Nat io1 Mroautits a*4 SAdiniitration

Co*tract .

ExtsraI Publication $o. 673

AczJIsITI AND TRJ((D BENAVICE

I I4LbSELKED LOON

A J. Vit.rbi

CASE FILE
COPY

CopyNo._
pp. iii, j_3

JET pROPULSION LABORATORY

Calif omit Instiwte of Technology
Pasadena 3, California
July 14, 199

t
 S

r •-

Extemal Puklication No. 613

coEKrs
Page

I. Introduction . . . . . . . . . . . . . . . . . . . . . . 1

II. Basic Operation . . . . . . . . . . . . . . . . . . . . 2

III. First-Order Loop With Constant Frequency Input . . . . . 4
IV. Second-Order Loop With Constant Frequency Input . . . . 6
V. Second-Order Loop With Linearly Varying Frequency
Input . . . . . . . . . . . . . - . . . . . . . . . . . . 15
VI. Secon&Order Loop With Imperfect Integrator . . . . . . 21
VII. ThirdOrder Loop Tracking Behavior . . . . . . . . . . . 26
A VIII. Noise and Loop-Noise Bandwidth . . . . . . . . . . . . . 30
IX. SuaryofResults. . . .. . .. .-. . . . .. . 55. 34

onclature . . . . . . . . . . . . . . . . . . . . . . . . . 38
References. . . . . . . . . . . . . . . . . . . • . • • • • • 40

F igures . a • • • • . . • • • • • • .............41

krtow1edgeaent . . . . . . . . . . . . • . • • . . . . . . . . 64

LFL
AAL

Page ii
 I. --

e# ropuJsion Laboratory
Laifiei- .i2.

_________ first line of. text

______ Oia.;sified - Title_or first iin cf t.exi

1. Phase - Locked Loop ...... . . . . . . . . . . . . . 4J.

2. First-Order Loop Pull-in Behavior (n even integer) . 41
3. Basic Analog Computer Mechanization . . ....... 41
4. = 1.414 (overdamped) ............... 42
5. 1.0 (critically damped) .. . . .......... 43
6. = 0.707 ........ - . . . . .......... 44
7. = 0.5 ....................... • 45
8. Limit of Frequency-Lock as a Function of Damping . 46
9. = 0.707, D/a = 1/4, sin (D. /u = 14.5°
= O.08,r rad * . . . . . . . . . . . ......... 47
10. = 0.707, 1/2, sin 1 (D/c) = 30.00

= /6 rad . . . . . . . .
7r, . . . . . . . • • • ...... 48
1
U. = 0.707, = ii72 = 0.867, sin 1 (/ci) = 60° -
.ulf = 17/3 rad . . . . . . . . . . . . . . . . . . . . 49
12. = 0.707, D/a = 0.95, sin 1 (D/c) = 72°
rad . . . . . . . . . . . . . . . . . . . . . . . 50
13 D/a 0.985,sin(D/c)=80.0°

14. - 0.707, a/2 n - 0.1, aQ/co - 0.4 . . . . . . . . . . 52

15. 0.707, a/2o = 01, czQ/c - 0.6, Q/oy - -'T . . . 53
16. = 0.707, u/2t = 0.1, uQ/o 0.7,
= 7/2 ./•• . . . . . . . . . , . . . . . . . . . . . 54
0.707, a/2cfi = 0.1, = 0.8, Q/O)g = 4/T . . . 55
18. = 0.707, /2c = 0.1, aQ/o = 0.9,
• Lse rine c' tet . . • • • .. . 56

1J$iiTFICXi2i (if any)

 9

Jet Propuson Laboratory

83Sified ReportprJ1
Unclassifi IG1
Classified - Tit.
rage
19. Relative Positions of Signal, Center, and Initial VCO
Frequencies . . . . . . . . . . . . . . . . . . . . . ..
. 57
20. 0. 707, D/co 1/2 . . . . . . . . . . . . . . . . .
. 58
21. t = 0. 707, 1.0 . . . . . . . . . . . . . . . . .
• 59
22. = 0.707, D/co = 5/4 . a . . • • . . . . . . . . . . .
. 60
23. 0.707,D/O)=3/2 . S SO •
• 61
24 . 0 • 707, D/a 7/4 • • • . . . . . . • . . • . • . .
. 62
25. = 0.707 • • • • • • . • • . • • • • • . . • .
. . . . . 63
26. = 0.5 • • • • • • • • • • • • • • • . • . • .
. S • . 63
27. 1.0 • . . . . . • • . • . . . • • • . . .
. . • . . . 63

b4
 External PublicM j onNc. 673

I. INTROIXCTI cR4

1'tase-locked or A loops have found increasing applications in

recent years as tracking filters, synchronizing devices, and narrow-

band FM discriminators. Considerable work has been performed to

determine the noise-squelching properties of the loop when it is
operating in or near phase lock and is functioning as a linear
coherent detector (Refs. I and 2). However, insufficient consideration
has been devoted to the non-linear behavior of the loop when it is out
of lock and in the process of pulling in. Experimental evidence has
indicated that there is a strong tendency for phase-locked loops to
achieve lock under most circumstances. However, the analysis
which
has appeared in the literature (Ref s. 3, 4, and 5) is limited to the
acquisition of a constant frequency reference signal with only one
phase-locked loop filter configuration.
This work represents an investigation of frequency acquisition
properties of phase-locked loops for a variety of reference-signal
behavior and loop configurations. Results are obtained concerning
the frequency pull-in and tracking behavior (for both constant and
linearly varying reference frequencies) of the following loop filter
transfer functions:

'This paper presents the results of one phase of research

carried out at the Jet Propulsion Laboratory, California Institute
of Technology, under Contract No. NASw-6, sponsored by the National
Aeronautics and Space Administration.

Page 2.
 External Publication No. 673

I. K (a wide-band amplifier or resistive pad)

2. K fi. + a/p ] (a single integration)

3. K[(p + a)/(p+a(a simple passive RC filter or

imperfect integrator)
4. K [1 + a/p + b/p 2 ](a double integration)

The general approach involves solution of the non-linear

differential equation which describes the system behavior, using

analog and graphical methods. The terminology adopted in the text

is the one in general use for closed-loop systems and servomechanisms.

However, the application of phase-locked loops to the filtering of

noisy signals has generated an alternate terminology in which the loop

- noise-bandwidth is the key parameter (Ref. i). The results will be

derived in terms of both sets of parameters and are summarized in

Sec. IX.

II. BASIC OPERATIJ

Figure 1 represents the loop configuration. The frequency of

the input reference signal, c 5 , may be either constant or time varying.

