IEEE Signal Processing - May 2018 - 71

open. In Table 1, all of the trained indoor events are listed. For
each transmitter-receiver location and each potential event, we
measured 3,000 CSI samples.
When the receiver was on location B, the transmission
between the transmitter and the receiver was NLOS because
of the absence of a direct LOS link. On the other hand, when
the receiver was on location C, no matter which green dot the
transmitter was on, it was transmitting under LOS, which led
to a dominant multipath component existing in the multipath
CSI. An example of the multipath CSI with respect to the open
and closed states of door D1 is shown in Figure 15(a) and (b),
where the dominant path remains the same and contains most
of the energy in the CSI, while other multipath components
are distinct under the open and closed states. In Figure 15(c),
we show the TRRS between different indoor events. As different colors represent different values of the TRRS, the TRIEDS
is capable of distinguishing between different indoor events
through the TRRS.
The overall false alarm and detection rate for the TRIEDS
and the RSSI-based approach are listed in Tables 2 and 3. We
find that the performance of the TRIEDS is superior to that
of the RSSI-based approach, as it achieves a better detection rate and a lower false alarm rate. Even in an office environment with human activities, the TRIEDS can maintain
a detection rate higher than 96.92% and a false alarm rate
lower than 3.08% under NLOS, as well as a detection rate
higher than 97.89% and a false alarm rate lower than 2.11%
under LOS.
We can see from the experimental results that the TRIEDS
captures the variations in the CSI, maps different multipath
profiles of indoor events into separate points in the TR space,
compresses the complex-valued features into a real-valued
scalar TRRS, and supports the simplest detection and clas-

Table 1. A state list for TRIEDS to detect.
State Index

All of the doors are open.

S i +1

Door D1 is closed and the others are open.

sification algorithms, producing a good performance. When
the surrounding environment is dynamic, changes introduced
to the CSI can be harmful to the TRIEDS's performance,
depending on how significantly the multipath profile changes.
For example, if there was a furniture rearrangement introducing great changes to the majority of the multipaths, the space
would need reprofiling, and the training database would require
updating to maintain detection performance. However, if the
environmental change affected only a few multipaths and the
distinct feature in the CSI was preserved-e.g., when changes
happened far away from the transmitter and the receiver or only
a small object moved-system performance could be guaranteed without reprofiling the space. In [62], the researchers utilized the CSI to detect door states in a single-family house for
two weeks with normal resident activities. As shown by these
long-term experiments, the system can achieve an accuracy of
over 90% without reprofiling all of the training events.

Human radio biometrics
Automated recognition of persons based on their biological
and behavioral characteristics involves well-known features,
including fingerprints, face, irises, and voice [63], [64]. Since
biometrics for individuals are inherent and unique, such traits
have become widely used in surveillance and authentication
systems for human identification. Because of the difficulty in

0.99

0.4

0.98

0.2

0.97

0.96

0.95

0.94

0.6

15
20
Index of Tap
(a)

0.8
0.6
0.4

0.93

0.92

0.2
0

Training: State Index

Amplitude

6 i = 1 , 2 , f, 8 .

0.8

Amplitude

Description

0.91

1
0

15
20
Index of Tap

4
5
6
7
Testing: State Index

(b)

0.9

(c)

FIGURE 15. A feasibility demonstration. The multipath profiles (amplitude part) for different states of door D1 under location C with (a) door open and
(b) door closed. (c) A resonance strength map with the receiver on location C and the transmitter on the first green dot (axis 1) [62].
IEEE Signal Processing Magazine

May 2018

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