2 MODULATION AND DETECTION Analog and digital modulation and detection are essential functions in commu- nication systems. While new methods of performing these functions continue to appear, certain schemes have matured over the years and are employed in many RF applications This chapter deals with basic modulation and detection techniques used in today’s RF transceivers, describing the merits and limitations of each method and emphasizing the implications in the design of RF ICs, Following a brief de- scription of important aspects of modulation schemes, we review analog amp! tude, phase, and frequency modulation. Next, we study binary and quadrature digital modulation techniques, focusing on their role in RF communications. Finally, we describe the power efficiency of various modulation schemes and consider noncoherent detection. 3.1 GENERAL CONSIDERATIONS The transmitted waveform in RF communications is usually a high-frequency carrier modulated by the original signal. Various reasons exist for modulation: (1) in wired systems, coaxial lines exhibit superior shielding at higher frequen- cies; (2) in wireless systems, the antenna size should be a significant fraction of the wavelength to achieve a reasonable gain; (3) in most cases, the communica- tion must occur in a certain part of the spectrum because of FCC regulations; and (4) in some applications, modulation allows simpler detection at the receive end in the presence of nonidealities in the communication channel In the context of modulation, it is common practice to define two types of signals, A “baseband” signal is defined as one whose spectrum is nonzero in the vicinity of = 0 and negligible elsewhere [1] [Fig. 3.1(a)], for exam- ple, the signal generated by a microphone or a video camera. A “passband”Sec.3.1 General Considerations 55 0 o =O, 0 +0 © @ (by Figure 3.1 (a) Baseband and (b) passband signals. signal! is a waveform whose spectrum is nonzero in a band around a “carrier” frequency «- and negligible outside this band [1] [Fig. 3.1(b)]. While mathe- matically imprecise, these definitions are adequate for our purposes because in RF design the signal bandwidths of interest are usually much less than the cartier frequencies. Modulation converts a baseband signal to a passband counterpart. From another point of view, modulation varies certain parameters of a carrier (e.g., a sinusoid) according to the baseband signal. These two definitions are equivalent ‘because a passband signal can always be expressed as x) = a(t) cosfwct + (1), GBD where a(t) and @(t) are generally functions of time [1]. In other words, we begin with a periodic carrier signal A cos(w.t + @) and vary its amplitude or phase to perform modulation. In (3.1), the argument wet + 0(t) is called the “total phase” and @(t) the “excess phase.” We also define the instantaneous fre- quency as the time derivative of the phase: «»-+d6//dt is the “total frequency” and d0/dr is the “excess frequency” or the “frequency deviation.” Note that. depending on the modulation scheme, a baseband signal with bandwidth B may occupy a different bandwidth when converted to a passband waveform. This will be clarified later. ‘The inverse of modulation is demodulation or detection, with the goal being to extract the original baseband signal with minimum noise, distortion, ISI, etc. Thus, as depicted in Fig. 3.2, a simple communication system con- sists of a modulator/transmitter, a channel (e.g., air or coaxial cable), and a Transmitter Receiver Modulator Channel Demodulator Figure 32 Simple communication system. 1 The term“ low-pass. vandpass” isalso used. Passband is consistent with baseband and bandpass with56 Chap. 3. Modulation and Detection receiver/detector. Since the overall performance depends on the design of both the modulator and the detector, modulation schemes are usually studied in conjunction with proper demodulators. We refer to the combination of the two as “modems.” Important Aspects of Modems Inanalyzing the system shown in Fig.3.2, the primary objective is to find the “quality” (e.g., SNR) of the output of the detector in the presence of attenuation and interference in the channel as well as noise at the input of the detector. For a given channel attenuation (also called “path loss”) and transmitted power, the quality depends on the type of the modem. Thus, if the modem achieves higher tolerance of noise, then either the transmitted power can be reduced, providing longer “talk time” in portable devices, or a higher path loss can be accommodated, allowing communication over a longer distance. Another important property of modulation schemes is the required band- width, Assuming a certain bandwidth for the baseband signal, we calculate that occupied by the modulated waveform in each case. Called the “spec- tral efficiency,” this criterion plays an important role in choosing a modem for limited-bandwidth applications such as