# Introduction

In order to process signals, we have to design and implement systems called filters. A filter is a device that transmits (or rejects) a specific range of frequencies. There are four basic filter types; low-pass, high-pass, band-pass, and band-stop.

Three Commonly Used Filters
MATLAB has a variety of functions in its signal processing toolbox which support the design and analysis of analog and digital filters. We shall begin with analog filters and take a look at the most commonly used approximations, namely,
□ Butterworth filters

□ Chebyshev filters □ Elliptic filters

Determination of the Minimum Order
These filters are typically specified by their cutoff frequencies and ripples on both the stopband and passband. The design process starts with the determination of the minimum order of the filter and the appropriate cutoff frequency to meet the desired specifications. These two parameters can be estimated via the following commands: >>[N,Wc]=buttord(Wp,Ws,Rp,Rs,’s’); >>[N,Wc]=cheb1ord(Wp,Ws,Rp,Rs,’s’); >>[N,Wc]=cheb2ord(Wp,Ws,Rp,Rs,’s’); >>[N,Wc]=ellipord(Wp,Ws,Rp,Rs,’s’); % Butterworth filter % Chebyshev Type-I filter % Chebyshev Type-II filter % Elliptic filter

N = lowest order of the filter Wc = cutoff frequency in rad/s Rp = passband ripple in dB Rs = minimum stopband attenuation in dB Wp, Ws = passband and stopband edge frequencies, respectively.

208

Practice - Analog Filter DesignUsing MATLAB, determine the order (N) and cutoff frequency (Wc) of a lowpass Butterworth filter with the following specifications: Wp =10 rad/s Ws = 20 rad/s Rp = -2 dB Rs = -20 dB % passband critical frequency % stopband critical frequency % passband ripple % stopband attenuation

>>Wp = 10; Ws = 20; Rp = -2; Rs = -20; >>[N,Wc] = buttord(Wp,Ws, -Rp, -Rs, ‘s’) N= 4 Wc = 11.2610

What Are Normalized Lowpass Analog Filters?
Analog filters are typically designed as normalized (cutoff frequency of 1 rad/s) lowpass filters and then transformed to the specific frequency and filter type (highpass, bandpass, etc.) with direct substituion.

Commands
- “buttap”, “cheblap”, “cheb2ap” & “ellipap” The functions buttap, cheb1ap, cheb2ap, ellipap, can be used to design normalized analog lowpass Butterworth, Chebyshev Type-I, Chebyshev Type-II, and elliptic filters, respectively.

209

Syntax >>[z,p,k]=buttap(n); %normalized Butterworth filter >>[z,p,k]=cheb1ap(N,Rp); %normalized Chebyshev Type-I filter >>[z,p,k]=cheb2ap(N,Rs); %normalized Chebyshev Type-II filter >>[z,p,k]=ellipap(N,Rp,Rs); %normalized elliptic filter Rp = ripple in the passband (in dB) Rs = ripple in the stopband (in dB) N = order of the filter z =vector containing the zeros P = vector containing the poles k = gain factor

Practice - Normalized Lowpass Analog Filters : Butterworth FilterUsing MATLAB, find the poles, zeros, and the gain factor of a normalized 5th-order Butterworth filter. >>[z, p, k] = buttap(5) z= [] p= -0.3090 + 0.9511i -0.3090 - 0.9511i -0.8090 + 0.5878i -0.8090 - 0.5878i -1.0000 k= 1

210

1049 .9580i k= 0.1634 Transfer Function .3968i -0. Syntax >>[num. where num contains the numerator coefficients and den the denominator coefficients.9580i -0.1049 + 0.k).p.k]=cheb1ap(4. find poles. den] = zp2tf(z.0.p.2532 + 0.0. >>[z.2) z= [] p= -0.Commands: “zp2tf” It is often the case to specify the designed analog filter in terms of its transfer function. zeros. Let us give it a try! 211 .3968i -0. poles.Normalized Lowpass Analog Filters : Chebyshev Type-I Filter Using MATLAB. The function zp2tf converts the zeros. and gain characterization of the filter to a rational function form (transfer function).2532 .Practice . and the gain factor of a normalized 4th-order normalized Chebyshev Type-I filter with Rp = 2 dB.

2058 % poles. % compute magnitude response >>phase=angle(H)*180/pi.grid % plot magnitude response >>%subplot(2.Practice .mag). % frequency vector >>mag=abs(H).phase).1).p.plot(w.den) % print the transfer function H(s) >>w=0:0. and gain specs of filter >>[num.005:3. zeros.Normalized Lowpass Analog Filters : Transfer Function - Using MATLAB.2). >>[z.[rad/s]’) % label the horizontal axis >>ylabel(‘|H(\omega)|’) % label the vertical axis >>title(‘Chebyshev Type-I normalized LPF’) (The graph is shown on the next page) 212 .k]=cheb1ap(4.plot(w.k).1634 -----------------------------------------------s^4 + 0. >>[z. >>[num.den) Transfer function 0. and gain specs of Chebyshev filter % Transfer function model % print the transfer function Practice .1.3). determine the transfer function of a 4th-order normalized Chebyshev Type-I lowpass filter with Rp=2 dB. Sketch its magnitude and phase responses.5168 s + 0.p.p.den]=zp2tf(z.p.k) >>tf(num.k] = cheb1ap(4. disp(num’) % display numerator coefficients >>disp(‘den=’).den]=zp2tf(z.Frequency Response of Analog Filters Design a 4th-order Chebyshev Type-I normalized lowpass filter with 3-dB ripple in the passband.disp(den’) % display denominator coefficients >>tf(num. % pole.grid % plot phase response >>xlabel(‘\omega. zero.7162 s^3 + 1.2).1. % compute phase response >>subplot(2. % numerator and denominator of H(s) >>disp(‘num=’).256 s^2 + 0.

Wc).den.“lp2lp”.Introduction The previous filters are designed for the normalized cutoff frequency (1 rad/s). >>[num. lp2bp”.den]=lp2lp(num. highpass.den]=lp2hp(num.den]=lp2bp(num.den]=lp2bs(num.den. >>[num. % lowpass-to-lowpass % lowpass-to-highpass % lowpass-to-bandpass % lowpass-to-bandstop Wc = cutoff frequency of the filter W0 = center frequency of the filter (bandpass and bandstop) Bw = bandwidth of the filter (bandpass and bandstop) 213 .Wc).den.den. >>[num. and lowpass-to-bandstop transformations. provide lowpass-to-lowpass. respectively. and lp2bs.W0. and bandstop filters. bandpass. Commands .W0. lp2bp.Bw).Bw). lp2hp”. lp2hp. & “lp2bs” The commands lp2lp. and obtain lowpass. We shall now consider a number of transformations that will allow us to alter the cutoff frequency. lowpass-to-bandpass. lowpass-tohighpass. Syntax: >>[num.

% compute magnitude response >>plot(w/2/pi. mag) % plot magnitude response versus frequency H(z) >>xlabel(‘frequency. % fourth-order Butterworth filter >>[num.k]=buttap(4).den) % print the transfer function H(s) >>wc=2*pi*5. >>[z. % define a frequency vector >>H=freqs(num. % specify cutoff frequency in rad/s >>[num.Practice .den]= lp2lp(num.Frequency Transformation (1) Design a fourth-order Butterworth lowpass filter with a cutoff frequency of 5 Hz. % convert to polynomial form >>tf(num. □ Apply the frequency transformation lp2lp to achieve the desired specifications. The procedure is as follows: □ Design a fourth-order normalized Butterworth filter using buttap.wc).k).den]=zp2tf(z.w).den.den.p. % compute frequency response >>mag=abs(H).p. [Hz]’) % label the horizontal axis >>ylabel(‘|H(f)|’) % label the vertical axis >>title(‘Fourth-order Butterworth LPF: fc= 5 Hz’) 214 . % desired filter >>w=0:pi/100:10*pi.

Practice - Frequency Transformation (2)

Design a fourth-order Butterworth bandpass filter with a center frequency W0= 24 rad/s and bandwidth Bw= 10 rad/s. >>W0= 24; Bw=10; >>[z,p,k]=buttap(4); >>[num,den]=zp2tf(z,p,k); >>[num,den]=lp2bp(num,den,W0,Bw); >>tf(num,den) >>w=0:pi/1000;20*pi; >>H=freqs(num,den,w); >>mag=abs(H); % specify cutoff frequency and bandwidth % specify the Butterworth filter % specify filter in polynomial form % convert LPF to BPF % print the transfer function of desired filter % specify frequency vector % compute the frequency response % compute the magnitude response

>>plot(w/2/pi, mag); % plot the magnitude response >>xlabel(‘frequency, [Hz]’) % label horizontal axis >>ylabel(‘|H(f)|’) % label vertical axis >>title(‘Fourth-order bandpass Butterworth filter’)

215

More Commands
The signal processing toolbox offers more straightforward command for the design of analog filters. A. Butterworth Analog Filters >>[num,den]=butter(N,Wc, ‘s’); >>[num,den]=butter(N,Wc,’s’); >>[num,den]=butter(N,Wc,’high’,’s’); >>[num,den]=butter(N,Wc,’stop’,’s’); >>[z,p,k]=butter(N,Wc,’s’); >>[z,p,k]=butter(N,Wc,’s’); >>[z,p,k]=butter(N,Wc,’high’,’s’); >>[z,p,k]=butter(N,Wc,’stop’,’s’); % Lowpass Butterworth filter % Bandpass Butterworth filter (order=2N) % Highpass Butterworth filter % Bandstop Butterworth filter % Lowpass Butterworth filter % Bandpass Butterworth filter % Highpass Butterworth filter % Bandstop Butterworth filter

N=order of the filter Wc=cutoff frequency. For bandstop and bandpass Wc=[Wc1,Wc2] z= zeros p= poles k= gain Practice - Frequency Transformation (3) Design an 8th-order Butterworth bandpass filter with Wc= [ 2π ,5π ] . Sketch the magnitude and phase responses. >>Wc=[2*pi,5*pi]; >>[num,den]=butter(w,Wc,’s’); >>[H,w]=freqs(num,den); >>mag=abs(H); >>phase=angle(H)*180/pi; >>plot(w,mag);grid % specify critical frequencies % specify the filter in polynomial form % computer the frequency response % extract the magnitude response % extract the phase response in degrees % plot magnitude response

