HILBERT TRANSFORM—CONTINUOUS AND DISCRETE TIME i Way coaeh cod After going through this chapter, students will be able to * define the Hilbert transform, explain its frequency-domain interpretation and also state and Prove its properties, define the analytic signal (or pre-envelop) of a given real-valued signal x(t) and sketch its magni- tude response, define the complex-envelope of a given bandpass signal, explain its meaning and its usefulness in bandpass signal /bandpass system representation, state and prove the relationship between the real and imaginary parts of the transfer function of a causal LTI system, and define a discrete-time Hilbert transformer in terms of its ideal frequency response and derive its unit sample response sequence h(n). 11.1 INTRODUCTION ‘We have already studied various types of transforms on continuous-time signals and discrete-t the Laplace and Fourier transforms on continuous-time signals and the DTFT and Z-transforme on the discrete-time signals. The transformation of a signal using any of these transforms brought about « change of variable—from 1 to sin the case of Laplace transform, from to fin the case of Fourier transform, from n to in the ease of the DTFT and from nto in the case ofthe Z-transform, Ther i always a change of the domain.ae arent rari [36] The Hilbert transform differs from all = riwous and Discree-time [561] time, oF, di the = eal pees does not bring eee the sense that Hilbert transforming a signal, continu- s ntinuous or discrete time, a change of domain, The Hilbert transformed signal also is @ I oti ‘case may be). As we will be seeing later in this chapter, when a " ane nty components suffer a phase-shif of -1/2 radians, while their This property ofthe Hilbert tansfers ilbert transformer is an all-pass 90° phase shifter. sentation of bandpass signals, espe, ig ‘makes it highly useful in a number of applications—in the repre~ deta cfiguals ail tye, pe ly the single-sideband signals; in the bandpass-to-lowpass transfor modulators, wherein, we ixagloy wae implementation of certain modulator circuits like the phase-shift ‘qanoemess: ind 90° phase-shifters, which are approximations of ideal Hilbert = ero AND FREQUENCY-DOMAIN INTERPRETATION Hilbert transform, £(1), of a signal x(0), is defined as the signal obtained by convolving x(#) with I/nt. 8) A xyet LD) mt p x(0) 1-1) ay = | aoa FAS i ay Note Although we have not explicitly stated any conditions to be satisfied by the signal, x(t), it may be noted that the above definition ofa Hilbert transform is applicable tall signals that are Fourier rar formable. Further, since all applications of Hilbert transform are concerned with real-valued signals, wwe shall assume in all our discussions that x(t) is real valued. 1 is best understood by looking at it in the frequency domain. We ‘The effect of Hilbert transforming a signal is bes g “ shall therefore take the Fourier transform of both sides of Eq, (1.1), and invoke the convolution theorem of Fourier Transform. If X(f)=F [&(0)], we then have * 1 Rf) = XN) [3] - (113) From Section 4.7, we know that F (e000) = Fp Hence, from the duality theorem, we Dave Fis an odd function off.) 1] 2 sqnt-f) = sem (since san F trl562| signals and systems : Ley pa it t Substituting this in Eq, (11.3) we get X(f) = ~ isan XP) - (115) : 1 for f>0 si A tne, RO 1 soe pie Equation (11.5) implies that one in -[ 2. (11.6) IX(f) 3 F< Since X(/) is the spectrum of £(t) and X() is the spectrum of x(, it follows that the effect of Hilbert transforming 2 signal x(0) is