Chapter 1: Introduction

1.1 INDUCTIVE PROXIMITY AND DISTANCE SENSORS

Inductive sensors have been developed commercially as proximity and

distance sensors for years. Based on Faraday’s law of electromagnetic induction, V

= -N(dφ/dt), these sensors transfer the magnetic information related to the

proximity or distance between the sensor and a target to an electrical signal.

Faraday’s law implies that the varying magnetic flux in a coil with N turns will

generate a voltage V at the two ends of the coil. Thus, the basic component of

inductive sensors is a coil which generates a magnetic field in front of the sensor

and monitors the magnetic field variation when a ferromagnetic or conducting

target approaches [1].

For a ferromagnetic target, the reluctance of the magnetic path consisting

of the coil and the target decreases and hence the inductance of the coil increases

when the distance between the sensor and the target decreases. The magnetic field

of a coil sensing a ferromagnetic target is illustrated in Figure 1.1(a). For a

conducting target, the magnetic field emerging from the coil induces an eddy

current in the approaching target surface. The eddy current creates an opposing

magnetic field which is absorbed by the coil according to Ampere’s law. Figure

1.1(b) shows the magnetic field of an eddy current sensor. This phenomenon

results in a decrease of the inductance of the coil. Therefore, we can measure the

inductance of the coil to tell the distance between the coil and the target. The eddy

current inductive sensors are more widely applied in industry since the magnetic

1
field of these sensors penetrates the target only to the level of the orders of skin
depth δ, which is approximately equal to 1 / πfσµ , where µ and σ are the

permeability and conductivity of the target, respectively, and f is the frequency of

the electromagnetic field. As a result, the sensing capability will not be restricted

by the thickness of the target if high frequencies are applied.

i i
N N

S S

Ferromagnetic target plane Conductiv e target plane

(a) (b)

Figure 1.1 Magnetic field of a inductive sensor sensing (a) a ferromagnetic and (b)
a conductive target.

1.2 INTERFACE CIRCUITRY OF THE INDUCTIVE SENSORS

In addition to the coil components, interface electronic circuits play an

important role in extracting the inductance variation with regard to changes in the

distance or proximity of the target. The sensitivity of the sensing system can be

highly restricted by the interface circuitry. Bridge circuits and oscillators are the

two most popular interface circuits for inductive proximity sensors [2].

Wheatstone bridge circuits are used due to their high linearity and

temperature compensation. Figure 1.2 is an example of this approach. Lx is the

2
inductance of the sensing coil, and it keeps its balance value of L1L2/L4 to generate

a zero output voltage until the presence of a nearby conducting target decreases its

inductance to 1-ε. The unbalanced bridge in turn provides an a.c. output voltage

the amplitude of which is proportional to the change of the coil inductance ε.

The oscillators are more widely used in industry because of their simplicity

[3][4]. A conventional inductive proximity sensor with an oscillator contains three

parts: an oscillator circuit, an air-around or ferrite-core inductive coil, and a

detector circuit. The oscillator circuit can be either a tuned or relaxation oscillator

consisting of a LC tank network in the feedback loop. The inductance and

capacitance of the LC circuitry determine the operation frequency of the oscillator.

The inductive coil forms the inductance portion of the LC circuit and carries a

current, the frequency of which is determined by the oscillator. When an

approaching conducting target causes a change in the magnetic field around the

sensor, the impedance of the coil is decreased, and therefore either or both the

frequency and the amplitude of the oscillator signal change. The change of

frequency or amplitude is converted to readable output information through the

detector circuit which can be either a Schimitt trigger level detector that generates

a digital output and provides target presence/absence information or an analog

readout circuit that provides more accurate distance information between the

sensor and the metal target. Recently, a differential relaxation oscillator has been

proposed [5]. It has the advantages of both the bridge and oscillator circuits and

also leads to a high sensitivity.

3
 L1L2
Lx = (1-ε)
L4
L1

Lx
AC voltage +
source VAC 2 VACL1 L4 ε

L4
L Vout ~ 2
(L1 + L 2)
-

Figure 1.2 A bridge interface circuit with Lx as the inductance of the sensing coil.

1.3 INTEGRATED INDUCTIVE SENSOR SYSTEM

As silicon technologies are developed rapidly nowadays, the compatibility

of proximity sensors with microelectronics circuits becomes an important issue.

The monolithic integration of the sensing element together with the interface

circuit on the same chip is desired to achieve a comparable price/performance ratio

with the standard silicon circuits. A miniaturized flat-coil inductor has been proven

to be a simple solution in several papers [6][7]. By applying standard process

techniques such as CMOS metalization, a coil can be fabricated with integrated

interface circuits. When the coil is scaled down, however, the inductance

decreases, and as a result, a much higher frequency about 1/ m2 of the original

operation frequency is required to maintain the sensitivity, where m is the scaling

factor [8]. Moreover, the downscaling causes a dramatic increase of the serial

resistance Rs of the coil due to Rs = ρl/A, where ρ is the resistivity of the

conductor making up the coil, and l and A are respectively the length and cross

section of the coil. From the interface circuits’ point of view, the serial resistance

will behave as a permanent damping element in the LC resonant circuit and

4
degrade the linearity of the bridge interface circuit because the sensing coil loses its

characteristics of a pure inductor with high Q factor.

1.4 TWO-COIL INDUCTIVE PROXIMITY SENSOR DESIGN

A two-coil eddy current proximity sensor has been proposed to avoid the

drawbacks of scaling down the sensor system [9]. The two coils build a flat

transformer structure in which one coil carries the periodic excitation signals to

generate the magnetic field and the other acts as a pick-up coil detecting the

magnetic field variation. The separation between the sensor and a metal target is

reflected in the phase shift between the electrical signals in the two coils. From the

a.c. simulation result of this sensing element model, it is found that the sensitivity

of this design is not sensitive to variation in the coils’ series resistance. Therefore,

the high serial resistance due to scaling down is no longer a concern.

In the conventional proximity sensor system design, there is only one

reading, and ideally this reading is a function of only one parameter (distance or

proximity). However, in the real world, several components are attributed to the

variation of the output reading, and therefore the sensing performance can be

degraded dramatically if only one parameter is considered as the variable. To

improve the sensitivity of the reading and to get more knowledge of the

environment which we want to sense, several studies [1][2] suggest making use of

the relationships of the many output variables rather than relying solely on the ideal

one-output function of one parameter. In the two-coil sensor design, for a fixed

distance between the sensor and a target plate of a certain material, there is an

operating frequency at which the sensor achieves its maximum sensitivity to target

5
proximity. This frequency is found to be related to the conductivity of the target

material. Therefore, the conductivity of the target and the frequency for the

maximum sensitivity can be regarded as the second input and output of this sensor

system, respectively.

The goal of this work is to design an integrated interface circuitry which is

capable of meeting the following requirements,

1. extracting both of the two outputs, the operation frequency for the

maximum sensitivity f and the phase shift at f with high resolution, in

order to get accurate information about both the proximity and the

conductivity of the target,

2. keeping the sensor’s characteristic independent of the coils’ series

resistance, and

3. being able to be fabricated monolithically with the two-coil sensor on

the same chip by the standard silicon process.

1.5 SUMMARY OF CHAPTERS

Chapter two describes the procedure of drawing the two-coil sensor’s

SPICE model. The phase response to the frequency of the sensor model as well as

the requirement of the input and output stages are also covered.

Chapter three explains how to design the electronic interface circuitry. Two

feedback controlling schemes based on a loop consisting of a VCO (voltage

controlled oscillator) and a FVC (frequency-to-voltage converter) are elaborated.

The performance of each circuit component involved is evaluated by the SPICE

simulation.

6
 Chapter four quantifies the characteristics of the sensor and the interface

circuitry in terms of sensitivity, resolution and accuracy. The factors causing

unsatisfactory results are discussed.

Finally, chapter five summarizes the thesis and addresses additional work to

be done in the future.

7
 Chapter 2: Two-Coil Inductive Sensing Element

One-coil inductive sensors have been shown to have serious drawbacks

that are caused by the increase in the coil resistance when the coil is scaled down.

The two-coil sensor design is a promising alternative.

The basic components of the two-coil inductive proximity sensing system

are an exciting coil, a pick-up coil, and a target plate. In this chapter, we will

discuss how to build a simple SPICE model out of these three components, and in

the following chapters, this model will be used to design and simulate the interface

electrical circuit.

2.1 EXTRACTION OF THE ELECTRIC PARAMETERS OF A COIL

In this section, only planar rectangular spiral coils are considered. This is

because the planar structure can be easily fabricated by the standard CMOS

process and because the rectangular spiral coil has been shown to achieve the

highest inductance and sensitivity to proximity compared to other geometric coils

with the same area [9].

2.1.1 Coil Resistance Calculation

The formula for the resistance of a straight homogeneous material with a

resistivity ρ, length l, and uniform cross section A for d.c. current is

l
R= ρ . (2-1)
A

The total coil resistance is calculated by summing the resistance of each segment

making up the coil. For a high frequency current, however, Eq. (2-1) will no

longer be valid due to the skin depth effect by which the current traveling in the

8
conductor is confined in a very thin layer, the thickness of which is approximately

proportional to the inverse of the square root of the frequency. In this sensor

model, the operating frequency range will be shown to be within the range low

enough for designers to keep the d.c. formula. Also, the skin depth phenomena has

little effect on planar thin-film coils.

2.1.2 Coil Inductance Calculation

The inductance calculation of the rectangular spiral coil is based on

Greenhouse’s method [10]. The self inductance Li of a straight conductor segment

i with length li in microns, thickness t in microns , width w in microns, and relative

permeability µ, is
  2 li  AMD µ 
Li = C li ln   − 125
. + + T , (2-2)
  GMD 
 li 4 
where C is 0.0002 nH µm − 1 and T is the frequency correction factor ( T = 1 for

microwave frequencies). GMD and AMD are the geometric mean distance and

arithmetic mean distance of segment i, respectively, and they can be found as

follows,
 
 
GMD  1 1 1 1 
ln( )= - + 4 + 6 + 8 + ........ (2-3)
d   d 2 d  d  d  
12  60  168  360  
 w  w  w  w  

AMD = w + t, (2-4)

where d is the distance between conductor filaments of the coil. The mutual

inductance between segments i and another segment with the same length a

distance d from segment i is

9
Mi = C •li •Qi , (2-5)

where Qi is the mutual-inductance parameter of segment i , obtained from the

equation

 li   li 2  
0.5 2 0.5
   GMD   GMD
Qi = ln  + 1 +     − 1 +    + . (2-6)
 GMD   
GMD     li   li
 

It can be shown that the mutual inductance between two segments is inversely

proportional to the distance between them.

segment j

d
segment m

P m q

Figure 2.1 Calculation of the mutual inductance between two coil segments with
different lengths [10].

So far, it is supposed that each element of the coil has the same length,

however, this statement is not valid for most cases. For two segments with

different lengths j and m as shown in Figure 2.1, the mutual inductance, Mj,m,

between them is given by the relationship [10]

2Mj,m = ( Mm+p + Mm+q ) - ( Mp + Mq ), (2-7)

10
where each M term on the right hand side is calculated by Eq.(2-5).

