IEEE Signal Processing - May 2018 - 77

There are other radio analytics approaches in the literature. One is to leverage the radar concept to utilize the time
of flight (ToF) for locationing [60], [71] and breathing monitoring [78]. There are a few drawbacks to obtaining the ToF,
including the need to perfectly synchronize the devices and
scan through a large spectrum, often with a range of several hundred megahertz or more. Thus, the hardware cost
can be higher than with simply using commodity Wi-Fi
or LTE devices. Furthermore, when in different environments, especially under NLOS conditions with different
construction materials, some calibration or training must
first be performed.
The presented TR techniques, however, do not require an
LOS path between the transmitter and the receiver and can
work as well in NLOS conditions. This leads to more flexibility in deploying the devices; e.g., only a few devices can cover
a large area for positioning and tracking.
Another common practice to obtain the multipath CSI
values is by means of ultrawide-band signal transmission
[1], [14], which offers an ultralarge bandwidth to obtain a
large number of multipaths. But because of regulations,
the very low transmission power requirement may significantly limit the coverage and increase the infrastructure
cost. As such, this solution is often used for close-range,
LOS applications.
In the following, we will discuss in detail the existing
approaches on indoor positioning and tracking, event detection, human radio biometrics, and breathing monitoring using
RF signals.

Indoor positioning and tracking
Although GPS has been widely deployed as a successful
solution to outdoor positioning, many questions still remain
concerning indoor positioning [1], [2], because the interior
structure of a building is much more complex than the outdoor environment. Various indoor positioning techniques
have been proposed, which can be mainly classified into four
categories, based on vision, acoustic signals, IMU, and radio.
The vision-based approaches, such as camera, laser, and so
forth, suffer from high deployment and hardware costs, the
need for sophisticated calibrations, and limited coverage,
although a very high accuracy may be achieved. The acoustic-based techniques have only a limited range and are not
scalable to a large number of users. The positioning accuracy
of IMU-based methods is limited mainly by gyroscope drifting and poor moving-distance estimation.
Because of the ubiquitousness and low cost of RF devices,
radio-based IPSs, especially those founded on Wi-Fi, have
become popular nowadays. Based on the principles they use,
the radio-oriented indoor positioning schemes can be further
categorized into two classes [1]: triangulation based and fingerprinting based. The features they utilize can be obtained
either from the medium access control layer information, e.g.,
RSSI readings and the time stamps of the received packets
at the receiver, or from the physical layer information, e.g.,
the CSI.

Triangulation based
In the triangulation-based schemes, either the distance [19]-
[21] or the angle [22]-[24] between a device and several anchors
is estimated, and the position of the device can be inferred by
performing geometrical triangulation. The distance between
the device and anchor can be estimated from the RSSI decay
[17] or from the ToF of the transmitted packets that can be
extracted from the time stamps of the received packets [25].
The angle between the devices can be obtained by examining
the features of the CSI values received by multiple receiver
antennas, and then the angle of arrival of the direct path to the
target can be found. However, the main challenges for the triangulation-based approaches are the blockage and reflection
of the transmitted signal, since only the signal coming from
the direct path between the device and anchor can be used
for positioning.

Fingerprinting based
In the fingerprinting-based schemes, the features can be
obtained either from the vector of the RSSIs [17], [26] or the
detailed CSI values [27], [28], [30] from a specific location to
all of the anchors in range. A major drawback of the fingerprinting-based approach is that the features of the mapped
locations are susceptible to environmental dynamics and the
database of the mapped fingerprints needs to be updated before
it can be used again. In addition, the computational complexity
of fingerprinting-based approaches is high, and thus they are
not feasible for low-latency applications.
Compared with these fingerprinting schemes, the centimeter-accuracy IPS discussed in this article fully exploits the
spatial and frequency diversity of the multipath CSI to increase
the effective bandwidth and enhance the robustness against
environmental perturbations. The complexity of the feature
calculation based on TRRS is also lower than most existing
approaches. With the centimeter-accuracy IPS, many applications that need precise positioning can be enabled. For example, one may be able to design industry production lines that
can accurately locate different parts, align them, and assemble
them with much less human intervention. The mapping-free
indoor tracking system can further reduce the implementation complexity, since it requires no feature matching as in
the fingerprinting-based schemes and no recalibration. Most
important, both of the two approaches can perform equally
well under LOS and NLOS.

Indoor event detection
Common features of RF signals to identify variations during
signal transmission for indoor event detection include the
RSSI and CSI. Because of its susceptibility to environmental
changes, the RSSI has been applied to indicate and further
recognize indoor activities [9], [10], [51], [52]. However,
because the RSSI is coarse-grained and can be easily corrupted by the multipath effect, RSS-based sensing systems often
require an LOS transmission, resulting in limited accuracy in
indoor activity detection. Furthermore, CSI, including the
amplitude and the phase, is now accessible in many commercial

IEEE Signal Processing Magazine

May 2018

Table of Contents for the Digital Edition of IEEE Signal Processing - May 2018