Professional Documents
Culture Documents
1977 - P Lin - Topologicalgenerationandanalysisofvoltagemultiplie (Retrieved-2015-08-30)
1977 - P Lin - Topologicalgenerationandanalysisofvoltagemultiplie (Retrieved-2015-08-30)
I. INTRODUCTION
has a much larger capacitance than the others in the Fig. 2. (a) Ladder C-D network. (b) Output tank.
ladder. Two tank connections for realizing even order and
odd order multipliers are shown in Fig. 3(a) and (b),
respectively [5].
As an aid for understanding the operating mechanisms
of voltage-multiplier circuits, let us first review some basic
Manuscript received November 17, 1976; revised April 15, 1977. This
work was supported in part by the Naval Electronic Systems Command
under Contract NBOO39-77-C-0022 and in part by the National Science
Foundation Grant ENG75-22282. L. 0. Chua’s research during the
1976-1977 academic year was supported by the Miller Institute as a
Miller Research Professor.
P. M. Lin is with the School of Electrical Engineering, Purdue Univer-
sity, Lafayette, IN 47907. The work of P. M. Lin was performed during
iris visit to the University of California, Berkeley, while on sabbatical
leave in 1976 From Purdue University.
L. 0. Chua is with the Department of Electrical Engineering and
Computer Sciences and the Electronics Research Laboratory, University
of Caliiornia, Berkeley, CA 94720.
‘We assume alF diodes are ided throughout this paper, i.e., the diode (b)
voltage oo and the diode current iD satisfy the constraints co $0, iD > 0, ,
and oDiD = 0. Fig. 3. N-Fold voltage multipliers derived from C-D ladder network.
518 IEEE TRANSACTIONS ON CIRCUITS F SYSTEMS, VOL. CAS-24, NO. lo, OCTOBER 1977
v,=Eo+-‘&++;“tlvctt,
Fig. 4. Simple rectifier circuit.
6
+_ 2Osinwt
n, C-branches, respectively, where no > 0, n, > 0, and n, + of components. With this practical consideration in mind,
n, = n. Now in the enumeration of all Tc, and T,, sub- we see that the diode D, in Fig. 6(c) is poorly located
trees, the nodes 0 and 1 are different from the rest because because its fundamental loop relative to the C-E tree
they serve as reference nodes and will henceforth be contains only capacitors (and D, of course). Since the ac
called the roots of T,, and Tc,, respectively. Given any n, voltage source is not included in the loop, we have a
and n, such that n= n,+ n,, our problem is to find all situation of charging some capacitors by other capacitors.
distinct combinations of rooted trees T,, containing n, We can not expect the highest dc voltage to be enhanced
branches, and all rooted trees T,, containing n, branches. by such a mechanism. Therefore, it is reasonable to con-
This problem can therefore be identified with the problem nect every diode only from one node of T,, to another
of enumerating “unlabeled rooted trees” for which well- node of T,,. In this way, every fundamental loop
known methods exist [6, p. 2461. In particular, if w, associated with the diodes relative to the C-E tree will
denotes the number of unlabeled rooted trees of k contain the voltage source. This procedure further implies
branches, then the value of wk corresponding to the first 9 that the fundamental cutset defined by the voltage source
values of k are listed below relative to the C-E tree will contain all the diodes [7].
Therefore, we have now justified the imposition of a third
k 1 234 5 6 7 8 9 hypothesis.
w, 1 2 4 9 20 48 115 286 719’ 3) D-E Cutset Hypothesis: The diodes and the voltage
source form a cutset of the network.
Applying this table to Fig. 7, we see that we have 1 rooted Next, let us review the action of the voltage multiplier
tree T,, of 1 branch, 2 rooted trees T,, of 2 branches, and of Fig. 3(a). When D, alone is connected to the C-E
4 rooted trees T,, of 3 branches. The value of w, for any k ladder, C, will be charged to Vc, = E. At this time, the
can be calculated by the use of a counting series [6, p. 2471 voltage drop from node 2 to node 1 is
which we will not elaborate here. The above list should
suffice for our present discussions. u*,(t)= E+ E sin wt
To get the total number of K-1 E trees we should:
1) consider all possible ways of partitioning n into which has positive peak of 2E. By connecting D, from
node 2 to node 3, we utilize the 2E peak voltage to charge
n=,n,+n,;
2) for each (n,,n,) pair, enumerate the unlabeled rooted C, and obtain Vc2 = 2E. At this point in time, if we
trees T,, and T,, ; connect another diode from node 2 to some other capaci-
3) consider all possible combinations of T,, and T,, tor, the voltage available for charging the capacitor is at -
most 2E. Hence, nothing is to be gained by connecting
which, together with E, form a C-E tree, and take care
to avoid duplications.
