Open navigation menu
Close suggestions
Search
Search
en
Change Language
Upload
Sign in
Sign in
Download free for days
0 ratings
0% found this document useful (0 votes)
3K views
27 pages
Tarun Kumar Rawat Module 1
Uploaded by
sumit sancheti
AI-enhanced title
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content,
claim it here
.
Available Formats
Download as PDF or read online on Scribd
Download now
Download
Save Tarun Kumar Rawat Module 1(1) (2) For Later
Download
Save
Save Tarun Kumar Rawat Module 1(1) (2) For Later
0%
0% found this document useful, undefined
0%
, undefined
Embed
Share
Print
Report
0 ratings
0% found this document useful (0 votes)
3K views
27 pages
Tarun Kumar Rawat Module 1
Uploaded by
sumit sancheti
AI-enhanced title
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content,
claim it here
.
Available Formats
Download as PDF or read online on Scribd
Download now
Download
Save Tarun Kumar Rawat Module 1(1) (2) For Later
Carousel Previous
Carousel Next
Download
Save
Save Tarun Kumar Rawat Module 1(1) (2) For Later
0%
0% found this document useful, undefined
0%
, undefined
Embed
Share
Print
Report
Download now
Download
You are on page 1
/ 27
Search
Fullscreen
CHAPTER, 14 Multirate Digital Signal Processing é 14.1 INTRODUCTION In our study of digital signal processing (DSP) techniques, so far, we have assumed that all signals in a given system have the same sampling rate. In Chapter 2 (Section 2.12), introduced the concept of sampling rate conversion: decimation (sampling rate decreas) and interpolation (sampling rate increase). A multirate system can inerease or decrease the sampling rate (and thus the sampling interval) of individual signals. These signals can be simultaneously processed in various components of the multirate system. Hence the sampling rate of the signal must be altered as it propagates from one component to another. The signal processing that uses more than one sampling rate to perform the desired operations is known as multirate signal processing. ‘There are two approaches to alter the sampling rate of a digital signal. In the first approach, we can use a digital to analog (D/A) converter to convert a digital signal into an analog signal and then resample it at the new rate using an analog to dig tal (A/D) converter. These D/A and A/D conversion processes introduces quantization noise. Therefore, this is not a right approach. The second approach is to alter the sas pling rate in the digital domain. In multirate systems, we change the sampling rate of a signal digitally downsamplers and upsamplers. Downsampling decreases the samplin rate and upsampling increases the sampling rate of a signal. Downsampling is usually pr’ ceded by 2 low-pass filter (LPF), and the combination is called decimation. Upsamplins is usually followed by a LPP, and the combination is called interpolation. 14.1.1 Advantages of Multirate Digital Signal Processing Multirate signal processing offers the following advantages: Reduced computational complexity Reduced transmission data rate Less memory requirements Lower order filter design Lower coefficient sensitivity and noise ee ek ee 14.2 DECIMATION Decreasing the sampling rate of a signal is called decimation, which consists of @” Es id. i - aliasing LPF followed by downsampling. We will analyze the input-output al downsampler in time domain and present the effect of downsampling, in the ¢-do™™” illMEE quency domain. ‘The frequeney domain analysis will point by a LPF. Multirate recedec time-domain Characterization rate of a signal 7(n) is decreased by a factor M if we pick only every Mth gual and discarding the rest. The process of keeping every Mth sample 1)" Gunal and discarding the rest is known as downsampling by a factor of M. Let he the input and y(n) be the output of a downsampler. The signal y(m) is given 14.2.1 ple of the si y(n) = 2(Mn) 10, 2h, (14.1) The block diagram of a downsampler is shown in Fig. 14.1. a(n) = x(n) {im Wn) = x(nT) = (MT) Input sampling Output sampling jp, 1 frequency = Fy = 7 frequency = F5= 95 ~F Fig. 14.1 Downsampler of a downsampler is related as follows: sampling rate of input 2() and output y(n) Sampling rate of the input signal 2(n) = Fs, Sampling interval of a(n) = Ts Fy ‘Sampling rate of the output signal y(n) = Fs = Wr