Professional Documents
Culture Documents
Part 4 IIR PDF
Part 4 IIR PDF
~ 2 logy) Qs (10) e By analytic continuation (jQ = s), equation (8) can be extended to the complex s-domain, giving H,(8)Ha(—s) (11) “T+ es" e The poles of the stable H,(s) are the left-half plane roots of 1 + e2(—s?)” = 0, whereas the poles of the unstable H,(—s) are the right-half plane roots of 1+ e(—s?)" =0. e The N poles of the stable H,(s) are given by ie ye ane L/2+h-N/2N) p= 19...N. (12) or Pk=O~n+ 52, k=1,2,...N, (13a) where 1 (one) mn n(2k— 4 ob = aw cos(5 + 7m) (136) 1. ym r(2k—1) % = ay sin(5 + oN) (13)are For above design, the passband criteria are just met, whereas ait 202N (14a) or a Ag = 10 logy)(1 + 02%). (14) These values are in most cases larger than the specified values A? and A,. All the poles of H,(s) lie on a circle of radius 1/e!/ centered at the origin of the s-plane. The pole with largest imaginary part appears at angle 7/2 + 7/(2N) relative to the positive real axis. The other poles ap- pear at angular increments of 7/N. This is exemplified in transparency 8. The transfer function for which |H,(j0)|?> = 1 or #A,(0) = 1 is then H,(s) = 2, (15a) N [[(s-»%) k=1 where My = [[(-»»). (158) ket All the N zeros are lying at infinity. e For the very original Butterworth filters, €2 = 1 so that |Ha(jQ,)|? = 1/2 and A, = 3 dB.EXAMPLE e Later on, when designing digital filters with the aid of analog classical filters, we end up with the following specifications: A, =0.2 dB, A,>60dB, , = 1.8944272. e Using the formulas described above, we obtain @ = 1040/1 _ 1 = 10°2/10 _ 1 — 0.047238748, AZ = 1048/19 — 4950/10 — 495 and ' > N > logyg[(A* — 1)/e"}/[2 - logyo 25) = 18.202340 > N=14. e The fourteen poles are located at —0.12487140 + j1.10826429, —0.36835261 + 71.05269129, —0.59336309 + 70.94433195, —0.78861987 + j0.78861987, —0.94433195 + 70.59336309, —1.05269129 + 70.36835261, and —1.10826429 + 70.12487140, whereas Ay = 4.60636100, and Ay = 10 logyy(1 + 292") = 64.4267e The pole-plot for this filter (those given by x) as well as the amplitude response are shown in transparency 8 (all the zeros are lying at the infinity (not visible)).CHEBYSHEV FILTERS OR CHEBYSHEV TYPE I FILTERS e The squared-magnitude function of this filter of order N is given by 1 |Ha(GQ)? = 1+ eR)’ (16a) where (N cos") iO] <1 _ J cos(NV cos ; < Tw(@) = eve cosh *Q) |Q|>1 (166) is the Nth-degree Chebyshev polynomial. e In the normalized passband 0 < 2 < Q, = 1, this func- tion alternatingly achieves the values of 1 and 1/(1+e?) at N +1 points such that |H,(j,)|? = 1/(1 + &). For N even, |Ha(j0)/? = 1/(1 + €?) and for N odd, |H,(j0)|? = 1 (equiripple passband). e At infinity, the value of |H,(j9)|? is zero and the first 2N — 1 derivatives are zero (maximally flat stop- band) (see the figure shown below). 1H, (iQ)? Nodd la, Gayl? Neven18- TRANSFER FUNCTION e In order to meet the criteria of the previous trans- parency, it is required that 1 Lo yy cosh VC = 1/4} < 14+ T2(Q,) — A? = cosh !(Q,) , (16a) where cosh™! a can be evaluated from cosh™! x = In(x + V2? — 1). (16b) e Like for Butterworth filters, the transfer function is of the form H,(s) = 2, (17a) [[(- k=l where N TI»), N odd Hy=¢ N (176) 1/(1 +e) [[ (2), N even. k=I e It can be shown (for details see a textbook on analog filters) that the poles H,(s) are located at Pr= K+ 902%, k=1,2,...N, (18a) where elie (2 = 1)r 2 NON on = - (186)19~ —atyt (Qk -1)r OQ, = 5 C05 naan (18¢) and rT l4¢vVi4e," ies Gee Bee (18d) € These poles are on an ellipse centered about the ori- gin in the s-plane. The ellipse has minor-axis length ar and major-axis length a and is given by the equation 2 2 1% + 1% il, (19) g-wtyY +7") For details see transparency 21, which gives the poles of H,(s) by x and the poles of H,(—s) by astrisk. Like for Butterworth filters, all the zeros are located at infinity. Futhermore, the squared-magnitude function of the above filter just meets the passband criteria, whereas & = 14 2T2(Q,) (20a) or As = 1logio[1 + PTH(Q)], (206) which are in most cases larger than the specified values of A? and Ag.