You are on page 1of 9

Analog Integrated Circuits and Signal Processing, 47, 233–241, 2006


c 2006 Springer Science + Business Media, Inc. Manufactured in The Netherlands.
DOI: 10.1007/s10470-006-5255-9

Phase-Compensated Time-Varying Butterworth Filters

JACEK PISKOROWSKI
Institute of Control Engineering, Szczecin University of Technology, ul. 26 Kwietnia 10, 70-313, Szczecin, Poland
E-mail: jacek.piskorowski@ps.pl

Received April 26, 2005; Revised June 25, 2005; Accepted September 29, 2005

Published online: 10 March 2006

Abstract. One of the important problems of signal processing is to design the filters which possess linear phase charac-
teristics. This paper presents the concept of time-varying Butterworth filters with linear phase. The compensation of the
phase characteristics is carried out with the aid of phase shifter which is cascade-connected to the structure of the original
Butterworth filter given by the transfer function. The parameters of the phase shifter were calculated in this way that the
group delay of the designed filter is possible constant in the filter pass–band. Time varying coefficients were introduced to the
phase–compensated filter structure for the purpose of minimization of the filter transient state. This paper contains simulation
results of proposed filters and comparison with classic circuits.

Key Words: Butterworth filters, phase shifters, time-varying systems, phase compensation

1. Introduction we can notice that all derivatives of the square of the gain
characteristics up to the 2N − 1 order are equal to zero for
The main goal of the Butterworth approximation is to obtain ω = 0. Hence Butterworth filters are named as filters with
a maximally flat gain characteristics in the filter pass–band. maximally flat gain characteristic in the filter pass-band.
The gain characteristics should begin maximally flat for fre- Low-pass Butterworth filters are forming as a re-
quency ω = 0 and inflexion of this characteristics should be sult of specified approximation of the gain characteris-
near by the limit frequency. The square of the gain charac- tics, however the phase characteristics is the secondary
teristics of the low–pass Butterworth filter can be written as effect of this approximation. In the result of this effect,
follows: the phase characteristics is nonlinear. Nonlinearity of the
phase characteristics causes that the dynamic properties of
1
G 2 (ω) =  ω 2N (1) designed filter may be undesirable. It especially concerns to
1+ ω0 the impulse engineering where the phase distortions of the
filter lead to the changes of the shape of the transmitted sig-
where: N—order of the filter, ω0 —3dB–limit frequency. nals. Ideal properties for transmission of rectangular signals
Together with the increase of Butterworth filter order, the are possessed by filters in which the delay doesn’t depend
gain characteristics approaches to the characteristics of the on the frequency which means that the phase shift is propor-
ideal filter: tional to the frequency. Linearity of the phase characteristics
or constancy of the group delay means that all frequencies
 ω from the pass-band are delayed by the same constant inter-
 1 if | ω0 | < 1
 val of time. The approximation of Bessel (Thomson) filters
lim G 2 (ω) = 12 if | ωω0 | = 1 (2) satisfies these properties. The purpose of the approximation
N →∞ 

0 if | ωω0 | > 1 of Bessel filters is to choose the coefficients in such a way
that the group delay below the limit frequency only slightly
By expansion of the square of the gain characteristics of depends on the frequency. Nevertheless Bessel filters pos-
Butterworth filter in the Maclaurin series: sess unfavourable course of the gain characteristics in the
 2N  4N filter pass-band as well as in the filter transition-band. The
1 ω ω comparison of Bessel filters and Butterworth filter is shown
G 2 (ω) = = 1 − +
1 + ( ωω0 )2N ω0 ω0 in Figs. 1 and 2. From these figures we can conclude that
 6N the maximally flat gain characteristics of Butterworth filter
ω
− + ... (3) causes nonlinearity of the group delay characteristics in the
ω0 filter pass-band.
234 Jacek Piskorowski

Fig. 1. Gain characteristics of 3rd order Butterworth filter and Bessel filter.

Fig. 2. Group delay characteristics of 3rd order Butterworth filter and Bessel filter.