It can be shown that the result is
t,, L number of distinct K-1 E trees
=
w,+w,w,-,+“*+W(n-I)/2W(n+1)/2, for odd n
wn + WIWn-, + . . . + W((n/2)- I)W((n/2)+ I)+tw(n,*)bqn,2)+
l), for even n. (2)
The first seven values of t, are listed below
nl 23 4 5 6 7
4 1 3 6 16 37 96 237.
Observe that for the example in Fig. 7, we have n = 3 and more diodes to node 2. Thus in order to make the best use
hence we expect to have 6 distinct 3 C- 1E trees. Once a of every diode, it is advisable not to have more than two
particular C-E tree configuration has been selected, there diodes connected to every node. This, together with the
are still many ways to connect the n diodes to form a D-E D-E cutset hypothesis, suggests that a) the number of the
tree. The combinatorial problem becomes extremely com- capacitors in the two subtrees T,, and T,, differ by no
plicated. Fortunately some practical considerations de- more than 1. b) The diodes are contained in a path. We
scribed below will further restrict the class of C-E trees have therefore justified the imposition of still another
and D-E trees for our purpose. topological constraint.3
Consider the three circuits shown in Fig. 6, each of 4) Aligned D-Path Hypothesis: The diodes, possibly with
which uses 4 capacitors and 4 diodes. Fig. 6(a) is a voltage the voltage source, form a path and are forward biased in
quadrupler whereas the other two circuits can at. most the same direction in the path.
function as voltage doublers. Naturally, we are only inter-
ested in voltage-multiplier circuits which produce the ‘Hypotheses I) through 4) imply that at most two diodes are con-
highest dc output voltage possible with the given number nected to each node.
LIN AND CIiUA: ANALhIS OF VOLTAGE MULTTPLIER CIRCUITS 521
0 and
6’ At-J
v,<O, i,>O, vziD=O.
‘b b3
k-5
(a)
4+5
(b)
Equations (4H7) must be satisfied for all t, although we
are looking only at t-co. Assume that the steady-state
capacitor voltages are constant, i.e., vc(oo)= V,. We shall
outline a procedure for finding V,. From (6), we have
i&co) =O. Since the D-E elements also form a tree TOE,
we can express iE and iD in terms of the link (for TDE)
currents ic. Then, ic =0 implies iE =0 and iD = 0. It follows
that all currents are zero in steady state, a result that
obviously satisfies KCL.
Ld)
Letting v, = Vc, vE = E sin wt (E > 0), we have from (4)
and (7)
Ot-&l v,(t)=-BEEsin&-BB,V,<O, for all 1.
(8)
E3 ‘i”s
It is easy to see that the inequality
and only if,
(8) holds for all t if,
m (9)
(j)
Fig. 9. Generation of voltage quadruplers by Algorithm VM. where U is an n-vector with l’s and O’s as its elements
such that
III. STEADY-STATE ANALYSIS-WITHOUT LOAD if the jth element of BE is either 1 or - 1
uj = 1,
Consider an nC-nD-1 E,, network N which satisfies the i 0, otherwise
C-E tree and D-E tree hypotheses. To facilitate the for-
mulation of the equilibrium equations for N, we construct We have succeeded in obtaining an inequality devoid of
G,, the directed graph associated with N. In Cd, let each the time variable t. We will show that Inequality (9) is the
diode branch direction be the same as the forward current crux to the steady-state analysis of N.
direction. Branch directions for the capacitors and the .The matrix B, is a square matrix of order n. Since the
voltage source may be arbitrarily assigned. Then the capacitors form a cotree for the D-E tree, B, is nonsingu-
fundamental loop matrix with respect to the C-E tree may lar [8, p. 931. At this point, it is rather tempting to ignore
be partitioned as follows (see [7] for notation): the “ > ” sign in Inequality (9) and to find V, by solving
B, V, = UE and accepting the result as the steady-state
E C,;..,C,, D,;..,D,, solution. Unfortunately, such a procedure has no theoreti-
B= [ BE ; B, ; 1, 1. cal foundation. It may or may not give the correct
(3)
answers, as the following two simple examples will dem-
C-E tree Dxee onstrate.
