On design and applications of digital differentiators

B. T. Krishna
Department of E.C.E
GITAM University
Visakhapatnam, 530045, India
Email: [email protected]
S. Srinivasa Rao
Department of E.C.E
GITAM University
Visakhapatnam, 530045, India
Email: [email protected]
AbstractThis paper deals with the design and applications of
digital differentiators. For real time applications it is mandatory
that a differentiator should have as small an order as possible.
Different procedures available for the design of FIR and IIR
type digital differentiators are presented. The IIR Type digital
differentiators are obtained by inversion and magnitude stabiliza-
tion of the existing digital integrators. In some applications like
controls, waveshaping, oscillators and communications require a
constant 90
0
phase for differentiators. In this paper an attempt
is made to study about the variation of phase angle of digital
differentiators with the application of fractional delay. The use of
digital differentiators for the detection of edges in an image, QRS
detection in an ECG signal is illustrated. It has been observed
that the digital differentiators have shown superior performance
compared to the well known gradient method. With the proven
efciency of the differentiators in various applications, they have
been implemented in hardware using Verilog.
Index TermsDigital differentiator, Digital Integrator, IIR
Filter, FIR Filter, Interpolation, Simpson Integration Formula,
Fractional delay, Rectangular rule, Edge detection etc.
I. INTRODUCTION
Digital differentiators are used to nd the time-derivative of
the incoming signal [1-3] and are dened as,
G(j) = j (1)
These devices are used in almost all elds of engineering
like instrumentation, control systems, digital signal and image
processing, bio-medical engineering and other allied elds. So,
the design of digital differentiators is of considerable interest.
For real-time dynamic applications, a differentiator should
have a wide band frequency response and should have a group
delay as short as possible. For easy practical Implementation,
a low-order is more favourable.
Rabiner and Steiltz,[3] has presented the design of recursive
and non recursive digital differentiators. They have made use
of square error criterion for the design of recursive differen-
tiators, while sampling technique is used for the design of
non-recursive digital differentiators.
Taylor series approximations[2] have been widely used
to derive differentiators.They are all in the form of central
difference approximations. It has been observed that the higher
the order, the closer that a Taylor series approximation is to
the ideal differentiator.
Digital differentiator can be either FIR or an IIR type.A
FIR lter of type III has an odd length and antisymmetric
impulse response.Theoretically, FIR lters of type III can be
designed to meet requirements at nearly all frequencies, with
increased lter order[6]. A FIR lter of type IV has an even
length and antisymmetric impulse response.The type IV FIR
differentiators are found to be superior to the type III FIR
differentiators in terms of the frequency response.
Kumar and Dutta Roy,presented optimal and maximally
linear FIR differentiators for low-frequency, mid-frequency,
and around specic frequency respectively[6]. They gave the
explicit formulae and efcient recursive algorithms to calculate
the impulse response of lters. Although different procedures
exist for the design of FIR type differentiators for different
frequency regions,the differentiators obtained suffer from the
problem of higher order,high cost and space making them
unsuitable for real-time applications. In 1992, Al-Alaoui has
proposed a procedure for the design of IIR type digital
differentiators, which is based on the inversion and magnitude
stabilization of the transfer function of digital integrators[9-
12]. In this paper an attempt is made to study about the design
of FIR and IIR type digital differentiators.
Fractional Delay lters have been widely used in areas
such as arbitrary sampling rate conversion, synchronization of
digital modems and speech coding etc.In this paper an attempt
is made to apply fractional delay to digital differentiators
to achieve constant phase angle, and is illustrated with an
example.
The organization of the paper is as follows. Section 2 deals
with the design of FIR type and IIR type digital differentia-
tor.Section 3 illustrates the application of fractional delay to
IIR type digital differentiators. The use of digital differentia-
tors in detecting edges in an image, QRS detection in an ECG
signal is illustarted in Section 4. Hardware implementation of
digital dif- ferentiators using Verilog is presented in Section
5. Results and conclusions are presented in section 6.
II. DESIGN OF DIGITAL DIFFERENTIATORS
A. Taylor Series Method
Taylor series approximations have been widely used to
derive differentiators [2].They are all in the form of central
difference approximations such as
y(n) =
_
N
k=N
c (k) x(nk) (2)
IEEE- Fourth International Conference on Advanced Computing, ICoAC 2012
MIT, Anna University, Chennai. December 13-15, 2012
978-1-4673-5584-1/12/$31.002012 IEEE
Fig. 1. Magnitude response of Taylor series based digital differentiators.
where N is the order of Taylor series approximation. The co-
efcients can be calculated as,
c (k) =c (k) =
(1)
k+1
N!
2
k (Nk) ! (N + k) !
, c(0) = 0 (3)
The magnitude response of the digital differentiators obtained
using Taylor series expansion is as shown in Fig.1.It is
apparent that the higher the order, the closer that a Taylor
series approximation is to the ideal differentiator.
B. FIR type differentiators
For a causal FIR differentiator[3,4,6,7],
y(n) =
_
N
k=0
c (k) x(nN) (4)
c
d
[n] =
_

