You are on page 1of 6

Stability of loadflow techniques for distribution

system voltage stability analysis

G.B. Jasmon, PhD


L.H.C.C. Lee, BEng

Indexing terms: Stability, Loadflow techniques, Power systems protection

P 8 ,Q, = scheduled net active and reactive powers


Abstract: Loadflow techniques are widely used in evaluated with the initial voltages
the planning and daily operation of power systems P o , Qo = active and reactive powers evaluated with the
including that of the on-line monitoring of dis- initial voltages
tribution system operation. The performance of
three loadflow techniques is investigated in their
ability to analyse distribution network loadflows. List o f symbols for FDLF m e t h o d
The criterion for voltage instability is presented
AP = difference between actual and scheduled real
and the ability of the three loadflow techniques to
predict it is discussed. power interchange
AQ = difference between actual and scheduled reac-
tive power interchange
H = submatrix of Jacobian matrix
List o f symbols f o r d i s t f l o w equations
L = submatrix of Jacobian matrix
= net active power injected at busbar i K = voltage at busbar i
= net reactive power injected at busbar i 6,j = load angle between busbar i and j
=line impedance in line connecting busbar i Qi = reactive power flow at busbar i
and i 1 + Xij = reactive between busbar i and j
=line resistance in line connecting busbars i Gij = real part of the admittance between busbar i
and i 1 + and j
= line reactance in line connecting busbars i Bij =imaginary part of the admittance between
and i + 1 busbar i and j
= voltage at busbar i B = submatrix of B matrix
= real power loss in line connecting busbars i s
' = submatrix of B matrix
and i 1 +
=reactive power loss in line connecting
busbars i and i + 1 1 Introduction
= real load at busbar i +
1 The operation and planning of power systems depends
= reactive load at busbar i + 1 heavily on the use of loadflow techniques [3, 51. The
primary objective is to check that any solution offered is
List of symbols f o r SONR method within the permissible (security) operating limits. In real
time operation, any system whose state variables fall
p, = net active power injected at busbar i
outside the permissible limit will be subjected to correc-
Qi = net reactive power injected at busbar i
Ei (or v) +
= ei jf, = complex voltage at busbar i
tive actions to alleviate the problem. In the event that the
system cannot be corrected, certain drastic actions such
J(x) = Jacobian evaluated for the initial estimates of
X
as load shedding or practical shutdown has to be exer-
cised. In such an event some or all of the customers will
H(x) = second derivatives evaluated for the initial
estimates x not have power supply.
In the real world, power systems are constantly being
fis) = vector of scheduled quantities, i.e. scheduled
threatened by various elements that could cause violation
load powers and generator voltages
fixe) = vector of estimated values, i.e. the power
of the safe or secure limits of operation. On some
equations evaluated for initial voltage estim- occasions, the system may be subjected to such an unex-
ates pectedly high demand and multiple faults that the
Ax = errors of corrections of unknown variables,
system states can go into extreme conditions. The
i.e. voltage corrections Aei; AJ occurrence of voltage collapse [6] has been observed and
= vector function of corrections in unknown
this problem is being actively investigated with the aim of
fiAx) determining suitable methods for predicting the onset of
variables, i.e. the power equations evaluated
for the voltage corrections such collapse. The ability to predict this collapse is most
vital as it can save the system from the unwarranted
Paper 8247C (H), first received 16th January and in revised form 23rd event of system collapse. It has been shown that the load-
May 1991 flow technique can be used to determine the voltage
The authors are with the Faculty of Engineering, University of Malaya, stability limit of power systems [SI. Voltage collapse
59100 Kuala Lumpur, Malaysia occurs when a loadflow solution is unachievable by a
IEE PROCEEDINGS-C, Vol. 138, N o . 6, N O V E M B E R 1991 419
loadflow technique. Loadflow techniques [2, 41 are pre- obtained in matrix form as
sented for predicting the voltage collapse point of power
networks. (4)
Many loadflow techniques have been developed and
the capability of some modern techniques have overtaken where the Jacobian matrix Jo is constructed from

