A universal oscillator analysis technique

During the last decades many different oscillator models and analysis techniques have been presented. The
negative impedance model, cavity type, DRO’s to the general feedback type, they have all one in common: As
oscillators are nonlinear positive feedback systems, their electrical behavior is very hard to predict. Modern
Harmonic Balance simulators are able to calculate parameters like oscillation frequency, power output, phase
noise or harmonic content, but they give little insight in oscillator function, during the design process. Many
circuit simulators have got analysis functions like oscillator-ports which mostly are specialized directional
couplers, but some parameters like termination impedances have to be estimated. This article will present a
method, with which it is possible to enlighten nearly any oscillator’s behavior, and so eases and shortens the
design procedure.

INTRODUCTION

Generally an oscillator can be seen as a positive feedback system. Usually it consists of an amplifier followed by
a resonator, which is fed back to the amplifiers input as shown in Fig.1.

RESONATOR

AMPLIFIER

Fig.1 Basic oscillator model.

The Amplifiers task is to withdraw the losses in the resonator, and the resonator does the frequency selection.
Now it is of interest how the components in Fig.1 should behave to raise proper oscillation. Fig.2 shows a
simplified mathematical representation of a feedback system. The amplifier is assumed to have no phase shift an
amplification of A and infinite input- and zero output impedance. The resonator is a simple phase shifter with an
amplification of B. The input signal I(j? ) is summed with the amplifiers output signal and fed into the phase
shifter. The output signal O(j? ) is the output of the phase shifter and the input of the amplifier.

I(j? ) + G(j? )=Bej? O(j? )

H(j? )=A

Fig.2 Simple mathematical model of the oscillator.

The transfer-function of the circuit in Fig.2 can then be stated as:

O ( j? ) B ?e j?
F ( j? ) ? ? (1)
I ( j? ) 1 ? A ? B ?e j?

where ? ? ? ?? ? ? f ?? ? as otherwise the oscillator‘s frequency is not defined.

The formula stated in [1] describes the behavior of a feedback loop by injecting a signal I(j? ) into the circuit
while observing the output O(j? ).
 So to get real oscillator conditions without injection the input signal I(j? ) has to fade to zero, while measuring
the output signal O(j? ). Mathematically stated as:

? B ? e j? ?
O( j? ) ? lim ?? I ( j? ) ?? (2)
I ( j? ) ? 0 1 ? A ? B ? e j?
? ?

It can be easily seen in (2) that if I(j? ) goes to zero the other term of multiplication must raise to infinity to get
an output which is different from zero. So A*B must be 1. Figure 3 shows the transfer-function stated in (1)
with A = 1 and B = 1, with respect to the phase shift ? .

25

20

15
Amplification [dB]

Fm 10

5
0 50 100 150 200 250 300 350 400
phim
Phase shift [degrees]

Fig.3 Transfer-function (dB) with A =1 and B = 1 dependent on phase shift ? (Degree).

It is easy retraceable in Figure 3 that a phase-shift of n x 360° raises the transfer-function to infinity.
For B = A = 1 (2) can be simplified to:

?? 1 ? ? ? 1? ?
O( j? ) ? lim ?? ?? j cot? ? ? ?? ? I ( j? ) ?? (3)
I ( j? ) ? 0
?? 2 ? 2 ? 2? ?
with the rule of De’Hospital and phase going to zero for oscillator condition the output can be calculated as:

?? 1 ? ? ? 1? ?
O ( j? ) ? lim ? ? j cot? ? ? ?? ? I ( j? ) ?? ? 1 (4)
0? ? 2
I ( j? ) ? 0 ,? ?
?? ? 2? 2? ?

Which is commonly known as the Barkhausen criterion, which states, that loop-amplification must be 1 and
loop-phase-shift multiples of 360° to raise oscillation.