2. Group Delay of the Filter function:

In the case of the filter with possible linear phase the char- ln[G(s)] = ln |G(s)| + j · ϕ(s) (5)
acteristic of group delay T (ω) is very useful. Group delay is
defined by the derivative of the phase characteristics ϕ(ω) After differentiation with regard to variable s we obtain:
with regard to the pulsation ω with the minus sign:
G  (s) d(ln |G(s)|) dϕ(s)
= +j· (6)
dϕ(ω) G(s) ds ds
T (ω) = − (4)
dω The phase characteristic is the odd function of fre-
quency however the gain characteristic and its loga-
In the general case of the group delay calculation we can rithm are the even function of frequency. It is also
start from the knowledge, that the phase characteristics of a known that the derivative of the even function is the
system is the imaginary part of the logarithm of the transfer odd function however the derivative of the odd func-
Phase-Compensated Time-Varying Butterworth Filters 235

tion is the even function. Therefore the derivative of Using (9), the group delay of the above mentioned filter
the phase characteristics may be expressed by the can be written as follows:
relation:

2ω4 + ω2 + 2
dϕ(s) G (s) T (ω) = (12)
− = j · Ev (7) ω6 + 1
ds G(s)
Let’s use the Taylor expansion of the group delay round
where: Ev[ f (x)]is the even part of function f (x). After about ω0 :
taking into consideration undermentioned relationship:

T (n) (ω0 )

T (ω) = T (ω0 ) + (ω − ω0 )n (13)
dϕ(ω) dϕ(s) n!
= j· (8) n=1
dω ds s= jω
After expansion the group delay given by (12) round
as well as the property which says that an arbitrary function about ω = 0 we obtain:
f(x) we can decompose on the sum of two functions f (x) =
g(x) + h(x), where g(x) is the even function (even part T (ω) ≈ 2 + ω2 + 2ω4 − 2ω6 − ω8 − 2ω10 + · · · (14)
of function f(x)) and is expressed by the relation: g(x) =
1
2
( f (x) + f (−x)), but h (x) is the odd function (odd part From Fig. 3 we can notice that the limitation of the group
of function f (x)) and is expressed by the relation: h(x) = delay expansion to a few first terms approximates the real
1
2
( f (x)− f (−x)). Finally, the group delay of the filter which function in the filter pass–band with sufficient accuracy. In
is given by the transfer function G(s) we can express as order to compensate of the group delay characteristics of
follows: the 3rd order Butterworth filter described by (11) let’s use

the 2nd order phase shifter described by (10). The group
1 G  (s) G  (−s)
T (ω) = − + (9) delay of (10) on the basis of (9) may be written as follows:
2 G(s) G(−s) s= jω
 2
1 2Q ωω0 p + 2Q
3. Group Delay Compensation TH2 (ω) = ·      2 (15)
ω0 p Q 2 ω 4
+ 1 − 2Q 2 ωω0 p + Q 2
ω0 p
For the purpose of the group delay compensation one can
use the phase shifters (all–pass filters). Transfer function of Expansion of the above mentioned function round about
the second order phase shifter can be written as follows: ω = 0 has the following form:
     
s 2
− 1 s
+1 2 3Q 2 − 1 ω 2
ω0 p Q ω0 p ω0 p · TH2 (ω) ≈+2· (16)
H2 (s) =  s 2   (10) Q Q3 ω0 p
ω0 p
+ 1 s
Q ω0 p
+1   4
5Q 4 − 5Q 2 + 1 ω
+2·
The above mentioned system is stable when Q > 0. Q5 ω0 p
The specificquality of the phase shifters is its constancy of  
the gain characteristics H 2 (ω) = const = 1. These sys- 7Q 6 − 14Q 4 + 7Q 2 − 1 ω 6
+2·
tems are only used to form the phase characteristics. Owing Q7 ω0 p
to these properties we know that if we connect in series
9Q 8 − 30Q 6 + 27Q 4 − 9Q 2 + 1
the phase shifter to the compensating filter then the gain +2·
characteristics of the resultant system remains without any Q9
 8
changes, however the group delay characteristics will be ω
× + ... (16)
the result from a summation of the phase shifter and the ω0 p
compensating filter characteristics. The procedure of the
group delay compensation is based on the choice of the
Sum of the Taylor expansions of the Butterworth filter
phase shifter(s) parameters in order that the resultant group
group delay and phase shifter group delay can be written as
delay characteristics will be as flat as possible in the filter
follows:
pass-band.
Let’s consider 3rd order low-pass Butterworth filter with  
2
limit frequency ωg = 1 described by the following transfer T (ω) + TH2 (ω) ≈ +2
Qω0 p
function:
3Q 2 − 1
G(s) =
1
(11) + 2· + 1 ω2
s3 + 2s 2 + 2s + 1 Q 3 ω03 p
236 Jacek Piskorowski

Fig. 3. Approximation of the group delay characteristics.