merely to give a phase shift of -1/2 to all ofits positive frequency components alter the magnitude spec- and a phase shift of +7/2 to all of its negative frequency components. It does not trum of x(¢) in any way, since we find from Eq, (11.5) that [ky] = OI .L7) Itis interesting to note from Eq, (11.1) that the Hilbert transform (1), of a signal x(1), may be obtained by giving x(#) as input to a Linear time-invariant system T whose impulse response function, h(2), is given by 1 iO - (18) Obviously, the ideal Hilbert transformer is not causal and hence is not physically realizable. However, it can be approximated. xo T Such a system T with impulse response (1) = I/nt, is therefore MO = tt called a ‘Hilbert Transformer’, since it produces at its output termi- open nals, a signal which is the Hilbert Transform of the signal given as input to it. A Hilbert transformer, therefore, has a transfer function, Fig. 11.1. Hilbert transformer H(f), given by 1 nin [2] =~ inn 2 (119) LH |t WHC $ a 1 0 7 =r -al2 @ o Fig. 11.2 (a) Magnitude resopnse ofa Hilbert transformer (b) Phase response of a Hilbert transformerume (583) (ie., one for which Mbert Transform—continuous and Disere Equation (11.9). like Fan. (11-7), tells us that the Hilbert transformer is an all-pass filter Hap =1 forall frequencies), However, its phase response is as shown in Fig. 11,2 (b) 11.3 PROPERTIES OF HILBERT TRANSFORMS 1 with {s is evident from Eq, (11.1), since the Hilbert transform ofa signal x(1) is obtained by convolving ee {iat itis also a time-domain signal. Thus, unlike the Laplace and Fourier transforms, the Hilbert does not change the domain of a signal, 1. Hilbert transform does not alter the amplitude spectrum of a signal, ic. the signal «(?) ant Hilbert transform, (1) have the same amplitude spectrum. It changes only the phase spectrum by producing 4.-90° phase-shift for all frequency components. The fact that the amplitude spectrum is unaltered, implies that if x(r) is band-limited to W Hz then (¢) is also bandlimited to W Hz. 3 If i(:) is the Hilbert transform of x(¢), then the Hilbert transform of (1) is fay =-x@)- proof \f x(t)<~++ X(f), we have already seen that #(1)< £7 »— jsgn( f)X(f). Hence the Fourier transform of *(t) is given by 5) ADs ~ joan NX) isen oP XC) = xf) ++ (11,10) jicates that to recover the original signal x(¢) from its Hilbert transform, %(¢), we sim- ply have to take the Hilbert transform of £(r) and reverse its sign. Thus, if syed a. 4) = MY Vie pit Then, the inverse Hilbert transform is simply 3 ‘ + xt) = 8-8 at) = ~H() = (0° SG Ic (LA) The result given by Eq, (11.10) could have been obtained intuitively without recourse to any mathematical equations. If Hilbert transforming x(1) merely produces a -90° phase-shift, Hilbert transforming it a second time should produce —180° phase-shift, i.e., it should give us — x(t). 4.A signal and its Hilbert transform are orthogonal to each other (@) Let the signal x(t) be an energy signal. Then, we have to show that J x@tnar - 0 ‘We know that the generalized Rayleigh’s theorem (Parseval's Theorem) applicable to Fourier transforms, Ei Says J xtoy' nar = J XN Ur ete the * indicates complex conjugation.S64] sgnakondSstms —— If-x(0) is real, i(r) is also real valued, since it is after all obtained by convolving 0 with another real- valued signal 1/x¢, Hence, applying generalized Rayleigh’s theorem to x(#) and (1), We Jamin = Fans oar f xenx nar j XUN Licen OUD] = J f sem NXE a However, since sgn( f) is an odd function of ‘f, while [X(/)? is an even