Continuous Wave Airport Surface Movement Radar
Author: Andre Adrian
Version: 6aug2006
Introduction
An Airport Surface Movement Radars (ASMR) is a simple radar. A rotating
antenna turns with 10 Revolutions per Minute (RpM) or faster. A
magnetron generates a Radio Frequency (RF) pulse of 10GHz. This is the
technology of 1960. The receiver is quite modern in terms of Moving
Target Indication (MTI) and clutter suppression, but there is only so
much you can do with a single frequency RF pulse that is more or less
coherent.
A big reason for this ASMR technology is cost. This paper discusses how
a modern ASMR can be build for the same price but with (hopefully)
better performance. The radar design is based on Continuous Wave (CW)
solid state transmitter and Field Programmable Gate Array (FPGA) hit
processor. The waveform is pseudo noise. The target identification is
done with a digital 1024 elements tapped-delay-line.
The Least Means Square (LMS) algorithm in fixed point is used.
At the moment the project is in concept phase: the market is scanned
for available Commercial of the Shelf (COTS) components to build a
proof-of-concept radar with a RF frequency of 2.4GHz and 100 taps.
Equipment for this RF frequency is very
cheap because of IEEE
802.11b Wireless LAN (WLAN) .
The project has no resources yet. No money, no manpower. But this
concept paper will hopefully open the money box...
ASMR Requirements
Range = 3Km
Range Resolution = 3m = minimum 50MHz bandwidth = minimum 100
Megasamples/s
Rotation = 30RpM = 2s scantime
Azimuth Resolution = 0.17° = 2048 steps per 360°
Radio Frequency = 10GHz
Minimum Target speed = 10km/h
Maximum Target speed = 270km/h
One range row is illuminated for 1ms. Therefore 100000 samples are
available for filter learning.
ASMR Proof-of-Concept Requirements
The Proof-of-Concept radar shall show the working of the LMS algorithm
for noise waveform CW radar in a near-live environment. Important
topics to test are:
- successful target detection of small targets (cars, vans)
- successful interoperation between hit processor (FPGA board) and
MTI processor (Computer running Linux)
- successful Moving Target Indication (MTI) with non-moving target
suppression
Range = 300m
Range Resolution = 3m = minimum 50MHz bandwidth = minimum 100
Megasamples/s
Rotation = 30RpM or slower
Azimuth Resolution = 0.17° or larger
Radio Frequency = 2.4GHz
Minimum Target speed = 10km/h
Maximum Target speed = 270km/h
Plant Simulation
For a radar, plant is the medium in which the radar signals travel on
their way from transmitter to target to receiver. Another term for
plant is channel, which is used in data communications.

The relation between transmitter, target and receiver can be simulated
with two target dependent properties and the superposition principle.
The distance between target and antennas determines the delay from signal transmit to signal
receive. The Radar Cross Section (RCS) together with the distance
determine the attenuation of
the signal. If there are several targets that reflect radar energy back
to the receiver, the receiver gets the sum
(superposition) signal.
Discrete Plant Simulation with Tapped Delay Line
The discrete plant simulation is based on a tapped delay line. The
distance between radar and target is simulated with a delay element
line. The attenuation is simulated with multiplication with a delay
element dependent constant factor w. The superposition is simulated
with addition of all values. Below is the typical structure of a
tapped-delay-line or Finite Impulse Response (FIR) filter. The top row
shows the attenuation values w, the second row contains the delay
elements, the third row performs the attenuation multiplications and
the last row calculates the superposition additions.

Radar as a System Identification Problem
The Least Means Square (LMS) algorithm can be used for system
identification: Estimate the w
values if only TX and RX signals are given. The w values form a "range
row". After processing of 30000 to 100000 TX and RX samples the w
vector has "learned" the delays and attenuations of the targets. The
LMS filter is a FIR filter plus the adaptation (learning) structure.
The output of the FIR filter is now the estimated RX signal. This is
compared to the received RX signal. The resulting error signal e is
attenuated by a constant value mikro. This mikro times e factor is now
used to update all w values with the formula new_w = old_w + mikro * e
* x. The value x is the delayed TX signal.
The LMS filter is the most simple adaptive filter. It has good
robustness, but slow learning speed. Because of the large number of
samples in the CW ASMR the slow learning speed is no problem.

