.

SWiFii

SWiFi1 SWiFi2 SWiFi3 ... SWiFim

SFMn

Nearest Neighbor
(NN, kNN)

The competition is a two-day event. During
the first day, all teams are given 7 h to set up
their indoor location technologies in the evaluSFM1 SFM2 SFM3 ... SFMn SWiFi1 SWiFi2 SWiFi3 ... SWiFim
ation area. During this time, teams are able to
deploy their custom hardware, if any, and also
Wireless Signal Fingerprint
perform any profiling of the space necessary
(a)
(i.e., fingerprinting, map construction, and so
forth). Each team that requires implementation
Check in
its own custom hardware is allowed to deploy
up to a predetermined (usually ten) maximum
number of devices (i.e., access points, custom
Wireless Signal
Business Info:
RF modules, magnetic field modulators, lightFingerprint
GPS/Address
modulating lamps, and so on) depending on
SFM1 ... SFMn SWiFi1 ... SWiFim + Position
the size of the evaluation area.
At the beginning of the first day, the orgaFM
Wi-Fi
nizers indicate an origin point for the reference
coordinate system that each team should use
Position
to report locations. Locations are reported as
SFM1 ... SFMn SWiFi1 ... SWiFim
Position1
two-dimensional (2-D) or three-dimensional
...
(3-D) coordinates [i.e., (2.12 m, 5.1 m)] with
Wireless Signal
...
S
S
respect to this origin point.
Position N
FMn SWiFi1 ... SWiFim
FM1
Fingerprint
At the end of the first day, the deployed
Fingerprint Database
SFM1 ...SFMn SWiFi1 ... SWiFim
hardware from all teams is turned off, and
all contestants leave the evaluation area. At
Training Stage
Positioning Stage
that time, the organizers mark 20 points on
the floor of the evaluation area and measure
(b)
the x, y, and z coordinates of these points with
respect to the predefined origin point. The FIGURE 1. An overview of wireless fingerprinting. (a) Creating a wireless fingerprint and
ground truth measurements of the evaluation (b) localization using wireless fingerprinting. SSID: service set identifier.
points are taken either using laser range finders or more advanced 3-D laser scanners, providing centimetertems are evaluated based on the average location error across
level or even millimeter-level accuracy.
all 20 evaluation points. The location error for a given point is
During the second day of the evaluation, each team shows
defined as the Euclidean distance between the true and reportup at a preassigned time slot, turns on its deployed system, and
ed coordinates for that point. Even though we recorded locahands the device over to the organizers to be localized. The
tion estimates only on the premeasured 20 evaluation points,
device can be a mobile phone, a tablet, or a laptop depending
the system under test was allowed to continuously perform
on the system under test. The organizers carry the device above
localization. For instance, the system under test could use inereach of the 20 evaluation points, wait for a couple of seconds, and
tial sensors to perform continuous path tracking to improve
record the location reported by the system under test. All syslocalization accuracy.
IEEE SIGNAL PROCESSING MAGAZINE

September 2017

131

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