Signal Processing and Linear Systems-B.P.Lathi copy

# 6902z 12 hzj z2 14678z 06298 0 15 f ig 1 215

This preview shows page 1. Sign up to view the full content.

This is the end of the preview. Sign up to access the rest of the document.

Unformatted text preview: 8 + (1.248)2 S tep 3: F ind H[z] b y u sing t he b ilinear t ransformation Finally, t he d esired transfer function H[z] o f t he b andpass filter is o btained from 1 i p(8) b y replacing 8 with np[z] from Eqs. (12.67) a nd (12.68). From Eq. (12.68), we o btain a= W P ,'WP2' - 1 = - 0.9485 = - 0.9021 1.0515 wp , ' w p , ' + 1 a nd b = wp , ' - w p , ' = 0.1665 wp, ' wp,' + 1 1.0515 = 0.1583 M ATLAB gives b =0.0296 0 - 0.0593 0 0 .0296 a nd a =l - 3.119 3 .9259 - 2.3539 0 .576 for B utterworth o ption. Therefore Hz_ - Z4 - 0.0296(z2 - 1)2 3.119z 3 + 3.9259z 2 - 2.3539z + 0.5760 a r esult, which agrees w ith t he a nswer found in E xample 12.8. To p lot t he a mplitude a nd t he p hase response, we c an use t he l ast 9 functions in E xample C I2.5. 0 o C omputer E xample C 12.S Using b ilinear t ransformation w ith prewarping, design a b andstop d igital filter t o m eet t he following specifications: Wp= [450 4 000], Ws= [1000 2000], Gp = - 2.1 d B , a nd Gs = - 20 dB. Use T = 1r/104. 756 12 F requency Response a nd Digital Filters As before, we give here MATLAB functions t o d esign t he f our t ypes o f a pproximations. T he i nput d ata a sks for frequencies so normalized t hat t he s ampling r adian frequency is 2. As explained i n E xample C12.5, all t he r adian f requencies c an b e n ormalized by multiplying each of t hem b y T/,Tr. I n t he p resent case, T = 11'/104 s o t hat T /lr = 1 /104 • T hus, t o n ormalize a ny r adian frequency, we j ust d ivide i t b y 104 • b nzn B utterworth [ n,Wn]=buttord(Wp,Ws,-Gp,-Gs) [ b,a]=butter(n,Wn,'stop') (12.69a) zn = bn Now, by definition, H[z] bn - bi bn -2 I bo + - z- + - z-2- + ... + -zn---I + z n is t he z -transform of h[k]: 00 Chebyshev [ n,Wn]=cheblord(Wp,Ws,-Gp,-Gs) [ b,a]=chebyl(n,-Gp,Wn,'stop') k=O = h[O] + hIll + h[2] + .,. + h[n] + h[n + 1] + ... z t. I nverse Chebyshev [ n,Wn]=cheb2ord(Wp,Ws,-Gp,-Gs); [ b,a] = cheby2 ( n,-Gs, W n , ' stop ' ) z2 zn zn+1 (12.70) C omparison o f t his e quation w ith Eq. (12.69b) shows t hat h[k] = 0 for k > n, a nd Eq. (12.70) becomes E lliptic [n,W n ]=ellipord ( W p , W s ,-Gp,-Gs); H[z] [ b,a] = ellip( n ,-Gp,-Gs, W n , ' stop') = h[O] h[l] h[2] h[n] + - z + -z 2 + ... + -z n M ATLAB r eturns b =0.3762 - 1.3575 1 .9711 - 1.3575 0 .3762 a nd a =l - 2.2523 2 .0563 - 1. 2 053 0 .4197 for Chebyshev option. Therefore H[z] = 0.3762(z4 - 3.6084z + 5 .2395z - 3.6084z + 1) Z4 - 2.2523z 3 + 2 .0563z 2 - 1.2053z + 0.4197 3 2 0 Nonrecursive Filters T he recursive filters are very sensitive t o coefficient accuracy. Inaccuracies in t heir i mplementation, especially t oo s hort a word length, m ay c hange their behavior drastically a nd e ven make them unstable. Moreover, t he recursive filter designs are well established only for amplitude responses t hat are piecewise constant, such as lowpass, bandpass, highpass, and b andstop filters. I n c ontrast, a nonrecursive filter c an b e designed t o have a n a rbitrarily s haped frequency response. In addition, nonrecursive filters can be designed...
View Full Document

## This note was uploaded on 04/14/2013 for the course ENG 350 taught by Professor Bayliss during the Spring '13 term at Northwestern.

Ask a homework question - tutors are online