Sampling interval of y() = 1, = MT. f © onl (oath sampling frequency F, = 10 samples/s, ea isles M = 2. The down- ; Fi = & = 5 samples/s, efor the input signal x(n) and (9).2(4), 2(5), 216), 207), 4 2(2), 24), 2} | y(n) for M = 2 is shown900 Digital Signal Processing chy y(n) = x(2n) xa) a(4) ‘ po x(2) x(6) (0) 1 ss) x(0) x(8) | | | at a a oor? Dns 7 F 0 te (a) (b) Fig. 14.2 Downsampling by M = 2 (a) Input signal x(n) (b) Downsampled signal y(n) 14.2.2 Frequency-domain Characterization We now derive the relationship between the input and output of a downsampler in =-domain and frequency domain, By definition, the z-transform of y(n) is given as ¥@= Do yar = Ss {Mn)2-" = > afk)2-W/M (14.2) ‘This step is invalid because the downsampler will retain only those samples of x(k) that oceur at k = 0,-+M,+2M,--» and discarding the express Y(2) in terms of X(2) using Eq. (14.2). ‘To overcome this problem, the downsampling can be done in two steps: (a) obtained a discrete-time sampled signal x,(n) by multiplying 2(n) by an impulse train s(n) and (b) the downsampled signal is obtained by simply leaving out the M — 1 zeros between ct sample (Section 2.12.1). igure 14.3(a) shows a discrete-time signal x(n), Figure 14.3(b) shows a discrete time impulse train, The signal «,(n) shown in Fig. 14,3(c) is obtained by discrete-time sampling, that is, by multiplying (n) by s(n). . Therefore, we cannot directly (143) a(n) = 2(n)s(n) = san nm OM ADM, ‘The downsampled signal in Fig. 14.3(d) can then be obtained ‘by simply leaving ot the M —1 zeros between each sample, y(n) = a4(Mn) = 2(Mn), n= 0,1,42,..+ (ua) a a 2 Now, Y(2)= D> ulnen™= D> atumerm = FS a(ayerve 0) 7 umes no oe ‘This step is valid because «,(k) is defined for all values of k as given in Eq. (14.3), ‘a Ye= meh = xe) ane keneoMul ultirate Digital Signa Processing 901 x(n)902 Digital Signal Processing where wy = 2n/M, and the DTFS coefficient is Mat Ma . 18 ir 1 = ay S sine re LY sine i | mao mao ‘Thus, the periodic sequence s(n) can be expressed as 1M é. Wea es Ot > ea = tty tet imo where Wy = e~ # is the Mth root of unity. We can now obtain Xz) = Yo a(n)z-" = YO 2(n)a(n)e™ e sero ma00 Palle om(ie3 Swe im) -* eco i=0 1M ison aE tae (Sx (Wk) ) = 37 Substituting Bq. (14.10) in Eq. (14.6), we get es Y@=7 = x (2/" wh) Substituting 2 =e, we ee ¥(e) =e zs X (elt e944) = For M = 3, we get 2 2 rie $B (sem) a =o iMe!) o 0% * in 3x 4x Sa 6x 393 ® (es ith aliasing is shown in Fig. 14.5(a)~ eee ia beniimiiel bla z stretched versions with overlapping, aye : es. ding frequency, that is, F< f, reduced sampling rate. After becomes T! = MT,, and the904 Digital Signal Processing Xe!) See a) ue8) ange) et eae ay ek TE) TSR Oe aa a ie x la BE Ey EH 3K 2g 4a ean oe ot (b) Hei”) o 4a Je BE SQ ne Se Oe 4x re © Fig. 14.5 Effect of downsampiing in frequency domain with M-— 3 in the presence of aliasing new sampling rate is F; = $. Hence, the folding frequency after downsampling becomes 2 = dig. Thats, after downsampling by a factor of M, the new folding [ee eee i ‘ frequency component F > 4. Therefore, aliasing due to a factor of M downsampling ca __beavoided if the signal is bandlinited to F < %, that is, F < Fi, or, 0 = 2xF = "f-Multirate Digital Signal Processing 905 w(n) ial Fig. 14.6 Decimator consisting of ant-alias fier and downsampler 14.2.4 Anti-aliasing Filter Specifications sume that z(n) has been obtained by sampling a continuous-time signal «,(¢) using sampling interval T, (or sampling rate F,). Then, 2(n) = aq(nT,), and let zat) —~ (0). The Discrete-time Fourier transform (DTFT) of 2(n) is given by [Eq. (2.120)] (0) erie w _ 2nn x(e (¢-%) (14.14) “eo Similarly, we can consider that the decimated signal y(n) has been obtained by sampling the same continuous-time signal a(t) using a sampling interval T; = M7, (or sampling rate F! = 4+). Then, y(n) = 2a(nT;) = a(nMT,). The DTFT of y(n) is given by [Eq. (2.120) wy Lis w I) 1 = worn Let r= k+nM, where —co