216

B. Chebyshev Analog Filters The following functions can be used for the design of Type-I and Type-II Chebyshev filters: B.1 Type-I >>[num,den]=cheby1(N, Rp,Wc,’s’); >>[num,den]=cheby1(N,Rp,Wc,’high’,’s’); >>[num,den]=cheby1(N,Rp,Wc,’stop’,’s’); >>[num,den]=cheby1(N,Rp,Wc,’s’); >>[z,p,k]=cheby1(N,Rp,Wc,’s’); >>[z,p,k]=cheby1(N,Rp,Wc,’high’,’s’); >>[z,p,k]=cheby1(N,Rp,Wc,’stop’,’s’); >>[z,p,k]=cheby1(N,Rp,Wc,’s’); B.2 Type-II >>[num,den]=cheby2(N,Rs,Wc,’s’); >>[num,den]=cheby2(N,Rs,Wc,’high’,’s’); >>[num,den]=cheby2(N,Rs,Wc,’stop’,’s’); >>[num,den]=cheby2(N,Rs,Wc,’s’); >>[z,p,k]=cheby2(N,Rs,Wc,’s’); >>[z,p,k]=cheby2(N,Rs,Wc,’high’,’s’); >>[z,p,k]=cheby2(N,Rs,Wc,’stop’,’s’); >>[z,p,k]=cheby2(N,Rs,Wc,’s’); % LPF: Rs=decibels down % HPF % BSF: Wc=[Wc1,Wc2] % BPF: Wc=[Wc1,Wc2] % LPF % HPF % BSF % BPF % LPF: Rp=ripples (dB) in passband % HPF % BSF: Wc=[Wc1,Wc2] % BPF: Wc=[Wc1,Wc2] % LPF % HPF % BSF % BPF

Two Kinds of Digital Filters
-FIR & IIR Two types of digital filters can be identified, usually classified according to the duration of their impulse response, which can be either of finite length or infinite length. Filters having a finite duration impulse response are called Finite Impulse Response Filters or FIR filters; and filters with an infinite duration impulse response are called Infinite Impulse Response Filters or IIR filters.

217

Ws.Wc]=cheb2ord(Wp. what is its digital frequency? ∴ Fd = 8 = 0.Ws.25 cycles/sample ⇔ Ωd = 2π ( 0. >>[N. cheby1.Rs).571 rad/sample 32 Infinite Impulse Response Filters The first step is to determine the minimum order and cutoff frequency of the filter required to meet the desired specifications.Rs). The following commands give estimates of the filter’s order and cutoff frequency: >>[N. we usually consider as ranging over the interval [0.Rs).0 corresponding to half the sampling rate. Ωd = ωa fs = ω a ts .Rp. Plots of frequency response obtained with MATLAB use the term “normalized frequency” for a frequency which is normalized with respect to half the sampling frequency.Digital Filter Design If a 8 Hz sinusoidal signal is sampled at 32 Hz.Rs).25) = 1. Just as analog frequency can be expressed in rad/s as well as Hz. or ellip. >>[N. % Butterworth filter % Chebyshev Type-I % Chebyshev Type-II % Elliptic 218 .Rp. with 1.Ws.Ws. Rp. cheby2. The second step is to pass this order and cutoff frequency to the design functions butter. 2π ] is specified in MATLAB digital filter function as [0.Rp. Fd = 1 Td Td : samples/radian Practice . The digital frequency axis which.Wc]=ellipord(Wp. >>[N.Wc]=buttord(Wp. digital frequency can be expressed in radian/sample as well as in cycles per sample.NOTE Normalized Frequency MATLAB uses a somewhat non-standard form for specifying frequencies for digital filters.2].Wc]=cheb1ord(Wp.

Bilinear transformation method 2.Digital Filter Design: Infinite Impulse Response Filter Find the order and cut off frequency of a Butterworth dignital lowpass filter with no more than 1 dB attenuation below 3400Hz and at least 40 dB attenuation at and beyond 3600 Hz. Practice . Bilinear Transformation Method A. Chebyshev Type II. The bilinear transformation method is a standard procedure of mapping the s-plane into the z-plane. The sampling frequency is 8000Hz. >>[N. cheb1ap. into a discrete equivalent. such that 0<Wp. respectively.1 Basic Commands In this section we start with the design of IIR filters based on the Bilinear Transformation Method (BTM). Then. namely. Chebyshev Type1. designed via conventional techniques. >>Rp=1. Impulse invariance method 3. 1. >>Fs=8000.Ws=passband and stopband edges frequencies. bandpass.Rp. The resulting transfer function is mapped into a digital filter transfer function via the bilinear command. Rs=40.Rs) % specify sampling rate % specify Nyquist frequency % normalized passband frequency % normalized stopband frequency % passband and stopband attenuation We shall now consider two techniques for designing IIR filters. >>Fn=Fs/2. It transforms an analog filter. and elliptic analog filters. The process begins with the design of a normalized lowpass analog filter.wc]=buttord(wp. 219 .Ws<1 where 1 corresponds to pi radians.N= lowest order of the digital IIR filter Wc=cutoff frequency Rp=passband ripple in dB Rs=minimum stopband attenuation in dB Wp. The command buttap. and ellipap provide Butterworth. frequency transformations are applied to the normalized filter to yield lowpass. >>ws=3600/Fn. >>wp=3400/Fn.ws. cheb2ap. The yule walker design method A. highpass or bandstop filter with the desired design frequency requirements.

2).p.1.den]=lp2lp(num. % plot the magnitude response >>title(‘Lowpass Butterworth digital filter’).1).2 –200 200]) >>ylabel(‘Phase response’) >>xlabel(‘Frequency.2 More Commands The signal processing toolbox includes butter.plot(wd/pi.p. % normalized 4-th order Butterworth filter >>[num.Fp).Infinite Impulse Response Filters: Bilinear Transformation Method (1) Design a 4-th order Butterworth lowpass digital filter with a cutoff frequency of 3 rad/s and sampling rate of 10 Hz.2 0 1]) >>subplot(2.dens.denz]=bilinear(nums.Fs). >>ylabel(‘Amplitude response’) >>axis([0 0. % scale the normalized filter by wc >>Fs=10.den]=zp2tf(z.den]=butter(n. % determine the frequency response >>magd=abs(Hd).dend). disp(dend’) % print denominator coefficients in column vector >>[Hd. denz: numerator and denominator of the transfer function H(z) nums.k).den.dend]=bilinear(num.grid >>axis([0 0. % analog to digital mapping >>disp(‘numd=’). [in units of pi]’) A.wc).k]=buttap(4).Syntax >>[numz. % LPF: 0<wn<1 220 . [num.grid. cheby2.wd]=freqz(numd.denz]=bilinear(nums. disp(numd’) % print numerator coefficients in column vector >>disp(‘dend=’).Fs. % polynomial specification of the filter >>wc=3.Fs). >>[numz. cheby1.1.angle(Hd)). >>[z. % specify the cutoff frequency >>[num. % Fs=sampling frequency % Fp=specifies prewarping in Hz numz.dens: numerator and denominator of the transfer function H(s) Practice . % sampling frequency >>[numd.wn).den. and ellip functions that automate the design of IIR filters via the bilinear transformation method.dens.plot(wd/pi. % compute magnitude response of digital filter >>subplot(2.mag).

wn).den]=cheby1(n.wn).den]=ellip(n.wn).Rs. 0<w1. [num. z: vector of zeros p: vector of poles k: gain of transfer function Practice .den]=ellip(n.’stop’).den]=cheby1(n.w2].den]=cheby1(n.wn.Rp. [num. >>[z.Rs.w2<1 % BPF: wn=[w1.w2<1 % BSF: wn=[w1.w2]. [num.w2].Rp. 300/500.[num.’stop’).Rp.wn).Rs.w2<1 % BPF: wn=[w1.den]=cheby2(n. [b.Rs. 0<w1. [num.Rp. [num.den]=ellip(n.k]=cheby1(n.wn.k]=cheby2(n.w2<1 % LPF: 0<wn<1 % LPF: 0<wn<1 % HPF: 0<wn<1 % HPF: 0<wn<1 % BSF: wn=[w1. 0<w1.wn).wn). zeros and the gain of the transfer function.p.w2<1 The following functions can also be used to determine the poles.w2].w2].128.0<w1. 0<w1.wn.Rp.’high’).Rp.den]=butter(n. [num.w2]. Freqz(b.Rp.a.wn.wn).den]=butter(n. [num. [num.a]=butter(9.wn).’high’).Rs.Rs.wn).Rs. >>[z.’high’).p.den]=ellip(n.wn).wn. 0<w1.w2].Rs.wn.’stop’).den]=cheby2(n. >>[z.’high’).Rs.’high’).Rs.Rp. [num. 0<w1. 0<w1.den]=cheby1(n.Infinite Impulse Response Filters: Bilinear Transformation Method (2) For data sample at 1000 Hz.’stop’). [num. >>[z.Rp.den]=butter(n. design a 9-th order highpass Butterworth filter with cutoff frequency of 300 Hz.k]=butter(n.k]=ellip(n.w2<1 % BPF: wn=[w1.w2<1 % BSF: wn=[w1.p.p. Rp: Passband ripple in dB Rs: Minimum stopband attenuation in dB num: Numerator of transfer function den: denominator of transfer function % HPF: 0<wn<1 % BPF: wn=[w1.den]=cheby2(n. [num. [num.Rp.w2].wn.wn).den]=cheby2(n. [num.wn.w2<1 % LPF: 0<wn<1 % HPF: 0<wn<1 % BSF: wn=[w1.1000) % 128 equidistant points from 0 to y 221 . [num.

