Signal Processing - March 2016 - 67

Panel

Discussion and conclusions
This article envisions accurate and robust indoor localization as
a key sensing modality of an AL system. It has been shown that
awareness to the signal propagation conditions enables the
robustness and allows to reduce the needed infrastructure.
Experimental, measurement-based results support the discussion of theoretical findings.
A geometry-based stochastic model of the received signal
allows the derivation of theoretical PEBs and thus provides
the theoretical background for a number of multipath-assisted
localization and tracking algorithms. More specifically, an
environment model, consisting of a geometrical model (based
on VA positions) and a measurement uncertainty model (based
on the SINR of MPCs), yields insight in the potential location
information that can be acquired at a certain position, in a
certain environment. Several algorithms have been discussed
that exploit such information: maximum likelihood localization, tracking filters with data association, and algorithms for
passive localization. The benefit of using this environmental
information has been shown.
Future 5G mm-wave communication systems could be an
ideal platform for achieving high-accuracy indoor localization with this concept. In addition to a large signal bandwidth,
beamforming capabilities are envisioned for such systems, which
can be exploited to make the localization and tracking more
robust and efficient. It becomes feasible to obtain accurate and
robust indoor localization with only a single anchor node in a

Panel

Anchor 1

0
0

2
1

Anchor 1

0
3 4
x [m]
(a)

3 4
x [m]

(b)

Panel

Cupboard

Right Wall

y [m]

Cupboard

y [m]

Cupboard

y [m]

has been computed by using a phased-array beamformer for
each exploited MPC. This is achieved by coherently adding the
signals at the agent-side array positions, taking into account the
relative phase shifts that correspond to the known arrival angles
of the MPCs. The figure exemplary shows that such a processing, envisioned for 5G mm-wave communication systems, can
greatly improve the robustness of the localization, since many
local maxima can be ruled out.

Anchor 1

Right Wall

for microwave band UWB measurements [33], highlighting
the need for online estimation (tracking) of the environment
model, as explained in the section "Simultaneous Localization
and Mapping Using Multipath Channel Information" and [34].
Figure 6(b) shows the PEB corresponding to the estimated
SINRs of Figure 6(a). The PEB is a measure of the potentially
achievable localization accuracy, hence, highly accurate single-anchor localization is possible in this scenario. The PEB
increases only slightly in the OLOS situations due to the still
significant SINR of the LOS component. Even if the LOS
component is not taken into account at all, (NLOS; the red
dash-dotted line), the agent is still localizable at centimeter
level, easily satisfying requirements of most AL applications.
A proper operation in total absence of an LOS indicates the
"good" robustness of the discussed techniques.
Figure 7 shows the likelihood (S1) for a sampled received signal r (t) as a function of position p, evaluated over the floor plan.
Figure 7(a) compares LOS and (b) OLOS conditions with (c)
OLOS with the use of beamforming. The bold black lines indicate
the directions to the anchor, thin black lines the directions to firstorder VAs, and black dashed lines the directions to second-order
VAs. The black diamonds mark the estimated positions of the agent.
Using a maximum likelihood positioning algorithm as in [24], an
error in the centimeter level is achieved (2 cm for the LOS and 3
cm for the OLOS situations). Only a small degradation results in
the OLOS case, as anticipated from the analysis of the SINR values.
The potential use of beamforming shows a different great
advantage: the multimodality of the likelihood function is
reduced, which reduces the risk of converging to a wrong local
maximum. Large modes at locations farther away from the
true agent position are suppressed due to the angular resolution of the array antenna. Note, however, that MPC delays are
still responsible for providing a high accuracy in a direction
orthogonal to the LOS path. Without the processing of multipath, we would see a smooth maximum (along the circle)
instead of a sharp peak. The likelihood function in Figure 7(c)

3 4
x [m]

(c)

figure 7. The likelihood function over the floor plan for (a) LOS, (b) OLOS situation, and (c) OLOS situation with phased-array beamforming. The position error of the MLE is 2 cm and 3 cm for LOS and OLOS situations, respectively. Bold black lines show the directions to the anchors, thin black line the
directions to first-order VAs, and black dashed lines the directions to second-order VAs. The black diamonds mark the estimated positions of the agent.
IEEE SIgnal ProcESSIng MagazInE

March 2016

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