wireless networks. ‘The third attribute of modulation techniques that is of great concern in RF design is the “power efficiency,” a reference to the type of power amplifier that can be used in the transmitter. As we will see later in this chapter, some mod- ulated waveforms can be processed by means of nonlinear amplifiers without spectral spreading or loss of information whereas some others require linear amplifiers. Since nonlinear power amplifiers are generally more efficient than their linear counterparts (Chapter 9). it is desirable to employ a modulation scheme that lends itself to nonlinear amplification. In summary, the choice of modems is based on three important aspects: quality, spectral efficiency, and power efficiency. Additive White Gaussian Noise Channel _ In order to compare the “noise tolerance” of different receivers, we assume the transmitted signal power is constant and the received signal is corrupted by additive white Gaussian noise (AWGN) having a power spectral density equal to No/2 (Fig. 3.3). The AWGN assumption proves adequate in taking into account the inherent noise of the receiver. However, the noise introduced by adjacent interferers is generally not white and must be considered separately, Furthermore, the noise in some demodulators (e.g., envelope detectors) is multiplicative rather than additive. Nol2 Figure 3.3 White spectrum assumed for additive Gaussian noiseSec.3.2. Analog Modulation 57 3.2 ANALOG MODULATION 3.2.1 Amplitude Modulation For a baseband signal x(t), an amplitude-modulated (AM) waveform can be constructed by writing (3.1) as xav(1) = Ac[1 + mxpa(t)] cos wet, where ‘m is called the “modulation index.” Illustrated in Fig. 3.4 is a method of gen- erating an AM signal along with the corresponding waveforms and spectra: multiplication by cos « Oe @ w Xealt) Xey( io & o5 > i | © “Om Om O-Oq Det Om, ) Figure 39 Narrowband FM with (a) random and (b) sinusoidal modulation.See.3.2. Analog Modulation 61 A special case of narrowband FM that proves quite useful is for a sinu- soidal baseband signal, ¢.g., x(t) = Am CO8@,t. In this case, we have m XpMNa() % AcCoS@.t — AmAc— sin wet Sin ®t (G3) On AmAcm = A, cosa,t — COS(@. — Wm ay AmAc 7 cos(we + Wm)t- (3.6) 20m Illustrated in Fig, 3.9(b), the resulting spectrum consists of impulses at tw, and “sidebands” at =£(d- + am) Equations (3.5) and (3.6) yield three important points, First, for the nar- rowband FM assumption to hold, we must have mAm/@m < 1 rad. Second, as @n increases, the magnitude of the sidebands decreases; i.c., higher mod- ulating frequencies have less weighting in the FM signal. Third, if xry(r) = Ac coset — (Am /dm) SiN yt], then the maximum frequency deviation is equal to mAm. Note that the spacing between the sidebands and the carrier is equal to @», and has no relationship with the maximum frequency deviation. While a useful concept in many communication circuits, narrowband FM has limited application in modems because of its low SNR, an issue described below. As the next step toward a realistic case, we consider an FM signal with a sinusoidal baseband waveform, but without the restriction m Am /@m 1 rad. If xpa (1) = Am COS@mt, then Xpy4(t) = Ac cos[ect — (MAm (tm) Sin Ont] This expression can be expanded as $5 xem(t) = Ac D> In(B) cose + rom)t, G7) where J,(+) denotes the nth order Bessel function of the first kind and 6 mAm/@m. If B < 1 rad (narrowband FM), then Jo(B) * 1, Jai(B) © 8/2. and J,(B) * 0 for |n| > 1. Thus, xpyy(t) = Ac cos wet + (A¢B/2) cos(w. + mt —(AcB/2) co8(w2e—m)t, the same result as thatin (3.6). As f approaches and exceeds unity, the coefficients J,() become significant for || > 1, intro- ducing components at @. + né&,. Shown in Fig. 3.10 are examples of the FM spectra for 8 = 1 and f = 2. More general cases of FM such as multiple- frequency or Gaussian modulation are treated in [2]. An important conclusion to be drawn from the above analysis is that the bandwidth of FM signals depends on m, Ay, and @,,. In fact, if the bandwidth, Bey, is defined as that containing 98% of the signal power, then Bry ~ 2(B + 1) Baga, where Byg is the bandwidth of the baseband signal [1]. This is called Carson’s rule.62 Chap.3_ Modulation and Detection Figure 3.10 FM spectra for two different values of f. In Eq. (3.6), we noted that as the frequency of the modulating signal in- creases, the magnitude of the corresponding component in the FM waveform decreases. The FM demodulator thus compensates for this rolloff by intro- ducing larger gain at higher frequencies, as can be seen in the differentiator of Fig. 3.8. However, this leads to amplification of noise at high frequencies, thereby degrading the overall SNR. For thisreason, most FM transmitters shape the baseband spectrum by amplifying the high-frequency end (“preemphasis”) (Fig. 3.11). In the receiver, the detected signal is applied to a low-pass filter to cancel the preemphasis (“deemphasis”), at