$1
some of the earlier ones. The earlier methods are inferior
in speed and storage requirements but are useful for illus- ap _
ap
tration of basic principles [3, 51.
_
One important new technique is the second order
Newton-Raphson method (SONR) [2, 71 which has Jo=[$
gained widespread application in many facets of power
systems. In many ways this method is comparable in per- In eqn. 4, notice that no voltage-controlled busbar equa-
formance with that of the fast decoupled loadflow
tions are present because the problem addressed is a dis-
method (FDLF) [4]. Another technique dedicated to the tribution network which has only one voltage-controlled
analysis of power distribution networks is the method busbar at the main input node or the reference node.
developed by Baran and Wu [I]. The development of These equations can easily be modified to take into
this method is revolutionary in view of its superior fea-
account systems with more than one voltage-controlled
tures for application in radial networks loadflow analysis.
busbars. Eqn. 4 forms the basis of the SONR method
The theory of the SONR and FDLF is briefly pre-
used for one method presented.
sented and the theory of the third loadflow method, i.e.
distflow method, is presented in detail. The criterion for
voltage collapse will then be presented and the ways in 2.2 Fast decoupled loadflow method [4]
which the loadflow methods can predict the voltage col- Consider the decoupled loadflow equations
lapse point are also discussed. AP= HAS

2 AC loadflow techniques AQ =L [7]


The theoretical basis for the three loadflow methods to
be used for comparison are briefly outlined. where the elements of the submatrices H and L can be
expressed as
2.1 Second order Newton-Raphson method [2]
Hij = K(Gij sin S i j - B , cos S i j ) i # j (7)
The loadflow equations, in rectangular co-ordinates, for
each node in a power network are H . . = - B . .,I V2I - Q.
I i =j (8)
L, = Y(Gij sin 6 , - Bij cos S i j ) = H , i #j (9)
Pi= CCe i e j G i j- e , f j B , + J A G i j +JejBij)
j= 1 L.. = - E . .I, V I? + Q I. i = j (10)
N
Qi = ( e , f j G i j- f i f ; B i j - e i f j G i j- e i e j B i j ) (1) Even though the developed loadflow method reduces the
j= 1 memory storage requirements considerably, it still
requires significant amounts of computational effort.
where N is the number of nodes in the network. Using
the Taylor series expansion of these equations, the Therefore, Stot and Alsac [SI have developed the fast
decoupled loadflow method. The basic assumptions used
general form of the expression is
are the following:
AS)= f i x ) + J ( x )Ax, + H(x)(Axi. Axj)/2 (2) (i) The power systems have high x / r ratios
where G i j sin 6, < Bij (1 1)
(ii) The difference between adjacent bus voltage angle
is very small
sin 6.. = sin (Si - 6.) E 6. - 6 . = 6 . .
J - 8 J 'J

cos Si,. = cos (Si - Sj) 1.0 (12)


and (iii) Also
Axl Axl Qi < E,, V: (13)
( A X ,. A x j ) = A x l Ax2 Therefore eqns. 5 and 6 can be further approximated as
CAP] = [ V x B x V][AS] (14)
Ax,, Axn
It can be shown [2] that the third and higher order terms
of the Taylor series expansion vanishes because the load-
flow equation is quadratic. According to Iwamoto [7],
[AQ] = [V x B x V ]
["V"l
-

where the elements of the matrices B' and B" are the ele-
the second order terms of the equation can be simplified, ments of the matrix -B. Therefore
thus resulting in the form
Bij = -l/X,j i #j (16)
Axkt' = J(xe)-'(y(s) - A x e ) - AX')) (3) U

where superscript k indicates the iteration number. Bii= 1 1 / X . . i = j (17)