Infinite Impulse Response Filters: Bilinear Transformation Method (5) Design a 4-th order lowpass Butterworth filter having a cutoff frequency of 3 rad/s and Fs=10 Hz. % provide the frequency response of digital filter >>magd=abs(Hd).magd).1.a]=cheby2(9.20. stem(t. % compute the magnitude response >>phase=angle(Hd)*180/pi. [b. %plot magnitude response >>subplot(2. % specify the sampling rate >>wc=3.dend).1).Infinite Impulse Response Filters: Bilinear Transformation Method (3) Design a 10-th order bandpass filter with a passband form 100 to 200 Hz and plot its impulse response.omegad).512. [y.a]=butter(n. %plot the phase response 222 .t]=impt(b.1000). % specify the digital filter transfer function >>[Hd.a. wn=[100 200]/500.dend]=butter(N. % normalized digital cutoff frequency >>[numd.1.wn).g) Practice .Practice .101). n=5. % specify the analog cutoff frequency >>omegad=wc/(pi*Fs). [b. Practice .phase). >>N=4.grid. % compute the phase response >>subplot(2.wd]=freqz(numd. % specify order of the filter >>Fs=10.300/500).2). design a 9-th order lowpass Chebyshev type II filter with ? ? of 20 dB down from the passband and a cutoff frequency of 300 Hz. Freqz=(b.grid.a.plot(wd/pi.plot(wd/pi.Infinite Impulse Response Filters: Bilinear Transformation Method (4) For data sampled at 1000 Hz.

>>Fs=4000. >>Fc=250. grid >>axis([0 2000 –90 10]) >>xlabel(‘Frequency. 2*Fc/Fs). >>plot(f.Rs.4000). 0. >>f=4*(0:500).2).’LineWidth’. (dB)’) >>title(‘Frequency response of the elliptic filter’) 223 . >>Rp=0. >>clear all. Rs=60.Practice . >>[H.512.Rp.mag. >>mag=20*log10(abs(H)). >>[numd.f]=freqz(numd.dend.dend]=ellip(8.5 dB of ripple in the passband and a minimum stopband attenuation of 60 dB to be used with a signal that is sampled at 4 kHz. (Hz)’) >>ylabel(‘Magnitude response.Infinite Impulse Response Filters: Bilinear Transformation Method (6) Design an 8-th order elliptic lowpass filter with a cutoff frequency of 250 Hz.5.

0. grid title(‘Frequency response of Bandpass filter’) ylabel(‘Magnitude response’) xlabel(‘Frequency’) subplot(2.2).1).a. Plot the magnitude response between 100 Hz and 300 Hz.2 dB of ripple in the passband. plot(F.Fs). and the pole-zero diagram.’LineWidth’.abs(H). Fs=1000.zplane(b.Infinite Impulse Response Filters: Bilinear Transformation Method – (7) A sequence has 1 kHz sampling.512.F]=freqz(b.[180/500.220/500]) [H.a) 224 .1.a]=cheby1(4.2.Practice . with 0. subplot(2.1. [b.2). Design an 8-th order Chebyshev Type I bandpass filter that has a bandwidth of 40 Hz centered at 200 Hz.

256.a3.a1]=butter(2.s]=freqz(b1.8192). >>order=4 >>ripple=0. >>title(‘Magnitude plot for IIR filter’) >>xlabel(‘Frequency (radians)’) >>ylabel(‘Magnitude (dB)’) Practice .0.Practice . >>H2=freqz(b2. 0.a.5dB of ripple in the passband to be used with a signal that is sampled at 8kHz.Infinite Impulse Response Filters: Bilinear Transformation Method (8) Design a 4-th order HR Chebyshev type bandpass filter with cutoff frequencies at 300 Hz. >>s. >>w1=300/4000.f.ripple.5.0. >>Fn=Fs/2.8192).Infinite Impulse Response Filters: Bilinear Transformation Method (9) Plot a family of three Butterworth lowpass filters: >>[b1.s) (The plot is shown on the next page) 225 .256. >>w2=3400/4000. >>[b2. >>w=0:pi99:99.256. >>phase=unwrap(phase.wc].plot=’mag’. >>Fs=8000. >>wc=[w1 w2].tol).a2]=butter(5.w). >>H=freqz(b. >>freqzplot(H. >>[b3.1). >>[b. >>H3=freqz(b3. >>H=[H1 H2 H3].0.8192).1).f. >>[H1.a3]=butter(10.1).a1.a]=cheby1[order.a2.

1.wn]=buttord(wp/pi. %Design digital Butterworth filter [b. ws=0.4π As = 40dB Solution: We will use a Butterworth filter for this design Matlab Code: >> %Given LPF specifications wp=0.Rp.25π R p = 0.ws/pi.a]=butter(N. be specified in %units of pi radian [N.25*pi.wn). Rp=0. (More matlab codes continue on the next page) 226 . As=40.Infinite Impulse Response Filters: Bilinear Transformation Method (10) Design an IIR Lowpass filter to meet the following specifications: ω p = 0.4*pi.Practice . %Calculate Butterworth filter parameters %Matlab requires that freq.1dB ωs = 0.As).

%Modified freqz command [db.2). title('Phase Response'). title('Magnitude Response in dB'). grd=grpdelay(b.'whole'). H=(H(1:1:501))'. grid.3). axis([0 1 -1 1]).1000.w). w]=freqz_m(b. plot(w/pi. pha.pha.2.w]=freqz_m(b.a) %db=Relative magnitude in dB computed over 0 to pi radians %mag=absolute magnitude computed over 0 to pi radians %pha=Phase response in radians over 0 to pi radians %grd=group delay over 0 to pi radians %w=501 frequency samples between 0 to pi radians %b=numerator polynimial of H(z) (for FIR:b=h) %a=denominator polynomial of H(z) (for FIR: a=[1]) % [H. w=(w(1:1:501))'. axis([0 1 -100 10]).pha. pha=angle(H).1). mag. pha/pi). grid. title('Magnitude Response'). subplot(2. ylabel('Phase in \pi units'). Required Routine function[db.2. grd. xlabel('frequency in \pi units').2.a). xlabel('frequency in \pi units').mag.4). %ylabel(‘Phase in \pi units’). db=20*log10((mag+eps)/max(mag)). plot(w/pi.mag). zplane(b.db).mag.grd.a.2.w]=freqz_m(b. ylabel('Decibels'). subplot(2. mag=abs(H). %Plot subplot(2.293π (The plots are shown on the next page) 227 . grid. %End of function Matlab Output: N= 12 wn= 0.w]=freqz(b. axis([0 1 0 1]). xlabel('frequency in \pi units'). title('Pole-zero plot'). subplot(2.a.a).grd.a) %Modified version of freqz sunroutine %---------------------------------------%[db. plot(w/pi.

w).NOTE To display the frequency-domain plots of digital filters. use the following routine.grd. H=(H(1:1:501))'.a) %db=Relative magnitude in dB computed over 0 to pi radians %mag=absolute magnitude computed over 0 to pi radians %pha=Phase response in radians over 0 to pi radians %grd=group delay over 0 to pi radians %w=501 frequency samples between 0 to pi radians %b=numerator polynimial of H(z) (for FIR:b=h) %a=denominator polynomial of H(z) (for FIR: a=[1]) % [H. pha=angle(H). grd=grpdelay(b.a) %Modified version of freqz sunroutine %---------------------------------------%[db.w]=freqz_m(b.'whole'). 228 . mag. the phase response.a.w]=freqz(b.mag.a. and the group delay response.1000. pha. db=20*log10((mag+eps)/max(mag)). w]=freqz_m(b. function[db. mag=abs(H). w=(w(1:1:501))'.pha. grd. %End of function It requires the magnitude response in absolute as well as in relative dB scale.

Infinite Impulse Response Filters: Bilinear Transformation Method (11) Design an IIR Highpass filter to meet the following specifications: ω p = 0. grid. ws=0.Rp. %Calculate chebyshev I filter parameters %Matlab requires that freq. plot(w/pi. be specified in %units of pi radian [N.2.pha/pi).grd. ylabel('Decibels').2.2dB ωs = 0.w]=freqz_m(b. xlabel('frequency in \pi units'). plot(w/pi.mag). subplot(2. title('Magnitude Response').mag.wn]=cheb1ord(wp/pi.3). axis([0 1 -1 1]). %ylabel('Phase in \pi units').pha.wn.2). %Modified freqz command [db.ws/pi. Rp=0. subplot(2.1). xlabel('frequency in \pi units'). grid. title('Phase Response'). subplot(2.4π As = 55dB Solution: Let us use a Chebyshev Type I filter for this example. plot(w/pi.a). As=55.2. xlabel('reequency in \pi units').a]=cheby1(N. Matlab Code: % Given HPF spcifications wp=0.5π R p = 0.Practice .Rp.a). title('Pole-zero plot').2.As) %Design digital Chebyshev I highpass filter [b.'high'). (The required rountine and matlab output are shown on the next page) 229 .db).4). grid. title('Magnitude Response in dB'). %plot subplot(2. Zplane(b. ylabel('phase in \pi units').4*pi.2. axis([0 1 0 1]). axis([0 1 -100 10]).5*pi.

db=20*log10((mag+eps)/max(mag)).5π 230 .grd. w=(w(1:1:501))'.w]=freqz(b.a) %db=Relative magnitude in dB computed over 0 to pi radians %mag=absolute magnitude computed over 0 to pi radians %pha=Phase response in radians over 0 to pi radians %grd=group delay over 0 to pi radians %w=501 frequency samples between 0 to pi radians %b=numerator polynimial of H(z) (for FIR:b=h) %a=denominator polynomial of H(z) (for FIR: a=[1]) % [H. %End of function Matlab Output: N= 11 wn= 0. grd=grpdelay(b.w]=freqz_m(b.pha.mag. mag=abs(H).a. pha=angle(H).Required Routine function[db. mag. pha.a.1000.'whole').w). grd. H=(H(1:1:501))'.a) %Modified version of freqz sunroutine %---------------------------------------%[db. w]=freqz_m(b.