TRANSFORMATION

The next question which arises is how numerous existing oscillator-circuits can be converted to a feedback loop
system. Stan Alechno showed in his article „The Virtual Ground in Oscillator Design - A Practical Example”,
that nearly every oscillator can be converted to a common-emitter circuit, by first connecting all ground and
power supply nodes, then letting them floating and then grounding the emitter. However, if feedback is
performed inside the transistor, the circuit can not be transformed.
 C1 C1 C1 L C
C C C
C2 C2 C2 C1
L RL L RL L RL RL C2

a) b) c) d)
Fig.4 Transformation of oscillator circuit: a) simplified common-collector oscillator, b) with floating ground c) with emitter grounded and d)
rearrangement of c)

Figure 4 shows an example, how a common-collector negative resistance oscillator can be transformed to a
common emitter circuit with the topology proposed in Figure 1. As it can be seen, the first thing of conversion in
a) is to connect all the ground and power-supply terminals together. This is possible, as power supply nodes
represent a AC-ground node, because AC-voltage is grounded by the large blocking capacitors between supply
voltage and the common ground. The next step is to remove ground from the circuit by letting all nets floating.
Now ground can be placed on every node in the circuit, on which it is advantageous for further analysis. In this
case the emitter of the bipolar transistor is grounded in c). If the circuit in c) is rearranged the circuit in d)
appears which shows a common-emitter circuit and a back fed resonator consisting of C1, C2, L and C.
This transformation becomes clearer, if the ground node is seen as an ordinary circuit node, which is in this case
for example the ground plane. The power supply as stated above is also an AC ground node. So it is possible to
think of any node in the circuit to be the ground node, and to re-sketch the circuit. But this transformation has
impact on components which are part of the bias-supply of the active device. As they have DC function, their
behavior changes, as the ground node is connected to another node. As seen in Figure 4a the load resistor RL is
connected between emitter and ground. A DC current through this resistor will result in a voltage drop over it,
and so in a voltage lift of the emitter and the base as well. A base bias resistor must take this voltage drop in
account. In Figure 4d the load resistor is connected to the transistors collector and the emitter is grounded. A
base bias resistor has now a much bigger voltage drop from power supply to the base as in the case of 4a) where
the supply-voltage to base-voltage is lowered through the voltage drop across the load resistor RL. So the bias
network has to be recalculated for equal operation of the circuits. An example later in this article will show the
difference between the two biasing cases. For further analysis the oscillator loop must be broken up, and a
transfer function can be recorded. A lot of oscillator theory also treating this problem is shown in Rheas book [4]
especially pp. 11-15 and chapter 2.

L C
P2

P1 C1
RL C2

Fig.5 oscillator from Fig.4 with open loop and input port P1 and output port P2

CALCULATION

The circuit in Fig.5 can be described by scattering parameters, but how should the ports P1 and P2 be
terminated? If they are standard 50 ? ports, this would not give the correct result for the transfer-function, as in
closed-loop condition the output of the oscillator is terminated with its input’s impedance and vice-versa.
As scattering parameters are defined for one characteristic impedance, it’s not easy to convert these parameters
to more generality. For the sake of understandability Z-Parameters will now be used for defining the problem.

The function:

Z ? TS ? Z ?S ? (5)
which converts S-Parameters to Z-Parameters with the identity:

? ?1 ? S11 ??1 ? S 22 ? ? S12 S 21 2S12 ?

? Z0 Z0 ?
Z? ?
?1 ? S11 ??1 ? S22 ? ? S12 S21 ?1 ? S11 ??1 ? S22 ? ? S12 S21 ? (6)
? 2S21 ?1 ? S11 ??1 ? S22 ? ? S12 S21 ?
? Z 0 ?1 ? S ??1 ? S ? ? S S Z0
?1 ? S11 ??1 ? S22 ? ? S12 S21 ??
? 11 22 12 21

will be used to convert the circuit from Fig.5 to the circuit in Fig.6, where Z2 indicates the correct oscillators
input impedance. More about two-port transformations and oscillators is shown in reference [2].

Z1 I1 1 Z11 Z12 2 I2
Z21 Z22
Vo V1 Z2 V2

Fig.6 Conventional view on the oscillator in open loop analysis

Following statements can be derived from Fig.6

V2 Z 2 ? Z 21
? (7) the forward voltage transfer-function
V1 Z11 ?Z 2 ? Z 22 ? ? Z12 Z 21

V1 Z1 ? Z12
? (8) the backward voltage transfer-function
V2 Z 22 ?Z1 ? Z11 ? ? Z12 Z 21

