Scribd
Upload
23 views

Bandpass Modulation and Demodulation

Bandpass modulation and demodulation

Uploaded by

Victor Tedesco
Copyright
© © All Rights Reserved
Available Formats
Download as PDF or read online on Scribd
23 views

Bandpass Modulation and Demodulation

Bandpass modulation and demodulation

Uploaded by

Victor Tedesco
Copyright
© © All Rights Reserved
Available Formats
Download as PDF or read online on Scribd
You are on page 1/ 70
CHAPTER 3 Bandpass Modulation and Demodulation From other sources ; 335 y f m A K 2 — Ba » a, as se 3 AO mn 125 8 °C.3.2 Signals and Noise ™ is taken to be the noise within the signal space, or the projection of the no, components on the signal coordinates i(t), - - » » W(t), and ise v(t) = n(t) — A(t) Bg is defined as the noise outside the signal space. In other words, f(#) may be thou, of as the noise that is effectively tuned out by the detector. The symbo, ty represents the noise that will interfere with the detection process. We can ex, . . ‘Press the noise waveform, n(1), as follows: 3 ” n(t) = D mbt) + ACE) 6.1 a where T ; = z ff moo dt for all j Gig) and 7 a= [i aongo ae 6.19) The interfering portion of the noise, (1), expressed in Equation (3.15) will hence. forth be referred to simply as n(t). We can express n(t) by a vector of its coef. ficients similar to the way we did for signals in Equation (3.6). n = (m,M2,...5 Mn) 3.20) where nis a random vector with zero mean and Gaussian distribution, and where the noise components nm, (i = 1, . . . » N) are independent. 3.2.2.4 Variance of White Noise White noise is an idealized process with two-sided power spectral density equal to a constant, No/2, for all frequencies from —% to +2, Hence the noise variance (average noise power, since the noise has zero mean) is o = var (aco =f () df=% 62) Although the variance for AWGN is infinite, the variance for filtered AWGN is finite. For example, if AWGN is correlated with one of a set of orthonormal functions ,(t), the variance of the correlator output is given by 5 o? = var (n) = E {0 nO) ai| } = Xe 62) The proof of Equation (3.22) is given in Appendix C. Henceforth we shall assume that the noise of interest in the detection process is the output noise of a correlator or matched filter with variance ¢? = No/2 as expressed in Equation (3.22)- 28) Bandpass Modulation and Demodulation cha? 4 3,3 DIGITAL BANDPASS MODULATION TECHNIQUES Bandpass modulation (ei , ither analog or digital) i mation signal i alog or digital) is the proc ich an i maton gl converted silver: gl malo, sch s referred to as a digital nodulation, su three feat to as a digital symbol. Th Reneaeytan aa LS used to distinguish it from other eae eee whereby the amplitude, is bandpass modulation can be defined as the process of them, is varied in eee Cid of an RF carrier, or a combination . e with tl in ic a general form of the carrier wave, s(t), is a jnienston to be transmitted. The s(t) = A(t) cos 6(1) (3.23) where A(1) is the time-varying amplitude and i i i one A Sere AE ipl and 0() is the time-varying angle. It is (1) = wot + (1) (3.24) so that s(t) = A(t) cos (wot + (01 (3.25) the carrier and (1) is the phase. The terms ¢ frequency. When f is used, frequency in per second is intended. where wo is the radian frequency of f and w will each be used to denot hertz is intended; when w is used, frequency in radians The two frequency parameters are related by w = 2nf. The basic digital modulation/demodulation types are listed in Figure 3.4. When the receiver exploits knowledge of the carrier's phase to detect the signals, the process is called coherent detection; when the receiver does not utilize such phase reference information, the process ‘iscalled noncoherent detection. In digital communications, the terms demodulation and detection are used somewhat in- terchangeably, although demodulation emphasizes removal of the carrier, and detection includes the process ‘of symbol decision. In ideal coherent detection, there is available at the receiver a prototype of each possible arriving signal. These prototype waveforms attempt to duplicate the transmitted signal set in