ELECTRODE LENGTH

MEASUREMENT IN
ELECTRIC 'ARC FURNACES

NORMAN BALLARD, B.Sc.(Eng)

n
w
26 April 1995 To
e
ap
C
of
ty
r si

Thesis submitted to the Department of Electrical

ve

Engineering of the University of Cape Town in partial

ni

fulfilment of the requirements for the Degree of

U

M.Sc.(Eng).
 n
The copyright of this thesis vests in the author. No

w
quotation from it or information derived from it is to be

To
published without full acknowledgement of the source.
e
The thesis is to be used for private study or non-
ap
commercial research purposes only.
C
of

Published by the University of Cape Town (UCT) in terms

of the non-exclusive license granted to UCT by the author.
ity
rs
ve
ni
U
ACKNOWLEDGEMENTS

The author would like to thank the following advisors, colleagues and friends
for all their assistance with this project:

Prof M. R. lnggs, UCT

n
Prof B. Downing, UCT

w
Mrs C. Pretorius, Mintek
Dr I. Barker, Mintek To
Mr G. M. Tattersfield, UCT
e
ap
C
of
ty
r si
ve
ni
U
SYNOPSIS

Accurate measurement of electrode length in electric arc furnaces will result

in decreased maintenance time, and improved plant productivity. This thesis
describes the development of a microwave-based Soderberg electrode length-

n
measurement system.

w
Various methods of electrode-length measurement were investigated, and

To
it was found that a microwave measurement system based on a conven-
tional frequency modulated continuous wave (FMCW) radar presented the
e
most feasible technique. In this system, microwaves are propagated down a
ap
waveguide placed in the electrode. As the waveguide melts, they continue
C

propagating in the resulting cavity until they are reflected by the discontinu-
ity at the bottom of the electrode. The time taken for the return journey to
of

the bottom of the electrode and back is measured, and the electrode length
ty

calculated.
FMCW radar measures return time by mixing a transmitted linear fre-
i
rs

quency sweep with the reflection to produce the difference frequency. The
ve

resulting beat frequency is then proportional to the distance from the re-
ni

flecting object. An investigation into the required linearity of the 111icrowave

source showed that the linearity is crucial to obtaining high signal-t~ .; pu..riou..~
U

ratios in the beat frequency and therefore also to achieving good measure-
ment accuracy. Although the designed radar had built-in linearisation capa-
bilities, it was found to be more cost-effective to use a temperature-stabilised
linear voltage controlled oscillator.
Ultimately, the accuracy with which electrode length can be determined
depends on how accurately the peaks in the beat frequency spectrum can
be determined. Various spectral estimation techniques were investigated,
and it was found that the fast Fourier transform in conjunction with zero
padding and weighted averages provided a good combination of fast, robust
and accurate results.
A steel pipe with the same dimensions as the one to be placed in the elec-
trode was used to test the measurement system. Electrode tip erosion was

ii
 modelled by moving a refl~ctive plunger up the waveguide. Length measure-
ments were performed between 4.5m and 6.9m at one-centimetre intervals.
The RMS error associated with the measurements was found to be 1.37cm,
and the linearity of the system was excellent. Further experiments up to 9m
long confirmed the accuracy and linearity. Finally, a 40cm-long section of
Soderberg electrode was placed at the end of the electrode to confirm that
the waves would continue to propagate after the waveguide had melted. The
length-measurement system performed as well as in previous experiments,
with the electrode section having no adverse effect on the measurements.
Various other experiments to determine the effects of waveguide joints, tem-
perature. changes and electrode terminations also had favourable results.
A radar-based system was thus ·designed to measure the length of Soder-

n
w
berg electrodes in electric arc furnaces. The system was then built and

To
tested, and it was found that it provided excellent measurement accuracy
under a range of conditions. The system must now be implemented in a real
e
furnace in order to evaluate the effect of the environment further.
ap
C

I
of
ity
rs
ve
ni

~
•
U

lll
 Contents

1 INTRODUCTION 1

2 REVIEW OF POSSIBLE MEASUREMENT TECHNIQUES 5

n
2.1 Previous Attempts at Electrode-Length Measurement 5

w
2.1.1 Erosion models . . . . . . . . . . . . . . . . . . 6

To
2.1.2 acoustical detection of electrode tips . . . . . . 7
2.2 New Electrode-Length-Measurement Ideas Considered 7
e
•
2.2.1 Weighing the electrode . . . . . . . . . . . . . . 8
ap

2.2;2 Using the mechanical resonant frequency of the elec-

C

trode to deduce its length . . . . . . . . . . . . . . . . 8

2.2.3 Inserting an optical fibre into the electrode . . . . . . 8
of

2.2.4 The insertion of a waveguide into the graphite electrode 9

ity

2.2.5 Ground-probing radar techniques . . . . . 10

2.2.6 Sensing the hottest point in the furnace 10
rs

I 2.3 Summary of Electrode-Measurement Techniques 11

ve

I 3 CHOICE AND DESIGN OF GUIDING STRUCTURE 12

ni

3.1 The Type of Waveguide to be Used . . . . . . . . . . . . . . 12

U

I 3.2
3.3
Frequency of Operation and Waveguide Dimensions. . . . .
Techniques of Launching Electromagnetic Energy Into the
14

Waveguide. . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.4 Group Velocity in the Circular Waveguide . . . . . . . . . 20
3.5 Attenuation in the Steel and Graphite Waveguide sections 20

4 RADAR SYSTEM DESIGN 24

4.1 A Pulsed Radar Measurement System 24
4.2 Multi-Frequency Continuous Wave Radar 26
4.2.1 MFCW radar accuracy analysis .. 28
4.2.2 Problems encountered with MFCW radar 28
4.3 Frequency-Modulated Continuous Wave Radar 30

iv
 4.3.1 FMCW radar accuracy analysis . 32
4.3.2 The two types of FMCW radar . 32
4.3.3 Phase-locked-loop FMCW radar 33
4.3.4 Conventional FMCW radar . . . . . . . 33
4.3.5 The effects of nonlinear modulation on FMCW radar. 34
4.3.6 A comparison between the two systems ........ 39

5 IMPLEMENTATION 40
5.1 Technical Details of the Implementation of the Conventional
FMCW Radar .................. 40
5.1.1 Conventional FMCW radar parameters 40

- 5.2 Signal Processing of the Radar Returns 47

n
5.2.1 The fast Fourier transform 47

w
5.2.2 Prony's method . . . . . . 48

To
5.2.3 The MUSIC method . . . 49
5.2.4 Application to real data ..
e 49
ap
6 RESULTS 54
C

7 CONCLUSIONS 58
of

A Waveguide launcher design 62

ty

B Low frequency circuit diagrams 63

r si

c Circuit diagram of the interface card. 64

ve

D Signal processing techniques 65

ni
U

E Program listing of RADAR.CPP 67

v
•
List of Figures

1.1 Tapped electrodes. . . . . . . . . . . . . . 2

1.2 Simplified form of a Soderberg electrode .. 3

2.1 Cross section through a three-electrode electric arc furnace. 6

n
2.2 Temperature profile of a Soderberg self-baking electrode. . 9

w
2.3 Conductivity of graphite as a function of temperature. . . 11

3.1
To
Electric and magnetic field patterns in guiding structures. 15
e
3.2 Attenuation of circular modes in copper waveguide. . . . 16
ap
3.3 X-band (8-12GHz) TE 1o mode launcher . . . . . . . . . . 19
3.4 Measured attenuation of various T Eu mode launchers. . . 19
C

3.5 Apparatus used to find the attenuation of graphite waveguide. 22

of

3.6 Attenuation of graphite waveguide as a function of frequency. 23

ity

4.1 Train of radar pulses. . . . . . . . . . . . . . . 25

rs

4.2 System layout for a homodyne system. . . . . 27

ve

4.3 System layout for a superheterodyne system. 27

4.4 In-phase and quadrature-phase components plotted on or-
ni

thogonal axes. . . . . . . . . . . . . . . . . . . . . . 29
U

4.5 Multiple reflections found in a real MFCW system. . 30

4.6 Linear frequency modulation to obtain range. . . . . 31
4.7 Phase-locked-loop implementation of an FMCW radar. . 33
4.8 Conventional FMCW radar. . . . . . . . . . . . . . . 34
4.9 Voltage tuning characteristic for a GUNN oscillator. . . 35
4.10 Modulation at 0% and 100% nonlinearity. . . . . . . . . 37
4.11 Beat frequency spectra at 0%, 5%, 10% and 20% nonlinearity. 38
4.12 Signal-to-spurious ratio and bandwidth as a function of non-
linearity. . . . . . . . . . . . . . . . . . . . . 38
4.13 Quality factor as a function of nonlinearity. . . . . . . . . . 39

5.1 Schematic representation of electrode-measurement system. 41

vi
5.2 Linearity graphs for the DAC and.ADC. . . . . . . . . . 46
5.3 Spectral plots of simulated sinusoids at 2Hz and 2.2Hz. . 50
5.4 Section of the sampled output from the FMCW radar. . 51
5.5 Length estimates as obtained using the three spectral-estimation
techniques. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

6.1 Length measurement with simple FFT signal processing tech-

nique. . . . . . . . . . . . .1. • • • • • • • • • • • • • • • • ·• • • 55
6.2 Length measurements with zero padding and a weighted av-
erage. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
6.3 Extended length measurements. . . . . . . . . . . . . . . . . . 57
6.4 Length measurements though a section of Soderberg electrode. 57

n
w
To
e
ap
C
of
ty
r si
ve
ni
U

Vll
List of Tables

3.1 Cut-off frequencies for a steel tube of diameter 2cm (GHz) . . 15

3.2 Cut-off frequencies (GHz) for circular waveguides of different
..
size . . . . . . . . . . . . . . . 17

n
5.1 System parameters 42

w
5.2 VCO specifications 42

To
5.3 Circulator specifications 43
5.4 Interface card specifications 46
e
5.5 Interface card error specifications 46
ap

5.6 Summary of the different spectral-estimation techniques .. 51

C
of
ty
r si
ve
ni
U

viii
Chapter 1
•
INTRODUCTION

Submerged electric arc furnaces have been used since the beginning of the

n
century as a means of melting scrap steel and smelting ores into their as-

w
sociated metals and alloys. As the name submerged electric arc furnace

To
suggests, graphite electrodes immersed into the burden (mixture of ores,
reducing agents and fluxes) [2, page 1] are used to create an electric arc
e
ap
beneath it. Electrical energy supplied to the electrodes is transferred to the
burden in a combination of arc and resistive heating. Since these are low
C

voltage mechanisms (100V-400V) [2, page37], the currents delivered have to

of

be large to provide the necessary energy for the melting process. The high
currents are fed to the electrodes through fixed bus bars and heavy duty
ity

cables from transformers situated close to the furnace. Contact between

rs

the conducting cables and the electrode is maintained by means of brass or

ve

copper contact shoes which are held against the electrode using hydraulics
or spring pressure.
ni

Traditionally, fixed graphite electrodes that required replacing as they

U

became eroded were used. This replacement was found to be prohibitively

expensive due to lost production time. The tapped electrode (Figure 1.1)
which was later introduced as the first continuous electrode did not require
replacing, but excessive wear and breakage were common at the joints be-
tween alternate sections.
In the early 1920s, C. W. Soderberg [6] invented the Soderberg electrode
(Figure 1.2), which, like the threaded electrode, is continuous. The Soder-
berg electrode consists of a tubular steel casing made up of one-to-two-metre
sections which are filled with blocks of carbon paste. As the electrode is
slipped further towards the furnace, the blocks of carbon paste melt, and
then bake, to form solid graphite. The steel casing then melts off, leaving a
solid graphite electrode. Steel fins are sometimes welded onto the insides of

1
 Figure 1.1: Tapped electrodes.

n
w
the casings to hold the hardening graphite and to improve the conductivity

To
of the electrode. Feeder pipes used to introduce fines into the furnace may
also be inserted.
e
ap

Knowledge of the length of the electrodes is crucial for many reasons:

C

1. The furnace operator must not slip the electrode too far, causing it to
of

hit the bottom of the furnace.

ty

2. Maintenance of the reaction zones beneath the electrodes requires

si

accurate knowledge of their length.

r
ve

3. Power factor correction which reduces electricity costs substantially is

simplified if electrode length is known.
ni
U

Various attempts have been made to estimate electrode length using math-
ematical models, erosion profiles and measurements of electrical variables
affecting the furnace. However, these have produced rough estimates only,
and have obtained limited success [2, page 34], [12].
The objective of this study is to investigate various methods of electrode-
length determination, and to develop a prototype system which can be used
under conditions similar to those encountered in a real furnace. The sys-
tem should be able to measure the length of the electrode with a resolution
greater than lOcm. Although various methods of length determination will
be investigated, the main thrust of the project will be towards the use of
radar-based techniques.