j=l
If these variables are replaced with the normal load-
flow variables, the second order loadflow equations are E!'. = - E . . (18)
480 IEE PROCEEDINGS-C, Vol. 138. No. 6, NOVEMBER 1991
The decoupling process in the fast decoupled loadflow The next iteration uses values of real and reactive
can be concluded after additional modifications based on power injections by using eqns. 30 and 31
the simplifying assumptions as
,
Pi+ = ( 2 x 2 P , - 2rxQ, + r)/(2(r2+ x2))
A P / V = IT A6 (19) + r)’ - 4(r2 + x2)
- ( ( 2 x z P , - 2rxQ,
AQJV = B” AV (20) x (x’P, + r 2 Q , - 2rxPlQ1 + rPl))1’2
This forms the basis of the fast decoupled loadflow
method used. - (2(r2 + x’)) (30)
Q i + , = (2r2Q, - 2rxPl + x)/(2(r2+ x’))
2.3 Distflow method [ l ]
The equations for the loadflow solution as used in the - ((2x2P1- 2rxQl + r)’ - 4(r2 + x’)
method by Baran and Wu [l] is derived. Fig. 1 shows a x (x2Pl + r Z Q l - 2rxP,Ql + rP1))112
i (2(r2 + x’)) (31)
“Le “1 L
P The values of req and xeq can be obtained by using the
0 Z& 1
Fig. 1 Distribution line
previous power injection of Pi j Q i . +
req = R/(PZ + Qt) (32)
line connecting bus 0 to bus 1 having a line impedance of xes = X / ( P t + Qz) (33)
+
r jx.
From Fig. 1, it can be shown that 2.3.1 Loadflow solution algorithm: The fundamental
equations for solving a loadflow problem of a distribu-
Real power loss: tion network using a single-line equivalent has been
derived. The loadflow algorithm using these fundamental
R = Z COS tr(P; + Q@/V’ equations can be set up using the following steps:
= r(P; + Qg)/V2 ( a ) Start the initial iteration by using the total real
loads and reactive loads as the initial power injection.
Reactive power loss: (b) Sum all the real and reactive losses and find the
equivalent resistance, req, and reactance, x e q , for a single
X = Z sin a(Pg + Q@/V’ line system from eqns. 32 and 33.
= x(Pi + Q@/V’ ( c ) Calculate the new power injection by using eqns. 30
and 31. If P i + l - Pi < e, then go to step (e) where e is a
and, set tolerance, else go to step (d).
V: = V; - 2(rPo + xQo) + (r’ + x’)(Pi + Qg)/Vg (23) (d) Iterate with new power injection from step (c), then
go to step (b).
Thus for a line as in Fig. 2 where there is a cumulative ( e ) Calculate other parameters required, e.g. voltages.
load at the end of the line, i.e. P,, j Q L l : + The voltage can be calculated from the losses and power
injections in the individual lines from eqn. 29.
P I = Po - r(P; + Q g ) / V 2- PL1 (24)
It has been found that the proposed loadflow requires
Qi = Qo - 4% + Qi)/v’- Q L ~ (25) fewer iterations than other known techniques.
V: = Vg - 2(rPo + xQ,) + (r’ + x2)(Pi + Q;)/Vg (26)
3 Criterion f o r stability [8]
r+jx
3.1 Mathematical formulation of technique
“0
3.1.1 Governing equations of a single-line system:
Pi *IQ1

po+ Pl*Ql
Before proceeding into the actual system, we first derive
the equations that characterise the behaviour of a single-
line system. Consider the single line in Fig. 1 which has
the parameters shown in Fig. 3.
Fig. 2 Example of distribution line

A more general form of eqns. 24-26, between two


+
busbars i and i 1 can therefore be derived [I] as
P i + , = Pi - r i p : + Q:)/V: - P , i + l (27)
Qi+ 1 = Qi - Xi(PZ + QZ)/V? - Q L I + I (28)
V;+ = V : - 2(riPi+ xi Qi) Fig. 3 Single line system