The current and voltage vectors are denoted by iE, ic, iD,
VE, v,, and vD, respectively. Throughout the remaining Example I
sections, we use small letters for functions of time, and Consider the voltage doubler shown in Fig. 1. We have
capital letters for constants. The network is governed by
three laws which are expressed by matrix equations as
follows.
B=[B, B, ln]= y; :, y .
I 1
Kirchhoff Voltage Law (KVL): Then U=[ 1 llT, and ec 2 B;‘UE=[E,2ElT which
B,I$E + BcVc + V, ~0.
is indeed the correct solution for Vc.
(4)
Kirchhoff Current Law (KCL) [ 7, p. 1453: Example 2
Consider the circuit shown in Fig. 10. We have
(5) ’ 1 0 0’ 1 0 0
B=[B, B, &I= -1 I-I 1 o’ o 1 o
Branch u-i relationships: 5 -1 IO 1 II0 0 1
dvc
ic=C- (6)
dt Then U=[l 1 l]‘, tic A B,-‘UE=[E,2E, - EIT,
which is obviodsly not a solution, as V,, can never be
51n the paper, cz> b means n, > b, for all i. negative in Fig. 10.
LIN AND CHUA: ANALYSIS OF VOLTAGE MULTIPLIER CIRCUITS 523
c2
Theorem I
Let N be an nC-nD-1 E,, network which satisfies the
‘ollowing topological constraints:
a) the capacitors and the voltage source form a tree;
b) the diodes and the voltage source form a tree;
c) the diodes and the voltage source form a cutset;
d) the diodes, possibly with the voltage source, form
a path and are forward biased in the same direction
Fig. 12. Voltage tripler defying intuitive method of solution.
in the path.
If u,(t)= E sin at, E >0 and the capacitors are initially
.
easily. The following lemma is of great usefulness in the uncharged, then in steady state the capacrtor voltages
present study. are constant, and are given by
Lemma I Vq=+kjE
Let A be an n X n nonsingufar matrix, and K an n-vector where kj denotes the number of diodes contained in the
where K > 0. If for each row of A -I, all nonzero elements Fundamental loop defined by the capacitor Ci with
have the same sign, then the inequality AX > K has one, respect to the diodes-voltage-source tree, and where the
and on& one, irreducible solution given by X= A _ ‘K: positive sign holds when the reference direction of Vci
Proof: See [9]. is aligned with the diode’s forward directions, and is
We shall now illustrate the use of Inequality (9) and rlegatiue otherwise.
Lemma I with an example.
B=[B,B,l,]=
-1;
1:
0
1; 01
The matrix B, and vector U associated with the Inequal-
1
00
0
1
0. 1 B= [ \ Bf
C-E tree
,: BE : ‘n
D-cotree
] (10)
The inverse of B, is
g=[ 1, I &! 6D 1. (12)
--
C-cotree D-E tree
From BQT=O [8, p. 971, we obtain
B,+Q&=O (13)
Since the nonzero elements in each row of B; ’ have the
same sign, it follows from Lemma I that the unique and
irreducible solution is given by B,+Q,T=O. (14)
V,= B,-‘z/E= [ -2E,2E,3E] From iQT=O, we obtain
and this solution is the steady-state so’ution of this circuit. l,+bDQ,T=O. (15)
We see that the circuit is a voltage tnpler. Observe that it Substituting QCT from (13) into (15), we obtain
is not possible to find the steady-state voltages of this
circuit by the intuitive reasoning described in Section I. 1, - f&B,=O. (16)
We shall now present a theorem which will once again Since B, is nonsingular [8, p. 931, (16) yields
make it possible for us to obtain the steady-state solution
by inspection. The theorem is a major result of this paper, pq. (17)
and therefore will be stated clearly in words.
