Aerospace and Electronic Systems Magazine May 2017 - 10

Collision Avoidance Radar System for the Bullet Train

Figure 10.

Typical situation (the radar facility is stationary).

However, it is pointed out in [31] that the application of this
algorithm to a practical real-time system is usually limited for
two reasons: phase and gain errors in the different receiving
channels can seriously degrade the performance of the MUSIC algorithm and computational complexity in searching the
whole space (e.g., −90° to 90°) completely might be unacceptable in a real-time system. Still, the MUSIC algorithm is
adopted here because the SNLA with such a big aperture is
not so sensitive to the modeling errors, and careful calibrating
of the system with self-calibration techniques [15], [32], [33]
and calibration techniques with calibrating sources [14] can be
used to eliminate or minimize phase and gain errors, which
can improve the performance of the DOA estimator in practice.
Another reason is that the radar observation region is narrow
(−3° to 3°), and there are only eight elements in the receive
array; hence, the computational complexity of target localization is acceptable using the high performance of digital signal
processors.
For low calculation complexity, a simple but effective calibration technique that uses a calibrating source is applied to calibrate
the SNLA. First, the detected angle of the calibrating source will
be obtained using the radar facility, then it is compared with the
known angle of the calibrating source, and as a result, the array
calibrating coefficient for that moment is obtained. It is reasonable
to suppose that the modeling errors of the SNLA are always stable
and stay almost unchanged for a long time, which means that once
the array calibration coefficient is obtained, it remains effective in
the following processing. To avoid the influence of other stationary
obstacles at a similar range, a moving calibrating source is usually
preferred.

DOA ESTIMATION SIMULATIONS
SIMULATION 1
There are many other techniques available to obtain the DOA estimation of the targets, such as the DBF technique [34]-[36] and
a superresolution algorithm like Capon beamforming [30], [37]-
10

Figure 11.

(a) Single target scenario. (b) Multiple target scenario. (SNR = 30 dB,
snapshots = 40).

[40]. To compare the performance of these techniques in distinguishing closely spaced targets in a practical situation with that of
the MUSIC algorithm, we show the result of a computer simulation in Figure 10.
Assumptions: The radar facility is fixed on the railway, and the
boresight of the SNLA overlaps with the center line of the railway.
After using the genetic algorithm [23], the optimized positions of
the eight elements in the receive array are set at 0, 0.4960, 0.6922,
1.2206, 1.4168, 1.9128, 2.1090, and 2.4412 m. There are two bullet trains on the parallel tracks and two catenary poles 3 km ahead
of the radar. The faulted bullet train A stops on the same track as
the radar facility, and the bullet train B is moving toward the radar
on the adjacent parallel track at a speed of 200 km/h. The DOAs
of these four targets relative to boresight of the radar facility are
[−0.1528, −0.0955, 0, 0.0573] in degrees, respectively. Suppose
the echoes of different targets have equal power, the noise is white
Gaussian but uncorrelated with the echoes, and there are no modeling errors in the receive array.

IEEE A&E SYSTEMS MAGAZINE

MAY 2017

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