every Te- spect, even RF phase. The receiver is then said to be phase Iocked to the incoming signal. During detection, the receiver multiplies and integrates (correlates) the incoming signal with each of its prototype replicas. Under the heading of coherent modulation/demodulation in Figure 3.4 are listed phase shift keying (PSK), fre- ift keyi ae spit Keying (ASK), continuous phase mod- quency shi keying Se pe basic bandpass modulation formats ulation (CPM), and hybrid combinations. are ee this Chapter. Some specialized formats, such as eT ae PSK (OQPSK), minimum shift keying (MSK) belonging Vr ee class, quadrature amplitude modulation (QAM), are treated in hap fore ae Noncoherent demodulation refers '0 systems employing demodic ing are designed to operate without f the absolute ce ce age signal's phase; therefore, phase i is not required. h te orice paid of noncoherent over coherent systems is re duced complexity. i i i ulation is increased probability of error (Pg). In Figure 3.4 the ‘modulation/demodt! 127 Sec.3.3 pigital Bandpass Modulation Techniques oo Bandpass Modulation/Demodulation Formatting/Source Coding Coherent Noncoherent Character coding Differential PCM (DPCM) Phase shift Differential phase Sampling Block coding keying (PSK) TR Keying Quantization Synthesis/analysis coding Frequency shit Frey oy wi ‘odulatt on reducing cod e quency shift Pulse code modulation (PCM)| | Redundancy reducing coding Amplitude shit Keying eect keying (ASK) Amplitude shift Continuous phase Keying (ASK) modulation Continuous phase (cpm) modulation Hybrids (cP) Hybrids Channel Coding Waveform Structured Synchronization Multiplexing/Multiple Acces Sequences Wary signaling Carrier Frequency division ‘Antipodal Block synchronization {FDM/FDMA) Orthogonal Convolutional Subearrier Time division Biorthogoral synchronization (ToM/TOMA) ‘Transorthogoral symbol Code division — synchronization (COM/COMA) Frame Space division synchronization (SOMA) . Network Polarization division Spreading synchronization (oma) Direct sequencing (0s) 5 Frequency hopping Encryption (FH) Time hopping (TH) Block Hybrids Data stream Figure 3.4 Basic digital communication transformations. types that are listed in the noncoherent column, DPSK, FSK, ASK, CPM, a" hybrids, are similar to those listed in the coherent column. We had implied tha! phase information is not used for noncoherent reception; how do you account for the fact that there is a form of phase shift keying under the noncoherent headins? It turns out that an important form of PSK can be classified as noncoherent (0° differentially coherent) since it does not require a reference in phase with (he received carrier. This “-pseudo-PSK,"" termed differential PSK (DPSK), utilizes phase information of the prior symbol ation ¢ r as a phase reference for detecting the rent symbol. This is described in Sections 3.6.1 and 3.6.2. Figure 3.5 illustrates examples of the most common digital modulation (07 mats: PSK, FSK, ASK, and a hybrid combination of ASK and PSK (ASKIPS or APK). The first column lists the analytic expression, the second is a Pi Pictorial of the waveform versus time, and the third is a vectorial schematic. the orthogonal axes labeled (4y(r)}. In the general M-ary signaling case. the P 128 $ Bandpass Modulation and Demodulation "#? “OlAV) ASAMISY (P) “ASV ©) “ASA (“sa (@) “suo! wen yo1aq ja That Guojanem (9 + 3!) s09 biav) asaisy (P) asv ) asa (9) Sd (©) 129 cessor accepts k source bits at a time and instructs the modulator to pr of an available set of M = 2* waveform types. Binary modulation, wh is just a special case of M-ary modulation. Each example shown in illustrates the set of signal waveforms with a particular value chosen ‘Oduce " ere k = 1 Figure 35 for M, 3.3.1 Phase Shift Keying Phase shift keying (PSK) was developed during the early days of the deep-space program; PSK is now widely used in both military and commercial Communi. cations systems. The general analytic expression for PSK is st) = [Een [wot + fo] = t r iM 6x where the phase term, ¢((t), will have M discrete values, typically given by 2ni At) = vd For the binary PSK (BPSK) example