Australian Government

Department of Defence
Defence Science and
Technology Organisation

*&&

i§P§t
llilllilli^«*.--.^^
"y^&

^*wSwSP^ ■'■ :'

ESS- Signal Processing
in a Semi-active Seeker
Luke Rosenberg, Yiding Hu
and Bill Moran
DSTO-TR-1606

DISTRIBUTION STATEMENT A
Approved for Public Release
Distribution Unlimited

}!
BEST AVAILABLE COPY

20041110 069
 Australian Government
Department of Defence
Defence Science and
Technology Organisation

Signal Processing in a Semi-active Seeker

Luke Rosenberg, Yiding Hu
Weapons Systems Division
Systems Sciences Laboratory
Bill Moran
Department of Electrical and Electronic Engineering,
Melbourne University

DSTO-TR-1606

ABSTRACT

In publicly available generic digital simulations of semi-active Radio Frequency

(RF) seekers, the combined effect of multipath, sea-clutter and other non-
linear noise sources are mainly studied in individual models, while the effect
in the seekers data processing has not been systematically investigated. This
report details these effects and how they have been combined to model the
signal processing blocks inside a semi-active seeker. A basic doppler detection
scheme is then simulated to investigate the consequence of these effects on
target detection.

RELEASE LIMITATION

This work was undertaken as part of a Masters degree in conjunction with

the Cooperative Research Centre for Sensor Signal and Information Processing
(CSSIP).

FöS"-öJ- öl 44-
DSTO-TR-1606

Published by
DSTO Systems Sciences Laboratory
PO Box 1500
Edinburgh, South Australia, Australia 5111
Telephone: (08) 8259 5555
Facsimile: (08) 8259 6567
© Commonwealth of Australia 2004
AR No. 013-164
August, 2004

Conditions of Release and Disposal

This document is the property of the Australian Government The information it con-
tains is released for defence purposes only and must not be disseminated beyond the stated
distribution without prior approval.
The document and the information it contains must be handled in accordance with secu-
rity regulations applying in the country of lodgement, downgrading instructions must be
observed, and delimitation is only with the specific approval of the Releasing Authority as
given in the Secondary Release Statement.
This information may be subject to privately owned rights.
The officer in possession of this document is responsible for its safe custody. When no
longer required the document should be destroyed and the notification sent to: Reports
Officer, DSTO Library, Edinburgh, SA.
 DSTO-TR-1606

Signal Processing in a Semi-active Seeker

EXECUTIVE SUMMARY

In a generic semi-active RF seeker in medium or long range missiles, the combined

effect of multipath, sea-clutter and other non-linear noise sources may cause detrimental
effects on target detection. This report describes models for each of the effects mentioned
and how they can be combined into a single model. A simulation of the combined effects
is then used in a simple detection scheme, so that any change in target detection can be
assessed.
DSTO-TR-1606

IV
 DSTO-TR-1606

Authors

Luke Rosenberg
Weapons Systems Division

Luke Rosenberg received his BE (Elec.) with Honours from

Adelaide University in 1999 and joined the Missile Simulation
Group at DSTO in January 2000. During this time he has com-
pleted a Masters degree in Signal and Information Processing
through Adelaide University and CSSIP. He is presently study-
ing his PhD at Adelaide University, looking into interference
suppression for multi-channel Synthetic Aperture Radar.

Yiding Hu
Weapons Systems Division

(B.Sc 1981, M.Sc 1984, Ph.D 1991, all in physics), 1991-1996,

Research Associate and Senior Research Associate at Physics
Department, University of Newcastle, Australia. Research area
was space physics. 1996-1998, worked on exploration geophysics
and mining technologies for Rio Tinto R&D Division at Mel-
bourne as Senior Scientist. 1998-1999, Senior Scientist at World
Geoscience, Perth, on R&D of airborne geophysical survey and
remote sensing. 1999-present, Senior Research Scientist at the
Weapons Systems Division of DSTO, working on analysis of
weapon systems.

Bill Moran
Dept. of Electrical & Electronic Engineering, Melb. University

Bill Moran has been a Professor of Electrical Engineering at the

University of Melbourne since July 2001. Previously was Pro-
fessor of Mathematics at Flinders University (1991-2001) and at
the University of Adelaide (1976-1991). He has a B.Sc. (Hons)
degree in Mathematics from the University of Birmingham in
the UK and a Ph.D. from the University of Sheffield. He was
elected a Fellow of the Australian Academy of Science in 1984.
From 1991 to 1998, he was a Program Leader in the Cooperative
Research Centre for Sensor Signal and Information Processing.
His research areas in mathematics include harmonic analysis,
representation theory and number theory. More recently he
has added research interest in several aspects of signal process-
ing. He is CEO of Eudoxus Consulting, through which he does
much of his current work with DSTO.
DSTO-TR-1606
 DSTO-TR-1606

Contents
Glossary xi

1 Introduction 1
1.1 Project Aim 1
1.2 Problem Description 1

2 The Semi-Active Seeker 2

2.1 Classic Model 2
2.2 Simplified Model 3
2.3 Doppler 4

3 Simulated Effects 6
3.1 Ideal Simulation 7
3.2 Multipath Reflections 8
3.2.1 Simulation Results 9
3.3 Sea Clutter 11
3.3.1 Main-lobe return 11
3.3.2 Side-lobe return 12
3.3.3 Simulation Results 12
3.4 Receiver Noise 14
3.4.1 Noise Fundamentals 14
3.4.2 Noise Statistics 15
3.4.3 Simulation Results 18
3.5 Complete Simulation 19

4 Doppler Detection 20
4.1 Detection Theory 20
4.2 Optimal Detection 21
4.3 Non-Optimal Detection 21
4.4 Detection Simulation 23
4.4.1 Doppler Filter Bank and the Overlapping Spectrum 24
4.4.2 Detection Range 24
4.4.3 Pause Length 25
4.5 Receiver Operating Characteristics 25

vn
DSTO-TR-1606

5 Conclusions and Future Work 30

References 32

Appendices

A Ideal Simulation Verification 33

B Sea-clutter Frequency Calculation 35

C Optimal Detection 36

D Range Equation 38
 DSTO-TR-1606

Figures
1 Problem scenario with multipath and sea-clutter reflections 2
2 Semi-active seeker block diagram 2
3 Simplified semi-active seeker signal processing block diagram 4
4 Semi-active homing target geometry 5
5 Convolution example 6
6 Ideal scenario 1 7
7 Ideal scenario 2 8
8 Rear multipath components 8
9 Front multipath components 9
10 Scenario 1 with multipath reflections 10
11 Scenario 2 with multipath reflections 10
12 Front and side-lobe clutter for a sea-skimming missile 11
13 Scenario 1 with sea-clutter 14
14 Scenario 2 with sea-clutter 14
15 Simulation probability distributions 17
16 Scenario 1 with receiver noise 18
17 Scenario 2 with receiver noise 18
18 Scenario 1 with all three effects combined 19
19 Scenario 2 with all three effects combined 19
20 Detection regions 21
21 Alternative methods of target detection and verification 23
22 Simplified block diagram for detection and verification 23
23 Comparison of non-overlapping and overlapping filters 24
24 Detection range 25
25 Comparison of pause lengths on detection 26
26 ROC curve for varying overlap - scenario 1 27
27 ROC curve for varying overlap - scenario 2 27
28 ROC curve for varying threshold pause - scenario 1 28
29 ROC curve for varying threshold pause - scenario 2 28
30 ROC curve for varying detection range - scenario 1 29
31 ROC curve for varying detection range - scenario 2 29
DSTO-TR-1606

32 ROC curve for varying sweep rates - scenario 1 30

33 ROC curve for varying sweep rates - scenario 2 30
Al Simulink model ,,,,..... 33
A2 Front and rear signals ..,,...,...... 34
A3 Ideal scenario 1 repeated . , , . . 34

Tables
1 Scenario parameters . 7
2 Detection parameters . 26
 DSTO-TR-1606

Glossary
AOA Angle Of Arrival
AC Alternating Current
AFC Automatic Frequency Control
CW Continuous Wave
DC Direct Current
IF Intermediate Frequency
LO Local Oscillator
LOS Line Of Sight
MLC Main Lobe Clutter
PDF Probability Distributions Function
ROC Receiver Operating Characteristic
RF Radio Frequency
SLC Side Lobe Clutter
VCO Voltage Controlled Oscillator
DSTO-TR-1606

xu
 DSTO-TR-1606

1 Introduction

1.1 Project Aim

In a generic semi-active RF seeker in medium or long range missiles, the signals at

the front and rear receivers are mixed to shift the carrier frequency to an Intermediate
Frequency (IF) where the target's doppler shift can be detected. When the missile is used
against low altitude targets in the presence of multipath in the rear receiver and sea-clutter
and multipath in the front receiver, the mixing of these two signals will significantly affect
the seeker's ability to detect a target correctly.