je
j(nM)
d =
cos ([nM] )
(nM)

sin ([nM] )
(nM)
2
(5)
Where M=N/2.The maximally at FIR approximation to the
ideal differentiator satises the derivative constraints.
|G(j)| = 0 at = 0 (6)

dG(j)
d

= 1 at = 0 (7)

d
k
G(j)
d
k

= 0 at = 0 2 = k = B 1 (8)
Where B is the number of the free parameters. Notice that the
approximation is performed at a single frequency point =
0. Kumar and Dutta Roy described how the solution can be
obtained from the maximally at low-pass FIR lter. Type III
and IV Low-pass digital differentiators are explained as below.
1) Type III and IV FIR Differentiators: A FIR lter of type
III has an odd length and antisymmetric impulse response.
c
d
[n] =
_
cos(nM)
nM
n = M
0 n = M
(9)
To eliminate the Gibbs phenomenon due to the nite trunca-
tion, a window function is required. Theoretically, FIR lters
of type III can be designed to meet requirements at nearly all
frequencies, as long as we increase the lter order.The Type
III transfer function can be written as,
H
4
(z) =
1
4
_
1 z
1
_ _
1 + z
1
_
H(z) (10)
That means a Type III transfer function always has a zero at
z = -1 and 1 whose frequency response can be written as,
H
4
_
e
j
_
= je
j
sin
_

2
_
cos
_

2
_
H(e
j
) (11)
A FIR lter of type IV has an even length and antisymmetric
impulse response.The type IV FIR differentiators are superior
to the type III FIR differentiators in terms of the frequency
response, since they have no disadvantageous characteristic of
being zero at = 1. The Type IV transfer function can be
written as,
H
4
(z) =
1
2
_
1 z
1
_
H(z) (12)
That means a Type IV transfer function always has a zero at
z = 1 whose frequency response can be written as,
H4
_
e
j
_
=
1
2
_
1 e
j
_
H(e
j
) = je
j/2
sin
_

2
_
H(e
j
)
(13)
C. IIR Type Digital Differentiators
An IIR type digital differentiator will be obtained from
a digital integrator. An ideal integrator is dened by the
following transfer function [8-11],
H
I
(j) =
1
j
(14)
The commonly used digital integrators are, Backward or
rectangular integrator,
H
1
(z) =
zT
z 1
(15)
Tustin or trapezoidal integrator,
H
2
(z) =
T(z + 1)
2(z 1)
(16)
Simpsons integrator,
H
3
(z) =
T(z
2
+ 4z + 1)
3(z
2
1)
(17)
Tick integrator,
H
4
(z) =
T(0.3585z
2
+ 1.2832z + 0.3584)
(z
2
1)
(18)
A Third order Digital Integrator is dened as,
H
5
(z) =
T(z + 2.3658)
_
z
2
+ 1.1752z + 0.047
_
2.7925z
2
(z 1)
(19)
IEEE-ICoAC 2012
0 0.5 1 1.5 2 2.5 3
0
0.5
1
1.5
2
2.5
3
Frequency
M
a
g
n
i
t
u
d
e
Comparison of Magnitude responses of Digital Integrators