the same time suppressing high- frequency noise. This method does not usually increase the required bandwidth significantly if the original baseband spectrum has a low-pass shape A [| Nhe o o a“ oO o fa) rose |= A shasi: emphasis a. (b) Figure 3.11 Preemphasis and deemphasis in FM,Sec. 3.3 Digital Modulation 63 For an FM detector without preemphasis and deemphasis, itcan be shown that SN Row SNRin = 66(B +1 (3.8) where Vp is the peak value of xpn(1) [2]. With preemphasis and deemphasis, SNRouc _ SNRn = 2B" ro+0 (2) ® VE 1 (3.9) where B is the bandwidth of the baseband signal and f; is the —3-dB corner frequency of the low-pass filter used in deemphasis. For typical applications, (3.9) yields a value 10 to 15 dB higher than (3.8) [1]. In both cases, the perfor- mance strongly depends on fi, revealing a trade-off between signal quality and bandwidth efficiency. ‘The last attribute of FM systems to be considered is the power efficiency. Since an FM signal has a constant envelope and carries the information in the zero crossings, it can be processed by a nonlinear power amplifier with no corruption of the baseband signal or increase in the required bandwidth. Thus, FM signals are well suited to high-efficiency (inevitably nonlinear) PAs, 3.3 DIGITAL MODULATION In “digital” RF systems, the carrier is modulated by a digital baseband signal. As explained in Chapter 4, digital modulation offers many advantages over analog modulation and is widely used in wireless systems. ‘The digital counterparts of AM, PM, and FM, are called amplitude shift keying (ASK), phase shift keying (PSK), and frequency shift keying (FSK), respectively. Figure 3.12 illustrates examples of these waveforms for a binary baseband signal. In RF applications, PSK and FSK find wider usage than does ASK because of their lower sensitivity to amplitude noise. As mentioned in Section 3.1, analysis of a modem entails quantifying three parameters: signal quality, spectral efficiency, and power efficiency. The quality of digital signaling methods is expressed in terms of the bit error rate (BER), defined as the average number of erroneous bits observed at the output of the detector divided by the total number of bits received ina unit time. Thus, the goal is to calculate the probability of error in the presence of noise and other interferers.64 Chap. 3 Modulation and Detection ASK PSK “NA A = FSK AAA © Figure 3.12 (a) Amplitude, (b) phase, and (¢) frequency shift keying 3.3.1 Basic Concepts Analysis of digital modems requires a number of conceptual tools that are not used in the more familiar analog modulation systems. We briefly review some of the essential concepts here. Binary and M-ary Signaling A natural choice of the digital baseband signal is a binary waveform, which can be expressed as xpa(t) = Dnpe —nTh), (3.10) where b,, is the bit value in the time interval [n7, (n + 1)T)] and assumes one of two values, ¢.g., 0 and 1 or —1 and +1. For a rectangular pulse shape, the receiver may simply decide whether the (peak) value of the waveform is less or greater than a threshold to detect the bit value. In practice, it is sometimes advantageous to employ multilevel “digital” signals to modulate the carrier, usually because bandwidth requirements can be relaxed. Shown in Fig. 3.13 is an example where the original binary data stream is subdivided into groups of two bits, and each group is converted to one of four possible levels by means of a D/A converter. We note that the resulting waveform exhibits fewer transitions per unit time but requires higher amplitude resolution in the detector. Called “M-ary” signaling, such a scheme can also be represented by (3.10) if by assumes all of the discrete levels. To avoid confusion with the binary case, each transmitted level here is called a “symbol” rather than a bit. In Fig. 3.13, the “symbol rate” is one-half the original bit rate.Sec.3.3. Digital Modulation 65 Binary Data 0 1 0 4 1 0 1 060 0 1.1 00 we SCS Data of oF Figure 3.13 Four-level digital representation of a binary data stream. Basis Functions With the baseband signal described by (3.10), we now consider the representation of digitally modulated waveforms. As an example, the binary FSK signal of Fig. 3.12(c) can be written as xesk(t) = Accosant, if by = 0, (.A1) = Accosant, if By = 1. (3.12) This indicates that every transmitted bit can be represented by an inner product, XFsK(t) = aigilt) + argr(t) G13) = [a a2] [Ai gai], G14) where [1 a2] = [0 Ac] or [A¢ 0], $i (t) = cosa yt, and g2(t) = cosw2t. In other words, x-s«(t) is expressed as a linear combination of two functions. ‘The idea of representing a waveform as a linear combination of “basis functions” should be familiar from Fourier series: a function x(¢) (with finite energy) can be written as 20 x (1) = YY aq cosnot + S> by sinnost, (3.15) nao n=O over the interval to <¢ < ty +7, where @ = 1/T. An important property of the Fourier