Signal Processing - March 2016 - 65

Reference [34] demonstrated that a 2-D-map can be constructed with no prior information about the scenario other than
the absolute positions of two fixed anchors. Figure 5 shows an
illustrative example of this SLAM approach, which has been
obtained from the same measurement data as the CDFs in Figure 4. Gray squares indicate the positions of some expected
VAs computed from the floor plan. Discovered VAs are shown
by red (Anchor 1) and blue (Anchor 2) square-cross markers;
their marginal position covariance matrices are indicated by
standard deviation ellipses, enlarged by a factor of 30 for better
visibility. The corresponding true agent trajectory is indicated
in gray. The current estimated agent position is shown by the
red dot; its standard deviation ellipse is in black (enlarged by
a factor of 100).
As shown in the figure-after 68 time steps-a number
of relevant VAs have been identified that match very well
with the geometrically computed VAs. Some of these VAs
have only been associated for a few time steps, corresponding to rather large variances due to large geometric dilution
of precision and/or low SINR values (e.g., MPC "A1 door
and left window"). On the other hand, some VAs already
have converged accurately to their true location (e.g., MPC
"A1 blackboard"). Falsely discovered VAs often show a very
large variance of their associated amplitudes, corresponding
to a low SINR. Thus, their influence on the tracking process
remains limited. The overall tracking performance almost
matches up the performance of the approach discussed in
Figure 4, and 90 % of the errors are within 4.4 cm. Assuming the availability of side information, e.g., from an inertial measurement unit, we expect that the robustness of this
SLAM algorithm against divergence gets even higher.

Passive localization exploiting multipath
As mentioned previously, passive localization has the great
advantage that no specific user compliance is necessary-in
other words, the person to be helped does not need to remember to carry a specific device. At the same time, the passive
principle makes it more challenging to handle multipath.
While in an active system, localization can be achieved based
on the triangulation with LOS paths, in passive localization
we have to base it on "direct paths" that go from the transmitter, via reflection at the target, to the receiver. Furthermore,
these "direct paths" are embedded in background paths-
paths that propagate from transmitter to receiver without participation of the target-and the delay of the background
paths can be larger or smaller than those of the direct path.
Second, there are also indirect paths, which involve reflection
at both target and additional objects. And analogously to
active localization, where the LOS path might be shadowed
off, the direct path might be blocked. This overall makes target localization much more difficult.
Despite these difficulties, passive vital sign monitoring has
a long history (the main motivation used to be in a military/
surveillance context, but the principles can be applied to AL
as well). Narrowband Doppler radar was already being used
to detect the presence of breathing beginning in the 1970s.

However, this does not allow the localization of the breathing
person and is of somewhat limited utility for AL applications.
A more promising approach seems to be the use of wideband
multiple-input, multiple-output (MIMO) radar. Reference [40]
demonstrated a prototype that could precisely localize a person
and track the small-scale movement of the chest that occurs
during breathing from a distance of several meters away. This
was enabled with a sounding waveform extending over 7-GHz
bandwidth (within the UWB band from 3 to 10 GHz), combined with an eight-element transmit array and high-resolution
(iterative maximum-likelihood estimation) evaluation. Most
noteworthy, the localization can be achieved without a direct
path, as long as the environment (location of walls, etc.) is
known. The figures in [40] demonstrate the relative location of
the echo reflected from the head and chest when the target is
breathing in/out.
The situation is more difficult when more than one possible target is present. In contrast to active devices that send
out unique signatures and allow identification of all associated signals, it is difficult (and often impossible) to distinguish between the MPCs belonging to different targets. Such
multitarget localization is another difficult but important
problem-obviously, in many AL situations (e.g., eldercare
homes), multiple targets might be present simultaneously,
and if they are moving, their trajectories might intersect.
From an algorithmic point of view, we have to distinguish the
cases where transmitter and receiver have multiple antenna
elements (and can resolve directions of the echoes), versus
the (much more difficult) case of distributed single-antenna
transceivers (e.g., [41]).
In addition to localization and tracking, radio signals may be
used for the reconstruction of a three-dimensional map of the
surrounding environment, e.g., to assist people with impaired
vision capabilities. This is, of course, strongly related to the
mapping task of the SLAM approach. The passive reflections
of the radio waves from the environment are exploited together
with additional reflections from targets and walls. A single sensor through-the-wall radar with data association is discussed
in [25], multipath-assisted through-the-wall imaging in [26].
The suitability of UWB radars for mapping, imaging, and also
breathing detection was shown in [42]. Recently, the concept
of personal radar has been proposed as a smartphone-centric
low-cost solution for the navigation and mapping problem
[43]. Personal radar could be an additional feature offered by
5G smartphones, exploiting mm-wave massive antenna arrays
with electronic pencil-beam steering capability and high ranging accuracy. The small wavelength of mm-wave technology
permits the packing of a massive antenna array in pocket-size
space [44]. In fact, mm-wave technologies could provide a
most suitable platform for the purpose of high-accuracy localization for AL, as discussed next.

Analysis of mm-wave localization systems
for assisted living
Insights gained so far show the promising features of a multipath-assisted indoor localization system. However, the

IEEE SIgnal ProcESSIng MagazInE

March 2016

Table of Contents for the Digital Edition of Signal Processing - March 2016