The frequency of the output of the voltage-controlled oscillator (YCO)

consists of a constant equal to the center frequency of the oscillator,

c plus a time-varying term proportional to the actuating signal.
The VCO output is 90 deg out of phase with . the input reference so

that the multiplier output is proportional to the sum of two sinusoids

of frequencies equal to the sum and difference of the two input
frequencies.

Page 2
 External Publication No. 673

Since the low-pass filter following the multiplier will not pass
tne sum frequency term, the multiplier output, ej, is proportional to

the sine of the difference phase, p. The linear transfer function of

the loop filter shall be termed K 1 F(p) where F(0) 1 and operational

notation is used throughout. Then

dc2 - K0K1K2F(p) sin

edK(jsinp= 2 sin (t) - p] (i)

Clearly, the proportionality constant, is one— half the product of

the input and return signal-amplitudes. Thus,

sin p = sin [ q's - )ct - KF(p) (2).

p

where K = 1(0 K1 K2 shall be defined as the loop gain. Differentiating

the arguments of both sides of Eq. (2) yields

+ KF(p) sin • p = - (3)

This then is the general equation for phase error of the phase-locked

loop. At this point the following approximation is generally invoked:

sinppforp<< Irad (4)

which permits linear analysis to be performed and particularly the

small-signal response of the system to noisy inputs to be considered.

To determine the pull-In behavior of the loop, however, the non-linear

Eq. (3) must be solved.

Page 3
 External Publication No 673

* 111. FIRSTORDER LOOP WITH Ca4STAJJT FREQUENc y

INIJT

The first and most simple loop which shall be Considered is one
in which the loop filter has no energy storage elements and thus

produces a constant transfer ratio (see Refs. 3 and 5). The reference
input shall be a constant frequency sinusoid. Thus Eq. (3) becomes
the first—order differential equation:

(5)

where

Q = - = Constant

'. Figure 2 shows

plotted against p. If the VCO is initially at its
center frequency, (o) Q and (o) = rnr (where n is an integer).
If the frequency error
is positive, the phase error p tends to
increese; for negative p, p decreases. Hence, if n is an even integei
•
1and Q <K, the system will travel along the sinusoidal trejectoy of
Fig. 2 until it reaches the p axis at p sin (Q/K) + rur. This is
stable point; can not b•co negative because , would then
tend to
decrease and return the system to the p axis. If n is an odd integer,
the system will go through a larger part of the sinusoidal trajectory
until it reaches a stable point at

(n + l)ir +

If Q >K, however, the trajectory never crosses the p axis and a

Page 4
 Externaj. Pu .licatjon 14*. 673

• stable or phase-lock point is never reached.. Thus the pull-in

frequency range of the loop is umax = K rad/sec. For the first-order

loop K is also equal to the closed-loop 3 db bandwidth.

When phase-lock does occur, the steady-state phase error is
shown to be sin-Q/K). The linear analysis which follows from the
approximation (4) arrives at a steady-state error Q/K, which is
clearly an accurate estimate when Q/K< < I.

Pull-in time may also be determined from Eq. (5). Since

= c/dt, this can be rewritten as

dt 1
d, Q-Ksinp

-- :F'" the pull-in time is given by

t = 1fina1 Q - K sin (6)

initial

fij cannot be taken as sin (w4K) + mr since for this value the

demoninator becomes zero and the integral becomes infinite. This is

correct since, in fact, the time required to reach the steady state
is infinite. To get a meaningful result 'final can be taken as
slightly less than the steady-state phase and the time required to
reach it can be evaluated. The integral is expressed in terms of
elementary functions in the standard tables.
The conclusions to be drawn regarding the first-order loop are
that it will acquire lock within one cycle provided the initial

Page5 I
 External Publication No. 673

frequency error is less than the 3 db loop bandwidth and that the
lock-in time is governed by the initial phase error.

IV. SECOND-ORDER LOOP WITH CONSTANT FREQUENCY INPUT

To overcome the limited pull-in range of the first order ioop,

and to provide for tracking linearly varying frequencies with narrow
bandwidths, an integrator is generally used in the loop filtez. Then

F(p) = I + - (7)
p

and Eq. (3) becomes

+ aK
xp+ (sin)Q (8)

Letting aK and K = 2C a and differentiating, the second-order

differential equation which results is

+
+ O) (sin ) sin , 0 (9)

Th. characteristic polynominal when the equation is linearized by

approximation (4) becomes

P2+ 2Z(0nP+(o 2= 0 (10)

In servo terminology co is the undamped natural frequency and

the damping factor for the linearized servo loop.

Page 6
 External Publication No. 673

Rewriting Eq. (9) without operational notation:

(11)

Making the substitution

t= (12)

2 2 cos+con2 sinO (13)

ci

Normalizing and letting

=.= I (14)
dt

(15)

Thus one of the parameters has been eliminated by normalizing the

time parameter. It would now be of interest to construct a plot
of as a function of p as was done in the previous Section. To this
aim let p x (the abscissa) and = y (the ordinate). This requires
that y = c and . Then Eq. (is) becomes

y+ycos x+sin.xuIo

or

Page 7
 External Publication No. 673

i=x=_cosx_sinx (16)
y x dx 42y

which relates the slope of the system trajectories to the instantaneous

frequency and phase errors and permits their plotting. Certain

observations can be made based on Eq. (16):

1. For large y (which corresponds to large frequency

error) the second term on the right becomes small and

the trajectories become nearly sinusoidal.

2. For x = 0 (, = 0), the slope is always - 1. Hence the

y axis is the - I isocline (line of constant slope).

The overall pattern can be determined by finding other

isoclines which in general will not be straight lines.

3. The equation is periodic in x with period 27r; that is,

the slope of the trajectories is the same at (x 0 , y0)

as it is at (x 0 + 2nTr, y0 ). Hence, to describe the

behavior it is only necessary to plot the trajectories

for - 7r< x< Tr.

4.. At y = 0, x = ± nw (n an integer or zero) the slope

dy/dx is indeterminate since the second term on the

right becomes 0/0. These points are called singu-

larities and are either stable points of the system

or centers of instability as will be discussed more

fully below.

The graphical procedure just.described for finding solutions of

the non-linear differential Eq. (15) is the so-called Nphase_planeN

method (Ref s. 6 and 7). The procedure of determining the isoclines

Page 8
 External PublicatioA No. 673.

and sketching the various trajectories is a long and tedious process,

however. On the other hand, Eq. (15) is easily programmed for an

analog computer as is shown in the simplified diagram of Fig. 3. The

voltage representing p is used to actuate the x axis of a plotter and

that representing goes to the y axis.