6π Solution: Let us use a Chebyshev Type II filter for this example.2). %Calculate Chebyshev II filter parameters %Matlab requires that freq. plot(w/pi. %Plot subplot(2.2*pi. grid. ws2=0.25π R p = 0.pha.4).a). ylabel('Phase in \pi units'). %Modified freqz command [db. axis([0 1 -1 1]). plot(w/pi.5*pi. xlabel('frequency in \pi units').5π As = 40dB ωs . %ylabel('Phase in \pi units') subplot(2.3).2.1 = 0. ylabel('Decibels'). ws1=0. zplane(b. wp=[wp1 wp2].25*pi.Practice .As.2.wn).2 = 0. (The required routine and the matlab output are shown on the next page) 231 . wp2=0.2.mag). subplot(2.grd. be specified in %units of pi radian [N. ws=[ws1 ws2].mag. xlabel('frequency in \pi units'). title('Phase Response'). title('Pole-zero plot').a]=cheby2(N. axis([0 1 -100 10]). grid.pha/pi). title('Magnitude Response in dB'). axis([0 1 0 1]).wn]=cheb2ord(wp/pi. grid. xlabel('frequency in \pi units'). title('Magnitude Response').w]=freqz_m(b.Infinite Impulse Response Filters: Bilinear Transformation Method (12) Design an IIR Bandpass filter to meet the following specifications: ω p .db). As=40.As) %Design digital Chebyshev II bandpass filter [b.2. Rp=0.2 = 0.2π ωs .ws/pi.1 = 0.a).Rp. plot(w/pi.2. Matlab Code: % Given BPF specifications wp1=0.6*pi.1). Subplot(2.2dB ω p .

grd.Required Routine function[db.1000. mag=abs(H).grd. w]=freqz_m(b.mag.a.w).564π 232 . db=20*log10((mag+eps)/max(mag)).a) %Modified version of freqz sunroutine %---------------------------------------%[db.w]=freqz(b.w]=freqz_m(b. grd=grpdelay(b. H=(H(1:1:501))'. pha. pha=angle(H). %End of function Matlab Output: N= 7 (Note: Filter order is 2N) wn= 0. w=(w(1:1:501))'.a) %db=Relative magnitude in dB computed over 0 to pi radians %mag=absolute magnitude computed over 0 to pi radians %pha=Phase response in radians over 0 to pi radians %grd=group delay over 0 to pi radians %w=501 frequency samples between 0 to pi radians %b=numerator polynimial of H(z) (for FIR:b=h) %a=denominator polynomial of H(z) (for FIR: a=[1]) % [H.a.208π 0. mag.'whole').pha.

plot(w/pi. be specified in %units of pi radian [N. grid.2. Matlab Code: % Given BSF specifications wp1=0.w]=freqz_m(b. %Modified freqz command [db.mag).2dB ω p .2 = 0.65π Solution: We will use the Elliptic filter for this example.2.3).'stop').Practice .7π As = 50dB ωs .Infinite Impulse Response Filters: Bilinear Transformation Method (13) Design an IIR Bandstop filter to meet the following specifications: ω p . (The required routine and matlab output are shown on the next page) 233 .grd.ws/pi. subplot(2. ws=[ws1 ws2].2. title('Magnitude Response in dB'). As=50. zplane(b.7*pi.wn.As. Subplot(2.2 = 0.3*pi.As) %Design digital Elliptic bandstop filter [b. title('Phase Response'). ylabel('Decibels'). ws1=0. grid.65*pi.Rp.2).a). title('Magnitude Response'). grid.3π ωs . xlabel('frequency in \pi units').pha.4). ylabel('Phase in \pi units'). %ylabel('Phase in \pi units').2π R p = 0.a). axis([0 1 -100 10]).mag. %Plot subplot(2. axis([0 1 0 1]).1 = 0.1).2. axis([0 1 -1 1]). xlabel('frequency in \pi units'). plot(w/pi.Rp. xlabel('frequency in \pi units'). plot(w/pi. Rp=0. subplot(2.2.db). title('Pole-zero plot'). wp2=0.2*pi.a]=ellip(N. ws2=0.1 = 0. wp=[wp1 wp2]. %Calculate Elliptic filter parameters %Matlab requires that freq.wn]=ellipord(wp/pi.pha/pi).

grd. mag.a) %Modified version of freqz sunroutine %---------------------------------------%[db.Required Routine function[db. mag=abs(H). %End of function Matlab Output: N= 6 wn= (Note: Filter order is 2N) 0. grd=grpdelay(b.a. pha=angle(H).pha. db=20*log10((mag+eps)/max(mag)).w).255π 0.7π 234 .mag.a.w]=freqz_m(b.grd. H=(H(1:1:501))'.'whole').w]=freqz(b. pha. w]=freqz_m(b.a) %db=Relative magnitude in dB computed over 0 to pi radians %mag=absolute magnitude computed over 0 to pi radians %pha=Phase response in radians over 0 to pi radians %grd=group delay over 0 to pi radians %w=501 frequency samples between 0 to pi radians %b=numerator polynimial of H(z) (for FIR:b=h) %a=denominator polynomial of H(z) (for FIR: a=[1]) % [H. w=(w(1:1:501))'.1000.

Fs1) >>[numdd.w]=freqs(num.3).Infinite Impulse Response Filters: Impulse Invariance Method (1) An analog filter is specified by its transfer function given by Ha (s) = 0. >>[numd. >>plot(Fd2.Fd1]=freqz(numd.B.’k’. >>[Hd2.dendd]=impinvar(num. one with a sampling frequency of 20 Hz the other with a sampling frequency of 50 Hz.’LineWidth’.’g’.(Hz)’) >>ylabel(‘Magnitude response. (dB)’) >>title(‘Frequency response: Analog vs digital (impulse invariance)’) 235 .den.’Digital: Fs=50 Hz’) >>xlabel(‘Frequency.20*log10(abs(H)). Impulse Invariance Method Syntax >>[numd.dendd.dend]=impinvar(num. Fs2=50.20*log10(abs(Hd2/Hd2(1))). >>num=0. >>Fs1=20.’Digital: Fs=20 Hz’.256.Fd2]=freqz(numdd.den.3). >>[H.20). Compare the analog filter to two versions of the digital filter.dend.’LineWidth’.[1 2]). dend]=impinvar(num.5 ( s + 4 ) ( s + 1)( s + 2 ) Use the impulse invariance method to design its digital counterpart. >>plot(w/(2*pi).’whole’.den).256.5*[1 4].den.grid >>hold on >>plot(Fd1.Fs2).’whole’. Fs) Practice . >>Hold off >>legend(‘Analog’.’LineWidth’.50).20*log10(abs(Hd1/Hd1(1))). >>[Hd1. >>den=conv([1 1].3).

Practice . bandpass.2 dB ripple in the passband. >>Fn=Fs/2.0056 0 0.dend]=yulewalk(N.1726 3. 180 Hz. >>[b.7288 3.5525 -3. and 400 Hz >>Fs=1e3.[100 180 220 400].2000 -69. with 0.0056 0 -0.7288 y [ n − 1] −3. >>fc=[180 220]/Fn. therefore we are not constrained to the standard types like lowpass.0019 x [ n − 6] + 1.fc) b= 0.1034 y [ n − 4] + 1.0000 -1. Find the attenuation at 100 Hz.5525 y [ n − 2] + 3.0056 x [ n − 2] + 0. Deduce the difference equation describing the filter 2.Infinite Impulse Response Filters: Impulse Invariance Method (2) A sequence has 1 kHz sampling frequency.0019 y [ n ] = 0. The Yule Walker Design Method The Yule walker method is a direct procedure for designing IIR digital filters.6646 0 -0. 220 Hz.f. etc.Fs))) HdB = -48.0019 a= 1.3156 y [ n − 5] − 0. The desired response is specified numerically in the frequency domain. Design a 6-th order Chebyshev bandpass filter that has a bandwidth of 40 Hz centered at 200 Hz.1034 -1.3156 0. Syntax >>[numd.0019 x [ n ] − 0.0056 x [ n − 4] − 0.4665 C.2.a.6646 y [ n − 6] >>HdB=20*log10(abs(freqz(b.a]=cheby1(3.0.m) 236 . 1.1943 -0.1726 y [ n − 3] − 3.2000 -0.

The vectors m and f must be of the same length. >>plot(f.returns row vectors numd. clc. >>[H. where f is a vector of frequency points.2). >>f=f/(fs/2). >>plot(w/pi*fs/2. and dend containing the (N+1) coefficients of the order N IIR filter whose frequency-magnitude characteristics approximately match those given in vectors f and m.(Hz)’) >>legend(‘Desired frequency response’.m.5 1 1]. where 1 corresponds to the Nyquist frequency.5 to 1 225 250 1 >>clear all. >>ylabel(‘|H(f)|’).’LineWidth’. >>f=[0 100 150 200 225 250]. corresponding to steps in the frequency response.f.’Yule walker approx.’) 237 .. >>m=[1 1 0.den).’LineWidth’. Duplicate frequency points are allowed.den]=yulewalk(n.2). >>xlabel(‘Frequency.abs(H).Infinite Impulse Response Filters: The Yule Walker Design Method Design an IIR filter having the following specifications with a sampling rate of Fs=500 Hz From To (Hz) Magnitude 0 100 1 100 150 decrease linearly from 1 to 0. >>hold off >>title(‘Yule walker approximate IIR filter’). Practice . >>n=6.5 0. with all intermediate points in increasing order.’r’. grid >>hold on >>fs=500.w]=freqz(num.5 150 200 0. ranging from 0 to 1.5 200 225 increase linearly from 0. >>[num. The vector m contains the desired magnitude response at the frequencies specified by f. The first point of f must be zero and the last point 1.m).