Ideal
H
1
(z),Backward
H
2
(z),Bilinear
H
AL
(z),Al-Alaoui
H
3
(z),Simpson
H
4
(z),Tick
H
5
(z),Ngo
Fig. 2. Comparison of magnitude responses IIR type digital integrators.
Comparison of magnitude response of the above mentioned
digital integrators is presented in Fig.2.
From the gure it can be observed that the magnitude
response of an ideal integrator lies between rectangular and
trapezoidal integrators and also between trapezoidal and simp-
son integrators. In[8-10], the following approach is proposed
by Al-Alaoui for the design of IIR type digital differentiators.
1) Design an integrator that has the same range and accu-
racy as the desired differentiator and Invert the transfer
function.
2) Reect the poles that lie outside the unit circle to inside,
in order to stabilize the resultant transfer function.
3) Compensate the magnitude using the reciprocals of the
poles that lie outside the circle.
1) First order Al-Alaoui Digital Differentiator: The First
order Integrator is obtained by the Interpolation of Rectangular
and trapezoidal Integrators [8].So,
H
AL
(z) = H
1
(z) + (1 ) H
2
(z) (20)
Where lies between 0 and 1. For =
3
4
the above equation
reduces to,
H
AL
(z) =
3
4
H
1
(z) +
1
4
H
2
(z) (21)
Substituting H
1
(z) and H
2
(z) in the above equation and
simplifying,
H
AL
(z) =
T (z + 7)
8 (z 1)
(22)
Reecting the zero z = -7 with its reciprocal -1/7, and com-
pensating the magnitude results in a minimum phase digital
integrator with the transfer function
H
AL
(z) =
7T (z + 1/7)
8 (z 1)
(23)
Inverting the above transfer function yields the Al-Alaouis
stabilized IIR differentiator of the rst order
G
AL
(z) =
8(z 1)
7T(z +
1
7
)
(24)
2) Second order Digital Differentiators: The Tick integra-
tor has two real poles located at z = 1 whose transfer
function is given by [9],
H
4
(z) =
T(0.3585z
2
+ 1.2832z + 0.3584)
(z
2
1)
(25)
By following the above mentioned Procedure the following
transfer function obtained for the digital differentiator is,
G
4
(z) =
0.852(z
2
1)
(z
2
+ 0.611z + 1)
(26)
The transfer function of the Simpsons integrator is
H
3
(z) =
T(z
2
+ 4z + 1)
3(z
2
1)
(27)
The corresponding transfer function of the digital differentiator
will be,
G
3
(z) =
0.8038(z
2
1)
T(z
2
+ 0.5358z + 0.0718)
(28)
By interpolating the Simpson and trapezoidal digital integra-
tors the following hybrid digital integrator is obtained
H
AL2
(z) = H
3
(z) + (1 ) H
2
(z) (29)
Substituting the expressions of Simpsons and the trapezoidal
integrators,the following is the expression for the new digital
integrator,
H
AL2
(z) =
T(3 )
_
z
2
+
2(3+)
3
z + 1
_
6(z
2
1)
(30)
for = 0.6,
H
AL2
(z) =
0.4T
_
z
2
+ 2.5z + 1
_
(z
2
1)
(31)
Following the above mentioned procedure, the digital differ-
entiator obtained will be,
G
AL2
(z) =
1.25(z
2
1)
T (z
2
+ z + 0.25)
(32)
3) Third order or wide-band Digital Differentiator: In
2006,N.Q.Ngo has derived an expression for wideband digital
integrator which is dened as [12],
H
5
(z) =
T(z + 2.3658)
_
z
2
+ 1.1752z + 0.047
_
2.7925z
2
(z 1)
(33)
Following the above Procedure and stabilizing the transfer
function the following digital differentiator is obtained.
G
5
(z) =
1.1804z
2
(z 1)
T
_
z
3
+ 0.168z
2
0.0607z + 0.0198
_ (34)
IEEE-ICoAC 2012
Fig. 3. Comparison of magnitude responses of IIR Type digital differentia-
tors.
Fig. 4. Comparison of phase responses of IIR Type digital differentiators.
The magnitude and phase responses of the digital differen-
tiators were compared in Fig.3 and 4 respectively.Percent
relative error of the magnitude responses of all the designed
differentiators is presented in Figure.5. It can be observed
from Figs.3,4 and 5 that,
1) Al-Alaoui rst order differentiator approximates the
ideal differentiator till 0.78 of the full band.The third
order digital differentiator can be used as a wide-band
digital differentiator.Differentiator obtained from the in-
version of the trapezoidal integrator has phase response
closer to ideal one.
2) The inverse Simpson differentiator has the poorest high
frequency response and has good low frequency re-
sponse up to 0.4 of the full band.Differentiator from
the tick integrator is linear till 0.5 of the full band. Al-
Fig. 5. Percent relative error.
Alaoui second order differentiator has exhibited good
low frequency response.
3) First order and Third order digital differentiators has
exhibited 10 percent relative error for higher range
of frequencies.The lower order of these digital differ-
entiators makes them suitable in real time applications
like radars, sonars, bio-medical engineering, speech
processing,Global positioning system etc.
III. APPLICATION OF FRACTIONAL DELAY TO DIGITAL
DIFFERENTIATORS
The transfer function of an ideal delay element may be
written as [15-16],
H (z) =
Y (z)
X(z)
= z
D
(35)
For the sake of discussion, assume that D is a positive real
number, dened as the sum of its integer part, N, and the
fractional part, d.
D = N + d (36)
In the frequency domain, the ideal fractional-delay lter can
be described as,
H
_
e
j
_
= e
jD
(37)
i.e., the magnitude response for an ideal delay element is unity
for all frequencies, while the phase response is linear with a
slope of -D. The delay can be calculated by using FIR and
IIR approximations.
A.Thiran All pass lter: With this approximation the total
delay D is approximated by an IIR lter as [15-16],
z
D
=
a
0
+ z
1
a
1
+ z
2
a
2
+ . . . z
m
a
m
b
0
+ z
1
b
1
+ z
2
b
2
+ . . . z
n
b
n
(38)
a
k
= (1)
k
C
N
k
N

