4 CHARACTERIZATION OF COMMUNICATION SIGNALS AND SYSTEMS Signals can be categorized in a number of different ways, such as random versus deterministic, discrete time versus continuous time, discrete amplitude versus continuous amplitude, lowpass versus bandpass, finite energy versus infinite energy, finite average power versus infinite average power, etc. In this chapter, we treat the characterization of signals and systems that are usually encountered in the transmission of digital information over a communication channel. In particular. we introduce the representation of various forms of Gigitally modulated signals and describe their spectral characteristics. We begin with the characterization of bandpass signals and systems, including the mathematical representation of bandpass stationary stochastic Processes. Then, we present a vector space representation of signals. We conclude with the representation of digitally modulated signals and their spectral characteristics. 4-1 REPRESENTATION OF BANDPASS SIGNALS AND SYSTEMS Many digital information-bearing signals are transmitted by some type of carrier modulation. The channel over which the signal is transmitted is limited im bandwidth to an interval of frequencies centered about the carrier, as in double-sideband modulation, or adjacent to the carrier, as in single-sideband modulation. Signals and channels (systems) that satisfy the condition that their bandwidth is much smaller than the carrier frequency are termed narrowband bandpass signals and channels (systems). The modulation performed at the 182CHAPTER 4. CHARACTERIZATION OF COMMUNICATION SIGNALS AND sysTEMS 153 lag FIGURE 41-1 Spectrum of a bandpass signal 7. ° % transmitting end of the communication system to generate the bandpass signal and the demodulation performed at the receiving end to recover the digital information involve frequency translations. With no loss of generality and for mathematical convenience, it is desirable to reduce all bandpass signals and channels to equivalent lowpass signals and channels. As a consequence, the results of the performance of the various modulation and demodulation techniques presented in the subsequent chapters are independent of carrier frequencies and channel frequency bands. The representation of bandpass signals and systems in terms of equivalent lowpass waveforms and the characterization of bandpass stationary stochastic processes are the main topics of this section. 4-1-1" Representation of Bandpass Signals Suppose that a real-valued signal s(¢) has a frequency content concentrated in a narrow band of frequencies in the vicinity of a frequency f., as shown in Fig. 4-1-1. Our objective is to develop a mathematical representation of such signals. First, we construct @ signal that contains only the positive frequencies in s(2). Such a signal may be expressed as SAP) = W(ASA) (4-1-1) where S(f) is the Fourier transform of s(t) and u(f) is the unit step function. ‘The equivalent time-domain expression for (4-1-1) is so= f° s.tne™tar F'Qu(f)) ® Fo '[S(A)) (4-1-2) The signal s,(1) is called the analytic signal ot the pre-envelope of s(t). We note that F~'[S(f)] =s(f) and Pu py)= ay + 2 (4-13)154 irra. COMMUNICATIONS, Hence, s9= [50 +L] «sty = 560 +700) 14) We define 5(0) as wt) elie) 1 al et (4-1-5) The signal (1) may be viewed as the output of the filter with impulse response A= saced =40 (f=0) (41-7) i (f<0) We observe that ‘H(f)|=1 and that the phase response @(f) = —4 for f>0 and @(f) = bx for f <0. Therefore, this filter is basically a 90° phase shifter for all frequencies in the input signal. The analytic signal s.(1) is a bandpass signal. We may obtain an equivalent lowpass representation by performing a frequency translation of S.(f). Thus, we define S,(f) as SP) = SAF + fe) ° (41-8) ‘The equivalent time-domain relation is S(t) =5.(eP™ [s@) + j8@]e?™ (41-9) or, equivalently, S(t) + j8(t) =5(e?™ (4-1-10) In general, the signal s(1) is complex-valued (see Problem 4-5), and may be expressed as sh) =x(0) + iv) 4)CHAPTER 4 CHARACTERIZATION OF COMMUNICATION SIGNALS AND S*sTEMs 15S, If we substitute for sf) in (4-1-1) and equate reat and imaginary parts on each side, we obtain the relations $(0) = 4(0) cos 2nf.¢~ y(0) sin 2afe 41-12) S(t) = x(t) sin 2arft + y(0) cos 2afet (41-13) The expression (4-1-12) is the desired form for the representation of a bandpass signal. The low-frequency signal components x(t) and y(t) may be Viewed as amplitude modulations impressed on the carrier components cos 2nft and sin 2nf-t, respectively. Since these carrier components are in phase quadrature, x(1) and y(t) are called the quadrature components of the bandpass signal (0). ‘Another representation of the signal in (4-1-12 s() = Re flx(9) + je?" = Re [se] (41-14) where Re denotes the real part of the complex-valued quantity in the brackets following. The lowpass signal s,(¢) is usually called the complex envelope of the real signal s(f), and is basically the equivalent lowpass signal. Finally, a third possible representation of a bandpass signal is obtained by expressing 57) as SA) = ate" (4-1-15) where ol = VEO TD (41-15) 10 a= tan (41.17) Then $(0) = Re [s/(e?™"] = Re fa(syell2nr-e = a(t) cos [mfr + Oe) (4-1-18) The signal a(t) is called the envelope of s(¢), and 6(t) is called the phase of s(0). ‘Therefore, (4-1-12), (4-1-14), and (4-1-18) are equivalent representations of bandpass signals. ‘The Fourier transform of s(t) is sin= [sie dt J Retsine2™rpe "a (41-19) Use of the identity Re (6) =H(E + &*) (4-1-20)456 piciraL, communications in (4-1-19) yields the result SUN =} foie sttne Pe Aa ae = USA - f+ SH-f - LD) (41-21) where S(f) is the Fourier transform of s,(/). This is the basic relationship between the spectrum of the real bandpass signal S(f) and the spectrum of the equivalent lowpass signal 5)(f). The energy in the signal s(f) is defined as sf s in) de = J ” tRe faite? de (41-22) ‘When the identity in (4-1-20) is used in (4-1-2), we obtain the following result: if =i wore +E sant costars +200) ai (41-23) Consider the second integral in (4-1-23), Since the signal s(t) is narrowband, the real envelope a(t) (s,(0)| or, equivalently, a°(¢) varies slowly relative to the rapid variations exhibited by the cosine function. A graphical illustration of the integrand in the second integral of (4-1-23) is shown in Fig. 4-1-2. The value of the integral is just the net area under the cosine function modulated by a(t). Since the modulating waveform a°(r) varies slowly relative to the cosine function, the net area contributed by the second integral is very small relative to the value of the first integral in (4-1-23) and, hence, it can be cod 16 280)) FIGURE 61-3 The signal o() cos [4p + 20()}CHAPTER 4 CHARACTERIZATION OF COMMUNICATION SIGNALS AND systems 157 neglected. Thus, for all practical pruposes, the energy in the bandpass signal '5(), expressed in terms of the equivalent lowpass signal s,(), is eat” P : =} fuera (4128) where (| is just the envelope a(t) of s(0). 