Aerospace and Electronic Systems Magazine May 2017 - 15

Liu et al.

Figure 20.

Stationary targets marked on the map.

MUSIC and the Capon algorithms may fail to obtain the right DOA
estimation of targets with a low SNR, while targets with a big SNR
might be still detected. Since there only four antenna elements in
SNLA, we can only find the three biggest targets at most; other
small targets are lost.
Catenary poles at a near distance have a high SNR and can
be estimated correctly, while catenary poles at a far distance are
lost. On the one hand, the power of the transmitter is relatively
low and reflections of catenaries at a far distance are so weak
that they are swallowed by the strong echoes of buildings and
other obstacles at a similar distance. Limited by the small freedom degree of the receive array, only the targets with strong
echo, such as buildings, are detected. On the other hand, catenaries at a far distance are blocked by buildings near the turning
of the track 1,000 m from the radar facility. Catenary poles at a
distance farther than 1 km are lost in this experiment, while the
strong echo of targets such as buildings at a similar distance can
still be detected. As shown in Figure 20, the white building at
a distance of about 1,300 m and the red building at a distance
of about 2,000 m are located in the direct line of sight from the
radar position, and only parts of these buildings are detected,
because other parts of these buildings are blocked by obstacles
between the radar facility and them. It seems to be possible to
detect objects with strong reflection ahead on the track in a builtup area with weak reflections coming in through the sidelobes,
while it is a problem to detect weak objects on the track when
there are too many big targets in a similar range, and it is a problem to detect targets that are blocked by other obstacles.
In order to obtain better identification of the detected targets in
complicated situations, the detected targets are marked on the map
(Figure 20). Compared with the positions of known objects such
as buildings and catenary poles that will be marked on the map in
our future work, we can have better identification of whether a detected target is an obstacle. Figure 20 shows that targets at a short
distance are mostly catenary poles near the track and targets at a far
distance are mainly buildings.

DISCUSSION AND CONCLUSION

ACKNOWLEDGMENTS
The authors thank the anonymous reviewers for their constructive
suggestions. The authors also thank Mr. Chen Li and his company
for financial support and cooperation.

REFERENCES

In this article, we have designed, implemented, and verified a lowcost CA radar system at a freight rail line. Because of the LFMCW
MAY 2017

signal with wide bandwidth, the CA radar can separate obstacles
range from a few tens of meters up to 3.0 km. Two key contributions have been verified above. One is that SULA technique has
been utilized in CA radar to overcome the angular ambiguity as a
result of a sparse array formed by only eight antennas. Another one
is that the MUSIC algorithm has been developed for the sparse array to obtain high angular resolution, which can provide the ability
to separate closely spaced targets. Experimental results show that
this system can provide a satisfying azimuth resolution and range
resolution in localizing both a moving train on the adjacent track
and stationary catenary poles along the railway. The maximum
monitoring distance of the proposed CA radar is 1 km, which is
limited by the transmitting power and the complicated environment.
However, several problems remain unsolved. First, an experiment on a moving platform has not yet been conducted. This
will be carried out at the next stage, and corresponding clutter suppression techniques will be presented. Second, antennas
used in the experiment are horn antennas, which are not suitable
for installation on the head of a bullet train in practice. Conformal antennas are being designed and will be used in the future.
Third, because only eight antennas are used in the receive array, the noncoherent targets in a range-Doppler resolution cell
should be fewer than seven so as to obtain correct azimuth estimation of the targets. This condition may be satisfied by appropriately improving the range and Doppler resolution and using a transmitting antenna with a narrow beamwidth to suppress
strong reflection from sidelobes. Finally, target map matching
for better recognition of the targets is an issue that we will study
in our future work.

[1]

IEEE A&E SYSTEMS MAGAZINE

Lal, S., and Chowdhury, S. An FPGA-based 77 GHZ MEMS radar
signal processing system for automotive collision avoidance. In Pro-

Table of Contents for the Digital Edition of Aerospace and Electronic Systems Magazine May 2017