LIN AND CHUA: ANALYSIS OF VOLTAGE MULT’IPLIER CIRCUITS 525
R”= -al,
(23)
where R, has the unit of resistance and will henceforth be
called the output resistance of this voltage multiplier. Not-
ice that the negative sign in (23) is due to our reference
direction of IL in Fig. 13(b) which is out of the one-port
seen by the load resistor R,. Note also that (23) implicitly
assumes C,-+ cc. I (d)
We shall now describe a method for calculating R, for
any voltage multiplier N yM generated by Algorithm VM. Fig. 14. Current waveforms in output circuit.
To facilitate the discussions we shall assume without loss
of generality that the reference polarity of the input volt-
during the time interval ($ -At,t,) (respectively, (t, -
age v,(t) has been assigned in such a way that D,,-the
At, t,)), it follows that
diode in series with the output capacitor C,,- conducts
during positive peaks of u,(t) = E sin wt. We will let $, and Aqo+-=0 (27)
t, denote the positive peak and negative peak instants of Aq;+ = 0.
q(t), i.e., vi($)= E and v,(t,J= -E. The calculation of R,
is accomplished in two stages. Our next task is to determine the remaining surge charges
Ad, A&, Aq; , and Aq; in terms of AqL. Consider a
Stage I-Calculation of Surge Charges typical node “rn” as shown in Fig. 13. Kirchhoff current
During each period T=2a/w, the charge passing law requires that
through the load R, is given by for all t.
ic,(t)-i,,(t)-iC2(t)=0, (28)
AqL = IL. T= IL/f (24) If we iutegrate both sides of (28) from $,-At to tp, we
where I, is a constant current in view of our assumption obtain
that C,,+co. It was shown in Section III that when there
is no ‘load (IL =O), all currents in NV, are zero in steady
Ad, - Ad, - Aq& = 0. (29)
state. However, when a load current is drawn (ZL#O) Equation (29) is exactly of the same form as (28). This
thereby causing the output voltage to drop slightly, cur- means that KCL must be satisfied by the surge charges,
rents will also be flowing in the other branches of the provided they are referred to the same time interval.
network for short time intervals At just before the positive Although the capacitor voltages uc,, vc2,. . . , v,-~_, now
and negative peaks of v,(t). The surge charges through a vary as a function of time, they must nevertheless have
branch with current i(t) are therefore given, respectively, constant average values in steady state. This implies that
by Aq,:i+Aqcj=O, j=l;..,n-1
or
Aq<j = -Aqzj, j=l;.*,n-I. (30)
and The condition prevailing at the output,tank- consisting
of the output capacitor C,, and the load resistor R,, Fig.
Aq- k
J/,-Al‘I i(7) dr. 14(a)-requires special attention. Refer to Fig. 14(b)-(d),
and observe that when the diode D,, conducts near tp,a
The diodes D,,,D,,-2,. ++ will be collectively denoted by surge current flows into C,,. But immediately after tp,the
D, to emphasize that they conduct near the positive peak diode D,, is cut-off. Since C,, is assumed to be infinite, the
of v;(t) from tp - At to tp. Similarly, the diodes voltage vcn is equal to a constant Vcn (the steady-state
D,,-,,Q,-3,. .. will be denoted by D- to emphasize that voltage). This voltage maintains the constant load current
they conduct near the negative peaks of q(t). Since the IL = - i, = V,,/R, throughout the whole period. ,Since
diodes belonging to D _ (respectively, D+) are cut-off there should be no net increase or decrease of the charge
LIN AND CHUA: ANALYSIS OF VOLTAGE MULTIPLIER CIRC”,TS 521
on C, in steady state, it follows from Fig. 14(a) that Using QAq’ =0 and Aqg- =0 from (27) we obtain
whatever surge charge Aq& injected into C,, during the
Aq,+ -(II,+ )‘AqD’, =O. (37)
short interval At near the positive peak tp must be trans-
ferred to the load resistor R,; namely, Substituting (35) for Aqg+ in (37) we obtain
Aq&= I,.T=Aq,. (31)
Aqc+ = (B,+ )‘. Aq;+ = (B,+ )=. AqL. (38)
Since D,, is connected in series with C,,, KCL requires that
Ad, = Ad,, =h.- (32) Finally, we observe from (30) that Aq; is simply equal to
To determine other diode surge charges, consider first the -Aq,+. Note that Aq,;, does not appear as a surge (see
case n is even. The network N,, has an aligned diode Fig. 14(d)).
path starting from node 1 and terminating at node n + 1.