~ 20 EXAMPLE e We again consider the specifications: A, =0.2dB, A, >60dB, 0, = 1.8944272. e Like for Butterworth filters, «2? = 0.047238748 and A? = 10°, whereas N > cosh7!{y/[(A? = 1)/2]}/ cosh7*(5) = 7.2808916 = N=8. e The eight poles are located at —0.05514327 + 71.01921190, —0.15703476 + 70.86404612, —0.23501912 + 70.57733716, and —0.27722396 + j0.20273385, whereas Ig = 0.035987195. e The pole-plot for this filter (those given by x) as well as the amplitude response are shown in the next trans- parency (all the zeros are lying at the infinity (not vis- ible)). e For this filter, A, = 67.8310 dB.a1- Eighth-Order Chebyshev Analog Filter with 2, = 1, Q, = 1.8944272, A, = 0.2 dB, and A, > 60 dB &th-order Chebyshev filter with Ap=0.2 1 xx * 0.8} xis the pole of H_a(s) is the pole of H_a(-s) Imaginary part x x 0 Real part ‘8th-order Chebyshev filter with Ap=0.2 50) Amplitude in dB Angular frequency Omega a 10° 107 ‘Angular frequency Omega22~ INVERSE CHEBYSHEV FILTERS OR CHEBY- SHEV TYPE IT FILTERS e The squared-magnitude function of this filter of order N is given by |Ha(i)/? = , (21) © 1+ (A? = 1)/[Tw(O,/2)" e Like for Butterworth filters, this function achieves the value of unity at 2 = 0 with the first 2N —1 derivatives being zero at this point (maximally flat passband). e In the stopband 2, < 2 < ov, |H,(jM)| alternat- ingly achieves the values of 1/A? and zero at N +1 points such that |H(jQ,)|?> = 1/A?. For N even, |Ha(joo)|? = 1/A? and for N odd, |Ha(joo)|? = 0 (equiripple stopband) (sce the figure below). lH, (jal? Nodd lH, (jal? Neven p---}------------s5= 23- TRANSFER FUNCTION e Like for Chebyshev filters, in order to satisfy 1/[1 + (A? = 1)/[Tw(Q5/Qp)]?] > 1/(1 +e), it is required that Ne cosh /{(4? = 1)/e]} (22) cosh!(Q,) e The transfer function is of the form N N Ho IIs ~ »)/T[s —pr), Neven H(s)=4 RT, Hy TT (s~ <)/[](s—p), N oda k=1 k=1 (23a) where N N [G/T N even Mayet kt (230) [[(-»)/ T] (2), N oda. k=1 k=1 e The zeros are located at 2 = FSR VON ay" k=1,2,---,.N. (24) e Note that if N is odd, then for k = (N +1)/2, the zero lies at the infinity (numerator order is N — 1).—24 e The poles are located at Pe =n +I%, K=1,2%...,N, (25a) where Qa; = b OK a me BR (251 ) =58 Q. = 25c Obra me with 1 (2k — 1) Y=? k= Ve =_ sino 25 aK sin[ aN ] (25d) yty! _(@k-1)r = ee (em Be Br 3g sl ar | (25e) and rN y=(A+VA?-1). (25f) e In this case, the poles do not lie on a simple geomet- ric figure as they do for Butterworth and Chebyshev filters. e Note that the above formulas for Chebyshev and in- verse Chebyshev filters have been constructed such that the Chebyshev filter just meets the passband cri- teria, whereas the inverse Chebyshev filter just meets the stopband criteria. e If it is desired that the inverse Chebyshev filter just meets the given passband criteria, then then the de- sired result is achieved by evaluating A? according to equation (20a).EXAMPLE e We again consider the specifications: Ap =0.2 dB, A, >60dB, , = 1.8944272. Like for Chebyshev filters, €? = 0.047238748 and A? = 10°, whereas N > cosh "{\/[(A? — 1)/e]}/ cosh7'(Q,) = 7.2808916 > N=8. e Knowing the fact that the Chebyshev and inverse Chebyshev filter mee + the same amplitude criteria, our filter can be forced to have A, = 0.2 dB by selecting A, = 67.8310 dB, giving A? = 6068718.6. The eight poles of the resulting filter are located at —0.18212766 + j1.16381690, —0.57926246 + 71.10192829, —1.03855485 + j0.88204869, and —1.42446611 + 70.36015085. The eight zeros are located at +71.93154121, tL j2.27840821, +73.40987886,26 and +79.71051342, whereas Ho = 0.00036795086. e The following transparency gives the pole-zero plot as well as the amplitude response for this filter.