www.ietdl.

org

Published in IET Circuits, Devices & Systems

Received on 3rd May 2010
Revised on 14th July 2010
doi: 10.1049/iet-cds.2010.0141

ISSN 1751-858X

Field programmable analogue array

implementation of fractional step filters
T.J. Freeborn1 B. Maundy1 A.S. Elwakil2
1
Department of Electrical and Computer Engineering, University of Calgary, Calgary, Alberta, Canada
2
Department of Electrical and Computer Engineering, University of Sharjah, P.O. Box 27272, Sharjah, United Arab Emirates
E-mail: [email protected]

Abstract: In this study, the authors propose the use of field programmable analogue array hardware to implement
an approximated fractional step transfer function of order (n + a) where n is an integer and 0 , a , 1. The
authors show how these filters can be designed using an integer order transfer function approximation of the
fractional order Laplacian operator s a. First and fourth-order low- and high-pass filters with fractional steps
from 0.1 to 0.9, that is of order 1.1–1.9 and 4.1–4.9, respectively, are given as examples. MATLAB simulations
and experimental results of the filters verify the implementation and operation of the fractional step filters.

1 Introduction theory is relatively new, and has shown applications in

transmission media [2], power electronics [8], integrator
Traditionally the Laplacian operator, s, is raised to an integer [9, 10] and differentiator circuits [11], oscillators [12],
order, that is, s, s 2, . . . , s n, when used in the design of multivibrator circuits [13] and filter theory [14 – 17] with
analogue circuits. However, it is also mathematically valid potentially many other applications.
to raise it to a non-integer order, s a, where a e <
representing the differintegrator [1] also referred to in this
work as the fractional Laplacian operator. A fractional The fractional Laplacian operator is especially useful in the
derivative of order a is given by the Grünwald– Letnikov design of filters with fractional step characteristics in the
approximation [2] as stopband, as the design of transfer functions can be done
algebraically rather than through solving the difficult time
domain representations of the fractional derivatives. The
(DT )−a 1
G(j − a)
Da f (t) W lim f (t − jDT ) (1) stopband attenuation of integer order filters has been
DT 0 G(−a)
j=0
G(j + 1) limited to increments based on the order, n, but using the
fractional Laplacian operator attenuations between these
where DT is the integration step and G(·) is the gamma integer steps can be achieved creating a fractional step filter
function. Applying the Laplace transform to (1) yields of order (n + a), where a is the fractional step between
integer orders n and n + 1 and is therefore limited to 0 ,
L{Da f (t)} = sa F (s) (2) a , 1. Previous work on fractional step filters [14, 15] has
shown these filters to be realisable with precise control of
Although there are no physical analogies to these derivatives, the stopband characteristics. However, using the transfer
like slope or area under the curve, the concepts of fractional function in [14] of order (1 + a) yields an undesired
calculus and fractional-order systems have slowly been peaking in the passband that increases with increasing a.
migrating into engineering showing many useful Using the fractional transfer functions proposed in [18] a
applications in diverse fields. These include materials fractional step through the stopband region is achieved
theory, diffusion theory, bioengineering, circuit theory and while eliminating the undesired passband peaking and
design [1], control theory [3, 4], electromagnetics [5] and maintaining a flat passband response through appropriate
robotics [6, 7]. The import of these concepts into circuit selection of the constants k2,3 in [18]. The elimination of

514 IET Circuits Devices Syst., 2010, Vol. 4, Iss. 6, pp. 514– 524
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cds.2010.0141
 www.ietdl.org

Figure 1 Comparison of magnitude responses of FLPFs of Figure 2 Magnitude and phase of s0.5 using second-order
order (1 + a) ¼ 1.9 from [14, 18] approximation (dashed) compared to ideal case (solid)