This results in the plots of Figs. 4 through 7 for various values

of . The = x axis extends from - ir to + ir. The following obser-

vations may be made from the figures:

1. For large values of y (positive and negative) the
trajectories are practically sinusoids. The trajectories

are traversed from left to right in the upper half

plane and vice versa in the lower. For large positive

y a small decay may be noted as the trajectory traverses

the strip from p - ir and p IT. The rate of decay

becomes greater as y decreases. Since the behavior

is periodic in each strip of width, 2,r, the decay will

continue and increase in each successive strip until at

x p = nir (where n is some odd integer) the value of

y will be below the line A - A. At this point the

system will stop skipping cycles and the phase and

frequency error will decay in toward 0, p = (n + 1),r

and phase-lock will have been achieved. The same

behavior ensues when the initial frequency error is

negative except that the movement of the trajectories

is from right to left. The system may be said to be in

frequency-lock when the errors lie within the lines

Page 9
 External Publication No. 673

A - A in a particular strip since when this condition

exists the loop stops skipping cycles.
2. The singular points at x * nit (n = 0, 1, 2, .1.) are
indeed points of stability or centers of instability
as mentioned above. When n is even there are stable
points where the ioop achieves phase-lock. These are
called nodes or foci depending upon whether the system
is overdamped (Fig. 4) or underdamped (Figs. 6 and 7).
At odd multiples of ii- there is a center of instability
known as a saddle point. Near this point, no matter
what the direction of the trajectories may. be , the
magnitude of the frequency error will decrease until
the saddle point is almost reached and then quickly
increase again. Further properties of the singular
points may be found in Refs. 6 and 7 on phase-plane
techniques, but these are not necessary for the
discussion.
3. It can be seen by comparing the value of y for which
the line A - A crosses the point x = - it on the
various figures that the region of frequency-lock is
extended as decreases. This, however, is a con-
sequence of the normalization of Eq. (12). Letting
the true radian frequency dcp/dt = co rad/sec, and using
Eq. (14) there results

o2ta2Zciy (17)

Figure 8 is a plot of the limiting value y and w/c&

Page 10
 — -'r'-t- e —','n- '---r-- -

Externa1 Publication No. 673

for frequency-lock as a function of obtained using

the analog computer. Clearly, the more damped the

system the wider is the region of frequency-lock for

a given loop natural frequency.

The qualitative intuition obtained by examining the phase-plane

plots may be put on a somewhat more quantitative basis. First, it

will be shown that the pull-in range is theoretically infinite. Then

approximate expressions will be derived for the frequency error decay

per cycle and the. pull-in time.

If both sides of Eq. (16) are multiplied by y and integrated

between the limits — ir and + r the result is

A7T
[y2 — 2c)] = — i
y cos x dx — 1 sin x dx (18)
2 4 2
.cy/7r J-ir

• The second term on the right is zero while the first may be
integrated by parts to yield:

[() — ()]
f sin x dy (19)

If the expression for dy from Eq. (16) is substituted into Eq. (19)
there results:

[2(ir) — Y2 ( 7T
sin x cos x dx — .-.i I
TT
dx (20)
= — /
4 ./-7T y

Page 1.1.
 External Ptèlicatioxi No. 673

• The first term is clearly zero and the second. can be rewritten as

2[
y2 (_,r)1 = - 1 - cos 2x dx (21)
J y

Clearly, for positive y the integrand is positive for all values of x

between the limits *ir, making the right side of Eq. (21) negative.
Therefore, during each cycle of width 2,r, the value of y must decrease
for all initial values of y, which implies that the pull-in range is
infinite when a perfect integrator is used in the loop filter.
To determine approximately the decrease in y per cycle of x it
may be assumed that for y above the line A - A the rate of decrease
per cycle is constant. This approximation is, of course, best for

large y. Then,

y(x) y(-w) - sin x - x (22)

for -r< x < ir where by/2w and by is the decrease per cycle.

Substitution into Eq. (16) yields

sin x
= - cos x -
dx - sin x - Px 1
42 y(ir)
L y(-r) •(y(-ir)i

If

- by = y(ir) - y(-ir)
2
-1
f-7r
____
(sin + sin
2x ________
y(-7r)
+ x sin x) dx
y(-ir)

Page 12
 External Publication No. 673

Substituting Eq. (22) for y,assurning y(-ir)> > .L and < <.1, and

preserving only the first and second order terms of the expansion
yields:

I (0
J1+sinx^sin2xl
bt
2 Ci) y(-Tr) L y(.i) y2(r)J dx

2co [y() + 3()] (26)

If by is small (which is certainly the case for y >3) the ratio

bt/by is a good estimate of the derivative of decay time with respect

to the magnitude of frequency error. Dividing Eq. (26) by Eq. (23)
yields:

dy
(27)
CA)fl

Integrating between the limits YO' the initial error, and

A' the
value at the line A - A,.

21(YO 1
C&) 0 YA (28)
A1 2y02 - 2yA2j
L

is an approximation of the time required for frequency-lock when the

initial frequency error is = 2t ciy0 rad/sec. Of course, the

derivative approximation becomes increasingly crude as y approaches

Page 14
 Exteraal PublicatioflL No. 673

Evaluation of the integral yields

6
42 y2(,r)

Substituting for from Eq. (22):

6y 1
(23)
4C [y(_,r) - 1]

This expression checks rather well with analog results down to

y(-ir) = 3.
By similar methods an expression may be obtained for the time
required to achieve frequency-lock (defined as the point at which
the trajectory dips below the line A - A). Since

= . = =
dt d

or

¶= (24)

using Eq. (12) the time per cycle is:

'07r

6t
_____
2Zcon
=
I
2ZcDfl J—iT
y (2)

Page i3
 External Publication No. 673

A However, if was large to begin with, terms of Eq. (28) are

ma11 with respect to the first, and asymptotically the pull-in time
- will be:

2 YO2 = _2o2 seconds

(29)
2t co3

It should be noted that if the integrator in the loop filter is

not perfect the pull-in range of the loop is no longer infinite, but
is proportional to the finite tine constant of the filter. This will

be shown in a later section.

V. SECOND-ORDER LOOP WITH LINEARLY VARYING FREQUENCY INPUT

When a phase-locked loop is used as a tracking filter or FM

discriminator it has to follow an input reference with variable

frequency. If this variation is due to a doppler shift it will be

almost linear over a large part of the tracking period. If it is due

to frequency modulation the maximum slope of the modulation will be

limited by its bandwidth. It follows that a loop which can follow

a frequency which varies linearly at this maximum rate can certainly

follow the modulation.

The input frequency to be considered is then c 5 + Dt. For a

second-order loop the general Eq. (3) becomes:

+ Kp + aK (sin p) = 2 + Dt (30)
p

Page 15
 - -
cç

External Publication No. 673

where Q is the initial frequency error which may be zero if the ioop

was originally in lock when the linear variation Dt of the input began.