4-1-2 Representation of Linear Bandpass Systems A linear filter or system may be described either by its impulse response h(t) ‘or by its frequency response H(f), which is the Fourier transform of A(t) Since h(t) is real, H-f)= HU) (41.25) Let us define H,(f ~ f.) as nar sy={o Gea 41-26) ‘Then I-D- ley gee ae Using (4-1-25), we have Hf) = AAS ~ f+ HE - fd) (4.1.28) which resembles (4-1-21) except for the factor }. The inverse transform of Af) im (4-1-28) yields h(s) in the form (0) = hi(e™+ hte Pm = 2Re [hte] (4-129) where A,(¢) is the inverse Fourier transform of H(f). In general, the impulse response fy(t) of the equivalent lowpass system is complex-valued. 41-3. Response of a Bandpass System to a Bandpass Signal In Sections 4-1-1 and 4-1-2, we have shown that narrowband bandpass signals and systems can be represented by equivalent lowpass signals and systems. In this section, we demonstrate that the output of a bandpass system to a158. picrrat. communications bandpass input signal is simply obtained from the equivalent lowpass input signal and the equivalent lowpass impulse response of the system. ‘Suppose that s() is a narrowband bandpass signal and 5(r) is the equivalent lowpass signal. This signal excites a narrowband bandpass system characterized by its bandpass impulse response h(t) or by its equivalent lowpass impulse response h(t). The output of the bandpass system is also a bandpass signal, and, therefore, it can ‘2 expressed in the form r(t) = Re [rhe] (41-30) where r(t) is related to the input signal s(¢) and the impulse response h(t) by the convolution integral ro) f s(Dat- ade (41-31) Equivalently, the output of the system, expressed in the frequency domain, is RY) = SCAMS) (41-32) Substituting from (4-1-21) for S(f) and from (4-1-28) for H(f), we obtain the result RUP)= ASUS —£) + SMF ~ AMS ~ f+ HA -f -f)) (41-33) When s(t) is a narrowband signal and A(s) is the impulse response of a narrowband system, 5(f ~ f-)=0 and H(f — f.)=0 for f <0. It follows from this narrowband condition that SF —AHMCS ‘Therefore, (4-1-33) simplifies to RU) = HSS ~ SHAS ~£.) + Sf ~ {HIF -f) = HRAS ~ f+ RIF -fI (4-1-34) =0, SM-f~- {HAS -f)=0 where RAF) = SHPHAS), (41-35) is the output spectrum of the equivalent lowpass system excited by the equivalent lowpass signal. It is clear that the time domain relation for the output (0) is given by the convolution of s,(¢) with h(t). That is, =| shhir- ede (41-36)CHAPTER 4. CHARACTERIZATION OF COMMUNICATION SIGNALS AND SYSTENS 159) The combination of (4-1-36) with (4-1-30) gives the relationship between the bandpass output signal r(r) and the equivalent lowpass time functions s;(¢) and h(t). This simple relationship allows us to ignore any linear frequency translations encountered in the modulation of a signal for purposes of matching its spectral content to the frequency allocation of a particular channel. Thus, for mathematical convenience, we shall deal only with the transmission of equivalant lowpass signals through equivalent lowpass channels. 4-1-4 Representation of Bandpass Stationary Stochastic Processes ‘The representation of bandpass signals presented in Section 4-1-1 applied to deterministic signals. In this section, we extend the representation to sample functions of a bandpass stationary stochastic process. In particular, we derive the important relations between the correlation functions and power spectra of the bandpass signal and the correlation functions and power spectra of the equivalent lowpass signal. Suppose that n(1) is a sample function of a wide-sense stationary stochastic process with zero mean and power spectral density ®,,(f). The power spectral density is assumed to be zero outside of an interval of frequencies centered around +f., where f. is termed the carrier frequency. The stochastic process 1n(0) is said to be a narrowband bandpass process if the width of the spectral Mae (42-14) Then, the orthonormal vector uy is (42-13) y= (42.15)‘CHAPTER «© CHARACTERIZATION OF COMMUNICATION SIGNALS AND sysTEMS. 165 By continuing this procedure, we shall construct a set of m, orthonormal vectors, where n, n then men 42.2 Signal Space Concepts ‘As in the case of vectors, we may develop a parallel treatment for a set of signals defined on some interval {a, 6]. The inner product of two generally complex-vatued signals x(t) and x,(0) is denoted by (x,(1), «,(0) and defined as cr, xx0 [e800 (4-2-16) ‘The signals are orthogonal if their inner product is zero. ‘The norm of a signal is defined as wxon=([xcorat)” (42.19 A set of m signals are orthonormal if they are orthogonal and their norms are all unity. A set of m signals is linearly independent, if no signal can be represented as a linear combination of the remaining signals. ‘The sriangle inequality for two signals is simply In) + 2:00 OH + ICO (4218) and the Cauchy=Schwartz inequality is 7 a a [ reor rl if veto? | (42:19) with equality when x,(¢ = axy(c), where a is any complex number. [ soeteoar| < 4-2-3 Orthogonal Expansions of Signals In this section, we develop a vector representation for signal waveforms, and, thus, we demonstrate an equivalence between a signal waveform and its vector representation. Suppose that s(t) is a deterministic, real-valued signal with finite energy &-f bora (4-220) Furthermore, suppose that there éxists a set of functions {f(¢), 0 = 1,2,..., N} that are orthonormal in the sense that [ seotato de= {2 rend ua (42.21)166 vicinal. communcaTions We may approximate the signal s(t) by a weighted linear combination of these functions, ie., $= D HAO (42-22) where {s,, 1 te seo} at (42-24) ‘The optimum coefficients in the series expansion of s(¢) may be found by differentiating (4-2-24) with respect to each of the coefficients {s,} and setting the first derivatives to zero. Alternatively, we may use a well-known result «from estimation theory based on the mean-square-error criterion, which, imply stated, is that the minimum of % with respect to the {54} is obtained when the error is orthogonal to each of the functions in the series expansion. Thus, ie [so - Sno fult)dt=0, n=1,2,...,K (42-25) Since the functions {f,(1)} are orthonormal, (4-2-25) reduces to Sa [sone dt, on Thus, the coefficients are obtained by projecting the signal s(t) onto each of the functions {f,(0)}. Consequently, $(¢) is the projection of s(t) onto the K-dimensional signal space spanned by the functions {f,(‘)}. The minimum ‘mean square approximation error is (42-26) Smu=f etosinae [wora-[§ stcosina =e-35 (42-27) me which is nonnegative, by definition.CHAPTER | CHARACTERIZATION OF COMMUNICATION SIGKALS AND SsTEMS 167 When the minimum mean square approximation error &y Sa fF [s(oP dt (4-2-28) Under the condition that ain = 0, we may express s(7) as st)= % sift) (4-229) where it is understood that equality of s(/) to its series expansion holds in the sense that the approximation error has zero energy. When every finite energy signal can be represented by a series expansion of the form in (4-2-29) for which %,,, = 0, the set of orthonormal functions {f,(¢)} is stid to be complete Example 4-2-1: Trigonometric Fourier Series A finite energy signal s() that is zero everywhere except in the range Os05 7 and has a finite number of discontinuities in this interval, can