series is the “orthogonality” of the basis functions, namely. # f cos not cosmat = 0, (3.16) 0 1 f cos net sinmet = 0, (17) 0 r f sinnotsinmat = 0, (3.18) 0 forn Am66 Chap. 3 Modulation and Detection With the above observations, we now state that each symbol in a digitally modulated waveform is represented by a linear combination of orthogonal basis functions, X(t) = angi (t) + ardo(t) + +++ + andn(t), G19) where Ts [ dm (NGx(Qdt = 0 G.20) Jo for m # k and Ts is the symbol period. The number N is called the “dimen- sion” of the set of basis functions. ‘The reader may wonder about the motivation behind the foregoing de- velopment. We justify this by two reasons. First, expressing waveforms as a linear combination of functions allows a unified treatment and fair compar- ison of many seemingly different modulation schemes. Second, orthogonal basis functions form a “minimum set” for representing waveforms, and, more importantly, they can be used in optimum detection of signals Signal Constellations For dimensions below 4, we can visualize the modulated waveforms in terms of the coefficients in their inner products, That is, in x(t) = [ay a2 + -]-[4i(t) @2(¢) ---], we simply ignore the basis functions and plot all possible values of the vector [or a ---] in cartesian coordinates. Thus, for the binary FSK signal of Fig. 3.12(c), we obtain the plot shown in Fig. 3.14(a). Called a “signal constellation” or a “signal space,” such a plot proves valuable in understanding various effects in digital modems. Oe, +A 2» Ae 0 tag OH (a) o) Figure 3.14 Signal constellations of (a) PSK, (b) ASK, 2 The basis functions are usually normalized so as to have unit energy over Tp, but here we ignore this detailSec.3.3 Digital Modulation 67 As another example, consider the binary ASK waveform of Fig. 3.12(a), where the modulated signal corresponding to each bit is expressed as Xask(t) = Accosaet, ifb, = 1, (321) = 0, if by = 0. (3.22) Consequently, x4sx(t) = aids (t), where a assumes a value of 0 or Ac and $1(t) = cos (eg. Ay = +1 and Az = —1), corresponding to x(t) = pi(r) and x(t) = p2(t), respectively. ‘The decision threshold is set at (Ay + A2)/2. In the presence of noise, on the other hand, the sampled value is equal to Ay + n(Tj) or A> + (Ts), where n(Tj) is the noise amplitude at the output of the integrator at ¢ = 7}. Note that (7) may be not be white, but it is still Gaussian because of the linearity of the correlator. Now recall from probability theory that if two independent random vari- ables are added, their probability density functions are convolved. Conse- quently, as illustrated in Fig. 3.28, the quantities Ay + n(Tj) and Az +n(T;) exhibit two overlapping Gaussian distributions, indicating that an error occurs if A +n(Zp) exceeds the threshold and p(t) is transmitted or if Az + (T;) falls below the threshold and p2(t) is transmitted. Since the probability that76 — Chap.3_ Modulation and Detection POF with no Noise POF of Noise . A; Az Sampled 0 ‘Sampled Value Value PDF with Noise » LV\.. Ay | Az Sampled | Value Threshold Figure 328 Calculation of PDF for binary data with additive noise. pa(t) is transmitted is equal to 1/2, the probability of the first error event is 1 1 —u- Ay Pa => exp du, (332) a a ) where o is the standard deviation of n(t). The second error event has the same probability, and hence the total probability of error is Pp = 2P.1. With a change of variable, v = (u — Ay)/on, we have es . ae z, f —*y dv (3.33) assy jtony Vom 2 oe = o(4—*), 634) 26, where Q(-) is Q(x) = fe + du (3.35) While not available in closed form, this function can be approximated as Q(x) ® 1 L xp—, ; Era (3.36) for x > 3. Note that the result in (3.34) holds for any filter, although (Az — ‘A,)/Qoq) depends on the type of the filter. The argument of the Q function in (3.34) is the ratio of two quantities (signal and noise) at the output of the integrator. It is more convenient to express this ratio in terms of signal and noise at the input of the detector. This is possible for a correlator (matched filter) implementation if we make two observations, First, linear superposition implies that the quantity Ay — Aj isSec. 3.3 Digital Modulation 77 the response of the receiver to an input x(t) = p;(t) — pp(t) (although such an input does not occur in actual transmission), Second, since the filter (or correlator) is matched to pi(t) — pa(t), if x(t) = pi(t) — po(t) is applied, the SNR at the sampling instant is equal to (Az — A1)’/a?, which reaches a maximum given by (3.30). Thus, (A? — ;)?