Since the right-hand side of Eq. (30) is the difference between the

reference and VCO frequencies, the time-varying frequency may just as•

well be that of the VCO, corresponding to a constant drift of the

oscillator. Following precisely the development and substitution of

Eq. (8) through (15) of the previous section there results:

D (31)

.
Again let p = x and p = y; the slope of the trajectories becomes

d = - + °n 2 (32)

The singular points at which the slope becomes indeterminate are now:

y = 0, x sin (;) ± 2 nir

and (n = 0, 1, 2, ...)

y = 0, x 7r - sin -if —f)

D \
± 2 nir (33)
(Un /

It is interesting to note that if D, the slope of the excitation,

equals the square of the natural frequency of the loop, the two sets

of singularities come together; and if

D > 2 (34)

no singular points exist.

Page 16
 External Publication No. 673

It will be shown that inequality (34) is the condition for

stability. That is, if the doppler or modulation slope ever

..atisfies this condition not only can the loop never achieve lock,

but even if it was initially in lock it will immediately fall out of

lock when the excitation is applied.

The phase-plane plot can be obtained by analog methods using the

computer configuration of Fig. 3 with the addition of a constant bias

on the first integrator. Figures 9 through 13 were obtained by

thus mechanizing Eq. (31) for. a damping factor = 0.707 and various
levels of the ratio D/a)2.

The following comments may be made concerning these graphs:

1. The singular points are of the same nature with the

linear excitation as without it. The only difference

is that the stable point is displaced to the right by

sin'' (D/w 2 ) while the saddle point has moved to the

left by this amount, as was predicted. Since the

stable point represents phase-lock, it is seen that

the steady phase-error of a second-order loop with

linear excitation is

(D
F55 = sin'4 \wfl2)

The linear approximation discussed in Sec. Ii predicts

a steady-state error of D/w 2 which is valid only for

small values of the ratio.

2. The trajectories are decidedly asymmetric about the

x axis. For low values of D/o 2 (Fig. 9), in the upper

Page 17
 External Pub1icat ion No. 673

half plane all trajectories above the line B-B are

div'ergent, slowly at first and more noticeably as the

error requency increases. On the other hand, all

trajectories below B-B converge toward a stable

point either in the same strip or some other strip to

the right or left depending on whether the trajectory

is in the upper or lower half plane. The line B-B

exhibits a periodic behavior (neither locking in nor

diverging) which is, however, unstable since the

slightest disturbance will tend to push the system

either into lock or into complete instability.

Figure 9 verifies the intuitive observation that

in order to acquire a positive doppler-shifted

frequency it is best to make the VCO frequency lead

the input reference (this corresponds to a negative

frequency error) and let the increasing doppler shift

decrease the error until lock is achieved.
3. As the ratio is increased the line B-B which
establishes the limit between stability and instability

in the upper half plane descends until at = 1/2

(Fig. 10) it practically coincides with the saddle

point asymptotes A-A. Thus, when the ratio reaches

this valueall trajectories in the lower half plane

are drawn into lock, while in the upper half plane lock

can never be achieved, unless achieved within the

original strip of width 2n-.

Page 18
 External Publication No. 673

- 4. For yet larger values of D/w 2 (Fig. ii) another

phenomenon becomes apparent. While most trajectories
which start in the lower half plane still achieve lock

in some strip, there are some that do not. In

particular, any trajectory that passes within the

corridor whose lines of demarcation are the extensions

of the saddle point &yptotes A-A will be whipped

past the x axis •irto the upper half plane. The width

of this corridor in the lower half plane is increased

as increases (Fig. 12) and many more trajectories

become unstable. As the ratio becomes almost . one

(Fig. 13) the corridor becomes so wide that only

trajectories which began in the immediate vicinity of

the stable singularity ever achieve lock.

- Finally, when D/a 2 ^1 the singularities

disappear and no stable point exists. Thus a loop

which was originally in lock before the excitation was
applied will immediately lose lock. The conclusion

which may be drawn from these results is that a

second-order loop can acquire a noise-free doppler-

shifted or modulated signal with certainty only if the

ratio of frequency slope to the square of the natural

frequency is less than 1/2, and it can track the signal

once lock is achieved only until the ratio becomes

unity.

Page 19
 _________________________________________________Liatin). 73

It isof nte _ive ai xiatepxesioxifar the...

p€riodic limiting behavior between stable, and divergent trajectories

for small D/co 2 (line B-B in Fig. 9) . If y is to be periodic
y(r) =
Multiplication of Eq. (32) by y and integration between the
lia.its ±7r yields:

irD
0 [() 2()] = - j y cos x dx +
2r 2

Integrating by parts and making use of Eq. (32):

1r / D irD
SiflXd+
0 r Sii_cosx+ 2 2
4 y

'For small D and consequently large ay of the periodic behavior this

leads to the approximation

iT
0ai ()
2 +
4C Yav

or

Yay

This can be taken as an approximate expression for pull-in range. In

texai of Q. the condition becomes
3
(36)

-- Page. 20,
 2: -- -
-
___ tiS pediçs
that Fif be noted t)tai a
The approximation improves as D/co 2 decreases but it is very poor as
D/c 2 approaches 1/2.

VI. SECOND-ORDER LOOP WITH IMPERFECT INTEGRAT(

If a passive RC loop filter2 is used or if the integrator in a

second-order loop is not perfect, the loop filter has a transfer
function of the form:

p+a
F(p) (37)
p+
Substitution of this function into Eq. (3) yields

+ a) sin p - °c (38)
+ K (:+a\
Assuming the input frequency to be constant and letting
this may be written as

+ (a + K)cosp+aKsjnpQ (39)

To obtain the form of Eq. (9) which represents a second-order loop

with perfect integrator, again let K 2t aK = cA;2.

• 2This was also investigated by Gruen (Ref. 3). Preston and

Tellier (Ref. 4) considered the simpler filter F(p) = a/(p + a), which
is a sub-class of this case.

Pa.21

-- .•.
 External 1blicatjQAN9. 67$

Then

(40)
dt 2 dt

If the time is again normalized, t 'r/(2( a), and the result is

+cos)+_sifl? 4(22 (4].)

where

= •
(2('G)fl)

With the convention = y, p = x

-
= - ( +
dx + (42)
\2(cDfl 4(2

This indicates that the singularities are at y o, x = sin_'

and x - sin-' (aQ/a n2 which implies that for
2
Q >.i_ (43)
U

there is no stable point and hence the system can not achieve lock.
Condition (43) is the upper bound for phase lock; that is, if the
input frequency is more than co 2/a radians away from the VO center,
frequency lock-on is impossible on the basis of the singular point

behavior.
Haw yer, for lesser Q pull-in may not occur even though a-
stable point exists in the phase-plane. Examination of Figs. l4-l8,

Page 22
 Extrnai'. b1ication . 67

the analog computer solut ions. of - Eq 41 . bears .thts out. I all cases

0.707 and a/2 a 0.1. The behavior varits with the value of Q.