The general expression of the total inductance of a coil is

Ltotal = ΣLi + ΣMj - ΣMk , (2-8)

where the first term is the sum of the self inductance of each segment in the coil,

and the second and third terms represent the sums of all the positive and negative

mutual inductance of each parallel pair of segments, respectively. The mutual

inductance is positive when the current flow in two parallel segments is in the same

direction or negative when the current flow is in opposite directions. Take a simple

rectangular spiral coil shown in Figure 2.2 with a current I flowing in the direction

of the arrows for example, the total inductance of the coil is

Ltotal = L1 + L2 + L3 + L4 + L5 - 2 (M1,3 + M2,4 + M3,5) + 2M1,5 , (2-9)

where each Li and Mi,j are calculated by Eq. (2-2) through Eq. (2-7).

1 5
3

Figure 2.2 An example of a planar rectangular spiral coil .

11
2.1.3 Calculation of the Mutual Inductance between a Coil and a Conducting
Plate

The main function of the inductive proximity sensor is to detect the

existence of a conducting plate. The magnetic field generated by the time-varying

current flowing in the coil excites the eddy current on the plate. The eddy current

then produces a magnetic field normal to the plate, which creates the magnetic flux

linkage in the coil. To reduce the complexity of solving the Poisson’s equation of

this system directly, the method of images is applied instead. By assuming that the

plate has zero resistivity and is infinite in extent, the conducting plate can be

replaced by the coil’s image being placed below the plate surface at a distance

equal to the distance between the coil and the plate surface, as drawn in Figure 2.3.

The current in the image coil flows in the opposite direction but at the same

magnitude as that in the coil. This can be done by imaging the signal source driving

the coil to the plate. Therefore, the calculation of the mutual inductance between

the coil and the plate placed at a distance d from each other is simplified by the

calculation of the mutual inductance between two identical inductors placed 2d

from each other. This calculation can be evaluated by Eq. (2.3) through Eq. (2.7) if

the coil structure is known.

12
 B1 B1

I1 I1
d d
Inductor Inductor

Metal sheet target

Metal sheet target d

I2

Reflected image inductor

B2

(a) Physical arrangement (b) Application of image theory

Figure 2.3 Illustration of the image theory for a system consisting of a sensing coil
and a metal target plate.

2.2 MODELING OF THE TWO-COIL INDUCTIVE SENSOR

The two-coil inductive sensor model is based on the derivation of the one-

coil inductive sensor model. So the one-coil modeling algorithm will be discussed

at first and then the two-coil model will be drawn out.

2.2.1 One-Coil Inductive Sensor Modeling

According to the method of images, the one-coil sensor with a conducting

target plate is present by the transformer circuit model in Figure 2.4. Vi, Ii, Ri, and

Li are the images of V1, I1, R1, and L1, respectively, so that Vi = V1, Ii = I1, Ri =

R1, and Li = L1. Besides, Mi is the mutual inductance between the coil and its

image. The effective inductance of the sensing coil thus is

13
Leff_i = L1 - Mi. (2-10)

The minus sign is a result of the image current traveling in the opposite direction to

that of the coil current.

I1 Ii

R1 Ri

V1 L1 Mi Vi
Li

Sensing circuit Image circuit

Figure 2.4 One-coil inductive sensor model based on the image theory.

A simplified circuit model [9] is drawn in Figure 2.5. In this case, the image

voltage source is removed, and Is, Ls and Rs are no longer necessarily equal to I1,

L1 and R1. The Kirchoff’s current loop equations for this circuit are

I1 ( R1 + jwL1) - Is ( jwMs ) = V1 (2-11)

-I1 ( jwMs ) + Is ( Rs + jwLs) = 0. (2-12)

Solving the two equations above, we obtain the effective impedance of the sensing

circuit,
V1  ω 2 Ms 2 Rs   ω 2 Ms 2 Ls 
zeff_s = = R1 +  + jω  L1 − . (2-13)
I1  Rs 2 + ω 2 Ls 2   Rs 2 + ω 2 Ls 2 

The effective impedance can be divided into the imaginary and the real parts as the

effective inductance and resistance of the coil, respectively. They are

ω 2 Ms 2 Ls
Leff_s = L1 - 2 (2-14)
Rs + ω 2 Ls 2
ω 2 Ms 2 Rs
Reff_s = R1 + 2 . (2-15)
Rs + ω 2 Ls 2

14
When the frequency ω is much higher than Rs / Ls, the dependence on frequency of

Reff_s and Leff_s is eliminated and thus Eq. (2-14) and Eq. (2-15) become
Ms 2
Leff_s’= L1 - (2-16)
Ls
Ms 2 Rs
Reff_s’= R1 + . (2-17)
Ls 2
In the following sections, it can be verified that ω is almost always much higher

than Rs / Ls for the two-coil sensor so that Eq. (2-16) and Eq. (2-17) can always be

applied.

I1 Is

R1

V1 L1 Ms Ls
Rs

Sensing circuit Simplified

Image circuit

Figure 2.5 Simplified one-coil inductive sensor model.

To force the simplified model in Fig. 2.5 to match the image theory model

in Figure 2.4, the effective inductance in two cases must be equal, which implies
Ms 2
Mi = . (2-18)
Ls
This is done by equating Eq. (2-10) and Eq. (2-17).

It is desirable to obtain the values of the parameters Ms, Ls and Rs based on

the image theory model the parameters of which are known already. According to

Eq. (2-18), it is found that a relation between Ms and Ls is needed to draw these

15
two parameters from the known Mi. This relation can be acquired as described in

the following.

A sensing coil segment with a grounded conducting plane can be connected

in two ways as shown in Figure 2.6(a) and (b). Figure 2.6(c) and (d) are the

equivalent circuits of the two connection schemes with the method of images

applied. The resistance of the coil is neglected to simplify the problem and the

image voltage source is removed as the simplified model in Figure 2.5. The

effective inductances seen by the voltage source in the two cases are
Mg 2
Leff_a = Lc - (2-19)
Lg

Leff_b = Lc + Lg - 2 Mg, (2-20)

respectively, where Eq. (2-19) is derived from Eq. (2-17).

If the grounded plane conductor has infinite conductivity, then the effective

inductance seen by the voltage source in the two cases should be the same, which

means Eq. (2-19) should be equal to Eq. (2-20). Thus,

Mg 2
- = Lg - 2 Mg. (2-21)
Lg

A simple relation between Lg and Mg can be calculated by rearranging Eq. (2-21),

Lg = Mg, (2-22)

which implies Ls = Ms in the simplified model in Figure (2.5). Moreover, Eq. (2-

18) becomes

Mi = Ms = Ls. (2-23)

16
 i1 i2
sensing coil
V segment
i1

V Lc Lg
i2
Mg

Grounded
plane conductor

(a) (c)

i1
sensing coil V
segment
i1

V i1
Lg Lc

Grounded Mg
plane conductor

(b) (d)

Figure 2.6 Two connection schemes for extracting the sensor model parameters.

For the simplified model in Figure 2.5, R1 and L1 can be obtained by the

methods indicated in Section 2.1.1 and 2.1.2, and Ls and Ms are equal to Mi which

can be calculated by the technique described in Section 2.1.3. The only parameter

still unknown is Rs , the resistance of the conducting plate. A quasi-static model

for microstrip transmission lines is utilized to calculate Rs. Figure 2.7 shows the

structure of this model which consists of a microstrip with a width of w over a

ground plane at the distance of h. The distance k is a measure of how long the

magnetic fields spread before reaching the ground plane. With k = 3h + w/2, the

impedance per unit length of the microstrip obtained using this quasi-static model

matches that obtained by the full-wave model within 3% [11]. Therefore, a

17
microstrip over a ground plane can be replaced by a microstrip over a conductor

with a width of 6h + w, and the resistance of the conducting plane below a

microstrip is the resistance of a conductor with a width of 6h + w. It should be

noted that the plane resistance depends on both the distance between the

microstrip and the ground plane and the conductivity of the ground plane.

microstrip

ground plane k

Figure 2.7 Using the model of a microstrip over a ground plane to calculate the
resistance of the target plate [11].

2.2.2 Two-Coil Inductive Sensor Modeling

The two-coil inductive proximity sensor has two identical coils, one of

which provides the excitation magnetic signals while the other picks up the

magnetic signals corresponding to the proximity information of a target plate.

Their individual impedance and the mutual inductance between them (Rprimary,

Rsecondary, Lprimary, Lsecondary and M12 as shown in Figure 2.8) can be calculated using

the method discussed in Section 2.1.1 and 2.1.2. The target plate is modeled based

on the simplified one-coil model in Figure 2.5, where Lplate is equal to Mplate based

on Eq. (2.23). The values of Mplate and Rplate are obtained by the image theory and

the quasi-static model mentioned in Section 2.1.3 and 2.2.1, respectively. Cg and

18
Rg are added as the gap resistor and capacitor between the sensor and the plate

because SPICE does not allow the existence of a floating circuit. A very large and

a very small value are assigned to Rg and Cg, respectively, in order to eliminate

their effects.

target plate
Rplate

Lplate

Cg Rg Rg Cg
Rprimary M plate Mplate Rsecondary

R1
Interface
V Lprimary M12 Lsecondary circuit

exciting coil pick-up coil

Figure 2.8 The schematic SPICE model of the two-coil sensor with a target plate.

For a certain two-coil sensor design, all the parameters are fixed for any

situation except Rplate and Mplate (also Lplate, which has the same value as Mplate).

Mplate relies on the distance between the target plate and the sensor while Rplate is

related to both the proximity and the conductivity of the target plate. Thus, it can

be affirmed that there is a unique solution set of proximity and conductivity of the

target for a given set of Rplate and Mplate. Also, this can explain another reason why

the simplified model (Figure 2.5) is preferable to the image model (Figure 2.4). By

using the image model, we can only extract the information about the distance of a

target, which is ideally a perfect conductor, through the single electronic variable

Mi. However, the effects of the nonideal conductivity of the target material are

19
seen as a reason for the degradation of sensing capability. With the simplified

model, the two physical parameters, distance and conductivity of the target, can

both be transferred to the magnitudes of the two electronic components, Rplate and

Mplate.

B
D

Figure 2.9 The mask design of a planar rectangular spiral two-coil sensor, where A
and B are the two terminals of one coil while C and D are the two
terminals of the other.

20
2.3 TWO-COIL INDUCTIVE SENSOR SIMULATION

Ref. [9] has proven that the simulation of the two-coil model is matched

quite well to the measurement results of a prototype sensor built on a copper PCB

(Figure 2.9). The simulation is done by adding the equivalent circuits of an

HP4194A Gain/Phase Analyzer, which is applied to measure the prototype sensor,

to the two-coil SPICE model in Figure 2.8. The driving circuit of the HP4194A

contains an ideal voltage source and a series resistor, and its detecting circuit is a

load resistor. The PSpice schematic circuit is shown in Figure 2.10. The
coupling factor between the two inductors L1 and L2 is defined as M12 / L1 L2 ,

so that the coupling factor between Lplate and Lprimary is

Mplate Lplate Lplate
k= = = (2-24)
LplateLprimary LplateLprimary Lprimary

Figure 2.10 The PSpice schematic circuit of the two-coil sensor together with the
equivalent circuit of the HP4194 Gain/Phase Analyzer.

21
 The variation of Rplate and Lplate is reflected in the phase difference between

the electric signals in the pick-up coil and the driving coil. This can be described

more precisely as the phase difference between node voltage Voutput and Vsource in

Figure 2.10. Despite the fact that the gain of Voutput / Vsource also depends on Rplate

and Lplate, the gain is generally very low ( below -50dB for frequencies lower than

1MHz ) and is thus difficult to measure accurately. Figure 2.11( a ), ( b ), and ( c )

show the influence of the variation of Lplate, Rplate, and Rprimary on the phase versus

frequency curves, respectively. It is observed that Rprimary has little effect on the

phase shift in the frequency range concerned so that this design is independent of

the resistance variation that arises from the process deviation. There exists a

minimum phase shift for each concave curve in the frequency range from 10KHz

to 1MHz. At the frequency at which the minimum phase shift occurs, the

sensitivity of the phase shift to Lplate achieves its maximum ( Figure 2.11 ( a ) ).