Example 4
Thus for each node j = 2,3,. . . , n - 1, there are exactly two
diodes connected to it, with one directed toward node j, Consider the voltage quadrupler circuit of Fig. 6(a) with
and the other away from node j. Besides, there are two or a load resistor connected across C,. The fundamental loop
more capacitors (the number of capacitors is unimportant matrix for this circuit is given by
in the argument) connected to node j. By way of illustra- C-E tree D, D-
tion, consider node 3 in Fig. 3(b). We have
C, G c3 Cd E DTD2 FD,
near $,: AqD+,-Aqc+, +Aq&=O (33)
near tn: - Aq,-, - Aq;, + Aq;s = 0. (34) B= i 0 0 -1 1’ -1; 1 0; 0 0 (39)
-1
------___ 1 00’ , -?I P- _‘c o-o_
Adding (33) and (44) and invoking (30) we obtain 0 -1 101 II 0 01 1 0
Ad, = Aq,. 1 0 0 01 II 0 01 0 1
Thus the two diode surge charges have the same magni- Hence we identify
tude although they occur at different time intervals. Ski-
lar situation applies at nodes 3,4,. . . ,n - 1 and also for
the case n is odd. Therefore we conclude that for NV,, all
-1
0
1
0 1
diode surge charges are equal to Aq,, i.e., if we define from (35): Aq;, = q&=Ab%,=Aq&=AqL
A!7D’+ g [Aq0+,, AqD+n-2, * . . 1’ and Aq& f from (38): Aq: =(Bg )‘AqL= [ -AqL,AqL, -Aq,,Aq,]’
W&i,-,,Aq,-,-3,. . . IT, then we have
from (30): Aq; = [ AqL, -AqL,AqL, -Aq,] ‘.
Aq;+ F Aq,_ = Aa. (35)
where AqL 42 [AqL, AqL,. . . , AqJT. Stage 2-Calculation of Capacitor Voltages
We shall now turn our attention to the calculation of In Section III, we have shown that the steady-state
AqC+. This is easily done by applying KCL to NV, near solution of NV, when IL = 0 must satisfy (18); namely,
t = $. During this time interval, all diodes in the D- group
B, Vc = UE. (40)
{Dn-,,Q-3,. . . > are cut-off, whereas each diode belong-
ing to the D, group { D,,, DnP2,. . * } behaves as a current Substituting this equation into (8), we observe that in
source. The capacitor surge charges Aq: can be expressed every scalar equation of (8), the equality sign will be
in terms of Aqz+ through a relationship similar to (5). An reached periodically, at positive peaks or at negative
explicit formula for Aqc will now be derived. peaks of u,(t). The physical meaning is that for this class
Further partition the fundamental loop matrix given by of networks, each diode reaches the breakpoint (iD =O,u,
(3) as follows: =0) periodically in steady state. Now when a load is
connected across C,, to draw current IL, the diodes will
C E D, D- not only reach the breakpoint, but will actually be for-
ward biased for a brief time interval. In particular, the
(36) diodes in the D, group {D,,, Dnm2,. . . } will be conduct-
ing currents from 5 -At to 5, while the diodes in the D-
C-E ‘tree D-&tree group {Dn-I,Q-3,.-- } will be conducting currents from
t, -At to t,,. Therefore, in applying (40), the scalar equa-
The associated fundamental cutset matrix is given by
tions should be separated into two groups, one applicable
CE D, D- at t = $, while the other applicable at t = t,,. The matrix B
-- in (36) has already been partitioned for this purpose.
Q=
-(B,+)T! -(B,-)’ Hence, (40) decomposes into two equations:
-(BE+)=; -(BE_)=
----I. B,+V,+=U+E (41)
D-co&& B,-V,-=U-E (42)
Let the surge charge vector Aq be similarly partitioned. where U+ and U- are associated with BE+ and BF in
528 IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS, VOL. CAS-24, NO. 10, OCTOBER 1977
(40) respectively. See the explanatory note in Inequality The explicit formulas (48) and (49) have been derived
(9) for the construction of U+ and U- from BE+ and B;, for VM circuits where the output capacitor C, is in series
respectively. with a diode D,,. Voltage-multiplier circuits generated by
For each capacitor q, the surge charge Aq; has been Algorithm VM may or may not have a series-connected
found earlier by the use of (38). The voltage difference C, - D,, output tank (see the tripler of Fig. 12 for an
VG - Vc; is therefore given by example of the latter case). It can be shown that (48) and
(49) are also valid for the latter case, and even for the case
V&J- Vc7 = Aq; / Cj, j= 1,2;. . ,n. (43) of finite (but large) C,,, provided that we consider V, to be
Again, the output capacitor C,, requires special attention. V+
Cfl’
Since C, is assumed to be infinitely large, we have VA = We shall now illustrate the use of the explicit formulas
Vc;l = V,. Equation (43) may be expressed more com- (48) and (49) with an example.