/o2 = 2E4/Np and, hence, (337) Equation (3,37) is an important result, suggesting that the error rate de- pends on only the (difference) signal energy and the noise spectral density. We emphasize the conditions under which (3.37) holds: additive white Gaussian noise channel and matched-filter (coherent) detection. The derivations considered thus far apply to all binary modulation schemes. We now study the implications of the above analyses in specific modems. BPSK In BPSK, the binary baseband data sclects one of two opposite phases of the carrier. Illustrated in Fig. 3.29(a), the modulated signal can be written as xppsx(1) = Ac cos(w.t + 9), where ¢ = 0 or 180°. To obtain the constellation, we write xg psx (t) = a coset, where @ = +A, ot —A, [Fig. 3.29(b)]. Since pi(t) = —p2(t), the correlating signal in the detector is Pi(t) — po(t) = 2p,(t) = 2A-cos wet [Fig. 3.29(c)]. Thus, in the absence of noise, the output of the integrator at t = T;, is equal to Ti, Vint = 2A? cos? wtdt. (3.38) hb For large @¢, Vin: * £A27;, yielding the baseband binary data. ‘To calculate the error rate with the aid of the results in (3.31) and (3.37). we recognize that +0 Eq = f (pitt) — poyPat 3.39) i, = (2Ac cos w.t)*dt (3.40) 0 = 24%, G41) and hence (3.42)78 — Chap.3- Modulation and Detection Baseband Pinan Xepsk(t) =§ —o——_L___» _» . “Ag 0 +g P,{t) =A, coset @ o) Te To xppsk(t) (2) fi in Baseband 6 [~-] Data 2A, coset © Figure 3.29 BPSK (2) modulator, (b) constellation, and (c) demodulator. ‘To make fair comparisons with other modulation schemes, we define the (av- erage) energy per bit as E, = A?7),/2, thus obtaining PeBPsK = o( 22) (3.43) Note that £ contains both the amplitude and the period of the signal, revealing that the error rate can be lowered by increasing the signal power or decreasing the data rate. The quantity £},/No is an indication of the signal-to-noise ratio in the receiver. We find the spectrum of a BPSK waveform by making two observations. First, the modulator of Fig. 3.29(a) can be modified as shown in Fig. 3.30, where the baseband data toggles randomly between +1 and —1 and is multiplied by Ac COS Wt: XBPSK = Xpp(t).Ac coswct. (3.44) Thus, the spectrum of xgpsx (t) is simply that of xg x(t) translated to tee Second, a random binary waveform, x(t), employing pulse shapes +p(t) and — p(t) to represent ONEs and ZEROs, respectively, has the following power spectral density, Sx(o) = A1PwP, (3.45) h where P() is the Fourier transform of p(t) {1}. If p(t) is a rectangular pulse with duration 7, and unity amplitude, then P (w) = [2sin(@T}/2)]/w, and in?[(w + w.)Ts/2] % AB sin?[(w — @)T;/21 (@ + we)* Ts (@ = @)* G46)Sec. 3.3 Digital Modulation 79 “1 xo0lt) Ff Qe rence) a i A, cosw,t Figure 330 Simplified BPSK modulator. The power efficiency of BPSK is discussed in Section 3.4. BFSK _ In BFSK, the binary baseband data selects one of two carrier frequencies with equal amplitudes [Fig. 3.31(a)].“ The modulated signal is writ ten as xgrsx(t) = a; COs ot + a2 COs wnt, where [a a2] = [0 Ac] or [Ac 0] For the two basis functions to be orthogonal over one bit period, we must have T f cos ait cosertdt = 0. (347) f For a +) >> «1 —an, the above equation reduces to [sin(@; — 9) To] /(w: — @) = 0. Thus, (@; — @))T), = nz, and for minimum spacing between «; and @), O — w = 1/Ts or fi — fp =1/Q2T). Baseband a nea coer +Ac xersk(t) ,{t) =Accosoat (—/ Decision Boundary 7 @ ) Baseband Data A. cos@;t-A, cosmzt © Figure 331 BFSK (a) modulation, (b) constellation, (¢) detection. 5 This approach is called “discontinuous” FSK because the phase changes abruptly.80 Chap.3 Modulation and Detection In contrast to BPSK, orthogonal BFSK is a two-dimensional scheme, exhibiting the constellation in Fig. 3.31(b). Here, p:(t) = Accosat and Pr(t) = Ac cos apt; that is, p(t) # —pp(t). As we will see below, this yields interesting differences between the error behavior of BPSK and BFSK. Shown in Fig. 3.31(c), a BFSK detector correlates the input with p(t) — pr(t) = Ae cost — Ae cosat. Consequently, the output of the integrator at f = J, in the absence of noise is equal to A?7},/2 for Xgrsx = Ac Cos wt and —A27j,/2 for xgrsx = Ac cos w2t. The bit error rate for coherent BFSK detection is calculated with the aid of (3.31) and (3.37). Since pj(t) and pp(t) are orthogonal, 1 Ea = | (pitt) + pi@lde (3.48) I >, > yt z (3.49) P=Q (3.50) As with BPSK, we define the average energy per bit as E, = A?7}/2, thereby obtaining E PoBeskK = o( ®). (3.51) Comparison of (3.43) and (3.51) indicates that, for a given probability of error and noise density, the bit energy in BFSK must be twice that in BPSK. What is the reason for this difference? From the constellations of Figs. 3.29(b) and 3.31(b), we observe that for a given value of carrier amplitude, Ac, the amount of noise that can corrupt a bit so much that the correlator output crosses the threshold erroneously is higher in BPSK than in BFSK. In other words, the minimum distance between the points in the constellation is greater in BPSK. From another point of view, recall that in SN Rmax = 2Ea/No. the value of Eq reaches its maximum if p;(t) = —p2(t), which is the case for BPSK but not for BFSK. The above difference is often stated as “BPSK has a 3-dB advantage over BFSK.” Nevertheless, BFSK is widely used in low data rate applications (such as pagers) where Ey, can be maximized by allowing a long T;,. The popularity of BFSK stems from the simplicity of its detection, especially in noncoherent, systems (Section 3.5), and its power efficiency (Section 3.4). ‘The spectrum of FSK signals is generally difficult to calculate [1]. From Carson’s rule, the 98% bandwidth is approximately equal to Br = 2(Af + 1/T,), where Af = fy — fr [1]. In orthogonal FSK, Af = 1/(27;) and Br ©3/Tp.Sec. 3.3 Digital Modulation 81 The power efficiency of BFSK is discussed in Section 3.4. 