the passband peaking in the magnitude response using the

proposed fractional low-pass transfer function in [18] of approximations with varying order and accuracy, with the
order (1 + a) compared to that previously proposed in [14] accuracy and approximated frequency band increasing as
is shown in Fig. 1 for the case when a ¼ 0.9 with the order of the approximation increases. Using the CFE
constants k2 ¼ 1.31 and k3 ¼ 0.99. method of [24] we obtain the following approximation for
the general Laplacian operator to second order
In this paper, we explore the feasibility of the new
approximation to the fractional step filter design using a (a2 + 3a + 2)s2 + (8 − 2a2 )s + (a2 − 3a + 2)
sa † (3)
field programmable analogue array (FPAA) as a means of (a2 − 3a + 2)s2 + (8 − 2a2 )s + (a2 + 3a + 2)
implementation and test. We use a second-order
approximation for the fractional Laplacian operator to Note that when a = 0.5, (3) reverts back to the
realise both fractional low-pass filter (FLPF) and fractional approximation of s 0.5 in [24]. Fig. 2 shows an example of
high-pass filter (FHPF) of orders 1.1  1.9 and the magnitude and phase of the approximation of the
4.1  4.9. MATLAB simulations are verified fractional Laplacian operator for the case when a ¼ 0.5
experimentally using the FPAA. The precise attenuation compared to the ideal case. It can be observed that for v [
control offered by these filters is highlighted using [0.032, 31.53] the magnitude error does not exceed
approximated fractional high-pass filters of order 4.1  4.9 1.375 dB while for v [ [0.142, 7.00] the phase error does
against fourth- and fifth-order standard highpass not exceed 3.28. Using a second-order approximation for
Butterworth filters [19–21]. the Laplacian operator results in an (n + 2) integer order
filter to approximate the (n + a) fractional step filter, which
is less expensive to implement in hardware over
2 Approximation of the fractional approximations of higher order [3, 25].
Laplacian operator
The use of the fractional Laplacian operator, s a, has
theoretically been shown to produce filters with a fractional 3 Fractional step filters
step through the stopband [14, 15, 17]. However, there are
no commercial fractance devices available for the physical Using the approximation of the fractional Laplacian operator
realisation of these filters. Although most capacitors do in (3), a fractional step low-pass transfer function of the form
exhibit fractional behaviour [22, 23] and should be
k1
modelled with impedance ZC ¼ 1/s aC, the value of a is LP
H1+ a (s) = (4)
very near to 1 preventing their use in implementing sa (s + k2 ) + k3
fractional filters with complete control of the stopband
attenuations. Therefore until commercial fractance devices changes to
become available to physically realise circuits that make use
of the advantages of s a, integer order approximations have LP k1 (a2 s2 + a1 s + a0 )
H1+a (s) † (5)
to be used. There are many methods used to create an a0 s3 + c0 s2 + c1 s + c2
approximation of s a that include continued fraction
expansions (CFEs) as well as rational approximation where a0 = a2 + 3a + 2, a1 = 8 − 2a2 , a2 = a2 − 3a + 2,
methods [3]. These methods present a large array of c0 = (a1 + a0 k2 + a2 k3 )/a0 , c1 = (a1 (k2 + k3 ) + a2 )/a0 and

IET Circuits Devices Syst., 2010, Vol. 4, Iss. 6, pp. 514– 524 515
doi: 10.1049/iet-cds.2010.0141 & The Institution of Engineering and Technology 2010
www.ietdl.org

c2 = (a0 k3 + a2 k2 )/a0 . Through careful selection of k2,3 the than a larger value.), R, of 0.15239 and 0.034421 for the
passband region can be shaped while maintaining the interpolated functions of k2 and k3 , respectively.
desired fractional step through the stopband. To create a
flat passband the constants k2,3 when k1 = 1 that yield the
minimum cumulative passband error when compared to the
Butterworth response (The Butterworth response was 4 FPAA implementation
chosen because it offers a maximally flat passband response. Anadigm FPAAs are ‘analogue signal processors’ consisting
However, the use of Chebychev or Bessel filters is also of fully configurable analogue modules (CAMs) surrounded
possible.) over a select frequency range, were calculated by programmable interconnect and analogue input/output
numerically. These cumulative magnitude errors, |EC ( jv)|, cells [26]. The signal processing occurs in the CAMs using
can be calculated as fully differential switched capacitor circuitry, which provide
specialised behaviours such as filtering, gain, sample and