be represented in a Fourier series as st) 2nkt 2nkr +b, sin2Ht) T T (a cos (42:30) where the coefficients {a,, b,} that minimize the mean square error are given by eau { * (090082 a a= Gp), (cos ae oe i (42.31) kt b= Fe] sto sina The set of trigonometric functions {V2/T cos 2xkt/T, V2/T sin 2xkt/T} is complete, and, hence, the series expansion results in zero mean square error. These properties ate easily established from the development given above Gram-Schmidt Procedure Now suppose that we have a set of finite energy signal waveforms {5,(1), (= 1,2,..., M} and we wish to construct a set ‘of orthonormal waveforms. The Gram-Schmidt orthogonalization procedure allows us to construct such a set. We begin with the first waveform s,(t), which is assumed to have energy @,. The first waveform is simply constructed as s(t) ve, Thus, f\(9 is simply s\(¢) normalized to unit energy. fat)168. DicrraL communicaTions The second waveform is constructed from s2(t) by first computing the projection of fi(r) onto s2(t), which is cn f sAt)fi( de (4-2-33) Then, c;,f,(0) is subtracted from s,(1) to yield FAN = SAD er) (4-2-34) This waveform is orthogonal to fi(¢) but it does not have unit energy. If & denotes the energy of f;(0), the normalized waveform that is orthogonal to Ald is HO Fg (4-2-35) In general, the orthogonalization of the kth function leads to (4-2-36) where fo=s0-'S Cafe) (4-2-37) and ca= [spo de (4238) ‘Thus, the orthogonalization process is continued until all the M signal waveforms {5,(f)} have been exhausted and N- si Based on the expression in (4-2-3), each signal may be represented by the vector sea vee Saal G24) or, equivalently, as a point in the N-dimensional signal space with coordinates {su. (= 1.2.....Nb The energy in the kth signal is simply the square of the length of the vector or. equivalently. the square of the Euclidean distance from the origin to the point in the N-dimensional space. Thus. any signal can be represented geomettically as 2 point in the signal space spanned by the orthonormal functions { f,(1)} Example 4-2-3 Let us obtain the vector representation of the four signals shown in Fig. 4.2-1(a) by using the orthonormal set of functions in Fig. 4-2-1(b). Since the dimensionality of the signal space is N = 3, each signal is described by three components, The signal s,(:) is characterized by the vector s, = (V2 0.0). Similarly, the signals sy(0). (0). and s,(t) are characterized by the vectors 0.¥3,0). $= (V2.0.1), and s.=(- 2.0.1), respectively. These vectors are shown in Fig. 4-2-2. Their lengths are [s,)= V2. js = V2. The fur signal ec epee spits fCHAPTER 4 CHARACTERIZATION OF COMMUNICATION SIGNALS AND svsteMs 171 {ss = V3, and js,| = V3, and the corresponding signal energies are & = {5,/°, =1,2,3,4 We have demonstrated that a set of M finite energy waveforms {s,(} can be represented by a weighted linear combination of orthonormal functions {f,(t)} of dimensionality N M=1 eae) for all m, n. Hence. the set of simplex waveforms is equally correlated and requires less energy. by the factor 1-1/M, than the set of orthogonal waveforms. Since only the origin was translated, the distance between any pair of signal points * maintained at d= V2@, which is the same as the distance between any pair of orthogonal signals. Figure 4-3-10 illustrates the simplex signals for M = the signal dimensionality is N= M ~ 1. 3, and 4. Note that Signal Waveforms from Binary Codes A set of M signaling waveforms can be generated from a set of M binary code words of the form CMe mz ve Gyale m =12,2 ME (43-36) where ¢,,,=0 or 1 for all m and j, Each component of a code word is mapped into an elementary binary PSK waveform as follows: Emmy = LP Smit) os 2mfr (O 5y(0) = -% where T.=T/N and & = €/N. Thus, the M code words (C,,} are mapped into a set of M waveforms {s,,(1)}. cos2afr (0818 T,) T.FIGURE 43:10 FIGURE 4311 (CHAPTER 4 CHARACTERIZATION OF COMMUNICATION SIGNALS AND svsreMs 185 Signal space diagrams for M-ary simplex signals. ‘The waveforms can be represented in vector form as oe eee ae (4-3-38), where sq; = VIN for all m and j. N is called the block length of the code. and it is also the dimension of the M waveforms. ‘We note that there are 2” possible waveforms that can be constructed from. the 2" possible binary code words. We may select a subset of M<2N signal waveforms for transmission of the information. We also observe that the 2” possible signal points correspond to the vertices of an N-dimensional hyper- cube with its center at the origin. Figure 4-3-11 illustrates the signal points in N=2 and 3 dimensions. 40 1 fo Signal space diagrams for signals ‘enerated from binary codes.186 biorac commesicattons Each of the M waveforms has energy & The cross-correlation between any pair of waveforms depends on how we select the M waveforms from the 2° possible waveforms. This topic is treated in Chapter 7. Clearly, any adjacent signal points have a cross-correlation coefficient €1=2/N)_N-2 a 39) € and a corresponding distance of a= VI8—p,) = Va¥IN (43-40) This concludes our discussion of memoryless modulation signals. 4-3-2, Linear Modulation with Memory FIGURE 43:12 ‘The modulation signals introduced in the previous section were classified as memoryless, because there was no dependence between signals transmitted in ‘non-overlapping symbol intervals. In this section, we present some modulation signals in which there is dependence between the signals transmitted in successive symbol intervals. This signal dependence is usually introduced for the purpose of shaping the spectrum of the transmitted signal so that it ‘matches the spectral characteristics of the channel. Signal dependence between signals transmitted in different signal intervals is generally accomplished by encoding the data sequence at the input to the modulator by means of ‘modulation code, as described in Chapter 9. In this section, we shall present examples of modulation signals with ‘memory and characterize their memory in terms of Markov chains. We shal! confine our treatment 10 baseband signals. The generalization to bandpass signals is relatively straightforward, Figure 4-3-12 illustrates three different baseband signals and the corres. ponding data sequence. The first signal, called NRZ, is the simplest. The binary information digit 1 is represented by a rectangular pulse of polarity A and the binary digit zero is represented by a rectangular pulse of polarity —A. wre rat Delay modulation Baseband signals bauFIGURE 43.13 CUAPIERC CHARACTERIZATION OF COMMUNICATION SIGNALS AND SETH 187. Hence, the NRZ modulation is memoryless and is equivalent to a or a binary PSK signal in a carrier-modulated system. ‘The NRZI signal is different from the NRZ, signal in that transitions from one amplitude level to another occur only when a 1 is transmitted. The amplitude level remains unchanged when a zero is transmitted. This type of signal encoding is called differential encoding. The encoding operation is described mathematically by the relation inary PAM. b= 4 BO, 34D where {a,} is the binary information sequence into the encoder, {b,} is the output sequence of the encoder, and @ denotes addition modulo 2, When 4, = 1, the transmitted waveform is a rectangular pulke of amplitude A. and when b, =0, the transmitted waveform is a rectangular pulse of amplitude ~A Hence, the output of the encoder is mapped into one of two waveforms in exactly the same manner as for the NRZ signal. The differential encoding operation introduces memory in the signal. The combination of the encoder and the modulator operations may be represented by a state diagram (a Markov chain) as shown in Fig. 4-3-13. The state diagram may be deseribed by two transition matrices corresponding to the two possible input bits (0. 