V1 Z 21Z12
? Z11 ? (9) the input impedance
I1 Z 2 ? Z 22

V2 Z Z
? Z 22 ? 21 12 (10) the output impedance
I2 Z1 ? Z11

From (7) to (10) it seems clear that all four parameters depend on the termination-impedances Z1 and Z2.
As calculated for the S-Parameter for correct oscillator function V2/V1 must be equal 1.
But as this transfer-function depends on Z2 and since its value is unknown, the transfer-function can not be
plotted correctly. If we assume for a while, that there is no reverse influence from the output (Z12 = 0) to the
input these parameters simplify to:

V2 Z 2 ? Z 21
?
V1 Z11 ?Z 2 ? Z 22 ?
(11) the forward voltage transfer-function

V1
? 0 (12) the backward voltage transfer-function
V2

V1
? Z11 (13) the input impedance
I1
V2
? Z 22 (14) the output impedance
I2

While (11) still depends on Z2 (13) and (14) are independent. Now a second view on the oscillator circuit in
Fig.7 brings more light into oscillator behavior.

Z11 Z12 2
1
Zin Z21 Z22
Zout

Fig.7 oscillator with feedback

As the output of the oscillator in Fig.7 is connected with its input, the impedance Z2 will result in V1/I1 = Z11.
This means that the oscillator looks back to its own input impedance. With this simplification and the feedback
in Fig.7, equation (11) reduces to:

V2 Z 21
? (15)
V1 ?Z11 ? Z 22 ?

With the oscillator condition (15) this will give:

! Z 21 ?? , Ps ?
1? (16)
?Z11 ?? , Ps ? ? Z 22 ?? , Ps ??
where it is noted that the Z-Parameters are a function of frequency and oscillator power Ps.

In the more general case, that if Z12 ? 0, equation (16) does not hold anymore, and a new way of determining the
impedances Z1 and Z2 must be gone formally. Oscillation can also be seen in the time domain as a wave
traveling circular through the oscillator. This means that the wave is traveling through the circuit in Fig.7 and is
fed back to its input. As the oscillator has group delay, the Barkausen criterion changes a little bit to:

? 0

?t 0
grf ?d? ? ? (0) ? n ?360? (17)

Where tgrf is the forward group delay and ? 0 the frequency of oscillation. Equation (17) just states that group
delay induces phase shift. That for the oscillator can be seen as a non-reciprocal transmission line. As continuous
reflection occurs in the oscillator through not correctly terminated impedances, a wave traveling back in the
oscillator is induced. But as the transistor has low backward amplification, this reverse wave is heavily damped.
In the case of a S12 = 0 in the transistor, the reverse wave is totally blocked by the transistor and the impedances
can be calculated as in equations (13) and (14). This case is very similar to an ideal distributed isolator.

V
MARK I1 1 2 I2 V
MARK

Oscillator
V1 V2
t=t1

t=t1+tgrf

Fig.8 Oscillator in the time domain.

If we imagine now that we could mark a certain voltage zero transition in the traveling wave at the input, and if it
were possible to follow it, we could see that the zero transition comes out of the oscillator circuit with a delay of
tgrf, as shown in Fig.8. If we assume that we already know the correct termination impedance we could state
following: The oscillator circuit from Figure 6 terminated with the correct impedance Z2, has the same input
impedance Zin=V1/I1 as the closed loop oscillator. So the traveling wave out of the oscillator will have the same
conditions of input impedances if it were fed back to the oscillator input or another oscillator circuit input, which
is terminated with the correct impedance Z2.

Zi1
I1 1 2 I2 I1 1 2 I2
Zi2
Oscillator Oscillator
V1 V2 V1 V2 Z2

Fig.9 Equal input impedances Zi1 and Zi2.

Figure 9 shows the equal input conditions. The impedance Zi1 stands for the closed loop impedance and Zi2
for the input impedance of another oscillator circuit with the same S-parameters, terminated with the correct
impedance Z2. As stated above Zi1 must be equal to Zi2. As the second oscillator circuit is of the same type as the
first one, the Barkhausen criterion is also fulfilled for the two oscillator circuits in series, as the second oscillator
is terminated with the correct impedance Z2. As we assumed that the impedance Z2 is a replacement of the input
impedance of the closed loop oscillator, the loop can be closed, as shown in Figure 10.

I1 1 2 I2 I3 1 2 I4
Oscillator Oscillator
V1 V2

Fig.10 Closed loop with two oscillators.