N
LP hold, summing, rectification and more. This provides a very
|EC ( jv)| = | |B1 ( jvi )| − |H1+a ( jvi )| | (6)
flexible architecture that can be easily reconfigured using
i=1
the AnadigmDesigner tools. These tools are a graphical
where |B1 ( jvi )| is the magnitude response at frequency vi of design environment to build circuits using the design
the first order low-pass Butterworth filter, |H1+ LP CAMs. In this design environment CAMs can be dropped
a ( jvi )| is the
magnitude response of the FLPF of order (1 + a) at in, wired together and configured for the desired design
frequency vi , and N is the number of samples. For our requirements. From the graphical implementation of a
analysis 200 samples were taken over the frequency range circuit, the AnadigmDesigner tools generate the
v e [0.01, 1]. For each value of a between 0.01 and 0.99 in configuration data file to program the FPAA.
steps of 0.01, the cumulative error was calculated for all
combinations of 0 , k2 , 2 and 0 , k3 , 1 in steps of Two CAMs that are particularly useful in the
0.001. From all the cumulative errors collected for each a implementation of approximated fractional step filters are
value, the combination that resulted in the minimum error the bilinear and biquadratic filter CAMs. The cascaded
was selected. These values that yield the minimum connection of these CAMs in the AnadigmDesigner
cumulative passband error when using the approximation as environment is shown in Fig. 4. These CAMs realise
well as the interpolated functions to match the collected data bilinear and biquadratic transfer functions given the pole
are shown in Fig. 3 as solid and dashed lines, respectively. and zero frequencies and quality factors making them ideal
The interpolated quadratic and linear functions from the for the realisation of filters that have been decomposed into
collected raw data for k2 and k3 , respectively, are found as biquadratic and bilinear sections. Previous discrete
component realisations of approximated FLPFs [18]
required the use of the design equations from [27] to
k2 = 1.0683a2 + 0.161a + 0.3324 (7) determine the component values to realise the biquadratic
transfer function using the STAR-SAB topology [28].
k3 = 0.29372a + 0.71216 (8)
Therefore using the biquadratic and bilinear filter CAMs
from the AnadigmDesigner tools greatly reduces both the
with a norm of residuals (The norm of residuals is calculated number of calculations and time required to physically
from the fit residuals, defined as the difference between the implement approximated fractional step filters.
ordinate data point and the resulting fit for each abscissa
data point; with a lower norm value indicating a better fit

4.1 Fractional low-pass filter

To realise the transfer function of (5) we decompose it into
first- and second-order transfer functions that can be
realised using bilinear and biquadratic filter CAMs,
respectively. For our realisation, the two transfer functions
generated were

1 e0 s2 + e1 s + e2
H (s) = H1 (s)H2 (s) = (9)
s + d0 s2 + d1 s + d2

where H1(s) and H2(s) represent first- and second-order

Figure 3 Plot of k2 and k3 against a that yields minimum transfer functions, respectively. Coefficients d0,1,2 and e0,1,2
passband error in (5) and the interpolated functions as can be determined through the numerical solution of the
solid and dashed lines, respectively system of equations obtained from equating the terms of

516 IET Circuits Devices Syst., 2010, Vol. 4, Iss. 6, pp. 514– 524
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cds.2010.0141
 www.ietdl.org

Figure 4 Cascaded connection of the bilinear and biquadratic filter CAMs in the AnadigmDesigner environment

(5) and (9), resulting in the following system of equations where T1 (s) and T2 (s) are the transfer functions of the bilinear
and biquadratic filter CAMs, respectively. Here G1 is the gain
a1 + a0 k2 + a2 k3 of T1 (s), f1 the pole frequency of T1 (s), f2p,z the pole and zero
d0 + d1 =
a0 frequencies of T2 (s) and Q2p,z the pole and zero quality factors
a (k + k3 ) + a2 of T2 (s). Before equating the biquadratic and bilinear
d0 d1 + d2 = 1 2
a0 transfer functions of (9)– (11) the frequency transformation
a0 k3 + a2 k2 s = s/vo = s/2pf0 must be applied to (9), where f0 is the
d0 d2 = denormalised frequency. The design equations to implement
a0 (10)
the approximated FLPF of order (1 + a) can therefore be
a
e0 = k1 2 summarised as
a0
a1
e1 = k 1
a0 f1 = d0 f0

e2 = k 1 e
f2z = f0 2
e0
The transfer function of (9) must also be arranged to match √
e0 e2
the CAMs, taking the form Q2z =
e1
 (12)
H (s) = T1 (s)T2 (s) f2p = f0 d2
2pf1 G1 
T1 (s) = d2
s + 2pf1 (11) Q2p =
d1
2
s + (2pf2z )/(Q2z )s + 4p2 f22z e
T2 (s) = − G1 = 0
s2 + (2pf2p )/(Q2p )s + 4p2 f22p d0