in Figure 3.5a, M is 2. The parameter Eis symbol energy, T is symbol time duration, and 0 < 1 = T. In BPSK modulation. the modulating data signal shifts the phase of the waveform, 51), to one of two states, either zero or 7 (180°). The waveform sketch in Figure 3.5a shows a typical BPSK waveform with its abrupt phase changes at the symbol transitions; if the modulating data stream were to consist of alternating ones and zeros, there would be such an abrupt change at each transition. The signal waveforms can be rer resented as vectors on a polar plot; the vector length corresponds to the signal amplitude, and the vector direction, for the general M-ary case, corresponds (0 the signal phase relative to the other M — 1 signals in the set. For the BPSK example, the vectorial picture illustrates the two 180° opposing vectors. Siz sets that can be depicted with such opposing vectors are called antipodal signal sets. 3.3.2 Frequency Shift Keying The general analytic expression for FSK modulation is n st) = FF cos (wt + 6) OS 1ST oe i=1....M h fm v ean seetueney term, w,, will have M discrete values, and the phase » is an constant. The F: sin Figure 3.50 illust@ the typical abrupt frequenc ‘SK waveform sketch in Figure 3.50 UT pe M has 5 y changes at the symbol transitions. In this types oer nasen equal to 3, corresponding to the same number of W# the mutually pereen at this M = 3 choice for FSK has been selected t0 oP 2. 4,8, 16. Pendicular axes. In practice, M is usually a nonzero Po "+ +). The signal set is characterized by Cartesian coordinates: cna? efor past ol * gust 130 3 | h Bandpass Modulation and Demodulation that each of the mutually Perpendic y icular axes repres inusoid with a di frequency. As described earlier, signal sets that al Be eiaaeie ed aon mutually perpendicular vectors are called orthogonal signals. The ve ee auency spacing between the orthogonal tones is discussed in Section teat 3.3.3 Amplitude Shift Keying For the ASK example in Figure 3.5c, the general analytic expression is PRE: sdt) = a cos (wor +4) PSST (3.28) where the amplitude term, V2E({)/T, will have M discrete values, and the phase term, , is an arbitrary constant. In Figure 3.5c, M has been chosen equal to 2, corresponding to two waveform types. The ASK waveform sketch in the figure can describe a radar transmission example, where the two signal amplitude states would be V2E/T and zero, The vectorial picture utilizes the same phase-amplitude polar coordinates as the PSK example. Here we see a vector corresponding to the maximum-amplitude state, and a point at the origin corresponding to the zero- amplitude state. Binary ASK signaling (also called on-off keying) was one of the earliest forms of digital modulation used in radio telegraphy at the beginning of this century. Simple ASK is no longer widely used in digital communication sys- tems; therefore, it will not be treated in detail. 3.3.4 Amplitude Phase Keying For the combination of ASK and PSK (APK) example in Figure 3.5d, the analytic expression 21 si) = jee al amplitude term illustrates the indexing of both the sign pea us phase APK waveform picture in Figure 3.54 ilutates some OPC Se, and amplitude changes at the sym wavefOrms (B-ary). The figure n chosen equal to 8, correspon the phase-amplitude plane. illustrate; ‘cal eight-vector signal set on are at a different ‘S a hypothetical eig! he other four vectors ‘M symbols Four of the itude; ¢ vectors are at one ampHlUuces © When the set of p amplitude; and each of the vectors is separated by 45°. Wir constellation, tr he two-dimensional signal space are arranged in odulation (QAM): examples © Signaling is referred to as quadrature amplitude m are considered in Chapter 7. ion types 4 i The vectorial picture for each of the modulate nose polar coordinates ro ©Xcept the FSK case) is characterized on 4 PONY sracterized in @ Carte- Present signal ampliuade and phase. The FSK case is OA (cas wut) from Sian : ting a frequen the ordinate space, with each axis representing “AY set of orthogonal tones. wt Bec a3 n Techniques general OstsT 3.29) i=1,...,M and the phase term. The cos [wot + ot)] bol transition ti ding to eight w scribed in Figure Digital Bandpass Modulatio! 