2
 -----.. .-.----.1'-'f--------
..._-~-------------electrode casmg
upper-level floor
. . . - - - - - - - - hydraulic hoists

n
w
To
Figure 1.2: Simplified form of a Soderberg electrode.
e
The requirements of the project are as follows:
ap

1. An investigation into the various electrode-length measurement tech-

C

niques must be performed, and a suitable technique chosen'.

of

2. A theoretical evaluation of the proposed system must be performed to

ty

determine whether it will provide the required resolution.

si

3. A prototype system must be built and tested to determine its perfor-

er

mance.
v

The length-determination system is to be designed for a general electric

ni

arc furnace. In other words, it should not matter to the length-determination

U

I system whether calcium carbide or scrap steel is being melted. Although ev-
ery effort will be made to consider all aspects of the furnace environment, this ·
will be extremely difficult due to the harsh conditions encountered therein.
The prototype developed in this thesis will therefore have to be ruggedised
and tested in the real environment.
Various electrode-length-measurement techniques including microwave,
ultrasonic and parameter estimation methods are discussed in Chapter Two.
Microwave length measurement, in conjunction with a waveguide inserted
into the electrode, is found to be the most feasible. The specific microwave
measurement technique is then investigated andit is decided that a frequency
modulated continuous wave (FMCW) radar mounted above the electrode
should be used.

3
 Chapter Three deals with the design of the FMCW radar and waveguide
to be inserted into the electrode. A trade-off between phase-locked loop
(PLL) and conventional FMCW radar is presented. Conventional FMCW
radar is chosen, the system parameters are calculated and the entire system
is specified.
The implementation of the hardware is discussed in Chapter Four. The
Components that were used, along with the integration process, are de-
scribed. An evaluation of various signal processing techniques including the
fast Fourier transform (FFT), Prony and multiple signal classification (MU-
SIC) techniques is presented, and it is found that the fast Fourier transform
best fits the requirements.
The various experiments performed to evaluate the system are described

n
in Chapter Five. The results show that the system performs well under a

w
variety of conditions, and provides the required resolution.

To
e
ap
C
of
ity
rs
ve
ni
U

4
Chapter 2

REVIEW OF POSSIBLE
MEASUREMENT

n
TECHNIQUES

w
To
e
This chapter will review previous attempts to measure electrode length, and
ap

will introduce some new concepts. The most promising technique from those
C

presented will be chosen for further research.

The most common electric arc furnace, and therefore the one to be as-
of

sumed in all discussion, is the circular three-electrode furnace shown in Fig-

ity

ure 2.1. In this arrangement, three Soderberg electrodes are held in a ring
above the furnace, and extend into the burden. The materials to be melted
rs

are carried above the furnace on a conveyor belt, and then gravity-fed into
ve

the furnace through chutes. Once melted, the burden produces a slag, which
is normally less dense than the melted ore and therefore lies on the surfa~y
ni

and can be removed. The ore is tapped off through tapholes in the sides of
U

the furnace.

2.1 Previous Attempts at Electrode-Length Mea-

surement
The most widely-used technique for electrode-length determination is called
sounding. During sounding, the flow of raw materials is stopped, and the
burden lowered, so that when the furnace is switched off, and the electrodes
raised, one can see their ends, and therefore measure their length. Unfor-
tunately, the technique of sounding can only be used periodically since it
results in a significant reduction in the performance of the furnace.

5
I
•

electrode
L:'!!!•••~---- slipping
device

4--+;o:t;.;~:,;o;.;.;i'l---+""""'.....,~1------ electrical
contacts
===~~§tjiif===ffif!teiffl'"""""f.f.f.#fffi~~=----highcurrent
conductors

furnace
shell

n
w
To
Figure 2.1: Cross section through a three-electrode electric arc furnace.
e
All the other methods of length determination to be discussed in this
ap

chapter are continuous methods of electrode determination. By continu-

C

ous, it is meant that the furnace operation should remain unaffected by the
length~determination system. The technique of sounding, therefore, does
of

not fall into this category.

ity

2.1.1 Erosion models

rs
ve

The length of an electrode is a function of the rate of slipping and the

rate of electrode tip erosion. The rate of electrode slip can be measured
ni

with relative ease; however, the rate of tip erosion is an unknown quantity.
U

Physical erosion models that relate erosion rate to power consumed by each
electrode have been suggested. The prediction accuracy of the models is
dependent on the accuracy with which phase-power measurements can be
performed. For still further predictability, the effects of arcing between the
electrodes can be taken into consideration.
The erosion models cannot, however, take electrode imperfections, elec-
trode non-homogeneity or tip breaks into account. The errors associated
with the electrode erosion models are additive, and should not therefore
be relied upon over extended periods of time. For this reason, electrode
sounding must still be carried out on a r_egular basis. ·
Measurements of electrical arcing, resistance and reactance were used by
Barcza [2] to model the length of graphite electrodes. Electrical parameter-

6
based models have the distinct advantage over erosion models of being able
to detect tip breakages. The life of these models can therefore be extended,
and the time between soundings increased. The models, however, require
certain furnace-specific constants such as the conductivity of the burden,
which need to be updated regularly.

2.1.2 Acoustical detection of electrode tips

Under acoustical methods, both audible and ultrasonic frequencies will be
included since the mechanisms behind both techniques are similar. Sound
waves would be propagated from below the furnace bath, through the fur-
nace shell and burden to be bounced off the electrode tips. Similar tech-
niques have been used by [1] and [8] to measure the thickness of walls and

n
steel plates respectively.

w
The furnace shell has to be heat resistant, and id designed to have a

To
long life, since renewal is costly both in terms of materials and of loss of
production. For this reason, it has a temperature-resistant carbon lining
e
between the burden and the refractory wall. The floor on the inside of the
ap

steel furnace shell is covered with up to half a metre of refractory brick, a

a level surface and then up to
C

thin layer of refractory concrete to provide

seven layers of refractory brick and two layers of carbon blocks [2, page 7,8].
of

A. Jongens [5] of the acoustical department at the University of Cape

ity

Town investigated the possibility of using acoustical measurements to deter-

mine the length of electrodes. He found, however, that the complex nature
rs

of the furnace shell, and the inconsistent and unpredictable nature of the
ve

materials contained within the furnace, would have prevented any accurate
ni

measurement from being made in either the audible or ultrasonic frequency

U

range.

2.2 New Electrode-Length-Measurement Ideas Con-

sidered
Before a final electrode-length-measurement technique was chosen, many
new ideas were considered. Some of the ideas, such as weighing the electrode,
could be discounted immediately, whereas others required more thorough
investigation. The aim of this chapter is to guide the reader through each
technique considered, and to ex:plain the final choice that was made.

7
 2.2.1 Weighing the electrode
The electrode could be suspended in a frame which is mounted on load
cells. The load cells would measure the weight of the electrode, from which
its length may be calculable. Unfortunately, the weight of the electrode is a
complicated function which is a:ffected by the addition of electrode materials,
by the buoyancy provided by the slag in the furnace and by erosion of the
electrodes. The buoyancy problem could be overcome by lifting the electrode
slightly before measurement, but as mentioned before with reference to the
technique of sounding, this is undesirable.
Therefore, due to the relative complexity and number of unknown quan-
tities, weight measurement as an indication of electrode length will not be
considered any further.

n
w
2.2.2 Using the mechanical resonant frequency of the elec-

To
trode to deduce its length e
All physical objects have a natural frequency of vibration, commonly called
I
ap

• the resonant frequency. The resonant frequency of a simple pendulum, for

• instance, is given by:
C

I ~ (2.1)
of

f,,. =
•
ity

where l is the length of the pendulum and g is the acceleration due to gravity.
This suggests that if the electrode were modelled as a pendulum, and
rs

its resonant frequency were measured, then Equation 2.1 could be used to
ve

• determine its length. An accelerometer placed on the electrode could be

ni

used to measure electrode vibration and, therefore, resonant frequency.

Unfortunately the structure holding the electrode and the slag in the
U

furnace would tend to dampen free oscillations, making detection of the res-
onant frequency difficult. In addition, the resonant frequency may depend
on the electrode mass, which, as stated previously, would be extremely dif-
ft.cult to measure.

2.2.3 Inserting an optical fibre into the electrode

If an optical fibre could be inserted into the electrode, laser light could be
propagated down the fibre to the bottom of the electrode. The electrode
length could then be found using laser interferometry.
The majority of optical fibres melt below 600°C, and would therefore
not survive the high temperatures existing in the furnace (2500°C). If an

8
 TOP Temperatures in°c

n
Figure 2.2: Temperature profile of a Soderberg self-baking electrode.

w
To
optical fibre were found which could survive the high temperatures in the
electrode, then it is improbable that it would melt as it exits the electrode,
e
since, as is shown in Figure 2.2, the hottest point in the furnace is inside
ap

the tip of the electrode. Inaccurate results could therefore be obtained.

C

. 2.2.4 The insertion of a waveguide into the graphite elec-

of

trode
ity

A waveguide could be inserted into the graphite electrode. Electromagnetic

rs

(EM) energy could then be propagated down the waveguide, reflected at

ve

the end and returned. IT the time taken for the electromagnetic energy to
be transmitted, reflected, and received were determined, then the electrode
ni

length could be calculated. Although the strength of the end reflection is

U

unknown, the electrode tip is maintained just above the burden and could
therefore be modelled to a first approximation as an open circuit,
The waveguide to be used in the electrode could either be a hollow
pipe or a dielectric rod. IT a hollow pipe were used, the EM waves would
propagate down the tube, bounded by the conductive steel walls. In the
dielectric waveguide, they would propagate in the dielectric (a thermally-
resistant _ceramic rod), being bounded by the graphite of the electrode.
The length of the waveguide, and therefore of the electrode could be
measured using any one of the following radar techniques:

• Pulsed radar

• Multi-Frequency Continuous Wave (MFCW) radar

9
 • Stepped frequency radar

• Frequency Modulated Continuous Wav,e (FMCW) radar

• 2.2.5 Ground-probing radar techniques

The principle used in ground-probing radar is to transmit electromagnetic
energy into the ground with an antenna. Any object or discontinuity in the
ground would cause the electromagnetic energy to be reflected. These reflec-
tions are received and interpreted to determine the depth of the reflecting
source.
The depth to which the ground-probing radar can detect reflecting ob-
jects depends on a factor called the skin depth of the material being probed.

n
The skin depth is the physical depth at which the power density is half

w
(-3dB) of the transmitted power and is given by:

Oskin == ~
y-:;µ; To (2.2)
e
ap
where
C

• Oskin is the skin depth

of

• w is the frequency of the electromagnetic energy in r~d

ity

• µ is the magnetic permeability of the material being probed

s

• u is the conductivity of the material being probed

v er

Using a frequency of lOGHz, a relative permeability of 1 and a con-

ni

ductivity of lOOS/m for graphite, the skin depth has been calculated to be
U

0.5mm. Therefore, after an electromagnetic wave has propagated 0.5mm

into a piece of graphite, its power is halved. From Figure 2.3 (courtesy of
Mintek), we see that the conductivity of the graphite electrode increases as
its temperature increases, further reducing the skin depth. Therefore, since
the ground-probing radar would only be able to detect objects a fraction of
a millimetre deep in graphite, it would certainly not be able to detect the
end of a 20m electrode.

2.2.6 Sensing the hottest point in the furnace

Since the electrode tip is the hottest point in the furnace, one could use
sensors situated around the furnace bath to determine the position of the
hottest point within the bath, and therefore of the tip of the electrode.

10
 .e-40
(/)
~
-30
.~
>
;;
~ 20
"C
c
0
0
10

n
0 540 1130 1672 2230

w
Temperature (degrees celcius)

To
e
Figure 2.3: Conductivity of graphite as a function of temperature.
ap
C
of
ity

Unfortunately, the time it takes for heat energy to reach the sides of the
rs

bath is far too long; and, in addition, convection currents in the ore could
ve

cause heat pockets to move around, causing spurious results.

ni
U

2.3 Summary of Electrode-Measurement Techniques

Those measurement techniques which require a regular sounding campaign
to be carried out do not fulfil the requirement that the system should be
on-line. From the other techniques suggested, acoustical methods, mass de-
termination, resonant frequency analysis, optical fibre insertion, heat sensing
and ground-probing radar could be discounted quickly.
So in conclusion, the use of a radar technique to measure the length of a
waveguide inserted into an electrode has been found to be the most feasible
method among a number of possibilities. More specific determinations, of
both the type of radar and of the waveguide to be used, will require further
investigation.

11
Chapter 3

CHOICE AND DESIGN OF

GUIDING STRUCTURE

n
w
To
Microwaves are to be propagated down a waveguide that has been inserted
in the electrode. At some point the waveguide will melt, and the microwaves
e
must continue propagating down the resulting cavity. The various aspects
ap

of the guiding structure to be investigated are:

C

• The type of waveguide to be used

of

• The frequency of the propagated energy

ty

• The form of electromagnetic energy to be propagated d~wn the elec-

si

trode
er

• The size and cross-sectional shape of the waveguide

v
ni

• Techniques of launc~ing electromagnetic energy into the waveguide

U

• The signal velocity in the waveguide

• The losses incurred when the electromagnetic energy propagates in the

cavity subsequent to the melting of the steel waveguide

3.1 The Type of Waveguide to be Used

Waveguides can be classified into two main types: dielectric waveguides
or hollow waveguides. A dielectric waveguide consists of a solid central
core surrounded by a material with different dielectric properties. In the
case of an electrode, the central core could be a heat-resistant ceramic, and
the surrounding material would be the graphite of the electrode. A hollow

12
waveguide would consist of either round or rectangular conductive tubing.
The various merits of the two waveguide types are discussed below:

• Low-cost steel piping of standard dimension could be used to form the

hollow waveguide whereas a more costly ceramic rod would be required
for the dielectric version.

• Slag or ore from the furnace may cause clogging of the hollow wave-
guide. It could however be flushed periodically with compressed air.
Since the dielectric guide is solid, it would be immune to this problem.

