Discretization Methods of Fractional Parallel PID

Controllers
Mohamad Adnan Al-Alaoui*, Senior Member, IEEE
Electrical and Computer Engineering department
American University of Beirut
Lebanon

Abstract— This work addresses employing direct and indirect fractional discrete system respectively [13]. The versatile S.
discretization methods to obtain a rational discrete C. Dutta Roy had employed it to obtain rational continuous-
approximation of continuous time parallel fractional PID time approximations to fractional analog systems[10-11]. The
controllers. The different approaches are illustrated by mathematical basis of the CFE are obtained in the classical
implementing them on an example. works of Wall [16] and Khovanskii [12]. Other proposed
continuous-time approximations were introduced by
Index Terms- Oustaloup et. al.[15], and recently by Aoun et.al. [6] Xue
Al-Alaoui operator, analog_to digital_conversion, bilinear et.al. [17]. The method of CFE was applied by Chen et. al. in
transformation, fractional order systems, PID control, a direct discretization approach [8-9]. Chen and Moore’s
transfer functions. work gave impetus to research employing CFE and the Al-
Alaoui operator [8]. In the sequel, the step response of the
unity feedback system of Figure 1, with a given fractional
I. INTRODUCTION order parallel PID (proportional, integral, and derivative)
Fractional systems are systems that are represented by controller, C(s), and an integer order plant, G(s), is
differential equations that allow non integer orders. This is a determined [7]. Different discretization methods for C(s) are
generalization of the integer order integration and employed while G(s) is discretized by using the bilinear
differentiation. The order can take on any real value, not transform. In the following s is approximated by using the
necessarily only fractional values. Thus the fractional bilinear transform [14], the Al-Alaoui transform [1-3], and
designation is not accurate, however the misnomer is now the the phase enhanced Al-Alaoui transform [4], represented
accepted designation. The corresponding transfer functions below by equations (1), (2), and (3) respectively. The phase
for fractional order systems resulting from applying enhanced Al-Alaoui transform is obtained by a half-sample
continuous or discrete time transforms will be non-rational. advance of the Al-Alaoui transform, which is achieved by
An analytical solution for the simulation and computation of multiplying (2) by z0.5, or dividing it by z-0.5. The half-sample
a fractional model’s output is often not simple to obtain. advance, shown in (3), changes the almost linear phase of (2)
Algorithms were developed to approximate the outputs of to almost 90 degrees while the magnitude remains the same.
fractional systems by using either continuous or discrete
rational models [5-9], [15], [17]. 2 (1 − z −1 ) (1)
s≈
T (1 + z −1 )
Indirect and direct discretization approaches to the time
domain simulations of the output of a fractional system are 8 (1 − z −1 ) (2)
the two major ones. The indirect approach first develops a s≈
7T (1 + z −1 / 7)
rational analog model approximation of the fractional system
then applies an s-to-z transform to the continuous time 8 z 0.5 (1 − z −1 ) 8 (1 − z −1 ) (3)
approximation. While, the direct approach first obtains a s≈ −1
=
7T z 7T − 0.5 z −1
fractional discrete system from the fractional analog system (1 + ) z (1 + )
by applying an s-to-z transform to the fractional analog 7 7
system, then a rational discrete approximation of the
fractional discrete system is obtained. The method of
continued fraction expansion (CFE) is one of the early
methods that was employed to obtain a rational analog or a
rational discrete approximation of a fractional analog or a
Figure 1. A Unity Feedback System

This research was supported, in part, by the University Research Board,

URB, of the American University of Beirut.

978-1-4244-5091-6/09/$25.00 ©2009 IEEE 327

 II. PID CONTROL ⎛ s ⎞
r

⎜ 1+ ⎟ (8)
PID controller is often employed as shown in Figure 1 to ω B (ω A + s ) ωB
s[−ωrA ,ω B ] ≅ ⎜ ⎟
improve the performance of the unity feedback loop shown (ω A ) n ( rs 2 + ω B s + (1 − r )ω Aω B ) ⎜ 1 + s ⎟
[7]. Two popular implementations of the PID controller are ⎜ ωA ⎟
the parallel and the series implementations. The transfer ⎝ ⎠
function of the parallel implementation of the PID controller Note that (8) is equation (22) in [6].
is expressed as Then equation (8) is expanded using CFE, Taylor series
2 expansion, or the method of recursive poles and zeros to
KI K s + KPs + KI (4)
Kp + + KDs = D obtain an analog approximation. The analog approximation
s s with recursive poles and zeros is described in detail in [17],
K p , K I , and K D designate the Proportional, Integral, and and will be used in this work. It is summarized below:
Derivative gains respectively with each one serving different ⎛ s ⎞
r