Condition. (43) requires Q/ 2 < . 1 for lock to occur. In Fig. 14

aQ/on2 =0.4 and pull-in is quite apparent. It should be noted that

gtven il G. I and 0 707 1/ J. 3 then 2 Thus, f

ttie V, was initially at its center frequency, th the initial value I

of y • = Q/21 o 2. the other hand, if initially the VCO was

not at its center frequency because it had been tracking a signal of

another frequency, then the initial value of y is not restricted. It

seems apparent from Fig. 14 that pull-in occurs for all initial values

of y when Q/co 2
Figure 16 represents the solution for (4/2 = Q.7 or Q/co

7 A limit cycle exists about Yav = 2.5 toward which all

higher trajectories converge. If the 'itO is at 0c Initially then

- initially y Q/2Z co 1 3.5 and the system converges to the limit

cycle and exhibits a periodically varying frequency error as long as

the input frequency remains constant. If the initial frequency of the

VCO is not 0C, one of three situations may exist (see Fig. 19).
1. The initial VCO frequency (co0) lies further from the
signal frequency (con) than does the 'ICC center
frequency (). Then the initial value of y >Q/2 a

and the trajectory will converge to the limit cycle.

u lies nearer to co than does cn ; then initially

0 < y< Q/2t cav and the system will lock in if y is

sufficiently small, or otherwise converge to the limit

cycle. Figure 16, in fact, shows this behavior in form

2açe23
 External Publication No. 673

of an unstable limit cycle below the aforementioned

stable one. If y initially lies below this unstable

limit cycle, lock-in occurs. If it lies between the

limit cycles it will converge to the higher one.

() o lies on the other side of from then initially

y is negative and the system seems to lock in always.

Figure 15 (aQ/a 2 =0.6 and Q/o 3 ñ) represents a behavior

intermediate to those already considered. Pull-in occurs throughout

but the rate is slower near the region in which a limit cycle occurs
for the case of Fig. 16. Thus, it appears that if the VCO is initially

at a, for 0.707 and a/22 0.-L the limit for pull-in lies in the

region

3 .12< < (44)

Figures 17 and 18 represent the solution for Q/co,. = 4 -f2 and

9/2 .12. If the VCO is initially at , the initial values of y are at

4 and 4.5 respectively and pull-in does not occur, of course. The

stable limit cycle is again in evidence and furthermore, if y is

initially negative (case 3 above), lock-in is not assured since those
trajectories which pass through a strip determined by the asymptotes

of the saddle point will converge to the limit cycle.

An approximate analytical derivation of the pull-in range follows.

Returning to Eq. (40), letting the frequency error dp/dt = c and

dividing by CD yields:

- 2 sin p
= - (a + 2 (D cos p) + (45)
dp Ci)

Page 24
 External Publication No. 673

Multiplication by o followed by integration from = - it to -fir yields

measure of the decay or rise in frequency error during one cycle.

i2 () - w - flit
= 2itaQ - / o(a + 2 co11 cos p) dp (46)
2(H

Integrating by parts using Eq. (45)

[o 2 - 2()] = aQ - ita[(it) +()]

(47)

+1 'iT
I/it

Making the assumption that the frequency error co is large compared

•to O), the terms in co in the integrand will not change appreciably
during one cycle, so that throughout the cycle co w( -ir) for -ir< p < it
Then:

2
- a (a +

J
i[co2(it) - o
c 2 ( it )] 2itaQ - a [co(it)]
co(-it)

(48)
2
co(-it) j[co(it)] - Q [co(-it)] + 2___
_ 2ini
(: + i) }

The trajectories will always decay when the right hand side of this

equation is negative, which occurs for the following range of

c2:3

3Gruen's
result for pull-in range was obtained by an eMpirical
technique for c/a > > 1. It is more pessimistic than condition
(49) by a factor of 2.

Page 25
 External Publication No. 673

a (49)

For the cases investigated a/2t 0.1 so that the approximate

condition for pull-in when the VCO
is initially at its center
frequency will be

< 2.f6 = (3.46) f2

This lies in the region determined on the analog computer (Eq. 44).

VII. THIRD-DER LOOP TRACKThR3 BEHAVIOR

Section V has demonstrated the limitations of the second-order

loop in tracking f requency_varjai signals. It is of interest to

investigate the possibility of extending some of these limitations

by inserting a second integrator in the loop filter. It develops

that pull-in behavior is less stable for a third-order loop than for

one of second order but it will be shown that the tracking range ror

a loop initially in lock can be extended by means of the second

integrator. The loop filter transfer function is

F(p) = I + .. + p2 (50)

and Eq. (3) becomes

pp + ici; + aKp + bK (sin cp) = 2 + Dt

- (51)

Page 26
 External Publication No. 673

Letting K 22 a; aK c 2 and differentiating

p + + 2n b)(sin p) = D (52)

Making the substitution

and dropping the operational notation

= 2

42 °n2 (d2 ) + 422 sin p (53)

P () +

+ b th = D

Norializing and letting =

.. . bfi p th = ()
P + P C°S p + sin p +

Conventional phase-plane techniques are not applicable to third-order

differential equations because there are three initial conditions

corresponding to each of the three dynamic variables to be considered:

phase, frequency, and frequency rate (or displacement, velocity, and

acceleration in mechanical systems). In principle, the trajectories

of a third-order equation could be described in three dimensions.

Hever, an attempt to project these trajectories for a multitude of

initial conditions on a plane will generate such a confused diagram

that little of a general nature can be deduced.