IET Circuits Devices Syst., 2010, Vol. 4, Iss. 6, pp. 514– 524 517
doi: 10.1049/iet-cds.2010.0141 & The Institution of Engineering and Technology 2010
www.ietdl.org

As an example the realised values of pole and zero frequencies implemented on silicon, the AnadigmDesigner tools select
and quality factors for both the bilinear and biquadratic the capacitor values with the best ratios to satisfy the design
CAMs, for approximated FLPFs of orders (1 + a) = 1.1, parameters entered. However, these best ratios do not
1.5 and 1.9, when f0 = 1 kHz, are given in Table 1a. In always meet the exact parameters which results in the
Table 1b the values of d0,1,2 and e0,1,2 computed from (10) variation between the theoretical and realised values.
that lead to Table 1a for the same values of a are shown.
Note the values of k2,3 shown yield the minimum passband
4.1.1 Higher-order FLPF design (n ≥ 2): From the
error with the second-order approximation of the fractional
stability analysis in [18] the transfer function of (4) is always
Laplacian operator when k1 ¼ 1. It should be mentioned
unstable for (n + a) ≥ 2 when n ≥ 2. Therefore this transfer
that the realised values given in Table 1a vary from the
function is not able to realise stable higher-order fractional
theoretical because of the limitations on the values that can
step filters. To overcome this, it was suggested in [18] to
be implemented by the FPAA. The biquadratic and bilinear LP
employ H1+ a (s), which is always stable when 0 , a , 1,
filter CAMs cannot realise all possible values because of
divided by higher-order normalised Butterworth polynomials
hardware limits as a result of the design parameters being
[20]. This creates a stable higher-order fractional step filter
interrelated to other parameters as well as the sample clock
of order (n + a) which can be written as
frequency. The corner frequencies of both poles and zeroes
are linearly related to the sample clock frequency, Fc , with LP
the absolute upper and lower values limited to Fc/10 and LP H1+a (s)
Hn+a (s) = , n≥2 (13)
Fc/500, respectively. Also, the corner frequencies, quality Bn−1 (s)
factors and gains are all interrelated based on the capacitors
of the internal switched capacitor circuits of the FPAA. where Bn (s) is a standard Butterworth polynomial of order n.
Since there are a finite number of capacitor values Using this method higher-order filters with a fractional step

Table 1 a Theoretical and realised biquad and bilinear CAM values for physical implementation of approximated (1 + a) order
FLPFs and b d0,1,2 and e0,1,2 values for decomposed first- and second-order transfer functions to realise an approximated FLPF
of orders 1.1, 1.5 and 1.9

a
Design value Order (1 + a)
1.1 1.5 1.9
Theoretical Realised Theoretical Realised Theoretical Realised
f1, kHz 0.305 0.309 0.455 0.456 0.697 0.699
f2p, kHz 1.81 1.81 1.48 1.47 1.20 1.21
f2z kHz 1.16 1.17 2.24 2.23 7.08 7.15
Q2p 0.447 0.445 0.618 0.619 0.653 0.637
Q2z 0.249 0.249 0.224 0.225 0.122 0.122
G1 2.43 2.40 0.439 0.429 0.0287 0.0287
b
Value (1 + a)kk32
(1 + 0.1)0.35
0.74 (1 + 0.5)0.69
0.86 (1 + 0.9)1.36
0.97

d0 0.3056 0.4511 0.6920

d1 4.0467 2.4109 1.8452
d2 3.2689 2.2125 1.4409
e0 0.7403 0.2000 0.0120
e1 3.4545 2.0000 1.1579
e2 1.0000 1.0000 1.0000
k2,3 shown are for minimum passband error

518 IET Circuits Devices Syst., 2010, Vol. 4, Iss. 6, pp. 514– 524
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cds.2010.0141
 www.ietdl.org

through the stopband while maintaining the flat passband following system of equations
response can be implemented. Realising these on the FPAA
requires cascading further combinations of bilinear and a0
d0 d2 =
biquadratic filter CAMs in the AnadigmDesigner a0 k3 + a2 k2
environment with those designed to implement (5). With a0 k2 + a1 + a2 k3
the added CAMs designed to implement a standard low- d0 d1 + d2 =
a0 k3 + a2 k2
pass Butterworth filter of order (n 2 1) frequency shifted to
f0 to realise the (n + a) FLPF. a1 (k2 + k3 ) + a2
d0 + d1 =
a0 k 3 + a 2 k 2 (14)