n=0
N n D
N n k D
(39)
IEEE-ICoAC 2012
Fig. 6. Fractional delay applied to AlAlaoui digital differentiator.
TABLE I
APPROXIMATIONS OF FRACTIONAL DELAY, D=0.5
Delay,d FIR Approximation IIR
Approximation
0.1
1+z
1
10
9+11z
1
11+9z
1
0.2
1+z
1
5
8+12z
1
12+8z
1
0.3
1+z
1
10/3
7+13z
1
13+7z
1
0.4
1+z
1
2.5
6+14z
1
14+6z
1
0.5
1+z
1
2
5+15z
1
15+5z
1
0.6
1+z
1
5/3
4+16z
1
16+4z
1
0.7
1+z
1
10/7
3+17z
1
17+3z
1
0.8
1+z
1
5/4
2+18z
1
18+2z
1
0.9
1+z
1
10/9
1+19z
1
19+z
1
Where C
N
k
=
N!
k!(Nk)!
and D = N + d.
B. Lagrange interpolation FIR Delay lter: With this ap-
proximation, a total delay D is approximated by a FIR lter
as [15-16],
z
D
=
L

n=0
h[n]z
n
(40)
The lter-order L is chosen such that
L1
2
= D =
L+1
2
. The
transfer functions obtained for FIR and IIR approximations
different values of d are tabulated in Table.1.The fractional
delay will be applied to the differentiator as follows. Multiply
the transfer function of the digital differentiator with
z
N
z
Nd
where N is the integer delay d is the fractional delay. By
doing so to an Al-Alaoui rst order digital differentiator the
phase angle variation is as shown in Fig.6. From the gure
it is evident that the phase angle can be made constant by
choosing different values of N and d. The same approach can
be applied to other differentiators with linear or approximately
linear phases.
TABLE II
RMSE FOR DIGITAL DIFFERENTIATORS
Digital differentiator RMSE
Backward 0.0024167
Bilinear 0.0025429
Alaloui 0.0024364
Simpson 0.0025993
Tick 0.0026045
Gradient 0.0025219
Ngo 0.0021188
IV. APPLICATIONS OF DIGITAL DIFFERENTIATORS
A. Edge detection using Digital Differentiators
Edge detection refers to the process of identifying and
locating sharp discontinuities in an image [13].The majority
of different methods may be grouped into two categories as
gradient method and Laplacian method. The gradient method
detects the edges by looking for the maximum and minimum
in the rst derivative of the image. The Laplacian method
searches for zero crossings in the second derivative of the
image to nd edges. The popular edge detection operators are
Roberts, Sobel, Prewitt, Frei-Chen, and Laplacian operators
etc [13,14,17]. In this paper rst and second order derivative
of the image is considered. The procedure for the edge
detection using IIR type digital differentiators is as follows.
The difference equation of third order digital differentiator
with input x[n] and output y [n] in time domain at T = 1
can be written as,
y[n] =
1
b
0
_
a
0
x[n] + a
1
x[n 1] + a
2
x[n 2] + a
3
x[n 3]
b
1
y[n 1] b
2
y[n 2] b
3
y[n 3]
_
(41)
Consider an image f(x, y) . The Gradient of the image can
be written as,
f (x, y) =
f (x, y)
x
u
x
+
f (x, y)
y
u
y
(42)
where u
x
, u
y
are the unit vectors in x and y directions. The
approximated magnitude of the Gradient is,
G = |G
x
| + |G
y
| (43)
Considering x[n] = f(x, y) and applying it to a digital
differentiator the outputs are calculated both in x and y
directions individually, and the gradient is to be calculated
using Eqn.43.The gure of merit of edge detectors Root-mean
square error E
RMS
is given by,
E
RMS
=