-27 Eighth-Order Inverse Chebyshev Analog Filter with 2, = 1, 2, = 1.8944272, A, = 0.2 dB, and A, > 60 dB ° 8 ot ‘ ° ke? 8 zd E E-2 8 ° 4 “6 | 3 -10) 2 -10 5 oO 5 10 Real pan Amplitude in dB 28 & ‘8th-order inverse Chebyshev filter with Ap=0.2 and Omegas=1.8944 8th-order inverse Chebyshev filter with Ap=0.2 and Omegas=1.8944 107 10° 10! ‘Angular frequency Omega 238 Amplitude in dB S 1 ‘Angular frequency Omega28- ELLIPTIC (CAUER) FILTERS e Like for Chebyshev filters, in the normalized passband 0logyg(16D)/logy9(1/q) ViFe+y Vire-1 1 (Nes ay al (27a) (27b) (27c) (27d) (27e) Q7f) (279)pan rg (— 1) mg" sinh{(2m + 1)A] | 1+2>21(-1)™q™ cosh[2mA] (27h) (1 + kaG)(1 + 49/k) (271) 2g so (1) 7g") sin{(2m + 1)]Jru/N] 1+ 250%_.(-1)"q” cos[2map/N] 7 (27)) where i, Nodd B {Pap Neven? @=b2-a7 Tk) 00 = Go/ VI: (271) (1 — k07)(1 — 03/k) (27m) 1 Aoi = ro (27n) (Vi)? + (Q3W)? OE 270 Boi KL + 6302) (270) 260V; li = ay (27) _ J % TTi=1[Boi/ Aoi), for N odd a { Tea TTizi[Boi/Aoi], for N even. (274) e In practice, three or four terms in the series of equa- tions (27h) and (27j) are sufficient. e The passband criteria are just met, whereas the result- ing minimum stopband attenuation is Ay = 10 logyo{e?/(16q™) + 1). (27r)EXAMPLE e We again consider the specifications: A, =0.2 dB, A, >60dB, 0, = 1.8944272. e < = 0.047238748 and A? = 10°, whereas k = 1/0, = 0.527864, k’ = \/1 — k? = 0.8493289 go = (1/2)(1 — Vk) /(1 + Vk’) = 0.0204022 = qo + 2g) + 15q§ + 15094° = 0.0204022 D = (A? —1)/e? = 21218562 N 2 logy(16D)/logyo(1/¢) = 5.046816 = N=6. e The six poles of our filter are located at —0.08205619 + 71.03019607, —0.25402886 + 70.79507992, and —0.39500663 + 70.30821324. e The six zeros are located at +71.95117116, +)2.57623214, and +j6.79458015, whereas Hy = 0.00015647808and As; = 76.1109. e The next transparency shows the pole-zero plot and the amplitude response for this filter.33 Sixth-Order Elliptic Analog Filter with 0, = 1 Qs = 1.8944272, A, = 0.2 dB, and A, > 60 dB ’ 6th-order elliptic iter with Ap=0.2 and Omegas=1.8944 a 6 i 6 2 3 E of e | £ % 4 a -al ° J ae 2 4 6 Real part 6th-order elliptic filter with Ap=0.2 and Omegas=1.8944 ‘Amplitude in dB Angular frequency Omega Amplitude in dB é of 1 ‘Angular frequency Omega 334 BILINEAR TRANSFORMATION ¢ The most efficient and popular analog-to-digial trans- formation is the bilinear transformation. e It transforms H,(s) to its digital equivalent H(z) via the relation H(z) = Hol) 5 = e(z — 1)/(z +1): (23) e Here, c can be selected arbitrarily. e Alternatively, H,(s) can be obtained from H(z) via the relation Hz) =H(2)\,—(14s/e)/(1—s/cy (24) e The transformation s = c(z—1)/(z+1) is a one-to-one mapping between the s-plane and the z-plane and has the following desired properties: e The left-half s-plane is mapped to the interior of the unit circle = A stable H,(s) is mapped into a stable H(z). e The right-half s-plane (unstable region for poles) is mapped to the exterior of the unit circle (unstable region for poles). e The imaginary axis s = jQ is mapped to the z- plane unit circle z = e/” > The analog frequency domain (imaginary axis) maps onto the digital fre- quency domain (unit circle), albeit, as we shall see, nonlinearly.-35- RELATIONS BETWEEN THE s- AND z-PLANES Y, s-Plane 2z-Plane Image of s=jQ Image of left half-plane ¢ Substituting z = e!” into equation (23) results in, after some manipulations, H(e!”) = Hy(jetan(w/2)). (25) ¢ Alternatively substituting z = e/ and s = jQ into s =c(z—1)/(z+1) we end up with the realtions Q = ctan(w/2) (26) or w = 2arctan(Q/c). (27) e s = 0 and s = oo are mapped to z = 1 and z = —1, respectively.