purposes like reducing rise time or steady state error, etc... ⎜1+ d ⎟
⎜ ωA ⎟
For the PIλDμ the transfer function of the parallel ⎜ b ⎟ =
k =N
1 + s / ωk' (9)
implementation is expressed as follows: ⎜ s ⎟ lim ∏ 1+ s / ω
⎜1+ b
N ⎯⎯→ ∞ k =− N k
⎟
KI K sλ +μ + K Psλ + K I (5) ⎜ ωB ⎟
Kp + λ
+ KDsμ = D ⎝ d ⎠
s sλ
r −2k
The fractional PIλDμ controller has five degrees of ⎛d ⎞ 2 N +1
ω = ⎜ ωA ⎟
'
freedom,( K p , K I , K D , λ , and μ ), versus three degrees of k
⎝b ⎠ (10)
freedom for the integer PID controller. The series PIλDμ
r +2k
controller is implemented as a cascade of PIλ(s) and PDμ (s). ⎛b ⎞ 2 N +1
In this work the parallel implementation of Cao et. al. [7] will ωk = ⎜ ω B ⎟ (11)
be addressed. The transfer functions of controller C(s), and ⎝d ⎠
the plant G(s) are presented below as equations (6) and (7).
⎛ ds 2 + bsωB ⎞ k = N s + ωk' (12)
C ( s ) = 2.7566 + 0.0029 s − 0.7908
+s 0.4848 sr ≈ K ⎜ ⎟∏
⎝ d (1 − r ) s + bsωB + dr ⎠ k =− N s + ωk
2
(6)
400 (7) ⎛d ⎞
r k =N
ωk
G (s) = (13)
2
s + 50s
K = ⎜ ωA ⎟
⎝b ⎠
∏ω
k =− N
'
k

III. INDIRECT AND DIRECT DISCRETIZATION METHODS We used b = 10 and d = 9 as in [17].

In this section direct and indirect discretization
The procedure for the approximation can be briefly
approaches are carried out to obtain rational discrete
summarized in the following steps.
approximations of C(s) and the different Bode plots of the
approximations are compared to the ideal response. a) The frequency range [ωA, ωB] and N are given.
Additionally G(s) is discretized using the bilinear transform b) Based on the fractional order r, calculate ω’k and ωk
and the resulting step response of the closed system with according to (10) and (13)
various C(s) discretizations are compared. c) Compute K from (13)
A. Indirect Discretization d) Obtain the approximate rational transfer function
from (12) to replace sr
Indirect discretization consists of two steps. In the first
step a rational analog transfer function that approximates the The following values are used:
irrational transfer function of the fractional controller is b = 10; d = 9; ωA = 0.01; ωB = 100; N = 3;
obtained. In the second step an analog to digital r1 = -0.7908; r2 = 0.4848; % Fractional exponents
transformation of the rational analog approximation is done.
The above analog approximation yields the following
1) The Analog Rational Approximation transfer function.
The approach described in [6] and [17] will be
implemented. The approximation is carried out in the 1629 s18 + 4.776e005 s17 + 5.253e007 s16 + 2.815e009 s15 +
following two steps. 7.908e010 s14 + 1.179e012 s13 + 9.335e012 s12 + 3.9e013 s11
+ 8.557e013 s10 + 9.798e013 s9 + 5.81e013 s8 + 1.77e013 s7 +
• Determine a frequency range [ωA, ωB] where we 2.722e012 s6 + 2.065e011 s5 + 7.255e009 s4 + 1.003e008 s3 +
need to approximate the fractional order transfer 1.128e005 s2 - 5606 s - 20.25
function. --------------------------------------------------------------------------
• Perform an analog integer-order approximation of 74.73 s18 + 3.205e004 s17 + 4.701e006 s16 + 3.196e008 s15 +
the fractional-order transfer function 1.103e010 s14 + 1.979e011 s13 + 1.836e012 s12 + 8.794e012
This is accomplished in [6] by letting: s11 + 2.154e013 s10 + 2.698e013 s9 + 1.714e013 s8 +