1}. We note that when a, = 0, the encoder stays in the same state Hence, the state transition matrix for a zero is simply 10 7 3-42) T le i (43-42) if a, results in a transition from state i to state j, i= i, 2, and j =1 2: otherwise, /, = 0. Similarly. the state transition matrix for ay = is n-[° 3] ws) ‘Thus, these two state transition matrices characterize the NRZI signal. Another way to display the memory introduced by the precoding operation is by means of a trellis diagram. The trellis diagram for the NRZI signal is State diagram for the NR2I signal. yetFIGURE 43:14 FIGURE 43:15 188. icra communications wey esi Ser way XK wan Oa aS i S20 ‘The uellis diagram for the NRZI signal illustrated in Fig. 4-3-14. The trellis provides exactly the same information concerning the signal dependence as the state diagram, but also depicts a time evolution of the state transitions. ‘The signal generated by delay modulation also has memory. As shown in Chapter 9, delay modulation is equivalent to encoding the data sequence by a run-length-limited code called a Miller code and using NRZI to transmit the encoded data. This type of digital modulation has been used extensively for digital magnetic recording and in carrier modulation systems employing binary PSK. The signal may be described by a state diagram that has four states as shown in Fig. 4-3-15(a). There are two elementary.waveforms s,(¢) and s3(0) and their negatives —5\(¢) and ~s,(t), which are used for transmitting the binary information. These waveforms are illustrated in Fig. 4-3-15(b). The mapping from bits to corresponding waveforms is illustrated in the state diagram. The state transition matrices that characterize the memory of this encoding and modulation method are easily obtained from the state diagram in Fig. 4-3-15. When a, =0, we have (43-44) coco cone State diagram (@) and basic waveforms (b) for delay modulated (Miller-encoded) signal 7 ' onan sy sins-agn OsisT AI sues Osis © raIAPIER CHARACTERIZATION OF COMMUNICATION SIGNALS AND SYSTEMS 189 and when a, = 1, the transition matrix is 0100 00 to = 4.3.45 Lal ne (43-45) 9010. Thus, these two 4x 4 state transition matrices characterize the state diagram for the Miller-encoded signal. Modulation techniques with memory such as NRZI and Miller coding are generally characterized by a K-state Markov chain with stationary state probabilities {p,i=1,2,...,K} and transition probabilities {py.i, j= 1,2,...,K} Associated with each transition is a signal waveform 5,(:), {=1,2,...,K. Thus, the transition probability p, denotes the probability that signal waveform s(t) is transmitted in a given signaling imerval after the transmission of the signal waveform s,(t) in the previous signaling interval. The transition probabilities may be arranged in matrix form as Pu Pave Pix Pe (43-46) ki Pra ss Pr where P is called the transition probability matrix. The transition probability matrix is easily obtained from the transition matrices {T,} and the corresponding probabilities of occurrence of the input bits (or, equivalently, the stationary state transition probabilities {p)}). The gene-al relationship may be expressed as 2 P=> aT; (43-47) where 4, = P(a, = 0) and q2 = P(ay = 1). For the NRZI signal with equal state probabilities p, =p. = } and transition matrices given by (4-3-42) and (4-3-43), the transition probability matrix is be p-[ J (4-3-48) iy ¢ ) Similarly, the transition probability matrix for the Miller-coded signal with equally likely symbols (q, = q2 = 4 or, equivalently, p, = p2 = ps = pa = 3) is oo} 00434 P= phoo 4040. The transition probability matrix is useful in the determination of the spectral (43-49)190 prorrat communtcarions characteristics of digital modulation techniques with memory, as we shall observe in Section 4-4, 4-3-3 Nonlinear Modulation Methods with Memory In this section, we consider a class of digital modulation methods in which the phase of the signal is constrained to be continuous. This constraint results in a phase or frequency modulator that hes memory. The modulation method is also nonlinear. Continuous-Phase FSK (CPFSK) A conventional FSK signal is generated by shifting the carrier by an amount f,=4 AfIy, [y= £1, £3,..., #(M-1), to reflect the digital information that is being transmitted. This type of FSK signal was described in Section 4-3-1, and it is memoryless. The switching from one frequency to another may be accomplished by having M=2* separate oscillators tuned to the desired frequencies and selecting one of the M frequencies according to the particular k-bit symbol that is to be transmitted in a signal interval of duration T = k/R seconds. However, such abrupt switching from one oscillator output to another in successive signaling intervals results in relatively large spectral side lobes outside of the main spectral band of the signal and, consequently, this method requires a large frequency band for transmission of the signal. To avoid the use of signals having large spectral side lobes, the information- bearing signal frequency modulates a single carrier whose frequency is changed continuously. The resulting frequency-modulated signal is phase-continuous and, hence, it is called continuous-phase FSK (CPFSK). This type of FSK signal has memory because the phase of the carrier is constrained to be continuous. In order to represent a CPFSK signal, we begin with a PAM signal a(t)= > hg -nT) (4-3-50) where {J,} denotes the sequence of amplitudes obtained by mapping k-bit blocks of binary digits from the information sequence {a,} into the amplitude levels +1, £3,..., £(M —1) and g(é) is a rectangular pulse of amplitude 1/27 and duration 7 seconds. The signal dit) is used to frequency-modulate the carrier. Consequently, the equivalent lowpass waveform v(t) is expressed as v= feo {ilsnns, fiw det 4} (43-51) where f, is the peak frequency deviation and do is the initial phase of the carrier. ‘The carrier-modulated signal corresponding to (4-3-51) may be expressed as s(= ft 208 [2nf.t + 6(t,1) + dol (4-3-2)CHAPTER 4 CHARACTERIZATION OF COMMUNICATION SIGNALS AND systEMS 191 where (t;1) represents the time-varying phase of the carrier. which is defined as or ant a(t)dt = 4nTf, f 5 [z Lge - »7)| dt (4-3-53) Note that, although d(‘) contains discontinuities, the integral of (0) i continuous. Hence, we have a continuous-phase signal. The phase of the carrier in the interval nT <1 (n + 1)T is determined by integrating (4-3-53). ‘Thus, iTS hy + aft — aT oeD= = 0, + 2mhlge~ nT) (43-54) where A, @,, and q(t) are defined as ha 267 (43-55) @=ah'S hk (43-56) 0 @T) We observe that 8, represents the accumulation (memory) of all symbols up to time (n ~ 1)T. The parameter h is called the modulation index. Continuous-Phase Modulation (CPM) When expressed in the form of (4-3-54), CPFSK becomes a special case of a general class of continuous-phase modulated (CPM) signals in which the carrier phase is OGD =2n SY hag(t-kT), nT T, the CPM signal is called full response CPM. If g(0) #0 for 1>T, the modulated signal is called partial response CPM. Figure 43.16 illustrates several pulse shapes for g(t), and the corresponding g(r). It is apparent that an infinite variety of CPM signals can be generated by choosing different pulse shapes g(t) and by varying the modulation index h and the alphabet size M. It is instructive to sketch the set of phase trajectories H(t: 1) generated by all possible values of the information sequence {/,). For example, in the case of CPFSK with binary symbols /, = +1, the set of phase trajectories beginning at time 1=0 is shown in Fig. 4-3-17. For comparison, the phase trajectories for quaternary CPFSK are illustrated in Fig. 4-3-18. These phase diagrams are called phase trees. We observe that the phase trees for CPFSK are piecewise linear as a consequence of the fact that the pulse g(¢) is rectangular. Smoother phase trajectories and phase trees are obtained by using pulses that do not contain discontinuities, such as the class of raised cosine pulses. For example, a phase trajectory generated by the sequence (1,—1,—1,—1,1,1,—1.1) for a partial response, raised cosine pulse of length 37 is illustrated in Fig. 4-319. For comparison, the corresponding phase trajectory generated by CPFSK is also shown. ‘The phase trees shown in these figures grow with time. However, the phaseCHAPTER 4) CHARACTERIZATION OF COMMUNICATION SIGNALS AND systems 193 FIGURE 4316 (Contimed), FIGURE 4317 Phase trajectory for binary CPFSK.FIGURE 43:18 194 oucrrat communications eS ae 7 Oe KEES Ss ISISLS oe Ee SESS RAI Phase trajectory for quatemary CPFSK. of the carrier is unique only in the range from = 0 to & =2 or. equivalently, from 6 = ~x to. = . When the phase trajectories are plotted modulo 2x, say in the range (—m, 2), the phase tree collapses into a structure called a phase trellis. To properly view the phase trellis diagram, we may plot the two quadrature components x,(:1) =cos 6:1) and x()=sin de.) as functions of time. Thus, we generate a three-dimensional plot in which the quadrature components x, and x, appear on the surface of a cylinder of unit radius, For example, Fig. 4-3-20 illustrates the phase trellis or phase cylinderFIGURE 43.19 FIGURE ¢3.20 Phase cylinder for binary CPM with h= } and a raised cosine CHAPTER # CHARACTERIZATION OF COMMUNICATION SIGNALS AND systems 195 Phase trajectories for binary CPFSK (dashed) and binary, paral response CPM based on raised cosine pulse of length 37 (solid. {From Sundberg (1986), © 1986 IEEE.] obtained with binary modulation, a modulation index A=, and a raised cosine pulse of length 37 : Simpler representations for the phase trajectories can be obtained by displaying only the terminal values of the signal phase at the time instants 1 = nT. In this case, we restrict the modulation index of the CPM signal to be rational. In particular, let us assume that h=m/p, where m and p are relatively prime integers. Then, a full response CPM signal at the time instants 1 = nT will have the terminal phase states (43-60) when m is even and (43-61) when m is odd. Hence, there are p terminal phase states when m is even and 2p states when m is odd. On the other hand, when the pulse shape extends plse of length 37. [From Sundberg (1986), © 1986 IEEE.)FIGURE 43.21 FIGURE 43.22, 196 osciTaL comMUNICATIONS « ‘State trellis for binary CPFSK with f= § over L symbol intervals (partial response CPM), the number of phase states ‘may increase up to a maximum of S, where s few (even m) “" Lapat=" (odd m) where M is the alphabet size. For example, the binary CPFSK signal (full response, rectangular pulse) with A = }, has S,=4 (terminal) phase states. The state ‘retlis for this signal is illustrated in Fig. 4-3-21. We emphasize that the phase transitions from one state to another are not true ptiase trajectories. They represent phase transitions for the (terminal) states at the time instants =n. An alternative representation to the state trellis is the state diagram, which also illustrates the state transitions at the time instants ¢= nT. This is an even more compact representation of the CPM signal characteristics. Only the possible (terminal) phase states and their transitions are displayed in the state diagram. Time does aot appear explicitly as a variable. For example, the state diagram for the CPFSK signal with h = 4 is shown in Fig. 4-3-22. (43-62) Minimum-Shift Keying (MSK) MSK is a special form of bit State diagram for binary CPFSK widCHAPTER 4 CHARACTERIZATION OF COMMUNICATION StONALS AND sysTEMs 197 (and, therefore, CPM) in which the modulation index A = |. The phase of the carrier in the interval nT <¢<(n + Tis o:D= tx Sh + ahgt—nT) =6,+ ant (2), nT b(n) z gti nT )g¢+t—nT mT) (4-4-6) The second summation in (4-4-6), namely, D gt -aT)gte+ t-nT- mT) is periodic in the ¢ variable with period T. Consequently, $.(t + 7:1) is also periodic in the 1 variable with period T. That is, balt + T+ ttt T= batt 0) (4-4-7) In addition, the mean value of u(:), which is Elo = 4. DS ge-at) (4-4-8) is periodic with period 7. Therefore v(t) is a stochastic process having a periodic mean and autocorrelation function. Such a process is called a eyclostationary process ot a periodically stationary process in the wide sense, as described in Section 2-2-6. In order to compute the power density spectrum of a cyclostationary process, the dependence of @q(1+ t;1) on the f variable must be eliminated This can be accomplished simply by averaging $,(¢+ 7:1) over a single period. Thus, - Gald= Ff bolt + edd = S Galen) 5 3] BMC AT g(t + nT — mT) de = > sam 5 7, -stoste+ t~ mT) dt (4-49) We interpret the integral in (4-4-9) as the time-autocorrelation function of g(t) and define it as bet) f st(Og(t+ ddr (4-4-10)206 oicrTaL communications Consequently (4-4-9) can be expressed as B.A) = 2D dulemdy(t mT) 411) ‘The Fourier transform of the relation in (4-4-11) yields the (average) power density spectrum of u(¢) in the form Palf) hice Bf) (44-12) where G(f) is the Fourier transform of g(1), and ®,(f) denotes the power density spectrum of the information sequence, defined as O(N = YL dilme Perr (4-413) The result (4-4-12) illustrates the dependence of the power density spectrum of u(t) on the spectral characteristics of the pulse g(t) and the information sequence {/,,}. That is, the spectral characteristics of v(#) can be controlled by design of the pulse shape g(¢) and by design of the correlation characteristics of the information sequence. Whereas the dependence of ,(f) on G(f) is easily understood upon ‘observation of (4-4-12), the effect of the correlation properties of the information sequence is more