Page 27
 External Publication. Mc. 673

( the other hand, if one limits oneself to a single set of

initial conditions, a meaningful projection of the trajectory onto the
- plane will result. The set of initial conditions of greatest

interest is p à4/dt 0; dp/dt 2 D; that is, the loop is initially

in lock so that the frequency and phase errors are zero when the
reference frequency begins to change linearly.
The analog computer program of. Fig. 3 can easily be modified to
include another integrator. Figure 20 represents a series of
trajectories projected on the p - plane. In all cases = 0.707

and = 1/2. In the hypothetical three dimensional 'phase space"

the trajectories initiate at D/4C2 and terminate on the

axis ( = 0). Figure 20 (a) shows th. behavior of a second-
order loop (b 0) with these initial conditions. The terminal or
steady-state phase is sin 1 (D/co 2 ) as was pointed out in Sec. V.
Addition of the second integrator causes the steady-state phase to
become zero with increasing rapidity as b/con2 increases. The peak
phase error also diminishes for increasing values of b/co 2 at the
cost, however, of decreased system damping which results in increased
rms phase error (Figs. 20 (b) - 20 (f)). Finally, as b/co 2 approaches
unity the loop becomes unstable.
The advantages secured by increasing the order of the loop are
more evident in Fig. 21. Here = .707 again but D/co 2 1. It
was shown in Sec. V that for this and higher values of linear
frequency shift the loop was incapable of tracking. Figure 21(a)

(b/co 2 0) substantiates this fact. On the other hand, even the

smallest magnitude of the second integrator produces zero steady-state

Page. 28
 External blicatIon No 673

• phase error. The pea.kphase error is decreased by increasing b/c2

but as b/a 2 approaches unity the rms error increases until near
= 1 the system is again unstable.
Similar behavior is apparent in. Figs. 22, 23, and 24 except that
as D/to 2 increases an increasingly greater value of b/w 2 is required
to keep the loop in lock. Finally, when D/a) 2 is near 2 a value of
b/a 2 near unity is required. However, it is evident from Figs. 20(g)
to 24 (f) that for this value of b/o 2 the system is unstable. The
range of values that b/o 2 may take on in order for the loop to remain
in lock is plotted as a function of D/a 2 in Fig. 25 - 27 for 0.707,
0.5, and I respectively. The cross-hatched area represents permissible
,alues of b/w 2 . Clearly, for linear frequency shifts the third-order
loop has extended the tracking range to approximately twice that of
the second-order loop for a damping factor, , of the latter equal to
0.707 and even further for higher values of .
A theoretical explanation of the oscillatory behavior for
b/a 2 near or above unity is possible. The characteristic polynomial
of the loop (Eq. 52) can be rewritten as

[3 + $jfl
CP2 + + 2t b) =0 (55)
P

The Routh-Hurwitz test on the polynomial yields the fact that for

. (so)

the loop is unstable. This indicates that if in a particular tra-

jectory the phase error reaches such a value that condition (56) is

Page 29
 External P1ict ion Wo, 673

• satisfied the trajectory will bome unstable. Clearly, if b/2

the condition is satisfied for all and the system is always unstable.
This fact is also borne out by the linear model which follows from
the approximation sin p. Application of the Routh-Hurwjtz test
to the linear equation sets the instability criterion as b/a) 2 2 1.
The tracking behavior of the third-order loop does not seem to
b. ich different than that of the second-order loop for D/o)
2 < 1.
For D/a 2 > I, the region for which the second-order loop can not

lock in, there is a limited range of D/o 2 for which the proper
choice of b/c 2 will secure lock. The pull-in behavior when the VCO

is not initially tracking the reference signal seems to be less stable

than for a second-order loop. However, generalizations can not be
made because of the complexity of the phase-space trajectories.

VIII. NOISE AND LOOP NOISEBANDWIDTh

Some comments are in order on the effect of noise on the pull-in

and tracking behavior. A small amount of phase and frequency jitter

due to noise will not affect the phase-plane trajectories significantly;
it will result only in hash being superimposed on the trajectories.
However, the bias due to the signal frequency error is still the
dominating factor which causes the loop to pull in and a small jitter
on the trajectories will not affect the number of cycles required to
• achieve lock. On the other hand, if the input noise is significant,
• particularly when the initial frequency error is large, the disturbance
on the trajectory may be such that a different number of cycles will be

required for the system to achieve lock.

Page 30
 External PLthliation No. 673

When the loop is in look. and tracking a noisy signal it is

convenient to consider the linear model which results from the
approximation sin p. This enables one to analyze the propagation
of noise through the system and even to derive the form of the loop
filter which will result in a Wiener optimum linear system (see Ref. 1).
A key parameter which results from the consideration of noise in the
system is the loop noise-bandwidth 3L defined as

2
L --
2irj r '°° dp (57)
IH(p)I

where H(p) is the transfer function of the linearized loop from input
phase to the output phase of the VCO. The basis of this definition
lies in the fact that a flat noise spectral density of magnitud

at the input of the linear system will produce at the output a noise
power 2'8L° Thus B L is the bandwidth of the ideal square-cutoff

low-pass filter which produces the same amount of noise power at its
output as does the linear system whose realizable transfer function
is H(p).

With this definition, the loop noise-bandwidths of the various

filters may be considered. For the first-order loop the linearized
transfer function follows from Eqs. (3) and (4).

H(p) K
p+K (58)

where = - , is the output phase of the VCO and F(p) = 1.

Applying the definition (57) results in a loop noise-bandwidth for

Paqe 3.1.
 External Publication. No. 673

the first-order loop of

BL= (59)

Then the transfer function (58) can be expressed in terms of BL:

4BL
H(p) = _______ (60)
p.+ 4 EL

Rechtin has shown (Ref. 1) that to minimize the output noise

power and simultaneously to maintain the rms phase error due to a
step change in frequency within a given bound, the optimum linear
filter will be of second order and of the form
2
( +J2Q)p
H(p) , (61)
nP + p2

From the denominator which is the characteristic polynomial it is

clear that the damping factor of this optimum loop is l/../= .707.
Application of Eq. (57) yields a

BL (62)
4 -,/

Then

1+ p
4 BL
H(p) . (63)
1+ p+
4 BL 32BL

Page 32
 External Rzblicatjoji No., 673

Also it is shown that to minimize the output noise power and

&i.iltaneously to maintain the rms phase error due to a linearly
varying frequency input within a given bound, the optimum filter will
be of third order and its transfer function in terms of the loop noise

bandwidth is

5 25 2
1+ p+
38L i8BL2
H(p) = (64)
5 25 2 125
1+ p.+ + 3p3
3 BL I8BL 2l6BL

The characteristic polynomial of'the system is then

P3+B L p2+BL2 P L3

Comparison of this with that of Eq. (55) for Which the linearizing
assumption is made results in the following values for the , a, and
b parameters in terms of BL.

6vr2
5 BL

C=- = . 707

b = BL2

Page 33
 External Publication No. 673

Also

4 (65)
O)fl
2

IX. SU1QI4ARY OF RESULTS

The various conclusions relative to the filters and signals

considered are summarized in terms of both the notations which have
been introduced
I. First-Order Loop (Sec. III.)
a. Pull-in range

Q< KorQ< rad

Br—
sec sec

b. Pull-in time for initial frequency error, Q

! final 'final

Q-Ksinq Q-4Bsin
t=f
initia1 'initial

The integral is evaluated in the standard tables.

