ICCAS2005 June 2-5, KINTEX, Gyeonggi-Do, Korea

Design of IIR Notch Filter for Removal of Baseline wander and

Power Line Interference from ECG Signal

Sorawat Chivapreecha, Kobchai Dejhan and Surapan Yimman*

Faculty of Engineering and Research Center for Communication and Information Technology
King Mongkut’s Institute of Technology Ladkrabang, Bangkok 10520, Thailand
Tel: +66-2326-4238, +66-2326-4242, Fax: +66-2326-4554
E-mail: {sorawat, kobchai}@telecom.kmitl.ac.th
*Department of Industrial Physics and Medical Instrumentation, Faculty of Applied Science
King Mongkut’s Institute of Technology North Bangkok, Bangkok 10800, Thailand

Abstract: This paper proposes a design of IIR notch filter with modified pole placements. The pole angle is derived from quadratic
programming to find the appropriate pole position on the unit circle in z-plane in order to achieve the symmetry of the amplitude
response. The simulation results are shown when compared with the conventional design technique. Moreover, it uses the
TMS320C31 DSP chip for hardware implementation. Finally, the hardware implementation can be applied to removal of ECG
baseline wander and elimination of AC power line interference in ECG signal.

Keywords: IIR Notch Filter, Quadratic programming, Optimal Pole placement, ECG

1. INTRODUCTION 3. OPTIMIZATION DESIGN METHOD USING

Normally, IIR notch filter can be designed by pole-zero QUADRATIC PROGRAMMING
placement method which in the conventional design technique In order to get symmetrical and can determine the pass-band
[1] which have many problems about asymmetric amplitude gain, the quadratic programming [2-3] which uses the iterative
response in pass-band. Also, this paper improves the pole scheme for solving the solution of quadratic equation using to
position from conventional design technique by using an search the optimal pole position. Improve the transfer function
iterative scheme. The optimal pole placements are found by in eq. (2) as follows,
solving the quadratic programming problem [2-3].
The results from proposed design technique will be
1 − 2 cos(ω 0 ) z −1 + z −2 B( z )
implemented on the TMS320C31 DSK board to shows the H ( z) = k =k (3)
removal of ECG (electrocardiogram) baseline wander and 1 − 2r cos(ω p ) z + r z
−1 2 −2
A( z )
elimination of AC power line interference.

2. CONVENTIONAL DESIGN METHOD FOR

Let a = cos(ω p ) , where ωp is pole angle. The
IIR NOTCH FILTER assumption in order to define the cost function is shown in Fig. 1.
The amplitude response specification of ideal IIR notch filter
is given by
1− H (e jω )
⎧ 0, ω0 ε
H (e jω ) = ⎨ (1)
⎩ 1, otherwise ε
1
And the transfer function is shown as follows

1 − 2 cos(ω 0 ) z −1 + z −2
H ( z) = (2)
1 − 2r cos(ω 0 ) z −1 + r 2 z − 2

ω
The zeros are constrained to locate on the unit circle at notch
frequency ω0 and poles are at the same radial line as zeros. ω0 π
For the stability, pole radius r must be smaller than one. When
r approaches unity, amplitude response becomes an ideal ω0 − ε ω 0 + ε
notch filter. Also, the radius r can be controlled the bandwidth Fig. 1 Assumption used for define cost function
of notch filter.
Follow by this design method, it gets the notch frequency
From Fig. 1 and eq. (3), assume k = 1. Also, the cost
which is exactly at ω 0 but it has the limitations on asymmetric function can be defined as
pass-band gain and cannot determine the pass-band gain
2
according to the desired specifications [4]. This problem causes
B (e jω )
2
A(e jω ) − B (e jω )
from the inappropriate pole position. Also, the quadratic J (a) = ∫ 1 − d ω = ∫R dω
programming can be used to search the new optimal pole R A(e jω ) A(e jω )
2

position to solve the asymmetric and undetermined pass-band

p (ω ) + q (ω ) a
2
gain.
=∫ 2
dω (4)
A(e jω )
R
 2 − 2 cos(ω 0 )
where the integral region R = [0, ω 0 − ε ] ∪ [ω 0 + ε , π ] and H ( e j 0 ) = H ( e jπ ) = 1 = k
1 − 2r cos(ω p ) + r 2
p (ω ) , q (ω ) are given by
1 − 2r cos(ω p ) + r 2
∴ k= (9)
p(ω ) = (r 2 − 1)e − j 2ω + 2 cos(ω 0 )e − jω and q(ω ) = −2re − jω 2 − 2 cos(ω 0 )
(5)
4. DESIGN EXAMPLES AND SIMULATION
The optimization problem in eq. (4) can be solved by the RESULTS
iterative scheme as follows, The design examples have specified the notch frequency at
ω 0 = 0.3π , pole radius r = 0.6, 0.7, 0.8 and 0.9 and
p (ω ) + q (ω ) a k
2