Why Noise waveform radar?
The LMS algorithm performs system estimation. One important LMS
application is echo cancellation. To speed up the learning of an LMS
echo canceller, often pseudo noise is transmitted in front of real
data. From information theory we know that white noise contains the
most
information of all wave forms. White noise gives the fastest learning
of an LMS filter.
Still we should see the differences between white noise we transmit and
noise we receive. The received signal is distorted by thermal receiver
noise and noise from the medium, e.g. other transmitters on the same
radio frequency. Signal distortion has a negative impact on range and
azimuth resolution and even on useable range. In this aspect noise
waveform radar is equal to pulse radar or CW-FM radar.
We can assume that for some distortions the CW-FM radar is superior,
and for other distortions CW-Noise waveform has the leading edge.
Important for the user of the radar are not all kinds of distortions,
but the distortions that happen in real live under working conditions.
The author assumes that for real live distortions CW-Noise waveform
outperforms CW-FM chirps because noise is maximum information. You
can't beat white noise!
CW-Noise waveform system diagram
The CW design needs a TX and a RX antenna. These antennas are mounted
on a rotating antenna platform. The encoder gives the azimuth
information to the hit processor. The noise waveform transmit signal is
generated with an pseudo noise generator that changes the frequency of
a voltage controlled oszillator (VCO). This design creates a single
sideband RF signal. The power amplifier (PA) feeds the TX antenna. A
small part of the transmit signal is coupled to the left down converter
mixer. Because of range resolution the intermediate frequency (IF)
amplifier bandwidth is 50MHz or even
more. The design is wideband, not ultra wideband (UWB).
IF amplifier (AMP) and low pass filter (LPF) prepare the downconverted
TX signal for the Analog Digital Converter (ADC) with is part of the
demodulator.
The receiver signal flow is very standard: low noise amplifier (LNA),
mixer, IF amplifier, low pass filter, ADC and demodulator.
The heart of the hit processor is the LMS filter. The 1024
LMS filter w values form the range cells. Together with the 2048
different azimuth
values there are 1024*2048 = 2.1 million radar bins. With a scan time
of 2 seconds (antenna rotates with 30RpM) and 2048 azimuth values one
range row is illuminated for 1 millisecond. In this time the w values
in the actual "range row" change. For the next 1999 milliseconds this
range row rests in memory.
The radar bin information in the hit processor is copied to the moving
target indication (MTI) processor, a dual CPU personal computer running
Linux as operating system. One CPU does the MTI process, the other CPU
does the radar bin to raster scan display conversion process.

Project Risks
At the moment there are no CW-Noise waveform ASMR on the marketplace to
buy as COTS equipment. Maybe this is a sign that the idea is not
working. Next to this "I don't want to be the first one" there are some
more risks:
- What kind of modulation? The modulation must preserve the white
noise nature of the transmit signal over upconverting to radio
frequency and downconverting to baseband.
- The CW approach changes "big pulse power" with "signal
integration over time". The integration gain is necessary to get the
wanted range with a low power solid state transmitter. How much
integration gain is in the LMS filter in a radar application is open.
- FPGA performance. A 1024 taps LMS running with 100 Megasamples
per second is on the edge of performance of today's FPGA chips. The
ASMR needs at least 200 Giga Multiplications per second.
- Hit processor to MTI processor interface. Without data
compression every millisecond 2048 bytes are copied on the interface if
every w value has 16bit size. This is 2MByte per second continuous.
- And the biggest risk of all: THE UNKNOWN RISK THAT NOBODY HAS
EXPECTED!