_
1
MN
M1

x=0
N1

y=0
[f (x, y) f
edge
(x, y)]
2
(44)
where f (x, y) is the original image of size M N and
f
edge
(x, y) is the edge detected image. The results obtained
by using the above mentioned procedure is shown in Fig.7 and
8.
IEEE-ICoAC 2012
Original Bilinear
Al-Alaoui Backward
Fig. 7. Edge detection using digital differentiators.
Gradient Tick
Simpson Ngo
Fig. 8. Edge detection using digital differentiators.
B. QRS detection
Electro Cardio Gram(ECG) is the electrical activity of the
heart is perhaps the most commonly known, recognized, and
used biomedical signal. A QRS de- tection technique designed
by Dobbs et al.which uses cross correlation was followed in
this paper[19,22,23]. The block diagram used for the detection
of QRS signal is shown in the Fig.9. The results obtained
using the procedure depicted in Fig.9 is shown in Figs.10-11.
Although Al-Alaoui differentiator is considered here, the same
can be extended to other differentiators also.
V. HARDWARE IMPLEMENTATION OF IIR TYPE DIGITAL
DIFFERENTIATORS
In general, Direct form-I, Directform-II, Cascaded form,
Parallel form are used for the implementation of any digital
Fig. 9. Block diagram of the Program ow
Fig. 10. QRS Template
lter. Because of the less hardware requirement, Directform-
II is preferred compared to other implementations. A RTL
schematic is shown in Fig.12. The above algorithm is im-
plemented in Verilog.The Corre- sponding input and output
signals are shown in Fig.13.
VI. RESULTS AND CONCLUSIONS
In this paper, theory of operation, applications and imple-
mentation of digital differentiators is presented. A study on
the design of digital differentiators reveals that type III FIR
differentiators have the inherent nature in frequency response
of approaching zero at Nyquist frequency. To extend the per-
formance of type III FIR lters in the higher frequency bands,
one has to increase the lter taps. Type IV FIR differentiators
IEEE-ICoAC 2012
Fig. 11. Detected QRS sequences using Al-Alaoui rst order digital differ-
entiator
Fig. 12. RTL Schematic for Simpson digital differentiators
Fig. 13. Simulation Result of Simpson digital differentiators
using Fourier series have been found to have out- standing fre-
quency response, however, they are noisy and biased. Higher
order central difference approxima- tions using Taylor series
might outperform windowed Fourier series since there is no
truncation and the associated Gibbs phenomenon. IIR digital
differentia- tors are more favourable for real-time application.
The rst-order IIR differentiator from Al-Alaoui is ideal in
terms of the frequency response. The IIR differentiators can be
adaptively used when systems experience high dynamics. The
digital differentiators of IIR type have been proved to be much
more efcient in detecting edges of an image, QRS detection
etc.
REFERENCES
[1] S.K.Mitra, Digital signal processing a computer based approach,Tata
McGraw Hill edition,Newdelhi, 2001.
[2] I.R.Khan,R.Ohba, Digital differentiators based on taylor series-IEICE
trans. fundamentals, Vol.E82-A,No.12,PP.2822-2824,Dec.1999.