3.3.3 Quadrature Modulation In many applications, it is beneficial to subdivide a binary data stream into pairs of two bits and represent each pair with one of four levels before performing modulation. For example, bits by, and by..1 can be impressed upon a single carrier as X(t) = by Ac cos wet — Bmsidc sin act. This is possible because cos «7. and sin wt are orthogonal functions. Called “quadrature modulation” or “quadrature multiplexing,” this op- eration is illustrated in Fig, 3.32, where a serial-to-parallel (S/P) converter sep- arates consecutive bits into one stream for the upper arm and another for the lower arm. Since each group of two bits (one in each arm) constitutes a symbol, the symbol rate is half the bit rate (Section 3.3.1), indicating that the required bandwidth is half that of BPSK. This is the principal reason for the wide use of quadrature modulation, <—-® Binary coset Baseband Data sino gt =~ a Binary Baseband Data Figure 332 Quadrature modulation. To obtain the constellation, we assume bits by, and bm+1 are rectangular pulses with a height -+1 and write the modulated signal as x(t) =e coset + a sin.f, where ey and a can cach take ona value of +Ac or —Ae. The constellation is shown in Fig. 3.33.82 xapsk(t)} Chap. 3 Modulation and Detection Figure 333 Signal constellation for quadrature modulation, Quadrature modulation encompasses two broad categories: quadrature phase shift keying (QPSK) and minimum shift keying (MSK). The first includes more specific types such as offset QPSK (OQPSK) and 7/4-QSPK, and the second has a widely used subset, Gaussian MSK (GMSK). Why are there so many types of quadrature modulation? Recall that three parameters determine the performance of a modulation technique for RF applications: BER, spectral efficiency, and power efficiency. The above schemes exhibit different trade-offs among these parameters. QPSK _Ifin (3.52), the bit waveform is a rectangular pulse, a QPSK sig- nal is obtained. From another point of view, in analogy with BPSK, one of four phases of a sinusoid is selected according to the symbol; that is, xgpsx(#) = V2Ac cos(wet + k/4), k =1,3,5,7. Coherent detection of QPSK signals can be performed as shown in Fig, 3.34. Here, the circuit correlates the input with cos wf and sin w +f Youre — mtoas]. (3.56)Binary Data Sec. 3.3 Digital Modulation 89 cos@yt coset Converter sino;t — sinogt Figure 343 MSK modulation and signal constellation, where the summation represents the binary baseband signal; that is, by, = +1 and p(:) is a rectangular pulse of width 7. InGMSK, on the other hand, p(t) isa smoother function. Note that (3.56) can be viewed as frequency modulation and, with a rectangular p(t), as FSK. In GMSK, the pulse shape p(t) is obtained by passing the baseband rectangular pulses through a filter with a Gaussian impulse response. h(t) = exp(—ar*). As depicted in Fig. 3.44, to obtain the output of the filter, x5 4(¢) * h(t), we can calculate (dxgy/dt) x h(t) and integrate the result. This method shows that for a narrow Gaussian response (large @), the filter output is close to a rectangular pulse with width 7;,, and as h(1) becomes wider, so does the output, The derivation of the GMSK power spectral density is difficult, but we intuitively expect the spectrum to decay faster if the time-domain modulating pulse, p(¢), becomes smoother. For example, if @ = 1, the 99% bandwidth is less than 1.2/7) [10].90 Chap. 3 Modulation and Detection To Gaussian —_— Filter xep(t) PC) | Frequency t dxpe(t) a \ at 7 aad apt) dt P(t) (on Do > “LN Figure 3.44 GMSK modulation. While our analysis indicates that the lower the value of e, the narrower the spectrum, the choice of a depends on another factor as well: intersymbol interference, In contrast to previous modulation schemes, GMSK allows the modulating pulse to be wider than 7), thus suffering from ISI. As decreases, ISI becomes more significant. In typical systems, ois in the vicinity of 0.3, requiring ISI mitigation in the receiver. The power efficiency of MSK and GMSK is described in Section 3.4. ‘The spectra and error probabilities of various modulation schemes are given in (11, 12, 13). 3.4 POWER EFFICIENCY OF MODULATION SCHEMES In our study of various modulation techniques thus far, we have frequently mentioned that power efficiency is an important aspect. In this section, we deal with this issue. 