• The joining of hollow waveguide sections could be performed with a

weld or simple clamp, whereas the dielectric waveguide would require
more precise alignment to prevent significant losses at joints,

n
w
• The attenuation suffered in the hollow waveguide would be simple to

To
calculate until the guiding structure me\ ts. Thereafter, the attenuation
due to the walls of the cavity in the electrode would be unknown. The
e
attenuation in the dielectric waveguide would depend on the nature of
ap

the graphite sheath surrounding the core, and would therefore also be
C

difficult to estimate.
of

e The dielectric waveguide relies on the graphite of the electrode being

able to provide a dielectric coating for the central core. In its first
ity

phase, however, the graphite in a Soderberg electrode is in the form of

rs

blocks of graphite paste. It is not until the blocks of paste melt that
ve

they would be able to act as a continuous coating for the waveguide

core. The dielectric waveguide would therefore have to be supplied
ni

with its own sheath already in place. This would further complicate
U

the joining of waveguide sections, and would increase the cost of the
system

• The ceramic core to be used in a dielectric waveguide would have to

be chosen with a melting point and erosion rate similar to that of
the electrodes in the furnace in question. Since the temperature and
rate of erosion of the electrodes vary substantially with the size of
the furnace and with the material being melted, the use of dielectric
waveguides would necessitate individual selection of the ceramic cores
at great cost.

From the ·above discussion it is clear that the hollow waveguide offers
a significant cost advantage over the dielectric waveguide, while being the

13
simpler of the two to implement. The hollow waveguide has therefore been
chosen as the guiding structure to be placed in the electrode.

3.2 Frequency- of Operation and Waveguide Di-

mens1ons.
.
Common RF oscillators (GUNN and YIG oscillators), are available with
operating frequencies in the X-band (8GHz-12GHz). The network analyser
to be used for testing is able to measure frequencies up to 12.4GHz, therefore
setting an upper limit. For reasons of availability and convenience, the
operating frequency will be centered around lOGHz. The rest of the system
must therefore be designed around this frequency.

n
The choice of cross-sectional shape is related more to the practical imple-

w
mentation of the system than to the theory of waveguides. The sections of

To
waveguide used will be joined together as the electrode is slipped. A cylin-
drical waveguide would be the easiest to align and connect together during
e
the joining phase. Cylindrical pipe is available cheaply in many sizes, and
ap

would be ideal as a waveguide. It was therefore decided to use cylindrical

C

rather than rectangular waveguide.

The transmission media determines the basic form of energy which is to
of

travel thro,ugh it. If we were inserting a coaxial cable into the electrode,
ty

then TEM [4, page 16] (Transverse Electromagnetic) waves would be propa-
si

gated. We are however using a waveguide, and are therefore restricted to TE

r

(Transverse Electric) or TM (Transverse Magnetic) modes of propagation.

ve

The various modes of TE and TM propagation are designated as T Emn,

ni

where m and n are integers specifying various modes. Certain of these modes
U

propagate in rectangular waveguides, and certain in cylindrical waveguides.

Figure 3.1 shows the electric and magnetic field pattern· in various rect-
angular and circular waveguides. For a full background on waveguides and
transmission of electromagnetic waves, see [7, 11, 4]. Results derived in these
texts will be quoted, and basic explanations given. However, derivations will
not be provided.
The various modes of TE and TM waves can only exist in waveguides
with certain cross-sectional dimensions. There is a minimum frequency
called the cut-off frequency which can be propagated in a waveguide with
given dimensions. Table 3.1 shows the cut-off frequencies for various modes
in a circular waveguide with a diameter of 2cm.
When one designs a waveguide to propagate a certain mode, the dimensions

14
 .....................................

························ ··········!.-
························ ··········!--
·············-···········--'···-•·I•

I
•

n
w
To
™11 e
Electric field lines Magnetic field lines
ap
C
of
ty

Figure 3.1: Electric and magnetic field patterns in guiding structures.

r si
ve
ni
U

Mode formula for cut-off cut-off freq (GHz)

c
TM01 ? 61~-n
11.48
c
TM11 'jl'Llh.11 18.29
c
TM21 1.224-a 25.51
c
TE01 1 ,:;,in.a 18.29
c
TE11 ~ <112-a
8.792
c
TE21 2 nc;7.a 14.58

Table 3.1: Cut-off frequencies for a steel tube of diameter 2cm (GHz)

15
I

I ATTENUATION (dB/m)
0.1~~~~~~~~~~~~~~~__;_~~~~~~~~~~~~~

o.oe
o.oe
0.04

0.02

0.01
0.008
0.006

0.004

0.002

4 a 10 12 14
FREQUENCY (GHz)

n
Figure 3.2: Attenuation of circular modes in copper waveguide.

w
To
of the waveguide must be such that the carrier frequency is higher than the
cut-off frequency for that particular waveguide.
e
Figure 3.2 shows the attenuation per metre of various modes in a cylin-
ap

drical copper waveguide. From Figure 3.2 we see that at higher frequencies
C

the attenuation per metre of T E 01 mode propagation drops off sharply. The
reason for this is as follows. The mechanism of attenuation in a waveg-
of

uide is the resistance of the waveguide walls. As the wave travels along the
ty

waveguide, the electric field couples into the waveguide walls, and power is
si

dissipated in the form of heat in the same way that a current flowing through
r

a resistor dissipates heat. Since the power dissipated in the waveguide walls
ve

is related to the density of the electric field being coupled into the walls,
ni

those modes which have a low density of electric field lines near the bound-
U

aries of the waveguide, experience less attenuation. Theoretically the T Eo1

circular mode has no electric field lines near the waveguide boundaries at
high frequencies, and therefore suffers little attenuation.
The physical dimensions of the waveguide must now be determined.
Since the different modes of propagation travel at different velocities along
the waveguide and cause ambiguous measurements, it is advantageous to
have pure modes propagating down the waveguide. Pure modes must there-
fore first be launched, and then maintained. In addition to causing atten-
uation, irregularities in the waveguide walls cause reflections of the prop-
agating wave, which may give rise to mode conversion. In other words,
modes are created by reflection which are different from the modes which
were launched, and this causes ambiguous measurements. Mode filters can

16
 Mode. diameter (cm)
1 2 4
TM01 22.96 11.48 5.74
TMu 36.59 18.29 9.14
TM21 49.02 25.51 12.25
TE01 36.59 18.29 9.14
TEu 17.58 8.792 4.39
TE21 29.17 14.58 7.29

Table 3.2: Cut-off frequencies (GHz) for circular waveguides of different size

be installed into the waveguide at reguJar intervals. However, these are

n
impractical for an electrode-length-measurement system.

w
One way to ensure pure mode propagation is to make sure that the mode

To
which was launched is the only mode which can exist in the waveguide. Ta-
ble 3.2 shows the cut-off frequencies of various modes in waveguides with
e
different radii. As can be seen from the table, if a lcm diameter waveguide
ap

is used at lOGHz, none of the modes will propagate. If a 4cm diameter

C

waveguide is used, all the modes except T M2 1 will propagate. A 2cm di-
ameter pipe allows only the T Eu mode to propagate. So if a 2cm diameter
of

pipe was used, pure T En mode would propagate, and no mode filter would
ity

be necessary to ensure unambiguous measurements. It is common practice

to operate a waveguide above its cut-off frequency, so the diameter of 2cm
rs

could be increased slightly if necessary.

ve

T En mode is also the easiest mode to launch into a waveguide. So

although, as can be seen from Figure 3.2, T Eu mode offers slightly more
ni

attenuation per metre than T Eo1, T Eu was chosen as the mode to be prop-
U

agated down the electrode due to the ease with which it can be launched
and maintained as a pure mode.
The waveguide is therefore to be cylindrical with an inner diameter of
2cm (or slightly more if necessary) to ensure pure mode propagation and
unambiguous measurements. A stainless steel pipe with an inner diameter
of 2.2cm was used for all further experiments due to availability.

17
3.3 Techniques of Launching Electromagnetic En-
ergy Into the Waveguide.
T En mode electromagnetic energy has been chosen to be propagated down
the electrode. A rectangular waveguide launcher (Figure 3.3) is used to
produce T E 10 rectangular mode electromagnetic energy. In order to convert
this mode to T E 11 circular, the rectangular waveguide can simply be placed
end-on-end with a circular waveguide. This is due to the fact that the electric
and magnetic field lines in T E10 mode rectangular and T En mode circular
line up, as can be seen in Figure 3.1. This technique is however inefficient
since not much T En mode energy is coupled into the circular waveguide.
Three possible techniques for maximising the power coupled into the circular

n
waveguide will be considered.

w
• Direct coupling between rectangular and circular waveguide

To
• A tapered rectangular-to-circular waveguide transition
e
ap
• A tapered rectangular-to-circular waveguide section including match-
ing screws
C

Direct coupling is the technique mentioned above in which rectangular

of

waveguide is simply attached to circular waveguide. The inefficiency of this

method is due to the impedance mismatch between the rectangular and cir-
ity

cular waveguide sections. This impedance mismatch results in a significant

rs

proportion of the transmitted energy being reflected at the transition point,

ve

and not being coupled into the pipe.

A tapered rectangular-to-circular waveguide transition presents a grad-
ni

ual impedance change from the circular to rectangular waveguide, and there-
U

fore increases the coupling into the circular waveguide. The length of the
taper should be ~ + ~ wavelengths long to produce the best match at the
center frequency (lOGHz). The tapered transition can be further enhanced
with tuning screws to match the two waveguide sections inductively.
The three methods were tested, and the results are shown in Figure 3.4.
The measurements were made by launching energy into a rectangular waveg-
uide with circular waveguide connected to it using the various techniques.
Microwave absorbent material was placed at the end of the circular waveg-
uide to minimise end reflections. The reflected power was then measured
relative to the transmitted power (Su reflection measurements).
As can be seen from the figure, the tapered section provides the best
match at the center frequency, as expected. However, this match deteriorates

18
 .-------mounting flange (WR-90)

• ~-t-----mounting
rr.:::««~««il4----ll-----
holes
aperture (22 .8mm x 1 O.2mm)
~--transmitting monopole antenna

• •

I FRONT VIEW

E------ mounting holes

_ _ _ _ aperture (22.8mm x 10.2mm)

n
---------transm1tt1ng monopole antenna

w
~---mountingflange (WR-90)
To
lllH-------coaxi al shaft section
e
ap

SIDE VIEW
C

Figure 3.3: X-band (8-12GHz) TE 10 mode launcher

of
ty
si

30
r

m
ve

"C
......
II)
25
--- ........
......--~~.. ·~-.-.:: ..~~
ni

·--~······-·--··-·····:·:~.... ,, ......""::::~······--
II) 20
0
-- - ---
U

1 5
...c
-
::I

Cl)

0::
10

1 0 .9 4 11 . 0 3 11 . 1 2
Frequency (GHz)

N o m a tc h in g
----------- Ta p e re d s e c t Io n
- - - Tapered section plus matching

Figure 3.4: Measured attenuation of various T En mode launchers.

19
as one moves away from the centre fre.qtiency. The tapered transition with
matching screws was chosen because it provides a good match across a wide
band of frequencies. The design of the launcher is shown in Appendix A.

3.4 Group Velocity in the Circular Waveguide

In order to measure elec~rode length, the time taken for EM energy to reach
the bottom of the electrode must be calculated. One of the radar techniques
discussed in Section 4 will be used for this purpose. To convert the calculated
time into a length measurement, the velocity of EM energy in the waveguide
must be .known.
The velocity at which EM energy travels in a waveguide is called the

n
group velocity, and is given by [7, page 528]:

w
To
e (3.1)

The cut-off frequency (Jc) can be calculated from the relationship given
ap

in Table 3.1 for T E 11 mode operation. It was found to be 8.99GHz in the

C

2.2cm diameter waveguide being used for investigation. The frequency of

operation (!0 ) is lOGHz, and the speed of light can be taken as the free
of

space value since Er (relative permittivity) and µr (relative permeability)

ity

are both unity in the hollow waveguide.

