The 47th IEEEInternational Mdwest

Symposium on Circuits and Systems

A Method for Circuit Analysis using Haar Wavelet Transform

Akira Ohkubo and Seiichiro Mor0 Tadashi Matsumoto
Department of Electrical and Electronics Engineering, Department of Electrical and Electronic Engineering,
University of Fukui Fukui University of Technology
Bunkyo 3-9-1 ,Fukui, 910-8507, Japan Gakuen 3-6-1, Fukui, 910-8505, Japan

Abstract-In this paper, we propose a method to analyze cir- 11. HAAR WAVELET
cuits using Haar wavelet transform. Recently, the method to an-
alyze the circuit using Daubechies wavelets has been proposed.
Haar function is defined on interval [0, 1 ) as follows,
Then a Fourier-like approach as well as Laplace-like one for the 1
solutions of transient problems by an algebraic system of equa- ho(t) = -
tions is obtained and numerical time stepping is avoided. In that
6
method, the matrices to express the integral and derivative are not
easy to handle. In the proposed method, we can easily treat such
matrices, thus the calculation and its comprehension become eas-
ier at the expense of more number of bases.

where a is positive integer and j and k are nonnegative integers

I. INTRODUCTION +
which satisfy i = 2j k , i.e., k = 0,1,. . . , 2 j - 1 for j =
0,1,2,....
For analysis of linear time invariant circuits, the analysis in The square integrable function y ( t ) defined on [0,1 ) is ex-
frequency domain using Fourier or Laplace transform is often panded to the following Haar series [ 2 ] ,
used, because it makes possible to analyze the differential equa-
tions in algebraic way. However, such analysis is not suitable
for time varying or nonlinear circuits. (3)
Therefore, for analysis of time varying or nonlinear circuits,
the time marching methods such as Euler method and Runge- (4)
Kutta method are commonly used. In such methods, however,
because the time step should be sufficiently small when the tra- Note that the set of the Haar bases is the set of the orthogonal
jectory moves rapidly at singular point, the calculation time functions. In matrix form,
will be wasted.
To overcome the problems described above, recently, much y" = F .H , or $ = H ~ c ' ( =~ -'g, (5)
attention have been paid to the method for circuit analysis us-
ing wavelet transform [7]-[ 151. The wavelet transform is often where y' is rn x 1-dimensional vector whose elements are the
used in signal processing because of its orthogonality and mul- discretized expression of y ( t ) and c'is m x 1-dimensional co-
tiresolution property. In particular, Barmada et al. have pro- efficient vector. H is m x rn-dimensional Haar wavelet matrix
posed the Fourier-like approach for the circuit analysis using
the wavelet transform [13]. In Ref. [13], the integral and dif-
ferential operator matrices are introduced to the analysis, and
the differential and integral equations are transformed into the
algebraic equations like as using Fourier or Laplace transform.
Moreover, the method can treat time varying or nonlinear cir-
cuits. Therefore, this method is useful for various circuit anal- where & is 1x m-dimensional Haar wavelet basis vector whose
yses. elements are the discretized expression of hi (t).
However, in that method, the use of Daubechies wavelet Equation (5)is called inverse Haar wavelet transform (IHWT)
makes the handling of the operator matrices complicated, es- and the following equation,
pecially, in the edges of the interval. Thus, in this paper, we
propose the circuit analysis method using Haar wavelet. The cq = ~ ' T H - ' = y + ~ T or c'= H$, (7)
Haar wavelet is easy to handle itself, and the operator matrices
is called forward Haar wavelet transform (HWT). Here, note
using the Ham wavelet are easily derived by introducing the
that
block pulse function [5], [6]. Moreover, the proposed method
can treat the nonlinear time varying circuits. The multiresolu- H-1 HT. (8)
tion property of the Haar wavelet makes the adaptive resolution 111. INTEGRAL A N D DERIVATIVE
OPERATOR MATRICES
control near the singular points possible, so that the efficient USING HAARWAVELET
calculation process will be achieved. In this paper, we confirm
the effectiveness of the proposed method using a simple exam- * The basic idea of the operator matrix has been firstly intro-

ple. duced by using Walsh function [5]. However, in logical way,

0-7803-8346-X/04/$20.00 02004 IEEE

I11-399
the matrices introduced by block pulse function are more fun-
damental [4], [5]. The block pulse function is the set of m
rectangular pulses which have l / m width and are shifted l / m
each other.
The integral operator matrix of the block pulse function ma-
trix B is defined as the following equation [5], 161,