The value of the minimum phase shift decreases when Lplate increases for a fixed

Rplate, while the frequency at which the phase shift achieves its minimum increases

when Rplate increases for a fixed Lplate (Figure 2.11( b )). It is desirable to determine

the values of Lplate and Rplate by the minimum phase shift Φ and the frequency f at

which the minimum phase shift occurs. However, this can be done only if the

system has a unique solution set of Lplate and Rplate for a measurement result of Φ

and f, which will be proved in Chapter 4.

22
 90
15nH
80 25nH

35nH
70
45nH
60 55nH

50

40

30
1E+3 10E+3 100E+3 1E+6 10E+6
Frequency(Hz)

Figure 2.11( a ) Simulation of the phase shift between Voutput and Vsource with
different values of Lplate. ( Rplate = 0.01Ω and Rprimary = 0.36Ω )

90
0.01Ω
85 0.03Ω

0.07Ω
80
0.1Ω
75

70

65

60
1E+3 10E+3 100E+3 1E+6 10E+6
Frequency(Hz)

Figure 2.11( b ) Simulation of the phase shift between Voutput and Vsource with
different values of Rplate. ( Rprimary = 0.36Ω and Lplate = 35nH )

23
 90
0 .1 Ω
85 1Ω

10Ω
80
100Ω
75 1 κΩ

70

65

60
1E+3 10E+3 100E+3 1E+6 10E+6
Frequency (Hz)

Figure 2.11( c ) Simulation of the phase shift between Voutput and Vsource with
different values of Rprimary. (Rplate = 0.01Ω and Lplate = 35nH)

2.4 REQUIREMENTS OF THE INPUT AND OUTPUT INTERFACES

In order to design the interface circuit of a two-coil sensor, we need to

understand the input and output requirements of the sensing element. For the

output requirement, it is obvious that the stage following the sensing element

should have a high input impedance to reduce its effect on the sensing

performance. For the input circuit, we notice that the driving source of an

HP4194A can be replaced by its Norton’s equivalent: a current source and a

parallel resistance of 50Ω . Since the input impedance of the sensing element is

much lower than 50Ω in the relevant frequency range, the parallel resistance of

50Ω can be neglected and therefore the required driving component turns out to

be an ideal current source. Moreover, the phase shift between Voutput and Vsource in

24
Figure 2.11 is equivalent to the phase shift between Voutput and the ideal current

source. Since it is impossible to measure the phase of a current directly, a voltage-

controlled current source, the current output of which is in phase with the

controlling voltage, is more appropriate than a simple current source.

By replacing the driving circuit with an ideal voltage-controlled current

source, the simulation results of the phase shift between Voutput and Vsource (Figure

2.12) are matched to those in Figure 2.11. The phase shift of Φ (Voutput) -

Φ (Vsource) can be divided into two parts,

Φ (Voutput) - Φ (Vsource) = [ Φ (Voutput) - Φ (Vin) ] + [ Φ (Vin) - Φ (Vsource) ]. (2-25)

The first part of Eq.(2-25), which is shown in Figure 2.13( a ) and is equivalent to

the phase shift between the source and the output voltages if an ideal voltage

source is used instead, deviates from the demanded characteristic of a concave-

shaped curve and decreases dramatically at higher frequencies. This explains why

an ideal voltage source is not desired. The second part of Eq. (2-25), which is
equivalent to the phase of the input impedance, increases from zero to 90§ as the

frequency goes up (Figure 2.13( b )), as a result of that the input impedance of the

sensing element is approximately Rprimary + jωLprimary. The summation of the curves

in Figure 2.13( a ) and ( b ) results in the concave-shaped curves as shown in

Figure 2.11( a ). Besides, both the first and the second parts of Eq. (2-25) vary

with Rprimary, but their summation is independent of it.

25
 Vin Voutput
+
Two-coil
Vsource I = AV source Sensing Rload
- Element

Figure 2.12 Circuit for simulating the sensing element with a ideal voltage-
controlled current source as the driving component. ( Rload = 50Ω )

90
15nH
80
25nH
70
35nH
60
45nH
50
55nH
40

30

20

10
0
1E+3 10E+3 100E+3 1E+6 10E+6
Frequency(Hz)

Figure 2.13( a ) Simulation of the phase shift between Voutput and Vin in Figure 2.12
with different values of Lplate for an ideal voltage controlled current
source as the driving circuit. (Rplate = 0.01Ω and Rprimary = 0.36Ω )

26
 90
15nH
80
25nH
70
35nH
60
45nH
50
55nH
40

30
20

10

0
1E+3 1E+4 1E+5 1E+6 1E+7 1E+8
Frequency (Hz)

Figure 2.13( b ) Simulation of the phase shift between Vin and Vsource in Figure 2.12
with different values of Lplate for an ideal voltage controlled current
source as the driving circuit. (Rplate = 0.01Ω and Rprimary = 0.36Ω )

27
 Chapter 3: The Interface Circuit Design

The interface circuit of the two-coil sensor is composed of three parts, a

driving circuit, a phase detector, and a feedback loop to adjust the operation

frequency to f, the frequency at which the minimum phase shift occurs. Two

feedback controlling techniques will be discussed in this chapter, and their

simulation results will be shown in Chapter 4.

All the circuits mentioned in this chapter are demonstrated by SPICE

simulation with the third-level 0.8µm CMOS process models available through the

MOSIS services ( Table 3.1 ). The 0.8µm process is chosen because its 5Volt

operating voltage is a reasonable magnitude for the analog integrated circuit design

compared with 3.3Volt of the 0.35µm process.

MODEL CMOSN CMOSP

TYPE NMOS PMOS
LEVEL 3 3
VTO 0.663 -1.009
PHI 0.6 0.6
TOX 1.73E-8 1.73E-8
XJ 0.2E-6 0.2E-6
TPG 1 -1
DELTA 9.0290E-1 2.012E+0
LD 7.273E-9 4.546E-9
KP 1.4669E-4 5.6847E-5
UO 734.9 284.8
THETA 4.666E-2 3.303E-1
RSH 2.191E+1 1.817E+1
GAMMA 0.5443 0.4558
NSUB 3.556E+16 2.493E+16
NFS 1.027E+13 1.799E+13
VMAS 1.251E+5 1.444E+5
ETA 8.828E-3 7.081E-2
KAPPA 1.803E-1 2.598E-1
CGDO 2.1776E-11 1.3611E-11
CGSO 2.1776E-11 1.3611E-11
CGBO 4.5116E-10 4.8015E-10
CJ 2.4537E-4 5.5486E-4
MJ 1.0036 0.5099
CJSW 6.8063E-10 1.0633E-10
MJSW 0.332 0.8009
PB 0.8 0.85

Table 3.1 SPICE level 3 parameters of the 0.8 µm CMOS process.

28
 The performance of each circuit component is characterized by its linearity

error or accuracy. The linearity error is defined as the maximum deviation of the

actual transfer function from the best fitted straight line that describes the output

signal, while the accuracy is specified as the ratio of the maximum error of the

output signal to the full-scale output signal in a percentage [12].

3.1 INPUT STAGE

As indicated in Chapter 2, the input stage includes a sinusoidal oscillator

which provides a frequency ranging from 10KHz to 1MHz and a voltage

controlled current source which creates enough power to excite the eddy current

in the target plate.

3.1.1 Voltage Controlled Oscillator

Voltage controlled oscillators (VCOs) are widely used for generating

untuned operating frequencies. Their outputs are typically square waves and

triangle waves, so a sine-shaping circuit is needed to reduce the harmonic contents

of the VCO output in this case. Several studies have proposed methods to design a

low voltage VCO [13][14][15][16]. According to the consideration of a VCO

which can produce a frequency range of two decades (10KHz ~ 1MHz) and can be

fabricated by the standard CMOS process, the circuit shown in Figure 3.1 is

chosen.

The VCO is based on the operation of the astable multivibrator. M1, M2,

M5 and M6 form a current mirror whose current value corresponds to the input

voltage Vin. The inverter of M3 and M4 is switched by the output of the

comparator and sources or sinks current of the magnitude of that in the current

29
mirror. Due to the basic relationship between the voltage and current on a

capacitor for a constant current, I = C ∆V/∆t, the oscillator’s frequency can be

derived as
1 IC1 IC1
f= = = , (3-1)
2 ∆t 2C1∆V 2C1(VTRP + − VTRP − )

where VTRP+ and VTRP- are the positive and negative trip points of the comparator

with hysteresis, respectively, and the capacitor charging current, IC1 is ideally

equal to i1. Because C1, VTRP+ and VTRP- are fixed for a certain design, the

oscillating frequency only depends on IC1.

Vdd = 2.5V

M1 M2
Vin

i2
i1 Comparator with hysteresis
M3

Vout
-
M4
C1
+

i3

R2
R1
M5 M6

Vss = -2.5V

Figure 3.1 Schematic diagram of a VCO compatible with CMOS technology [16].

30
 As the input voltage decreases from Vdd ( = 2.5Volt), M1, M2, M5 and M6

change from the turn-off region to the saturation region and the current in the

current mirror increases from zero amperes. This small current guarantees a two

decade difference between the minimum current and the maximum current. For the

purpose of obtaining a symmetric triangle waveform in the output, the transistors

in the current mirror need to be in the saturation region which makes I1, I2 and I3

in Figure 3.1 equivalent to each other via transistor matching. To reduce the

magnitudes of VGS, M3 and M4 should be assigned higher β, which stands for

µeffCoxW/L, where µeff is the effective mobility of electrons or holes, Cox is the gate

oxide capacitance per area and W/L is the width-to-length ratio of the gate area.

Also, the β3/β4 ratio must be adjusted to set the trip point of the CMOS inverter at

zero volt. When the input voltage continues to decrease, M1 enters its linear region

and the relation between the frequency output and the voltage input is no longer

square but linear. Referring to Figure 3.2, the output frequency is roughly a linear

function ( with linearity error of 18KHz ) in the input range from -2.5Volt to

0.6Volt (corresponding to 130KHz to 980MHz in the output frequency) and is

proportional to the square of the input voltage for higher input voltage.

For the VCO in Figure 3.1, the gate of either PMOS M1 or NMOS M5 can

be chosen as the input. Since the active loading MOS (M5 in our case) of the input

stage is always in the saturation region, it should be assigned the kind of MOS

which is less susceptible to the channel-length modulation effect. Therefore, the

gate of PMOS is chosen to be the input node since the channel-length modulation

31
parameter of p-channel devices (λp) is higher than that of n-channel devices (λn) in

the model that is applied.

1000
900
800
700
600
500
400
300
200
100
0
-2.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5
input voltage ( Volt )

Figure 3.2 Simulation result of the VCO circuit.

(C1=35pF, V TRP+ = -VTRP- = 1.25Volt)

During the SPICE simulation, the VCO output signal may not be a periodic

waveform but a d.c. voltage. This problem can be solved by restricting the

maximum time step of the transient analysis to a small number via setting the

ceiling time step.

3.1.2 Sine-Shaping Circuit

Since the phase shift of the two-coil sensor is very sensitive to the

operating frequency, the driving signal needs to have a pure sinusoidal waveform.