4.2 Fractional high-pass filter e0 = k 1

a1
To obtain a high-pass filter from its low-pass counterpart is e1 = k 1
a0
straightforward and implies use of the LP-to-HP
transformation [21]. That is, replacing s with 1/s in the LP a
e2 = k 1 2
transfer function, which when applied to (5) results in a a0
high-pass filter with a fractional step through the stopband
region. These FHPF can also be physically realised using The realised values for both the bilinear and biquadratic
the FPAA using the same pole and zero frequency and CAMs, for approximated FHPFs of orders (1 + a) = 1.1,
quality factor design equations of (12) and bilinear and 1.5 and 1.9, when f0 = 10 kHz are given in Table 2a.
biquadratic filter CAMs configured in the high-pass Correspondingly, using (14) the values of d0,1,2 and e0,1,2
configuration. However, while the same design equations for the same order to the values of k2,3 that yield minimum
can be utilised, the values of d0,1,2 and e0,1,2 are different passband error are presented in Table 2b. Similar to how
than the low-pass values and must be calculated from the stable higher-order FLPFs are obtained, higher-order

Table 2 a Theoretical and realised bilinear and biquad CAM values for physical implementation of approximated (1 + a) order
FHPFs and b coefficients d0,1,2 and e0,1,2 values for decomposed first- and second-order transfer functions to realise an
approximated FHPF of orders 1.1, 1.5 and 1.9

a
Design value Order (1 + a)
1.1 1.5 1.9
Theoretical Realised Theoretical Realised Theoretical Realised
f1, kHz 34.1 32.8 21.99 21.9 14.4 14.4
f2p, kHz 5.52 5.22 6.75 6.76 8.35 8.35
f2z, kHz 8.60 8.66 4.47 4.27 1.41 1.42
Q2p 0.447 0.448 0.618 0.632 0.654 0.657
Q2z 0.249 0.247 0.224 0.228 0.122 0.123
b
Value (1 + a)k3 k2

(1 + 0.1)0.355
0.737 (1 + 0.5)0.681
0.863 (1 + 0.9)1.352
0.973

d0 0.3410 2.1988 1.4352

d1 4.1726 1.0918 1.2768
d2 2.9332 0.4552 0.6968
e0 1.0000 1.0000 1.0000
e1 3.4545 2.0000 1.1579
e2 0.7403 0.2000 0.0200

k2,3 shown are for minimum passband error

IET Circuits Devices Syst., 2010, Vol. 4, Iss. 6, pp. 514– 524 519
doi: 10.1049/iet-cds.2010.0141 & The Institution of Engineering and Technology 2010
www.ietdl.org

FHPFs are also possible by dividing by the Butterworth

polynomial after applying the low-pass to high-pass and
frequency shifting transformations. This creates a stable
higher-order high-pass fractional filter of order (n + a) that
can be written as

HP
HP H1+a (s)
Hn+a (s) = HP
, n≥2 (15)
Bn−1 (s)

where BnHP (s) is a standard Butterworth polynomial of order n

after applying the LP to HP transformation.

5 Simulations and experimental

results
To verify the proposed fractional step filters, both low- and
high-pass filters of orders (1 + a) and (4 + a) for a = 0.1,
Figure 5 Test setup to measure the magnitude response of 0.5 and 0.9 were designed in the AnadigmDesigner
circuits implemented on the AN231E04 with a HP4395A environment using the values from Tables 1a and 2a
network analyser and realised on an AN231E04 FPAA development kit
from Anadigm. The magnitude responses of the

Figure 6 MATLAB simulation and FPAA experimental results of the magnitude response of the approximated (1 + a) order
FLPF, shown as solid and dashed lines, respectively, compared against the first- and second-order standard Butterworth
low-pass filters

Algorithm 1 MATLAB code to simulate and plot FLPF of order (1 + a) ¼ 1.1 frequency shifted to 1 kHz

520 IET Circuits Devices Syst., 2010, Vol. 4, Iss. 6, pp. 514– 524
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cds.2010.0141
 www.ietdl.org

Table 3 Theoretical, simulated and experimental stopband Table 4 Theoretical, simulated and experimental stopband
attenuations of approximated (1 + a) order FLPF realised attenuations of approximated (4 + a) order FLPF realised
using FPAA using FPAA

Order Theoretical, Simulated, Experimental, Order Theoretical, Simulated, Experimental,