x (een) 50. = 5%) +x) = 5 [*(e7) + xe] =$[x(e2)+x(-2%)] 2 li ee aSeee “Tate {c) The DTFT of y(n) = 2(2n) = a?" u(n) ig as ve= Fo ue Se neere WT Multirate Digital Signal Processing 907 =x) [pon] » nextel) n=0,41,22 a(n) = 407) | n=xn=)"\'G) n= 041,421, a {2 (0 Otherwis Input sampling E Ourpar sampling = frequency =F frequency = F’ = LF, Fig. 14.7 Upsampler L, Here, 1 is an integer. Let 2(n) be the input and y(n) be the output of an fcr © npsampler- The output yu(n) is given as _ fa(g) 2= 0.4L, £21, re block diagram of upsampler is shown in Fig. 147 Upsampling is identical to the finexpansion property of z-transform (Section 6.9.8) ‘and DTFT (Section 5.5.5). rhe eampling rate of input 2(n) and output y(") of a” ‘upsampler is related as follows: Sampling rate of the input signal 2(n) = Fe, Sampling interval of 2(n) = Ts Sampling rate of the output signal y.(n) =F; = LF, Sampling interval of y(n) = Ts = For example, let x(n) be the input signal frequency F, = 10 samples/s, tad sxmpling interval = T, = 0.1 s) and the ‘upsampling factor Z = 2. The upsampled signal is y4(n) = 2(3) (with sampling frequency Fi = 2F, = 20samples/s, and sampling = 0.05 8). Therefore, the input signal x(n) and upsampled signal are (with sampling a(n) ={--, 2(-2), 2(-) 2(0) 2(1), 2), 2) 24) 2(3),--} nln) =2(B) = 6D 0, 2(-1), 4 (3), 0, 2(4), 0 (6), 77} (0), 0, 2(1), 0, 2(2) 0 of information. Figure 14.8(a) 2 is shown in oes not cause loss ed signal yu() for ae lenis epptororsni 20) # HT in Fig. 14.(0). The nae sae 1400) sos te se ame ee anole interval 1 he sep Hoga Fig. 143(0) t Fig. 14304) It ig gyi '8 evident that the upsampler 4 ri a(n). The ‘upsampl908 Digital Signal Processing y(n) i o't234 56 789 li2isiaisig—" &) O 12345678 @) Fig. 14.8 Upsampling by L = 2 (a) Input signal x(n) (b) Upsampled signal y,(n) x(n) 0 Pal) ieee eee n OleTeoresT:) or, t 0) a(n) ‘a . S. ot lads n eee t ©) Fig. 14.9 Upsampling by L = 3 (a) Input signal (b) Upsampled signal y,(n) (c) pap ea 14.3.2 Frequency-domain Characterization — We now derive the relationship between the input and output of an upsampler i domain and frequency domain. By definition, the z-transform of y(n) is givet 8° Yu(2)= DO wu(n)er” = s 2(f)s= 3 a(r)eas Multirate Digital Signal Processing 909 = >> ar)(z")* = x(24) (04.21) = &*, we get ¥a(e™) = X(e**) (14.22) por b= 3, we get Yale) = X(e™*) Figure 14.10(a)-(b) demonstrates the effect of upsampling for I = 3. Part (a) shows: the spectrum of the original signal. Note that this original spectrum X(e’~) has one spectral component in the interval (—x, x). The spectrum of the upsampled signal iz ae x 0k Oe an 6x910 Digital Signal Processing is shown in part (b). It is obtained by dividing each label on the frequeney axis py L=3. sampling compresses each spectral component; therefore, aliasing do not exist in upsampling. Note that the upsampled signal spectrum Y,(e”) has three spectral compo. nents in the interval (~, 7). However, X (e!) has only one spectral component. These (L-1) =8—1 =2 extra components are called images and the phenomenon is called imaging. These L—1 = 2 images must be filtered out by passing Y,(e!) through a loy.- pass anti-imaging filter with a cutoff frequency ¥ = %, as shown in Fig. 14.10(c). The spectrum Y(e!) of the filter output is shown in Fig. 14.10(d). The filtered output has one spectral component in the interval (—m, 7). The low-pass anti-imaging filter is also known as interpolation filter. In the time-domain, filter output y(n) = yu(n)*h(n). The effect is that the zero-valued samples introduced by the upsampler are filled with inter- polated values as shown in Fig. 14.9(c). The complete process, that is, upsampling and then filtering, is referred to as interpolation, This is shown in Fig. 14.11. The spectrum of the filtered output is given as ¥(e*) = Yu(el*) H(e!*) LPF x(n) vuln) Ae) tL ; 7p | Bain ue Fi=LE, frat Fig. 14.11. Upsampling followed by filtering (interpolator) The input output relation of interpolator in the time domain is given as u(n)= > a(k)h(n — kb) 14.3.3 Anti-imaging Filter Specifications Assume that x(n) has been obtained by sampling a continuous-time signal zq(t) using a sampling interval 7, (or sampling rate F,), Then, 2(n) = aa(nT,), and let aq(t) — X(Q). The DTFT of z(n) is given by (Eq. (2.120)] 1 1 < w 2nw =e as eee. (14.23) (x! n=) Th er. . i ) ‘ Note that X(e) is periodic with period 2m and X, (e%") is periodic wit iod T vote x periodic with period 7 Similarly, we can consider that the interpolated signal y(n) has been obtained by 54 pling the same continuous-time signal a(t) using a sampling interval T! = 2 (° sampling rate Fy = LF,). Then, y(n) = 