The triangle waves generated by the VCO have significant odd harmonic contents

that are undesirable. It is not practical to apply filters to remove them due to the

difficulty of accurately tuning the variable frequencies. Sine-shaping techniques are

more commonly used for reducing the harmonic contents of the VCO output.

32
Instead of diode and bipolar shaping networks which require more sophisticated

design and are not appropriate for standard CMOS process, a simple CMOS sine-

shaper, shown in Figure 3.3, is used. The square-law differential property of M3

and M4 performs the sine-shaping function. The total harmonic distortion (THD),

defined as [17]
THD = 100 ( D2 2 + D 32 + D 4 2 + ......) , (3-2)

where Dk ( k = 2, 3, ..) are the ratios between the amplitude of the kth harmonic

and that of the fundamental in the Fourier series representation of the output

waveform of the sine-shaping circuit coupled to the VCO mentioned in Section

3.1.1, is plotted as a function of the VCO frequency in Figure 3.4. If a BiCMOS

process is available, a refined sine-shaping circuit based on the same differential

algorithm but providing a THD under 5% and being independent of the VCO

offset voltage, device characteristics and temperature can be built [18]. The

transient simulation results of the sinusoidal VCO are plotted in Figure 3.5.
V dd

M3 M4

Sinusoidal Output

VCO Signal M1 M2

Vbias

Vss

Figure 3.3 Sine-shaping circuit.

33
 12

10

0
10 100 1000
Frequency (KHz)

Figure 3.4 THD at the sine-shaper output as a function of VCO frequency.

Figure 3.5 The waveform at the sine-shaper output, VCO output, comparator with
hysteresis output and the current flowing in the VCO capacitor.

3.1.3 Transconductance Amplifier

The driving circuit for the primary coil requires an ideal voltage-controlled

current source which is also called the transconductance amplifier. The class AB

output amplifier in Figure 3.6 ( a ) was formed utilizing discrete components and

34
was shown to provide enough power to the two-coil sensor by David Lee and

David Onsongs [19]. Since the maximum current the two Darlington BJTs

(NTE245 and NTE246) can provide is 1 Ampere and the load impedance of the

transconductance is below 1Ω , the maximum output swing is ±1Volt – an output

swing much lower than the ±15Volt power supply voltages required for the power

BJTs to generate 1 Ampere. In this case, the two transistors rather than the

primary coil consume most of the power.

V dd

V dd Vdd

M1

Vg1 VB1

NTE245
M2

1N5226B + VB2
io
io io
Vin Vin Vin
M3
1N5226B +
VB3

NTE246
M4

Vg2 VB4

Vss Vss

Vss

(a) (b) (c)

Figure 3.6 Transconductance amplifiers: ( a ) Class AB BJT output stage ( b )

Class AB CMOS output stage ( c ) A linear transconductance
element [20].

35
 The two most widely used approaches to designing a transconductance

amplifier compatible with CMOS process are a differential stage approach with

clever linearizing techniques and a class AB output stage approach illustrated in

Figure 3.6 ( b ). The latter is far simpler than the former; however, its linear

performance depends critically on the matching between the NMOS and the

PMOS, i.e., βn = βp. By replacing each transistor in Figure 3.6 ( b ) with an

NMOS-PMOS pair, as shown in Figure 3.6 ( c ), the matching between n-channel

and p-channel transistors no longer affects the linearity. With identical NMOS

devices, M1 and M3, and identical PMOS devices, M2 and M4, the current flowing

through M2 and M3 can be derived as [20]

1
IM2 = βeff (Vg1 − Vin − VTN 1 − VTP 2 )
2
(3-3)
2
1
IM3 = βeff (Vin + Vg 2 − VTN 3 − VTP 4 ) ,
2
(3-4)
2

where
βeff = βnβp / ( βn + βp ) 2 , (3-5)

VTN1, VTN3, VTP2, and VTP4 are the threshold voltages of M1, M2, M3 and M4,

respectively. Output current Io is thus

Io = IM2 - IM3 = -βeff (Vg1 + Vg2 - ΣVT) Vin + βeff (Vg1 + Vg2 - ΣVT) ∆VT/2, (3-6)

where
ΣVT = VTN3 + VTN3 + VTP2 + VTP4 (3-7)

∆VT = (VTN3 - VTN3) + ( VTP4 - VTP2 ) + (Vg1 - Vg2). (3-8)

If four wells are used ( VBS = 0 for each transistor ) and Vg1 = Vg2, then Io has an

ideal linear relation with Vin. However, with a low operation voltage ( ±2.5Volt ),

it is difficult to isolate each well from the substrate. So, the two-well structure is

36
applied here, i.e., VB1 = VB3 = Vss and VB2 = VB4 = Vdd, and thus the body effect

causes a nonlinear effect in Eq. (3-6). The d.c. offset and second-order current

can be eliminated by appropriately choosing the device dimensions of the n-channel

and p-channel devices.

The biasing circuit providing Vg1 and Vg2 can be implemented by either a

resistive or a MOS voltage divider. If a resistive voltage divider is employed, the

resistors of the biasing circuit together with the parasitic capacitors CGD and CGS of

the transistors add high order poles and zeros to the phase shift curve in the

frequency range discussed in Chapter 2. The additional poles and zeros destroy the

concave-shaped curves and are sensitive to the change of the magnitude of the

biasing resistors caused by process variation. On the other hand, the equivalent

resistance of an MOS voltage divider with the parasitic capacitors of its four

transistors can also add a pole, which moves down to the relevant frequency range

when the parasitic capacitors increase as a result of the fact that large area

MOSFETs are used to provide the 500mA output current. A simple way to avoid

adding extra poles or zeros to the phase shift curve is to connect Vg1 and Vg2 to Vdd

and Vss, respectively.

Following the procedure described in [20] to optimize linearity with Vg1 =

2.5Volt and Vg2 = -2.5Volt , the relation, (W/L)p ≅ 5(W/L)n, is obtained for the n-

channel and p-channel devices in the linear transconductance element. Besides, for

the standard CMOS process, the large area transistors can be implemented by

connecting several identical small area components in parallel instead of using

power devices such as DMOS or VMOS, which require special geometric

37
configuration or various doping values for different regions [21][22]. This parallel

arrangement would result in an output current with the magnitude of the sum of

the currents flowing in each channel. The transconductance amplifier consisting of

one hundred elements in parallel each with W/L of 1180µm/0.8µm for NMOS and

200µm/0.8µm for PMOS leads to a linearity error of 6.7mA for input voltage

within ±0.6Volt and a transconductance of 726mA/Volt. For the a.c. analysis, the

phase of the current output is almost equal to that of the voltage input with a

maximum error of 0.5 degree.

3.2 OUTPUT STAGE

The output stage is a phase detector which generates a d.c. voltage

corresponding to the phase difference between the output signal of the sensing

element and the input signal of the transconductance amplifier. A prototype phase

detector built in the discrete-component level is analyzed first. Then, a transistor-

level phase detector compatible with 0.8µm CMOS process is developed to

achieve a monolithic sensor device.

3.2.1 Prototype Design of Discrete Components

The prototype configuration of the output stage, which is shown in Figure

3.7, includes an operational amplifier ( LM6265 with a resistive feedback loop ),

two identical comparators ( LM6265 ), an exclusive-or ( XOR ) gate ( 74AC86 ),

a buffer ( AD9630 ), and an integrator (a resistor and a capacitor). LM6265 is

chosen to perform the amplification function for the small signal in the pick-up coil
because its close-loop frequency response of phase drops only 1§ from 0§ within

38
1MHz. This small amount of the phase drop implies that the phase difference

between the signal at the pick-up coil and the transconductance input

Figure 3.7 Phase detector built by discrete components (Vdd = -Vss = 15Volt).

Figure 3.8 Transient simulation results of the prototype phase detector. From
bottom to top: waveforms at the pick-up coil, the op-amp output, the
first comparator ( U2 in Figure 3.7 ) output, the second comparator
out put ( U3 in Figure 3.7 ), and the XOR output.

39
signal can be taken as the phase difference between the pulses at the outputs of the
two comparators with a maximum error of 1§. The XOR gate then produces the

absolute difference of the pulses at the two comparator outputs. The ratio of

average pulse width over the pulse period of the pulses at the XOR output is

proportional to the phase difference between the two comparator output signals.

Since the XOR gate is a simple digital component, a buffer is needed to provide

enough power to charge the capacitor in the integrator. Finally, the integrator

averages the pulse train and generates a d.c. voltage in its settling time. Figure 3.8

explains the operation of the phase detector by showing the waveforms obtained

by the circuit simulation with prototype CMOS transconductance element in

Figure 3.6 ( b ) fed by an ideal a.c. voltage signal, Rplate = 0.015Ω , Lplate = 50nH

and Rprimary = 0.36 Ω .

A good phase detector should perform accurately, linearly and fast.

Relating these factors to the prototype circuit described above, it is clear that,

ideally, the input offset voltage and the resolution of the comparators and the

ripple and the settling time of the integrator should ideally be zero.

3.2.2 Integrated Phase Detector Circuit

The transistor level phase detector structure is generally equivalent to the

prototype except that the op-amp of the first stage is taken away because

comparators with high gain and accuracy are available.

The open-loop comparator is based on the two-stage structure in Figure

3.9. Three inverter pairs are added to increase the gain and the drive capability,

and thus the rise and fall times of the output can be reduced. The first two stages

40
are connected from -2.5Volt to 2.5Volt because a larger power supply is needed to

keep each transistor biased in the saturation region, while the sources of n-channel

transistors of the inverter pairs are connected to the ground since the XOR gate

requires a non-negative pulse train input. For the XOR gate described in the next

paragraph as the load and with one of the inputs connected to the ground, the

input offset voltage and resolution of the comparator are 0.55mV and 0.1mV,

respectively, and the fall time and rise time are within 10ns in the frequency range

of concern. The gain is 25000V/V.

Vdd=2.5V

VN VP Vout

Vbias

Vss=-2.5V

Figure 3.9 Two-stage comparator followed by three inverter pairs.

Vout = A(VN-VP), where A is the gain of the comparator.

Two schemes to implement XOR gates are examined. The novel one in

Figure 3.10 ( a ) inhabits a much smaller area than that of the conventional XOR

gate in Figure 3.10 ( b ), however, the input impedance of the two input nodes is

different so that the phase response of frequency for the two comparators is

different. As a result, the phase difference between the two comparator output

signals deviates from the phase difference between the transconductance input and

41
the sensing element output signals. On the other hand, the conventional CMOS

XOR gate has approximately the same input impedance for both inputs, and in this

case, the two comparator outputs will not interfere with each other during a.c.

analysis as they will in the tiny XOR because of the isolation of the high gate

impedance at the XOR inputs. The buffer stage following the XOR gate can be

easily implemented with several inverter pairs.

Vdd

A
B
A B
A+B

A+B

(a) (b)

Figure 3.10 Schematics of XOR gate: ( a ) Transmission gate XOR (tiny XOR)
[23] ( b ) Conventional XOR.