3.3.5 Waveform Amplitude Coefficient The waveform amplitude coefficient appearing in Equations (3.26) to B29 the same general form, V2E/T, for all modulation formats. This express)" derived as follows: ‘ons s(t) = A cos wt 05, where A is the peak value of the waveform. Since the peak value of a Sinusoidg waveform equals V2 times the root-mean-square (rms) value, we can write s(t) = V2Arms COS wt = V2Aims COS wt Assuming the signal to be a voltage or a current waveform, AJns represents a. erage power P (normalized to 1 9). Therefore, we can write s() = V2P cos wt G3) Replacing P watts by E joules/T seconds, we get s(t) = V = cos or 3 We shall use either the amplitude notation, A, in Equation (3.30) or the designation V2E/T in Equation (3.32). Since the energy in a signal is the key parameter in determining the error performance of the detection process, its often more convenient to use the amplitude notation in Equation (3.32) becavs it facilitates solving directly for the probability of error, Pr, as a function of sgt! energy. 3.4 DETECTION OF SIGNALS IN GAUSSIAN NOISE 3.4.1 Decision Regions Consider that the two-dimensional signal space in Figure 3.6 is the locus of noise-perturbed prototype binary vectors (s) + n) and (s; + n). The noise vee 1, is a zero-mean random vector; hence the received signal vector, r. is ard Vector with mean s, or 2. The detector's task after receiving r is to decide "Mt of the signals, sy or so, was actually transmitted. The method is uswally to on the signal classification that yields the minimum expected Pe, although OU", strategies are possible [4]. For the case where M equals 2, with s, and #0 equally likely and with the noise being an additive white Gaussian noise (AN the sonal ea see that the minimum-error decision rule is equivalent 10 xf ieeatiog ag uch that the distance d{r, s1) = |r — 5; |] is minimized, "rs Of decisnan ine term or magnitude of vector x. This rule is often stated it regions. In Figure 3.6, let us construct decision regions in the f0 ‘oositt og tiowité 132 ap Bandpass Modulation and Demodulation Yett) Scatter values’. Ofsg tn We son Scatter values ofsy tn wilt) Region 1 Figure 3.6 Two-dimensional si sy and s2. ignal space, with arbitrary equal-amplitude vectors way. Draw a line connecting the tips of the prototype vectors, s; and sp, Next, Construct the perpendicular bisector of the connecting line. Notice that this bi- Sector passes through the origin of the space if s, and s, are equal in amplitude. For this M = 2 example in Figure 3.6, the constructed perpendicular bisector Tepresents the locus of points equidistant between s, and s2; hence the bisector describes the boundary between decision region 1 and decision region 2. The decision rule for the detector, stated in terms of decision regions, is: Whenever the received signal r is located in region 1, choose signal s,; when it is located in Tegion 2, choose signal s2. 3.4.2 Correlation Receiver In Section 2.9 we treated the detection of baseband binary signals in Gaussian Noise. Since the detection of bandpass signals employs the same Seat shall summarize the key findings of that section. We eee ae ‘o reer de realization of a matched filter known as a correlator. In ‘ ae Nea “tion, we also consider the more general case of eon le sceived signal, (0), atthe only performance degradation is due to AWGN. The reoziv® ‘Sthe sum of the transmitted prototype signal plus the ranco een 633) Hth= stm j24,...,M . ists of nvo basic steps. Given Such a received signal, the detection process consists of 1 e received waveform, (1 ‘duced to a single ran dom vari+ In i ‘m, r(t), is re " Sutpy ge’, OF 2 Set of random variables, zAT) Ue the symbol duration. In the Mt Of t ' Correlator(s) at time ¢ = 7, where T'S 133 See, Detection of signals in Gaussian Noise second step, a symbol decision is made, on the basis of comparing 2 to threshold or on the basis of choosing the maximum z(7). Step I can be thoy, ft of as transforming the waveform into a point in the decision Space, Step 2.6 " be thought of