24(nT!) = x(n). The DTFT of y(n) is 8°? by (Ea. (2.120)] hence QT: oo Cd =a (HB) ard Sx i (420 Note that Y(e*) is periodic with period 2; Therefore, ¥ (el) £ X(eieL), These interval (—z, ). (€L) is periodic with period 7: spectrum X(ei*") has (L~1) extra images in th? Bie a:Multirate Digital Signal Processing 911 23) and (14.24), we can see that the spectrum X(e/”) has @ sealing h is inversely proportional to the sampling interval T,, and Y(¢”) oe = L,, which is inversely proportional to the sampling interval Ty. + hand, using upsampler, we get (Eq. (14.22)] ov) = x(a) Joes not change the scaling factor and it inserts (— 1) extra images in ie x. 7). If we pass upsampled signal y,,(n) through a LPF with gain = ca & frequency = #, the output of the LPF will be y(n) and its spectrum Y(e?*) is in Eq. (14.24). Therefore, the magnitude response of the anti-aliasing filter is jad ee (14.25) 0. otherwise 5 xe”) ssider the sequence x(n) with Fig. 14.12. Let y(n) = x(2n). recover 2(n) from y(n) using filters ey er blocks. a Ye”) 3 mn ee zon 30” ) i) o in tg ee TBR © c He!) ete 5%. 2x z. @) ore) eee YS ae gl ge Ox co} for Example 14.3 (a) I ee ()Decimsted signal specrum ) (c) Upsampled signal spectrum V(o c rapes He} aL ‘scheme is shown in Fig. 14.13. It works the frequency domain, the output912 Digital Signal Processing V(e%) of the upsampler is a compressed version Figure 14.17(a)-(e) demonstrates the of ¥(e™). By using a LPF H(e!“), we can, there- (n) in frequency domain, fore, eliminate the images and extract the original spectrum X(e) wn) = BPF x(n) 2 x(2n) to vin) He) V(e*) = ¥(e™*) and G(e) = V(e*)H(e*) | a i TecoverY of Bandwidth = Figure 14.14(a)-(e) demonstrates the recovery of centre freq = 5 (n) in frequency domain, 6 —_——_ Fig. 14.16 Recovery scheme of x(n) from its Example 14.4 Consider the sequence 2(n) with decimated version y(n) = x(2n) for Example 14.4 X(e’*) as shown in Fig. 14.15. Let y(n) = 2(2n) be the downsampled version of x(n). Show how we can Ne") recover =(n) from y(n) using filters and multirate building blocks, : Solution: i o = Se de in Note that input signal x(n) is a complex-valued 6 o band-pass signal whose bandwidth is and the @ downsampling factor M = 2. In frequency domain, Nel" the output of the downsampler given as os 1 ¥(e%) => x(ee9 7) & = 5 [Xe +x(e*)] =5[X(e8)+xC2)] Xe!) 2. ee Bet 6k es, Fig. 14.15 Input signal spectrum X(e!) for ‘example 14.4 2 V(e*) = ¥(e), and Gle) = V(e)H(e%) bs- =“ Multirate Digital Signal Processing 913 , SAMPLING RATE CONVERSION BY A RATIONAL FACTOR jj \i have discussed, so far, about the decimation and interpolation where the sarapling rate conversion factor is an integer (such as M or L). Now, we consider the sampling rate conversion by @ rational factor 4. The basic approach is to first increase the sampling rate by L and then decrease it by M. In other words, the sampling rate conversion by 4, is achieved by a cascading factor of L interpolator and a factor of M decimator as shown in Figure 14.18(a). The sampling rate of the output signal y(n) given as This scheme (interpolation followed by decimation) is preferred for the following reasons: (a) The interpolation is done first in order to preserve original spectral characteristics of a(n). (b) The two LPF H,,(z) and Hg(z) can be combined into a single equivalent LPF “with system function H(z) = H,(z)Ha(z) because they operate at the same sampling The simplified scheme is shown in Fig. 14.18(b). The frequency response of the quivalent LPF is given as Z (ol min (Es Hi) (14.26) He { 0 otherwise916 Digital Signal Processing oe Solution: | car Fig. 14.23, we get ). From ay; (n) + bya(n) Vi(z) = X(2°) Since ya(n) = ayi(n) + bya(n), the system is linear Time-invariant | ‘The input 2(n) and output y(n) of the system are related Vo(2) = 27% Vi(z) = 27" X (25) as ale ¥(2) = 5 Vaz! wht) yln) = 2 a(r)g((n=r)L) kno For a system to be time-invariant, if m(n) = 21(n - =x} = a-N/S yeu ng), then ya(n) = y(n ~ no). Thus, consider bo . n(n) = S m(rja((n- nL) In time pee we get roa a(n) = 2(5) Bam io— m)e((n= 794) a(n) = 11(n) g(n) = o1(n) #6(n.