328
5.509e012 s7 + 8.816e011 s6 + 6.886e010 s5 + 2.465e009 s4 + s = Di ( z ) , where i =1, 2, or 3 corresponding to equations (1)-
3.434e007 s3 + 3.207e004 s2 - 2017 s - 7.347 (3) and Di ( z ) correspond to the right hand side of equations
(14)
(1)-(3). Then apply CFE to approximate the fractional-order
2) Discretizing the Analog Approximation
generating function, Gi ( z ) = ( Di ( z )) r , by a rational disctret
To discretize the analog rational approximation of (14)
select suitable s-to-z transforms. The bilinear transform, time , IIR, transfer function [5], [8-9]
equation (1), Al-Alaoui transform, equation (2), and Al- B. CFE
Alaoui-1 transform, equation (3), are employed. Replacing s
in (14) by the right hand side of equations (1) or (2) yield The second step obtains a rational discrete transfer
two rational discrete time transfer functions, while replacing s function of the controller that approximates the irrational
in (14) by the right hand side of equation (3) yields an discrete time function that was obtained in the first step. The
‘almost’ rational discrete time function, a discrete time second step could employ a variety of methods such power
function with additional delay multipliers. The resulting series expansions, or CFE. Replace s by (1) - (3), then use the
three discrete time functions are plotted in Figure 2. CFE to approximate the obtained digital fractional-order
Concerning Al-Alaoui-1 indirect method resulting from transfer function.
substituting (3) in (14), the transfer function can be written in We let: NT = 50 (Order of Taylor Series
the following form: [ N1 ( z) + z 0.5 ( N2 ( z)]/[D1 (z) + z0.5 D2 ( z)] , where Expansion);m = 9; n = 9; T = 0.01;
N1 ( z ) , N 2 ( z ) , D1 ( z ) and D2 ( z ) are rational polynomials. This
1) The CFE of the bilinear discretization
is due to the fact that when s has an even power this will yield
Applying the bilinear transform to equation (6) then
N1 ( z ) in the numerator and D1 ( z ) in the denominator. applying the CFE approximation method yields the following
Whereas when the power is odd then z0.5 can be factored rational transfer function, if we only considered it up to its
outside yielding N 2 ( z ) in the numerator and D2 ( z ) in the 18th order.
denominator. An alternative to the above approach is to
substitute for the half sample advance or delay the 15.8 z18 - 17.49 z17 - 56.59 z16 + 64.15 z15 + 81.94 z14 - 96.07
approximations proposed in [4]. z13 - 61.36 z12 + 75.58 z11 + 25.15 z10 - 33.47 z9 - 5.439 z8 +
8.297 z7 + 0.5236 z6 - 1.073 z5 - 0.008252 z4 + 0.06111 z3 -
0.0008988 z2 - 0.00105 z + 5.231e-005
30
Analog Ideal --------------------------------------------------------------------------
z18 - 0.306 z17 - 4.214 z16 + 1.197 z15 + 7.369 z14 - 1.917 z13 -
25 Analog Approximation
Magnitude (dB)

Bilinear
20 Al-Alaoui 6.918 z12+ 1.619 z11 + 3.766 z10 - 0.7735 z9 - 1.197 z8 +
0.2078 z7 + 0.2124 z6 - 0.02933 z5 - 0.01873 z4 + 0.001838 z3
Al-Alaoui-1
15

10 + 0.0006293 z2 - 3.267e-005 z - 5.083e-006

5
-2 -1 0 1 2 3
(15)
10 10 10 10
Frequency (rad/sec)
10 10
2) The CFE of Al-Alaoui and Al-Alaoui-1 discretizations
40
Applying the Al-Alaoui transform, equation (2), to
Analog Ideal equation (6) then applying CFE method to the resulting
30 Analog Approximation
irrational disctrete time function yields the following rational
Phase (deg)