subtle. First of all, we note that for an arbitrary autocorrelation ¢,(m) the corresponding power density spectrum ®,(f) is periodic in frequency with period 1/7. In fact, the expression (4-4-13) relating the spectrum ®,(f) to the autocorrelation ¢,(m) is in the form of an exponential Fourier series with the {¢,(m)} as the Fourier coefficients. As a consequence, the autocorreletion sequence ¢,,(m) is given by daimy= Tf aye" ap (4414) Second, let us consider the case in which the information symbols in the sequence are real and mutually uncorrelated. In this case, the autocorrelation function ¢,(m) can be expressed as ao? +p? (m=0) b,(m) = { 2 ie eo (44.15) where a} denotes the variance of an information symbol. When (4-4-15) is used to substitute for ¢,(m) in (4-4-13), we obtain euN= oi tat 3 eenmr (4-4-16) ‘The summation in (4-4-16) is periodic with period 1/7. It may be viewed asCHAFIER 4 CHARACTERIZATION OF COMMUNICATION SIGNALS AND SYSTENS 207 the exponential Fourier series of a periodic train of impulses with each impulse having an area 1/T. Therefore (4-4-16) can also be expressed in the form b> ap = o (4-4-17) apaoir® S alf-F) (44.17) Substitution of (4-4-17) into (4-4-12) yields the desired result for the power density spectrum of v(t) when the sequence of information symbols is 3 lo(F)/ ar The expression (44-18) for the power density spectrum is purposely separated into two terms to emphasize the two different types of spectral components. The first term is the continuous spectrum, and its shape depends ‘only on the spectral characteristic of the signal pulse g(¢). The second term consists of discrete frequency components spaced 1/T apart in frequency. Each spectral line has a power that is proportional to |G(f)} evaluated at f = m/T. Note that the discrete frequency components vanish when the information symbols have zero mean, i.e., 4, = 0. This condition is usually desirable for the digital modulation techniques under consideration, and it is satisfied when the information symbols are equally likely and symmetrically positioned in the complex plane. Thus, the system designer can control the spectral characteris- tics of the digitally modulated signal by proper selection of the characteristics of the information sequence to be transmitted. 2 2 (9) = TG Ae +S (4-418) Example 4-4-1 ‘To illustrate the spectral shaping resulting from g(/), consider the rectangu- lar pulse shown in Fig. 4-4-1(a). The Fourier transform of g(¢) is sin mT sey Of ate FIGURE 441 Rectangular pulse and its energy density spectrum (G(f)? one Ph are ° T a a) a iw o208. DIGITAL COMMUNICATIONS. Hence any = IGP = (ATY ar ) (4-419) This spectrum is illustrated in Fig. 4-4-1(b). Note that it contains 2er0s at multiples of 1/T in frequency and that it decays inversely as the square of the frequency variable, As a consequence of the spectral zeros in G(). all but one of the discrete spectral components in (4-418) vanish, Thus, upon substitution for |G(f)i? from (44-19), (4-4-18) reduces to sin fT) oe ) + Ate) (4-4-20) (N= 0141 Example 4-4-2 AS a second illustration of the spectral shaping resulting from g(t), we consider the raised cosine pulse 22h . s0=4[1 +0084 (1-D)], o LT. The autocorrelation function of the equivalent lowpass signal vyeesen bul + = seferp 2mm S hlate> r-kT)-qt-«7)] (4-430) First we express the sum in the exponent as a product of exponents. The result is aot tin) = SE TT expt i2abila+ r= &7)— qk) (4430) Next, we perform the expectation over the data symbols {/,}. Since these symbols are statistically independent, we obtain dbudt t= 4 TT 2D Prexpli2ahniq(s + ¢~ kT) “aan (4-432) Finally, the average autocorrelation function is 7 S(t) = +[ by At + ED dt (44.33) Although (4-4-32) implies that there are an infinite number of factors in the product, the pulse g(1) = 9°(1)=0 for 1<0 and 1> LT, and g(1) =0 for <0. Consequently only a finite number of terms in the product have nonzero exponents. Thus, (4-4-32) can be simplified considerably. In addition, if we let t= E+ mT, where 0 € LT. In this case, (44-34) may be expressed as SoulE+ mT) =[WUNNY" ACE), mel, OSEST — (4-4-35) where (jh) is the characteristic function of the random sequence {f,}, de- fined as wih) = Ele") = DS Pete (4-4-36) nh» node and A(é) is the remaining part of the average autocorrelation function. which may be expressed as wo=a[ AY = Patan -at-ar) wetne | a2 Grn = tl ( Ss, reine) meL (44-37) Thus, 4,,.(t) may be separated into a product of A(é) and (jh) as indicated in (4-4-35) for t= & + mT > LT and 0< €< T. This property is used below. ‘The Fourier transform of 6,.(t) yields the average power density spectrum as = [date Phar ore [ff dune ae] (4-4-38) But f batnemtar= [duinne ear +[ danerrar (43.39 With the aid of (4-435), the integral in the range LT<1 + => vue all Age POEM |E — (4.4.41) A property of the characteristic function is |¥(jh)| = 1. For values of h for which |y(jh)| <1, the summation in (4-4-41) converges and yields > 1 20 NO = Ga (4442) In this case, (4-4-41) reduces to : i aes : J batne aide mae [ Gol E+ LT) PNET de (4-443) By combining (4-4-38), (4-4-39), and (4-4-43), we obtain the power density spectrum of the CPM signal in the form Tamer]. Bale Par] Af) =2Re [doce Pe d+ (4-4-44) This is the desired result when |y(jh)|<1. In general. the power density spectrum is evaluated numerically from (4-4-44). The average autocorrelation, function $,,(t) for the range 0< r= (L + 1)T may be computed numerically from (4-4-34), For values of h for which |¥(jh)|= 1, €.g.. k= K, where K is an integer, we can set WUhy=e", OSvE) (44-45) ‘Then, the sum in (4-4-41) becomes Serene urate ~jhcot ar(r - %) (44-46)CHAPTLRE 4 CHARACIERIZATION OF COMMUNICATION SIGNALS AND sysTEMS 213 ‘Thus, the power density spectrum now contains impulses located at frequencies +y ae be Osv 1 the spectrum becomes much broader. In communica- tion systems where CPFSK is used. the modulation index is designed to conserve bandwidth, so that A <1 ‘The special case of binary CPFSK with h=} {or f,=1/4T) and y=0 corresponds to MSK. In this case, the spectrum of the signal is 16A°T ( cos 2afT y 1=16P 7? OA f)= (4-453) r where the signal amplitude A= 1 in (4-4-52). In contrast the spectrum of four-phase offset (quadrature) PSK (QPSK) with a rectangular pulse g(t) of duration T is 4 Pl f) = (inal y ny af If we compare these spectral characteristics, we should normalize the frequency variable by the bit rate or the bit interval T,. Since MSK is FSK, it follows that 7 = 7, in (4-4-53). On the other hand, in OOPSK. so that (4-4-4) becomes =2ar7( 22D) bf) afl (44-55) ‘The spectra of the MSK and OOPSK signals are illustrated in Fig. 4-4-6. Note thai the main lobe of MSK is 50% wider than that for OOPSK. However. the side lobes in MSK fall off considerably faster. For example. if we compare the bandwidth W that contains 99% of the total power, we find that W = 1.2/T, for MSK and W = 8/7, for OQPSK. Consequently. MSK has a narrower spectralCHAPTER 4 CHARACTERIZATION OF COMMUNICATION SIGNALS AND sysTEMs 215 Special easy for wo level CPESK. Spec eos for twokevel CPFSK. 