In publicly available generic digital simulations of semi-active RF seekers, there has

been substantial research into both multipath, [l]-[3] and sea-clutter, [4]-[5]. The aim of
this project is to combine these effects into a single model, simulate them and investigate
the mixing of the front and rear signals. A target detection scheme is then used to deter-
mine the effect on target detection.

The report starts by describing a classical model for a Continuous Wave (CW) seeker
and a simplified version used for simulation. The second section describes three effects:
multipath, sea-clutter and receiver noise, which are used to increase the realism in the
model. The final section focusses on doppler detection and outlines an optimal detection
scheme as well as a classical non-optimal detector described in Skolnik, [6]. This is exten-
sively modelled and Receiver Operating Characteristics (ROC) are compared for both an
ideal simulation and a realistic one using the three effects.

1.2 Problem Description

A typical encounter with a missile containing a semi-active RF seeker against sea-

skimming targets is shown in Figure 1. The seeker system has two receivers, one at the
front and the other at the rear. As a semi-active system, it contains no active radar and
is completely dependent on the launch platform to illuminate the target and provide rel-
evant information about the frequency of the illuminating beam and its modulation, if any.

As the target and the missile approach each other, two doppler shifts arise. The first
is from the launch platform to the missile and the second is from the launch platform
reflected off the target and incident on the seeker. This differential doppler allows the
seeker to identify and lock onto the target by carefully combining and filtering the two
incoming signals. To calculate the position of the target, a monopulse receiver at the
front of the seeker measures the angle of arrival (AOA) of the target signal in azimuth and
elevation. This information will be noisy however, due to multipath reflections, sea-clutter
and non-linearities from the hardware in the seeker.
DSTO-TR-1606

Semi-Active Seeker Target

— Ray path of signal
— Multi path reflection RearRx I^^^.sji JFront Rx
— Sea-Clutter reflection

****
Launch Platform
with Active RADAR

JLL i*-i-
Refiections off sea surface

Figure 1: Problem scenario with multipath and sea-clutter reflections

2 The Semi-Active Seeker

2.1 Classic Model

The block diagram in Figure 2, [6], is representative of the earliest systems developed
in the late 1940's and early 1950's, It is the simplest representation of a semi-active CW
seeker and consists of a rear receiver, a front receiver, a signal processor (speedgate), and a
tracking loop to control the gimballed front antenna. The missile also contains an autopi-
lot to guide it and stabilize the airframe, a fuze to detonate the warhead at the optimum
time, and a source of electrical and (in most missiles) hydraulic power.

BORESIGHT
ERROR
TO
AUTOPILOT

DOPPLER
TRACKER
(SPEEDGATE)

AFC
SPEEDGATE
LOCAL
OSCILLATOR

tANTENNA

Figure 2: Semi-active seeker block diagram

 DSTO-TR-1606

The purpose of the rear reference receiver is to provide a coherent reference for the
detection of the front or target signal. The rear signal, after conversion to IF, closes the
Automatic Frequency Control (AFC) loop around the microwave Local Oscillator (LO)
and acts as the reference for the IF coherent detector. The target signal, received at the
front antenna, is heterodyned to IF and amplified in a relatively wideband amplifier. It is
then converted to the baseband by mixing with the rear signal in the balanced mixer.

The doppler signal, now at baseband, is amplified in the video (doppler) amplifier,
which has a bandwidth equal to the total range of possible doppler frequencies. It is then
mixed with the speedgate LO, which is controlled by an AFC loop to keep the desired
signal centered in the narrow speedgate (sometimes called the velocity gate or doppler
tracker). Target detection is covered more thoroughly in section 4.

2.2 Simplified Model

The semi-active seeker model detailed in the previous section is now simplified so the
effect of different noise sources in the doppler output can be studied.

• The AFC has been removed as both front and rear signals are created from ideal
frequency sources.

• The gimbal servo and antenna control has been removed, as the front receiver is
assumed to point to the target and the rear receiver is fixed at the exit direction
with a broad view angle.

• The first mixing stage including the LO has also been removed, as simulating an
RF signal in the Gigahertz band is not feasible. Also, the effect of amplifying the
IF signals in the Front and Rear IF blocks may not have a big effect on the final
doppler measurement. A low pass filter has been included in its place and is placed
after the mixer to remove unwanted high frequency signals.

• The section after the balanced mixer is addressed separately in section 4.

The resulting model is shown in Figure 3.

The mixing of the front and rear signals, SR and SF is either a multiplication in the
time domain, or a convolution in the frequency domain.
In the time domain:
Sn(t) = SR(t)SF(t) (1)

In the frequency domain:

SD(f) = SR(f)®SF(f) (2)

The low pass filter is then used to remove unwanted high frequency components (see
section 2.3).
DSTO-TR-1606

Mixer Equivalent

Combining block

Rear Receiver Front Receiver

SR

Lowpass Filter

Spectral Measurement

Figure 3: Simplified semi-active seeker signal processing block diagram

2.3 Doppler
The doppler effect is a shift in the frequency of a wave radiated, reflected or received by
an object in motion. In the scenario for this project, a stationary illuminator is tracking a
moving target and the radio waves either compress or stretch when they come in contact
with the target. The doppler frequency equals the rate of change of the distance to the
target d, divided by the wavelength.

d
f - (3)

The minus sign accounts for the fact that, if d is negative (closing target), the doppler
frequency is positive.

The seeker makes use of the differential doppler frequency to distinguish small but fast
moving airborne targets from each other and from any background terrain and clutter.
Early doppler homers used pure CW illumination of the target, but recently the tendency
has been to employ some type of modulation such as pulse or CW frequency modulation.
A system with a modulation scheme is able to determine both range and range rate
simultaneously, but for simplicity the simulation in this project is a pure CW system. The
geometry for an ideal doppler shift of the front and rear receivers is shown in Figure 4,
[73-
Let the illuminator (I) carrier frequency be /c, the missile and target velocities, um and
ut respectively. The target (T) receives a frequency fc + (^) cos/? due to a movement of
T with respect to I. This then re-radiates towards the missile (M) with a further doppler
shift of (^) cos a due to M travelling towards T. The missile is also affected by that motion
and has a doppler shift of (^-)cosi5. Hence the front receiver on the seeker receives a
signal with frequency
um cos 8 + ut (cos a + cos ß)
fc + (4)
The mixer inside the seeker extracts the target's doppler, fa, by comparing this frequency
with that of the rear reference, which is
umcos7
fc~ (5)
 DSTO-TR-1606

Figure 4-' Semi-active homing target geometry

due to a movement of M away from I. Now, define

um cos 5 + Ui(cos a + cos /?)

IF =
A
Um COS 7
/K =
A

and when two CW signals at these frequencies are multiplied in the time domain as in
equation 1, the following occurs:

SD(t) = AFsm(2Tr(fc + fF)t)ARsm(2ir(fc + fR)t)

= "V^ lcos(MfF ~ fR)t) - cos(27r(2/c + fF + fR)t)\ (6)

where each signal has zero phase. After low-pass filtering the result, and sampling high
enough to see only the bottom part of the spectrum where the final doppler will be, the
final amplitude of the doppler signal, SD is:

AFAR
AD = (7)

with corresponding doppler frequency:

ID = IF - IR
um (cos S) + ut (cos a + cos ß) um cos 7

um(cos 5 + cos 7) + ut(cos a + cos ß)

Ä ' (8)

The maximum value of /p occurs when I, M and T are collinear and M and T are flying
towards each other, giving

2(um + ut)
J Umax (9)
DSTO-TR-1606

3 Simulated Effects
Equation 8 represents ideal line spectra. However, in real life the spectra spreads in
frequency, making computation more complicated. Effects such as main lobe sea-clutter
which will cause an extra peak, multipath reflections which spread the front and rear sig-
nals and receiver noise will alter the ideal spectrum. To explore these effects, simulation
is required.

The simulation for this section was written in MATLAB, using Simulink for the sim-
ulation and signal processing. The CW or carrier frequency is chosen as 10GHz which
corresponds to an X-band radar. However, the concerned positive and negative doppler
effects will be around 10GHz. Therefore, the CW frequency was used as a reference and
a sampling rate cover range of 4MHz was used to give a spectra range of 10GHz±2MHz.