Because the oscillators in Figure 10 are from the same type and the Barkhausen criterion is still fulfilled V1 must
be equal to V2 and I1 must be equal to I3. Now it can be easily seen that an infinite number of oscillator blocks
can be inserted in the circuit of Figure 10, by keeping the input currents and the node voltages equal. In this case,
the wave would travel through an infinite number of circuits, which is essentially the same as traveling circular
through one closed loop oscillator circuit. The infinite oscillator block chain circuit can now be analyzed to get
the correct termination impedance. That for it is very advantageous to use ABCD parameters, as chaining two-
ports can be accomplished by matrix multiplication. A new transform function will be introduced, to convert the
oscillator’s open loop S-parameters to ABCD parameters as shown in equation (18)

A ? TS ? A ?S ? (18)
So the two two-ports of Fig.10 transform to its ABCD equivalents A1 and A2. Now the combined Matrix is just:

Atotal ? A1 ? A2 (19)

infinite blocks
Z1 A11 A12 A11 A12 A11 A12 between A11 A12
1 21 21 2 21 2
A21 A22 A21 A22 A21 A22 A21 A22
Vo Z2

Fig.11 oscillator with infinite inserted two-ports A and broken up loop.

To get some insight into the infinite circuit in Fig.11 the Z-matrix of the infinite chained oscillator A-matrices
will be calculated as:
Z o ? lim T A ? ? ?? ? ???? ZZ
n ? ? A? Z
n o11

o 21
Z o12 ?
? (20)
Z o 22 ??

This means that an infinite chain of A matrices can be chained mathematically as an infinite matrix-
multiplication with it self stated as An, which stands for A1 x A1..... x An It is very interesting that the coefficients
Zo11 and Zo22 converge to a fixed value which will be analytically derived later. As an amplifier has usually a
backward transfer-function lower than 1, which means forward gain and backward isolation, the parameter Zo12
goes to zero, so with the equations (11)-(14), equation (20) reduces to:

?Z 0 ?
Z o ? ?? o1 ? (21)
? Z o 21 Z o 2 ??

That means that Zo1 is the input impedance and Zo2 the output impedance of the network in Fig.11.
With the determination of the impedances Zo1 and Zo2 the voltage-transfer-function can be defined and set to 1.
For oscillator analysis like in Figure 6, Z1 will be equal to Zo2, which actually won’t be used in oscillator forward
analysis, and Z2 will be equal to Zo1, which is the interesting part for the transfer-function. If the loop in Fig.10 is
closed and broken up anywhere again, the calculated Zo Matrix will have the same coefficients. If the input
impedance of an infinite chain of oscillator two-ports is Zo1 , it will be still Zo1 if only one two-port is removed as
there will still remain an infinite chain of oscillators. This shows that the input impedance of one single
oscillator two-port must be Zo1 if it’s output is terminated with Zo1 as shown in Figure 12.

infinite blocks
Zo1 A11 A12 A11 A12 A11 A12 between A11 A12
1 21 21 2 21 2
A21 A22 A21 A22 A21 A22 A21 A22
Zo1

Zo1 A11 A12

1 2
A21 A22
Zo1

Fig.12 Two-port input impedance with termination of port 2 with Zo1.

From this approach it seems clear that the oscillators closed loop impedances are Zo1 and Zo2.
To calculate the impedances Zo1 and Zo2 directly, the lower sketch in Fig.11 can be used with equation (9) and
(10) to set up a solution. The definition out of (9) and (10) will be then.

Z 21Z12 Z 21Z12
Z o1 ? Z11 ? and Z o 2 ? Z 22 ? (22)
Z 01 ? Z 22 Z 02 ? Z11

If the definition in (22) is transformed with respect to Zo1 and Zo2, two quadratic equations will appear.
The solution of these equations will be:

1?
Z o1(1, 2 ) ? ? Z11 ? Z 22 ? ?Z11 ? Z 22 ? ? 4 ? det ?Z ? ?? (23)
2

2? ?

1?
Z o 2(1, 2 ) ? ? Z 22 ? Z11 ? ?Z 22 ? Z11 ? ? 4 ? det ?Z ? ?? (24)
2

2? ?