12 10} os 07 06 os 03 02 pea 3 ° oH Oe TT 1s Normale requncy 7 o T z Normalized frequency Spectral demsiy for tavevel CPFSK Spec densi Fortwo evel CPFSK. Spectral density Spec deiy 16 ° 07 va 7 is Normaied feqency (7 Normalized equeny fT ‘el o FIGURE 443 Power density spectrum of binary CPFSK, ‘occupancy when viewed in terms of fractional out-of-band power above {T= 1. Graphs for the fractional out-of-band, power for OQPSK and MSK are shown in Fig. 4-4-7. Note that MSK is significantly more bandwidth-efficient than QPSK. This efficiency accounts for the popularity of MSK in many digital communications systems. Even greater bandwidth efficiency than MSK can be achieved by reducing216 o1crrat commesicaTions os oe os Spectral dey Spec dey Special dey fe fourteet CPESK Spectral densy fr for level CPSK os ox n=09 ae eet go heayr gos ko 0: a1 T + 3 ° 7 3 Nomateedtequensy (7 [Nomalzed egueay 7 1 Spel emf foclesel PSK Nowmslired trquency (F FIGURE 4-44 Power density spectrum of quaternary CPFSK. the modulation index. However, the FSK signals will no longer be orthogonal and there will be an increase in the error probability. Spectral Characteristics of CPM In general, the bandwidth occupancy of CPM depends on the choice of the modulation index h, the pulse shape g(/), and the number of signals M, As we have observed for CPFSK, small values of ‘t result in CPM signals with relatively small bandwidth occupancy, while large values of A result in signals with large bandwidth occupancy. This is atso the case for the more general CPM signals.CHAPTER 4 CHARACTERIZATION OF COMMUNICATION SIGNALS AND SySTEMS 217 Spectral deny for vgn eve CPPS. Special density fr eight ievel CPESK. 10 026 on pean Normalized fequency/7 (oh FIGURE 445 Power density spectrum of octal CPFSK. FIGURE ¢46 Power density spectra of MSK and offset QPSK. [From Gronemeyer and McBride (1976): © IEEE, 0 -100 -20 = -200 $ $ emai pomer spec den g 010 200A 80 Normalized frequency ofa from case (JT efits}218 vicrrat communicanions, i on es 5 2 og FIGURE 447 Fractional outot-band power (normalized — | sy frovnded bandwidth 281). [Pron Gronemever and McBride (1976); ©1976 * “#25 Fa EE RO 10120 TAO 160 180 200 EEE] 2WT = two-sided normed bandwidth |(Mafbit)/s| The use of smooth pulses such as raised cosine pulses of the form 1 2a0 (1 -cos2™) ere) woud Fi z(t cost) ( 1 for partial response, result in smaller bandwidth occupancy and, hence, greater bandwidth efficiency than the use of rectangular pulses. For example, Fig. 4-4-8 illustrates the power density spectrum for binary CPM with different partial response raised cosine (LRC) pulses when A= 4. For comparison, the spectrum of binary CPFSK is also shown. Note that as L increases the pulse g(t) becomes smoother and the corresponding spectral occupancy of the signal is reduced. Pose trum OWI) FIGURE 448 Power density spectrum for binary CPM with r= § and diferent pulse shapes, [From Aufin eta. (198%) 0) as esa OUI EEE} Normalized frequen 7FIGURE 449 FIGURE 44:10 CHAPTER 4. CHARACTERIZATION OF COMMUNICATION SIGNALS AND systES 219) Powe spect (Wa) & Power density spectrum for M = 4 CPM with 3RC and different modulation indices. [From Auli et al. (1981) ae eaoneeae (©1981 EEE) Neematiedegaersy The effect of varying the modulation index in a CPM signal is illustrated in Fig. 4-4-9 for the case of M=4 and a raised cosine pulse of the form given in (4-4-56) with L= 3. Note that these spectral characteristics are similar to the ones illustrated previously for CPPSK, except that these spectra are narrower due to the use of a smoother pulse shape. Finally, in Fig, 4-4-10, we illustrate the fractional out-of-band power for two-amplitude CPFSK with several different values of h. Fractional out-of-band power for two-component CPFSK. (Mulligan, 1988.) 00) 1000] 1500] 2000) 250 -200| suo Fria ato band power (8) 500 sown 0 OS Tie os a 1s 150 15 Fe Normale reqensy/7220 biciraL, COMMUNICATIONS. 4-4-3, Power Spectra of Modulated Signals with Memory In the last two sections, we have determined the spectral characteristics for the class of linearly modulated signals without memory and for the class of angle-modulated signals such as CPFSK and CPM, which are nonlinear and possess memory. In this section, we consider the spectral characteristics of linearly modulated signals that have memory that can be modeled by a Markov chain. We have already encountered such signals in Section 4-3-2, where we described several types of baseband signals. The power density spectrum of a digitally modulated signal that is generated by a Markov chain may be derived by following the basic procedure given in the previous section. Thus, we can determine the autocorrelation function and then evaluate its Fourier transform to obtain the power density spectrum. For signals that are generated by a Markov chain with transition probability matrix P, the power density spectrum of the modulated signal may be expressed in the general form (see Titsworth and Welch, 1961) 3. |Sps(4)f a(r-2)+ 5p. sine of) + Re {= Ses: snsinean)| (44-57) where $f) is the Fourier transform of the signal waveform s,(t). silt) = 5,(1) ~ = Pst) P,(f) is the Fourier transform of the discrete-time sequence p,(m), defined as PAA = = Pilaje 2eet {4-4-58) and K is the number of states of the modulator. The term p,(r1) denotes the probability that the signal s(t) is transmitted m signaling intervals after the transmission of s,(1). Hence, {p,(n)} are the transition probabi transition probability matrix PY. Note that p,(1) =p, When there is no memory in the modulation method, the signal waveform transmitted on each signaling interval is independent of the waveforms transmitted in previous signaling intervals. The power density spectrum of the resultant signal may still be expressed in the form of (4-4-57), if the transition probability matrix is replaced by PL Pros Pw Be) ee os (4-4-59) 1 Pe PxCHAPTER 4 CHARACTERIZATION OF COMMUNICATION SIGNALS AND SYSTEMS 221 and we impose the condition that P" = P for all n > 1. Under these conditions, the expression for the power density spectrum becomes a function of the Stationary state probabilities {p,} only, and, hence, it reduces to the simpler form on=7,&,|Ss(§)) alr-§) + £3 papas 28 Spy ReISinsrn) (4440) We observe that our previous result for the power density spectrum of memoryless linear modulation given by (4-4-18) may be viewed as a special case of (4-4-60) in which all waveforms are identical except for a set of scale factors that convey the digital information (Problem 4-30). We also make the observation that the first term in the expression for the power density spectrum given by either (4-4-57) or (4-4-60) consists of discrete Frequency components. This line spectrum vanishes when : ps(4) =0 (4-461) The condition (4-4-61) is usually imposed in the design of practical digital communications systems and is easily satisfied by an appropriate choice of signaling waveforms (Problem 4-31), Now, let_us determine the power density spectrum of the baseband- modulated signals described in Section 4-3-2. First, the NRZ signal is characterized by the (wo waveforms 5\(1) = g(¢) and s3(1) = —g(¢), where g(r) is a