as determining in which decision region the point is located, r the detector to be optimized (in the sense of minimizing the error Probability), is necessary to optimize the waveform-to-random-variable transformation, by using matched filters or correlators in step 1, and by also optimizing the decision it in step 2. as insects 2.9.2 and 2.9.3 we found that the matched filter provides the maximum signal-to-noise ratio at the filter output at time ¢ = T. We described g correlator as one realization of a matched filter. We can define a correlation receiver comprised of M correlators, as shown in Figure 3.7a, that transforms received waveform, r(z), to a sequence of M numbers or correlator outputs, 2(7) (i= 1,...,M). Each correlator output is characterized by the following product integration or correlation with the received signal. 7 AT) = [Frise .M 639 The verb ‘‘to correlate’? means ‘‘to match.” The correlators attempt to match the incoming received signal, r(¢), with each of the candidate prototype waveforms, s,(t), known a priori to the receiver. A reasonable decision rule is to choose the waveform, s(t), that matches best or has the largest correlation wih (1). In other words, the decision rule is: Choose the s(t) whose index corresponds to the max z(T) ee Following Equation (3.4), any signal set, {s(1)} (i = 1 , M), can be expressed in terms of some set of basis functions, {W(1)} (j = 1 » N), where N = M. Then the bank of M correlators in Figure 3.7a may be replaced with bank of NV correlators, shown in Figure 3.76, where the set of basis functions {UA} form reference signals. The decision stage of this receiver consists of logic circuitry for choosing the signal, s,(t). The choice of s,(t) is made according to the best match of the coefficients, ay, seen in Equation (3.4), with the set of outputs {<{T)}. When the prototype waveform set, {sAz)}, is an orthogonal sel. Hhelcecive implementation in Figure 3.7a is identical to that in Figure 3.7b (dit ting perhaps by a scale factor). However, when {s2)} is not an orthogonal set the receiver in Fi i i {4A0), can represent 3.76, using N correlators instead of M, with reference signs iver with reference slnals {o(7) ter, we shall assume a correlator recei”®! In the case of bir , } erinary detection, the correlation receiver can be configu! integrator, as shown in Figure 3.83. 7 n . ‘gal he utpur of thence between the binary prototype si! silt) — 52(1). TI lator, 2(7), is fed directly to the decisio* stage, 134 Bandpass Modulation and Demodulation chs Reference signals s(t T Decision Hlth (eban tage 0 jt att) = it) + alt) . : : Srecesio | _- sm(t . . natty po sto its (thdt Reference signals vit) Decision stage elthy Ct) dt Logie circuit selects s(t) whose components aj best match ull Ht) = 5) + alt} Le awe dy(t NSM aylT= | ddl at {, (bo) Figure 3.7 (a) Comrelator receiver with reference signals {540}. © Correlator Teceiver with reference signals {0} iver can also be drawn, as shown . For binar 7 . in Fi, 'y detection, the correlation rec nN men gure 3.8b, as two matched filters or product integrators, each of which is x ae waged to one of the prototype reference signals, s1(t) oF 50. The aoe Can then be configured to follow the rule in Equation (3.35), oF the utp "'S, 24T) (i = 1, 2), can be differenced to form 30 aD = aT) - 2D " test statistic, in the absence of " 8S sho * ae fed to the decisio Stage, ac Figure 3.8b. Then, 2(7), called the ied oa & the case of the single correlator. I "a4 135 Detection of Signals in Gaussian Noise Reference signal sy(t) ~ p(t) Decision stage J] 2tT)=a(T) + ng (TD . rit) i, Pr Si ° fa) Reference signals sy (t) TY Jc al Decision ° a stage + Hi a(T) + ng(T 1 . rt) ——>t salt) 217) By Pr Ho 7] zt) >| f, (o) Figure 3.8 Binary correlator receiver. (a) Using a single correlator. (b) Using two correlators. he waveform, s(t), yields the output, <(T) = a,(T), a signal-only component: 7 input noise, n(t), is a Gaussian random process. Since the correlator is 4 yt device, the output noise is also a Gaussian random process [4]. Thus the 04? of the cosrelator, sampled at ¢ = T, yields X(T) = afT) + nT) i 