The group velocity of T Eu circular mode in a 2.2cm diameter steel
rs

waveguide has been calculated in to be 180 · 106 m/s. This is the velocity of
ve

EM energy which will be used in all further waveguide calculations.

ni
U

3.5 Attenuation in the Steel and Graphite Waveg-

uide sections
Each of the radar techniques to be considered for the measurement system
relies on a similar principle: reflected microwaves are interpreted to deter-
mine distance. In an electrode measurement system, microwaves will have
to propagate down an electrode and back up. It is important that the atten-
uation suffered by the microwave on its return trip should not be suffidently
large to prevent reception.
When microwaves are launched into an electrode, they will travel in a
steel waveguide. The attenuation suffered in ·the steel waveguide will be
negligible due to the high conductivity of the waveguide walls (see Section

20
3.2), and will be ignored. When the steel waveguide melts (above~ 800°C),
the wave must continue propagating in the resulting graphite-walled cavity.
This is where the majority of attenuation will occur. Figure 2.3 shows the
conductivity of graphite over a range of temperatures. One can see from this
graph that the conductivity of graphite is above 20kS/m at temperatures
above 800 °C. The approximate attenuation of T Emn modes [4] in dB per
metre is given by:

8.6861¥- (
a = --_,1==aw=2 . k2 . ~ - p2
2 + w-:-2
2) (3.2)
2·ayl-7
where

n
• a is the diameter of the graphite waveguide section (a = 0.022m)

w
To
• w is the frequency of operation in rad/s (2 · 7r • lOGHz)

• We is the cut-off frequency in rad/s (2 · 7r • 7.99GHz)

e
ap

• CT is the conductivity of graphite (20kS/m)

C

• €0 is the permittivity of free space ( 36 .;. 109F/m)

of

• k is the wave-number at the cut-off frequency, and is given by: k = 7'-

ity

• p is the mode number (p = 1)

rs

From Equation 3.2, the attenuation of a circular graphite waveguide with a

ve

diameter of 2.2cm was found to be 1.76dB/m.

ni

So, if the wave had to propagate down the last five metres of the electrode
U

and back, 2 · Sm · 1. 76dB /m= 17 .6dB attenuation would be realised. Since

a typical microwave source can deliver 10 dBm, and a typical. receiver can
detect -90 dBm, a maXimum attenuation of 100 dB can be tolerated. Other
factors will however reduce this figure. 17.6dB attenuation in the graphite
section is therefore large, but not prohibitive.
Unfortunately, it could not be determined at what frequency the con-
ductivity measurements had been made, and since conductivity is dependent
on frequency, the theoretical result could not be relied upon. Therefore, an
experiment (Figure 3.5) was performed to determine the low-temperature
attenuation of a section of graphite waveguide. Figure 3.6 shows the mea-
sured attenuation as a function .of frequency. It can be seen from the figure
that the measured attenuation ranges from O.ldB/m to 0.8dB/m, and is
therefore lower than the theoretical value. Heating the graphite waveguide

21
 I+- waveguide
launcher

I
D
hole in
--------it--graphite
sample

I.-------------~-~coa~al
cable

121.

n
w
_ _ _ _ network
HP8410

To
analyser
e
ap

Figure 3.5: Apparatus used to find the attenuation of graphite waveguide.

C
of

to 265°C made no detectable change in the attenuation. At the physical

position in the electrode where the wave would propagate in graphite, the
ity

temperatures are much higher than those in the experiment. The conduc-
rs

tivity would therefore be greater, and the attenuation would be lower. If the
ve

attenuation of the graphite were found to be too large, then the diameter
of the tube could be increased slightly; or alternatively, operation of T E 01
ni

mode instead of T E1 l mode propagation could be considered.

U

Other losses in the system would include:

• mismatches in the waveguide launcher, and in the transition between

rectangular and circular waveguide

• reflections at the transition between the steel and graphite waveguide

in the electrode

• attenuation of high frequency cables and connectors

The majority of these could, however, be kept low with good design and
efficient implementation.

22
 n
w
-- To
e-o.5
m
e
'C
ap

c -1
~
C

:::J
c
of

CD
:i 1.5
ty
si

-2
r

...._.__...____._..___.__~~~-'--......__.......___.___,__.___._.
ve

9.750 10.250
Frequency (GHz)
ni
U

Figure 3.6: Attenuation of graphite waveguide as a function of frequency.

23
 Chapter 4

RADAR SYSTEM DESIGN

I

n
w
To
There are various radar techniques which fall into the two broad categories
of pulsed radar and continuous wave radar. Each of the two methods will
e
be considered with reference to the electrode-length-measurement problem.
ap
A suitable radar will then be designed using the chosen technique.
C

4.1 A Pulsed Radar Measurement System

of

The principle used in a pulsed radar is that the time (t) taken for a single
ity

pulse to be transmitted, reflected by a target and received is measured. The

rs

distance to the target, or the length of the electrode (R) is then given by
ve

Equation 4.1, (13, page 109]:

ni

R = Ug ·t (4.1)
U

2
where u 9 is the group velocity in the waveguide (electrode), and the factor
!
of a accounts for the measured time representing the full return time to
and from the target. In practice, a train of pulses as shown in Figure 4.1
would b'e transmitted.
The time between pulses is set so that the transmitted pulse has had
time to return before a new one is sent. The maximum range of a pulsed
radar is therefore set by the time between pulses, or the rate at which pulses
are transmitted (pulse repetition frequency or PRF).
For the electrode measurement system, the maximum length of the elec-
trode is assumed to be 20 metres. From Section 3.4, the velocity of energy
transmission (group velocity) is known to be 180 · 106 m/s. The maximum

24
 Tp\

Figure 4.1: Train of radar pulses.

n
PRF is given by (13, page 110] as:

w
To
PRF = _3!:L_ (4.2)
2·R e
So the maxj.mum PRF for the electrode measurement system would be
ap
4.5MHz which means that 4.5 million electrode-length measurements could
be made per second. However, a few measurements per second would suffice.
C

It can be shown that the resolution of a pulsed radar system depends on

of

the width of the transmitted pulse, with higher resolutfon being obtained
with narrower pulse widths ( T ). The range resolution is given by:
i ty
rs

l:l.R = u9 • r (4.3)
2
ve

In order to obtain the high resolution required in the e~ectrode mea-

ni

surement system, a very narrow pulse width is required. For 5cm range
resolution, the pulse width needs to be 490 picoseconds, which is extremely
U

narrow. The approximate bandwidth of the system required to transmit

and receive such a pulse is given in hertz by the inverse of the pulse width in
seconds, and is found to be 2GHz. This may be realisable if the transmission
medium were free space, as is the case with most radars. The bandwidth
of the waveguide, which is the transmission medium in this case, should
be kept as low as possible to prevent dispersion, and to ensure single-mode
propagation.
The high bandwidth of a pulsed radar system therefore makes it imprac-
tical for use in the electrode-length-measurement system; other methods will
now be considered.

25
 4.2 Multi-Frequency Continuous Wave Radar
Multi-frequency continuous wave (M:fCW) radar is widely used in survey-
ing equipment such as tellurometers. It is hoped that it will offer some

- advantages in an electrode-length-measurement system.

In MFCW radar, a high frequency signal is transmitted, reflected off a
target, and received. The phase shift incurred during this journey is then
measured to determine the distance travelled by the signal.
Range (R) as a function of phase shift(</>) is given by:

R= </>. Ug (4.4)
2(211". !)
Unfortunately, the phase is ambiguous, having a period of 211". Therefore,

n
ranges of larger than half a wavelength (:::::1.5cm) will be ambiguous. This

w
problem <;:an be overcome if the phase shift due to more than one frequency

To
is measured [9]. Equation 4.4 then becomes:

= 11</> •
e
R Ug (4.5)
2 . (2 . 1r • 11!)
ap

where 11/ is the difference between the two measurement frequencies, and
C

11</> is the difference in measured phase. From Equation 4.5, the unambigu-
of

ous range is found when (11</> = 2 · 7r) or:

ity

(4.6)
rs

A system must therefore be built that can be used to measure the

ve

phase difference between two high-frequency signals. There are various well-
ni

documented systems, which can be broken down into two main categories:
U

• Homodyne or zero intermediate frequency (IF) systems

• Superheterodyne or non-zero IF systems

In zero-IF systems, phase difference is measured at high frequency, whereas

in non-zero-IF systems, phase information is modulated down in frequency
and measured at a standard intermediate frequency. This allows the use of
high-quality, narrow-band amplifiers and filters, which results in improved
noise performance. The superheterodyne receivers therefore offer improved
performance when compared with homodyne systems. Figure 4.2 shows the
system layout for a homodyne receiver, and Figure 4.3 the system layout for
a superheterodyne system.

26
 Circulator Antenna
Power

Oscillator
splitter

00
00

0
Quadrature
power 900
splitter 1L..:;..:;:___ _ _- - {
00

Quadrature

n
w
In phase To
e
ap

Figure 4.2: System layout for a homodyne system.

C
of

Circulator Antenna
ity

Power

• Oscillator
splitter 00

0
rs

00
ve
ni
U

Filter
Mixer
Coherent
local
oscillator

Quadrature
power o0
splitter 0
90

In phase Quadrature

Figure 4.3: System layout for a superheterodyne system.

27
As can be seen from Figures 4.2 and 4.3, the superheterodyne system re-
quires more hardware than the homodyne system.
Both systems use a similar principle. A signal is propagated down the
electrode and reflected at the bottom. This reflected signal is then mixed
with the transmitted signal to determine their phase difference. In practice,
the received signal is mixed with in-phase and quadrature components of
the transmitted signal. In this way, the in-phase (I) and quadrature-phase
(Q) components are formed. The phase can then be calculated as:

<P = arctan( ~ ) (4.7)

4.2.1 MFCW radar accuracy analysis

n
Partial differentiation of Equation 4.5 yields the range accuracy using MFCW

w
radar:

1
6R(~¢,~f)
6( ~</J)
I
+ To
16R(~</>,~f)
6( ~!)
I
(4.8)
e
fR = · €t,q, · fAJ
ap
From which we get the normalised range accuracy as:
C

(4.9)
of

Where:
ity

• R is the length of the electrode

rs

•
ve

~</>is the difference in measured phase at the two measurement fre-

quencies
ni

• ~f
U

is the frequency difference between the two measurement frequen-

cies

• ER, Et,q, and fAJ are the respective errors in the above quantities

For an electrode length of 20m, a frequency span of 500MHz, a phase

error of 5 degrees and a frequency span error of lOOkHz, the range error is
found using Equations 4.5 and 4.9 to be 6.5mm. This value is well within
the system requirements.

4.2.2 Problems encountered with MFCW radar

Some simple tests were done using a system similar to that shown in Figure
4.2 to verify the principles used in MFCW Radar.

28
 GI
Ill
~ 2 t---.1T----------------'t'r--i
c.
I

-~ Qf-+~~~~~~~~~~-'-'-~-*'---1
,, -2
I!
111
::I
t--+--------------+--i
a

n
-4t---=~~~~~~~~7L-~-l

w
To
-6'--_.__~~-..___._____.i_~-"'--_._____.i_~----'

-6 -4 -2 0 2 4 6
In-phase e
ap
C
of

Figure 4.4: In-phase and quadrature-phase components plotted on orthogo-

nal axes.
ity

A reflective plunger was placed in a steel pipe, and microwaves were

rs

propagated down the pipe. As the plunger was pushed up the pipe, the
ve

in-phase and quadrature-phase components were expected to trace out sine

ni

and cosine functions, or a circle if plotted on orthogonal axes. A homodyne

U

MFCW radar system was implemented to measure the phase components.

The resulting phase was found to be inaccurate, and time-varying. After
some experimentation, it was found that the microwave mixer was prone to
DC drift over time, and that this drift was causing distortion of the phase
components.
A heterodyne system overcame the temperature-drift problems associ-
ated with the mixer. The in-phase and quadrature components were once
again plotted against each other, and the result is shown in .Figure 4.4.
As can be seen from the figure, the shape of the graph is fairly close to
that of a circle. Applying the phase measurements to Equation 4.5 gave an
accuracy of approximately lOcm using this system.
There was however another source of error. The MFCW radar does

29
 Q

Target .
Resultant., ..

n
w
To
e
ap

Figure 4.5: Multiple reflections found in a real MFCW system.

C
of

a vector measurement of the reflected wave (amplitude and phase). This

: vector will be made up of all the small reflections which occur along the
ty

pipe. Under laboratory conditions, these minor reflections would be static,

si

and could be calibrated out. In the real system, however, there will be larger
er

reflections (Figure 4.5), and they may be dynamic (e.g. the reflection from
v

the transition between the steel pipe and the inner core of the electrode). So
ni

although the MFCW Radar offers excellent measurement accuracy in some

U

cases, it cannot be used in an electrode-length-measurement system.

4.3 Frequency-Modulated Continuous Wave Radar

FMCW radar is used in aircraft altimeters and radar speed traps. It is one
of the simplest and cheapest radars to implement, while being robust and
reliable.
In an ideal FMCW radar, a linear ramp of frequencies as shown in Fig-
ure 4.6 is transmitted., The frequency difference at any instant in time
between the transmitted and reflected signals determines the target range.
The time -delay between transmitted and received waveforms is given by:

30
 fo + B/2 Tran sm itte d
>-
Received ······························
0
c:
Q)
::I
~ fo IG-~-"'-~~~~~~~_,._~.....:....,.~~~~~~~~~~-=-~__,-

~ Time
u.
........ -----------··
fo -B /2

>-
~ f Difference (beat)
~equency
!...___l~_v~~~-v~~~~.
-

Time

n
Figure 4.6: Linear frequency modulation to obtain range.

w
T=-
2·R
To (4.10)
e
Ug
ap

where u 9 is the velocity of electromagnetic energy in the transmission medium.

C

The frequency difference between transmitted and received signals is

. called the beat frequency, fb, and can be found from Figure 4.6 as:
of

= T . !' ~ 2RJ'
ty

fb (4.11)
Ug
si

where f' is the rate of frequency change, and is given by:

r
ve

!' = !. (
4.12)
ni

2
U

: where B is the maximum frequency deviation and Tm is the period of the

modulation. The modulation frequency Um) is given by :zt.
Using Equations 4.11 and 4.12, we get an expression for range as:

R= Ug. fb
(4.13)
4·B·fm
The values of fm and B can be chosen once the range and beat frequency
have been selected. In the case of the electrode-length-measurement radar,
the range R falls within the range 5m to 20m. The beat frequency will
depend on the type of FMCW radar being implemented. The bandwidth
(B) should be maximised, since it determines the resolution of the radar.