(1 + a) dB/dec dB/dec dB/dec (4 + a) dB/dec dB/dec dB/dec
1.1 222 221.34 221.33 4.1 282 281.39 280.80
1.5 230 230.75 230.61 4.5 290 290.76 290.40
1.9 238 238.81 239.00 4.9 298 298.3 296.80

implemented fractional step filters were measured using a In a second test the magnitude response of an
HP4395A network/spectrum analyser with the test setup approximated fourth-order FLPFs with fractional steps of
being shown in Fig. 5. In the sections that follow, we 0.1, 0.5 and 0.9 was obtained and is shown in Fig. 7 as
examine first the approximated fractional step filters dashed lines, with the MATLAB simulation results of the
followed by an application of the fractional step filter in theoretical transfer functions presented as solid lines. The
two tone detection. experimental results show close agreement with the
simulations confirming that the method of implementing a
higher-order FLPF by dividing by a higher-order
Butterworth polynomial is valid. Also, this method of
5.1 Approximated fractional step filters implementing a higher-order filter maintains the fractional
In the first of the experimental tests the results of the step through the stopband without the passband peaking.
magnitude response of the approximated first-order FLPFs Comparing the results, we find that the passband
with fractional steps of 0.1, 0.5 and 0.9 are shown in Fig. 6 attenuation of the experimental and simulated filters are
as dashed lines, with the corresponding MATLAB very close to their theoretical values of −20(4 + a) dB/dec.
simulation results of the theoretical transfer function For comparison, these attenuations are listed in Table 4. In
presented as solid lines. The MATLAB code to simulate a third and final experimental test the results of the (1 + a)
and plot the theoretical transfer function of (5) when and (4 + a) FHPFs are also shown in Figs. 8 and 9,
a = 0.1 frequency shifted to 1 kHz is presented in respectively. MATLAB results are shown alongside as solid
Algorithm 1. The results indicate that the stopband lines, with measured stopband attenuations given in
attenuation of the experimental and simulated filters are Tables 5 and 6, respectively. As before close agreement in
very close to their theoretical values of 220(1 + a) dB/dec. the values can be easily observed.
For comparison, these attenuations are listed in Table 3.
Note that in both the simulation and experimental results
the stopband is not perfectly linear, which is a result of the
5.2 Application of a fractional step filter
deviation from the ideal Laplacian operator associated with To highlight the precise filtering achieved by a fractional step
using the second-order approximation of (3). filter, two tones are applied to an approximated high-pass

Figure 7 MATLAB simulation and FPAA experimental results of the magnitude response of the approximated (4 + a) order
FLPF, shown as solid and dashed lines, respectively, compared against the fourth- and fifth-order standard Butterworth
lowpass filters

IET Circuits Devices Syst., 2010, Vol. 4, Iss. 6, pp. 514– 524 521
doi: 10.1049/iet-cds.2010.0141 & The Institution of Engineering and Technology 2010
www.ietdl.org

Figure 8 MATLAB simulation and FPAA experimental results of the magnitude response of the approximated (1 + a) order
FHPF, shown as solid and dashed lines, respectively, compared against the first- and second-order standard Butterworth
high-pass filters

Figure 9 MATLAB simulation and FPAA experimental results of the magnitude response of the approximated (4 + a) order
FHPF, shown as solid and dashed lines, respectively, compared against the fourth- and fifth-order standard Butterworth
high-pass filters
Table 5 Theoretical, simulated and experimental stopband filter of orders (4 + a) = 4.1 to 4.9 in steps of 0.2 with the
attenuations of approximated (1 + a) order FHPF realised output of the filter measured by the HP4395A network/
using FPAA spectrum analyser to determine the output signal power.
Tones at 3 kHz and 10 kHz with peak-to-peak voltages of
Order Theoretical, Simulated, Experimental, 500 mV were applied to the approximated FHPFs shifted
(1 + a) dB/dec dB/dec dB/dec to a frequency of f0 = 10 kHz. The peak value of the tones
1.1 222 221.43 221.20 for each filter are presented in Table 7. Note that the use