~ B ( T ) E
~ TQ B . B ( ~ ) , (9)
+4-:
V
I

Qf,'LJO

(a) (b)
I, I!

where B(t)is m x m-dimensional matrix whose elements are

the discretized expression of the block pulse function b i ( t ) , i =
0,1,... ,m- land

Next, we show the operator matrix of the set of the orthogonal

functions [6]. Let @ be the set of the orthogonal functions as
follows,
Fig. 1. Equivalent circuit of linear time invariant (a) capacitor, (b) inductor, (c)

[ :]
coupled inductor, (d) resistor, (e) current controlled voltage source.
6"
wavelet domain expression of the voltage v ( t ) and the cur-
= [G>
8:.
' . G-11' = ' (I2) rent i ( t ) ,respectively, and L,, C,, R,, M , and Li, denote
-
@ 1

diag[l], diag[C],diag[R],diag[M] and diag[li], respectively.

$TI-1

-.
where $%, i = 0 , 1 , . . . m - 1,is 1 x m-dimensional vector. In
~
Capacitor:
~

this case, the integral matrix of Q, can be written as

Qo = Q,'QBQ,-~.
As the Haar wavelet matrix H is also the set of the orthogonal
(13)
v(t) =

v =
,U(O-) +-
V~+C;~Q~I,
:l i ( T ) d T > 110 := v(O-),

functions, the integral matrix of H is given as follows: or I = CwQH1[V- V i ] ,

Q H = HQBH-' = H Q B H ~ . (14) where, for example, V = Hi?, v' =

[v(to),v ( t l ) ,. . . , v(tm-1)IT, and w ( t k ) is the sampled
Similarly, the derivative matrix of H can written as voltage at t k . See Fig. 1 (a) for the expression in wavelet
domain.

Inductor:
Iv. HAARWAVELET SOLUTIONS O F NONLINEAR
TIME
VARYING CIRCUITS i(t) = i(0L) + l'v(r)d.r, io := i ( O - ) ,
A. Haar Wavelet Expression of Branch Characteristics of Lin-
ear Time Invariant Circuit Elements I = I~+L;'Q~V,
or V = Q;lLw[I-Io]. (17)
As the Haar wavelet is defined on interval [O: l ) , the generic
interval [tmin,tmax) can be rescaled by a new variable T on See Fig. 1 (b) for the expression in wavelet domain.
+
[O: l),where t = ( t w x - tmin)T tmin.In this paper, tmin= 0
without loss of generality, then capacitance c (F) and induc- Coupled Inductor:
tance 1 (H) are scaled to C = c/tmaxand L = l/tmax,respec-
d . d .
tively. v1(t) = L1-21(t)
dt
+ M--zz(t),
dt il(0-) =: i l 0 ,
Next, we show the Haar wavelet expression of the linear time d . d .
invariant circuit elements. In the followings, I/ and I are the vz(t) = L2-22(t)
dt
+ M-21(t),
dt
iz(O-) =: 220.

111-400
 K = QG1(hwIi + M d 2 ) - Q ~ l [ L i z u I+~Mo d m ] , K I

V2 = QN1(L2d2 + W J i ) - &21[L2w12~ + MJio].

(18)
See Fig. 1 (c) for the expression in wavelet domain.

Resistor and Current Controlled Voltage Source:

v ( t ) = Ri(t), L -
A
Fig. 2. Example circuit in wavelet domain.
I

V = R,I, R, =diag[RJ. (19)

See Fig. 1 (d) and (e) for the expression in wavelet domain.
B. Time Varying Elements and Nonlinear Elements
Consider the linear time varying elements such as R(t),L(t),
C(t).For example, for time varying resistor,
R, = Hdiag[R(to), R ( t l ) , . . . , R(t,-l)]HT, (20)

where t k is sampled time, while R, = diag[RJ in linear time Fig. 3. Time varying resistor in Fig. 2.
invariant resistor. Similarly, for capacitor and inductor, we de-
rive, V. CALCULATION
RESULTS

C, = Hdiag[C(tk)]HT, (21) In this section, we show an example of the circuit analysis.