[3] Lawrence R.Rabiner,K.Steiglitz, The design of wide-band Recursive and
non recursive digital differentiators- IEEE transactions on audio and
electroaccoustics, vol.AU-18.no.2 June 1970.
[4] Lawrence R.Rabiner, Ronald W.Schafer,Recursive and non recur-
sive realizations of digital lters designed by frequency sampling
techniques- IEEE transactions on audio and electroaccoustics,vol.AU-
19.no.3 September 1971.
[5] B.Kumar,S.C.Dutta Roy, Design of digital differentiators for low fre-
quenciesProc.IEEE, vol.76, pp.287-289, 1988.
[6] C.K.Chen, J.H.Lee, Design of high-order digital differentiators using
L
1
error criterion- IEEE transactions on circuits and systems, vol.-
42.no.4,April 1995.
[7] S.C.Pei, J.J.Shyu, Analytic closed-form matrix for designing higher
order digital differentiators using Eigen approach-IEEE transactions on
signal processing,vol.-44.no.3, March 1996.
[8] M.A.Al-Alaoui, Novel Digital integrator and Differentiator-IEEE Elec-
tronic Letters, Vol.29, no.4, pp.376-378,Feb.1993.
[9] M.A.Al-Alaoui, Novel approach to designing Digital Differentiators-
IEEE Electronic Letters, Vol.28, no.15,pp.1376-1378, Jul.1992.
[10] M.A.Al-Alaoui,Novel IIR Digital Differentiator From Simpson Inte-
gration Rule- IEEEtransactions on Circuits Systems.I ,Fundam Theory
Appl., Vol.41,no.2, pp.186-187, Feb.1994.
[11] J.L.Bihan,Novel class of Digital integrators and Differentiators-IEEE
Electronic Letters,Vol.29, no.11, May.1993.
[12] Nam Quoc Ngo,A New Approach for the Design of Wideband Dig-
ital Integrator and Differentiator- IEEE Transactions on Circuits and
Systems-II,Vol.53, No.9, September 2006.
[13] Gonzalez, Woods, Digital Image Processing,Prentice Hall, 2008.
[14] L.S.Davis, Edge detection techniques,Computer Graphics Image
Process.,Vol.4,pp.248-270, 1995.
[15] T.I.Laakso,V.Valimaki,M.Karjalainen,U.K.Laine, Splittingthe Unit De-
lay, IEEE Signal Processing Magazine,vol.13,no.1,pp.30-60,Jan.1996.
[16] V.Valimaki, PhD Thesis,Discrete-Time Modeling of Acoustic Tubes
Using Fractional Delay Filters, Helsinki University of Technology,
Dec.1995.
[17] J.A.Canny, Computational Approach to Edge Detector, IEEE Transac-
tions on PAMI, pp679- 698, 1986.
[18] B.T.Krishna,Design of digital differentiators and integrators of order 1/2-
World Journal of Modelling and Simulation,Vol.4, no.3, PP.182187,
2008.
[19] B.T.Krishna, Studies on fractional order differen- tiators and integrators:
A survey, Elsevier signal processing, Vol.91,No.3, pp.386-426,2011.
[20] L.S.Davis, Edge detection techniques, Com- puter Graphics Image
Process., Vol.4,pp.248-270, 1995.
[21] J.A.Canny, Computational Approach to Edge De- tector, IEEE Transac-
tions on PAMI, pp.679- 698, 1986.
[22] W. J.Tompkins,Biomedical Digital Signal Pro- cessing C-Language
Examples And Laboratory Experiments For The Ibm Pc, Prentice Hall,
Up- per Saddle River, New Jersey-2000.
[23] Dobbs S.E, Schmitt N.M, Ozemek H.S,QRS de- tection by template
matching using real-time cor- relation on a microcomputer, Journal of
Clinical Engineering ,1984,Vol.9,pp.197 - 212.
IEEE-ICoAC 2012