1.5 230 229.49 229.80

Table 7 Signal power of tones at 3 and 10 kHz after
1.9 238 238.82 237.58 application to approximated FHPFs of orders (4 + a) ¼ 4.1
to 4.9 in steps of 0.2
Table 6 Theoretical, simulated and experimental stopband
Order Power @ 3 kHz, Power @ 10 kHz,
attenuations of approximated (4 + a) order FHPF realised
(n + a) dBm dBm
using FPAA
4.1 232.7 3.94
Order Theoretical, Simulated, Experimental,
(4 + a) dB/dec dB/dec dB/dec 4.3 235.9 3.37
4.1 282 281.43 280.4 4.5 238.2 3.5
4.5 290 289.1 289.6 4.7 238.4 5.01
4.9 298 298.74 299.2 4.9 240.7 4.45

522 IET Circuits Devices Syst., 2010, Vol. 4, Iss. 6, pp. 514– 524
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cds.2010.0141
 www.ietdl.org

Figure 10 Frequency spectrum of the approximated 4.5 order FHPF (dashed) compared to fourth (dotted) and fifth (solid)
order high-pass Butterworth filters

of the approximated fractional Laplacian operator results in high-pass filters. The precise attenuation control of these
the deviation of the linear spacing between the powers of filters was highlighted in a simple two tone test.
the tone at 3 kHz as a increases. This control of the
attenuation is not possible using integer order filters. High-
pass Butterworth filters, frequency shifted to f0 = 10 kHz,
would result in signals of 232.5 and 243.4 dBm for a
7 References
3 kHz tone using fourth- and fifth-order filters,
[1] ORTIGUEIRA M. : ‘An introduction to the fractional
respectively. This precise control is also shown in the
continuous-time linear systems: the 21st century systems’,
frequency spectrum of Fig. 10. The spectrum of the fourth-
IEEE Circuits Syst. Mag., 2008, 8, (3), pp. 19 – 26
and fifth-order highpass Butterworth filters, as dotted and
solid lines, respectively, are compared to that of a 4.5 order
[2] OLDHAM K., SPANIER J.: ‘The fractional calculus: theory and
FHPF, shown as a dashed line. All of the filters maintain
applications of differentiation and integration to arbitrary
the tone at 10 kHz with the attenuation of the 4.5 order
order’ (Academic Press, New York, 1974)
filter at 3 kHz clearly between those of the standard
Butterworth filters. These results reinforce the experimental
[3] PODLUBNY I., PETRAS I. , VINAGRE B., O’LEARY P. , DORCAK L. :
results of the magnitude response in Fig. 9 showing that
‘Analogue realizations of fractional-order controllers’,
these filters can be used to precisely control attenuation
Nonlinear Dyn., 2002, 29, pp. 281– 296
characteristics.
[4] HABA T., LOUM G., ZOUEU J., ABLART G.: ‘Use of a component
with fractional impedance in the realization of an analogical
regulator of order 1/2’, J. Appl. Sci., 2008, 8, pp. 59 – 67
6 Conclusion
In this paper, we have presented and demonstrated a valid [5] MACHADO J., JESUS I., GALHANO A. , CUNHA J.: ‘Fractional
method of fractional filter design using the approximation order electromagnetics’, Signal Process., 2006, 86,
of (3) and (4) – (5) in extending the design of filters with a pp. 2637 – 2644
flat passband from the integer order to the more general
fractional order domain. The implementation of these [6] MIGUEL F. , LIMA M., MACHADO J., CRISOSTOMO M. :
filters using an FPAA verified this design theory. We have ‘Experimental signal analysis of robot impacts in a
also simulated and physically realised integer order filters fractional calculus perspective’, J. Adv. Comput. Intell.
that demonstrate a fractional step through the stopband for Intell. Inf., 2007, 11, p. 1079
both high-pass and low-pass filters of orders (1 + a) and
(4 + a), all of which show very close agreement between [7] FERREIRA N.F., DUARTE F., LIMA M., MARCOS M., MACHADO J.T.:
the simulated and experimental results. These results ‘Application of fractional calculus in the dynamical
further serve to verify that the low-pass to high-pass analysis and control of mechanical manipulators’,
transformation does yield a high-pass fractional step filter Fractional Calc. Appl. Anal., 2008, 11, (1), pp. 91 – 113
without peaking in the passband, and that the selection of
k2,3 for minimum passband error also applies to the [8] CALDERON A., VINAGRE B., FELIU V.: ‘Fractional order control
FHPFs. Finally, that the method for implementing higher- strategies for power electronic buck converters’, Signal
order fractional step filters is suitable for both low- and Process., 2006, 86, pp. 2803– 2819