Figure 2 shows the example circuit used in this section. In the
L, = Hdiag[L(tk)]HT, (22)
wavelet domain, from KCL and KVL, we derive
(k = 0,1,**.,m-l).
(32)
Moreover, in this case, we have to consider the initial condi-
tions as (33)
and branch equations,
CO = Hdiag[C(0)]HT, (23)
LO = Hdiag[L(0)JHT, (24)

where tmjn = to = 0. Therefore, the branch characteristics of a (34)

linear time varying capacitor and a linear time varying inductor ICC = CY, IR.
can be written as follows,
Therefore, the wavelet expression I L of the current through the
v = c,-~c~K,+c,-~Q~I inductor i ~ ( is
t )described as follows when the circuit is linear
or I = QH1[CwV- CoK,], (25) and time invariant,
I = L,lLoIo+L,lQ~V
or V = QH1[L,I - LoIoO]. (26)

On the other hand, for nonlinear time varying elements

R ( z ,t ) , C(z, t ) , L ( z ,t ) where 5 can be the voltage v or the In this paper, we show the results when the inductor is a non-
current i, we derive the parameter matrices as follows, when linear time invariant one whose characteristics are described as
we consider the current controlled elements,
( f l ( i L )= (1 + O.O2iL2)iL, (36)
R, = Hdiag[R(ik, & ) ] I T T , (27)
and the resistor is a linear time varying one as shown in Fig. 3.
C, =Hdiag[C(ik, &)]IfT, (28) In this case, because we cannot the current I L as the form
L, = Hdiag[L(ik,tk)]HT, (29) shown in Eq. (35), we have to solve numerically the nonlin-
( k = 0 , 1 , . . *, m - l ) , ear algebraic equation with respect to I L . In this paper, we take
tmx 1, tmin= 0,C = 0.006F1, (Y = 2, i ~ ( 0 - )= lO[A],
and
VC(O-) = 5O[V], e ( t ) = 20t, R m i n = O.O1[f2], and R ~ =x
CO = Hdiag[C(io, O)]HT, (30)
1OO[f21.
LO = Hdiag[L(io, O)]HT, (31) First, we show the results for rn = 64 (i.e., cy = j = 6) as
shown in Fig. 4. In this figure, the dots show the result using
where tmi, = t o = 0. The branch characteristics of a non- fast HWT and the solid line shows the result using SPICE. Fig-
linear time varying capacitor and a nonlinear time varying in- ure 5 shows the result of multiresolution analysis f o r m = 64,
ductor can be the same as Eqs. (25) and (26). In this case, as where 9 ( 3 )is the finer level result of a = j = 3 and g ( 5 )
the branch characteristics become the nonlinear algebraic equa- is the finer level one of cy = j = 5. From these results, we
tion, we have to use some numerical methods such as Newton- can see that the amplitude in higher resolution is larger near
Raphson method to solve the nonlinear algebraic equations. the switching points of time varying resistor. It indicates that
However, from the properties described above, we can analyze more precise analysis is needed only at these points, so that the
even nonlinear time varying circuits. higher resolution analysis is required only near these points.

111-401
 TABLE I
RESOLUTIONI N EACHINTERVAL OF THE RESULT
DETAILED
SHOWN IN FIG. 6.

(1 02 0 2 Ob ilR I
# !*Wl

Fig. 4. Calculation result f o r m = 64

VI. CONCLUSIONS
In this paper, we have proposed a method to analyze circuits
using Haar wavelet transform. In the proposed method, the op-
erator matrices are easily derived and handled due to the use of
the block pulse function [SI, [6]. And, it can treat even non-
linear time varying circuits. Moreover, it is suggested that the
adaptive error control will be possible by using the multiresolu-
tion analysis. The complete adaptive error control method will
be reported in the near future.