2. Practice . plot(fq. plot(f. >>subplot(2. thus yielding some deviations from the wanted frequency response.fq]=freqz(bk.f]=freqz(b.ak) E.a.Infinite Impulse Response Filters: Effect of Coefficient Quantization >>[b.512. >>[H.2 0. Cascade implementations have better quantization properties than their finite-precision canonic equivalents. >>bk=round(b/max(b)*2^n)/2^n*max(b). the coefficients of the designed filters are typically quantized.30.2). >>n=8. >>ak=round(a/max(a)*2^n)/2^n*max(a).20*log10(abs(H))).[0.4]). Similarly.2.a) >>subplot(2.zplane(b. 238 . This can cause the filter coefficient to be slightly inaccurate.zplane(bk. The overall transfer function is simply the product of the individual transfer functions.2.3).2.20*log10(abs(Hq))).a]=ellip(4. Cacade Implementation of IIR Filters IIR filters are typically implemented by cascading low order sections such as secondorder section (biguads). the function tf2sos converts a digital filter transfer function parameters to second-order section form. >>subplot(2.4).1).D. MATLAB function zp2sos converts pole-zero-gain representation of a given system to an equivalent second section representation. >>subplot(2. and even instability.2.ak.8000).512. Effect of Coefficient Quantization in IIR Filters In the design of digital filters.8000). >>[Hq.

g]=zp2sos(z.. 3......g]=tf2sos(num..p.’scale’) Returns a matrix sos in second-order section form the gain g equivalent to the zero-polegain system represented by input arguments z. 4. >>[sos. . ⎡ b01 ⎢b ⎢ 02 sos = ⎢ .... ( z − zn ) ( z − p1 )( z − p2 )( z − pm ) Where n and m are the length of z and p..k) >>[sos. ...p... and k.g]=zp2sos(z. 2. H ( z ) = g∏ k =1 L b0 k + b1k z −1 + b2 k z −2 1 + a1k z −1 + a2 k z −2 239 . >>[sos.’order’) >>[sos. ⎥ ⎥ .. a1L a21 ⎤ a22 ⎥ ⎥ . b2 L 1 1 . ⎢b0 L ⎣ b11 b12 . The matrix sos is of size L-by-6.. ..den) Returns a matrix sos in second-order section form with gain g equivalent to the system specified by input arguments num and den... and k is a scalar gain.. ⎢ ⎢ . ..... The poles and zeros must be real or complex conjugates.Syntax 1.k.p. H (z) = k ( z − z1 )( z − z2 ) . ⎥ a2 L ⎥ ⎦ Whose rows contain the numerator and denominator coefficients b1k and a1k of the second order section of H(z).p. b1L b21 b22 . 1 a11 a12 .k.’order’...g]=zp2sos(z. respectively.

>>[n..0000 1.A]=butter(N. The sampling frequency is 10000 Hz. >>[B.ws. FIR filters are popular for digital filter design because of the following reasons: □ They are inherently stable.3455 1.0000 0.p.0000 -1..3083 0.0043 1.6407 Finite Impulse Response (FIR) Filters A FIR filter is a recursive filter that has a transfer function of the form: H ( z ) = b0 + b1 z −1 + .k]=td2zp(B. 240 .k) sos= 0.Wn]=buttord(wp. >>sos=zp2sos(z.0043 1.0000 0. an FIR filter is completely specified by a finite set of coefficients.Rp.0087 2. elsewhere ⎭ As can be seen.0729 -1. 0 ≤ n ≤ N − 1⎫ Hence the impulse response h[n] = ⎨ ⎬ ⎩0. □ They can achieve peruse phase linearity in the passband □ They are easily implementable In this section we consider the techniques based on the window method and the Parks and McClellan algorithm (minimax).Cacade Implementation of IIR Filters - Design a Butterworth Digital lowpass filter with no more than 1 dB attenuation below 800 Hz and at least 56 dB attenuation at and beyond 3200 Hz. + bN −1 z1− N = ∑ bn z − n n=0 N −1 ⎧bn .0000 1. >>[z.p. Find the coefficients of second-order sections.Practice .Rs).A).Wn).

Chebwin.Finite Impulse Response (FIR) Filters: Window Method: Hamming (1) >>b=hamming(N) This command is used to truncate the infinite impulse response of an ideal digital filter with the result being an FIR filter with length N. They are used to truncate the infinite impulse response of an ideal filter with the result being an FIF filter. Repeat part (1) using the MATLAB built-in function (The matlab codes and the plots are shown on the next page) 241 .46 cos ⎜ ⎟. There are many types of window proposed for use in spectral analysis and filter design. Window Method Windows are used to manipulate data (original signal) in such a way. The Hamming window is defined by ⎧ ⎛ 2π n ⎞ ⎪0. Blackman. Boxcar (uniform). Bartlett. Write a piece of code to plot it 2. MATLAB has several built-in window functions.A. namely. wm [ n ] = ⎨ ⎝ N −1 ⎠ ⎪0. Hamming.54 − 0. Hann. Window Functions Rectangular Hanning Hamming Kaiser Chebyshev Blackman Bartlett Triangular MATLAB Keywords rectwin hanning hamming kaiser chebwin blackman bartlett triang A. that the desired information can be extracted from the spectrum.1 Window Method: Hamming & Hanning Practice . Kaiser. elsewhere ⎩ 0 ≤ n ≤ N −1 1. and Triang.

required routine & plots on the following pages) 242 . stem(index.2 ⇒M ≥ = 41. we can use a Hanning window (which gives a minimum attn.hamming(41).Finite Impulse Response (FIR) Filters: Window Method: Hanning (2) Design an FIR lowpass filter to meet the following specifications: ω p = 0.’filled’) Practice . Δω = ωs − ω p = 0.54-0.4π + 0.4π As = 40dB Solution: For a 40dB stop band attenuation.stem(index. say M=42 M 0.1). index(n+1)=n.N=40.1. of 44 dB).ham.25π = 0.’filled’) subplot(2.25π R p = 0.15π ≥ 6.1.1dB ωs = 0.46*cos((2*pi*n)/N). for n=0:N ham(n+1)=0.33.2π 6.2). end subplot(2.15 = 0.325π 2 ωc = ωs + ω P 2 (Matlab codes.

w]=freqz_m(h. (The required routines and matlab output are shown on the next page) 243 .3]). grid. h=hd. subplot(2. %Modified freqz command [db. title('Actual Impulse Response').2). stem(n. ws=0. stem(n.mag.*w_han. As=-round(max(db(ws/delta_w+1:1:1:501))) %Plots subplot(1.5*(ws+wp). stem(n. wc=0.2.2. since As=40 dB M=ceil(6.hd).2. axis([0 1 -100 10]).1 0. %Compuate actual passband ripple Rp=-(min(db(1:1:wp/delta_w+1))) %Compute actual stopband attn. axis([0 M-1 -0. title('Ideal Impulse Respose').ylabel('hd(n)').M).2*pi/tr_width) n=[0:M-1]. %Use Hanning window.w_han). xlabel('n').4*pi. tr_width=abs(ws-wp). ylabel('h(n)'). w_han=hanning(M)'.3]).1). xlabel('frequency in \pi units'). subplot(2. axis([0 M-1 -0.1) subplot(2.1 0. %Compute Ideal filter impulse response hd=ideal_lp(wc.4).db).h). axis([0 M-1 0 1. %Ideal LPF cutoff freq.pha.25*pi. ylabel('Decibels').3).1]). %frequency spacing (freqz_m computes 1000 samples) % of DTFT on the unit whole circle) delta_w=2*pi/1000. subplot(2.[1]). plot(w/pi. title('Magnitude Response in dB'). ylabel('w(n)').2. title('Hanning Window'). xlabel('n'). xlabel('n').grd.1.Matlab Code: wp=0.

grd=grpdelay(b.mag.w).M) % hd=ideal impulse response etween 0 to M-1 % wc=cutoff frequency in radians % M=length of the ideal filter % alpha=(M-1)/2.1000.a) %Modified version of freqz sunroutine %---------------------------------------%[db. mag=abs(H). %add smallest number to avoid divide by zero hd=sin(wc*M) . mag.Required Routine I function[db. pha=angle(H).w]=freqz(b. grd. n=[0:1:(M-1)] M=n-alpha+eps. an ideal impass impulse response hd[n] is required. db=20*log10((mag+eps)/max(mag)). w]=freqz_m(b.M) %Ideal LowPass filter computation %---------------------------------------%[hd]=ideal_lp(wc.a) %db=Relative magnitude in dB computed over 0 to pi radians %mag=absolute magnitude computed over 0 to pi radians %pha=Phase response in radians over 0 to pi radians %grd=group delay over 0 to pi radians %w=501 frequency samples between 0 to pi radians %b=numerator polynimial of H(z) (for FIR:b=h) %a=denominator polynomial of H(z) (for FIR: a=[1]) % [H.grd. w=(w(1:1:501))'.a. H=(H(1:1:501))'.w]=freqz_m(b.'whole')./ (pi*M). pha.a. %End of function Required Routine II To design FIR filters based on the window technique.pha. Therefore it is convenient to have a simple routine that creates hd[n] as follows: function hd=ideal_lp(wc. (Matlab output and the plots are shown on the next page) 244 .

Matlab Output M=42 Rp=0.0825 As=44 245 .

2.mag.hd). %Plot subplot(2. stem(n.h).w]=freqz_m(h. plot(w/pi. hd(a+1)=0. ylabel('h(n)'). %Modified freqz command [db. if(a==round(a)) %M odd hd=((-1).3]). axis([0 M-1 0 1. subplot(2.1 0.Finite Impulse Response (FIR) Filters: Window method: Hamming (3) Design a M-point FIR differentiator using a Hamming window for M=16 and M=17. (The required routine and the plots are shown on the next page) Required Routine 246 .3). end w_ham=hamming(M)'./(2*(n-a)). a=(M-1)/2. h=hd. title('Magnitude Response'). Matlab Code : M=16.^2). ylabel('Decibels').2. ylabel('w(n)'). title('Actual Impulse Response').1 0. title('Ideal Impulse Response').4).mag)./(pi*(n-a).^((n-0.3]).1). title('Hamming Window').pha. subplot(2.2).^(n-a)).grd. stem(n.[1]). axis[(0 M-1 -0. grid.2. ylabel('hd(n)').Practice . stem(n. xlabel('n'). xlabel('frequency in \pi units').1]).w_ham).5-a))).* w_ham. xlabel('n').2. n=[0:1:M-1]. else hd=((-1). xlabel('n'). subplot(2. axis([0 M-1 -0. axis([0 1 0 4]).

grd=grpdelay(b. db=20*log10((mag+eps)/max(mag)).w]=freqz_m(b. pha=angle(H).function[db.mag. mag=abs(H).a.1000.'whole').w]=freqz(b. grd. %End of function 247 . H=(H(1:1:501))'. mag.grd.w).pha.a) %db=Relative magnitude in dB computed over 0 to pi radians %mag=absolute magnitude computed over 0 to pi radians %pha=Phase response in radians over 0 to pi radians %grd=group delay over 0 to pi radians %w=501 frequency samples between 0 to pi radians %b=numerator polynimial of H(z) (for FIR:b=h) %a=denominator polynomial of H(z) (for FIR: a=[1]) % [H. pha. w]=freqz_m(b. w=(w(1:1:501))'.a.a) %Modified version of freqz sunroutine %---------------------------------------%[db.