— ¥) n-N Y alh)a((n - k= m9)L) = n(n — m0) oda 19 =2(25%) ke 00 (n) = m(6n) =2(®=%) 22 (n-F) Since yo(n) = y(n — no), the system is time- ee pau ers een aa) 5 invariant. x(n) y(n) PO) (n) Example 14.7 In the system of Fig. 14.23, find vi in te of X(z). In addition, find Se ae i) pee Fig. 14.23 Structure for example 147 14.5 IDENTITIES (CASCADED EQUIVALENCES) To realize a computationally efficient multirate system, we need to swap the postion a filter or a multiplier or adder with a downsampler or an upsampler. The pro?" that govern these operations are called noble identities, 14.5.1 Identities for the Downsampling Some identities for the downsampling of signals are as follows: First Identity Downsampler commute with addition and multiplication. his" in Fig, 14.24. The outputs y(n) and ya(n) are given aa n(n) = yo(m) = a (Mn) +b 2(Mn)Multirate Digitat Signal Processing 917 xm @ Jim) a pain) oo Soe @ gt b) Fig. 14.24 First identity y; n) a 0 Pe mpe? = fa ® ©) Fig. 14.25 Second identity jay of M sampling intervals before a downsampler is the same as the delay of one sampling interval after the downsampler, This is shown in Fig. 14.25(a) and (b). _ Consider Fig. 14.25(a), we get Vi (2) ae £ ee Ma zs i = Ne = 9 ne Wi) =F Ss ee Wet Xe wey = = ag LSS gM ye Z 1 aoe DSC wk) = consider Fig. 14.25(b), we get the second identity. It is possible to swap ‘if we modify the system function of the918 Digital Signal Processing a y(n) yi) xn) vin) xt Es _a- (a) o) Fig. 14.26 Third identity Now, consider Fig. 14.26(b), we get SS xe wh) ; Vale) = 7 Yo(2) = Va(z)H(2) =918 Digital Signal Processing “Fs: x) [in| dem) (b) Fig. 14.26 Third identity Now, consider Fig, 14.26(b), we get 1M2 Vale) = 37 x X(2 wh), 1M Yo(2) = Va(2)H(2) = 9p YX" Wi) HG) = 4) imo 14.5.2 Identities for the Upsampling — Some identities for the upsampling of signals are as follows: ; Fourth identity Upsampler commute with addition and multiplication. ‘This is shown in Fig, 14.27. The outputs yi(n) and yo(n) are given as w(n) = (7) 2 a(n) = wn) =02(F), a(n) =Multirate Digital Signal Processing x(n) i} » i-a— () Fig. 14.28 Fifth identity Ml yy) vi) in ce a oi * fe fae” @ &) Fig. 14.29 Sixth identity Consider Fig. 14.29(a), we get Vaz) = X(2)H(2) ¥a(@) = Vaz") = X(2)H(") Now, consider Fig. 14.29(b), we get Volz) = X(z") Yala) = Vale) H(2") = X(2)H4) = HE) ‘Solution: that L=3, M =2, and k=0,1. Therefore, Note that for = 3 and M = 2, Wir920 Digital Signal Processing From Fig. 14,30(b), we get a(n) = {1, 2,8, 4, 6 va(n) = 2(2n) = {1, 8, 6} n va(n) = (3) = (1,0, 8, 0, 6} # m(n) Now, if L=3 and M = 2, that is, both L and M are re any input x(n), we get from Fig. 14.30(a) a(n) = {t, 2, 8, 4, 6} w(n) =2(5) 3 (nr) = 1 (2n) = {1, 0, 0, 8, 0, 0, 6} {1,0,0,2,0,0,8, 0,0, 4,0 From Fig. 14.30(b), we get #(n) = {1, 2, 8, 4, 6} (n) = 2(2n) = {L, 8, 6} n w2(n) = » (5) = (1,0, 0,8, 0, 0,6) Therefore, the two structures in Pig. 14.30 ar relatively prime, that is, L and M do not than 1. Consider Fig. 14.30(a), we get Vi(z) = X(z4) Mat : Yi(z) = + XY vie wh) = fo Now, consider Fig. 14.30(b), we get oa Vale) = 3p XW) a i i Fx Ya(z) = Va(z") = = ie) = Valet) = a It follows from Bgs (14.29) and (14.30) Mat 2 XG 0 The equality holds if Wit = Wy and the two structures in Fig.> Fas Wer noble identities to simplity the Fig, 14.82(c), the upsampler of factor 4 is brought joomle 149 J gi(a) and find Y(z) in terms of ‘the way to the end and the filters are changed Poo oe find y(n) in terms of 2(n). ing to fifth identity. In Fig. 14.32(d), the upsamm i ition of factor 4 followed by the downsampler of factor BS sl 4 cancel each other. iol 431(0), the downsampler of factor 2 is shead of filter (given by 27) and the fil- ett according to second noble identity. In . the upsampler of factor 2 followed by Multirate Digital Signal Processing 92% Example 14.11 Use noble identities to simplify the system of Fig. 14.33(a) and find Y(=) in terms re HC iplr of factor of 2 cancel each other. In of X(z). In addition, find y(n) in terms of x(n). es weet Solution: r “1x (s) In Fig, 14.33(b), the upsampler and downsampler ¥(e) have interchanged. In Fig. 14.33(c), the upsampler sine domain, we get y(n) = 2(n 1), is brought ahead of filter (given by 2~*) and the fil- ter is changed according to fifth noble identity. From an) p r 2 wn) Fig. 14.33(c), we get Vaz) = X(2°) (a) Va(z) = 271? Va(z) = 27? X(2°) stn) eel n) 1s tr } 7 Y(2)= 4 rae wh) ae 12/4 we1Bk x (29/4 wi) xn) = ; 28x (25/4 we") 2. In time domain, we get in time