2. Second-ckder Loop (ierfect Integrator); Constant Reference
Frequency (Sec. IV.)
a. Pull-in range is infInite
b. Pull-in time for large initial frequency error, Q

Page 34
 Exterzal I.thiIcat io No 673

2Zü

A more accurate expression for smaller Q is given

by Eq. (28)
For optimum loop (using Eq. 62)

27Q2
t - .sec,
256 BL'

3. Second-Order Loop (Perfect Integrator); Linearly Varying

Reference Frequency or Constant VCO Drift at Rate D raci/sec2
(Sec. v)
a. Approximate pull-in range for Wo 2 1/2

0 3
- Q<.(Dn
D sec

for optimum loop when D/BL2< 16/9

_____ rad
27D sec

b Maximum rate which can be tracked

D cD2 rad
sec
for optimum loop

rad
sec 2

Peg. :
 External Publicatica N. 673

- 4. Second-Ordr Leep (Ipiifeet ntegrat.ox) CGn.stant Reference

Frequency (Sec. VI)
Pull-in range

Q < 2ui0+1L
a sec

where a is inverse time constant of filter.

For optimum loop

Q< LV L^1rad
3 3a sec

5. Third-Order Loop (Sec. VII)

Tracking range can be extended beyond D/co 2 1 by means of a
second integrator in the loop filter. Figure 25 indicates that if in
a second order loop, whose damping factor =0.707, a second integrator
is introduced, the tracking range can be extended to = 1.83 by

Making the normalized value of the second integrator constant

= 0.63.

The optimum third order loop in t#rrs of the other set of

paraieters has C 0.707, con = 6 8L and b/con2 = 1/4. Its
tracking range extends to

= 1.5 or =
BL2 144

Pge36
 - -

Exte.ri.al Publication No.- 673

The reason that b/c 2 is chesefl in this case ta yield a non-

aaxt.i tracking range is that the optiui loop must strike a

compromise between tracking accuracy and output noise power, so that

Qptimizing e composite of the two is preferrable to optimizing either
one.

Page 37
[iet PopuL'iion Laboratory ExternaUn tjjt
lassified - eportri.
_______ Unclassified_NLkTEst ilne of text ____
Classifie3 - tfitle or fIrst line of' text
A - A saddle point asymptotes

a = first-order integrator gain

B - B periodic limit of lock-in

loop noise-bandwidth

b second-order integrator gain

C1 reference signal amplitude
C2 = VCO output amplitude

D = linear rate of frequency variation

steady-state phase error

multiplier output

F(p) filter transfer function

H(p) loop transfer function

K0C1C2

K1 filter DC gain

K2 = VcX gain

K = loop gain

p = Laplace operator

t = tii variable

X = phas. error

y frequency error

yAva1ueofya1onAA

y0initialy

'1av = average value of periodic 'trajectory

a = inverse time constant of filter with imperfect integrator

linear decrease in y per cycle of x

_____________ Last line of _text or fuotnote _____

iChTt ____ ___ _____

ets.onLoboratory __.___________
Cassifi-ecrtN.
_____ ctt !4 of L
Classified Title or first line of texc

loop damping factor

normalized time variable

phase error

reference signal phase

VO output phase

= 4 frequency error
s reference signal frequency
V) center frequency

loop natural frequency

initial V(X) frequency

initial

_____________ Last line of text or footnot.e _____

 4 4

External Publication No. 673

REFERFJCES

I. Jaffe, R. M. and Rechtin, E., "Design and Performance of Phase-

Lock Circuits Capable of Near-Optimum Performance Over a Wide
Range of Input Signal and Noise Levels," IRE Transactions on
Information Theory, IT-i, No. 1: 66-76, March 195.
2. Gilchriest, C. E., "Application of the Phase-Locked Loop to
Telemetry as a Discriminator or Tracking Filter," IRE Transactions
on Telemetry and Remote Control, TRC-4, No. 1: 20-3, June 1958.
3. Gruen, W. J., "Theory of A.F.C. Synchronization," Proceedings of
IRE, 41, 8:1043-8, August, 1953.
4. Preston, G. W. and Tellier, J. C., "The Lock-in Performance of an
A.F.C. Circuit," Proceeding s of IRE, 41 2:249-51, February 1953.
5. Richman, D., "Color'-Carrier Reference Phase Synchronization in
NTSC Color Television," Proceedings of IRE, 106-133, January 1954.
6. Andronow, A. A. and Chaikin, C. E., Theor y of Oscillations,
Princeton University Press, Princeton, New Jersey, 1g49.

7. Truxail, J. G., arta.tiC Feedback Cntrol Syste* Sysi,

(Chapter ii), raw-(iil Book Ce.,Inc., New York, New '[ork, 1955.

Pag. 40
 L\ L1

Wborotory -_____________ jj: L7

Ci sin 4's :K0 Sin 4' FILTER

x K1F(P) -

c2cos[.t +/(0K1

I_________
_________ K0 K F(p) sin 4'

Fig0 1 PhaseLocked Loop

(nw+sin')

Fig0 2 First-Order Loop Pull-in Behavior (n even integer)

 - - 1n CtND. 73

-A

-'If

- ---------------------
Fig. 4. 1.4L4 (overdaMped)

I.

- -----.---.--. --.- Page 42

 Ext.rnaJ

V - -- I
4-- -
-' 4

/1

Fiq. .. 4-- O-4crtia14y da.p.d)

Pag43
tV(jt
, 73

:i€ - --
- -

4
--4-

Fig. 6. 0.707

Pag 44
- . External PublicatiQn No. 673

L. . age45
 - Ext 673

-' :r
-a------ -- p - -

—1 r.: 'r

_7 .

---,_ __•_'••_'____p____. - - ------- -- .--------- ----.--- - .-- -

Fig. 8. Limit of Frequency-Lock as a Function of Damping

L3t ir r' f -

i- - - ---------- ---- -------- agt 46

 Extea.Ptj $o_.

- D.2Q7 = 1/4.

= 14.5° O.O8irrad
P*47
 External

-- - I'-tr:Lr3t

Th

: - Pa48
 Exteznal Publication No. 673

A "p

• Fig. 11. Q,7O7 D/c& ./"2 0.867,

 External Publjcatjo No. 67

-j

Fig. 12. 0,707, D/u = 0.95,

72!,- 2/
- -

I•\

i.. _-._____ --

Fig. 13. C---.Q-.-7O7. = 0.985,

sin(b/)i'8O.Od 4r19 rad

Pag 1
 External Pubjjcôtj to. 67

-.., 4 .•-•-•,-,._-.-

F&g. 14. Z O.7O7 G/2o 0.1,

= 0.4
Page 52
 External Publication No. 673

Fig. 15. 0.707, a/2u 0.1,

aQ/w 0.6, Q/'CD = 3-/i

Page 53
 S

Extena1 Publication No. 673

Fig. lfj 0.707, a/2o 0.1,

a2/cü 0.7, Q/a 7/2-/
Page 54
 Exterrj1 Pibjjtj No. 673

. _._.__.__ •4•-*-4 -•-_-.-, -

p
-f

I-
-rn

Fig. 17. 0.707, a/2,U)1?1 0.1,

ac2/w = 0 • 8, Q/co = 4 iT
Page §5
 Exteir. Publication No. 673

- - _- __, 'J _____t__, ,___

Fig. 18. O.7&7, a/2co 0.1,

aQ/cn = 0.9, Q/c:r = 9/2-'h

Page 56
 61J

(ci)
LN
B

c (AJ)

w
(A)5
(1)0

Fig. 19. Re1atve Positions of Signal,

Center, rind Initial VCO Fr jcn

2
pa "
 xterna1 2u1 atin No 673

- -

- b IQ ___
iii
iifIf iiia

(a) (b)

U -- __
(d) (e)

f: b
&Tz4

(f)

(g)

D _I
-2
C =.707

Fig. 2O.