’time’) ylabel(‘\fontname{times}\itw_{\rmkaiser}\rm[\itn\rm]’).A.2 Window Method: Kaiser Window & Its Spectrum Practice .1).’m’.’filled’).stem(0:M.1. b=Kaiser(M+1.grid h=get(gcf.20*log(abs(H)/abs(H(1))). beta=3.’Current Axes’). subplot(2.’LineWidth’. title(‘\fontname{times}\itKaiser window-M\rm=20.grid h=set(gcf. subplot(2.b.1.’Current Axes’).1. ylabel(‘\fontname{times}20 log_{10}|\itW_{\rmkaiser}\rm(\ite^{j\omega}\rm)|’) xlabel(‘\fontname{times}\omega(\times\pi) rad \cdot sample^{-1}’) 248 . xlabel(‘fontname{times}\itn’) [H. M=20. set(h.’b’.N).Finite Impulse Response (FIR) Filters: Window Method: Kaiser Window & Its Spectrum (1) N=1024.2).2).beta).\beta=3’).’FontName’.w]=freqz(b.plot(w/pi.

4*pi. Δf = ωs − ω p 2π = 0. %Ideal HPF cutoff freq. M=2*floor(M/2)+1 %Ensure the M is odd for HPF/BSF if (M <= 21) beta=0 elseif (M < 50) beta=0.95)/(14.Finite Impulse Response (FIR) Filters: Window Method: Kaiser Window & Its Spectrum (2) Design an FIR Highpass filter to meet the following specifications: ω p = 0. wc=0.7) = 5.2dB ωs = 0. As=55. β = 0. %Use Kaiser window M=ceil((As-7.05 ⇒ M ≥ As − 7.95 = 65.52.36*tr_width/(2*pi))+1)+1. say M=67 14.1102* (As-8.4π As = 55dB Solution: Let us use a Kaiser window for this example.5*pi.7) end n=[0:M-1].25π R p = 0. M must be odd (cannot use a type II filter).1 ωc = ωs + ω p 2 = 0.4π + 0. ws=0.5*(ws+wp).45π 2 Matlab Code: wp=0. (More matlab codes continue on the next page) 249 .4+0.5842*(As-21)^0.Practice .36Δf Note: For HPF and BSF.07886*(As-21) else beta = 0. tr_width=abs(ws-wp).1102( As − 8.5π = 0.

a) %Modified version of freqz sunroutine %---------------------------------------%[db.mag.ideal_lp(wc.pha.2. %Modified freqz command [db.grd.mag. axis([0 M-1 -0. mag=abs(H).a.w_kai). title('Magnitude Response in dB'). stem(n. %frequency spacing (freqz_m computes 1000 samples) %of DTFT on the unit whole circle) delta_w=2*pi/1000.pha. axis([0 M-1 -0.grd. stem(n. title('Actual Impulse Response'). pha. As=-round(max(db(1:1:ws/delta_w+1))) %Plot subplot(2. axis([0 M-1 0 1. subplot(2.hd). subplot(2. h=hd.3]). pha=angle(H).db).w]=freqz_m(b.a.1]). ylabel('w(n)').1 0. ylabel('Decibels'). %Compute actual stopband attn. ylabel('h(n)'). grid. Required Routine I function[db. title('Ideal Impulse Response').[1]).xlabel('n'). %End of function (Required Routine II. grd. stem(n. title('Kaiser Window').1). xlabel('frequency in /pi units'). mag.2. subplot(2.3).* w_kai.axis([0 1 -100 10]).w).beta))'. xlabel('n'). w=(w(1:1:501))'.4).2.1 0.2). xlabel('n').w]=freqz(b. db=20*log10((mag+eps)/max(mag)).'whole').w]=freqz_m(h. w_kai=(kaiser (M.a) %db=Relative magnitude in dB computed over 0 to pi radians %mag=absolute magnitude computed over 0 to pi radians %pha=Phase response in radians over 0 to pi radians %grd=group delay over 0 to pi radians %w=501 frequency samples between 0 to pi radians %b=numerator polynimial of H(z) (for FIR:b=h) %a=denominator polynomial of H(z) (for FIR: a=[1]) % [H.h). plot(w/pi.M) . ylabel('hd(n)').3]).M). grd=grpdelay(b.%Compute Ideal filter impulse response hd=ideal_lp(pi.2.1000. H=(H(1:1:501))'. matlab output & the plots are shown on the next page) 250 . w]=freqz_m(b.

Required Routine II function hd=ideal_lp(wc./ (pi*M). %add smallest number to avoid divide by zero hd=sin(wc*M) .M) %Ideal LowPass filter computation %---------------------------------------%[hd]=ideal_lp(wc. Matlab Output M= 69 beta= 5.M) % hd=ideal impulse response etween 0 to M-1 % wc=cutoff frequency in radians % M=length of the ideal filter % alpha=(M-1)/2. n=[0:1:(M-1)] M=n-alpha+eps.102 Rp= 0.024 As= 55 251 .

28.6π = 0.25π = 0. ωc .025 ⇒ M ≥ As − 7.2 2 wp1=0.1 2π = 0.7) end n=([0:M-1]).07886*(As-21) else beta=0.1102*(As-8. w_kai=(kaiser(M.1 + ω p .225π 2 = 0.2 = 0.M) .beta))'.Practice .2π ωs . wc2=0.6*pi.1 2 = 0.36 f Note : We use the more stringent of the two transition band widths. wp2=0.25*pi.2dB ω p .Finite Impulse Response (FIR) Filters: Window Method: Kaiser Window & Its Spectrum (3) Design an FIR Bandpass filter to meet the following specifications: ω p .2π + 0.2 = 0. f = ωs . ws1=0.5π + 0. %Use Kaiser window M=ceil((As-7. %Compute Ideal filter impulse response hd=ideal_lp(wc2.55π 2 ωc . As=40. if (M <= 21) beta=0 elseif (M < 50) beta=0.1 − ω p .5π As = 40dB ωs .25π R p = 0.95 = 89.2 = Matlab Code : ωs .36*tr_width/(2*pi))+1)+1. ws2=0. %Ideal BPF cutoff freq.ideal_lp(wc1.abs(ws2-wp2)).5*(ws2+wp2). tr_width=min(abs(ws1-wp1).4+0. h=hd.M).95)/(14.5842*(As-21)^0.2*pi. (More matlab codes are shown on the next page) 252 .1 = ωs .6π Solution : Let us use a Kaiser window for this example. say M=90 14.1 = 0.*w_kai.5*(ws1+wp1). wc1=0.1 = 0.5*pi.2 + ω p .

stem(n.w]=freqz(b. grd. ylabel('hd(n)').w).grd.mag. xlabel('n'). w]=freqz_m(b.1]). ylabel('Decibels').w]=freqz_m(b.2.max(db(ws2/delta_w+1:1:501)))) %Plot subplot(2. axis([0 M-1 0 1.w]=freqz_m(h.'whole').w_kai).2. subplot(2. %Compute actual passband ripple Rp= -(min(db(wp1/delta_w+1:1:wp2/delta_w))) %Compute actual stopband attn. %frequency spacing (freqz_m computes 1000 samples) %of DTFT on the unit whole circle) delta_w=2*pi/1000. ylabel('w(n)').pha. H=(H(1:1:501))'. Required Routine I function[db.3]). mag.[1]). title('Magnitude Response in dB').a) %Modified version of freqz sunroutine %---------------------------------------%[db. subplot(2. pha=angle(H). title('Ideal Impulse Response').grd.mag. stem(n. matlab output & plots are shown on the next page) 253 .h). stem(n.hd).db). db=20*log10((mag+eps)/max(mag)). title('Actual Impulse Response').1 0.4). xlabel('frequency in \pi units').3). title('Kaiser Window'). grd=grpdelay(b.round(max(max(db(1:1:ws1/delta_w+1)). As= .1 0. axis([0 M-1 -0. ylabel('h(n)'). pha. plot(w/pi. xlabel('n'). mag=abs(H).2). xlabel('n').a.pha.a. w=(w(1:1:501))'.2. axis([0 1 -100 10]).a) %db=Relative magnitude in dB computed over 0 to pi radians %mag=absolute magnitude computed over 0 to pi radians %pha=Phase response in radians over 0 to pi radians %grd=group delay over 0 to pi radians %w=501 frequency samples between 0 to pi radians %b=numerator polynimial of H(z) (for FIR:b=h) %a=denominator polynomial of H(z) (for FIR: a=[1]) % [H.3]). grid.1).1000.%Modified freqz command [db.2. subplot(2. axis([0 M-1 -0. %End of function (Required Routine II.

/ (pi*M). Matlab Output : M= 92 beta= 3.45 Rp= 1.166 As= 40 254 .M) % hd=ideal impulse response etween 0 to M-1 % wc=cutoff frequency in radians % M=length of the ideal filter % alpha=(M-1)/2.Required Routine II function hd=ideal_lp(wc.M) %Ideal LowPass filter computation %---------------------------------------%[hd]=ideal_lp(wc. %add smallest number to avoid divide by zero hd=sin(wc*M) . n=[0:1:(M-1)] M=n-alpha+eps.

the frequency response at the Nyquist frequency is zero.wn. bandpass.’high’). + h [ N + 1] z − N If no window is specified. highpass.’stop’). Syntax: 1. >>h=fir1(N.Wc). 255 . are ordered in ascending powers of z −1 . Ex:>>b=fir1(N-1. with 1 corresponding to the Nyquist frequency.wn. >>h=fir1(N. and bandstop configurations.’high’.A.rectwin (N)) creats a highpass filter with cutoff frequcny Wn using a rectangular window.window).. N is the order of the filter. the default is the hamming window. wn=[w1. 2. h.3 Window-based FIR Filters: The “fir1” Command MATLAB has many commands for designing FIR filters..wn).’stop’.w2]. 3. >>h=fir1(N. The function fir1 implements the classical method of windowed linear-phase FIR digital filter design. >>h=fir1(N.window). and Wc is a normalized cutoff frequency. If wn (unlike other methods. It designs filters in standard lowpass. H ( z ) = h [1] + h [ 2] z −1 + .. >>h=fir1(N. This function uses the hamming window by default.Wn. >>b=fir1(N. The cutoff frequency must be between 0 and 1. %lowpass or bandpass filter %lowpass or bandpass with a custom window Generates an N-th order lowpass FIR filter with cutoff frequency wn (normalized by pi.. >>h=fir1(N. This is because for odd orders. it is the frequency in radians/sample divided by pi) using the specified window and returns the filter coefficients h in length N+1 vector.’high’. wn=[w1. here Wn corresponds to the 6 dB point) is a two-element vector. %highpass FIR filter %highpass with a custom window %bandstop FIR filter.w2] %bandstop with a custom window Note The function fir1 always uses an even filter order for the highpass and bandstop configurations.wn. The output coefficients.window).wn. fir1 returns a bandpass filter with passband w1<w<w2. which is inappropriate for highpass and bandstop filters.wn.