J k (a k ) = ∫ 2
dω = ρ k −1 a + 2 β k −1 a k + α k −1
2
k
normalized the pass-band gain is equal to 1. In case of the
Ak −1 (e jω )
R
conventional design method using eq. (2) and the optimization
design method using eq. (3-9) where given ε = 0.0001π
(6) and terminate the iterative procedure if J k − J k −1 ≤ 0.00001 .
where Fig. 2 shows the convergence of algorithm that uses the initial
q(ω ) values a 0 = cos(ω 0 ) .
2

ρ k −1 = ∫ 2
dω
Ak −1 (e jω )
R

Re al[ p (ω )q ∗ (ω )] 2

β k −1 = ∫ 2
dω 1.5

Ak −1 (e jω )
R r = 0.6

1
r = 0.7

p(ω )
2
0.5

α k −1 = ∫ dω
Values of Jkmin(ak)

2
Ak −1 (e jω )
0
R
r = 0.8
-0.5

-1

Therefore, the parameter a k can be determined by solving -1.5

the quadratic programming problem as follows, -2

Minimize J k ( a k ) = ρ k −1 a k2 + 2 β k −1 a k + α k −1
-2.5
(7) r = 0.9
-3

− 1 ≤ ak ≤ 1
0 2 4 6 8 10 12 14 16 18
Subject to No. of Iterations (k)

The cost function of this optimization problem [5] is a (a) Values of minimized cost function.
parabola curve. Also, the unique closed form optimal solution
can be obtained as follows,

− β k −1
>1
0.72
⎧ if
⎪ 1
ρ k −1
− β k −1 (8) 0.7
⎪
ak = ⎨ − 1 if < −1
⎪ − β k −1 ρ k −1 0.68
r = 0.6

⎪ ρ − β k −1
⎩ k −1 if − 1 ≤ ≤1 0.66

ρ k −1
Values of ak

0.64
r = 0.7

After get ak it means the optimal pole position can be 0.62

r = 0.8
obtained. The amplitude response of notch filter becomes 0.6
r = 0.9
symmetrical. However, the pass-band gain cannot been
0.58
controlled according to the desired specification. Therefore, it
must be computed the scaling constant k which is used for
0 2 4 6 8 10 12 14 16 18
controlling the pass-band gain to meet the specifications. No. of Iterations (k)
Since, the amplitude response is symmetry already. Also,
the magnitude of transfer function at ω = 0 is equal to at
(b) Values of cos(ω p ) .
ω = π . Consider in eq. (3) with normalized maximum pass-
band gain, it can be get
Fig. 2 Convergence curve of algorithm
 TABLE I
RESULTS FROM CONVENTIONAL DESIGN AND OPTIMIZATION DESIGN METHOD FOR IIR NOTCH FILTER ω 0 = 0.3π

Zero Position Pole Position Pole Angle (rad) k

Conventional Design 0.58778 ± j0.80901 0.35267 ± j0.48541 ± 0.94247 1
r = 0.6 Optimal Design 0.58778 ± j0.80901 0.39969 ± j0.44748 ± 0.84175 0.68000
Conventional Design 0.58778 ± j0.80901 0.41144 ± j0.56631 ± 0.94247 1
r = 0.7 Optimal Design 0.58778 ± j0.80901 0.43790 ± j0.54611 ± 0.89493 0.74500
Conventional Design 0.58778 ± j0.80901 0.47022 ± j0.64721 ± 0.94247 1
r = 0.8 Optimal Design 0.58778 ± j0.80901 0.48198 ± j0.63850 ± 0.92419 0.82000
Conventional Design 0.58778 ± j0.80901 0.52900 ± j0.72811 ± 0.94247 1
r =0.9 Optimal Design 0.58778 ± j0.80901 0.53194 ± j0.72597 ± 0.93843 0.90500

The computed values of pole-zero position, pole angle and 5. APPLICATION TO REMOVAL OF ECG
scaling constant that is used for pass-band gain controlling is BASELINE WANDER AND POWER LINE
shown in Table I and Fig. 3 shows the amplitude response that INTERFERENCE
obtained from conventional design method when compared In particular, medical signal processing of ECG signals,
with the optimization design method using quadratic there is a need to eliminate the baseline wander that caused
programming. from very low frequency components and the AC power line
frequency [6-7]. Also, in this experiment will be divided to 2
1.6 experiments. The first, for removal of ECG baseline wander
r = 0.7
r = 0.6
which assumed very low frequency is 1 Hz. Therefore, the first
1.4
IIR notch filter has notch frequency at 1 Hz, sampling
1.2 frequency 800 Hz and 3dB bandwidth 1 Hz which correspond
to pole radius r = 0.996 with optimal pole angle ± 0.007124
Normalized Amplitude