The simulation is done entirely in the frequency domain to simplify the addition of the
external effects. Consequently, the mixing is now a convolution instead of a multiplication
in the time domain. The effect of convolving two signals is a lengthening of the vector,
a final amplitude dependent on the length of the input and triangulation. For example,
if two uniform signals Xi and X2 have lengths N and M respectively, then X\ <g> X2 has
length M + N — 1 and the spectrum is shown in Figure 5.

i.a

1.S

1.2
<3>
1

o.a

o.e

0.2

Figure 5: Convolution example

The CW signal at the illuminator is modelled as a Gaussian shape with a small fre-
quency spread OQW and a mean frequency, flow equal to 10GHz.

(f-ßcw)21 (10)
~>cw (/) = exp
y 27rfT
CTK 2°~cw

The front and rear CW signals are then created in the simulation by adding a further
doppler shift specified by the parameters in the scenario. The frequency offsets for the
front and rear signals are defined as \i\ and [1%.
 DSTO-TR-1606

To increase realism, three noise sources have been included before mixing, (1) multipath
in the front receiver, (2) main-lobe sea-clutter in the front receiver and (3) receiver noise
in both the front and rear receivers. The simulation runs for a single time period, so the
missile and the target are at a constant velocity and do not change their position. Two
test scenarios were created for the analysis and have the following parameters:

Table 1: Scenario parameters

Scenario aO 0O HI in Missile velocity, Target velocity, Doppler - Doppler -
no. um (m/s) ut (m/s) front (kHz) rear (kHz)
1 0 0 0 0 900 300 50 -30
2 60 90 60 60 900 300 20 -15

The first set of parameters was chosen as an ideal case where the target, missile and
illuminator are in a straight line. The missile velocity is set to three times the target
velocity and the target is assumed to be incoming. The second set of parameters are a
little more realistic and have the target flying straight up trying to avoid the incoming
missile.

3.1 Ideal Simulation

The ideal simulation results, without any extra effects, are shown in Figures 6 and 7.
They are based on equation 11, which is represented in the frequency domain.

SD = SR, (8> SF
(11)

For both scenarios, the results clearly show the target's doppler at 80kHz and 35kHz
respectively. A verification for this simulation is given in Appendix A.
tdeal_case1

100
Frequency (kHz)

Figure 6: Ideal scenario 1

DSTO-TR-1606

fdeal_ease2

ffi-100
as
1 80
§
60h

40

20

80 100 120
Frequency (kHz)

Figure 7: Ideal scenario 2

3.2 Multipath Reflections

Multipath effects come about by reflections on the surface of the sea and by ducting,
[2], Multipath reflections from the, sea can be either specular or diffuse. The specular
component is coherent with respect to the direct signal, and is well defined in terms of
amplitude, phase and incident direction. The diffuse component has a random nature, and
arises from scattering sources from many directions. These paths are shown in Figure 8, [3].

ILLUMINATOR
DIRECT PATH;
MISSILE
m . ..jp, u-

SPECULAR
PATH

DIFFUSE
PATH

DIFFUSE
SURFACE

Figure 8: Rear multipath components

A duct is a layer of air with different properties than its surroundings and acts as a
waveguide to trap electromagnetic energy. These refractive gradients in the atmosphere
can cause over-the-horizon fields to be tens of decibels higher than expected. More detail
 DSTO-TR-1606

about ducting can be found in [2] and [6]. For this project, only the effect of optical
interference has been included.

Due to the small grazing angle in the scenario for this project, multipath reflections are
likely to be strong, even with high sea states. According to plane earth reflection theory,
the one-way voltage reflection factor is 2sin(27r^), where hi and h2 are the radar and
target heights respectively and R is the range. Practical values and a more comprehensive
explanation are given by James in [7].

In the semi-active scenario, there are three ways for multipath forward scattering: (1)
the transmit path from illuminator to missile rear receiver, (2) the transmit path from
the illuminator to target, and (3) the receive path from the target to missile in the front
receiver (see Figure 9, [3]). In this simulation, we consider only multipath corruption of
the rear signal.
ILLUMINATOR
MISSILE RECEIVE
PATH

TRANSMIT
PATH

DIFFUSE
SURFACES

Figure 9: Front multipath components

It has been shown, [1] that the effect of multipath reflections distorts the final doppler
signal by widening the spectrum. For the simulation in this project, the rear signal, SR
will change from a line spectrum to a broad spectrum, where the spread in rear signal
accounts for both the specular and diffuse paths.

Multipath in the front receiver was studied as a special case of the multiple target
problem, where resolving two closely spaced targets is very difficult. In tracking radar,
the presence of unresolved targets will degrade the quality of the data obtained. In a
guidance system, the approach used to overcome this is to minimize the reflected (image)
signal and also prevent the 'noise' in the pitch (elevation) plane guidance channel from
causing the missile to impact the surface during low-altitude intercepts. In the case of
the sea-skimming anti-ship missile, this is done by implementing a radar altimeter for
guidance in the elevation plane, and by only using the target angle data in azimuth.

3.2.1 Simulation Results

There are many models that can be used to simulate the multipath signal. The mul-
tipath signal for this project is created by the original CW signal with no doppler shift,
DSTO-TR-1606

Sew, convolved with a multipath function, M. M is a Gaussian centered at the rear

doppler and the result of this convolution is another Gaussian centered with a frequency-
shift H2, the same as the rear signal. The resultant is modelled with amplitude AM and
a very broad user specified frequency spread aM. In reality, the spectrum would skew to
the left as the doppler shift in the diffuse path is smaller.
SMU) = Scw(f)®M(f)
i-M
■exp
C/-M2) 21 (12)
J ™M
2 %&
The rear signal can then be represented as the sum of direct and diffuse components:
SR = SR + SM (13)
The spread rear signal is then further spread by the convolution and the equation for the
doppler output signal can now be updated:
SD = (SR + SM) ® SP (14)
Figures 10 and 11 demonstrate the multipath reflections for each scenario.
muWpath_case1

100 120
Frequency (Hz)

Figure 10: Scenario 1 with multipath reflections

mulUpaih_ease2

lOO
Frequency (Hz)

Figure 11: Scenario 2 with multipath reflections

In each of the above figures, it is clear that the original signal is greatly spread in
frequency. This amount of spread was chosen to give a better comparison to real life.

10
 DSTO-TR-1606

3.3 Sea Clutter

As the sea-skimming target is being illuminated, there is a large amount of background
clutter. This sea clutter is then seen by the front receiver on the missile which is looking
down at the target (see Figure 12).

Semi-Active Seeker

Illuminator with
Active RADAR

Sea surface Target

Figure 12: Front and side-lobe clutter for a sea-skimming missile

The return from the sea is made up of two main parts, main-lobe and side-lobe return
and is typically different for different sea states and atmospheric conditions. In general,
the power of the return from a small patch of ground, at a given transmitter frequency is
given by

PgygG OAg

where

Pavg = average transmitted power

G — gain of radar antenna in the direction of the patch (G2 = two way gain)
a = factor called the incremental backscattering coefficient
Ag — resolvable area of ground (ground patch)
R = range of ground patch

The backscattering coefficient, a is the radar cross section of a small increment of ground
area, A A. This parameter is explained further in [8].

3.3.1 Main-lobe return

Main-lobe return or Main-lobe Clutter (MLC) is produced whenever the main-lobe

intercepts the sea. Because the area intercepted by the main-lobe can be extensive and
the gain of the main-lobe is high, main-lobe return is generally quite strong.

The spectral characteristics of main-lobe return are best understood by visualizing

the ground area illuminated by the main-lobe as consisting of a large number of small

11
DSTO-TR-1606

individual patches. The doppler frequency of each patch, fa is proportional to the cosine
of the angle a, between the missile velocity and the Line of Sight (LOS) to the patch.
umcosa
fd = ; (16)

where

um = velocity of missile
a = angle between the missile and LOS to the ground patch
A = wavelength

The angle a is not the same for every part of the patch. As a result, the collective return
occupies a band of frequencies.