% windowed FIR coefficients (Blackman window) >>hk=fir1(N. % windowed FIR coefficients (uniform window) >>hb=fir1(N.1. % compute frequency response of filter >>mag_dB=20*log10(abs(H)). order=50. [rad/sample’]’) >>ylabel(‘Magnitude response’) % specify order of filter % specify normalized frequency (digital frequency 0.65]).wn. >>wn=0.Blackman(N+1)). % plot the magnitude response >>title(’37-th order FIR filter with uniform window’) 256 .w]=freqz(coef.35 0.Finite Impulse Response (FIR) Filters: Window-based FIR filters: The “fir1” Command (1) Design a 48-th order FIR bandpass filter with passband 0.grid. >>N=37. >>N=50.4.Finite Impulse Response (FIR) Filters: Window-based FIR filters: The “fir1” Command (3) Design a 37-th order lowpass FIR filter with a normalized frequency of 0.512).’).kaiser(N+1. freqz(b.4 pi) % provide filter coefficients % print the coefficients % determine the frequency response % plot magnitude response Practice .w]=freqz(hu.wn.4*pi’). grid >>title(‘FIR filter. >>plot(w/pi. disp(coef) >>[H.4. >>disp(‘coef: ‘).wn=0.65: b=fir1(48.512) Practice . % specify the cutoff frequency in units of pi >>hu=fir1(N.Finite Impulse Response (FIR) Filters: Window-based FIR filters: The “fir1” Command (2) Design a 50-th order lowpass FIR filter with normalized frequency of 0.[0. >>coef=fir1(N. % specify order of filter >>wn=0. >>xlabel(‘Frequency.% windowed FIR coefficients (Kaiser window) >>[H.6.35<w<0. abs(H).1).6.wn. Use the uniform window.mag_dB).1.rectwin(N+1)).653)).’.wn). % compute magnitude in dB >>plot(w/pi.5.Practice .

% sampling frequency >>Fc=1e3.’LineWidth’.w]=freqz(b. % cutoff frequency vector >>b=fir1(order. % coefficients of FIR filter >>[H. Use a sampling rate of 8000 Hz.grid % plot magnitude response >>title(‘FIR filter: window method’) % add title >>xlabel(‘Frequency (Hz)’) % label horizontal axis >>ylabel(‘Magnitude’) % label vertical axis (The plots are shown on the next page) 257 .1.Finite Impulse Response (FIR) Filters: Window-based FIR filters: The “fir1” Command (5) Design a 64-th order lowpass FIR filter for a sampling frequency of 8kHz and cutoff frequency of 1kHz.512. >>N=64.3). % filter coefficients >>[H.hanning(N+1)).abs(H). % frequency response >>plot(f. Use a Hanning window. % normalized lower frequency >>wu=3400/Fn.abs(H)).fc.grid % plot magnitude response >>title(‘FIR bandpass filter’) % add title >>xlabel(‘Frequency (Hz)’) % label horizontal axis >>ylabel(‘Magnitude response’) % label the vertical axis Practice .1.Practice .Finite Impulse Response (FIR) Filters: Window-based FIR filters: The “fir1” Command (4) Design a FIR bandpass filter of order 63 with bandwith extending from 300 Hz to 3400 Hz. % filter order >>Fs=8e3. % frequency response >>plot(w*Fn/pi. % order >>Fs=8000. % sampling rate >>Fn=Fs/2. % Nyquist frequency >>fc=Fc/Fn.512]. >>order=63.Fs). % folding frequency >>wl=300/Fn.wc).f]=freqz(b. % normalized cutoff frequency >>b=fir1(N. % normalized upper frequency >>wc=[wl.wu]. % cutoff frequency >>Fn=Fs/2.

w]=freqz(b.1.Practice . >>Fn=Fs/2.grid >>xlabel(‘Frequency (Hz)’) >>ylabel(‘Magnitude’) >>title(‘FIR filter: window method’) % filter order % sampling frequency % Nyquist frequency % normalized cutoff frequencies % filter coefficients % frequency response % plot magnitude response % label horizontal axis % label the vertical axis % add title 258 .2 kHz and upper cutoff frequency of 2. 2100]/Fn.512).1 kHz.fc. >>b=fir1(N. >>plot(w%Fn/pi.abs(H)).Finite Impulse Response (FIR) Filters: Window-based FIR filters: The “fir1” Command (6) Design a 20-th order band-stop filter for a sampling frequency of 16 kHz. >>[H. >>fc=[1200. >>N=20. lower cutoff frequency of 1. >>Fs=16e3.’stop’).

fir2 uses a Hamming window.m. with sampling frequency 800 Hz and cutoff frequency 65 Hz. >>wc=65/Fn. Vectors f and m specify the frequency points and magnitude values at the specified frequency points. >>Fn=Fs/2. >>b=fir1(N.f. with the first frequency being 0 and the last frequency point being 1 (Nyquist frequency). arranged in ascending powers of z −1 . >>Fs=800. % sampling frequency % Nyquist frequency % normalized cutoff frequency % order % filter coefficients A.window).Practice .f. 259 .The “fir2” Command The function fir2 designs a linear phase FIR filter using the window method with arbitrary shaped magnitude responses.m).Finite Impulse Response (FIR) Filters: Window-based FIR filters: The “fir1” Command (7) Design a 100-th lowpass FIR filter. in the range 0 to 1. %FIR filter with Hamming window %FIR filter with custom window designs an N-th order FIR digital filter with the frequency response specified by vectors f and m and returns the vector h of length (N+1) corresponding to the filter coefficients. The vector window must be (N+1) elements long. The frequency points are arranged in an increasing order. corresponding to steps in the frequency response.4 Windowed Method with an Arbitrary Shape ( Multiband FIR Filter Design) . Duplicate frequency points are allowed. Syntax >>h=fir2(N.wc). By default. >>N=100. >>h=fir2(N.

’LineWidth’.3).Practice . m. >>Fs=1000.256).Finite Impulse Response (FIR) Filters: Window Method w/ an Arbitrary Shape - Design a 30-th order lowpass filter and overplot the desired frequency response with the actual frequency response. >>m=[1 1 0 0]. >>plot(f*Fn.w]=freqz(h.m). >>h=fir2(30. >>[H. ‘o’.Fn*w/pi. ‘Designed’) >>xlabel(‘Frequency (Hz)’) >>ylabel(‘Magnitude’) >>title(‘FIR filter’) 260 . abs(H). >>f=[ 0 400 400 500]/Fn. >>Fn=Fs/2.1.grid >>legend(‘Ideal’.f.

f is a vector of pairs of frequency points. Frequency represents a vector of frequencies specifying the start and endpoints for each band.512). Here is an illustration of the procedure. The bands are specified in terms of the Nyquist frequency. Magnitude specifies the corresponding magnitude at the edges of the bands. specified in the range vetween o and 1. This method is carried out in MATLAB using the remez command.f.co]=freqz(b.6 0. but not restricted to having zero slopes.a.Finite Impulse Response (FIR) Filters: Park-MacClellan Optimal Design (1) Graph the desired and actual frequency responses of a 17-th order Parks-McClellan bandpass filter: >>f=[0 0.optims) Returns row vector b containing the (n+1) coefficients of the order in FIR filter where frequency-amplitude characteristics match ? given by vector f and a. a is a vector containing the desired amplitudes at the points specified ? and a must be an even number Practice . and must always start with 0 (DC) and end with 1 (Nyquist frequency).7 1].a.a) >>[n.plot(f. because it produces a design that gives the smallest error between the actual and desired responses. The frequencies must be in increasing order.4 0.1.f. The bands must be linear. Syntax >>b=remez(n. A FIR filter designed via this procedure is specified via the order of the filter and the location of its passbands and stopbands. This method is considered optimal. a=[0 0 1 1 0 0] >>b=remez(17. Parks-MacClellan Optimal FIR Filter Design The most widely accepted method for designing FIR filters is based on an algorithm devised by Parks and McClellan based on the Remez exchange algorithm. Filters designed this way exhibit an equiripple behavior. To specify the bands two variables.3 0. hwere 1 corresponds to half the sampling frequency. frequency and magnitude are given.w/pi.B.abs(h)) 261 .

>>gain=[0 0 1 1 0 0]. To implement this algorithm. Assume a 5 kHz sampling frequency.freq. 1. we apply the remez command to derive the filter coefficients. one containing six frequencies each normalized to Nyquist frequency. Lastly. % frequency vector % amplitude vector % order of filter % coefficients % plot responses 262 .gain). >>order=31.Finite Impulse Response (FIR) Filters: Parks-MacClellan Optimal Design (2) Design an optimal 32-coefficient FIR bandpass filter that aims to achieve unity gain between 600 Hz and 900 Hz. we need to specify two vectors. >>freq=[0 400/2500 600/2500 900/2500 1100/2500 2500/2500]. The order of the filter is taken to be the number of coefficients less one. the other containing the desired amplitude response at the six specified frequencies. >>freqz(coef.5000). and zero gain below 400 Hz and above 1100 Hz. >>coef=remez(order. 400.Practice .