5Q 4 − 5Q 2 + 1 The group delay of the compensated Butterworth filter
+ 2· + 2 ω4 may by express by the following expansion:
Q 5 ω05 p
T (ω) + TH2 (ω) ≈ 5.5481 + 1.3866ω6 − 3.3638ω8 − . . .
7Q 6 − 14Q 4 + 7Q 2 − 1
+ 2· − 2 ω6 (21)
Q 7 ω07 p The group delay characteristics of the phase–
compensated Butterworth filter is shown in Fig. 4. How
9Q 8 − 30Q 6 + 27Q 4 − 9Q 2 + 1
+ 2· −1 one can notice, the compensation of the group delay char-
Q 9 ω09 p acteristics brings good results, because compensated char-
acteristics is considerably more flat in the filter pass-band
× ω8 + . . . (17)
than in the original Butterworth filter. The group delay value
of the phase–compensated filter is obviously greater than in
Parameters Q and ω0 p should be selected in this way, in
the original filter, however this value does’t play a greater
order to eliminate the term of second and fourth order of
part. For the systems from impulse engineering the im-
T (ω) + TH2 (ω). For this purpose it is necessary to solve the
portant feature is constancy of the group delay in the fil-
undermentioned system of equations:
ter pass-band. Compensation of the group delay was car-
 3Q 2 −1 ried out at the cost of the extension of the filter transient
 2· Q 3 ω03 p
+1=0
state. Deterioration of the dynamic properties of the phase–
(18)
2 · 5Q 4 −5Q 2 +1
+2=0 compensated Butterworth filter is shown in Figs. 5 and 6.
Q 5 ω05 p
Figure 5 shows the response for the unit step of the Butter-
worth filters and Fig. 6 shows the response for the sinusoidal
After short transformations:
input x(t) = sin(ω0 t).

Q 3 ω03 p + 6Q 2 = 2
(19)
2Q 5 ω05 p + 10Q 4 − 10Q 2 = −2 4. Time-Varying Parameters in
Phase–Compensated Filters
After taking into consideration the conditions of stability
and conditions of filter realizability we obtain the following For constant parameter filters there are only small possibil-
solution: ities of shortening the transient state since the filter param-
eters are calculated on the basis of the assumed approxima-
Q ≈ 0.5509, ω0 p ≈ 1.0232 (20) tion method of the frequency characteristics. This guaran-
tees that the frequency requirements are satisfied without
Phase-Compensated Time-Varying Butterworth Filters 237

Fig. 4. Group delay of the phase–compensated Butterworth filter and the original Butterworth filter.

Fig. 5. Step responses of classic and phase-compensated 3rd order Butterworth filters.

taking into consideration the character of the transient state. filter bandpass in low-pass filters. This procedure is con-
If the requirements on the frequency characteristics are im- nected with the change of value of parameters ω0 , β and
posed, we can slightly influence the shortening of duration T which leads to non-realization of the imposed filter fre-
of the transient state of the n-th order filter by choosing dif- quency characteristics.
ferent methods of approximation this characteristics. The The classic filter structure is described by the transfer
indeterminacy principle says that it isn’t possible to achieve function which is the product of 2nd order systems for even
a shorter rise time of the low-pass filter output signal when filter orders or the product of 2nd order systems and one
the filter pass-band is constant. One can obtain significant 1st order system for odd filter orders. This can be written as
changes of duration of the transient state by variation of follows:
238 Jacek Piskorowski

Fig. 6. Responses for the sinusoidal input x(t) = sin(ω0 t) of classic and phase–compensated 3rd order Butterworth filters.

– for even order n = 2i Next, let’s create the set F(t) of functions of filter and
phase shifter parameters:
ku
K u (s) =    (22)  
i
1 2
2 s
ω0i
+ 2βi
ω0i
s+1 1
F(t) = βi (t), ω0i (t), (t), β p (t), ω0 p (t) (26)
T
– for odd order n = 2i + 1
The main assumption which was imposed on func-
ku
K u (s) =    (23) tions from set F(t) is the necessity of settling these
(T s + 1) i
1
ω0i
2 s 2 + 2βi
ω0i
s+1 functions during the transient state of the original
time–invariant filter. This condition can be written as
Time varying filter design is the result of modeling of follows:
system of ordinary differential equations with varying co-
efficients described as follows: lim F(t) = F ± α (27)
t→tuα
 2

 a1 (t) d dty12(t) + b1 (t) dydt1 (t) + y1 (t) = k1 x(t) where: tuα —settling time of a constant parameter filter with



 a (t) d 2 y2 (t) + b (t) dy2 (t) + y (t) = k y (t) an accuracy of α.
2 dt 2 2 dt 2 2 1
(24) Relation (27) is responsible for keeping frequency as-