The peak-to-peak ripple voltage and the settling time of the single-pole
integrator in the prototype circuit are derived as

Vripple ≈T × IR / C1 (3-9)

Tsettling ≈R3 × C1, (3-10)

where T is the period of the operating frequency, C1 and R3 are the capacitor and

the resistor of the integrator as shown in Figure 3.7, respectively, and IR is the

average current that the buffer provides and is approximately inversely

proportional to the integrator resistor R3. To increase the resolution of the

42
integrator, low ripples are required; however, decreasing the ripple increases the

settling time. This trade-off can be minimized by filtering the single-pole integrator

output by a high-order low-pass filter [24]. By this method, the single-pole

integrator is designed with high ripples but a fast settling time, and the ripples are

then reduced by the high-order low-pass filter. Except for the frequency range

below 30KHz, during which the ripples increase dramatically, this approach results

in a resolution within 0.004% ( with 0.1mVolt of ripple and 2.5Volt full scale

output ) and a settling time of 2ms, compared with 0.4% ( with 60mV of ripple

and 15Volt full scale output) and a settling time of 0.5ms for the single-pole design

in the prototype circuit. The linearity error is 5mV.

3.3 SUCCESSIVE FEEDBACK LOOP

3.3.1 Theory of Operation

The successive feedback scheme is based on the operation of an HP4194A

Gain/Phase Analyzer. The operating frequency scans from an initial value ( 20KHz,

in our case) and increases by a fixed amount in logarithmic scale at each rising

edge of the clock pulse until the polarity of the slope of the phase response to the

frequency becomes positive. The block diagram and the waveform plotted in

Figure 3.11 and Figure 3.12, respectively, illustrate how this feedback loop

functions.

The phase detector output is sampled during Clock1 or Clock2 by sample-

and-hold circuit 2 or 1, respectively, to store the phase information V(φn-1) that

comes in response to the operating frequency in the previous clock cycle (a clock

cycle means half of the period of Clock1 and Clock2). Two control clocks, Clock1

43
 Main Circuit Clock2 OR Reset
Reset

Vinitial
(frequency=20KHz) Sinusoidal Inductance Phase Sample
Reset Transconductance
VCO shaping
amplifier Sensor detector & hold
circuit 1

Clock1OR Reset
1+δ
D (Clock3orReset)and D

Sample
Phase Readout
Frequency
& hold
Voltage Sample to 2
D Level & hold voltage
Translator converter
Clock1

Comparator+
Frequency Readout 1
-

D Clock2
Clock1NORClock2 Clock1ORClock2
Comparator+
2 -
Latch Latch D
Control signalD 2 1
Logic 0
Reset

Figure 3.11 Block diagram of the sensor system with successive feedback loop.
The dashed line indicates the main circuit which will be discussed in
Chapter 4.

47.5
Reset

Clock1
37.5
Clock2
Clock3
27.5
Phas_dec_out

SH1
17.5
SH2
comp1
7.5
comp2

Dcontrol
-2.5
0 2 4 6 8 10
Time (ms)

Figure 3.12 Clock and control signal waveforms (From top to bottom: Reset,
Clock1, Clock2, Clock3, the outputs of phase detector, sample-and-
hold circuit 1, sample-and-hold circuit 2, comparator 1, and
comparator 2, and the control signal D as illustrated in Figure 3.11.)
44
and Clock2, of the same period but with 180§ phase difference are needed for

sampling the phase detector output because the transit time will result in

undesirable charge redistribution in the capacitors of the sample-and-hold circuits

and the phase detector if only one clock is used here. The voltage outputs of the

two sample-and-hold circuits, V(φn) and V(φn-1), one of which corresponds to the

present operating frequency fn and the other to the previous one fn-1, are compared

by the following comparators. If V(φn) - V(φn-1) is negative, i.e., the slope of the

frequency response of phase shift is negative, the ‘D’ control signal in Figure 3.11

will be given the value of logic high. On the other hand, if V(φ(n)) - V(φ(n-1)) is

positive, i.e., the slope of the frequency response of phase shift becomes positive,

the ‘D’ control signal will change to logic low and keep logic low during all the

following clock cycles. The ‘D’ control signal plays the role of determining

whether the operating frequency fn+1 of the next cycle should continue to increase

in the sub-loop consisting of a VCO and a frequency-to-voltage converter ( FVC ).

During Clock3, which rises right after each pulse of Clock1 and Clock2 with a

period determined by the settling time of the phase detector and the FVC, the

sample-and-hold circuit in the sub-loop samples the FVC output. Then the sampled

voltage increases by timing 1 + δif the ‘D’ control signal is logic high or keeps the

same value if the ‘D’ control signal is logic low. A voltage level translator follows

to translate the voltage level from the FVC scale (0 to 2.5Volt) to the VCO scale

(-2.5Volt to 2.5Volt) and feeds the new voltage value to the VCO, which then

converts the voltage to the operating frequency of the next cycle.

45
 Note that the frequency of Clock1 and Clock2 is half of that of Clock3 as a

result and the three clocks do not overlap at any time. The pulse width of Clock3

should be short enough to keep the sampled FVC output unchanged with the

variation of the VCO input through the feedback loop during the sampling time.

3.3.2 Frequency-to-Voltage Converter

The pulse-integrating FVC is applied here to produce a d.c. voltage linearly

proportional to the input frequency. The FVC circuit is based on the configuration

of the charge-balancing VFC device, VFC320 of Burr-Brown Corporation

[17][24][25], as shown in Figure 3.13. The waveforms are drawn in Figure 3.14.

The input comparator ( comparator A ) converts the sinusoidal waves at the sine-

shaping output to square pulses. By suitable biasing, a signal which is insured to

cross the zero threshold at each falling edge of the input pulse for any frequency

concerned is generated through the coupling capacitor C1 and triggers the flip-flop

built with two NAND gates. The one-shot output of the flip-flop, in turn, closes

the switch S1 and thus pulls a constant current Iref out of the capacitor Cos until Cos

is charged to the reference voltage Vref. The flip-flop then changes its holding value

and waits for the next trigger. During this period, Cos is discharged to zero through

the switch S2, which is closed by the flip-flop output. The duration Tos of the

current pulse flowing through Cos does not depend on the input frequency but is

determined by Cos, Iref, and Vref,

Cos × Vref
Tos = . (3-11)
Iref

Also, the voltage output Vo is obtained by integrating the one-shot pulse train,

which has the duration of Tos and the frequency of the sine-shaper output signal.

46
 Vdd
Vdd
Vdd
R2
sine shaper
output + C1
Co mpa rato rA -
- Co mpa ra to r1

R1 + One-shot R
Vo

Vs
s
Vs
s
Vdd
VRef One-shot
+
Compa ra to r2
-

Co s Vs s

S2 S
1

IRe f
V ss

Figure 3.13 Schematic of the FVC.

Figure 3.14 Transient simulation results of the FVC. (Form top to bottom: the
FVC input, the comparator A output, the comparator 1 input and
output, and the one-shot.)

47
Thus,
Tos
Vo = Vdd × = Vdd ×Tos × f , (3-12)
T

where T and f are the period and frequency of the sine-shaper output signal,

respectively, and the one-shot signal has the pulse height of Vdd, which is 2.5Volt

in this case. All the comparators and the integrator are constructed by the same

circuit techniques mentioned in Section 3.2.2.

Tos has to be no longer than the period of the maximum frequency but long

enough to achieve good resolution, i.e., high dVo/df. By setting Cos = 50pF , Vref = -

0.5Volt, and Iref = 28µA, Tos has the value of 0.886µs and dVo/df is 2.2mV/KHz.

Moreover, the simulation of this FVC results in a linearity error below 2mV.

3.3.3 Sample-and-Hold Circuit

The simple structure of the sample-and-hold circuit applied is shown in

Figure 3.15. The capacitor CH is charged to the input voltage when switch φ is

closed and holds the voltage when φ is open. Since the pulse width of the three

clocks controlling the sample-and-hold circuits is 10µs and their periods are 2ms

or 4ms, the capacitor CH has to be small and the switch’s driving capability has to

be high in order to attain a charging time within 10µs. On the other hand, CH has

to be large enough and the switch’s capability of turning-off has to be good

enough to hold the charges for a comparably longer time of approximately 2ms or

4ms. Due to this trade-off, the value of the capacitor and the area of the switch

should be selected carefully to satisfy both the requirements. Besides, the gate

current of the positive input of the op amp should be eliminated to reduce the

leakage current during the holding time.

48
 To insure the stability of the feedback loop, the op amp should be

compensated properly to attain a phase margin of at least 45°. Moreover, the

common mode range ( CMR ) of the op amp should include the whole possible

output range of the phase detector and the FVC. The CMR is defined as the range

of common-mode values of the input signals that the op amp continuously senses

and amplifies the difference between the two input signals with the same gain [16].

-
A1 Vout
Vin +
φ
CH

Figure 3.15 Sample-and-hold circuit.

A single NMOS is used as the switch because the sampled voltage is

always positive. If a PMOS alone or a CMOS pair is applied instead, the voltage

difference between the body, which is connected to the most positive voltage,

2.5Volt, and the drain/source of the PMOS may not be high enough to maintain

the inverse-biased p-n junction between them, and thus the undesirable leakage

current will become very high. Dummy transistors are used to couple with each

single-transistor switch for the purpose of avoiding the clock feedthrough effect.

All the switches of sample-and-hold circuits are closed during ‘Reset’ to precharge

the capacitors gradually, otherwise the capacitors may not be able to follow the

sharp voltage increase during the first clock. For the following clock cycles, the

49
changes at the outputs of the phase detector and FVC are comparably small due to

the small frequency step, so the sample-and-hold circuits can accurately follow the

changes.

By carefully designing this system with the 0.8µm process parameters, the

sample-and-hold circuit decays by only 1mV during a holding time of 4ms with the

accuracy of 1% in the input range from -2.5Volt to 2.3 Volt. Input signals above

2.3Volt exceed the CMR of the op amp, so that the accuracy is degraded to 5%.

For the same reason, the voltage level translator has the linearity error of 10mV for

the whole input range except for input signals above 2.05Volt for which the

linearity error is as high as 0.8Volt.

3.4 ITERATIVE FEEDBACK LOOP

Instead of the fixed frequency steps in logarithmic scale for the successive

feedback scheme, the iterative feedback loop generates frequency steps that are

determined by the outputs of the phase detector and FVC during the last two clock

cycles. Besides, the frequency step can be either positive or negative as opposed to

the unidirectional characteristic of the successive feedback loop.

3.4.1 Theory of Operation

The block diagram of this design is drawn in Figure 3.16. Both the phase

detector and FVC have two sample-and-hold circuits that sample their output

voltages in turn by applying the same clock signals, Clock1 and Clock2, as those

used to control the switches in the successive feedback loop. Thus the phase

information Φ n-1 and Φ n-2 and the frequencies fn-1 and fn-2 for the last two clock

cycles can be stored in the form of voltage. An analog computer takes the outputs

50
of the four sample-and-hold circuits as its input signals and Clock1 and Clock2 as

its control signals and calculates the frequency step ∆ fn for the next clock cycle.

At the end of Clock1 or Clock2, the summation of ∆ fn and the last frequency fn-1 is

saved as a voltage value in the sample-and-hold circuit 6. During Clock3, which is

the same as Clock3 in the successive feedback design, the sample-and-hold circuit

5 samples the voltage containing the frequency information for the next clock cycle

and feeds it into the VCO.

Clock2 OR Reset2

Sinusoidal Phase Sample

Transconductance Inductance
VCO shaping
amplifier Sensor detector & hold
circuit 1

Clock1OR Reset1

Clock1OR Reset1 Sample

Phase Readout
& hold
2
Sample
& hold
3 Clock1
OR
Frequency Clock1Clock2 Clock2
to
voltage Clock2 OR Reset2
converter
Analog Sample
Sample Computer & hold
& hold ∆f 6
4 Frequency Readout
Clock3

Reset1
Sample
1.3Volt & hold
(20KHz) Reset2 5

1.05Volt
(50KHz)

Figure 3.16 Block diagram of the sensor system with iterative feedback loop.