For the case when the antenna is looking straight ahead, the doppler frequency of the
return from the patches near the center of the illuminated area very nearly equals its
maximum possible value:

JClmax — ~T~* U')

3.3.2 Side-lobe return

The clutter signal received from the antenna side-lobes, also called Side Lobe Clutter
(SLC) is always undesirable. It tends to be less concentrated (less power per unit of doppler
frequency) than the main-lobe clutter, but covers a much wider band of frequencies. Side-
lobes extend in virtually all directions, therefore regardless of the antenna look angle,
there are always side-lobes pointing ahead, behind and every angle in between. The
band of frequencies covered by the side-lobe clutter extends from a positive frequency,
corresponding to the radar velocity (/^ = yfL)1 to an equally negative frequency. The
extent to which side-lobe clutter is a problem, depends on many things:

• Frequency resolution provided by the radar

• Range resolution provided by the radar

• Gain of the side-lobes

• Altitude of the radar

• Backscattering coefficient and angle of incidence

• Man-made objects in the terrain

3.3.3 Simulation Results

In the simulation, only MLC has been included, as the effect of SLC on doppler is
negligible compared to a peak that is near or overlapping a target signal. The sea-clutter

12
 DSTO-TR-1606

signal is created in the frequency domain with a Gaussian shape, user defined amplitude
Ac, broad frequency spread a\ and frequency offset /c;.

Sc(f) =
Ac
, exp U-fcif
(18)
2*ä
This distribution is not an exact representation of real life, as there are many parameters
involved which shape the MLC. It was chosen as a simple model to use in this simulation
and could be improved in future work. In the simulation, the sea-clutter signal is added
to the front signal, which now becomes:

S'P = SP + SC (19)

where SF is the original front signal. The equation for the doppler output signal can now
be updated:
SD = SR <g> (SF + Sc) (20)
where the effect of the convolution is to further spread the sea-clutter signal. For the
scenarios in this project, the desired frequency offset, fd is set to the corresponding doppler
between the missile and the ground offset by 20° to represent the peak sea-clutter signal
not being directly at the line of sight.

umcos(20°)
fd = Y (21)

For both scenarios, the missile's velocity um is 300m/s, and equation 21 gives a fre-
quency of 28.21kHz. The mixing then causes the center frequency to shift to the right by
subtracting the rear doppler component. Hence, the expected frequency for scenario 1 is
28.21kHz - (-30kHz) = 58.21kHz and 28.21kHz - (-15kHz) = 43.21kHz for scenario 2.
This calculation is detailed in Appendix B.

Figures 13 and 14 show the results from the Simulink model. The amplitude of the
sea-clutter signal is set to be slightly lower than the target signal, and the spread is large
to give a better comparison to real life. In each case, it is clear that the added signal is at
the correct frequency and the characteristic of the sea-clutter signals are correct.

13
DSTO-TR-1606

sea_clutter_case1
160

140

M 80-•

80 100 120
Frequency {kHz)

Figure IS; Scenario 1 with sea-clutter

sea_jciutter_case2

St
J 80
1
* 60

40

20

1QO 12G
Frequency (kHz)

Figure 14: Scenario 2 with sea-clutter

3.4 Receiver Noise

3.4.1 Noise Fundamentals

Receivers generate thermal noise which masks weak echoes being received from the
radar transmissions. This noise is one of the fundamental limitations on the radar range
and is shown in the range equation for a bi-static radar, where the transmitter and receiver
are separated (see Appendix D).

Most of this noise originates in the input stages of the receiver. The reason is not that
these stages are inherently more noisy than others but when amplified by the receiver's full
gain, noise generated there swamps out the noise generated further along. Since the noise
and the received signals are thus amplified equally, in computing signal-to-noise ratios, the
factor of receiver gain can be eliminated by determining the signal strength at the input
to the receiver and dividing the noise output of the receiver by the receiver gain. Thus,
the receiver noise is commonly defined as noise per unit of receiver gain.
Noise at output of receiver
Receiver noise (22)
Receiver Gain

14
 DSTO-TR-1606

More commonly, the noise performance Figure of merit is known as the 'noise figure',
Fn. This is the ratio of the noise output of the actual receiver to the noise output of a
hypothetical, 'ideal' minimum-noise receiver providing equal gain.
_ Noise output of actual receiver
Noise output of ideal receiver
Thermal noise is spread more or less uniformly across the entire spectrum, [8]. So, the
amount of noise appearing in the output of the ideal receiver is proportional to receiver
bandwidth. The mean power per unit of receiver gain, of the noise in the output of the
the hypothetical ideal receiver is thus:
Mean noise power (ideal receiver) = UTQB (W) (24)
where B is the receiver bandwidth and by convention, T0 is taken to be 290°K, which is
close to room temperature and makes kT0 a round number (4 x 10~21 Ws).

When the internally generated noise is considerably greater than the external noise,
the noise figure, Fn, multiplied by the previous expression for mean noise power per unit
of gain for an ideal receiver is commonly used to represent the level of background noise
against which target echoes must be detected.
Mean noise power (actual receiver) = FnkToB (W) (25)

Typically though, the noise temperature of the radar receiver is not the most important
characteristic, since choosing a low noise amplifier will sacrifice other important perfor-
mance characteristics such as dynamic range, instantaneous bandwidth, phase amplitude
stability and cooling in the receiver.

3.4.2 Noise Statistics

It is important that the noise modelled in this section accurately produces the correct
statistics, otherwise the target detection schemes in section 4 will not work properly. To
this end, the following hypothesis for the output signal is defined and a derivation of the
Probability Distribution Functions (PDF) for each case follows:
HQ : noise only is present —> n(t)
H\ : signal and noise are present —> s(t) + n(t)
The additive zero mean white noise n(t), is spatially incoherent and uncorrelated with vari-
ance a2. It is added to both magnitude and phase and hence has a complex representation.

Consequently, the front and rear signals can be written as a combination of real (in-
phase) and imaginary (quadrature) parts, separated by a phase of 90°.
u(t) = i//(t)cos(27r/jnt) + i/Q(t)sin(27r/int)
= r{t) cos{2irfint-(f>(t))
or simply
!//(*) = r(i)cos(0(t))
uQ(t) = r(t)sm(<f>(t))

15
DSTO-TR-1606

where fin = fc + IR and fc + fp for the rear and front signals respectively, r(t) is the
amplitude of u{t) and the phase, (f){t) = arctan(-^).

If we run a test scenario where there is no target, then the in-phase and quadrature
components are:

Vl{t) = n/(t)
VQ&) = nQ(t)
and likewise, if we know there is a target present with an amplitude A:

uj{t) = A + ni(t) =» nr(t) = r(t)cos(<p(t)) - A

^QW = nQ(t) = r(t) sinm)),
where the noise in-phase and quadrature components rij{t) and nq{t) are uncorrelated
zero mean Gaussian with equal variances a2. The joint PDF of the two random variables
nr,nq is
n
1 ( ! + ng\
f(nr,nQ) = -—^exp
2
2ira la2 J
1 (r cos 4> — A)2 + (r sin 4>)2
■exp (26)
27TCT 2 2^2

Now, the PDF's of the random variables r(t) and <f>(t), represent the modulus and phase
of i/(t). The joint PDF of these two random variables r(t); <j)(t) is given by

f(i"](f>) = /(«/; nQ)\J\ (27)

where the Jacobian is:

dti[ dnj
8r 8<f> cos (f> —r sin <f>
J 9riQ dng
sin (j) r cos (p
. dr &4>
with determinant \J\ = r. Substituting this result and equation 26 into equation 27 gives:
,. ,, r / (r.cosS — A)2 + (rsmd))2
f(r'><t>) = ^ZZ2
2TT(T
2
ex
P(- 2a2
r r2 + A' r^Lcos<
exp - l exp (28)
2a
Then to obtain the PDF for r alone, we need to integrate over <p.
r2TT

f(r) = /"/(r;#)#
Jo
r ( r2 + A2\ 1 /2- / Mcos^)\

where the integral inside equation 29 is known as the modified Bessel function of order
zero:
1 f2ir
eßcm6d9 (30)
^ Jo

16
 DSTO-TR-1606

Thus,
2
rA + A2
f(r) = -r/o exp (31)
a' 2<r2

which is the Rice PDF. If the scenario contains noise only, then -\ = 0 and equation 31
becomes a Rayleigh PDF:

/(r) = (32)
^eXP("^)
and if there is a target with the noise, then 4j- is very large and equation 31 becomes a
Gaussian PDF, with mean A and variance a2:

1 ( (v-Af
m= y/2-Ka2
exp
2<72
(33)

The PDF for the random variable (f> is found likewise, but not described here. If we look at
the noise statistics of the simulation over 200 runs, the amplitude distributions of target
with noise and noise alone match the theoretical distributions. One is a Gaussian, while
the other is a Rayleigh distribution.