REFERENCES
[I] A. Haar, “Zur Theorie der orthogonalem functionen Systeme,” Math.
Ann., vol. 69,pp. 331-371, 1910.
,) ......................................
121 G . Strang, “Wavelet transforms versus Fourier Transforms,” Bull. Amer.
I,! ! Math. Soc., vol. 28, no. 2, pp. 288-305, Apr. 1993.
n j 131 C. F. Chen and C. H. Hsiao, “Haar wavelet method for solving lumped
and distri buted-parameter system,” IEE Proc. of Control Theory Appl.,
vol. 144, no. I , pp. 87-93, Jan. 1997.
14) J. L. Wu, C. H. Chen, and C. F. Chen, “A model reduction via Haar
wavelet,” Int. J. Confrol and Intelligent Sysfems. vol. 29, no. 2, pp. 29-
32,2001.
[SI C. F. Chen, Y. T. Tsay, and T. T. Wu, “Walsh operational matrices
for fractional calculus and their application to distributed systems,” J.
Franklin lnsfifute, vol. 303, no. 3. pp. 267-284, Mar. 1977.
[6] J. L. Wu, C. H. Chen, and C. F. Chen, “Numerical inversion of Laplace
Transform using Haar wavelet operational matrices,” IEEE Trans. Cir-
cuitsSyst. I , vol. 48, no. I , pp. 12@122,Jan.’2001.
Fig. 5. Multiresolution analysis for rn = 64. 171 D. Zhou, X. Li, W. Zhang and W. Cai, “Nonlinear circuit simulation
based on adaptive wavelet method.” Proc. of IEEE ISCAS ’97,pp. 1720-
1723, June 1997.
Figure 6 shows the results that the equivalent higher resolu- [8j T. Matsumoto, H. Shimode, and S. Moro, “A simple and direct wavelet
tion analysis with mixed resolutions is carried out at the switch- collocation method for nonlinear time evolution partial differential equa-
ing points of the time varying resistor for m = 64. Detailed in- tions,” Proc. ojlnternalional Technical Conference on CircuitslSystems.
terval [a,b) and converted resolution m’ are shown in Table I. Computers and Communications (ITC-CSCC ‘99), pp. 747-750, 1999.
191 T. Matsumoto, H. Shimode, and S. Moro, “Solving initial boundary
In this table, m’ is the converted resolution of each sub-interval
[a,b) where m’ = e.
In this example, CY’(= log, m’) is cal-
value problems via wavelet collocation methods in nonlinear time evo-
lution PDEs,” Proc. of International Symposium on Nonlinear Theory
and its Applications (NOLTA ’99). pp. 503-506, Nov. 1999.
culated as CY’= 8.19 while a = j = 6 in the result shown
101 T. Matsumoto, T. Araki, and S. Moro, “Solving numerically stiff differ-
in Fig. 4. In this case of Fig. 6, more precise analysis can be ential equations based on adaptive wavelet collocation methods,” Proc.
achieved with smaller calculation costs than that with single of IEEE International Symposium on Intelligent Signal Processing and
resolution of Fig. 4.It is suggested that the adaptive error con- Communication Systems (ISPACS ‘99). pp. 669472, Dec. 1999.
trol will be possible by using the results of multiresolution anal- I 1 I D. Zhou and W. Cai, “A fast wavelet collocation method for high-speed
circuit simulation,” IEEE Trans. Circuits Syst. I , vol. 46, no. 8, pp. 9 2 s
ysis. In this example, though we choose sub-intervals manually 930, Aug. 1999.
from the result of the multiresolution analysis shown above in 121 D. Zhou, W. Cai, and W. Zhang, “An adaptive wavelet method for non-
Fig. 5, our future problem is to control the sub-intervals auto- linear circuit simulation,” IEEE Trans. Circuits Syst. I , vol. 46, no. 8, pp.
matically. 931-938, Aug. 1999.
131 S. Barmada and M. Raugi, “A general tool for circuit analysis based
on wavelet transform,” Int. J. Circuit Theory Appl., vol. 28, no. 5, pp.
461480,2000.
141 X. Li, B. Hu, X. Ling, X.Zeng,“A wavelet-balance approach for steady-
state analysis of nonlinear circuits,” IEEE Trans. Circuits Syst. I , vol.
49, no. 5, pp. 689494, May 2002.
[ 151 G. Antonini and A. Orlandi, “A wavelet-based time-domain solution for
PEEC circuit,” IEEE Trans. Circuits Syst. I , vol. 47, no. 11, pp. 1634-
1639.Nov. 2000.
[ 161 Y.Nievergelt. Wavelets Made Eusv, Birkhauser, 1999
-I,!? I
rj 0.2 0.4 I1 h 03 I

I I,LCI

Fig. 6. Calculation result with mixed resolutions

I11-402