It is obvious that the magnitude of the frequency step ∆ fn should be

proportional to the last frequency fn-1 and increases when the difference between Φ n-

1 and Φ n-2 increases according to the Newton-Raphson iteration. ∆ fn also increases

51
with the last phase shift Φ n-1. This will be explained in Chapter 4. Thus, the transfer

function of the analog computer has the form of

∆fn = C × ( fn − 1 + A) × ( Φ n − 2 − Φ n − 1 + B) k × (Φ n − 1 + D) l , (3-13)

where A, B,... and l are unknown coefficients and constants. The polarity of ∆ fn is

determined by that of Φ n-2 - Φ n-1 and that of fn-1 - fn-2. And for the same Φ n-1, Φ n-2, fn-1

and fn-2, the magnitude of positive ∆ fn should be larger than that for negative ∆ fn

because the approximately symmetric concave curve of the phase shift is drawn in

a logarithmic scale of the frequency. Besides, we should prevent the operating

frequency from sticking to the maximum frequency (1MHz).

The coefficients and constants of the transfer function are obtained by trial-

and-error on the spreadsheet until the lowest number of the average clock cycles

required for ∆ fn to converge and for the frequency variation to be below the

variation tolerance is achieved. The output ∆ fn is thus

∆fn = ∆fno × IF ( ∆fno > 0, 1 , 1/2.45) × IF ( fn-1 = 1MHz and ∆ fno > 0 , -1/2.45 , 1 ),

(3-14)

where
∆fno = 0.0028 × ( fn − 1 + 10) × Φ n − 2 − Φ n − 1 × (Φ n − 1 + 35)

× IF (Φ n − 2 − Φ n − 1 > 0, 1 , -1)
× IF ( fn − 1 − fn − 2 > 0, 1 , -1 ), (3-15)

and IF(A, x, y) is x if A is true and y otherwise. The circuit implementation of this

analog computer is shown in Figure 3.17. Φ n-1 - Φ n-2 is calculated by the difference

amplifier shown in Figure 3.18. This instrumentation amplifier configuration

ensures that the input impedance seen by each input is ideally infinite, and its

52
accuracy is 0.1% in the phase shift range concerned ( 25§ to 85§). The square

root and multiplication computation are performed by the multifunction

converters, which will be described in the next section. Note that all the frequency

and phase terms in Eq. (3-14) and Eq. (3-15) have to be converted to voltages

through the transfer functions of the FVC and the phase detector to propagate in

the voltage-based analog computer. And the output ∆ f in Figure 3.18 is an analog

voltage signal corresponding to the frequency step through the VCO transfer

function. As a result, the linearity of the transfer function of the FVC, VCO and

phase detector will have a serious influence on the accuracy of the analog

computer.
(1meg)AND(dF0)

(1meg)AND(dF0)
Clock1

dF0

Clock2 dF0

dF

dF
Clock1
-1/1.5 1.99V
SH2_out

Clock2
Vφ(n-1) Vx_2 × Vy_2
-1 -1 1/2.45 -1/2.45
SH1_out
1meg
Vy_2 Multifunction
Clock1 2.5V Converter
∆f
Vx_2 2
SH3_out 16.2mV 1/2.5

Clock2 Vy Multifunction
Vf(n-1) Converter sqrt (Vx)× VY
SH4_out -6
Vx 1
Clock2

Difference 6
Amplifier
Clock1

Vφ(n-1) -Vφ(n-2)

dP

dP

Figure 3.17 Schematic of the analog computer.

53
 V1 -
+ R4
R3
R2 -
R1 + +
R2 R3
R4 R2
R4 Vo= (1+2
R3 R 1 )(V2-V1)
-
-
V2 +

Figure 3.18 Schematic of the difference amplifier.

3.4.2 Multifunction Converter

The multifunction converter is based on the fundamental architecture of

two devices, LH0094 of National Semiconductor and 4302 of Burr-Brown

Corporation [17][26], as shown in Figure 3.19 ( a ). Four external discrete bipolar

transistors ( MM5262 ) with very low threshold voltage to satisfy the low power

supply requirement are added to the integrated circuit simulated. Knowing that the

negative input node of each op amp can be taken as virtual ground, we get
Vx Vy Vx Vo
Ix = , Iy = , Ix = , Io = . (3-16)
Rx Ry Rx Ro

By setting all the resistors of the same magnitude ( 10KΩ ), the ratio between any

two currents is equal to the one that exists between their corresponding voltages.
The exponential I-V characteristic of the bipolar transistors gives the following

relations,
V 1 − VAy _ out 
Io = Is exp  (3-17)
 VT 
 − VAy _ out 
Iy = Is exp . (3-18)
 VT 

Dividing Eq. (3-17) by Eq. (3-18) and rearranging the terms, we obtain
 Io 
V 1 = VT ln   . (3-19)
 Iy 

54
By the same method, V3 can be written as
 Iz 
V 3 = VT ln   . (3-20)
 Ix 

With R1 = R2 = 50Ω in Figure 3.19 ( b ), V3 = 0.5 V1, so that Eq. (3-19) and Eq.

(3-20) together yields ( Io / Iy ) = ( Iz / Ix ) 1/ 2 , which implies

1

Vz  2

Vo = V y

 . (3-21)
Vx 


The square root of Vz can therefore be computed by setting Vz and Vx equal to

1Volt. For the multiplication computation, letting V3 = V1 as shown in Figure 3.19

( c ) and thus equalizing Eq. (3-19) and Eq. (3-20), we obtain

 
Vz 
Vo = Vy   . (3-22)
Vx 

Ry Iy Io Ro
Qy VAy_ou t Qo V1 V2 V3
Vy Vo

- -
Ay Ao
+ + R1
R2

V1

Ix Rx (b)
Vx
Rz Iz
Qz Qx
Vy
- - V1 V2 V3
Az Ax V2
+ +

V3

(a) (c)

Figure 3.19 ( a )General schematic of the multifunction converter and node

arrangements for ( b ) square root extractor and ( c ) multiplier.

Figure 3.20 and 3.21 show the accuracy of the circuit in computing the square root

and in the multiplication modes, respectively. The accuracy can be better than 1%

55
if the input signals are converted to 1Volt full scale before being fed to the

multifunction converter.
9

8
7

5
4

3
2

0
0 0.5 1 1.5 2 2.5
Vz ( Volt )

Figure 3.20 Performance of the multifunction converter in square root

computing mode.

1x102
Vy=0.05V
1
1x10 Vy=0.5V

0 Vy=1V
1x10

-1
1x10

-2
1x10

-3
1x10

-4
1x10

-5
1x10
0 0.5 1 1.5 2 2.5
Vz ( Volt )

Figure 3.21 Performance of the multifunction converter in multiply mode.

56
 During the SPICE simulation, stability problems can occur in the iterative

feedback circuit even though each op amp employed is compensated properly. The

use of large bypass capacitors with one end connected to the negative inputs of the

op amps Az and Ay in Figure 3.19 ( a ) and the other connected to ground can

enhance the stability.

57
Chapter 4: Characteristics of the Sensor and the Interface Circuits

In this chapter, we will describe the performance of the sensor and the

interface circuits by referring to the SPICE simulation results. The sensitivity of the

sensing element with an ideal voltage-controlled current source as the input will be

quantified at first, and then the resolution of the main circuit in Figure 3.11 with an

input of an ideal voltage source will be determined. Finally, the successive and

iterative feedback schemes will be examined by the transient analysis.

4.1 SENSITIVITY OF THE SENSOR WITH AN IDEAL INPUT STAGE

The definition of sensitivity is the ratio of the change of the electronic

output signal to the change of the physical measurand [12]. Generally, it is

desirable that the change of an output signal is caused by only one measurand so

that the output readout can be converted to a physical quantity easily and

accurately. The two-coil sensor system that we are interested in has two output

signals, the minimum phase shift and the frequency at which the minimum phase

shift occurs, and two measurands, resistance and inductance ( Rplate and Lplate) of

the target plate, as indicated in Chapter 2. The a.c. analysis is applied to the circuit

configuration illustrated in Figure 2.11 in order to understand how these output

parameters relate to the measurands. The simulation results are given in Figure 4.1

and Figure 4.2. It is observed that the minimum phase shift is independent of the

resistance of the target plate. This means that if a pair of electrical output signals is

measured, a unique pair of Rplate and Lplate measurands can be determined. The

average sensitivity of the minimum phase shift to the target plate inductance is 1.14

degree/nH with a linearity of 8 degrees. The linearity is not so ideal, but in this
58
case, this phenomenon will not have any significant effect on the sensing

performance nor does it degrade the functioning of the interface circuit.

90
0.01Ω

80
0.03Ω

70
0.05Ω

60 0.07Ω

50 0.09 Ω

40 0.1Ω

30
10 15 20 25 30 35 40 45 50 55
Lplate (nH)

Figure 4.1 The minimum phase shift vs. the target plate inductance for diverse
fixed values of the target plate resistance. The input stage is an ideal
voltage-controlled current source.
1000
10nH
900
15nH
800
20nH
700
25nH
600 30nH
500 35nH

400 40nH

300 45nH

200 50nH

100 55nH

0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
Rplate (Ohm)

Figure 4.2 The frequency at the minimum phase shift vs. the target plate resistance
for diverse fixed values of the target plate inductance. The input stage
is an ideal voltage controlled current source.

59
 For a specific Lplate, which is extracted from the minimum phase shift

readout via Figure 4.1, Rplate can be derived from the frequency at which the

minimum phase shift occurs by referring to Figure 4.2. The sensitivity of the

frequency at which the minimum phase shift occurs to Rplate ranges from 69.6 to

174 KHz /0.01Ω , and the linearity is within 700Hz for each fixed value of Lplate.

4.2 RESOLUTION OF THE MAIN CIRCUIT

The resolution is defined as the minimum detectable change in the

measurand that can cause a change in the output signal [12]. According to this

definition, the resolution of the circuit configuration discussed in the previous

section is extremely small. However, it is impossible to obtain the phase shift

directly and to build an ideal voltage-controlled current source. Thus, to evaluate

the resolution of the practical sensing system before any feedback is involved, the

main circuit shown in Figure 3.11 is simulated through the a.c. analysis. The result

is given in Figure 4.3. Due to the fact that the output impedance of the

transconductance amplifier is not infinite, the minimum phase shift changes with

the target plate resistance for a fixed target plate inductance and this change can be

as high as 0.8 degree. Because the magnitude of this change is not random but

depends on other known parameters, this undesirable result can be minimized by

modifying the minimum phase shift at the output Φ to Φ ′through the following

equation,
 f − fo 
Φ ' = Φ + A ×  + ∆Φ ′, (4-1)
 k 

where f is the frequency readout at which the minimum phase shift occurs and

60
the coefficients and constant are listed in Table 4.1. The variation of the minimum

phase shift due to the change of Rplate for a fixed Lplate is thus below 0.03 degree as

listed in the ‘ Φ ′error ’ in Table 4.1. The resolution of Lplate is also shown in the

table and has a maximum value of 0.04nH.