Comparison of Probability Distributions

0.07
Signal + Noise
Noise Only

0.06

0.05

£0.04
D

I 0.03 b

0.02

0.01

50 100 150 200 250

Amplitude

Figure 15: Simulation probability distributions

17
DSTO-TR-1606

3.4.3 Simulation Results

In the simulation, the receiver noise N, used is additive zero-mean coloured noise with
an amplitude determined by the user defined noise power level. This signal is added to
both the front and rear receiver blocks in the Simulink simulation. The equation for the
doppler output signal can now be updated:

SD = (SR + N) ® (SF + N) (34)

Scenario's 1 and 2 are shown in figures 16 and 17. They both clearly show the effect
of the randomness over the entire spectrum.
recefver_nofse„case1

m 100 ■
$r$fjM ■>Ytyf#tfls#^ Wftv^jff^U^,^^
.§ 80 -
E

i i i i i i
BO 100 120
Frequency (kHz)

Figure 16: Scenario 1 with receiver noise

receiver_nolse_case2
—I 1— i i i 1 1 1
160

140

120

M so -
£
ffV^-^W L^^^-^
i i i i I i i i i
60 100 120
. Frequency <kHz)

Figure 17: Scenario 2 with receiver noise

 DSTO-TR-1606

3.5 Complete Simulation

The last stage of the simulation is to combine the three effects together. The final
equation for the doppler output is:

SD = (SR + SM + N) ® (SP + Sc + N) (35)

where the different amplitudes are determined by the user. Simulation results are shown in
figures 18 and 19. The spectra's clearly show the sea-clutter peak at the correct frequency,
the spread target signal and random noise over the entire spectrum. When the combination
of the sea-clutter and multipath merge together, only the tip of the target is clear.
mixed„case1

80 100 120
Frequency (kHz)

Figure 18: Scenario 1 with all three effects combined

mtxed_case2

80 100 120
Frequency (kHz)

Figure 19: Scenario 2 with all three effects combined

19
DSTO-TR-1606

4 Doppler Detection
Target detection or 'lock on' is a post-launch function since many missiles are not in a
position to view the target before launch. Those that are able to lock onto a target may-
break lock due to a launch shock, plume effects, or extremely high feedthrough, which is
an unwanted DC component in the spectrum. Thus, the seeker must accomplish target
detection at some prescribed time during flight.

Commitment to fire on a target implies that a missile begins its task with the knowledge
that a target exists. For a successful firing, the probability of finding a target must be
95 percent or higher. To achieve this, the initial detection threshold is set relatively low,
ensuring a high probability of seeing the target. The probability of false alarm is then
high, and a suitable detection scheme must be employed to compensate. After a brief
summary of detection theory, two different doppler detection schemes are outlined. The
first is an optimal scheme for calculating the doppler frequency and amplitude, while the
second is a system employed in many semi-active systems. A simulation of the second
scheme has been done and the results are summarised in the final section. In a semi-active
system, the target's AOA is calculated using monopulse principles, but that is beyond the
scope of this report.

4.1 Detection Theory

A target is detected when its amplitude exceeds a threshold value r. The aim of a
good detector is to maximise the probability of detection for a given probability of false
alarm. Using the hypothesis and pdf's derived from 3.4.2, we can define the detection and
false alarm probabilities for a point X on the spectrum, as:

Pi = P(X>rm - jT-^«p(-^)* (36)

Pfa = P(X > T\HO) = jf ^exp[-~)dr

- ■*(-£)■ (37)

where the variance, a2 is the mean noise power and the threshold, T can be now calculated
based on a given false alarm probability,

T=y/-2o-2ln(P/o). (38)

The case where the noise causes the target's amplitude to be smaller than the threshold
is called a miss.

IW(*<™-£^«»(-^)* (39)
The following figure shows the regions where each of these definitions are defined.

20
 DSTO-TR-1606

SIGNAL PLUS NOISE

CO
©

MAGNITUDE (VOLTS/<r)

Figure 20: Detection regions

4.2 Optimal Detection

To find an optimal doppler detector, one approach is to maximise a likelihood function
containing an accurate model of the signal. This approach has been looked at, but not
completed. The derivation is included in Appendix C.

4.3 Non-Optimal Detection

Many different techniques for doppler extraction and tracking have evolved in the
last 30 years. The most common systems use combinations of the following three stages:
narrow-banding, doppler frequency extraction and doppler tracking.

Narrow-banding

• 'Late narrow-banding' segregates the wanted target signal in doppler frequency from
all other signals by a narrow bandpass filter at a late stage in the receiver chain
incorporated within the main IF amplifier.

• 'Early narrow-banding' does the segregation as early as possible in the receiver chain,
typically before the IF amplifier.

Doppler frequency

• 'Explicit' doppler frequencies are translated to baseband and each doppler frequency
is explicitly represented by an AC frequency.

• 'Implicit' doppler frequency remains always as a shift on the carrier frequency and
its presence is implied by the change of frequency.

21
DSTO-TR-1606

Doppler tracking

• 'Frequency locked loop5 - The doppler frequency of the wanted target signal is
changed to a second intermediate frequency which is compared with that set by
a fixed frequency discriminator. Any maladjustment in frequency is sensed and used
to retune the associated LO.

• 'Phase locked loop' - A phase discriminator is used which senses phase difference.
This difference is used to retune the LO via an integrator.

Target acquisition is accomplished by sweeping the frequency of the speedgate LO over

the designated portion of the doppler bandwidth. The speedgate LO is typically some form
of voltage-controlled oscillator (VCO). The linearity of its frequency-volt age characteristic
must be carefully controlled to allow precise sweep positioning and a constant, sweep rate
over the full doppler frequency range of interest. Also, since the VCO forms part of the
doppler tracking circuit, control linearity is required to maintain a constant loop gain over
its total frequency coverage range. The sweep rate is based on the doppler filter bandwidth
and must be slow enough to ensure signal buildup in the filter. The extent of the sweep
will differ from system to system, depending on the accuracy of the designation and on
operating frequency. Together, the rate and extent of the sweep will determine the num-
ber of looks at the target during the available acquisition time and hence the cumulative
detection probability.

The actual detection process consists of programming the speedgate LO with a saw-
tooth of triangular sweep voltage. When the difference between the LO and the target
doppler frequencies equals the speedgate filter center frequency, an output is produced at
the speedgate discriminator or at a separate amplitude detector. The output signal is then
detected and if it exceeds the detection threshold, the search is stopped for a few tens of
milliseconds while the signal is examined to verify that it is a coherent target and not a
false alarm due to noise. This process of verification examines the signal in the gate for
persistence, since a target will remain above the verification threshold while noise will not.
Typically, the false alarm rate and the verification time required for each false alarm can
be optimised for a given system.

Once the verification threshold has been passed, the speedgate tracking loop is closed
and the speedgate is said to be 'locked on'. A valid target is then tracked in frequency using
the discriminator, and guidance commands can be extracted from it. A block diagram
representing the system is shown in Figure 21, [6].

22
 DSTO-TR-1606

(ALTERNATE
I METHODS)
COHERENCY
CHECK
(VERIFICATION)
TRACK
AFC

Figure 21: Alternative methods of target detection and verification

4.4 Detection Simulation

A simulation based on Figure 21 was created in MATLAB to test the effect of noise,
multipath and sea-clutter on doppler detection. The tracking discriminator is not present
in this simulation, since we are only interested in the doppler detection. Consequently,
the system is simplified and can be represented by the block diagram in Figure 22.

Doppler Narrowband Check

Spectrum Filter Hypothesis
t i

MJL^T, it JL>x
New Filter
Parameters
i L
'
Pause for
confirmation

If
afl er pause
i r
Detection

Figure 22: Simplified block diagram for detection and verification

Since the worst case centre frequency of the MLC is at the velocity of the missile, the
'doppler spectrum' or 'passband' can be determined as a portion of the total spectrum
coming from the mixer, with the start point set by moving a fixed distance away from the
MLC and the end point set at a harmonic of the MLC frequency. In this simulation, the
third harmonic was chosen as the end point.

23
DSTO-TR-1606

There are a number of important parameters in this simulation. These include the
sweep rate or number of steps through the filter bank over a given time, the bandwidth
of the filter, whether to overlap the spectrum from one filter block to the next, the detec-
tion range, pause length and the detection threshold. The consequence of a non-optimal
detection scheme, is that the relationship between the false alarm probability and thresh-
old calculated in section 4.1 no longer holds. The theoretical calculation was based on a
system using the entire spectrum and not filtered sections of it. The best analysis that
can be done is a comparison of the parameters and how an ideal simulation compares to
a realistic one.