31
4.3.1 FMCW radar accuracy analysis

The range resolution ( 6.R) can be found by taking partial derivatives of

Equation 4.13:

Simplifying Equation 4.14 gives:

f.fb f.Jm f.JB) R

f.R= ( -+-+-
fb fm B
. ( 4.15)

From Equation 4.15, we see that the only way to increase the resolution

n
is to decrease the percentage error in each measurement. For a range of 15m

w
and a resolution of 5cm, the percentage error allowed for each parameter is

To
only about 0.13. This is a stringent requirement which will necessitate
investigation into various error-reducing techniques.
e
Although the error requirements associated with the FMCW radar are
ap
stringent, they should be realisable. Furthermore, since in FMCW radar, a
C

reflection is represented by a frequency output proportional to its distance

from the radar, any spurious reflections can simply be filtered out. So an
of

FMCW radar system provides target differentiation and does not require
ity

calibration as did the MFCW radar. In addition, the FMCW radar allows
a large degree of freedom when choosing system parameters. For these
rs

reasons, it was decided to use FMCW radar, in conjunction with a waveguide

ve

inserted into the electrode, as the electrode-length-measurement technique

to be investigated further.
ni
U

4.3.2 The two types of FMCW radar

There are two main methods of implementing an FMCW radar. The first
method will be termed conventional FMCW radar, and the other method
phase-locked-loop FMCW radar. Block diagrams of the two methods will
be,shown, and their various merits discussed. Looking at Equation 4.13,
one sees that the range is dependent on the beat frequency, the modula-
tion frequency and the frequency deviation. So, in principle, any two of the
three dependent variables could be kept constant. A measurement of the
third variable would then provide the information to determine the elec-
trode length. For practical reasons, the frequency deviation is always kept
constant, and either of the other two variables is measured to determine
range. When the beat frequency is measured to determine range, we call

32
 Voltage 3-port
controlled l+-......-'H circulator,___ ___,
oscillator

Waveguide
r-- -----------, launcher

ii
1 I
I I
I I

n
I I

w
I Phase
I

To
comparato ,._---~ Waveguide
I Loop
10.7 MHz (electrode)
I filter
IL _____________ _JI oscillator
e
ap
C

Figure 4.7: Phase-locked-loop implementation of an FMCW radar.

of

the radar a conventional FMCW radar, and when the modulation frequency
ity

is the measured variable, we have a phase-locked-loop radar.

rs

4.3.3 Phase-locked-loop FMCW radar

ve

In a phase-locked-loop (PLL) FMCW radar, the beat frequency is maintained

ni

using a PLL by changing the modulation frequency., The inverse modulation

U

frequency is then measured to determine target range.

Advantages of PLL FMCW radars include:

• The hardware required is fairly simple and cheap.

• A PLL could track a very noisy beat frequency accurately.

4.3.4 Conventional FMCW radar

In conventional FMCW radar, the beat frequency is measured directly to
determine target range. Various techniques, ranging from frequency coun-
ters to spectrum analysers and signal processing techniques, can be used
to measure this frequency. Figure 4.8 shows a conventional FMCW radar
system.

33
 Triangle Voltage 3-port
wave controlled ~......--;Hci rculatori--~---.
generator oscillator

Waveguide
launcher

Spectrum 1-+~~~~~~~--'

n
analyser Waveguide

w
(electrode}

To
e
Figure 4.8: Conventional FMCW radar.
ap
C

Advantages of the conventional FMCW radar include:

of

• '.fhe system is easy to interface to a digital computer.

ity

• Complex signal processing can be done on the heat frequency spectrum

1
if necessary, making the system robust
rs
ve

• The modulation frequency remains constant. This makes linearising a

microwave VCO easier if it is found necessary.
ni
U

• It is conceptually easier to understand than the PLL FMCW radar,

and easier to implement.

• The beat frequency is directly proportional to target range.

4.3.5 The effects of nonlinear modulation on FMCW radar

All the radar equations rely on the components within the system being
linear. It can be seen 'from Figure 4.9 that the tuning characteristic of a
GUNN voltage controlled oscillator (VCO) is nonlinear, and therefore the
equations do not strictly apply. The standard approach is to linearise the
oscillator digitally. However, Skolnik [9, page 84] mentions that any periodic
waveform can be used in an FMCW radar, and that the range is then given
by the average beat frequency over a modulation period. Skolnik's approach

34
 11.04 .------------------~

11.0'2
'N·

-6 11
~ 10.98
g 10.96
~ 10.94

n
C"
e

w
LL 10.92

To
10.90
10 20 30 40
Tuning voltage M
e
ap
C

Figure 4.9: Voltage tuning characteristic for a GUNN oscillator.

of

will be investigated in this section before any digital solutions are considered.
ity

If the linearity is found to be crucial, a YIG (Yttrium Iron Garnet) oscillator

with high linearity could be used. However, the YIG oscillators are expensive
rs

and will be avoided if possible.

ve

The phase-locked-loop (PLL) implementation relies on the PLL being

ni

able to lock onto the beat frequency and keep it constant. This requires that
the beat frequency be clearly distinguishable from any additional harmonics
U

that might be present due to nonlinearities in the system. A conventional

FMCW radar measures the beat frequency directly as a measure of range.
Although digital signal processing can be used to enhance the beat frequency,
the resolution obtainable will ultimately depend on the available signal to
noise ratio. So if nonlinear modulation is to be used, the spectral purity
of the beat frequency needs to be investigated. Secondly, it was found in
Section 4.2 that target differentiation is necessary. In a linear FMCW radar
system, different targets give rise to different beat frequencies and can be
differentiated; the unwanted frequencies due to reflections other than from
the target can simply be filtered out. It must _be ensured that if nonlinear
modulation is to be used, the spectral components from different targets do
not overlap, or target differentiation will no longer be possible.

35
 As mentioned before, an ideal FMCW radar would transmit a continuous
linear frequency sweep. The frequency difference at any instant in time
between the linear sweep and a delayed version results in the constant beat
frequency. As a first look into the effect of nonlinearity, we can see that if the
transmitted signal were not an exact delayed replica of the reflected signal,
there would be small deviations in the difference between the two frequencies
and hence in the beat frequency. The beat frequency would therefore no
longer appear as a single spectral line, but would have a smeared frequency
distribution.
A real FMCW radar could at best transmit a sawtooth or triangularly-
modulated waveform. Figure 4.6 shows a triangularly-modulated signal, a
delayed version and the resultant beat frequency. Since the beat frequency

n
is the difference between the transmitted signal and a delayed version, tri-

w
angular dips are seen to occur at the peaks and troughs of the modulation.

To
One would therefore expect the beat frequency spectrum to contain a strong
component representing the distance to the target. In addition, there would
e
be harmonics at twice the modulation frequency centered around this main
ap

frequency.
C

Since the signal spends most of its time at the beat frequency, the power
of

contained in this frequency component would be far larger than that con-
tained in the harmonics due to the triangular dips. However, with increasing
ty

nonlinearity in the modulation, increased spectral power would occur in the

i
rs

harmonics.
These ideas were tested by simulating an FMCW radar with the program
ve

Mathcad. The modulation used took the form of a ramp with a varying
ni

nonlinearity added. Equation 4.16 was used to create the modulation.

U

1
y(x) = --b(ax
a+
+ bx 2 ) O<x<l ( 4.16)

The factor ah was necessary to keep the bandwidth (B) of the frequency
sweep constant. With reference to the equation, nonlinearity was defined as
!, so that 100% nonlinearity corresponds to a = b. Figure 4.10 shows the
modulating sawtooth wave with 0% nonlinearity and 100% nonlinearity.
Figure 4.11 shows the beat frequency spectra for 0%, 5%, 10% and 20%
nonlinearity. As can be seen from the figure, the linearity of the modulation
is crucial to obtaining a clean, easily discernible beat frequency. In order to
describe and quantify the effects of nonlinearity, two terms must be intro-
duced. The signal-to-spurious ratio is defined as the average power in the
beat frequency divided by the average noise or spurious output power, and

36
 100%
_...,.._
~0.8
-!
~
0.6
0%
---·
-I:; o.4
a.

n
'5 0.2

w
0

To
Vmax
Input voltage M
e
ap
C

Figure 4.10: Modulation at 03 and 1003 nonlinearity.

of

can be given in dB as:

ty

signal = 10 ·lo (beat) ( 4 .17 )

si

spurious g noise
r
ve

The bandwidth is defined as the 3dB bandwidth in the standard manner.

The signal-to-spurious ratio and the bandwidth were measured for non-
ni

linearities ranging from 03 to 1003. As can be seen from Figure 4.12, there
U

is a strong dependency on the nonlinearity factor. A form of quality factor

which is given by the signal-to-spurious ratio divided by the bandwidth is

t~~ ) - -~ L-~
~-~--)--~;'_"_}
pJ ).}µ)..
"' ~AU1~)
l plotted in Figure 4.13.
The accuracy of the distance measurements which can be obtained with
an FMCW radar is directly related to how accurately the beat frequency
• , _,_t-,.£ \' ~- · can be measured. A signal with a high signal-to-spurious ratio and low
1
v

Yt, 'Jr..J.~ bandwidth can be measured far more accurately than one with a low signal-
~ J..D(J,_°\_e,,~4-.) · to-spurious ratio and high bandwidth. It can therefore be seen that to obtain
_ )\ accurate measurements from an FMCW radar, the modulating source has
to be highly linear.
A quadratic regression to the GUNN tuning curve showed that it exhibits
a 613 nonlinearity factor. The GUNN oscillator would therefore have to be

37
 O% nonlinearity 5 % nonlinearity

... ....., .. ....

7500
Frequency (Hz) Frequency (Hz)

10 % nonlinearity 20 % nonlinearity

n
w
To
e 7500
ap
Frequency (Hz) Frequency (Hz)
C

Figure 4.11: Beat frequency spectra at 0%, 5%, 10% and 20% nonlinearity.
of
ity
rs
ve

6,000 16
ni

bancmdth signal-to-spurious 14
U

12~
g4.000
10l
13,000
:6
8 Zl
c 6 ~
~ 2,000
4 i
.!iZI
1,000
2 "'
0
100
Nonlinearity (%)

Figure 4.12: Signal-to-spurious ratio and bandwidth as a function of non-

linearity.

38
 N

...~ cG
.
2
0.1

.....
u
0.03
>-
..
=::s
0.01

0.003
a
0.001

0.0003
0 100
Nonlinearity (%)

n
Figure 4.13: Quality factor as a function of nonlinearity.

w
To
linearised if either a PLL or conventional FMCW radar system were to be
used.
e
ap

4.3.6 . A comparison between the two systems

C

The PLL FMCW radar offers simple, cheap hardware and the ability to track
of

a fairly noisy beat frequency signal. The problem, however, occurs when the
beat frequency becomes smeared due to nonlinearities in the system or be-
ty

cause of multiple targets. At this stage some 'intelligence' would be of great

i
rs

advantage. The FMCW radar's compatibility with digital computers, and

ve

the consequent ability to perform signal processing, give rise to a versatile

and robust system. In addition, changes can be made to the operating of
ni

the system without a significant change in hardware being required.

U

The FMCW radar measurement system will have to operate in a harsh,

unpredictable environment. Although PLL FMCW radar does offer signif-
icant advantages over conventional systems, the robustness of the conven-
tional system makes it more suitable for this application.

39
Chapter 5

IMPLEMENTATION
(

In this chapter, the technical aspects of the implementation of the radar will

n
be discussed. The hardware and software dedicated to signal processing will

w
also be described.

To
e
5.1 Technical Details of the Implementation of
ap

the Conventional FMCW Radar

C

The conventional FMCW radar technique offers a large degree of flexibility

of

that is not present in any of the other techniques discussed so far. Con-
ventional FMCW radars such as those used in altimeters employ simple
ity

frequency counters to determine the beat frequency. However, an aircraft

rs

altimeter operates in a fairly predictable environment, where it is known

ve

that the land surface will be the major source of reflection. In a furnace,
no such prediction can be made about the environment. A microprocessor
ni

linked to the radar would give the system a certain amount of 'intelligence',
U

which will be necessary in the furnace environment. The electrode mea-

surement system to be implemented will therefore consist of a conventional
FMCW radar linked to a computer. The computer will have the dual task
of providing input to the radar, and of analysing the radar output.

5.1.1 Conventional FMCW radar parameters

The various subcomponents of the radar system to be discussed are the
microwave, analogue and digital sections (Figure 5.1).
For reasons that will be explained in the sections below, the parameters
of the FMCW radar were chosen as follows:

40
 M~CWAVE SECTIOll

ANAl.OOUE SECTION

n
(

w
To
'---1----1 LEVELAll.I
e
ap
C

LEVEL All.I
of
ity

llGITAL SECTION
rs
ve
ni
U

RAM RAM

I
~---l
-~
.. _
181111 I

Figure 5.1: Schematic representation of electrode-measurement system.