90
0.01Ω
80
0.03Ω
70
0.05Ω

60 0.07Ω

50 0.09 Ω

40 0.1Ω

30
10 15 20 25 30 35 40 45 50 55
Lplate (nH)

Figure 4.3 The simulation results of the main circuit: the minimum phase shift vs.
the target plate inductance for diverse fixed target plate resistance.
i Φ (degree) k (KHz/Ω) A(degree/Ω) fo (KHz) ∆Φ ’(degree) Φ ’ error (degree) Lplate Resolution (nH)_
2 ≥82 214 0.209 181 0.193 0.011 0.02
3 79~82 140 0.137 125 0.13 0.02 0.03
4 76~79 106 0.103 99 0.099 0.01 0.01
5 72~76 88 0.08 84 0.082 0.03 0.04
6 67~72 78 0.709 75.5 0.07 0.005 0.005
7 62~67 73 0.0626 71 0.124 0.01 0.009
8 56~62 71 0.057 70 0.057 0.01 0.007
9 49~56 73 0.0538 72 0.055 0.002 0.001
10 37~49 81 0.0535 80 0.054 0.003 0.001
11 ≤37 109 0.0616 107 0.061 0.002 0.001

Table 4.1 Adjustment parameters for the minimum phase sift readout of the main
circuit. A is the slope of the minimum phase shift vs. Rplate for each
Lplate of i×5nH, k is the slope in Figure 4.2 for each Lplate of i×5nH, fo
is the frequency at which the minimum phase shift happens for each
Lplate of i×5nH and Rplate=0.01Ω , and ∆Φ ′is the fitting constant. Φ ′
error is the magnitude of the maximum variation of Φ ′caused by the
change in Rplate for each fixed Lplate of i×5nH. Lplate resolution is
derived from dividingΦ ′error by the differential of the curve in
Figure 4.1 at each Lplate of i×5nH.

61
 So far, it is supposed that the transconductance amplifier is a linear device

so that the small signal model used here is valid. However, the input signal of the

transconductance amplifier has an amplitude of 0.2Volt for the purpose of

providing a large current at the output and enhancing the power efficiency. With

this large signal of 0.2Volt, the nonlinear effect of the transconductance amplifier

must be taken into account. The method used to evaluate the degradation of

resolution caused by the nonlinear effect is illustrated in Figure 4.4. With -0.2Volt

and +0.2Volt as the input offset voltage, the a.c. performance deviates from the

expected performance with the ideal input offset voltage of 0V. The magnitudes of

the deviation, which depend on the values of Rplate and Lplate and can be taken as

the minimum detectable changes in the phase ( ∆Φ ) and the minimum effective

changes in the frequency ( ∆f ), respectively, are shown in Table 4.2 as well as the

resolution of Rplate and Lplate if only the nonlinear effect is considered. Three values

of Rplate are chosen to test three different frequency ranges, and for each Rplate, two

values of Lplate, one of which corresponds to the sharpest phase response ( low Φ )

and the other relates to the smoothest one ( high Φ ), are applied. We observed

that the resolution degrades with higher Rplate and lower Lplate, i.e., higher

frequencies and smoother phase response.

It is important to bias the output of the transconductance amplifier at zero

Volt, otherwise there will be a d.c. current flowing through the primary coil, the

magnitude of which will be sensitive to the variation of the coil resistance. As a

62
 85.5 Voffset=-0.2V

Voffset=0V

84.5 Voffset=0.2V

83.5

82.5

∆f
81.5
∆Φ

80.5
5 6
1x10 1x10
Frequency ( Hz )

Figure 4.4 Illustration of the nonlinear effect of the transconductance amplifier.

Rplate = 0.03Ω and Lplate = 15nH.

Rplate 0.01 0.03 0.1

(Ω ) (Low Freq. (Middle freq. (High Freq.
50KHz~120KHz) 120KHz~600KHz) 600KHz~1MHz)
Lplate 50 20 55 15 50 30
(nH) (low Φ ) (high Φ ) (low Φ ) (high Φ ) (low Φ ) (high Φ )

∆Φ 0.01 0.02 0.034 0.07 0.1 0.13

(degree)

∆Lplate 0.004 0.03 0.013 0.1 0.04 0.11

(nH)

∆f 2K 6K 17K 60K 72K 104K

(Hz)

∆Rplate 3E-4 6E-4 2E-3 4E-3 9E-3 1.3E-2

(Ω )

δ 0.025 0.053 0.06 0.16 0.089 0.13

Table 4.2 Minimum detectable changes in phase shift (∆Φ ) and minimum effective
changes in frequency (∆f) of the main circuit due to nonlinear effects
of the transconductance amplifier. The resolutions of Lplate (∆Lplate)
and Rplate (∆Rplate) are then derived from dividing ∆Φ and ∆f by the
differential of the curve in Figure 4.1 and k at the specified Lplate and
Rplate, respectively. Coefficient δis calculated by Eq. (4-2).

63
result, the variation of the coil resistance will change the d.c. biasing condition of

the transconductance amplifier and thus cause deviation in the a.c. performance.

The phase detector and the FVC do not influence the a.c. performance, but

the ripples at the d.c. output of the integrator restrict the resolution. The amplitude
of the ripples is 0.1mV, which can be converted to 0.0073§ and 45Hz via the

transfer functions of the phase detector and the FVC, respectively.

Summing up the effects drawn by the three sources causing the degradation

of the resolution, the finite output impedance and the nonlinear effect of the

transconductance amplifier, and the ripple at the integrator output, it is evident that

the resolution of the main circuit is dominated by the nonlinear effect of the

transconductance amplifier.

4.3 COMPARISON BETWEEN THE SUCCESSIVE AND THE ITERATIVE

FEEDBACK SCHEMES

4.3.1 Performance of the Successive Feedback

In Section 3.3, the coefficient which determines the frequency step, δ, was

not quantified. Now, since the minimum effective changes of the frequency have

been obtained by the simulation of the main circuit, δcan be given in Table 4.2 by

applying
∆f
δ= , (4-2)
f

where ∆f is as listed in Table 4.2 and f is the frequency at which the minimum

phase shift occurs for the specified Lplate and Rplate. It is observed that the smoother

the phase response to frequency is, the larger δis. Since the smoother phase

response to frequency corresponds to the longer distance between the sensor and

64
the metal target, it is clear that there is a tradeoff between the resolution of the

detection for the shorter distance and the maximum detectable proximity if a fixed

δvalue is chosen.

Rplate 0.01 0.03 0.1 Simulation

(Ω ) (Low Freq. (Middle freq. (High Freq. method
50KHz~120KHz) 120KHz~600KHz) 600KHz~1MHz)
Lplate 50 20 55 15 50 30
(nH) (low Φ ) (high Φ ) (low Φ ) (high Φ ) (low Φ ) (high Φ )

Φ of Sensor a.c.
with Ideal 43.388§ 78.321§ 30.453§ 81.691§ 43.388§ 70.255§ analysis
Input Stage
(Section 4.1)
Φ of Main a.c.
Circuit†† 43.388§ 78.321§ 30.453§ 81.68§ 43.389§ 70.257§ analysis
(Section 4.2)
Accuracy of Φ
of Successive 0.73% -1.5% 1.0% * 2.6% * transient
Feedback analysis
Scheme
Accuracy of Φ transient
of Iterative -0.79% -5.6% 0.30% 1.8% 0.89% -3.6% analysis
Feedback
Scheme
Resolution of
Lplate with 0.06 0.1 0.04 0.1 0.07 0.1 transient
Iterative analysis
Feedback
Scheme (nH)†

Table 4.3 Comparison of the accuracy of phase readouts of the successive and
iterative feedback schemes, where 85§ is taken as the full-scale
output signal. Φ is the minimum phase readout. * The circuit could
not sense a change in the polarity of the slope of the phase response
to the frequency in the scanning frequency range. †A refined VCO
with the level one model applied is used here. ††The phase readings
shown have been converted to Φ ′by Eq. ( 4-1 ).

Letting δbe fixed to 0.025, the desired δvalue for Lplate = 50nH and Rplate

= 0.01Ω , in the circuit shown in Figure 3.11, the successive feedback circuit is

simulated for the six conditions given in Table 4.2 through transient analysis. Table

4.3 and Table 4.4 list the accuracy of the phase and frequency outputs. Due to the
65
nonlinear characteristic of the VCO and the voltage level translator, the average

effective magnitudes of δare 0.025, 0.05 and 0.075 in the three frequency ranges

concerned, respectively, and are very close to the desirable values of δfor lower Φ

in Table 4.2. Therefore, the accuracy for lower Φ is within 2% while the change of

the polarity of the slope of the phase response to the frequency may never be

sensed in the scanning frequency range for higher Φ .

Rplate 0.01 0.03 0.1 Simulation

(Ω ) (Low Freq. (Middle freq. (High Freq. method
50KHz~120KHz) 120KHz~600KHz) 600KHz~1MHz)
Lplate 50 20 55 15 50 30
(nH) (low Φ ) (high Φ ) (low Φ ) (high Φ ) (low Φ ) (high Φ )

f of Sensor with a.c.

Ideal Input Stage 80K 98K 319K 368K 800K 754K analysis
(Hz)
(Section 4.1)
f of Main Circuit a.c.
(Hz) 80K 99K 321K 386K 812K 780K analysis
(Section 4.2)
Accuracy of f of
Successive 1.3% 1.4% 1.1% * -2.7% * transient
Feedback Scheme analysis
Accuracy of f of
Iterative 0% -4.9% 1.0% -25% -18% -22% transient
Feedback Scheme analysis
Variation of f of
Iterative ±2K ±4K ±18K ±7K ±15K ±30K transient
Feedback Scheme analysis
Resolution of
Rplate with 1.2E-3 4.8E-3 2.5E-3 2.9E-3 8.9E-3 9.0E-3 transient
Iterative analysis
Feedback Scheme
(Ω )†

Table 4.4 Comparison of the accuracy of the frequency readouts of the successive
and iterative feedback schemes, where 1MHz is taken as the full-scale
output signal. f is the frequency readout. * The circuit could not
sense a change in the polarity of the slope of the phase response to
the frequency in the scanning frequency range. †A refined VCO with
the level one model applied is used here.

66
4.3.2 Performance of the Iterative Feedback

δis desirable as a variable rather than as a constant in order to avoid the

tradeoff described in the previous section. Eq. (3-14) and Eq. (3-15) are thus

applied to determine the frequency steps. The accuracy of the iterative feedback

circuit is shown in Table 4.3 and Table 4.4 to compare with that of the successive

feedback scheme. Due to the 1mV decay at the outputs of the sample-and-hold

circuits during a clock cycle, there is a ±0.5mV variation at each phase readout,
which is 0.037§ after being converted by the transfer function of the phase

detector. This is the reason why the operation frequency still varies after

convergence is achieved. The frequency variations observed are listed in Table 4.4

and are generally lower than the minimum effective frequency steps in Table 4.1,

so this phenomena can be ignored. But the ±0.5mV variation at the phase readings

restricts the resolution of Lplate.

The iterative scheme is capable of extracting a smoother phase response

and generating a generally lower error compared to the successive scheme. But the

accuracy in the frequency output is still unsatisfactory for higher Φ because the

circuit converges at a much lower frequency than the desired value. Since the

phase response to the frequency is quite flat for higher Φ , the accuracy of the

phase reading does not reflect this phenomenon. One of the possible factors

contributing to this problem is the high THD of the sinusoidal signal at the output

of the sine-shaping circuit, which is frequency-dependent and, in turn, causes the

phase detector to be sensitive not only to the phase shift but also to the operating

67
frequency. Hence, a decrease in the phase shift due to an increase in the operating

frequency may be reflected as an increase at the phase detector output. Even

though a good sine-shaping circuit can be built by using the 0.8µm CMOS model,

the serious channel length modulation effect of the model limits the performance of

the VCO. Referring to the VCO configuration illustrated in Figure 3.1, the channel

length modulation effect destroys the equivalence among i1, i2 and i3 and also

makes the sourcing and sinking currents of the VCO capacitor, i2 and i3, vary with

the output voltage ( Vout ). In the time domain, the varying current results in a

nonlinear increase or decrease of the output voltage, while the unequal sourcing

and sinking currents generate asymmetric output waveforms. These phenomena

are reflected in the form of higher order harmonics in the Fourier series

representation of the output waveform of the sine-shaping circuit, which are then

amplified by the transformer with a gain higher than that for the fundamental

harmonic.

