Aerospace and Electronic Systems Magazine May 2018 - 15

three synchronized networked nodes and has an operational frequency of 2.4 GHz. The three nodes were configured in a receiveonly mode, while a commercial AP was the transmitter of opportunity. The master node (Node 1) distributes a 100-MHz clock signal
to Slave Nodes 2 and 3.
Authors of [10] presented RF capture, a system that captures
the human figure, i.e., a coarse skeleton, through a wall. RF capture uses a combination of a two-dimensional antenna array and
frequency-modulated continuous wave radar chirps to scan the surrounding three-dimensional space for RF reflections.
In [11], the authors implemented a simple radar method to determine Doppler frequency shift of a moving person. They transmitted a 1-kHz tone by using a 5.5-GHz carrier frequency and sampling the received signal at a rate of 200 ksamples/s.
The Wi-Vi system is presented in [4] to track people behind
walls. Wi-Vi is essentially a three-antenna multiple-input and
multiple-output (MIMO) device using 20-MHz WiFi orthogonal
frequency division multiplexing signals. The system consists of
three Universal Software Radio Peripheral (USRP) platforms connected to an external clock so that they act as one MIMO system.
Two of the USRPs are used for transmitting and one for receiving.
It also uses directional antennas to focus the energy toward the
wall or room of interest. MIMO nulling is implemented directly
into the USRP hardware driver so that it is performed in real time.
Postprocessing is performed offline in MATLAB by using the
smoothed MUSIC algorithm to compute the power received along
a particular direction. The Wi-Vi approach requires full control of
both transmitter and receiver; hence, it currently cannot be applied
to commercial WiFi devices.
In [3], a TTW human detection system using CSI from commodity WiFi devices is presented. They apply a principal component analysis-based filtering to clean the collected CSI, and then
they exploit the correlated changes over different subcarriers and
propose a subcarrier dimension-based feature, i.e., the mean of
the first-order difference of eigenvectors. Experiments have been
conducted in two different indoor environments, achieving both a
true-positive rate and a true-negative rate of up to 99%. However,
the experiments have been carried out with the strong assumption
that volunteers walk with a slow or normal speed. No stationary
humans are considered.
MAY - JUNE 2018

EXPERIMENTAL SETUP AND DOPPLER SPECTRUM
ESTIMATION
Experiments have been carried out in three different setups involving three adjacent environments, i.e., Setup 1, Setup 2, and Setup
3, in Room A, Room B, and Room C, as shown in Figure 1. Each
experiment has been performed in a TTW configuration. Particularly in Setup 3, the double TTW scenario has been tested, i.e., the
transmitter and receiver are placed in different rooms from the one
in which the human activities are performed. During the experiments, the following conditions have been considered:
1. Empty: the considered room is empty during the experiment.
2. Presence (static): one person is inside the considered room in
a sitting or standing fixed condition and in different positions.
3. Presence (dynamic): one person is inside the considered room,
walking over different patterns.
The considered rooms have different sizes and are separated by
12 cm of drywall. In each setup, the transmitter AP is located on
the opposite side of the wall of interest.
The proposed system includes a commercial 2.4-GHz WiFi
AP with three antennas acting as transmitter and a laptop having
a WiFi card Intel 5300 with three antennas acting as receiver. The
laptop, running Ubuntu 10.04 LTS, operates by sending Internet
Control Message Protocol (ICMP) echo request packets every Tp
= 10 ms to the AP and waits for an ICMP echo reply from the AP.
CSIs are extracted from the received ICMP echo reply packets
by using a customized firmware and an open-source Linux wireless driver for the Intel 5300 WiFi card [12], [13]. The AP uses a
double-stream transmission; hence, a total of Nch = 6 CSI estimac
c
tions H m = H m (l ) of length Nsub = 30, l = 1, 2, ..., 30, are collected
for each ICMP echo reply packet received at time m. Each element
of a CSI vector H cm represents the complex channel gain for a particular subcarrier, given the channel index c and the time index m.
Then, a unique CSI is obtained by concatenating Nch CSI vectors,
resulting in a global CSI vector H = Hm(l) of length Nel = 180, l = 1,
2, ..., Nel. To remove the power fluctuations of the AP, each global
CSI vector is normalized by its mean value.

IEEE A&E SYSTEMS MAGAZINE

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