domain, we get Fig. 1431 Structures for Example 14.9 is as win =a(2) i. ple 14.10 Use noble identities to simplify v(m) = (nm) * 9(m) = v1(n) + 5(m — 12) ‘stem of Fig, 14.32(a). n—12 4 a i =n(n=12)=2("5 = wn) * ) ; id t=} fee} ft] Sea @) ©) x(n) on) "tea @ () ton, Structures for Example 14.10 xu) YC) v2) An) of factor 2 is © yy 1+ 22) and the fil- d noble identity: In Fig. 14,33 Structures for Example 14.14922 Digital Signal Processing n=0,4L,421, Example 14.12 In the system of Fig. 14.34, find ¥(z) in — of X(z). In addition, find y(n’ terms of 2(n in 0 otherwise ‘Therefore, in time domain, x(n) |vi(r) | wn) S ” 4 ae 5 as Wz *z(n) = 2( iso ie i 7a e Fig. 14.34 Structures for Example 14.12 a(n) n=O4 1,421.0 ~ (0 otherwise Solution: From Fig. 14.34, we get Example 14.13 In the system of find Y(z) in terms of X(z). In additie terms of x(n). Vile) = Solution: Bele We) In Fig. 14.35(b), the downsampler of tored into two downsamplers of f yet Fig. 14.35(c), the upsampler of pt] = 5 >> X(we) the downsampler of factor of 3 cai i= Using the results of Example 14.12 4 ‘We know that vee z y x(ews) pin el a and [Bptlasa(n) — x(=) a =; W™ un) = 5 We a(n) {Spt](Wr*)"2(n) — x(—= a(n) n=045, 4 1 We 2(n) i Sx (ewe) ee. io vuln) ¥(@) Thus, y(n) ne a(n) = a(n) > eer Le Consider the summation a ts n=0,+L,+2L,--- eee & =F otherwise {: n=0,4L,+42L,-- 22S, otherwisewe © “Se ,OMPUTATIONAL REQUIREMENTS Multirate Digital Signal Processing 923 ion or interpolation filter can be either finite impulse response (FIR) or an infinite impulse response (IIR) digital filter. Infinite impulse response filters are omputationally more efficient than the FIR digital filters for single-rate DSP, whereas FIR filters are computationally more efficient than the ITR digital filters for multirate DsP. 14.6.1 Efficient Direct Form Structure of a Decimator ig. 14.36(a). The decimation filter Then, ‘The low-pass decim ator as shown in Consider a factor of M = 2 de H(z) is an FIR filter of order N = 2 (or length = N +1 = N 2 v(n) = So A(k)2(n — kb) = SO A(k)2(n — b) =o to = A(O)x(n) + A(1)x(n — 1) + A(2)a(n — 2) Let the input sampling rate be F, = 4 samples/s. Then, Number of multiplications per second (MPS) = (NV +1) x Fy =3 x 4 = 12 simple but computa omputed at the input structure The direct form filter is shown in Fig. 14.36(b). Thi tionally inefficient. This is because the filter output v(n) must be F, EB Fy r xm) (0) {ny z wn) ACD) @ © Decimator (b) Direct form structure followed Fie, 1496 tices poeta tata (Oe Yor N= ard M = 2 (a) An efficient ‘Structure for N and M_924 Digital Signal Processing sampling rate and we keep every Mth sample only to get y(n) = v(Mn) = u(2n). Since the filter output is a function of the current input and past inputs only, the outputs that are ignored by the downsampler need not be computed. We can avoid the unn calculation of the values v(n) by using the first identity shown in Fig. 14.24, ‘This first identity is used to develop the structure shown in Fig. 14.36(c) for M = 2. Interchanging the order of operations between the multipliers and the downsampler results in the multiplication operation being performed at the changed rate. Now, Fy 4 Number of multiplications per second (MPS) = (N +1) x 4p =8% 5 =6 ‘Thus, in the case of FIR filter, we save the computations by a factor of M. Figure 14.36(d) shows the efficient structure for any value of N and M. Now, assume that the decimation filter H(z) is an IIR filter of order N = 2 with system function a ou vu) X(z) H(z) = = 1+) ope-* it Let the input sampling rate be F, =4 samples/s. Then, Number of multiplications per second (MPS) = N x F, + (N +1) x Fy =(2x4)+(8 x4) = 20 Its direct form II implementation is given as —a,w(n— 1) ~ agw(n— 2) + 2(n) dgw(n) + byw(n = 1) + b2w(n — 2) ‘The direct form II structure is shown in Fig. 14.37(a), which is computationally ineffi- cient. Since w(n) is a function of its past values and current input, it must be computed for all values of n. For IIR filters, the feedback portion cannot be commuted with the downsampler. However, we keep every Mth sample of the filter output u(n) to get y(n) = v(Mn) = v(2n). Since the filter output v(n) is a function of the current and past values of w(n) only, the outputs that are ignored by the downsampler need not be924 