+ +
(r: xZ)(P? Q:)/V: + (29)
The proposed method can also be used for loadflow From Fig. 1, the real and reactive power equations
analysis simply by iterating on the loss terms. The dis- have been derived in Reference 1 as
tribution network can be reduced into its single line
P = r(P2 + Q 2 ) / V 2+ P I (34)
equivalent by calculating the equivalent resistance, rer ,
and reactance, xeqrfrom the total real and reactive losses. Q = x(P2 + Q 2 ) / V 2+ Q 1 (35)
IEE PROCEEDINGS-C, Vol. 138, No. 6, NOVEMBER 1991 48 1
From eqns. 1 and 2, the (P’ + Q 2 ) / V 2 terms can be From eqns. 45 and 46
eliminated thus obtaining
R i + , / R i = (ri+l/riXPZ+l+ Q f + J
x(P - Pi) = r(Q - Q i ) (36)
By rearranging eqn. 6, and eliminating Q in eqn. 4, a Similarly for reactive losses
quadratic equation in terms of P is obtained
X i + i / X i =(xi+i/xiXP?+i + Q ? + i )
(r’ + x’)P’ - (2x2P1- 2rxQl + r)P
+ (x’P, + r’Q1 - 2rxPlQl + r P l ) = 0 (37)
For a given distribution network
The voltage at the sending end is the reference voltage
and its magnitude is kept constant, and in this case (49)
V’ = 1 p.u. (50)
Hence, from eqn. 37
From eqns. 47 and 48,it can be seen that the losses in the
P = ( 2 x 2 P 1- ZrxQ, + r)/(2(r2+ x’)) distribution network are ratios of the losses in the first
+ r)’ - 4(r2 + x’)
- ( ( 2 x z P 1- 2rxQl line of the network.
Hence
x (x’P, + rZQ1- 2rxPlQl + rPl))li’ P = r,(P’ + Q2)+ 1 P l i
x (2(rZ + x’)) (38)
Q = xeq(PZ+ Q’) + Q i i
Similarly for reactive power Q , because of symmetry of
where req and xeq is the equivalent resistance and reac-
equations, the reactive power equation can be derived as
tance in the single line. Hence the real distribution
Q = (2rZQ1- 2rxPl + x)/(2(r2+ x’)) network consisting of many lines has been reduced into a
system with only one line.
- ((2x2P1- + r)’ - 4(rZ + x’)
2rxQl
x (x’P, + r’Q, - 2rxPlQl + rP1))’/’ 3.1.3 Voltage collapse: By using the single line method
for reducing a distribution network, the occurrence of
x (2(r2 + 2)) (39) voltage collapse can be studied easily as every line in the
network does not have to be considered.
The above equations are quadratic in form and for P and Recall from eqn. 40, for P and Q to have real roots
Q to have real roots
(2x2P1- 2rxQl + r)’ - 4(r2 + x’) L = 4((xPl - r e l ) ’ + XQl + r P l )
where L < 1.0.
x (X’P, + r2Q1- 2 r x P l Q l + r P l ) > 0 Hence for the reduced network
which on simplification can be reduced to L=4((xeqPi -reqQi)’+xeqQi +reqPi) (53)
4((xPl - rQ1)’ + x Q , + r P l ) -= 1 (40) If the network is loaded beyond this critical limit, the
power becomes imaginary and it is at this point that
3.1.2 Reduction of real network to single line equiva-
voltage collapses.
lent: A given power distribution network can be reduced
to a single-line equivalent. From Reference 1, the real and 4 Performance test results b e t w e e n t h r e e
reactive power flows in any line are given by l o a d f l o w methods