rectangular pulse of amplitude A. For K = 2, (4-4-60) reduces to a= FETE § |o(2)) a(r-2) +2 =Dicine assay Poets T ee where 2 carp AFTY? IGF = (aT) (or ) (4-463) Observe that when p =4, the line spectrum vanishes and df) reduces to a=") (4-464)222 o1arrat commentcations ‘The NRZI signal is characterized by the transition probability matrix Lea Pali i (4-465) Notice that in this case PY = P for all n> 1. Hence, the special form for the power density spectrum given by (4-4-62) applies to this modulation format as well. Consequently, the power density spectrum for the NRZI signal is identical to the spectrum of the NRZ signal. Delay modulation has a transition probability matrix o1o4 O14 ae) bolo. (4-4-6) and stationary state probabilities p, =} for i=1, 2, 3, 4. Powers of P may be obtained by use of the relation Pip=-lp (4-467) where p is the signal correlation matrix with elements i: pra sacistorae (4-468) and where the four signals {5,(1), (= 1, 2,3, 4} are shown in Fig. 4-315. It is easily seen that 1 0 0-1 o 1-1 0 7 4-4 er} o-1 1 0 i) -1 0 01 ‘Consequently, powers of P can be generated from the relation Pifp=— IP, k>1 (4-4-70) Use of (4-4-66), (4-4-69), and (4-4-70) in (4-4-57) yields the power density specirum of delay modulation. It may be expressed in the form = race 3 - = I2eos3y +5. PUN = SaT weonthy 23 ~ 2008 — 22 €05 24 — 12c0834 + Scos 44 + 12-cas Sy +2 cos bu - 8cos 7y + 2cos BU] (4-4-71) where w= af7.CHAPTER 4 CHARACTERIZATION OF COMMUNICATION SIGNALS AND SySTEMS 223 Delay madtaon (Milercae FIGURE 44:11 Power spectral density (one-sided) ‘of Miller code (delay modulation) and NRZ/NRZI baseband signals. [From Hecht and Guida (1969); Yo Osos Os Os TO TP ia 16 18 30 (01969 IEEE.) Normalized fequerey 7 The spectra of these baseband signals are illustrated in Fig. 4-4-11. Observe that the spectra of the NRZ and NRZI signals peak at f = 0. Delay modulation has a narrower spectrum and a relatively small zero-frequency content. Its bandwidth occupancy is significantly smaller than that of the NRZ. signal. These two characteristics make delay modulation an attractive choice for channels that do not pass de, such as magnetic recording media. 45 BIBLIOGRAPHICAL NOTES AND REFERENCES The characteristics of signals and systems given in this chapter are very useful in the design of optimum modulation/demodulation and coding/decoding techniques for a variety of channel models. In particular, the digital modula. tion methods introduced in this chapter are widely used in digital communica- tion systems. The next chapter is concerned with optimum demodulation techniques for these signals and their performance in an additive, white gaussian noise channel. A general reference for signal characterization is the book by Franks (1969). ‘OF particular importance in the design of digital communications systems are the spectral characteristics of the digitally modulated signals, which are presented in this chapter in some depth. Of these modulation techniques, CPM is one of the most important due to its efficient use of bandwidth. For this reason, it has been widely investigated by many researchers, and a large number of papers have been published in the technical literature. The most comprehensive treatment of CPM, including its performance and its spectral characteristics, can be found in the book by Anderson et al. (1986). In addition to this text, the tutorial paper by Sundberg (1986) presents the basic concepis and an overview of the performance characteristics of various CPM techniques. ‘This paper also contains over 100 references to published papers on this topic. There are a large number of references dealing with the spectral charac- teristics of CPFSK and CPM. As @ point of reference, we should mention that MSK was invented by Doelz and Heald in 1961. The early work on the power224 pIcitat COMMUNICATIONS spectral density of CPFSK and CPM was done by Bennett and Rice (1963), ‘Anderson and Salz (1965), and Bennett and Davey (1965). The book by Lucky etal. (1968) also contains a treatment of the spectral characteristics of CPFSK. Most of the recent work is referenced in the paper by Sundberg (1986). We should also cite the special issue on bandwidth-efficient modulation and coding Published by the EEE Transactions on Communications (March 1981), which contains several papers on the spectral characteristics and performance of CPM. The generalization of MSK to multiple amplitudes was investigated by Weber et ai. (1978). The combination of multiple amplitudes with general CPM was proposed by Mulligan (1988) who investigated its spectral characteristics and its error probability performance in gaussian noise with and without coding. 41 Prove the following properties of Hilbert transforms: a Ifx(e) =x(=0) then (0) = =2(-0): b Ifx() = ~x(-0) then 2) = 2(-0): € Ifx(¢) =c05.wyt then £(0) = sit @ If.x(e) =sin ays then #(t) £0 = =x: fe Odr= PP de gS. (OR) di=0. 42 If x(0) is a stationary random process with autocorrelation function 4,,( Elx(Ox(r + 2)] and spectral density ®,.(f) then show that c(t) = (7). s(t), and @:(f) = (Ff) 43 Suppose that (1) is @ zero-mean stationary narrowband process represented by either (4-1-37), (4-1-38), or (4-1-39). The autocorrelation function of the equiv- alent lowpass process 2(1) =.x(t) +/9(0) is defined as sea 2} 48 Show that Elz(nzte + 1) =O Suppose ¢,.(1) = .8(7), and let v= [soa Determine E(V") and E(VV*) = E(iVE). ‘44 Determine the autocorrelation function of the stochastic process x()=A sin (2nfr + 0) where f, is a constant and @ is a uniformly distttvsted phase, PO)=. OB) Determine the power spectral density and the autocorrelation fanction of y(t) = 200. 414 Consider an equivalent lowpass digitally modulated signal of the form u(t) =D [ag(t ~ 2nT) = jb,g(t - 2nT - T)] where {a,} and {b,} are two sequences of statistically independent binary digits and ‘8(0) is a sinusoidal pulse defined as sin (27) (0<1<27) so (otherwise) This type of signal is viewed as a four-phase PSK signal in which the pulse shape is one-half eycle of a sinusoid. Each of the information sequences {as} and {b.} is transmitted at a rate of 1/27 bits/s and, hence, the combined transmission rate is L/T bis/s. The two sequences are staggered in time by T seconds in transmission Consequently, the signal u() is called staggered four-phase PSK. 1 Show that the envelope \u() isa constant, independent of the information a, on the in-phase component and information b, on the quadrature component. In other words, the amplitude of the carrier used in transmitting ihe signal is constant. b Determine the power density spectrum of u(0). € Compare the power densty spectrum obtained from (b) with the power densi spectrum of the MSK signal. What conclusion can you draw from this comparison? 415 Consider a four-phase PSK signal represented by the equivalent lowpass signal u(t) = lagen) where 1 takes on one of the four possible values VI(21 2/) with equal probability. The sequence of information symbols (is statistically independent. 4 Determine and sketch the power density spectrum of u(t) when ‘A (0<18T) 60-(0 (eae b Repeat (a) when (ave (m/T) (O f.