.......................................

 sumptions of designed time-varying filter. On the basis of

 2
ai (t) d dtyi2(t) + bi (t) dydti (t) + yi (t) = ki yi−1 (t) [4] time-varying system is stable when functions from set
F(t) satisfy undermentioned condition:
where: k1 · k2 · . . . · ki = ku , ai (t) = ω21(t) , bi (t) = 2βi (t)
ω0i (t)
. In
0i lim F(t) = F (28)
the case of odd filter order: a1 (t) = 0 and b1 (t) = T (t). t→∞
Dynamic properties of arbitrary low-pass filter can be
described by means of the damping factor βi , characteristic
frequency ω0i and time constant T (only in the case of In order to shorten the filter transient state one assumes
odd filter orders). Let’s assume, that all those parameters functions from set F(t) on the basis of [7, 8, 10] as follows:
belong to certain set F f = {βi , ω0i , T1 }. The 2nd order

phase shifter is described by ω0 p and β p , where: β p = 2Q 1 d −1


. F(t) = d · F · 1 − · h(t) (29)
These parameters form second set F ps = {β p , ω0 p }. Sum d
of the above mentioned sets is described as:
  where: F—value of parameter from set F following from
1 the Butterworth approximation or phase shifter calculus,
F = F f ∪ F ps = βi , ω0i , , β p , ω0 p (25)
T d—variation range of the functions from set F(t) described
Phase-Compensated Time-Varying Butterworth Filters 239

Fig. 7. Step responses of 3rd order Butterworth filters.

Fig. 8. Group delay of 3–rd order phase-compensated time-varying Butterworth filter in the function of time (for ω = 0).

by the relation: where: L−1 is the inverse Laplace transform.


For β f < 1 :
F(0) 
d= (30)   
F(∞)
h (t) = 1 −  cos ω0 f t 1 − β 2f
Function h(t) describes the step respond of the second
order system: 
1   
  + sin ω0 f t 1 − β 2f  · exp(−β f ω0 f t)
h(t) = L−1 
1 1  (31) 1 − β 2f
2β f
s 1
ω02 f
s2 + ω0 f
s+1 (32)
240 Jacek Piskorowski

it is necessary to design a special system to detect any step


changes of the input signal. For this purpose one can make
use of the circuit which delays input signal by a given period
f time τ . This time should be chosen in this way in order to
not omit any significant changes of the level of the input sig-
nal x(t). The absolute value of the difference between input
signal and delayed input signal ξ (t) = |x(t)−x(t −τ )| is the
Fig. 9. Block diagram of continuous running of phase–compensated signal which is delivered to the input of comparator. In this
time-varying Butterworth filter.
comparator the actual value of the signal ξ (t) is compared
where: β f — parameter which determines oscillations of with the sensitivity c. The value of the sensitivity is given
functions from set F(t) (on the basis of [7, 8] β f = 0.9), by user of the filter. In many cases of signal processing one
ω0 f — parameter which determines the range of the speed of can estimate the level of the noise signal. By the knowledge
above–mentioned functions. The best results in shortening of this level of the noise signal we have possibility to define
the transient state of phase–compensated Butterworth filters the sensitivity of filter. If ξ (t) ≥ c at that time on the output
were obtained while parameters ω0i and ω0 p were varied of comparator the signal ζ (t) = 1 is generated, otherwise
according to the same function F1 (t). Parameters βi and β p ζ (t) = 0. Then signal ζ (t) is delivered to the reset inputs
were varied also according to the same function F2 (t). In the of integrators which are responsible for generation all func-
second case variation range d of function F2 (t) is equal to tions of filter parameters. Any rising edge of signal ζ (t)
1.2. Detailed description of optimal selection of all function is synonymous with restarting all integrators what causes
parameters was presented in [7, 8, 10]. cyclic generation of functions. Block diagram of this pro-
Fig. 7 presents step responses of phase–compensated cess is shown in Fig. 9. Figure 10 shows that application
time-varying Butterworth filter, phase–compensated time– of time-varying filters to processing the noised rectangular
invariant Butterworth filter and original time–invariant But- signals gives much better results than time-invariant filters.
terworth filter. Figure 8 presents how the time varying prin- This system doesn’t require a precise knowledge about the
ciple alters the group delay of the filter. We can notice that level of noise in the input signal. The fact that the sensitiv-
the group delay isn’t constant only in the duration of the ity of system c was chosen incorrectly doesn’t cause faulty
filter transient state for the individual frequency ω. work of filter. This situation is illustrated in Fig. 11. Incor-
rect selection of sensitivity will cause frequent restarting of
5. Continuous Running of Phase–Compensated integrators. By too frequent restarting of integrators func-
Time-Varying Butterworth Filters tions of the filter parameters will have no time to settle its
correct value in steady state. This fact may cause that in
So far introduction varying in time parameters has only con- some intervals of time filter will work with greater limit
cerned initial phase of filter work. However, in order that frequency, what is illustrated by some oscillations in the
time-varying filters were useful during continuous running output signal (Fig. 11).