As it can be seen from equations (23) and (24) four solutions are possible, but only two of them are meaningful.
The one with the positive real part is the preferred one, as the impedances should be passive in feedback designs.
The main difference between the oscillator and the transmission line is that the oscillator contains an active
device, in which the input power is amplified with some gain. So it is possible to get reflections and |S21| = 1,
which is not possible in ideal transmission lines as there S11 must be zero for lossless operations. These
reflections in the oscillators travel backwards and have their own propagation properties. Through the reverse
isolation of the oscillator, this reverse wave is damped, but not extinguished. As this wave is generated in every
two-port in Figure 11, it influences the input impedance of the oscillator, which has been taken into account in
the equations (23) and (24). The impedances Z01 and Z02 from equation (21) can bee seen as the characteristic
impedances of the traveling waves (forward and reverse), as the node voltages and input currents are equal over
the whole infinite two-port chain in Figure 11 like in Figure 10.

SIMULATION

The oscillator from Figure 4a should be designed for an oscillator-frequency of 2.9 GHz and a loop-power of 0
dBm. The bipolar transistor will be a BFP 420 from Infineon.

+5V

RB

Lb
C3
BF P420 Cb
C1
C3
RE RV

L C2

Lb
RL

Fig.13 Investigated oscillator circuit.

The resistors RB and RE are for biasing, while Lb are AC- and Cb DC-blocking components. RV is a coupling
resistor to prevent oscillation from quenching, through a too low output load. The capacitor C3 has been
introduced to get more flexibility in frequency adjustment. After the transformation of the circuit in Figure 13,
as described in Figure 4, the circuit can be drawn like in Figure 14.

2
REST
ID=IN1
R=100 Ohm
T

T=23.85 DegC
SRLC
SUBCKT ID=RLC1
PORT ID=S2 1 R=1 Ohm
P=1 NET="SP0" L=5.3 nH
Z=50 Ohm C=0.79 pF
1 2

CAP PORT
ID=C5 P=2
C=0.3 pF Z=50 Ohm
CAP CAP
ID=C1 ID=C3
C=8.37 pF C=2.245 pF

Fig.14 Transformation of the circuit in Fig.12

The subcircuit-block “SUBCKT” in Figure 14 is a replacement of a BFP-420 transistors common-emitter large-

signal-S-parameters for a input power of 0 dBm. Some harmonic-balance-simulations have been done with the
pure transistor amplifier to get the different large-signal-parameters at certain power levels. These large signal
parameters are only a modeling help, but they do not really exist by definition of scattering parameters, as
through the strong non-linearities of the active device many harmonics will be generated, and the parameter are
also dependent on the used power, so this part of the modeling will produce the largest error in the transfer-
function behavior. The resistor with 100 Ohms replaces the sum of Rv + RL in Figure 12. Blocking components
like Lb and Cb have been removed, as they have no influence on the oscillator analysis. Next the transfer-
function is calculated using Z-Parameters with the result of equation (21) put into (7) and plotted over frequency
as shown in Figure 15.
NZ11 = Schematic 1:Z[1,1]
NZ12 = Schematic 1:Z[1,2]
NZ21 = Schematic 1:Z[2,1]
NZ22 = Schematic 1:Z[2,2]

ZK = 0.5*(NZ11-NZ22+sqrt((NZ11-NZ22)*(NZ11-NZ22)+4*(NZ11*NZ22-NZ12*NZ21) ))
VK=(ZK*NZ21)/(NZ11*(ZK+NZ22)-NZ12*NZ21)

Table 1: Formulas to calculate the voltage-transfer-function VK in Microwave Office

Transfer-Function
10 150
2.901 GHz
-0.3581 dB
0 90
attenuation [dB]

DB(|Eqn|) (L)

phase shift [°]

-10 transfer-function 30
Ang(Eqn) (R, Deg)
transfer-function 2.901 GHz
-20 -30
0.01455 Deg

-30 -90

-40 -150
0.5 1.5 2.5 3.5 4
Frequency (GHz)

Fig.15 Plot of the transfer-function

A good proof, that the plotted transfer-function is correct, is to move components in Figure 13 from the out put
to the input. As the oscillator-analysis should be independent on the resonator location, the position of amplifier
and resonator can be changed without altering the transfer-function. Even partial resonator components can be
transferred from input to output and vice-versa without effect. With the component-values from Figure 14, the
transfer-function in Fig.15 is evaluated. It can be extracted that the phase shift crosses zero at 2.901 GHz while
the magnitude is –0.358 dB at the same frequency. To verify the results in Figure 15, a harmonic balance
analysis has been set up and simulated.
 DCVS PORT
ID= V1 P=2
V=5 V Z=50 Ohm

