Aerospace and Electronic Systems Magazine July 2018 - 56

LTE CommSense for Object Detection in Indoor Environment
1. Four distances of the human target from the SDR platform in
indoor environments were chosen.
2. LTE DL signals were captured with and without the human
target for those four distances, keeping the indoor background
fixed. Figure 12 shows the two distances of the human target
from the SDR in the indoor environment.
3. Five readings were taken, viz. one for only the background
with no human target present and the other four corresponding
to a human target present at four distances: 0.5, 1, 2, and 3 m.
4. The channel estimates for all cases were evaluated.
5. PCA was performed on the dataset for dimensionality reduction.

Figure 12.

Human target present in the indoor environment at different distances
from the SDR platform.

in the case of rotation angle a = 0.6. Therefore, TFR for a = 0.6
can be used to visually distinguish the presence of a human in an
indoor environment from the absence of a human.

Effect of Varying the Distance of the Object from the Sensing
System
The objective of this experiment is to find the effect of changing
the distance of the object, i.e., the human target in this experiment,
on the distinguishing capability of LTE CommSense in an indoor
environment. The following experimental procedure was adopted
for this purpose:
56

To select how many principal components are sufficient for
this case, the sorted eigenvalues and their cumulative energy
contribution are plotted in Figure 13a. The largest two principal
components are plotted in a 2D scatterplot and shown in Figure
13b. The largest three principal components are plotted in a 3D
scatterplot and shown in Figure 13c. In the scatterplots in Figure
13, "black dots" and "blue circles" represent the lowest distances
(i.e., 0.5 and 1 m). "Cyan crosses" represent a distance of 2 m, the
"green X " is for a distance of 3 m, and "magenta squares" are representative of no human target present in the scene. Five clusters
can be observed for the five cases. It can be further observed that
the distance between the cluster centers decreases as the distance
of the human target from the SDR platform increases.
Next, SVM-based classification was performed on the collected data, corresponding to the two datasets, i.e., with human
target and without human target. For the different datasets, the distance of the human target from the SDR platform varies. Separate
training and testing datasets were prepared to find the detection
accuracy and corresponding confusion matrix. The nomenclature
of different test cases generated from the database is as follows:
Case$b-r$r. Here, $b is the sequence number of the indoor background being considered and r denotes the number of readings for
the considered background. For example, Case3-r100 is the third
dataset of 100 readings with and without a human target. Here,
the human target is at a particular distance from the CommSense
receiver platform.
For each of the preceding cases, the performance accuracy and
confusion matrix were evaluated. An average recognition accuracy
of 88.9% was achieved in this case.
The two standard measures to indicate the identifying power
of a detection methodology are false rejection rate (FRR) and false
acceptance rate (FAR) [20], [21]. FRR is considered a type I error,
and FAR is considered a type II error. FAR is fraction of the falsely
accepted patterns (i.e., human target not present but classified as
human target present) divided by the number of all impostor patterns. The fraction of the number of rejected client patterns (i.e.,
human target present but not detected) divided by the total number of client patterns is the FRR. According to Bolle et al. [20],
false accept (FA) is an act of deciding that a category is legitimate
when it is an impostor in reality. The frequency at which FA errors are made can be denoted as FAR. If NFA signifies the number

IEEE A&E SYSTEMS MAGAZINE

JULY 2018

Table of Contents for the Digital Edition of Aerospace and Electronic Systems Magazine July 2018