Figure 10 The output signal of time-varying filter and time–invariant filter.


Phase-Compensated Time-Varying Butterworth Filters 241

Fig. 11. The output signal of time-varying filter and time–invariant filter (sensitivity c was chosen incorrectly).

6. Concluding Remarks 2. W.K. Chen, The Circuits and Filters Handbook. CRC Press, Boca
Raton 1995.
As it has been proven, application of time varying coeffi- 3. S. Darlington, “A History of Network Synthesis and filter the-
ory for circuits composed of resistors, inductors and capacitors,”
cients in low–pass phase–compensated Butterworth filters
IEEE Trans. On Circuits and Systems, vol. 46, no. 1, pp. 4–13,
causes considerable shortening of the settling time. The best 1999.
results of shortening of the settling time were obtained by 4. B.P. Demidowic, Lekcij po matematicieskoj tieorij ustoiciwosti. Iz-
varying in time all groups of parameters according to the datwielstwo, Moscow, 1980.
same function. It seems that further examinations of phase– 5. J. Izydorczyk and J. Konopacki, Filtry analogowe i cyfrowe.
Wyd. Pracowni Komputerowej J. Skalmierskiego, Katowice
compensated time-varying Butterworth filters are needed.
2003.
Especially, the problems of continuous running of these 6. M. Jafaripanah, B.M. Al-Hashimi, and N.M. White “Application of
filters are open. Nevertheless already this paper proves pos- Analog Adaptive Filters for Dynamic Sensor Compensation,” Proc
sibilities and practical usefulness of the proposed filter con- IEEE Trans. On Instrumentation and Measurement, vol. 54, no. 1, pp.
cept as a signal processing instrument. Now research of 245–251, February 2005.
the application of analog time-varying filters to the area of 7. R. Kaszynski and J. Piskorowski, “The Proposal of Bessel Filters
with Varying Parameters,” Proc. Int. Conference on Instrumentation,
dynamic sensor compensation is carried out. Preliminary Control and Information Technology (SICE), Sapporo, Japan, June
examinations show that the application of described time- 2004, pp. 1267–1270.
varying filters can improve the sensor response to speed up 8. R. Kaszynski and J. Piskorowski, “Bessel Filters with Varying Pa-
the process of measurement. rameters,” Proc. IEEE Conference on Analog and Digital Techniques
in Electrical Engineering (TENCON), Bangkok, Thailand, November
2004, pp. 641–644.
Acknowledgments 9. B.P. Lahti, Linear Systems and Signals. Berkeley-Cambridge Press,
Carmichael 1992.
This work was partially supported by Polish State Commit- 10. J. Piskorowski, “Influence of Varying in Time Parameters in
tee for Scientific Research under grant 3 T10C 032 28. Low–pass Bessel Filters on the Settling Time of Output Signal,”
Proc. Int. Conference on Methods and Technics of SignalProcess-
ing in Physical Measurements, Rzeszow, Poland, September 2004,
References
pp. 107–113.
11. J. Szabatin, Podstawy teorii sygnalow. WKiL, Warsaw, 1990.
1. A. Carusone and D.A. Johns, “Analogue adaptive filters: past and 12. U. Tietze and Ch. Schenk, Uklady polprzewodnikowe. WNT, Warsaw,
present,” IEE Proc.-Circuits Devices Systems, vol. 47, no. 1, pp. 82– 1987.
90, February 2000.

You might also like