Pi+ = Pi - r,(P? + Q?)/VZ - P l i + (41) The three loadflow methods have been tested on several
distribution systems. The loadflow data and results for
Qi+i = Qi - xAPZ + QWZ- Q,i+i (42)
the 32-node test system from Reference 1 are given as
The real and reactive loss terms are Tables la and lb. The following comparisons can be
made :
Ri = r i p ? + QZ)/VZ (43)
4.1 Speed of convergence
X i = x,(P? + QZ)/V? (44) Both the SONR and distflow methods require about the
Using eqn. 42, the ratio of real losses between line i and same number of iterations to converge to a tolerance of
+
preceeding line i 1 can be computed as O.OOO1 on the real and reactive power accuracy, but
FDLF requires about three times more. The relative
Ri+ i/Ri = (ri+ i(P?+1 + Q?+ iW?+ 1) speed is given in Table 2.
x (rdp? + Q?)/V?) The distflow method does not require any matrix
inversion and as a result the computation time is con-
= (ri+i(P?+i + Q?+ 1)) siderably less.
x (rXPZ + QZ)Xv?/v?+i) (45) 4.2 Accuracy of solution
By considering the current flow in the line The accuracy of the methods depend on the tolerance
specified and for all the results presented here a tolerance
(P? + QZYVZ = ((Pi+1 + PI)’ + ( Q i + 1 + Ql)’)/VZ+ 1 of O.OOO1 on the power is used. All the results are exactly
which gives the same and exact (depending on the tolerance).
4.3 Storage requirements
VZ/vZ+i = U?
‘ + QZ)/((Pi+i + Pi)’ + ( Q i + i + Q i ) ’ ) It is obvious that SONR requires more storage than dist-
(46) flow because the Jacobian matrix has to be formulated
482 IEE PROCEEDINGS-C, Vol. 138, No. 6 , NOVEMBER 1991
Table l a : Test system
Branch Receiving Sending P, (kW) 0,(kVar) Branch parameters
number node node
R (D.u.) X (D.u.) SusceDtance (D.u.)
1 0 1 100 1318 0.0058 0.0029 0.0052
2 1 2 90 750 0.0308 0.0157 0.0277
3 2 3 120 479 0.0228 0.0116 0.0206
4 3 4 60 660 0.0238 0.0121 0.0214
5 4 5 60 630 0.0511 0.0441 0.0460
6 5 6 200 1049 0.0117 0.0386 0.0105
7 6 7 200 1425 0.1068 0.0771 0.0961
8 7 8 160 1106 0.0643 0.0462 0.0578
9 8 9 60 602 0.0651 0.0462 0.0586
10 9 10 45 298 0.0123 0.0041 0.0110
11 10 11 60 895 0.0234 0.0077 0.0210
12 11 12 60 961 0.0916 0.0721 0.0824
13 12 13 120 599 0.0338 0.0445 0.0304
14 13 14 60 621 0.0369 0.0328 0.0332
15 14 15 60 947 0.0466 0.0340 0.0419
16 15 16 60 936 0.0804 0.1074 0.0724
17 16 17 90 371 0.0457 0.0358 0.0411
18 1 18 90 970 0.0102 0.0098 0.0092
19 18 19 90 1099 0.0939 0.0846 0.0845
20 19 20 90 658 0.0255 0.0298 0.0230
21 20 21 90 431 0.0442 0.0585 0.0398
22 2 22 90 776 0.0282 0.0192 0.0253
23 22 23 420 1053 0.0560 0.0442 0.0504
24 23 24 420 673 0.0559 0.0437 0.0503
25 5 25 60 495 0.0127 0.0065 0.0114
26 25 26 60 772 0.0177 0.0090 0.0160
27 26 27 60 963 0.0661 0.0583 0.0595
28 27 28 120 684 0.0502 0.0437 0.0452
29 28 29 100 732 0.0317 0.0161 0.0285
30 29 30 150 909 0.0608 0.0601 0.0547
31 30 31 21 0 408 0.0194 0.0226 0.0174
32 31 32 100 261 0.0213 0.0331 0.0191

Table 16: Loadflow results


Branch Receiving Sending Sending node Power loss
number node node
p, (P.U.) Q, (P.U.) I VI P (kW) (kVar)
1 0 1 0.0100 0.1318 0.9967 0.3859 0.2401 0.0015 0.0008
2 1 2 0.0090 0.0750 0.9824 0.3354 0.2173 0.0053 0.0027
3 2 3 0.0120 0.0479 0.9747 0.2273 0.1725 0.0022 0.0011
4 3 4 0.0060 0.0660 0.9670 0.2193 0.1685 0.0020 0.0010
5 4 5 0.0060 0.0630 0.9477 0.2091 0.1629 0.0042 0.0036
6 5 6 0.0200 0.1049 0.9436 0.1006 0.0533 0.0002 0.0008
7 6 7 0.0200 0.1425 0.9279 0.0791 0.0421 0.0016 0.0011
8 7 8 0.0160 0.1106 0.9203 0.0625 0.0297 0.0006 0.0004
9 8 9 0.0060 0.0602 0.9144 0.0561 0.0275 0.0004 0.0003
10 9 10 0.0045 0.0298 0.9135 0.0515 0.0244 0.0001 0.0000
11 10 11 0.0060 0.0895 0.9120 0.0454 0.0209 0.0001 0.0000
12 11 12 0.0060 0.0961 0.9058 0.0392 0.0172 0.0003 0.0002
13 12 13 0.0120 0.0599 0.9035 0.0271 0.0091 0.0001 0.0001
14 13 14 0.0060 0.0621 0.9020 0.0211 0.0081 0.0000 0.0000
15 14 15 0.0060 0.0947 0.9007 0.0150 0.0060 0.0000 0.0000
16 15 16 0.0060 0.0936 0.8986 0.0090 0.0040 0.0000 0.0000
17 16 17 0.0090 0.0371 0.8980 -0.0000 -0.0000 0.0000 0.0000
18 1 18 0.0090 0.0970 0.9962 0.0271 0.0121 o.oo00 0.0000
19 18 19 0.0090 0.1099 0.9926 0.0180 0.0080 0.0001 0.0001
20 19 20 0.0090 0.0658 0.9919 0.0090 0.0040 0.0000 0.0000
21 20 21 0.0090 0.0431 0.9913 -0.0000 -0.0000 0.0000 0.0000
22 2 22 0.0090 0.0776 0.9790 0.0846 0.0305 0.0003 0.0002
23 22 23 0.0420 0.1053 0.9728 0.0421 0.0201 0.0005 O.OOO4
24 23 24 0.0420 0.0673 0.9695 -0.0000 -0.0000 0.0001 0.0001
25 5 25 0.0060 0.0495 0.9459 0.0820 0.0737 0.0002 0.0001
26 25 26 0.0060 0.0772 0.9436 0.0758 0.0635 0.0002 0.0001
27 26 27 0.0060 0.0963 0.9344 0.0691 0.0579 0.0007 0.0006
28 27 28 0.0120 0.0684 0.9280 0.0566 0.0525 0.0005 0.0004
29 28 29 0.0100 0.0732 0.9251 0.0464 0.0504 0.0002 0.0001
30 29 30 0.0150 0.0909 0.9188 0.0310 0.0200 0.0003 0.0003
31 30 31 0.0210 0.0408 0.9177 0.0100 0.0100 0.0000 o.oo00
32 31 32 0.0100 0.0261 0.9171 -0.0000 -0.0000 0.0000 0.0000