pactly as
Example 5
Vc’ - V; = SAq; (9 Let us continue with the voltage quadrupler circuit of
where Example 4 [Fig. 6(a)]. Suppose that C, = C,= C,= C and
c,+oo
Substituting
-1 0 01 0’ 0
the above matrices into (48) and (49) we
1
obtain
(46) 31,
(50)
S$ce B, is nonsingular, and its inverse is given by Bc ’ = V”=4E- fc
BD [see (17)], we obtain the following explicit formula for and
V +.
C’ RO=+ (51)
0 By the same method, we find the output resistance for the
V,+=i$JJE+i$, (47)
B,-StB,+ IT4 voltage quadrupler circuit of Fig. l(c) to be R, =6/fC.
II
The output resistance of Fig. l(c) can also be obtained
where from the equation given in [5, (3)] by letting n = 4. It is
important to note that RO for the new voltage quadrupler
Bg and B; are defined in (36)
circuit of Fig. 6(a) is only one-half of the conventional
U is defined in (9)
quadrupler of Fig. l(c).
S is defined in (44)
Using (49) we can prove the following theorem which
kqL is defined by (35) and (24)
BD is defined in (12). constitutes the second major result of this paper.
“=fE,
n-1 (1 T
lq (52)
the following explicit formula for the output resistance:
where mj is the number of diodes contained in the funda-
R, = - ( WTB;SBc+ TW)/f. (49) mental.cutset defined by Cj with respect to the C-E tree.
LIN AND CHUA: ANALYSIS OF VOLTAGE MULTIPLIER CIRCUITS 529
Proof In the derivation of (32) we have assumed that ply consists of all diodes connected between one node of
the output capacitor C,, is in series with the diode D,. This T s,= and another node of T,,.-,. Upon reviewing the
condition can be relaxed. By a reasoning similar to that properties a+z) described above, we immediately con-
given in (33~(34) we can easily see that Aq& =AqL, as clude that the number of diodes connected to T,-, is even,
long as one node of C,, has only one diode D,, and and are equally divided in the D, and D- groups. Thus
possibly several other capacitors connected to it. Since (56) is valid. As an example, consider Fig. l(c). For C,,
this is true for all circuits generated by Algorithm VM, we the fundamental cutset contains all four diodes, with
can repeat the derivation of the subsequent equations ml+ =2, m,- =2, and m, =4.
(33)<49). Therefore the explicit formulas (48) and (49) are Putting (56) into (55), we have (52). This completes the
valid for all circuits generated Algorithm VM. proof of Theorem 2. cl
Equation (49) has a topological interpretation. First, We observe that the expressions for the output resis-
note that from (49) tance as shown in [5, (1) and (2)] are just special cases of
Theorem 2. With Theorem 2, the problem of determining
R,=-~~[allelementsofB;S(B~)T]. (53) the output resistance is reduced to that of finding the
fundamental cutsets defined by the capacitors (with re-
Since S is a diagonal matrix (see (44)) the triple product spect to the C-E tree) and counting the diodes in the
B; ‘S(B,‘)T can be expanded quite easily to give the cutsets. In particular, if we assume C, = C, = . . . = C,- , =
contribution of 1/C, in the output resistance. We have C, then Theorem 2 leads to (see also [5, (3) and (4)]) the
6 term in (53) following expressions for the conventional ladder voltage
J multipliers:
= - -&, [jth column of BF][jth row of (B,+)T]
J
for n even, Fig. 3(a)
= L[no. of diodes in D- group whose f-loops contain Cj]
fq * (57)
. [no. of diodes in D, group whose f-loops contain Cj] for n odd, Fig. 3(b)
REFERENCES