2
REST
IND
ID=IN2 ID=L4
R=1.2e4 Ohm L=0.58 nH

T
T=23.85 DegC

IND
ID=L1
CAP L=0.56 nH
ID=C2
C=0.0069 pF
IND
P_METER3
ID=L2
CAP ID=P1
ID=C5 L=1e6 nH 2 C
IND IND
C=0.3 pF W ID=L6 ID=L2
PORT 2 C
1 I 2 1 P=1 L=0.53 nH L=0.47 nH
GBJT
Z=50 Ohm ID=GP_BFP420_1
B CAP
V ID=C1 1 4 CAP CJE=0.001806 pF
SUBCKT ID=C1
ID=S1 C=1e6 pF S VJE=0.8051 V
3 E B C=0.134 pF MJE=0.4658
CAP 3 NET="Schematic 7"
ID=C6 VTF=0.2379 V
C=2.245 pF 2 E ITF=1 mA
3
REST
SRLC ID=IN3
ID= RLC2 R=250 Ohm
CAP
T

R= 1 Ohm T=23.85 DegC

ID=C3
L= 5.3 nH
C= 0.79 pF C=0.136 pF
2 IND
ID=L3
CAP 1
ID=C7 REST L=0.23 nH
ID=IN1
C=8.37 pF R= 50 Ohm
T

T=23.85 DegC
A OSCAPROBE
IND ID= X2
ID=L1 Fstart=2 GHz
1
L=1e6 nH Fend=3.5 GHz
Fsteps=500
Vsteps=200 IND
ID=L5
L=0.05 nH

PORT
P=1 PORT
Z=50 Ohm P=3
Z=50 Ohm

Fig.16 Oscillator circuit prepared for harmonic-balance simulation using Microwave-Office

One thing on which one must take care is the different transistor biasing in Figure 16 compared to a common-
emitter circuit. It is very important that collector-current and collector-emitter voltage stay equal to the common
emitter large-signal equivalent circuit. On the right side of Fig. 16 the basic SPICE-model of the bipolar
transistor is presented with its package parasitic components.

The circuit in Fig.16 gives following result compared to the transfer-function approach

transfer-function harmonic-balance Error

Frequency 2.9 GHz 2.9078 GHz 0.3 %
Power 0 dBm (in base) -0.2 dBm (in base) 0.2 dB

Table 2: Comparison of the transfer-function-method with harmonic-balance-simulation

CONCLUSION

It has been shown, that for many types of oscillators which can be converted to a feedback system, an accurate
estimation of oscillator-frequency and output-power is possible. Through linking an infinite number of oscillator
open-loop circuits the true input- and output termination impedances are defined. With the formulas (23) and
(24) these impedances can be calculated, and inserted in an open loop analysis with a single oscillator circuit.
With this result a transfer-function can be plotted, in which the oscillation points can be determined by general
oscillator conditions (gain = 1 and phase-shift = n x 360°). Using small-signal parameters the start-up condition
can be checked, and with the transistor’s large-signal-S-parameters at a certain input- or output-power, the
oscillator can be designed the way that the true loop power will meet the large-signal estimation. As the large-
signal-S-parameters of the transistor are measured without the resonator circuit, the different termination of
harmonics as well as unknown reflections of the different harmonics and the-fundamental will influence the
termination impedances. Further the model which describes the active device must be accurate enough, as
otherwise the results of the transfer-function approach will meet the harmonic-balance simulation results quite
good, but probably not the physical implementation. Good parasitic- and nonlinear-modeling are obligatory. The
comparison in the table above shows that the estimation-error will be less than 1% for frequency and less 0.5 dB
for power, which will be satisfying for the most applications.
REFERENCE

1. S. Alechno, „The Virtual Ground in Oscillator Design - A Practical Example”, Applied Microwave &
Wireless, July 1999.

2. George D. Vendelin et al., “ Microwave Circuit Design, Using Linear and Nonlinear Techniques”, Wiley
Interscience, 1990

3. AWR Research, Microwave Office 2002, educational version.

4. Randall W. Rhea, ”Oscillator Design And Computer Simulation”, Noble Publishing, Atlanta 1995