4.4.1 Doppler Filter Bank and the Overlapping Spectrum

In Figure 21, a LO is used to mix the doppler spectrum down to baseband, so it can
be used in a single filter. For the purposes of modelling the detector, this is simplified
by creating a new bandpass finite impulse response filter at each iteration through the
detection loop. The overall effect is the same as if a box is drawn around the LO, mixer
and filter.

In this simulation, the bandwidth is altered depending on the sweep rate and there is
an option for using a 50% overlap between one filter block and the next. This increases
the sweep rate and minimises the loss of output when a signal lies between the center
frequencies of two filter blocks. For example, if there are N filter blocks in a given time
period, then there are 2N — 1 when using a 50% overlap.

FILTER ii FILTER In +'

fILTefin FILTER

t-

3dB a. 3dE
t-
3
0

Figure S3: Comparison of non-overlapping and overlapping filters

4.4.2 Detection Range

The detection range is required to test Hi, the case where there is a signal and noise
present. It defines the bounds where a correct detection takes place relative to the center
frequency. If a detection takes place outside the detection range, it is deemed a miss.

24
 DSTO-TR-1606

Filtered Input Signal

Figure 24-' Detection range

4.4.3 Pause Length

The pause length determines how long the system waits before deciding if a signal
contains a target and not noise. If the pause length is longer, there will be less noise
that can be mistaken for a target, but the miss rate will be higher since there is a longer
time interval where a target can not be detected. The following figure has a target with
noise present for the first 15 time periods, then noise only for 15 time periods. There
are four pairs of images, each pair having the thresholded only scenario on the left and
the thresholded and delayed scenario on the right. If a filter block is white, then it has
exceeded the threshold and is a detection.
The target is clear in each of the scenarios as the constant line in the bottom left. As
the pause length increases, more noise is removed until only the target is seen in the final
scenario on the bottom right image.

4.5 Receiver Operating Characteristics

To do a complete comparison of the different parameters, a common technique is to
vary the threshold, and measure the false alarm and detection rates. This is called a
Receiver Operating Characteristic (ROC). Recall the two hypothesis:

HQ : noise only is present

Hi : signal and noise are present

The axes of the ROC curve relate Ho to the false alarm rate and H\ to the detection rate
using the definitions from section 4.1.
The following table describes the default variables used in the simulation for all pa-
rameters except the one being varied.

25
DSTO-TR-1606

Comparison of Delay Lengths (Time vs. Filter Block)

Pause=1

Pause=2

Pause=3

Pause=4

10 15 20 25 5 10 15 20 25

Thresholded Only Thresholded and Paused

Figure 25: Comparison of pause lengths on detection

Table 2: Detection parameters

| 50% Overlap Threshold Pause (time periods) Detection Range Sweep Rate |
| No 2 3kHz 50 |

The following ROC curves were then obtained by running both scenarios 75 times
for H§ and 75 times for Hi. The sea-clutter and multipath effects were included in the
simulation, while the receiver noise was re-simulated each time to generate the randomness
required.
The effect of overlapping filter banks caused an increase in the false alarm rate for
both scenarios. This is a consequence of the increased sweep rate required to compensate
for the overlapping filter bank. The second scenario has approximately twice the false
alarm rate than the first due to the closer proximity of the sea-clutter peak. Further work
could be done to verify this by changing the position of the sea-clutter and observing the
simulation results.

26
 DSTO-TR-1606

Ideal Simulation ROC Multipath Simulation ROC

Figure 26: ROC curve for varying overlap - scenario 1

Ideal Simulation ROC Multipath Simulation ROC

Figure 27: ROC curve for varying overlap - scenario 2

The curves look very similar in both scenarios, though the second one has a slightly
higher false alarm rate. There does not appear to be many differences between the ideal
and the multipath cases indicating that this variable is independent of the scenario. Clearly
though, as the pause length increases, the false alarm rate decreases in each case.

27
DSTO-TR-1606

Ideal Simulation ROO (ylultlpath Simulation ROC

Pause Length = 1
Pause Length = 2
Pause Length = 3 Pause Length =* 1
Pause Length = 2
Pause Length = 3

Figure 28: ROC curve for varying threshold pause - scenario 1

ideal Simulation ROC Multipath Simulation ROC
1
Pause Length = 1 Pause Length = 1
Pause Length = 2 Pause Length = 2
Pause Length «= 3 0.9 Pause Length = 3 -
/
0.7 - o.e /
O.T' - / -
0.6 -
M o-s -1 / / _
0.6 -1 / J -
0.5
- ( / -
|0.4 - -
C3 0.4
- -
0.3 - - 0.3 - -
0-2 - - -
- / 0.2

-
O.l

"I - O.I -

Figure 29: ROC curve for varying threshold pause - scenario 2

There are many interesting conclusions that can be drawn from these results. Firstly,
when there is a small detection range, the multipath case performs better due to its larger
frequency spread. However, in the second scenario, the 1kHz detection range proved to
be too narrow to pass the detection threshold. The trend is that larger detection ranges
give lower false alarm rates until a limit is reached and then they have an adverse affect
in the multipath case of Figure 30.

28
 DSTO-TR-1606

Idea 1 Simulation ROC Multipath Simulation ROC

1
^-^—^ ^ '/
0.9

O.S
/^Jr~^ 0.9

O.S
V /}
0.7"

0.6
/ 1 0.7

o.e - / f

J
O.S
Detection range = 1kHz
Detection range = 2kHz
0.4 Detection range = 3kHz 0.4
Detection range = 4kHz

0.3 0.3

0.2 0.2
Detection range = 1 kHz
Detection range = 2kHz
O.I O.I / Detection range = 3kHz
Detection range = 4kHz
,

Figure 30: ROC curve for varying detection range - scenario 1

Ideal Simulation ROC Multipath Simulation ROC

O.S -
O.S

0.7
///'-■

O.S

J
Detection rang© = 1 kHz
Detection range = 2kHz
- Detection
Detection
rang©
range
=
=
3kHz
4kHz
.

Figure 31: ROC curve for varying detection range - scenario 2

All but the multipath case in scenario 1 were able to give distinctive results, as the
curves overlap. The trend is as the sweep time increases, the false alarm rate gets higher.
This is due to the bandpass filter having a smaller bandwidth as the sweep time increases.
There is also a slight increase in false alarm rate for scenario 2. Prom these results there
are a few conclusions that can be drawn:

• The multipath case did not give significantly worse results, as the broad sea-clutter
overlapped the multipath component.

• Using a 50% overlap in the filter bank gives a higher false alarm rate.

• A longer threshold pause gives a lower false alarm rate and is independent of the
scenario.

• The trend for the detection range shows that larger detection ranges give lower false
alarm rates until an upper limit is reached.

• A greater sweep time (filter bank) gives a higher false alarm rate.

29
DSTO-TR-1606

Ideal Simulation ROC IVlMltlpath Simulation ROC

/^-^

or
g / i -as o.7f-
s
n
<3 oe
s

1 Filter Length *= 30
Filter Length — SO
Filter Length = TO
Filter Length = 30
Filter Length = SO
Filter Length = 7Q

Figure 32: ROC curve for varying sweep rates - scenario 1

ideal Simulation f^OO [Vluitipath Simulation ROC

Filter Length = 30 Filter Length = 30

Filter Length = 50
Filter Length = 7Q

Figure 33: ROC curve for varying sweep rates - scenario 2

5 Conclusions and Future Work

The Simullnk simulation worked well and demonstrated the effects of multipath, sea-
clutter and receiver noise on an ideal simulation. It is clear that mixing the multipath and
sea-clutter caused the spectrum to spread widely and this in turn gave some interesting
detection results. There are a number of improvements that could build on this section:

• Multipath with a reduced amplitude could be added to the front signal.

• A small randomness in the frequency offset for the multipath signal could be used to
more accurately model real life. This would add extra frequency components during
the convolution.

• A comparison on different amounts of frequency spread could be included.

• The models for multipath and sea-clutter could be improved to more accurately
model different sea states and other atmospheric conditions.

30
 DSTO-TR-1606

The optimal detection scheme needs some more work to provide an alternative means
to determine the doppler frequency and amplitude of the filtered spectrum. The non-
optimal detection scheme gave good results and led to a number of observations. These
are repeated for completeness:

• The multipath case did not give significantly worse results, as the broad sea-clutter
overlapped the multipath component.