function, the integrand in the last integral is odd and hence the integral is zero. J xitnar =0 Hence, x(0) and i(¢) are orthogonal over the entire interval ~ = to + = (b) In the case of power signals, orthogonality condition is stated as if Hag | Moin =0 This also can be proved in the same way as in the case of energy signals. BEES Find the Hilbert transform of x(t)=sinedg1. | fren ay a[ dle HW @ v)] 3B f)-8F + fy} ~ssen XP) = XUN = 3 {8 ~fo)- 5S + fy)} sen f) 4 = Flours h)+87-K)) ‘Taking the inverse Fourier transform of the above on both sides, we get HO) = ~cos2x fot = —coseys ae Solution PRE Find the Hilbert transform of x(t) = cost, OU +h) +8 f.)] = = Seem fIX(P) XD) = STB + f0)+ 8 - fe emf)ibert Transform —commoonn 004 WAIL [xs] Ky af WS + $y) 4S -W) ag toe roverse Forcier temas on tots ses, Gt) = sin 2a fy ~ sining RIDE 0 1 site steal =~ — AD) = contagg + joining it) = coming + jong ~ 30) fe) Now Floonens) ~ UBS fy)+8I + ho) Hence. HS) = PAASVZOS Lo) I +500) 2 WLPO. iL = Has f-65 +0) Who. Et) =sinog Bat 4-0, KD) = FLASH 8S +50) 2 hfs, fe) ~ ~sineony et) = sing 2. HAN PLAS LO + fo] EAS) = PAD) FE) OS + HN) 2 hee 4D) ~ FS +f) 6-H) 2 if ro, i,t) = -coseng bat <0, GS) ~ 28S) +8 +0) 1 KO, ft) = coneng Hence, oP! = sina - joosens ~~ fleosens + jsintns}=— Je" , provided fy> 0 feet <0, eA = —simung + joostang + fcortag + jsinny |= je 7 wemmey write566 Signals and Systems = In communication engineering, we often come across modulated signals which may piel, = oy sented by the product of a lowpass signal like speech and a high-frequency signal, called the carrier signal Let the lowpass signal be denoted by x(¢) and the highpass signal by y(#). Then A= 310-0) + (11.12) ‘We shall presently consider the effect of taking the Hilbert transform of such a signal in which the spectra of the two signals are non-overlapping, ‘Taking the Fourier transform of Eq. (11.12) on both sides, we have ZP) = X(N*V(f) X(AY(f-A)da 2 UD) = F [HO]=- sen NZN ==) J XANS-A) sen faa and 40 = [4] if j X(AY(S -A)sgn( ye" dhaf = -if j X(Aje?™ Vf —Ael*UM son f) afar 2-11.13) Since x() is a lowpass signal, bandlimited to say, W Hz, the range of values of A for which X(A) is non-zero, are |A < W. But, (1), being a highpass signal, the range of values of ‘f for which ¥(/) is non-zero are typi- cally |f| >> W. Hence, in the integrations involved in Eq. (11.13), we will be interested in small values of the variable A and very large values of the variable ‘f°. Hence (f— A) may be approximated by f without any appreciable error and we may re-write Eq. (11.13) as 3) = ff xe -Y Ne jogn(N df dd = fF xbe aa. J De? joom nay = x0) () Hence, if (0) isa lowpass signal, (0) isighpass signal, and ifthe spectra of x) and (0) are non-overlapping, then HHO =H) HO + (L14) SUT REE #19 152 speech signal band limited toa frequency of W Hz, ie, XQ) = 0for ff >W Ifo the carier frequency is very much greater than Wind the Hilbert ransform ofthe modulated signal x(0) = x{0) cos Inf.Solution cos 2n¢ ; transform of this signal cai 1 We know, from Hence, from the result of Example 11.2 ang Fa. (11.14), we *O =20 cosday x@ sin ot Table 11.1 Hibert transform pairs S.No, Time function ~~ Hilbert transform 1 | cos amp Sin 2p 2 | sin ape ~c0s 2afee S| MO c08 2nf-15(0) is bandstimited to Wand f.>> Ww x) sin 2nfer 4 | Osinzzgeg os my “XG 008 2afit s. | ~ 25() 6 | Ginny (1 ~cos aye 7 | & a) 8 | va+ey +?) 