Bilinear
20 Al-Alaoui transfer function if we only considered it up to its 18th order.
Al-Alaoui-1
10
12.7 z18 - 105.7 z17 + 398.4 z16 - 897.1 z15 + 1343 z14 - 1403
z13 + 1044 z12 - 553.3 z11 + 203.3 z10 - 47.97 z9 + 5.545 z8 +
0

-10
10
-2
10
-1
10
0 1
10
2
10
3
10 0.318 z7 - 0.1912 z6 + 0.01836 z5 + 0.0007462 z4 - 0.000199
Frequency (rad/sec)
z3 + 4.167e-006 z2 + 3.157e-007 z + 1.669e-009
Figure 2. The frequency response of the ideal, the analog approximation, --------------------------------------------------------------------------
the discretization of the analog approximation using the bilinear , the Al- z18 - 7.889 z17 + 28 z16 - 58.94 z15 + 81.62 z14 - 77.79 z13 +
Alaoui and the Al-Alaoui-1 transformations. 51.74 z12 - 23.66 z11 + 6.994 z10 - 1.077 z9 - 0.02814 z8 +
0.0433 z7 - 0.005877 z6 - 0.0001575 z5 + 8.696e-005 z4 -
IV. DIRECT DISCRETIZATION 2.923e-006 z3 - 3.441e-007 z2 + 1.238e-008 z
The direct discretization approach consists of two steps. + 8.385e-011
In the first step an s-to-z transform is applied to the irrational (16)
transfer function of the fractional order controller to obtain an The CFE corresponding to Al-Alaoui-1 transform is
irrational discrete time transfer function, a generating obtained, as in Chen and Moore [8], by applying phase
function, of the controller. The direct discretization in this compensation to (16). Figure 3 plots the corresponding
paper applies the CFE method to the generating function, to frequency responses of the direct discretization approach
obtain digital rational approximations, IIR filters, to the employing CFE.
continuous fractional order controller C(s) [5], [7-9].
V. CONCLUSIONS
A. Generating Functions Figures 2-4 demonstrate that the Al-Alaoui and the Al-
Direct discretization of s r can be expressed by the Alaoui-1 transforms approach the ideal magnitude response
generating function obtained from an s-to z transform better than the bilinear transform at high frequencies.

329
However, the bilinear transform phase is an ideal 90 degrees indirect, is superior to the other in terms of accuracy. This
while the Al-Alaoui's presents an almost linear phase starting will depend on the different approximations that evolve over
from 90 degrees at low frequencies with that disadvantage time for both approaches in addition to the problem at hand.
overcome by Al-Alaoui-1. Figure 2 presents the results of the
indirect discretization approach of controller, C(s) and ACKNOWLEDGMENT
reflects clearly the properties of the transforms used. Figure 3 This research was initiated during my visit to the
presents the results of the direct discretization of C(s) using Adaptive Systems Laboratory at UCLA in 2007-2008. It is a
the different transforms and CFE. Figure 4 shows that the pleasure to acknowledge Prof. A. H. Sayed for his invitation
lowest absolute magnitude error is obtained by using direct and help. I am grateful to Cassio Lopes and Qiyue Zou for
discretization using Al-Alaoui+CFE. On the other hand it is their help during my stay at UCLA.
observed that all the responses of the indirect discretization
methods perform better than the direct discretization I am grateful to the highly talented graduates of the ECE
approach using bilinear + CFE. If we consider the step Department at AUB. It is a pleasure to acknowledge Elias
responses of the unity feedback system using the analog Yaacoub, Jimmy Azar, and Mohammed Baydoun for their
approximation and the responses due to direct discretization it help in this work.
should be known that direct discretization methods yield
REFERENCES
shorter rise time than the analog approximation which the
indirect approaches can at best approximate. [1] M. A. Al-Alaoui, "Novel Digital Integrator and Differentiator"; IEE
Electronics Letters, Vol. 29, No. 4, pp. 376-378, 18 February 1993.
(See also ERRATA, IEE Electronics Letters, Vol. 29, No. 10, pp.934,
1993.)
[2] M. A. Al-Alaoui, " Filling the Gap Between the Bilinear and the
60 Backward Difference Transforms: An Interactive Design Approach";
Ideal International Journal of Electrical Engineering Education. Vol. 34, No.
Magnitude (dB)