Digital Signal Processing sampling rate and we keep every Mth sample only to get y(n) = v(Mn) = v(2n). Since the filter output is a function of the current input and past inputs only, the outputs that are ignored by the downsampler need not be computed. We can avoid the unnecessary calculation of the values v(n) by using the first identity shown in Fig. 14.24. This firs identity is used to develop the structure shown in Fig. 14.36(c) for M = 2. Interchanging the order of operations between the multipliers and the downsampler results in the multiplication operation being performed at the changed rate. Now, F, 4 mM °*3 ‘Thus, in the case of FIR filter, we save the computations by a factor of M. Figure 14.36(d) shows the efficient structure for any value of N and M. Now, assume that the decimation filter H(z) is an IIR filter of order N = 2 with ‘system function Number of multiplications per second (MPS) = (NV + 1) x 2 A= YO - an X@) z 14 Sart Let the input sampling rate be F, = 4 samples/s. Then, Number of multiplications per second (MPS) = N x F, +(N +1) x Fy = (2x 4) + (3x4) =20 Its direct form II implementation is given as w(n) = —a,w(n — 1) ~ agw(n — 2) + 2(n) v(n) = bow(n) + byw(n ~ 1) + b2w(n — 2) The direct form TI structure is shown in Fig. 14.37(a), which is computationally ineff- cient. Since w(n) is a function of its past values and current input, it must be computed for all values of n. For IIR filters, the feedback portion cannot be commuted with the downsampler. However, we keep every Mth sample of the filter output v(n) to get y(n) = v(Mn) = v(2n). Since the filter output v(n) is a function of the current and past values of w(n) only, the outputs that are ignored by the downsampler need not be (a) Fig. 14.37 IIR Decimator structures (a) Direct form II structure followed by a downsampler (b) Structure obtained by using the first identity for N — 2 and M - 2
You might also like
Dsp 5th unit part 1
PDF
100% (1)
Dsp 5th unit part 1
53 pages
DSP Important Questions Unit-Wise
PDF
100% (7)
DSP Important Questions Unit-Wise
6 pages
Signals & Systems 2016-23 - 8 Years PYQ - @yuvra - Jsingh
PDF
0% (1)
Signals & Systems 2016-23 - 8 Years PYQ - @yuvra - Jsingh
14 pages
Explain Generation and Demodulation of PAM, PPM and PWM With Waveforms
PDF
100% (1)
Explain Generation and Demodulation of PAM, PPM and PWM With Waveforms
6 pages
Narrowband Noise
PDF
100% (3)
Narrowband Noise
5 pages
AM Generation and Detection Methods
PDF
100% (3)
AM Generation and Detection Methods
26 pages
SS Notes - VMTW
PDF
No ratings yet
SS Notes - VMTW
67 pages
Filter Realization
PDF
100% (1)
Filter Realization
67 pages
Multirate Signal Processing
PDF
No ratings yet
Multirate Signal Processing
147 pages
Low Pass Sampling Theorem
PDF
No ratings yet
Low Pass Sampling Theorem
24 pages
Unit-5 Multirate Updated
PDF
No ratings yet
Unit-5 Multirate Updated
83 pages
Narrowband and Wideband FM
PDF
100% (2)
Narrowband and Wideband FM
18 pages
PTSP Notes Final
PDF
No ratings yet
PTSP Notes Final
57 pages
PROBLEMS - DIT FFT Part 3
PDF
No ratings yet
PROBLEMS - DIT FFT Part 3
10 pages
Digital Signal Processing by Nagoor Kani PDF
PDF
No ratings yet
Digital Signal Processing by Nagoor Kani PDF
817 pages
Advanced Digital Signal Processing
PDF
No ratings yet
Advanced Digital Signal Processing
71 pages
Multirate_part1
PDF
No ratings yet
Multirate_part1
13 pages
D.S.P Book N.G Palan PDF
PDF
79% (14)
D.S.P Book N.G Palan PDF
692 pages
Multirate DSP Part 1 (Read-Only)
PDF
No ratings yet
Multirate DSP Part 1 (Read-Only)
38 pages
Multi Rate Signal Processing
PDF
No ratings yet
Multi Rate Signal Processing
21 pages
Examples On Radix-2 DIT-FFT Algorithm 1) Use The 8-Point Radix-2 DIT-FFT Algorithm To Find The DFT of The Sequence X (N) (0.707,1,0.707,0, - 0.707,-1,-0.707,-0)
PDF
No ratings yet
Examples On Radix-2 DIT-FFT Algorithm 1) Use The 8-Point Radix-2 DIT-FFT Algorithm To Find The DFT of The Sequence X (N) (0.707,1,0.707,0, - 0.707,-1,-0.707,-0)
2 pages
What Are Twiddle Factors
PDF
No ratings yet
What Are Twiddle Factors
5 pages
Prep For TI - Digital (With Sample Questions)
PDF
No ratings yet
Prep For TI - Digital (With Sample Questions)
7 pages
Digital Signal Processing by J.S. Katre (Tech Max) PDF
PDF
100% (6)
Digital Signal Processing by J.S. Katre (Tech Max) PDF
964 pages