41
 Centre frequency lOGHz
Sweep bandwidth 500MHz
Modulation frequency 60Hz
Modulation type sawtooth

Table 5.1: System parameters

The high-frequency component of the measurement system

A high-frequency voltage controlled oscillator (VCO) with a centre frequency

of lOGHz (Section 3.2) acts as a microwave source. Radar resolution is
known to be inversely proportional to bandwidth, as will be shown later in
Equation 5.6. The bandwidth must therefore be maximised, while remem-
bering that higher bandwidth results in greater waveguide dispersion. It

n
was found experimentally that a bandwidth of 500MHz resulted in the best

w
overall system performance. The various VCOs investigated were:

Manufacturer
Avantek
Model no.
VT08950 To
Frequency range (GHz)
9.5GHz - 10.5GHz
Cost (R)
4095
e
Avantek AV74010 8.0GHz - 12.4GHz 14000
ap

Hewlett Packard HP8530b O.lGHz - 14GHz 20000

C

Table 5.2: VCO specifications

of

It was found in Section 4.3.5 that the microwave VCO linearity was
ty

crucial. The VT08950 is a low-cost varactor-tuned oscillator with a nonlin-

si

ear tuning characteristic that is temperature-dependent. It would therefore

r

require temperature-dependent linearisation over the vast range of temper-

ve

atures present in the furnace environment. Both The AV74010 (YIG) and
ni

HP8530b are temperature-stabilised and extremely linear. The HP8530b

U

sweep oscillator was used because it was rela,tively easily available. How-
ever, in a practical version of the system, a YIG tuned oscillator would be
substituted. The oscillator had a power output of lOdBm, which was flat
across the system bandwidth. A regular power output is important to pre-
vent harmonics due to amplitude modulation occurring in the beat frequency
spectrum.
A 3dB Wilkinson power splitter delivered half the power to the mixer
(7dBm), and half the power to the electrode through a T E 11 mode waveguide
launcher. An ordinary power splitter could have been used, but the Wilkin-
son power splitter has the advantage that it does not allow power reflected
by mismatches in the system to pass back through the splitter. Reflected
power arriving back at the oscillator causes an effect called frequency pulling,

42
 Model number MRI W-90 WJX-NC
Centre frequency lOGHz
Bandwidth 500MHz
Isolation min 35dB
VSWR max 1.3
Insertion loss max 0.4dB
Waveguide type WR-90
Flange type UG 135/U

Table 5.3: Circulator specifications

during which the frequency of the oscillator can change substantially.

An X-band (8GHz to 12.4GHz) waveguide circulator was used. The
measured isolation was above 34dB over the band 9.75GHz to 10.25GHz,

n
and the insertion loss negligible. Table 5.3 gives the full specifications of the

w
circulator.

To
A double-balanced mixer was used to find the difference in frequency
betwe;n the transmitted and received signals because of its superior inter-
e
modulation product rejection capabilities [16]. The RF and LO inputs of
ap

the mixer covered the range of X-band frequencies, and the IF output was
C

DC-coupled. It is important that the beat frequency should fall within the
\J
bandwidth of the IF output port. The LO input required an optimum of
of

6dBm to operate without producing excessive harmonics. If the power out-

ity

put of the VCO had been greater than lOdBm, a 3dB power splitter could
not have been used because the LO port of the mixer would have become sat-
rs

urated and would then have produced harmonics. In this case, a directional
ve

coupler could be used to deliver optimum power to the LO port.

ni

The waveguide launcher used has been described in Section 3.3.

U

The low-frequency analogue section of the radar.

The analogue circuitry used to interface the microwave and digital sections
consists of amplifiers, gain sections and filters. All of these units were imple-
mented using operational amplifiers, and some accompanying passive com-
ponents. The details of each unit will not be given here, but the circuit
diagrams are shown in Appendix B.
The beat frequency produced by the microwave mixer will contain un-
wanted harmonics and both high-frequency and low-frequency noise which
must be filtered out. Because a sawtooth waveform instead of a triangu-
lar waveform is being used to modulate the microwave VCO, the definition
given in Equation 4.12 becomes:

43
 !' = !!_ (5.1)
Tm
The FMCW range equation is then given by:
R = Ug. fb
(5.2)
2·B·fm
Using Equation 5.2, it has been calculated that the beat frequency will
fall into the range 1. 7kHz to 6. 7kHz depending on the length of the electrode
(5m-20m). A 6th-order Chebyshev bandpass filter with a centre frequency
_, of 4.2kHz and a bandwidth of 5kHz was therefore placed at the output of the
mixer to attenuate unwanted harmonics a~d noise. Gain and offset sections
were plated after the filter to optimise the signal levels before analogue-
to-digital conversion. The input filter also prevents aliasing of the beat

n
frequency signal due to sampling.

w
The computer produces the modulation for the VCO through a DAC.

To
The output from the DAC is filtered to eradicate its stepped nature and
then amplified and level-shifted to make it compatible with the input of the
e
VCO. A simple low-pass filter in the feedback of the gain section provides
ap

the necessary smoothing.

C

\J
Interface between the radar and the computer
of

The computer performs the dual task of sampling the output from the mi-
ity

crowave section, whilst providing the modulation for the VCO. An interface
rs

between the radar and the computer to perform these tasks had to be de-
ve

signed and built.

Since a computer was to be used to sample the output from the radar,
ni

the beat frequency had to be kept fairly low. A lower beat frequency also
U

means a lower modulation frequency, and this has the advantages of sim-
plifying linearisation of the VCO, and of reducing capacitive effects on the
tuning port. The modulation frequency was chosen as 60Hz, which, as men-
tioned before, results in a beat frequency of 1. 7kHz to 6. 7kHz, depending
on electrode length.
In order to satisfy the Nyquist sampling theorem, the sampling rate
must be at least twice the maximum beat frequency, or at least 13.3kHz.
The sampling duration must be one period of the sawtooth modulation or
1/60Hz=16.7ms. The timing between samples must be extremely accurate,
·and must not be affected by system interrupts.
With reference to the microwave section of Figure 5.1: EM energy enters
port 1 of the circulator, and should be directed through to port 2. However,

44
 some of the energy 'leaks' through to port 3, and can be viewed as noise.
The isolation associated with a particular circulator is its ability to isolate
port 3 from port 1. Higher isolation results in less leakage and therefore
a lower system spurious noise-floor. The system's ability to measure small
reflections returning to port 2, or the system's dynamic range is therefore set
by the isolation provided by the circulator. Although the specified isolation
is 40dB, it has been measured to be more like 35dB. Since 7 bits provides
42dB dynamic range, a seven bit system would have provided the required
I

35dB dynamic range in this case; however, an eight bit system was chosen
because of the computing standard.
An interface card with these specifications was therefore designed and
built.
The steps involved in the operation of the interface card are:

n
w
• Calculate the modulation waveform using the computer;

To
• Apply a linearisation matrix to the modulating waveform;
e
• Load the modulation waveform into a memory on the interface card;
ap

• Give the instruction for the card to start sampling;

C

I
of

• The interface then samples and outputs the modulation waveform at

I
the same time;
ity

• The interface tells the computer that it has completed sampling;

rs
ve

• The computer loads the sampled values from a memory on the interface
into the computer;
ni

• The computer analyses the beat frequency spectrum and determines

U

the electrode length.

The interface has a high-speed DAC and ADC on board. There are two
RAMs which are used to store the modulating waveform and the sampled
data. Buffers isolate the computer from the card and therefore from the
radar. An oscillator provides a clock which determines the sampling fre-
quency. A counter, being incremented by the clock, logs data in and out
of the RAMs. A full circuit diagram and timing diagrams are given in Ap-
pendix C. The final specifications of the interface card are shown in Table
. 5.4.
The corresponding curves for tests performed on the DAC and ADC
sections of the interface are shown in Figure 5.2. An error analysis of the

45
 Sampling frequency 62.66kHz
Sampling period 16.3ms
Number of samples 1024
Number of bits 8

Table 5.4: Interface card specifications

-
"'
.
"'"'
250
-
~
2 .5

0 200 ;
; 0
.,
0 150 I .5

~
"'
~
.!'!' 100 0
c >

n
so 0 .5

w
1.31 2.5 50 100 150 200 250

To
Voltage In (V) Olgltal In (0-255)
e
Figure 5.2: Linearity graphs for the DAC and ADC.
ap

interface was undertaken, and the errors, which are defined by Horowitz and
C

Hill [10, page 614-615] are shown in Table 5.5.

of

These errors are comparable to those reported by the manufacturers of

ty

the ADC and DAC chips. The guaranteed linearity set by National for the
si

ADC (ADC0820) is 0.5 LSB. The final measured linearity of 0.43 bits is
r

therefore within the limits specified by the manufacturer. National specifies

ve

the maximum linearity associated with the DAC (DAC0832) as 0.05% of

ni

the reference voltage or 0.00125 volts. The measured value of 0.006 volts
U

compares well with this value. A source of extra error is that associated
with stability of the power supply used as a reference for these experiments.

Digital to Analogue Converter Analogue to Digital Converter

(volts) (bits)
Linearity error 0.006 0.43
Scale error 0.00 0.20
Offset error 0.00 1.92

Table 5.5: Interface card error specifications

46
 5.2 Signal Processing of the Radar Returns
Since the length of an electrode is proportional to the beat frequency, it
is of extreme importance that the beat frequency be determined as accu-
rately as possible. Traditionally, spectral estimation has been achieved using
fast Fourier transforms (FFT). Although the FFT is able to provide quick
and robust results, more modern techniques have been found which provide
better frequency resolution. It was therefore decided to investigate some
modern spectral-estimation techniques. Mathcad documents describing each
of the techniques can be found in Appendix D.
The following spectral-estimation techniques were compared:

• Fast Fourier Transform (FFT)

n
w
• Prony's method

• Multiple Signal Classification (MUSIC) method

To
e
5.2.1 The fast Fourier transform
ap

The FFT provides us with a computationally-efficient, reliable method of

C

spectral estimation. There are however some inherent limitations associ-

of

ated with the FFT, the most prominent of these being the limit on spectral
resolution i.e. the ability to distinguish two similar frequencies in a signal.
ity

,' According to the Rayleigh criterion, two signals should be considered to be

rs

'>just resolved when the first maximum from one component sits at the first
ve

minimum of the other (14, page 450]. Using the FFT, the obtainable reso-
ni

lution in Hz is approximately inversely proportional to the time in seconds

U

over which the signal was sampled.

(5.3)

Therefore, if an FFT is used to analyse the beat frequency spectrum, the

resolution with which the beat frequency can be determined will be given
by the inverse of the sampling time. Assuming the modulation frequency
error (EJm) and the bandwidth error (EB) in Equation 4.15 to be negligible
(this is reasonable, since with good hardware design they can be kept low),
we obtain:

fR = Efb • R (5.4)
fb
or, substituting Equation 5.2 for R:

47
 (5.5)

Then, by realising that Efb = J..

= fm, it can be shown using Equation
5.3 that the range resolution obtainable when using an FFT is given by :

u9
tr 2·B (5 -6 ) A
where Ug is the velocity of the electromagnetic energy in the waveguide and ~~
B is the bandwidth of the transmitted signal in Hz. The aim of this chapter ~~ ~ ~ 1
is to try to improve on the result obtained in Equation 5.6. J:JW..- ~
It is important at this point to introduce the signal processing terms bias ~ J./l '
and variance. The bias of a spectral estimator is the average error in the
estimator. Therefore, a low bias gives an accurate absolute spectral estimate,

n
w
as is required in this application. The variance of a spectral estimator is the

To
mean-squared error associated with the estimator. A smooth curve has a
low variance and a spiky curve a high variance. The higher the variance, the
easier it is to differentiate two closely-spaced frequencies i.e. higher variance
e
ap
corresponds to better resolution. It must however be remembered that it is
not resolution but absolute accuracy which is required in this application.
C

The first method of beat frequency estimation tested entailed squaring

-
of

the magnitude of an FFT to obtain a power spectral density. The spectral

component with the greatest amplitude was then taken as the beat fre-
ity

quency. Since the FFT is a discrete estimator, the beat frequency can only
rs

be determined to the nearest frequency bin in the estimator. Therefore,

ve

when using the FFT, the bias is limited to a minimum value of half the

- distance between frequency bins, or half the Rayleigh frequency. Averag-

ni

ing of the FFT spectrum decreases the bias slightly, but accuracy is rarely
U

improved beyond the limit imposed by the Rayleigh criterion.

For means of resolution comparison, it was seen whether the various
spectral-estimation techniques tested could resolve two sinusoids embedded
in white noise. The frequencies of the sinusoids were 2Hz and 2.2Hz respec-
tively, and the signal-to-noise ratio was 20dB. Figure 5.3 (a) shows that the
FFT periodogram is unable to resolve the two frequencies due to its low
variance.

5.2.2 Prony's method

The FFT assumes that a signal is composed of a set of harmonically-related
sinusoids and it is therefore a discrete estimator. Prony's method is a tech-

48
 nique for modelling data of equally-spaced samples using a linear combi-
nation of complex exponentials. Since the exponentials can have arbitary
amplitude, phase and frequency, the spectral estimate becomes continuous
and is no longer bound by the Rayleigh criterion.
The Prony method requires that the IJ,umber of exponentials to be used
in the model be known prior to spectral estimation. In cases where only
an approximate fit to the data is required, the number of exponentials can
be less than half the number of data points. In practice, however, far fewer
terms are used in the model since the rough computational complexity is
related to the squared power of the number of exponentials used in the
model.
A complex rooting algorithm is required to determine the frequencies
contained in the spectral estimate. This has the disadvantage of being com-

n
w
putationally intensive, but it does produce results without having to search

To
for maxima, as with the FFT. As can be seen from Figure 5.3 (b ), the
. Prony method is well able to resolve the two sinusoids of 2Hz and 2.2Hz
e
(see Appendix D for the Mathcad spreadsheet).
ap

5.2.3 The MUSIC method

C

The MUSIC (multiple signal classification) [14] algorithm is an eigenvector-

of

based projection approach to super-resolution spectral estimation. The basic

ty

idea is to partition the observation space spanned by the eigenvectors of a

i

correlation matrix into two subspaces denoting signal-plus-noise and noise

rs

respectively. It can then be shown [14, page 454) that, under ideal conditions,
ve

the complex frequencies contained in the input data can be found from the
ni

subspaces. Under realistic conditions, spectral peaks in an approximate

U

spectrum must be located.