- Page8
' 5:çfl
Exter ik1iction . 13

iIiiTi -

t±'t

II?
(0)

(b)
UT
Ic) Id)

E .707

Fig. 21. .= 0.707, D/c 1.O

.. S -

:.. :;

Pss 59
-. - -, -. External Publiatin Na. 67

4-

(b)

Cd)

..... I.

)f)
:TiT-j

--
cJ fll 4
--n-
-.707

Fi Z2. 0,707, D/cc 5/4

Page 60
 ti 1iciti N 673

V - - -

_____
I
- xr1 b1ati No. 64

it -. cr I i •- L •

-I - - - -ii- 475

Ii

•.ff

Fiq.. -2&.. 1 £L7Q..7., i/ 7/4

Li' ( ' n :) ______

 External Publication No.673

c4 4o(

= 0.707

I
* Fig. 26. r =

17

Page 63
 xteIn atik 67

-
0

1'DWLFDiMENT

:ie uthox is indebted to L. R. Welch ar

R. C. Titsworth for their helpful suggestior.

to G. A. Faist for the analog com puter proqrar

:. (
a44
 t

o.

0.E
c.J

0.4

1.0 2.0 2.2 2.4

D/c2

- . i. - . . .
QE

0.4

0.2

D/w2

0
2 ' ; •i

oo
g o
3j D -_______ _______

2. 5
N
__ N __ _ ____
2. 4

I. 5 __ _ __

o _

0. 5

o I - 0
0 U. ' 'U C

•11 X

-'1 rn
cD_u

-J
 A

hpt.

i-_Ia.

r1. -

}-ig. -
icr

1(\j

H 1•

c\J
C
r 3

H--

csJ
a

0 (0 c'J
- 0 0 0 0
 •
O
j1H roi-

0
0 10
O\HIr0'0
0
0 1(J)
HcD
So
0, -

.4 i+ lID '0
Pd

•
-v.flD
j (A -a. (P
o 'jr\) Si) 0J4I\)
'J1 (P
J:
(P 0
OO\j$\) (00) O)ii0)0)
¼)iU) 0,0)
))i)1 SO)
(A
'0(1)
-I: -
S.
0, H'.4 '-
H CUPi
SP U)

9
)(1
0

H
H

d() tJC) -Uo C 0C)

0(A O(A r (fl -(A p+(A H (A
H H i-i H H H

U) (1) U) U) U)
çF + çf
U U U U U
0) 0) 0)
(A (A
'1 1 '1
0) 0) 0) a

) Q(A (A cr(A
- 0
OH
H
5) .-) HH
F')
(A x
(A
(+H H H
H 0 0 0
0 Hi

)1d0(A i-J 0).i I H (A ro U- 0H o i (A

OUD U
I - (A
rF 0 \H. U 0 H (P U-H H

(A
I-'cYOO '1 U (A H(7) H cf 0 (A
i-O(DU(A(-.4 (A0 (Al PU (A(AHW
(A
O .40 C)
H )rfl. EH l-- 00 (O(Aro ()(A
U(A HO (A (7-) 0(A i+f P- .IZ1 I(AO
rf O OOU ii OH C0-IH H 0
OlD (A(D (A lO silO)))) U--

[I
c+l +o U • U(Dct ID(A
0)) c1 (+(A HH I. ,) AID 0 00 (AC)
(DU H(D(A OH OH - ,+ OHc$'i (A
.4(H OlD H (P ') H
H (A '0 (+CO'1 (A(AO cPU.
HID U0p+ (Aa • (A ID(A
U (7((D(A OlD -4
(70 '00 OH O' (AcP (A(A
ID .400 (10 ID I(A'I(A cl-H
H H(P(D rf(7 C) .H A—) H(A(D'l (DO
(A 00 (PH I ç- . Si) .00 (ACID 'dO
g i--rf(A HH (A (A )1j •rl-.(D ID
01D0 HO C) )jlH )+ 00.
+H HO (A 0 U
U ID (A OlD (A U-
C) (A 00 C)') 0) I U L'H
(A (A HU çfF 10lb -
(00.

0. cf (P

0 0 0 0 0
U U U U
(P (P ID (P
H
U

4
1:1
H
(A

a
(A
U.
 0 CD
C-tCD

Cf
CD

P.o cCCD C)) 4)

C-(
C)) 00
CD CD CD
ro C')
C') C']
CD CD
CiC) +
CfCD
Ca

I-' 1'C:j 4-.. ,-

Ci CD CD CD
'PP
'1 P. (C CD
ID 4-' 0
Cf 00
mm Cf
CD CD '10
o CD '.s c+ . CD P.
(C
IC) jo C) C) C)) l
4CD 4Cfl CFCD -CD 0
'-4 4-1 '-4 I-1 Ci Ci 0 I-.. (C
CD DC Cf
0

('CD

C)) C)) C)) 0CD

C-f
CD
Ci Ci '< I-'
0. P.
CD 0CD I-'
CD 0)
0' a a '1 0
Ci

0. C) 0 CD '-1
0 0
C) Cf
CDf-'
CDCDCi0
00 CDCD CD
(+4_C f
Qrf
(+0 H
I-'Ci
CD
('Ci c+Cf 0
CD ID4

fC
0 0 ()
Ci Ci 4-' CD(D
Ci CD -.
(DID

C) 0\
I- ICDI-' CD I-'
) )0 Ci I m
)CD()
Dcf+
+CD C) ,
C) 4-0 CD
C) " t')CD
j ' CD CD jf
CD CD

lCD DC

I
'I-
..Ci
0 OCD
cfQs
0.

xt —I
0 0I-'
0 0
m (DC) CD

q 4_C
4-' 4-
C) ()0
Cf ID
4-
0
Ci

4-
)Ci 4-' CDCf
('C) 0
CD
4-C-f
4-'
0
Ci