Table 2: Speed comparison and subsequently inverted for SONR. The distflow
SONR: 1.5 method does not require the formation of a Jacobian
Distflow: 1 matrix or any matrix inversion. In fact, this is the greatest
FDLF: 3 disadvantage of SONR particularly because the Jacobian
I E E PROCEEDINGS-C, Vol. 138, N o . 6, N O V E M B E R 1991 483
matrix for radial networks are usually sparse. Both as its ability to produce a loadflow solution close to the
SONR and FDLF requires more storage as given in voltage collapse point.
Table 3.

Table 3:S t o r a g e comparison 6 References

SONR: 8 1 BARAN, M.E., and WU, F.F.: ‘Network reconfiguration in distribu-


Distflow: 1 tion systems for loss reduction and load balancing’, IEEE Trans.,
FDLF: 4 1989, PWRD-4, pp. 140-1407
2 JASMON, G.B., AMIN, R.M., and CHUAN, C.Y.: ‘Performance
comparison of two exact outage simulation techniques’, IEE Proc. C ,
1985,132, (6), pp. 285-293
4.4 Effect of loading up io voltage collapse point 3 LAUGHTON, M.A., and HUMPHREY DAVIES, M.W.: ‘Numeri-
All three loadflows have been tested to see whether they cal techniques in the solution of power system loadflow problems’,
fail at the voltage collapse point. From the test run, it IEE Proc., 1964,111, (9), pp. 1575-1587
4 STOTT, B., and ALSAC, 0.:‘Fast decoupled loadflow’, IEEE
was found that all the three methods gave the same Trans., 1974, PAS-93, pp. 859-869
results, i.e. they can produce a solution for any condition 5 STOlT, B.: ‘Review of loadflow calculation methods’, Proc. IEEE,
up to the point when the stability factor approaches close. 1974.62, (7), pp. 916-929
to unity. 6 BROWNELL, G., and CLARK, H.: ‘Analysis and solutions for bulk
system voltage instability’, IEEE Comput. Appl. Power, 1989, 2, (3),
pp. 31-35
5 Conclusion 7 IWAMOTO, S., and TAMURA, Y.: ‘A fast loadflow method
retaining nonlinearity’, IEEE Tram., 1978, PAS-97, pp. 1586-1599
The performance of three AC loadflow techniques has 8 JASMON, G.B., and LEE, L.H.C.C.: ‘Distribution network reduction
been compared with respect to their ability to solve dis- for voltage instability analysis and loadflow calculations’, Int. J.
Electr. Power Energy Syst., 1991,13, (I), pp. e 1 3
tribution network voltage stability problems. The dist- 9 BEGOVIC, M.M., and PHADKE, A.G.: ‘Voltage assessment
flow method has been demonstrated to be the most through measurement of a reduced state vector’, IEEE Trans., 1990,
suitable method in view of its speed and storage as well PAS-5, pp. 198-203

484 IEE PROCEEDINGS-C, Vol. 138, No. 6, NOVEMBER 1991

I-

You might also like