The MUSIC method assumes prior knowledge of the number of complex
sinusoids contained in the input data. The MUSIC algorithm was able to
differentiate the two simulated sinusoids in Figure 5.3 ( c ).

5.2.4 Application to real data.

The three spectral-estimation techniques and some hybrids of them were
tested on some real data that was measured using the FMCW radar system.
Figure 5.4 shows a plot of a section of the radar return as measured using
the system in Figure 5.1. Ideally one would expect a pure sinusoid as output
from the radar. However, for reasons discussed in Section 4.3.5, this does not
occur in practice. The measured result is seen to resemble a rough sinusoid,

49
 FFT

2 3

iI
Frequency (kHz)

ID
:!:!.
., Prony
"~
CL
E
11 I.
<
!\---'11·----------------------------------------------------------··············
___________________________________________________ )

'Frequency (kHz)'
.,,ID MUSIC
""

n
:E
Ci.

w
E
<

To
' 3
Frequency (kHz)
e
Figure 5.3: Spectral plots of simulated sinusoids at 2Hz and 2.2Hz.
ap

as one would expect.

C

The three spectral-estimation techniques were applied to sets of sampled

of

data taken at different electrode lengths. As can be seen from Figure 5.5,
ity

the three different methods worked with varying success.

Zero padding was applied to the FFT in order to smooth the result and
rs

to provide interpolation between the spectral bins. The figure shows that
ve

the FFT provides the most accurate absolute measurement of frequency,

ni

and therefore of length. Although the other spectral-estimation techniques

U

provide better variance than the FFT, they tend to add a varying bias to the
absolute measurement depending on the number of terms used in the model.
This result agrees with [3, page 89] who mentions a trade-off between bias
and vari.ance. Table 5.6 gives a summary of the different techniques [15,
page 1410] where N is the number of data samples and Mis the order of the
model used in the Prony and Music methods.
Therefore, on the grounds of its computational efficiency and robust-
ness, the FFT was chosen as the spectral-estimation technique to be used to
determine the beat frequency from the sampled FMCW radar data signal .
•
As mentioned previously, zero padding was used to interpolate between val-
ues. A weighted average of a few spectral lines around the maximum also
provided a simple means of further interpolation between spectral bins.

50
 ...... 180
...
e' 160
.,,•
..
:I
:I
140

Q, 120
E

"' 100

80 0 50 100 150 200 250 300

S a m p le n u m b er

n
w
Figure 5.4: Section of the sampled output from the FMCW radar.

7
To
e
ap

6.8
C
of

E 6.6
.......
..
ity

.J:.
Cl
c
rs

..! 6 .4
..
i::I
ve

GI

.!!
::I
..!:!
ni

ta 6.2
0
U

5.8

5.6
0 20 40 60 80 100
Known relative length (cm)

Prony M uslc FFT FFT+Ave

----e-- ----- ............. ..

Figure 5.5: Length estimates as obtained using the three spectral-estimation

techniques.

51
 Technique Discrete or Complexity Advantages and disadvantages
Continuous
Fourier D Nlog2N Output directly proportional to power
transform Most computationally efficient
Resolution roughly the inverse of sampled
time
Prony's c M 2 +MN Requires prior knowledge of order
method Requires polynomial rooting
Resolution as good as autoregressive (AR)
techniques
Resolution exceeds that obtainable with FFT
Requires high /J for low bias
MUSIC c M"l.+MN Requires prior knowledge of order

n
method Resolution exceeds that obtainable with FFT
Requires high f;. for low bias

w
Sensitive to changes in spectral model

To
0

e
Table 5.6: Summary of the different spectral-estimation techniques.
ap

The signal processing techniques as well as the data capture, the VCO i"
C

linearisation and the interface card control procedures, were written in Bor-
of

land C. The program, called RADAR, is listed in Appendix E. nU

'\ ~~l VJ(J~JJI.
ty

..;- \]t.tJ!.-

~().A)e ~
si

\,l,cfl,~~~ .
v er
ni
U

52

•
 Chapter 6.

RESULTS

n
The FMCW-radar-based measurement system was set up as described in

w
Chapter 5. A stainless steel pipe was used to represent the electrode, and a

To
reflective plunger was inserted into the pipe to represent changes in electrode
length. e
The measurements were made using the HP3050 sweep oscillator as a
ap
frequency source, being swept by the FMCW radar to PC interface card.
The results were logged by the interface card, interpreted by the PC and
C

displayed as a length measurement on the PC display. Measurements were

of

made over the approximate range 4.5m to 6.9m at one-centimetre intervals.

M first, only a simple FFT was used to determine the largest frequency
ty

component, and therefore the beat frequency could only fall into a discrete
i
rs

frequency bin. For this reason, the length measurements were stepped and 0 1 L
(71AR. ~"""~re r
\t- 1" ~ ~4efr.J ~
ve

not continuous as 2-ne might expect. Figure 6.1 shows a graph of measured
length versus the known length. Although the stepped nature of the graph
ni

causes a large overall error, a linear regression showed that the gradient
U

of the graph was almost ideal, having a slope of 1.001:1. Since the trend
followed by the measurements was so accurate, but the overall accuracy was
being limited by the discrete nature of the FFT (RMS error~ 7cm), it was
decided to do some signal processing on the measured data. It should be
noted that, for absolute measurements, the distance between the microwave
components must be subtracted from the measured distance, but that for
relative measurements this has no effect.
As mentioned in section 5.2, it was found that zero padding in con-
junction with the FFT was the most successful signal processing technique.
) Weighted averaging was also found to improve the accuracy with which
\ the beat frequency could be .measured. This process was applied to the
computer-based interpretation program, and the measurements were re-

53
 -E

-
J:. 6.5

n
Cl
c

w
.! 6
,,
To
...
G)

; 5.5
ca
e
G)
ap
:!: 5
C

50 100 150
of

Known relative length (cm)

ity
rs

Figure 6.1: Length measurement with simple FFT signal processing tech-
ve

nique.
ni

peated. Figure 6.2 shows the repeated set of measurements. From the
U

figure, it can be seen that the stepped nature of the measurements has been
vastly reduced, but that accuracy has been maintained. When compared
with the known length, it was found that the RMS error associated with the
measurements was 1.37cm - well within the required limit.
Further measurements were then performed between eight and nine me-
tres to determine whether the system was linear over large ranges. In order
to perform these measurements, an extra section of pipe had to be added to
the existing length. The discontinuity formed at the junction between the
two pipes was found to have a very small effect. Alignment of the pipes was
not critical to the performance of the system. The extended set of measure-
ments lined up perfectly with the initial set (Figure 6.3), and proved that
the system remains linear over extended ranges.

54
 7

-·
E
..... 6.5
z:
.,,
~

c
GI
6
"O
...
GI
::::s
UI
Cl 5.5
GI
E

n
w
4.5

To
0 50 100 150
known relative length (cm)
e
ap

Figure 6.2: Length measurements with zero padding and a weighted average.
C
of

Finally, a 40cm thick section of Soderberg electrode was attached to the

end of the pipe. The electrode had been pre-bored with an inner diameter
ty

equal to that of the pipe. Due to the crystalline nature of the electrode, the
i
rs

inner surface was rough, just as would be expected in a real application in

ve

a furnace. It was found that Sl,irprisingly little attenuation and virtually no

wall reflections occurred in the section of electrode. The length-measurement
ni

system performed as well as in previous experiments, with the electrode

U

section having no adverse effect on the measurements (Figure 6.4).

55
--
E 10

..
.i::.
en
c
9

CD 8
'O
...
CD
7
::::s
en
CG 6
CD
:E
5

n
4

w
0 100 200 300 400

To
Known relative length (cm)
e
ap
C

Figure 6.3: Extended length measurements.

of

E
__. 6 .8

-
ty

.i::.
i

en ··1
rs

c 6.6
'··~.
CD
ve
ni
U

20 40 60 80
Known relative length (cm)

P ip e S ode rb erg

Figure 6.4: Length measurements though a section of Soderberg electrode.

56

/
Chapter,7

CONCLUSIONS

n
Based on the findings of this report, the following conclusions can be drawn.:

w
• A microwave system in which a radar placed above the electrode is

To
used to determine the electrode length has been identified as the most
promising electrode-length-measurement technique.
e
ap
• Both practical aiid theoretical evaluations have show that microwaves
will be able to continue propagating in the cavity left in the electrode
C

once the inserted waveguide has melted.

of

• A conventional frequency-modulated continuous wave radar provides

ity

the best overall performance in the furnace environment due to its abil-
ity to differentiate between targets, and the 'intelligence' with which
rs

it can be interfaced.
ve

• The linearity of the microwave oscillator was found to be of critical

ni

importance in obtaining a high signal-to-noise ratio at the radar out-

U

put.

• The fast Fourier transform provided both the resolution and robustness
needed in the signal analysis of radar returns.

• The length of a hollow waveguide could be measured with an accuracy

of 1.38cm over the range 4m to 9m using the system.

• The termination at the end of the pipe was found to be non-critical,

with the system working well whether the pipe was terminated or not.

• The addition of a section of Soderberg electrode at the end of the steel

tube did not affect the accuracy of the measurements. This shows
firstly that the electromagnetic waves can propagate in a cavity in

57
 the graphite electrode, and secondly that the reflection caused by the
transition between the waveguide and the graphite cavity is small in
comparison with the reflection caused at the end of the electrode.

The system now needs to be tested on a real furnace in order to evaluate

the effects of the furnace environment more accurately.

n
w
To
e
ap
C
of
ity
rs
ve
ni
U

58
Bibliography

[1] Ultrasonic wall-thickness measuring instrumentt Materials Evaluation,

cu
47:140, February 1989.

[2] A. B. Stewart. The Measurement of Electrical Variables in a

n
Submerged-Arc Furnace. Technical report, National Institute for Tech-

w
nology, April 1981.

To
[3] A. Bas. An Ultra High Resolution FMCW Radar. MSc Thesis, Univer-
e
sity of Cape Town, 1992.
ap

[4] A. E. Karbowiak. Trunk Waveguide Communication. Chapman and

C

Hall LTD, 1965.

of

[5] A. W. D. Jongens. Personal communication. March 1994.

ity

[6] C. W. Soderberg. Soderberg Self-Baking Continuous Electrodes. Chem-

ical and Metallurgical Engineering, 26(25):1178-1182, June 1922.
rs
ve

[7] David K. Cheng. Field and Wave Electromagnetics. Addison-Wesley

Publishing Company, 1989.
ni

[8] John A. Dorr. Ultrasonic width measurement of continuously cast slabs.

U

Iron and Steel Engineer, 69:33-36, October 1992.

[9] Merrill Skolnik. Introduction to Radar Systems. McGraw-Hill Interna-

tional editions, 1981.

[10] P. Horowitz and W. Hill. The Art of Electronics. Cambridge University

Press, 1987.

[11] R. A. Waldron. Theory of Guided Electromagnetic Waves. Van Nos-

trand Reinhold Company, 1969.

[12] R. Innvaer, L. Olsen, A. Vatland. Operational Parameters for Soder-

berg Electrodes from Calculations, Measurements, and Plant experi-
ence. Technical report, Elkem R&D Center, Norway.

59
[13] S. A. Hovanessian. Radar System Design and Analysis. Artech House,
1984.

[14] Simon Haykin. Adaptive Filter Theory. Prentice Hall, 1991.

[15] Steven M. Kay and Stanley L. Marple. Spectrum Analysis-A mod-

ern Perspective. Proceedings of the IEEE, 69(11):1380-1419, November
1981.