Module 5 - MWA-18EC63
PDF
No ratings yet
Module 5 - MWA-18EC63
45 pages
Signals and Systems Lab Manual Print PDF
PDF
100% (2)
Signals and Systems Lab Manual Print PDF
27 pages
GATE Signals and Systems by Kanodia
PDF
83% (6)
GATE Signals and Systems by Kanodia
128 pages
UNIT - 7: FIR Filter Design: Dr. Manjunatha. P Professor Dept. of ECE
PDF
No ratings yet
UNIT - 7: FIR Filter Design: Dr. Manjunatha. P Professor Dept. of ECE
94 pages
Unit 4 DFT and FFT
PDF
No ratings yet
Unit 4 DFT and FFT
42 pages
Topic 5:: Multirate Digital Signal Processing
PDF
No ratings yet
Topic 5:: Multirate Digital Signal Processing
27 pages
R-22 Eca Lab Manual
PDF
No ratings yet
R-22 Eca Lab Manual
59 pages
Experiment No. 04 Experiment Name-QPSK Modulation & Demodulation Aim Equipments Needed
PDF
No ratings yet
Experiment No. 04 Experiment Name-QPSK Modulation & Demodulation Aim Equipments Needed
10 pages
DSP Question Bank With Solutions
PDF
73% (22)
DSP Question Bank With Solutions
46 pages
2-Mark Questions Anna University Signals and Systems
PDF
100% (2)
2-Mark Questions Anna University Signals and Systems
11 pages
Digital Signal Processing Two Marks
PDF
No ratings yet
Digital Signal Processing Two Marks
11 pages
Gate Question LIC
PDF
No ratings yet
Gate Question LIC
46 pages
Ec8562 Digital Signal Processing Lab
PDF
100% (3)
Ec8562 Digital Signal Processing Lab
73 pages
Questions & Answers On Discrete Fourier Transform - Properties and Applications
PDF
50% (2)
Questions & Answers On Discrete Fourier Transform - Properties and Applications
13 pages
DSP Notes Unit1 and 2
PDF
No ratings yet
DSP Notes Unit1 and 2
45 pages
Transistor at Low Frequencies
PDF
100% (2)
Transistor at Low Frequencies
40 pages
Transistor H-Parameter Analysis
PDF
70% (10)
Transistor H-Parameter Analysis
51 pages
Linear Filtering Methods Based On The DFT
PDF
67% (3)
Linear Filtering Methods Based On The DFT
50 pages
Ecao912 Digital Signal Processing (Gate Questions)
PDF
No ratings yet
Ecao912 Digital Signal Processing (Gate Questions)
14 pages
Analogy BW Vector ND Signals
PDF
50% (2)
Analogy BW Vector ND Signals
7 pages
Experiment 6 Implementation of LP Fir Filter For A Given Sequence
PDF
0% (1)
Experiment 6 Implementation of LP Fir Filter For A Given Sequence
25 pages
MDSP 1
PDF
No ratings yet
MDSP 1
48 pages
Emtl Scan Notes
PDF
96% (77)
Emtl Scan Notes
153 pages
DSP Lab Sample Viva Questions
PDF
No ratings yet
DSP Lab Sample Viva Questions
10 pages
FIR Filter Design
PDF
No ratings yet
FIR Filter Design
19 pages
Pce Mini Project Report: "RF Transmitter and Receiver"
PDF
No ratings yet
Pce Mini Project Report: "RF Transmitter and Receiver"
9 pages
Optimum Linear Filter (Wiener Filter) Tarun Kumar Rawat
PDF
No ratings yet
Optimum Linear Filter (Wiener Filter) Tarun Kumar Rawat
26 pages
Selected Answers CH 9-13
PDF
100% (1)
Selected Answers CH 9-13
15 pages
8051 Lab Dac
PDF
No ratings yet
8051 Lab Dac
6 pages
Electronic Voting Machine Based On Biometrics.: Be Project Group-2 Project Guide: Mrs. Neeru Pathak
PDF
No ratings yet
Electronic Voting Machine Based On Biometrics.: Be Project Group-2 Project Guide: Mrs. Neeru Pathak
14 pages
Experiment - Amplitude Shift Keying
PDF
80% (5)
Experiment - Amplitude Shift Keying
3 pages
Linear Convolution Program in C Language Using CCStudio
PDF
80% (5)
Linear Convolution Program in C Language Using CCStudio
3 pages
Fir-Iir Filters Viva-Questions PDF
PDF
No ratings yet
Fir-Iir Filters Viva-Questions PDF
5 pages
Balanced Modulator Using Diodes
PDF
No ratings yet
Balanced Modulator Using Diodes
5 pages
Biometric Based Electronic Voting System: International Research Journal of Engineering and Technology (IRJET)
PDF
No ratings yet
Biometric Based Electronic Voting System: International Research Journal of Engineering and Technology (IRJET)
5 pages
Project Management
PDF
No ratings yet
Project Management
23 pages
Viva Questions For Analog Communication Lab
PDF
100% (1)
Viva Questions For Analog Communication Lab
2 pages
Lattice Structure For FIR Filter
PDF
100% (1)
Lattice Structure For FIR Filter
10 pages
DSP Viva Questions PDF
PDF
No ratings yet
DSP Viva Questions PDF
4 pages
DSP Lab Model Questions
PDF
0% (1)
DSP Lab Model Questions
3 pages
EXTC Advanced Digital Signal Processing
PDF
No ratings yet
EXTC Advanced Digital Signal Processing
7 pages