Another possible reason for the unsatisfactory accuracy is that the

amplitude of the desired signals at the secondary coil is so low (several millivolts)

that a significant error can be generated by the offset voltage of the comparator in

the phase detector. Thus an amplifier should to be added between the secondary

coil and the comparator, while another amplifier of the same design should also be

put before the other comparator in the phase detector to avoid any phase error

produced by the amplifiers.

If a simple level one model with very low channel-length modulation

parameters ( λp = λn = 0.005 ) is applied to the transistors in the VCO and two

68
amplifiers are added in the phase detector, the accuracy of the frequency and the

phase can be improved to be below 8% and 0.6%, respectively, for all the high Φ

conditions specified in Table 4.3 and Table 4.4. The resolution of the refined

interface circuitry is also listed in the tables and is approximately of the same order

as that shown in Table 4.2. Therefore, it can be concluded that the nonlinear

transfer function of the transconductance amplifier dominates the resolution.

The iterative feedback scheme is also superior to the successive one with

respect to sensor response time. The total number of clock cycles needed for the

successive feedback scheme to sense the minimum phase shift is proportional to

the frequency output and can be as high as fifty, compared with that of the iterative

feedback scheme which has an average of fifteen and does not depend on the

frequency output.

4.4 EFFECT OF VARIATION IN COIL RESISTANCE

Although the a.c. analysis of the two-coil sensing element has been shown

to be unchanged as the coil resistance increases, the sensor system is inclined to

lose sensing capability for coil resistance higher than 1Ω when the transient
analysis is performed in the interface circuit. This is a result of the output voltage

swing of the transconductance amplifier being higher than the range in which the

linear transfer function can be held if the coil resistance increases above 1Ω and the

output current is still kept higher than 200mA to excite the eddy current in the

target plate. For the higher coil resistance, say 200Ω , the reason for the loss of

sensitivity is simply that the output voltage swing exceeds the power supply

voltage. Therefore, we have shown that the circuit discussed is immune to process

69
deviations, which cause a resistance variation of several tens of percentage points,

but the circuit would require some redesign if it were to be used with a scaled

sensor having much higher resistance.

70
 Chapter 5: Summary and Conclusion

An interface circuit for the two-coil proximity sensor has been designed

and simulated using the 0.8µm CMOS model available through the MOSIS

service. Except for several bipolar transistors, which can be fabricated on the same

chip if a BiCMOS process is available, the whole circuit with the sensing element

can be integrated monolithically by the conventional CMOS process. The periodic

signal driving the primary coil is generated by an on-chip VCO and conditioned by

a transconductance amplifier. For the output stage, a phase detector and a FVC

extract the phase shift and the operation frequency, respectively, in the forms of

d.c. voltage. Feedback schemes are applied to make the operation frequency

converge to the value at which the minimum phase shift happens. The two

measurands, Lplate and Rplate, can thus be determined by the two d.c. voltage output

readings representing the minimum phase shift and the operation frequency at

which the minimum phase shift occurs, respectively. It is observed that the

accuracy can be significantly enhanced by applying the iterative feedback scheme.

However, the transformer structure of the sensing element amplifies the

undesirable higher order harmonics in the Fourier series representation of the input

signal so that the sinusoidal output signal is seriously distorted. The configuration

of the sinusoidal VCO needs to be refined or a feedback-controlled tunable

bandpass filter has to be added to the output stage to reduce this problem.

71
 Another main factor degrading the resolution and accuracy of the circuitry

is the nonlinear effect caused by the transconductance amplifier as mentioned in

Section 4.2. Further investigations are necessary to avoid the tradeoff between the

linearity of the transfer function and the power efficiency of the transconductance

amplifier.

An alternative way to more accurately evaluate the frequency step in the

iterative feedback loop is to convert the analog signals at the outputs of the FVC

and the phase detector into the digital domain by means of analog-to-digital

converters (ADCs) and digital-to-analog converters (DACs). The analog computer

described in Chapter 4 will be replaced by digital programming circuits, and hence

the nonlinear effect and offset voltage of the analog computing components, such

as the multifunction converters and difference amplifiers, for example, can be

eliminated.

The characteristic of the proximity sensor makes it possible to measure the

distance independent of the conductivity of the target material. The conductivity of

the target can also be identified by using multiple output readings. Further work

needs to be done to quantify the relationship between the two measurands, Lplate

and Rplate, and the two physical parameters, distance and conductivity of the target

plate.

A planar sensing element based on micromachining technology has been

fabricated with a mask design similar to the one described in Figure 2.9 [27]. The

silicon substrate underneath the transformer consisting of two inductors was

removed to eliminate the effect drawn by the parasitic capacitance and the resistive

72
loss. The coil resistance and inductance are 5.5Ω and 10nH, respectively. Based on

an approximate and reasonable assumption for scaling down the geometrical

dimension of the coils, the resistance of the target plate ( Rplate ) is scaled up and

the inductance of the target plate ( Lplate ) is scaled down proportionally to the

scale factor, if the target distance is scaled down as well. The coupling factor

between the two coils remains the same [8]. The a.c. analysis of the sensor model

in Figure 2.10 with the parameters specified above shows that the characteristic of

the phase response to the frequency remains approximately the same except that

the operation frequency range increases by the inverse square of the scale factor.

Therefore, the interface circuit can be applied to the micromachining sensor if the

transconductance amplifier is modified to provide more power and to keep zero

phase shift between the output current and input voltage. Also, the operation

frequency range of the FVC and the VCO needs to be enhanced by the inverse

square of the scale factor.

73
 Bibliography

[1] H. Clergeot, D. Placko and J. M. Detriche, Electrical Proximity Sensors,

Sensors and Sensory Systems for Advanced Robots, vol. F43, pp. 295-307,
1988.

[2] R. S. Popovic, J. A. Flanagan and P. A. Besse, The Future of Magnetic

Sensors, Sensors and Actuators, vol. A56, pp.39-55, 1996.

[3] D. Dixon and R. Hoehener, Inductive Prox: An Old Sensor with New
Wrinkles, Chilton’s Instruments and Control Systems, vol. 62, n 10, pp.
55-58, 1989.

[4] C. Loughlin, Proximity Sensors: Close to You, Sensor Review, vol. 8, n 4,

pp. 198-202, 1988.

[5] P. A. Passeraub, G. Rey-Mermet, P.A. Besse, H. Lorenz and R. S.

Popovic, Inductive Proximity Sensor with a Flat Coil and a New
Differential Relaxation Oscillator, Sensors and Actuators, A: Physical
Proceedings of the 1996 10th Eurosensors conference, part 1, pp. 122-126,
1996.

[6] A. Billat, A. Lecler, L. Hardy and G. Villermain-Lecolier, Study and

Application of a Flat Eddy-Current Proximity Sensor, IECON’87, pp.
1123-1128, 1987.

[7] L. Hardy, A. Billat and G. Villermain-Lecolier, Flat Eddy-Current Matrix

Sensor for Detecting Metallic Object, Sensors and Actuators, vol. A29,
pp. 13-19, 1991.

[8] P. A. Passeraub, P. Besse and R. S. Popovic, Scaling Down an Inductive

Proximity Sensor, Sensors and Actuators, vol. A52, pp.114-118, 1996.

[9] V. I. Gupta, Design of a Proximity Sensor Using Planar Inductors,

Compatible with Integrated Circuit Fabrication, Master Thesis of U. T.
Austin, 1995.

[10] H. M. Greenhouse, Design of planar rectangular microelectronic

inductors, IEEE Trans. Parts, Hybrids, Pachaging, vol. PHP-10, pp.101-
109, June 1974.

74
[11] E. Tuncer and D. P. Neikirk, Highly Accurate Quasi-Static Modeling of
Microstrip Lines Over Lossy Substrates, IEEE Microwave and Guided
Wave Letters, vol. 2, no. 10, pp.409-411, 1992.

[12] Lj. Ristic and R. Roop, Sensor Technology and Devices, Norwood, MA:
Artech House, 1994.

[13] R. R. Pal, H. N. Acharya, S. Kai and N.B. Chakrabarti, Voltage Controlled

Oscillators Using Complementary Current Mirror Inverter, Electronics
Letters, vol. 28, n 18, pp. 1701, 1992.

[14] B. Keeth, R. J. Baker and H. W. Li, CMOS Transconductor VCO with

Adjustable Operating and Center Frequencies, Electronics Letters, vol.31,
pp.1397, 1995.

[15] P. Horowitz and W. Hill, The Art of Electronics, Cambridge, England:

Cambridge University Press, 1989.

[16] P. Allen and D. Holberg, CMOS Analog Circuit Design, New York, NY:
Oxford University Press, Inc. , 1987.

[17] S. Franco, Design with Operational Amplifiers and Analog Integrated

Circuits, New York, NY: McGraw-Hill, Inc., 1988.

[18] H. Eaton and J. Lesho, A Low Voltage, Low Power, BiCMOS, Audio-
Frequency Voltage Controlled Oscillator with Sinusoidal Output, IEEE
Symposium in Circuits and Systems, vol. 2, pp. 1463-1466, 1993.

[19] D. Lee and D. Onsongs, Senior Project in Electrical Engineering

Department of U. T. Austin, 1996.

[20] C. S. Park and R. Schaumann, A High-Frequency CMOS Linear

Transconductance Element, IEEE Trans. on Circuit and Systems, vol.
CAS-33, n 11, pp. 1132-1138, 1986.

[21] P. Rossel, Power M.O.S. Devices, Microelectron. Reliab., vol. 24, n 2,

pp.339-366, 1984.

[22] B. J. Baliga, Power Semiconductor Devices for Variable-Frequency

Drives, Proceedings of the IEEE, vol. 82, n 8, pp.1112-1116, 1994.

75
[23] N. H. E. Weste and K. Eshraghian, Principles of CMOS VLSI Design: A
systems perspective, Addison-Wesley Publishing Company, 1992.

[24] R. M. Stitt and R. Burt, Get F-V Converters to Settle Fast Without Ripple,
Electronic Design, pp. 73-80, 1991.

[25] Voltage-to-Frequency and Frequency-to-Voltage Converter, VFC320 Data

Sheet, Burr-Brown Corporation.

[26] LH0094 Multifunction Converter, LH0094 Data Sheet, National

Semiconductor.

76
 Vita

Huei-Ching You was born in Taichung, Taiwan on June 24, 1974, the

daughter of Lih Liu and Hon-Fai Yau. After completing her work at National

Normal University Affiliated High school, Taipei, Taiwan, in 1992, she entered

National Tsing Hua University, Hsin Chu, Taiwan. In June, 1996, she received the

degree of Bachelor of Science in Electrical Engineering from National Tsing Hua

University. She entered the Graduate School at The University of Texas at Austin

in August, 1996.

Permanent address: 2F, No. 17, Aly. 5, Lane 171, Kun-Yang St.
Nang-Kung, Taipei 115
Taiwan

This thesis was typed by the author.

77