114 ANALYTIC SIGNAL Let x(Q be a real-valued signal. Then we define its analytic signal or Pre-envelope as l, which has *(P) itself as the f x0), namely, $(t), a8 its imaginary part. An ‘we will be seeing in the mi nt sections, this concept of an analytic signal, or, Pre-envelope, plays an important role +, ie atc signals and in the analysis of bandpass syercer The usefulness of the pre-envelope stems from the perils feature of its spectrum. To investigate this, ‘et us take the Fourier transform of both sides of Eq. (11.15). Representing the Fouree transform of x, by %1D, we have KD) =X) +L jsant XY] XN+XSP) 5 f>0 UkKN-xX) + f<0 2X(f) for f>0 xo-{i for f<0 = 5 (0, the analytic signal of the real signal x0 will be twice Equation (11.16) tells us thatthe spectrum ot ed will be 2er0 forall ne IB: Positive frequency part of the spectrum 20) has its Spectrum X(/) with its magnitude as shown in Fig. 11.3(a), then the ‘ative frequencies, analytic signal, will be as shown in Fig. 11.3(b) with only the Hence, if magnitude of the spectrum of Posive frequency par of XY) burt the (0), itsSignats and Systems 7 ny t ot ©) Fig. 11.3 (a) Magnitude spectrum of x(t) and () Magnitude spectrum of x,(t), the analytic signal of x(t) magnitude being twice. The shape of [X()] shown in Fig. 11.3(a) has no particular significance and is completely arbitrary, except that it has got to have even symmetry since x) i a real-valued signal. By now, the reader might have realized the significance of the subscript + in the symbol x,(t) used for de- ‘noting the pre-envelope, or analytic signal, as defined in Eq, (11.15). It is because its spectrum has a similar Shape as that of x0) only fr the positive frequencies, It is for this reason that x,() is sometimes referred vo as the positive frequency pre-envelope. tis clear that we can also define a negative frequency pre-envelope x (t) as follows. X04 x0)- 72) (1.17) Obviously, x)= (0) eee (i) where, * indicates complex-conjugation. If x(0) has a magnitude ‘Spectrum as shown in Fig. 1 1.3(a) then x (1), the ti of x() will have a magnitude spectrum as shown in Fig. 11.4, ), the negative frequency pre-envelope benny + —r FlG-114 Magnitude spectrum of x(t)569 Ite ronrm- comand arate [309] By taking the Fourier transform on both sides Of Eq. (11.17), we get Pq. (11.17), we ge ae {oe r>0 119% 2X(S) for f<0 Ie may be noted that although 340) has been asa Figs. 11.3(a), (b) and 11.4, it; ee bbe a lowpass signal for the sake of drawing the "an, in fact, be any real- rier transformable. ‘valued signal, lowpass or bandpass, provided it is Fou- 11.5 COMPLEX ENVELOPE AND BANDPASS SIGNAL REPRESENTATION tain frequency f~ called the centre frequency. Wis generally very small compared to the center frequency f- and therefore they are narrowband signals. For atypical amplitude modulated audio broadcast signal, 2 is 10 kHz while f- may be typically of the order ofa few megahertz. Let x(9 be a real bandpass signal centered on f, and with an amplitude spectrum as shown in Fig. 11 5(a). From Figs. 11.5 (b) and (c), itis clear that 3(1) is a lowpass signal of bandwidth W whose spectrum ¥( f) is obtained by shifting X,(/) to the left by an interval of frequency equal tof. Thus, . RN = Xf) (11.20) ‘Hence, from the modulation theorem of Fourier transforms, we know that HO) = x,(e 4 +121) X(Q= (De +++ (122) Since x. (9) = x(0) + J8(0) , we have, x() = Refx. (#)] = Ref i(t)e/*/] 11.23) Because of the above equation, i() is called the complex envelope of the bandpass signal x(t).