IET Circuits Devices Syst., 2010, Vol. 4, Iss. 6, pp. 514– 524 523
doi: 10.1049/iet-cds.2010.0141 & The Institution of Engineering and Technology 2010
www.ietdl.org

[9] VARSHNEY P., GUPTA M., VISWESWARAN G.: ‘New switched [18] FREEBORN T., MAUNDY B., ELWAKIL A.: ‘Towards the realization
capacitor fractional order integrator’, J. Act. Passive of fractional step filters’. IEEE Int. Symp. on Circuits and
Electron. Devices, 2007, 2, pp. 187 – 197 Systems (ISCAS), May 2010, pp. 1037 – 1040

[10] SANTAMARIA G., VALVERDE J., PEREZ-ALOE R., VINAGRE B.: [19] SCHAUMANN R., VALKENBURG M.E.V.: ‘Design of analog filters’
‘Microelectronic implementations of fractional-order (Oxford University Press, New York, 2001)
integrodifferential operators’, J. Comput. Nonlinear Dyn.,
2008, 3, p. 021301 (6 pp.) [20] DELIYANNIS T., SUN Y., FIDLER J.: ‘Continuous time active filter
design’ (CRC Press LLC, New York, 1999)
[11] BISWAS K., SEN S., DUTTA P.K.: ‘Realization of a constant
phase element and its performance study in a [21] ZVEREV A.I.: ‘Handbook of filter synthesis’ (John Wiley
differentiator circuit’, IEEE Trans. Circuits Syst. II, Express and Sons, New York, 1967)
Briefs, 2006, 53, (9), pp. 802– 806
[22] WESTERLUND S.: ‘Dead matter has memory! [capacitor
[12] AHMAD W., EI-KHAZALI R., ELWAKIL A.: ‘Fractional-order wien- model]’, Phys. Scr. (Sweden), 1991, 43, pp. 174 – 179
bridge oscillator’, Electron. Lett. (UK), 2001, 37, pp. 1110–1112
[23] WESTERLUND S., EKSTAM L.: ‘Capacitor theory’, IEEE Trans.
[13] MAUNDY B., ELWAKIL A., GIFT S.: ‘On a multivibrator that Dielectr. Electr. Insul., 1994, 1, (5), pp. 826 – 839
employs a fractional capacitor’, Analog Integr. Circuits
Signal Process., 2010, 62, (1), pp. 99– 103 [24] KRISHNA B., REDDY K.: ‘Active and passive realization of
fractance device of order 1/2’, Act. Passive Electron.
[14] RADWAN A., SOLIMAN A., ELWAKIL A. : ‘First-order filters Compon., 2008, 2008, p. 5
generalized to the fractional domain’, J. Circuits Syst.
Comput., 2008, 17, (1), pp. 55– 66 [25] DJOUAMBI A. , CHAREF A., BESANCON A.: ‘Optimal
approximation, simulation and analog realization of the
[15] RADWAN A., ELWAKIL A., SOLIMAN A.: ‘On the generalization of fundamental fractional order transfer function’,
second-order filters to the fractional-order domain’, Int. J. Appl. Math. Comput. Sci., 2007, 17, (4), pp. 455– 462
J. Circuits Syst. Comput., 2009, 18, (2), pp. 361 – 386
[26] Anadigm: ‘3rd Generation dynamically reconfigurable
[16] LAHIRI A., RAWAT T.K.: ‘Noise analysis of single stage dpASP’. AN231E04 Datasheet Rev 1.1, 2007
fractional-order low pass filter using stochastic and
fractional calculus’, ECTI Trans. Electr. Eng. Electron. [27] HUELSMAN L.P.: ‘Active and passive analog filter design:
Commun., 2009, 7, (2), pp. 136– 143 an introduction’ (McGraw-Hill, Inc., 1993)

[17] MATOS C. , ORTIGUEIRA M.D. : ‘Emerging trends in [28] FRIEND J., HARRIS C., HILBERMAN D.: ‘Star: an active
technological innovation’ in CAMARINHA-MATOS L.M., PEREIRA biquadratic filter section’, IEEE Trans. Circuits Syst., 1975,
P., RIBEIRO L. (EDS.): ‘Fractional filters: an optimization CAS-22, pp. 115 – 121
approach’ (Springer, 2010), pp. 361– 366

524 IET Circuits Devices Syst., 2010, Vol. 4, Iss. 6, pp. 514– 524
& The Institution of Engineering and Technology 2010 doi: 10.1049/iet-cds.2010.0141