3.4.1 Constant- and Variable-Envelope Signals A modulated waveform x(#) = A(#) cos[wct +¢(t)] is said to have a constant envelope if A(r) does not vary with time. Otherwise, we say the signal has a variable envelope.Sec. 3.4 Power Efficiency of Modulation Schemes 91 Constant- and variable-envelope signals behave differently in a nonlinear system. Suppose A(t) = A, and the system exhibits a third-order memoryless nonlinearity: y(t) = o3x3(t) + --- (3.57) = asA3cos [oct + o(t)] + ++ (3.58) _ @3A2 =& (3.59) The first term in (3.59) represents a modulated signal around w = 3w.. Since the bandwidth of the original signal, Ac cos{at + (¢)], is typically much less than «-, we note from Carson's rule that the bandwidth occupied by cos[3a.t-+ 3¢(t)] is also quite small. Thus, the shape of the spectrum in the vicinity of « remains unchanged. Now consider a variable-envelope signal applied to the above nonlinear system. Writing x(t) as x(t) = x1(0) cos wet — xg(t) sin wet, (3.60) where x(t) and xg(t) are the baseband I and Q components, we have YD) = as[xz (2) coset — x(t) sinw.tP + +++ (3.61) cos3eret + 3c0s at = 3x7) 3.608 3act + 3sinace = 5x9) (3.62) ‘Thus, the output contains the spectra of x} (1) and x}(1) centered around ae. Since these components generally exhibit a broader spectrum than do x; (1) and Xo(t), we say the spectrum “grows” when a variable-envelope signal passes through a nonlinear system. Anexception to the above observation occurs if the baseband pulse shape isrectangular. Ashigher powers ofa rectangular waveformare still rectangular, the modulated signal spectrum has a sinc*w7}, shape for any nonlinearity. 3.4.2 Spectral Regrowth Consider the simple transmitter illustrated in Fig. 3.45. Here, a low-pass filter precedes the modulator to limit the bandwidth of the signal, thereby suppress- ing spectral leakage into adjacent channels. The PA amplifies and butters the result, delivering the required signal level to the antenna. What is the effect of filtering on a digitally modulated signal? Intuitively, we expect limiting the bandwidth tends to smooth out the abrupt transitions in the time domain, For example, a QPSK waveform appears asshown in Fig. 3.45 after filtering, exhibiting greater variation in its envelope as the filter bandwidth92 Chap. 3 Modulation and Detection PA xoo(t)-—»] LPF > xoult) 180° 180° 30° Unfiltered QPSK Filtered aPsk Figure 348 Effect of bandlimiting on the envelope of a QPSK waveform. decreases. We also note that envelope variation due to filtering becomes more pronounced at greater phase steps. What is the implication of envelope variation? The one-to-one cor respondence between the band-limited spectrum and the time-varying en- velope indicates that if the power amplifier is to maintain the spectrum to the limited bandwidth, then it must also amplify the envelope variations lin- early. This can be seen by expressing the modulated waveform as x(t) = x; (1) cos wt — xg(¢) sin wt. If the PA exhibits significant nonlinearity, then the shape of x/(f) and xg(t) is not preserved and the spectrum is not limited to the desired bandwidth. ‘This effect is called “spectral regrowth” [14, 15] and can be quantified by the relative adjacent channel power [15] The above discussion can be summarized as follows. Abrupt phase changes in a digitally modulated waveform, for example, QPSK, result in envelope variations if a filter limits the bandwidth, Such variations in turn require a linear power amplifier so as to avoid spectral regrowth. As explained in Chap- ter 9, linear PAs are typically less efficient than nonlinear PAs. For a 40% efficieney—which is rarely achieved in linear PAs—and a 1-W output power, the PA consumes a total of 2.5 W, a considerable amount with respect to the power dissipation of the rest of a portable phone. Nonlinear PAs, on the other hand, exhibit efficiencies as high as 60%. Thus, it is desirable to employ mod- ulation schemes that do not experience spectral regrowth when processed by nonlinear amplifiers. This aspect of modulation formats is called power effi- ciency. A simple example of power-efficient modulation is FM. From xpyy (1) = Ac cos[et +m J xpp(t)dt], we see that FM waveforms have no abrupt phase change and exhibit a constant envelope. FM and FSK signals can therefore be amplified by means of nonlinear PAs with no spectral regrowth.See.3.5 Noncoherent Detection 93 From the illustration of Fig. 3.45, we conclude that filtered QPSK is quite susceptible to spectral regrowth. For example, if root raised-cosine filtering is used in the baseband so as to achieve a narrowband, the resulting envelope variation requires a relatively linear power amplifier [17, 16]. This drawback is less severe in OQPSK and 2 /4-QPSK by virtue of smaller phase steps allowed in these schemes, More power efficient than the QPSK family are MSK signals because their constant envelope allows processing by nonlinear PAs, However, MSK signals typically