—36- DIGITAL FILTER DESIGN USING THE BI- LINEAR TRANSFORMATION e The following figure exemplifies the design process, where the bandedges of the digital filter, w, and ws, as well as allowable passband and stopband variations are specified. e Here, the design of the digital filter is converted to that |H(e)| of the analog filter for which the required passband and stopband variations are the same. 3 w37 SYNTHESIS PROCEDURE Step 1 Step 2 Step 3 Step 4 Step 5 Determine c such that w, is mapped to Q, = 1. The condition for cis Q, = 1 = ctan(w,/2), giving oe (28) Determine Q, = ctan(w,/2). Determine ¢? and A? from the passband and stop- band ripples A, and A, using equation (7). Select the analog filter type (Butterworth, Cheby- shev, inverse Chebyshev, or elliptic) and syn- thesize the minimum-order filter transfer func- tion H,(s) whose squared-magnitude resonse stays within the limits unity and 1/(1 +?) in the pass- band 0 < Q < 1 the limits zero and 1/A? in the stopband 2, < Q < c. The desired digital filter is then A(z) = Hals)i5 = e(z-1)f(z+ 1) 29)8 In practice, the desired H(z) can be conveniently gen- erated by first determining the poles and zeros of the digital filter using the relation z = (1+ s/c)/(1— s/c) to each pole and zero of the analog filter. This gives the poles and zeros of the digital filter, denoted by 6, and a, for k = 1,2,---, N, respectively. The resulting H(z) is then N N H(z) = ko [](1- ane) /T]G- &e2)). (80) k=l k=1 The constamt kp can be then determined from the con- dition that H(1) = 1 (|H(e*)| = 1 at w = 0) for But- terworth and inverse Chebyshev filters as well as for Chebyshev filters and elliptic filters for N even. For N odd, H(1) = 1/V1+€? for Chebyshev filters and elliptic filters. Note that the zero at infinity (N zeros for Butter- worth and Chebyshev filters and one zero for inverse Chebyshev and elliptic filters for N odd) are mapped toz=-l.39 - ILLUSTRATIVE EXAMPLE e It is desired to synthesize a Butterworth, Chebyshev, inverse Chebyshev, and an elliptic filter such that the sampling frequency is F, = 10 kHz, the minimum pass- band ripple A, = 0.2 dB in the passband 0 < f < 2 kHz and the minimum stopband attenuation is A, = 60 dB in the stopband 3kHz < f <5 kHz= F,/2. Step 1 w and the ‘real’ frequency are related via w = 2nf/Fs so that in terms of w the passband and stopband edges are w, = 0.47 and w, = 0.67, respectively. Step 2 w, = 0.47 is mapped to Q, = 1 in the bilin- ear transformation by selecting c = cot(w,/2) = 1.3763819. Then, Q, = ctan(w,/2) = 1.8944272. Step 3 Applying equations (7a) and (7b) gives 2 = 104/19 _ 1 = 0.0471285 and A? = 1049/19 = 108 Step 4 These are the same criteria we considered previ- ously for all the four classical analog filter types. All what is left is to apply the bilinear transfor- mation.40- BUTTERWORTH FILTER e By applying the substitution z = (1+ s/c)/(1 — s/c) with c = 1.3763819 to the poles of the anaolg filter given in transparency 15, we end up with the following fourteen z-plane pole locations: 0.89585800erp(+j0.433121817) 0.7152600leap(+j0.429702197) 0.56158624exp(+j0.421939757) 0.42686164erp(+j0.407317997) 0.30642203exp(+j0.379346827) 0.19959206exp(+j0.318418707) 0.11888906eaxp(+j0.154722697) 1 e All the fourteen zeros of the analog filter lying at in- finity are mapped to z = —1. e This filter can be implemented in following the cascade form (see the next transparency) R - i 1+ agiz + az ‘| = Eo Gove + ane 1 M(2) = ko Ih bye! — bz (31) ¢ Combining complex-conjugate pole pairs, seven (R = 7) second-order denominator sections become 1 — 0.210236982"! + 0.01413460z~? 1 — 0.21556526271 + 0.03983699z~?