>>N=43. >>m=[0 0 1 1 0 0]. >>F4=2200/Fn.Fs).1.m).f]=freqz(b. >>b=remez(N-1.20*log10(abs(H))).Finite Impulse Response (FIR) Filters: Parks-MacClellan Optimal Design (3) Design a bandpass filter using Remez >>Fs=8000.F. >>plot(f. >>[H. >>F3=2000/Fn.512. >>F=[0 F1 F2 F3 F4 1].Practice . >>F1=1000/Fn. >>Fn=Fs/2.grid 263 . >>F2=1200/Fn.

ws1=0. say M=73 14.Practice .1 2π = 0. wp2=0. ws2=0.2*pi.6*pi. Rp=0.2dB ωP . %Calculate delta_1 and delta_2 d1=(10^(Rp/20)-1)/(10^(Rp/20)+1). (More matlab codes continue on the next page) 264 . %vector of tolerances tolerance=[d2.2 = 0.2. ( R / 20) ⎛ 1 − δ1 ⎞ 10 p − 1 Rp = −20 log10 ⎜ = 0.1 = 0.1 − ω p .01 ⎝ 1 + δ1 ⎠ f = ωs .Finite Impulse Response (FIR) Filters: Parks-MacClellan Optimal Design (4) Design an FIR Bandpass filter to meet the following specifications: ω p .d2].1 = 0.025 ⇒M ≥ −20 log10 δ1δ 2 − 13 + 1 = 72. As=40.2π ωS . Matlab Code : %Given BPF specifications wp1=0.6π Use the equiripple/minimax design method Solution : First we find the tolerance δ1 and δ 2 .5π As = 40dB ωS .012 ⎟ ⇒ δ1 = ( Rp / 20) 10 +1 ⎝ 1 + δ1 ⎠ ⎛ δ ⎞ As = −20 log10 ⎜ 2 ⎟ ⇒ δ 2 = (1 + δ1 )10( − As / 20) = 0.2.2 = 0.25*pi.25π R p = 0. We will use the matlab function remezord to estimate M. d2=(1+d1)*(10^(-As/20)).6 f Empirical estimate of M is usually an underestimate.d1.5*pi.

Ass=-round(max(max(db(1:1:floor(ws1/delta_w)+1)). (Required Routine.Ao.1]).grd.grd. grid. %Estimate order of filter for Remez algorithm [N. title('Passband Ripple').ws2]/pi. %vector of nominal amplitudes amplitude=[0 1 0]. matlab output and plots are shown on the next page) 265 . plot(w/pi. stem(n. axis([0 1 1-2*d1 1+2*d1]). xlabel('frequency in \pi units').amplitude. [db.[1]). %freqency spacing (freqz_m computes 1000 samples) %of DTFT on the unit whole circle) delta_w=2*pi/1000.w]=freqz_m(h.mag). title('Actual Impulse Response'). grid. xlabel('frequency in \pi units').w]=freqz_m(h.4).wp1.W). %compute actual passband ripple %Rp= max(abs(mag(floor(wp1/delta_w)+1:1:floor(wp2/delta_w)+1))) %Compute actual stopband attn. axis([0 1 0 2*d2]).max(db(floor(ws2/delta_w)+1:1:501)))) %If specs.mag).h).W). %Modified freqz command [db. axis([0 1 0 1. plot(w/pi. xlabel('freqnency in \pi units').1).Ao. Ass= -round(max(max(db(1:1:floor(ws1/delta_w)+1)). not met.2.[1]).mag. grid. title('Stopband Ripple').mag).Fo. must increase N and repeat process while (Ass<As) N=N+1 h=remez(N.Fo. subplot(2. %Plot subplot(2.2).pha. title('Magnitude Response').W]=remezord(f.mag. axis([0 M-1 -1 1]).pha.Ao. so M=N+1 is filter length M=N+1 n=[0:1:M-1].%vector of band edge frequencies (normalized in units of pi) f=[ws1. %Call the remez function for filter design h=remez(N.Fo.max(db(floor(ws2/delta_w)+1:1:501)))) end %N is filter order. ylabel('h(n)').wp2. subplot(2. xlabel('n'). subplot(2.2.2.3).2. plot(w/pi.tolerance).

a) %Modified version of freqz sunroutine %---------------------------------------%[db.'whole'). H=(H(1:1:501))'. %End of function Matlab Output : Ass= 40 M= 77 Note that the filter length of 77 is considerably less than 90 for the window design case. grd.mag. pha.a.a) %db=Relative magnitude in dB computed over 0 to pi radians %mag=absolute magnitude computed over 0 to pi radians %pha=Phase response in radians over 0 to pi radians %grd=group delay over 0 to pi radians %w=501 frequency samples between 0 to pi radians %b=numerator polynimial of H(z) (for FIR:b=h) %a=denominator polynomial of H(z) (for FIR: a=[1]) % [H.pha. w=(w(1:1:501))'. w]=freqz_m(b. mag.w).a. grd=grpdelay(b.w]=freqz(b.grd. 266 . mag=abs(H).w]=freqz_m(b.Required Routine function[db. db=20*log10((mag+eps)/max(mag)).1000. pha=angle(H).

[1]). %Call the remez function for filter design h=remez(N. title('Actual Impulse Response'). title('Magnitude Response in dB').Practice .9*pi]/pi. title('Phase Response'). N=M-1 is filter order M=21.pha).2.pha. axis([0 1 0 1. grid. subplot(2. Matlab Code : %M=filter length. axis([0 1 -40 0]).1*pi to 0. title('Magnitude Response'). axis([0 M-1 -1 1]). subplot(2.mag). xlabel('n'). xlabel('frequency in \pi units'). 267 . grid. stem(n.mag.2. amplitude=[1. axis([0 1 -pi pi]). xlabel('frequency in \pi units').grd.w]=freqz_m(h.2.db).Finite Impulse Response (FIR) Filters: Parks-MacClellan Optimal Design (5) Design a 21-point FIR Hilbert transformer using the equiripple design method.h). %Want Hilbert transformer to work in the %freq. 0. N=M-1.1*pi.amplitude.2). ylabel('h(n)'). %Modified freqz command [db.1). %Plot subplot(2. n=[0:1:M-1].9*pi f=[0. plot(w/pi.1]). plot(w/pi.1]. plot(w/pi.2. 'Hilbert').f. subplot(2.3). xlabel('frequency in \pi units'). band 0.4). grid.

Five Realizations MATLAB has a built-in filter realization wizard that automatically generates filter realizations with specific architecture.0102 a= 1.0000 -1.0408 0.9684 1.1204 Configure the Wizard You configure the wizard to use the coefficients b and a of the designed filter of a DirectForm II structure: □ Select Direct-FORM II from the Type menu □ Type b in the numerator text field □ Type a in the denominator text field 268 . The Wizard’s interface allows you to select from the following realizations: □ □ □ □ □ Direct form I & II Symmetric FIF Lattice (MA) Lattice (AR) Lattice (ARMA) Launch the Wizard To launch the filter realization wizard.0408 0.25) b= 0.7245 0.a]=butter(4. lowpass Butterworth filter >>[b.0.Filter Realization Wizard- Direct Form II Realization: Design a fourth-order.7359 -0. quarter-band.0613 0.0102 0. simply type dspfwiz in the command window: >>dspfwiz <enter> Practice .

The direct-Form II filter realization is depicted below: Syntax >>y=filter(b.a. Practice -MATLAB Filter- An LTI filter is described by a difference equation y [ n] − 0. >>a=[1 0 -0. filter normalizes the filter coefficients by a(1). >>stem(y) %vector corresponding to the coefficients of y’s %vector corresponding to the coefficients of x’s %impulse input %impulse response %plot impulse response 269 . filters the data in vector x (input) with the filter described by coefficients vectors a (den) and b (num) to create the filtered data vector y (output). >>b=[0. >>y=filter(b.634 x [ n − 2] Determine the impulse response and sketch it.634 0 0.634].634 x [ n] + 0. If a(1) ≠ 0. you double-click the created block to see the Direct-Form II filter realization that the Wizard created. >>imp=[1 zeros(1. The command filter works for both real and complex inputs.31)].268].x).imp).268 y [ n − 2] = 0.a. Finally.The graphical use interface with these setting is depicted below: Press the Build button to create the specified filter subsystem in a new model window.

cutoff). [num. clc.t]=impz(b.t. y=filter(num.n.’ro’.05.a. cutoff=0. t=(1:350).50). plot(t.[ ]. and then plots the results.den.a.Fs) % compute n samples of the impulse response % Fs: sampling rate Compute the impulse response of the filter with numerator coefficients b and denominator coefficients a.’LineWidth’.x)..Fs) >>[h.den]=butter(1.2]) xlabel(‘Time’) ylabel(‘Amplitude’) title(‘Low-pass filtered squarewave’) % time base % an array of 1’s % an array of 0’s % signal with three pulses % cutoff frequency of filter % Butterworth filter % filtered output 270 .t]=impz(b.a) >>[h.t]=impz(b.x. close all.50). o=ones(1. The function impz chooses the number of samples and returns the response in column vector h and times in column vector t Practice -MATLAB impz(1) This script file takes a series of 3 pulses and passes them through a Butterworth filter. clear all. x=[z o z o z o z]. z=zeros(1. grid axis([0 350 0 1.2).y.Syntax >>[h.

(dB)’) >>title(‘Frequency response: Anaolg vs digital (bilinear)’) >>legend(‘Analog’.5*[1 4].20*log10(abs(Hd/Hd(1))).256. >>[Ha.Fd]=freqz(numd. (Hz)’) >>ylabel(‘Magnitude response.dend]=bilinear(num. >>num=0.Practice -MATLAB impz(2) An analog filter is described by its transfer function given by Ha (s) = 0.10.5 ( s + 4 ) ( s + 1)( s + 2 ) Assume the sampling frequency of 10 Hz and a matching frequency of 1 Hz.20*log10(abs(Ha/Ha(1))).’k’. >>plot(wa/(2*pi).3).wa]=freqs(num.’b’.3) >>hold off >>xlabel(‘Frequency.dend. >>[numd.10). >>hold on >>plot(Fd. ‘Digital: bilinear’) 271 .’LineWidth’.[1 2]).’whole’.1).den). >>[Hd. Use the bilinear transformation method to find a digital filter equivalent.’LineWidth’. >>den=conv([1 1].den.