require a wider band than OPSK waveforms with raised-cosine filtering. Insummary, digital modulation schemes exhibit a trade-off between spec- tral efficiency and power efficiency. As the signal bandwidth is more limited— by filtering or pulse shaping—the power amplifier must achieve a higher lin- earity so as to avoid spectral regrowth. Different wireless systems have aimed at one of these extremes: pulse-shaped QPSK modulation with narrow chan- nels but linear PAs or MSK modulation with wider channels but nonlinear PAs (Chapter 4) 3.5 NONCOHERENT DETECTION In Section 3.3.1, we saw that matched filters, an example of which are coherent detectors (correlators), provide the highest signal-to-noise ratio and hence the lowest probability of error. Coherent detectors, however, require that the phase of the local oscillator in the receiver bear a strict relation to that of the received signal. Called “carrier recovery.” establishing such a phase alignment necessitates substantial circuit complexity and becomes especially difficult at lowsignal levels in the presence of interferers and signal fading, For this reason, many RF systems employ noncoherent detection despite its somewhat inferior performance. Noncoherent FSK Detection _In Section 3.3.1, we briefly studied a non- coherent FSK detector. Shown in Fig. 3.46, this configuration employs two bandpass filters (for binary modulation), each followed by an envelope detec- tor, to determine which frequency is received. Since « — «2 is typically much less than « and a, the input spectrum is first translated to a lower frequency so as to relax the Q and the precision of the center frequency required of the bandpass filters. Derivation of the probability of error for this detector is rather involved. We only state the result for the practical case of sufficient difference between @; and @2 so that their spectral overlap is negligible [1] ee = 3 °PO7,B,Ny P. (3.63)94 Chap. 3 Modulation and Detection Figure 346 Noncoherent FSK detection, where B, is the (effective) bandwidth of each bandpass filter and E;, is the bit energy, Ey = A27;/2. If B, = 1/T,, then -Ey 1 Py Sneepe” 2° IN (3.64) For error rates on the order of 10°, (3.64) requires an E,/No that is 1.5 dB greater than that in coherent FSK detection [Eq. (3.51)]. In many applications, the lower complexity justifies this loss in the performance. Differential Phase Shift Keying _ While in FSK itis possible to determine the frequency corresponding to each bit (using bandpass filters or frequency de- modulators), in PSK the phase relates to the time origin and has no “absolute” meaning. For example, a 90° phase shift in a QPSK waveform converts the constellation to a similar one, but with all the symbols interpreted incorrectly. ‘Thus, simple PSK waveforms cannot be detected noncoherently. However, if the information lies in the phase change from one bit (or symbol) to the next, then a time origin is not required and noncoherent detection is possible. This is accomplished through “differential” encoding and decoding of the baseband signal before modulation and after demodulation, respectively. Let us consider binary differential PSK (DPSK). The rule for differen- tial encoding is that if the present input bit is a ONE, then the output state of the encoder does not change. and vice versa. This requires an extra starting bit (of arbitrary value). The concept can be better understood by consider- ing the implementation depicted in Fig. 3.47(a). An exclusive-NOR (XNOR) gate compares the present output bit, Dow(m 7), with the present input bit, Diq(mTp), to determine the next output state, Dou l(t + 1)Th] = Din(MT) ® Dow(mTs). (3.65)Sec. 3.5 Noncoherent Detection 95, or [_— DoE Paulnty) P| cK Input Data o1110011014 EncodedData 10000100011 DecodedData 0111001101 © Figure 3.47 (2) Differential encoding, (b) differential decoding, (c) example of encoded and decoded sequence. implying that if Din(m7) = 1, then Dou[(m + 1)T] = Dou(mT,), and if Din(mT) = 0 then Dou[(m + 1)T}] = Dow(mT,). The extra starting bit mentioned above corresponds to the state of the flipflop before the data sequence begins. ‘The DPSK signal received by a system can be decoded by a reverse op- eration, depicted in Fig. 3.47(b). It can be easily shown that phase shifts in the encoded signal do not alter the differentially decoded output of the receiver. Differential signaling does not provide the lowest bit error rate because it is not based on matched filtering. From another point of view, in coherent demodulation the signal is compared with a low-noise reference (oscillator), whereas in differential detection two noisy signals are compared with each other [12]. It can be shown that the probability of error for DPSK is 1 -E P. (3.66) where E, = A?7},/2. This result indicates that for an error rate of less than 10~*, DPSK requires approximately 3 dB higher Ey/Np than does coherent PSK [Eq. (3.43)].