-The reason for calting is valued lowpass signal 3() as the complex-envelope of the real bandpass signal x(*), isas follows. Suppose x(1) = a(t) cos [@2,t + (9), a real-valued bandpass signal, where a(t) and @(¢) are real- valued lowpass signals. Then we know that x(t) = a(#) 60s [at + O(0)) = Re{ faire} erm] ee In Eq, (11.24), (a(t) &%} is obviously the complex envelope with e/*" being the complex carrier. A comparison of Eqs (11.23) and (11.24) reveals that in this case, (1) = a(r)e/% aepani + 2w 2, 5 ea ° be eh @ beh 2W. 2 0 fe aT (b) BRI | 2 -w 0 Ww ai (o) Fig. 11.5 (a) Amplitude spectrum of the bandpass signal x(t) (6) Amplitude spectrum of pre-evnvelope of x(t) (©) Amplitude spectrum of complex envelope of x(t) The complex-envelope representation of a bandpass signal is a very convenient tool that is widely used in the representation of radar and sonar signals as well as in the analysis of bandpass systems. ‘We will now use the complex-envelope to arrive at the familiar ‘inphase and quadrature component’ rep- resentation ofa real-valued bandpass signal x) with center frequency .. For this purpose, let the complex- envelope ¥(¢) be written in terms of its real and imaginary parts as XO) = 1O+ixg0 +++ (11.25) Since i(1) isa lowpass signal of bandwidth, say, W, 40) and xo() are also lowpass signals of the same bandwidth W, but are real valued. From Eq, (11.23), we have x(t) = Ref ei] Re[ fx syotO}foose.+ js] [x() = x; (0) 008 @.1— x9() sin o,f --- (11.26) This representation of the bandpass signal x(),is called the canonical representation of x(t), The lowpass real-valued signal, 2/() i called the ‘in phase’ component of the bandpass signal x(t), while the tealonrecd lowpass signal, xo(), is called the ‘quadrature ‘component ofthe bandpass signal, (0) This because, while 21(0) multiplies cos a, (¢) multiplies sin «1 which isin phase quadrature withthe carrie: signal cos a. In the foregoing, we have used three different representations of the real-valued bandh 1 th centre frequency f... These different representations are: aeserene [572] ynvous and en 7 Hilbert Transform—Continucs 127) x40) = a(t cos faye + BO] (11.28) (0 = Re[A(e/™” ]= Re[x.00] 129) and X(1) =x1() 608 @.t~ x9 (1) sin @t 11.27) and ‘Te entities used in these three representations are obviously related. By expanding RHS of Fa ( ‘comparing with RHS of Eq. (11.29), we get x,(1)= a(e)cos (0) 11.30) Xo(t)= a(s)sino(e) 8), By writing cos (af + 6(0)] of Eq. (11.27) as Re[e/®°*"! ] and comparing with RHS of Eq, (11.28), we get Hy = aie” ai3p so that a(t) = |X) Further, from Eq. (11.22), we have X60 = He? Ix.) = FO] =a --- (11.32) From Eq. (11.30), we have ao =[so+30]° --41.33) and (0) = tan"! [22] 3,0, + (1.34) Given a bandpass signal x(¢) = a(¢) cos {@.1 + (0) it is possible to obtain, except for a constant multiplier, its in-phase and quadrature components. Conversely, given the in-phase and quadrature components, it is possible to obtain/generate the signal x(¢) from them. (9) = a(0) cos [eet + O(0)] x(0) c08 @.1 = a(f[cose,1 cos (1) —[email protected] sin (1) kos@,t = [a(t)cos6(1)]cos? 4 —[a(1)sinO(.)|[email protected]@.t i 1 = 78UMI+ 608201) xo(4)sin 2,1 Hence, ‘when (cos ais lowpass feed the high-frequency terms. (7608 ene and g(t) sin 2 i " ‘sh and (0) is obtained at the output of the filter. Similarly, auhvens, if.x(0) is multiplied by sin Pe sin @ f and the c 5 roduet signal fered, only the low-frequency component + 9(t) is obtained at the perations are shown in block schematic form in Fig. 11.6(a) ‘output of the LPR. di572! signals and Systems LPF > 1/2 x(t) COS Wet Oscillator x(t) COS Wet 90° phase shifter sin Wet LPF > 112 x(t) (a) x(t) Oscillator cos ast 90° phase shifter (b) x(t)