• Using a 50% overlap in the filter bank gives a higher false alarm rate.

• A longer threshold pause gives a lower false alarm rate and is independent of the
scenario.

• The trend for the detection range shows that larger detection ranges give lower
false alarm rates until an upper limit is reached. More extensive simulations of this
variable could give more conclusive results and be used to optimise the detection
range for this system.

• A greater sweep time gives a higher false alarm rate, due to the decreasing bandwidth
of the bandpass filter. Again, further simulations could be used to find an optimal
sweep time for this system.

The detection simulation could be improved by increasing the length of time that each
hypothesis was tested over and increasing the number of points on the ROC curves. Fur-
ther simulations could also be used to optimise the detection range and the sweep time
for this system. There are also a few alternative detection schemes that could be used to
compare with this one, such as the 'n out of m detector', [6], detectors based on a 'constant
false alarm rate' and schemes which incorporate both detection and tracking.

Finally, putting the seeker in the real world would involve the target moving in different
directions and at different velocities. The seeker would need to compensate by changing
its trajectory and velocity. Both of these changes would vary the doppler measured at the
front receiver. In the future, this model could be introduced into a larger system where
there are moving targets and existing sea-clutter and multipath models. Testing the effect
under these dynamic conditions is a much larger project, but could easily extend this work.
The ultimate outcome would then be to verify 'Hardware In the Loop' simulations with
the measured output doppler and AOA.

31
DSTO-TR-1606

References
1. W.V.Andrew, The effects of Multi-path Propagation on Low-Altitude Detect Pulsed-
Doppler RADAR Systems, December 1990, Arizona State University.

2. W.Lippincott, Sea Surface Multipath Effects on Ship Radar Radiated Power Determina-
tion, Advanced Systems Technology Branch, Space Systems Development Department.

3. R.M.Smith, J.Y.Yee, C.S. An and A.L.Haun Simulation of Multipath for Semi-active

Missiles, AGARD CP-473, Turkey 1990.

4. D.Lewis Principles of Naval Architecture, 1983, Society of Naval Architects and Marine
Engineers, August 1997,

5. C. Beard and I.Katz The Dependence of Microwave Radio Spectra on Ocean Roughness
and Wave Spectra, Transactions on Antennas and Propagation, 1957

6. M.Skolnik, 'RADAR Handbook - Second Edition, 1990, McGraw-Hill Inc.

7. D.A.James, 'Radar Homing Guidance for Tactical Missiles, 1986, Royal Military Col-
lege of Science Shrivenham.

8. G.W.Stimson, Introduction to Airborne RADAR, 1983, Hughes Aircraft Co., El Se-

gundo, CA.

9. R.Smith, Multipath Math Model and Implementation Guide - Revision No. 1, July 29,
1986, Missile Software Branch, China Lake, California USA.

10. B.Moran Detection, Estimation and Classification, MSIP Lecture notes, March 2000.

11. B.Mahafza Radar Systems Analysis and Design Using MATLAB, Chapman and
Hall/CRC, 2000

32
 DSTO-TR-1606

Appendix A Ideal Simulation Verification

The Simulink model used for the simulation is shown in Figure Al. The low pass filter
is implemented in Simulink by first creating it in MATLAB, then passing its frequency
response to be multiplied in the frequency domain.

H pioLmodel
•Eäe. Edft ijern Simulation Format look Help

DüDii % rtü e ft (H •• 'S-- > ■ I'lviri.-bl 3

Rear Receiver Front Receiver

Rear Signal Signal In Rear Signal Front Signal Signal In Front Signal
V V

front Front out

^ mixed_out

FIR Frequency Response Mixed Signal

FIR LöwpäsrFiiier
i±

Fina| Output output

Low pass filter in
freq-domaih

Ready 100% 'ode45

Figure Al: Simulink model

Using this model, scenario 1 was run with a rear doppler of -30kHz and a front doppler
of 50kHz. With the carrier reference set to 1MHz rather than 10GHz for the convenience
of demonstration, the frequency of the front signal should be lMHz+50kHz=1.05MHz and
the frequency for the rear signal should be lMHz-30kHz=0.97MHz. The front and rear
signals are shown in Figure A2:

33
DSTO-TR-1606

Ftesr Signal Front Slgnel

SO

**o

20

00
-8
so

eo

**o

20

SOO 6SO 900 950 1 OOO 1 OSO 1 1 OO 1

Frequency (kHz) Frequency <RI-tz>

Figure A2: Front and rear signals

It is clear that the signals are at the correct frequencies, and hence the convolution
places the final doppler at the difference of the front and rear frequencies, which is 80kHz,
Figure 6 has been reproduced to show this.

100
Frequency (kHz)

Figure AS: Ideal scenario 1 repeated

34
 DSTO-TR-1606

Appendix B Sea-clutter Frequency Calculation

To calculate the sea-clutter center frequency shift, a Gaussian shaped spectrum is ap-
proximated by the Fourier Transform of a cosine function, where the mean is approximated
by the frequency of the cosine. The exact frequency was then determined by the following
signal definitions in the time domain:

SF(t) = sin(27r(/c + fF)t)

SR(t) = sin(27r(/c + fR)t)
Sc(t) = sin(27r(/c +/d)t)

where the new front signal is now represented as in equation 19 and the mixing is a
multiplication, since the signals are represented in the time domain. The final spectrum
becomes:

SD(t) = SR(t)[SF(t) + Sc(t)]

= - cos(2n(fF - fR)t) + 1 cos(27r(/d - fR)t) +

^ COS(2TT(2/C + fR + fF)t) + l- COS(2TT(2/C + fR + fd)t) (Bl)

where the desired sea-clutter frequency is now shifted by the rear doppler frequency.

35
DSTO-TR-1606

Appendix C Optimal Detection

iProm the complete simulation in section 3.5, we can define a signal s(f), in the
frequency domain with all three effects added, sea-clutter, multipath and noise. Math-
ematically, s(f) is multiplied by a Gaussian with unknown amplitude ay, mean fr and
known variance uT and contains additive zero mean white noise, n(f) with variance a2:

8(f) = aTexp ( -{fnJ_f2 ) + n(f) (Cl)

2-KOj,

The probability density of s(f) is given by

"aT;fT\S) — r
1 — exp ( —-—u
1 v-> s(/)-arexp (-
(f-hf (C2)
TroZ)N v \ 2TRT2 y
(v27TCr' 2o\

and the likelihood function is obtained by taking the log of the probability density:

Ls(aT-fT) = --—2^ s(/)-arexp -

(f~hfs 2
(C3)
2ira2 2<4
f
If we differentiate this with respect to the two unknowns, ax and fr, we get:
dLs
s(f) ~ &T exp
(f-hf exp -
(f-hf (C4)
OUT TftJ u 2e4
/
dLs _ 2 Y^ (f-fT ,2 \ 1 ay exp (f-hf\(f-h
4f) - aT exp - (C5)
1a\ 2<J\ 2crf,

If these two equations are then equated to zero, they become

2
(/-/T)
E s(f) ~ aT exp
2<rf,
eXpl ,=0
2äf- (C6)

X1
(f-hf
E s(f) ~ aT exp
2c4
expl-^^H/-/r) = 0 (C7)

Equation C6 can then be rearranged

^s(/)exp (f-hf aT Y^ exp ( - (/-/T)

;
(C8)
2o\ 2of
/ /
to show that the RHS represents a Gaussian probability density function, with mean fa
and variance er|>.

ex (f-hf (C9)
E-7==T
J2-KO-J,
f 2of

Hence, equation C8 becomes

7TCT,? (CIO)

36
 DSTO-TR-1606

and can be rearranged to give an expression for a?-

aT = J2 s(f) exp ( -
U-hf (Cll)
27r V 2o\
V 4 /
Now, if equation C7 is rearranged in a similar way,

(/ JT)2)(/-/T) (C12)
2

and we use the fact that E(f — fr) = 0. Then,

2
(/-/T)
^2 exp (/-/T) = 0 (C13)
2CJ2
f
and equation C12 becomes

£*(/)(/-/,)exp(J^)=0. (C14)

The next step is to define

;
(/~/T)
g{f ~ IT) = exp
2erf,
and notice that equation C14 is the same as

E s(/)s'(/ - /r) ~ / *(/)</(/ - /r)d/ = 0 (C15)

Using integration by parts, this becomes

*(/)<?(/ -h) - ( s'(f)g(f -fT)df = 0 (C16)

/ Jf
where the first part tends to 0. Thus

fs'(f)g(f-fT)(V = 0 (C17)

Now, using Parsevals Identity:

J nnf))HGU)}df = 1-J f(t)9(t)dt (C18)

we can take the Fourier Transform of this expression, where

T{s'(f)} = ts(t) (C19)
HaU-fr)} = e-^f^igif)}
a2t2
= e-2^*Cexp (C20)

Thus, equation C17 becomes

a2t2
/ is (i) exp e-^fTtdt = 0 (C21)

At this stage, a numerical solution in MATLAB is required to solve this equation.