[16] Watkins-Johnson Company. RF and Microwave Components Design-

ers' Handbook. Watkins-Johnson, 1990. l'

n
w
To
e
ap
C
of
ity
rs
ve
ni
U

60
Appendix A

Waveguide launcher design

__.3mm
-
~

-
- 20mm ~
18.?mm - ~
15mm ...
l
I I

I .
I

n
41 .2mm

w
22.Bmm 22.5mm 25.3mm

.
,
To 1•
.
e
ap

l
C

TOP VIEW
of

-
ity

3mm
rs

--.. "'4--
ve

18.?mm 15mm
ni
U

20mm

41.2mm 10.2mm 22.5mm 25.3mm

SIDE VIEW

61
 Appendix B

Low frequency circuit

diagrams

A 8 c D

n
+15V

w
INPUT
Jl Rl

To
lk
c: LM32 J
;::10 LPF

CNO
Ul:A
e
"1
2 ~i:~ 2
ap
CND

-15V
,....
C

HPF

"' +15V

16 CB w
R7
of

LPF
LM3 4J
100n 100n 100n HPF

CS> IX)

...
IX)
,..., Ul:C
ity

a:: cw
3 a:: 3
CD
rs

0
ve
ni
U

4 4

A 8 c D

62
 A 8 c D

U4

•
... ......
m - u
g rs;
...
~
:::>
...
a
a ...... >
~
I ~~~~~~~~~ (t) '"1 ~
~ 7~~~mmmmm n (1)
'"1
Pl §I SN74HC6BBN
•'l.....2Lvcccte1~ DAC eur ~ ......
e+-
= ~
.....
U5 n n
2 2 (t) Cl,. ><
LSB •~LSB
ow n
...... 0
T ~
~ C1Q
SN74HC4040N
:1&.ll-vcc ...,.....L....!i!!I
e '"1 '"1
p Cl,. ~
0) Q
~ ,;; ADC-RAM OAC-RAM
:::>
a
U13 C U11
8
3
3 of 0
Ii-+)
ity e+-
74HC157
,.,,, --ll-wx:CHDJ..- CND
rs ~
e (t)
i v
n OAC0B32

4 TC5563AL-15PN .~63AL-15PN 4
Mil "'!11..--li- >CC A-lo-0 and i>-lo-A
u~ CHO
U
18M62 ~·
SN74HC4040N
+,_,_11....YCC CNO_a__GND
27""'''"
D
-
t ..:>
A B c
 Appendix D

Signal processing techniques

Sampling frequency: f s := 10

Number of time samples : .N := 64 n := 0 .. N - 1

Frequencies to be resolved: f1 := 2 f2 := 2.2

Creating the time

wave1 n := sin 2·7t·f f1
s · ") + .1·rnd(1 ) - .05 + sin ( 2· 7t·f f2·
s n)
funtion with added (
noise

L:wave1

n
tot1 .- wave1n := wave1 0 - tot1

w
N+1

To
e
ap

wave1no
C
of

0 10 20 30 40 50 60
ity

n
rs
ve

Apply the MUSIC method to the time domain function

ni
U

Filter length M := 5 (0-4)

x := 0 .. N - M - 1 y := O.. M spy := Y

A1 h := wave1(M _ Y >
(y,x) +x

A1 h := wave1( ) r
(y,x+N-M) x+y

T 1 .
A1 = A1 h R1 := .,1
2·(N - M)

64
 ei1 := sort( eigenvals( R1 ) ) v11 := eigenvec(R1, ei1 0)

V1 n := v11 __ T
(y,O) y V1 nh := V1 n
_ _T
S 1 ( ro ) := e <- j · ro· sp > S 1 h ( ro ) := S 1 ( ro )

1
1
5 music< f) := ( )
DEN1 2·1t·f
fs

Spectral plot of the time domain function using the MUSIC method

n
w
spec := 0.01, .05 .. 5

1·1~6
5 To
e
1·10
1·104
ap

1000
S1 music(spec)
C

100
of

10
1
ity

0.1
0 1 2 3 4 5
rs

spec
ve

------------------------------------'
ni

PRONY'S METHOD
U

Number of filter coefficients P := 15

Apply Prony's method to the time domain function

s := 0 .. N - P - 1 p := 0 .. p - 1
/'

pp:= 0 .. p popp := P - PP

b<5 > := - ( wave 1 ) <5 + P >

A(s, p) := wave\ 5 + ( P _ 1 _ p))

1
T )- T
a:= ( A ·A ·A ·b

65
 ~

0
zs(z) := zP poly(z) := zs(z)·c

Solving the spectral estimation function to obtain the frequency content :

2·n·f2·j
2·n·f1 ·j
fs - fs
q1 := e q2 := e

z1 := root( poly( q1 ) , q1 ) z2 := root(poly(q2).q2)

z1 = 0.3094 + 0.9509j z2 = 0.1871 +0.9817j

atan (-lm_(_z_1_) ) atan( Im( z2))

f1 := Re(z1) ·f s f2 := Re(z2) ·fs

n
2·7t 2·7t

w
To
Frequencies f1 =1.9993 f2 = 2.2002 e
Iz1 I = 0.9999 Iz2 I = 0.9994
ap
Magnitude of the spectral lines

_______
C

--------------·--- ---·--- ,_,..

of

FAST FOURIER TRANSFORM

ity

Apply the FFT to the time domain function :

rs
ve

j := 0 .. N
2
ni
U

fff := Ifft( wave1 ) I

fff;
_J 2

o1-====:±::::===--1-=~='- ............L..~---1
0 1 2 3 4 5

66
 #Include <lo.h> lstep=nmax << 1·
#include <stdio.h> '( theta=isign*(6.~83185307179591nmax);
I Include <stdl ib.h> oldhandler = getvect(INTR); II Save the old interrupt wterrp=s i n(0.5*theta);
llnclude <dos.h> => vector. wpr = ·2.0*wterrp*wterrp;
llnclude <conio.h> setvect(INTR, handler>; II Install the new Interrupt wpi =sin( theta);
#Include <math.h> => vector. wr=1.0;
#Include <gra{lllics.h> Original INT mask = l~rtb(Ox21); II Set up 8259 PIC wi=O.O;
llnclude <str1ng.h> => chip-for-interrupt 7. for <m=1;mcnmax;m+=2> c ~ ~
outportb(Ox21, Original INT mask & Ox7f >; for <l=m\i~=n;l+=lstep> c >
#define INTR OxOf II ADCDA card Interrupts on 7. 1=1+nmax; ~
ldef Ine CPPARGS ••• while·(flag == 0) II Waiting for an Initial interu terrpr=wr*data(jJ·wi*dat t-1
#define 'SQAP(a,b) terrpr=<a>;<a>=<b>;<b>=teq:>r II Fin:tl => pt for synchronisation. · => a(j+1J;
=> on to swap a and b. (
inportb(Ox300J; II Read trash data and Increment t => ata(j); :' ·
terrpi =wr*data (j+1J +wi*d
> ~
C'D
=> he interface counter. data q J =data Ci J · t~r; t1 ~
II················································· } data[J+1J =data[i+1J ·tem
II GLOBAL VARIABLE DECLARATIONS I> => pi;
II················································· data[iJ += teq:>r; ~
n data(i+11 += teq:>i; >~ = .....
int mldx midy; II To store screen coords. v1················································· w >
const no_elements = 4096; II No of time sa~les. wr=(wteirpswr)*wpr·wj*wpl+wr; ><
const no_aves = 20; II The ~r of elements In wl=wl*wpr+wterrp*wpl+wl; ~ s
=> the FFT to average.
o } •
volatile int flag= Ol· II Indicates that an lnterrup old close_lnt() T nmax=lstep; ttj
=> t has occured (f ag=1). }
}
float address first c O; II Array address of maxlna111 e I Resets orlgional values to the Interrupt
=> lement. - I controller before exittlng the program.
e 0 ::::-:
float ave freq = 0; II Average beat frequency. t-0 ~
Int DAC DitaC1023J;' II Dota used to linearise the 11·························-~·-····················
=> oscillator. outportb(Ox21, Original INT mask & Oxff); II Disable
p ~ t-1•
int ADC Data[1023l; II Data read In from the ADC, => the Interrupts - -
0)
=> FRCW radar Info. setvect(INTR, oldhandler>; II Restore the orlglonal I
a void DAC_Flle_load()
~
float ave_array[no_avesJ; II T~rary storage to perfo => nterrupts as stored. C
=> rm averaging. II loads data for linearising the RF oscillator from crq
float transformed((no elementa*2)·1J; II Used to store f II a file called IN DAC.DAT. The data is In an ASCII
II file of floats In a list.
=> freq data after fFT.
void Interrupt < *oldhandler>< CPPARGS);
VI·················································o (
=
U1Slgned Int Orlglnal_INT_mask'f" FILE *fp;
0
!void four1(float data[), unsigned long nn, Int !sign) fl oat terrp; ~
I Does a fast fourier transform on the s~led time
II····························-·····-·············
ity
I domain data. This procedure comes from the C
s fp=fopen<"IN_DAC.DAT", 11 r 11 ) ; / / points to open file.
I Recipes collection of algorithms. for(int l=O; I< 1024; I++)
(
void interrupt handler(_CPPARGS) ~ r
IC fscanf(fp, 11 l'1 11 ,&teq:>); II load a single
II This procedure. Is called each time an interrupt unsigned long n,nmax,m j,lstep,I; => time s~le.
II occurs. The variable flag Is triggered to double wteq:>,wr,wpr,wp1,wi,theta; DAC_Data(IJ=t~; II Store the data val
II Indicate that an Interrupt has been received.
i ve
float teq:>r, teq:>i; •> ue In an array.
n }
c n=nn << 1; fclose( fp);
dlsable<>i II Disable all Interrupts. j=1 • }
outportb(ux20, Ox67); for'<i=1l•i<ni1+=2> C
flag=1; II Set interrupt flag to affirmative.
U f (J > i) c
delay(10); SWAP(dataCtl data(il); II·················································
enable(); II Re·enable the Interrupts. SWAP(data[J+h ,data[i+1J >;
} }
m=n » 1;
while (m >= 2 && j > m> C
II················································· j ·= m·
m »= ;;
>
j +•
 forCint i=O; i< 2*no elements; i++) => 8 elements are tot=totlno aves;
< - < II zero padded gcvt(tot*3:3338,3,range); II Convert from freq to ra
fprintf(fp,"Xf\n", transformed Ci]>; => to provide => nge, and characterise.
> transformed[iJ=O; II interpolatio strcat(range,meter); II Add the word "meters" t
fclose(fp>; II Store frequency data. => n of data once => o the measurement.
> > II FFT has been cleardeviceO;
=> done (ZERO PADDING). outtextxyCmidx,midy,range>; II Display the range in
=> the mid of screen.
11------------------------------------------------- for Cf = O; i <1024; i++) delay(100);
< transformedCt~J = ADC DataCiJ; II Fill in zero
>
void load_DAC() => imaginary values -
tempt-+; II for the time 111-------------------------------------------------
II Loads the linearising function into the DAC RAM. => domain function
II The VCO is then linearised with this function. transformed[templ=O; ll as required
=> by the FFT procedure. void main()
< temp++; < clrscro;
flag = O;
> init int(); II enable and initiate interrupt 7.
for (int i = O; i<1024; i++) four1(transformed-1,no elements,1); II Perform the Four DAC file Load(); II Load the linearisation data.
=> ier transform. - loaa DAC(); II Send the modulation to the DAC RAM on
< inportb(OX300); II Read from the interface is done
n => interface card.
=> to increment counter. ~etlJ)=O;
outportb(Ox300,DAC Data[il>; II Load function int ddress first=O;
w clrscr();
=> o RAMS on interface. -
printf("Data out = Xd\n11 ,DAC_Data[i] >;
o
for < i = O; i <(no_elementsl2>; i++) II Convert from
T II CHANGING TO GRAPHICS MODE
> => Re, Im to ABS co-ords.
> < int gdriver =DETECT, gmode, errorcode; II Request aut
transformedCil = sqrt(powCtransformed[t~J,2)+po => o detection
=> w(transformed[temp+1J,2>>; pe int size = 6;
if (CtransformedCiJ > big)&(i>50)) II Ignore firs initgraph(&gdriver, &gmode, "">; II Initialise graphic
11------------------------------------------------- => t 50 due to large DC a => s.
0) 11 component i errorcode = graphresult(); fl Read result of initial is
00
<
=> n FFT. C => ation. ·
void Read_RAMO big=transformedCiJ; II Choose the largest ele if (errorcode I= grOk) II An error occurred, so displa
=> ment from the spectrum. => ythe type of error.
II Reads s~led time domain values from ADC RAM.
f
address_first=i; II Store the address of the <
=> largest element.
o printf("Graphics error: Xs\n", grapherrormsg(errorco
< > => de));
tetl1)++; printf("Press any key to halt:">;
flag= O; II Reset interrupt indicator to negative. temp++; ity getchO;
> exit(1); II Terminate with an error code.
for (int i = O; i <1024; i++)
ddress
s
+= transformed[address first-1J*(address first-
r >
< ADC Data[i] = inportb(Ox300); => 1); e - - midx = getmaxx<> I 2; fl Get middle screen co-ords.
> II lead a value from the ADC RAM. ddress += transformed[address first]*address first; midy = getmaxy() I 2;
> ddress += transformed[address-first+1]*(addriss first+
I => 1); - - settextjustify(CENTER TEXT, CENTER TEXT); II Set text
,total = transformed[address first-1l+transformed[addres => attributes. - -
~ niv
=> s_first]+transformed[aCJdress_first+1l; settextstyle(DEFAULT_FONT, HORIZ_DIR, size);
11-------------------------------------------------
U address=address/total·
VI A weighted ave of largest 3 elements. while (lkbhit()) II Run until any key is pressed.
c
void Calc_FreqO ~ve freq=31.33*addressl(0.5*no elements>; II Convert th if (flag == 1) II If an interrupt has occured, t
=>·e FFT bin no to a freq. - => hen the ADC RAM is
II Calculates the dominant frequency contained in tot = ave freq; CRead_RAMC>; II full (Sampling.has been comple
II the fourier transformed data. Displays length. 1ave_arraylno_aves-1] =tot; => ted). Read the
Calc_Freq<>; II contents of the RAM and calcul
< for Ci = O; i < ~no_aves-1>; i++) II Average no_aves ti => ate the range.
=> mes to obtain >
int tefll>=O· C II more accurate beat >
float big=b,total=O,address=O,tot=O; => freq. ADC File Save(); II Save the last FFT data (just for
char range[15J, *meter= 11 meters."; ave_arrayCil=ave_arrayCi+1l; => -later inspection). ,-
tot=tot+ave_array[1J; close_ int(); II This· is useful for checking inte ~
> => rface performance. ~
for Ci~~-~-=2048; i<2*no_elements; i++) II The last 204 > . ~