40
Bilinear 4, pp. 331-337, October 1997.
Al-Alaoui
[3] M. A. Al-Alaoui, “Al-Alaoui Operator and the α-Approximation for
Al-Alaoui-1
20
Discretization of Analog Systems,” Facta Universitatis (NIS),
Ser.:Elec. Energ. , Vol. 19, No. 1, pp. 143-146, April 2006.
[4] M. A. Al-Alaoui, “Using Fractional Delay to Control the Magnitudes
0
-2 -1 0 1 2 3 and Phases of Integrators and Differentiators”, IET Signal Processing,
10 10 10 10 10 10
Frequency (rad/sec)
Vol. 1, issue 2, pp. 107-119, June 2007.
[5] M. A. Al-Alaoui, " Simulation and Discretization of Fractional Order
60 Systems " The 2009 International Conference on Genetics and
Ideal Evlutionary Methods, (GEM'09), Las Vegas, July 13-16 2009.
40 Bilinear
Phase (deg)

[6] M. Aoun, R. Malti, F. Levron, A. Oustaloup, “ Numerical Simulations

Al-Alaoui
20 of Fractional Systems,” Proceedings of DETC’03, September, 2003.
Al-Alaoui-1
[7] J.-Y. Cao, J. Liang, B.-G. Cao, “ Optimization of Fractional Order PID
0 Controllers Based on Genetic Algorithms,” Proceedings of the 4th
International Conference on Machine Learning and Cybernetics,
-20
10
-2
10
-1
10
0
10
1 2
10 10
3 Guangzhou, China, August, 2005.
Frequency (rad/sec) [8] Y.Q. Chen and K.L. Moore, “Discretization for Fractional Order
Differentiators and Integrators”, IEEE Trans. on Circuits and Systems I,
vol. 49, pp 363-367, March 2002.
Figure 3. CFE comparison results for bilinear and Al-Alaoui
[9] Y.Q. Chen, B.M. Vinagre, and I. Podlubny. "Continued fraction
discretizations.
expansion approaches to discretizing fractional order derivatives-an
expository review", Nonlinear Dynamics 38 (1-4), pp. 155-170, 2004.
10 [10] S. C. Dutta Roy, “ On the Realization of Constant-Argument
Analog Approximation + Bilinear Immittance or Fractional Operator,” IEEE Trans. Circuit Theory, Vol.
9 Analog Approximation + Al-Alaoui CT-14, pp. 264-274, Sept. 1967.
Analog Approximation + Al-Alaoui-1
8
CFE + Bilinear [11] S. C. Dutta Roy, “A Rational Discrete Approximation of Some
CFE + Al-Alaoui Irrational Functions Through a Flexible Continued Fraction
Absolute Magnituded Error

7
Expansion,”Proc. IEEE (Letters), vol. 70, no. 1, pp.84-85, Jan 1982.
6
[12] A. N. Khovanskii, “The Application of Continued Fractions and Their
5 Generalizations to Problems in Approximation Theory”, Transl.
4
Groningen, The Netherlands: Noordhoff, 1963.
[13] L. Lorentzen, and H. Waadeland: Continued Fractions with
3
Applications, Addison- Wesley, North-Holland, Amsterdam, 1992.
2 [14] S. K. Mitra: Digital Signal Processing, 3rd Edition: New York,
1 McGraw-Hill, 2005.
0
[15] A. Oustaloup: La derivation non entiere. Hermes, Paris, 1995. .
0 50 100 150 200 250 300 350
Frequency (rad/sec)
[16] H. S. Wall: Analytic Theory of Continued Fractions: New York, Van
Nostrand, 1948.
Figure 4. Error Comparison of Direct and Indirect Discretization Methods. [17] D. Xue, C. Zhao, Y. Q. Chen,“ Fractional Order PID Control of a DC
Motor with Elastic Shaft: A Case Study,” Proceedings of the 2006
The above shows that there is no inherent reason to think American Control Conference, June, 2006.
that any of the two discretization approaches, direct or

330