1
r = 0.9 r = 0.8 radian. Follow by proposed method can be got the transfer
0.8 function as follows,
0.6

0.996078 − 1.99209 Z −1 + 0.996078Z −2

H (Z ) =
1 − 1.99209 Z −1 + 0.992156 Z − 2
0.4

0.2

0
and the amplitude response can be shown in Fig. 4.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Normalized Frequency
1.1

(a) Conventional design method 1

Normalized Amplitude

1.2 0.9

1
0.8

r = 0.6
Normalized Amplitude

0.8
r = 0.7 0.7

r = 0.8
0.6
r = 0.9 0.6

0.4

0.5
0 50 100 150 200 250 300 350 400
0.2 Frequency (Hz)

0
Fig. 4 Amplitude response of removal of ECG baseline wander
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Normalized Frequency filter

The ECG signals obtained from the ECG simulator of

(b) Optimization design method Lionheart Technologies, Inc. by setting the heart beat is 90
BPM (beat per minute). The result of removal of ECG baseline
Fig 3 Amplitude response at notch frequency at ω0 = 0.3π wander can be shown in Fig. 5. Signal in upper trace shows
ECG with baseline drift and signal in lower trace shows filtered
for various pole radius signal.
 Fig. 5 Removal of ECG baseline wander Fig. 7 Elimination of AC power line interference

In fact, the recommendations of the American Heart 6. CONCLUSIONS

Association state that the frequency components above 0.05 Hz The optimal IIR notch filter design using quadratic
should not be removed. Also, the very low frequency should be programming for modified pole placement result in symmetric
closed to DC. The filter coefficients for notch frequency close and can determine the pass-band gain according to the desired
to DC have high sensitivity which cannot implement using specifications. The obtained results optimal design are better
Direct Form II structure but the filter realization used for than when compared with the conventional design technique
implementation is Direct Form II structure. Also, this but more complicated design procedures. Furthermore, it can
experiment will be assumed very low frequency is 1 Hz. This be applied to removal of ECG baseline wander and eliminate
problems can be cleared using low sensitivity structure or the AC power line noise frequency from ECG signal.
reduce the sampling rate.
The second experiment, to eliminate of AC power line
interference which assumed power line frequency is 50 Hz.
REFERENCES
[1] J. G. Proakis and D. G. Manolakis, “Digital Signal
Therefore, the second IIR notch filter has notch frequency at 50
Processing Principle, Algorithm and Applications,”
Hz, sampling frequency 800 Hz and 3dB bandwidth 5 Hz
Prentice-Hall, 1996.
which correspond to pole radius r = 0.98 with optimal pole
[2] R. A. Roberts and C. T. Mullis, “Digital Signal
angle ± 0.39223 radian. Follow by proposed method can be got
Processing,” Addison-Wesley, 1987.
the transfer function as follows,
[3] C. C. Tseng and S. C. Pei, “Stable IIR Notch Filter Design
with Optimal Pole Placement,” IEEE Trans. Signal
0.980755 − 1.8122 Z −1 + 0.980755 Z −2
H (Z ) = Processing., Vol. 49, No. 11, pp. 2673-2681, November
1 − 1.8122 Z −1 + 0.96151Z − 2 2001.
[4] S. Yimman, W. Hinjit, S. Sriboonsong, M. Puangpool and
and the amplitude response can be shown in Fig. 6. K. Dejhan, “IIR Notch Filter Design with Modified Pole-
Zero Placement Algorithm,” Proc. IEEE Int. Symp. Signal
Processing and InformationTechnology (ISSPIT 2003),
1 Darmstadt, Germany, December 2003.
[5] L. V. Fausett, “Applied Numerical Analysis Using
0.8
Matlab,” Prectice-Hall, 1999.
[6] J. A. Van Alste and T. S. Schilder, “Removal of Baseline
Normalized Amplitude

Wander and Power Line Interference from ECG by an

0.6
Efficient FIR Filter with a Reduced Number of Taps,”
IEEE Trans. Biomedical Engineering, Vol. 32, pp. 1052-
0.4 1060, December 1985.
[7] J. P. Marques de Sa, “Digital FIR Filtering for Removal of
0.2 ECG Baseline Wander,” J. Clinical Engineering, Vol. 8,
pp. 235-240, 1982.
0
0 50 100 150 200 250 300 350 400
Frequency (Hz)

Fig. 6 Amplitude response of elimination of AC power line

interference filter

The result of elimination of AC power line interference can

be shown in Fig. 7. Signal in upper trace shows ECG with 50
Hz AC power line frequency and signal in lower trace shows
filtered signal.