Work on this method is still in progress.

37
DSTO-TR-1606

Appendix D Range Equation

Here is the range equation for a bi-statlc radar, as used in the semi-active case in this
report.

-a -n2\ Pio-GjAn
(RlR2)max - (AX2TQ (D1)

where,

Ri — range illuminator-target (m)

i?2 = range target-missile (m)
Pi — illuminator power (W)
a = radar cross section (m )
Gj = illuminator gain (dB)
Am = effective aperture of the receiving antenna (TO )
L = loss factor for miscellaneous Losses (dB)

and, Smin = Fn(kT0B)(S : N)min, where

(S : N)min = minimum detectable signal level at the receiver output (dB)

Fn = receiver noise figure
B = receiver bandwidth (Hz)
k = Boltzmann's constant 1.38 x 10~23(Ws/°K)
TQ = absolute temperature of the resistor representing the external noise ( K)

38
 DISTRIBUTION LIST

Signal Processing in a Semi-active Seeker

Luke Rosenberg, Yiding Hu and Bill Moran

Number of Copies
DEFENCE ORGANISATION

Task Sponsor
Director General Air Development
S&T Program
Chief Defence Scientist
FAS Science Policy
AS Science Corporate Management
Director General Science Policy Development
Counsellor, Defence Science, London Doc Data Sheet
Counsellor, Defence Science, Washington Doc Data Sheet
Scientific Adviser to MRDC, Thailand Doc Data Sheet
Scientific Adviser Joint 1
Navy Scientific Adviser Doc Data Sheet
Scientific Adviser, Army Doc Data Sheet
Air Force Scientific Adviser Doc Data Sheet
Scientific Adviser to the DMO M&A Doc Data Sheet
Scientific Adviser to the DMO ELL Doc Data Sheet
Information Sciences Laboratory
Bill Moran
Systems Sciences Laboratory
Chief, WSD Doc Data Sheet
RLAWS WSD 1
RLMWS WSD 1
HMS WSD 1
HRFS WSD 1
HWSA WSD 1
Luke Rosenberg 1
Yiding Hu 1
Ninh Duong 1
Peter Simms 1
Richardo Nazar 1
DSTO Library and Archives
Library, Fishermans Bend Doc Data Sheet
 Library, Edinburgh 1
Library, Sydney Doc Data Sheet
Library, Stirling Doc Data Sheet
Library, Canberra Doc Data Sheet
Australian Archives 1
Capability Systems Division
Director General Maritime Development Doc Data Sheet
Director General Land Development 1
Director General Information Capability Development Doc Data Sheet
Office of the Chief Information Officer
Deputy Chief Information Officer Doc Data Sheet
Director General Information Policy and Plans Doc Data Sheet
AS Information Strategies and Futures Doc Data Sheet
AS Information Architecture and Management Doc Data Sheet
Director General Australian Defence Simulation Office Doc Data Sheet
Strategy Group
Director General Military Strategy Doc Data Sheet
Director General Preparedness Doc Data Sheet
Navy
SO (SCIENCE), COMAUSNAVSURFGRP, NSW Doc Data Sheet
Director General Navy Capability, Performance and Plans, Doc Data Sheet
Navy Headquarters
Director General Navy Strategic Policy and Futures, Navy Doc Data Sheet
Headquarters
Army
ABCA National Standardisation Officer, Land Warfare Devel- Doc Data Sheet
opment Sector, Puckapunyal
SO (Science), Deployable Joint Force Headquarters (DJFHQ)(L), Doc Data Sheet
Enoggera QLD
SO (Science), Land Headquarters (LHQ), Victoria Barracks, Doc Data Sheet
NSW
Air Force
Intelligence Program
DGSTA, Defence Intelligence Organisation 1
Manager, Information Centre, Defence Intelligence Organisa- 1
tion
Assistant Secretary Corporate, Defence Imagery and Geospa- Doc Data Sheet
tial Organisation
 Defence Materiel Organisation
Deputy CEO Doc Data Sheet
Head Aerospace Systems Division Doc Data Sheet
Head Maritime Systems Division Doc Data Sheet
Chief Joint Logistics Command Doc Data Sheet
Head Materiel Finance Doc Data Sheet
Defence Libraries
Library Manager, DLS-Canberra Doc Data Sheet
Library Manager, DLS-Sydney West Doc Data Sheet
UNIVERSITIES AND COLLEGES

Australian Defence Force Academy Library 1

Head of Aerospace and Mechanical Engineering, ADFA 1
Deakin University Library, Serials Section (M List), Geelong, 1
Vic
Hargrave Library, Mdnash University Doc Data Sheet
Librarian, Flinders University 1

OTHER ORGANISATIONS

National Library of Australia 1

NASA (Canberra) 1
Government Publications Librarian, State Library of New South 1
Wales
INTERNATIONAL DEFENCE INFORMATION CENTRES

US Defense Technical Information Center 2

UK Defence Research Information Centre 2
Canada Defence Scientific Information Service 1
NZ Defence Information Centre 1

ABSTRACTING AND INFORMATION ORGANISATIONS

Library, Chemical Abstracts Reference Service 1

Engineering Societies Library, US 1
Materials Information, Cambridge Scientific Abstracts, US 1
Documents Librarian, The Center for Research Libraries, US 1

INFORMATION EXCHANGE AGREEMENT PARTNERS

National Aerospace Laboratory, Japan 1

National Aerospace Laboratory, Netherlands 1
SPARES
DSTO Edinburgh Library 5

Total number of copies: 43

 Page classification: UNCLASSIFIED

DEFENCE SCIENCE AND TECHNOLOGY ORGANISATION 1. CAVEAT/PRIVACY MARKING

DOCUMENT CONTROL DATA

2. TITLE 3. SECURITY CLASSIFICATION

Signal Processing in a Semi-active Seeker Document (U)
Title (U)
Abstract (U)
4. AUTHORS 5. CORPORATE AUTHOR
Luke Rosenberg, Yiding Hu and Bill Moran Systems Sciences Laboratory
PO Box 1500
Edinburgh, South Australia, Australia 5111
6a. DSTO NUMBER 6b. AR NUMBER 6c. TYPE OF REPORT 7. DOCUMENT DATE
DSTOTR-1606 013-164 Technical Report August, 2004
8. FILE NUMBER 9. TASK NUMBER 10. SPONSOR 11. No OF PAGES 12. No OF REFS
AIR 99/133 DGAD 38 11
13. DOWNGRADING / DELIMITING INSTRUCTIONS 14. RELEASE AUTHORITY
Not Applicable Chief, Weapons Systems Division
15. SECONDARY RELEASE STATEMENT OF THIS DOCUMENT

This work was undertaken as part of a Masters degree in conjunction with the Cooperative Research
Centre for Sensor Signal and Information Processing (CSSIP).
OVERSEAS ENQUIRIES OUTSIDE STATED LIMITATIONS SHOULD BE REFERRED THROUGH DOCUMENT EXCHANGE, PO BOX 1500
EDINBURGH, SOUTH AUSTRALIA 5111
16. DELIBERATE ANNOUNCEMENT

Approved for Public Release

17. CITATION IN OTHER DOCUMENTS

No Limitations
18. DEFTEST DESCRIPTORS

19. ABSTRACT

In publicly available generic digital simulations of semi-active Radio Frequency (RF) seekers, the com-
bined effect of multipath, sea-clutter and other non-linear noise sources are mainly studied in individual
models, while the effect in the seekers data processing has not been systematically investigated. This
report details these effects and how they have been combined to model the signal processing blocks
inside a semi-active seeker. A basic doppler detection scheme is then simulated to investigate the
consequence of these effects on target detection.
Page classification: UNCLASSIFIED
 M
c]

Australian Government
Department of Defence
Defence Science and
Technology Organisation

DSTO SYSTEMS SCIENCES LABORATORY

PO BOX 1500 EDINBURGH SOUTH AUSTRALIA 5111 AUSTRALIA
TELEPHONE (08) 8259 5555