1,2 . Xe Where no(7) is the noise component. To shorten the notation we sometimes, Press 2(7) as a, + no. The noise component, ng, is a zero-mean Gaussian "A 9, variable, and thus z(7) is a Gaussian random variable with a mean of ei Or az depending on whether a binary one or binary zero was sent. 3.4.2.1 Binary Decision Threshold i itio” For the random variable, 2(7), Fi i wo COM aad or : » 2(7), Figure 3.9 illustrates the ! a Probability density functions (pdfs) p(als,) and p(zles). with mean V8 0 a 42, respectively (these pdfs are also called the likelihood of s, and the 3 136 om? Bandpass Modulation and Demodulation Decision line Region 2 Region 1 Likelihood of sy plzi sy) Liketinood of s, pli s,) cd ay 2m) _ arta w= 0 Figure 3.9 Conditional probability density functions: p(z|s1), p(els:). of sz, respectively): a _1(z-a)* Dial) = Yor and hy- Sec. 94 | 137 Detection of Signals in Gaussian Noise i iding that so(0) was sent i d [equivalent to deciding tl re Gy he Secon Ka be an arbitrary one. For eaeener fal a = le IMU tikely antipodal signals, where s(t) = —sa(t) anda; = ay, optimum tule becomes if an equally lecision mn = y= 0 6.3% xT) Z Yo or decide s(t) if z(T) > 227) 6.3%) decide s2(¢) otherwise d filles In the next section we illustrate the use of correlators a . for the coherent detection of PSK and FSK modulation. In lat * ice of various Consider noncoherent detection, and we treat the error performan modulation types. 3.5 COHERENT DETECTION 3.5.1 Coherent Detection of PSK 5 tion of The detector shown in Figure 3.7 can be used for the coherent detec! aft digital waveforms. $ maxima uch a correlating detector is often referred to tpl. likelihood detector. Consider the following binary PSK (BPSK) ex: i) oe n() = = cos ton +6) Osr=T bE s(t) = COS (wot + + mr) 34! & = ~ yZ cos (wor + 4) OstsT ™O = zero-mean white Gaussian random process ste ig js UN nt where the phase term, . the analysis is UV Fist by setting boo cas an arbitrary constant, so that the i bol, AN ayes = 0. rameter, E, is the signal energy per symbol nis Oe q7mbol duration. For this antipodal case, only a single basis fan ie D, we eonetmal signal sparene assumed in Equations (3.4) and G-5 » We can express a basis function, W(t), as follows: wi) = iE Osr1sT " POS wot — for "7 138 ation Bandpass Modulation and Demodu! Thus we may express th 5 fe transmit i ficients a(t): transmitted signals si(1) in terms of w(t) and the coef- is ¢ coef st) = aay!) (3.42a) $0 = auntn() = VEC) aan) $21) = aan) = — VEW,(t) tte) ee at FiSEra TON wal ts cae gas Wry ee Ta oun 7D, Wil is - follows: reference signals qx(t) and — (2), are found as 1 Blain} = E { ff VBuD + aon ai} 6.430) 5 Bala) = 8 { [7 —VEHO ~ mento a} 43) r2 E{zilsi} = E {f B VE cos? wot + n(t) cos wot ar} = VE (3.44a) 7 12 Efzds:} = e{ [" — 2 VE cox aot ~ nto) {bcos oot ar} = -VE (3.44) where E{-} denotes the ensemble average, referred to as the expected value. Equa- tion (3.44) follows because E{n()} = 0. The decision'stage must decide which signal was transmitted by determining its location within the signal space. For this example, the choice of Wilt) = AIT cos wot normalizes E{z(7)} to be ‘VE. The prototype signals {s,(#)} are the same Bs the reference signals {W)(0)} except for the normalizing scale factor. The decision stage chooses the signal with the largest value o! in this example is judged to f z(1). Thus, the received signal t be s,(t). The error performance for such coherently detected BPSK systems 's treated in Section 3.7.1. 3.5.2 Sampled Matched Filter In Section 2.9.2 we discussed the basic cimrctts el namely, that its i ‘esponse is a delayed version > je mirror image | nthe nae put i Porm, Therefore, ifthe signal waveform is s(t), its mirror image is s(- 0) and the mirror image delayed by 7 seconds S(T ~'1), The impulse response, Hs ofa filter matched to 18 by he matched filter— ic of t image (rotated onan (3.45) elsewhere ented using digital be implem jer can be imp signal, (0s Pls Figure 3. 10a illustrates how @ matched fill . A s totype szomorised of (RE PTOCUN ist sampling ardware. The input si : . put signal, r(1), ' ca where the noise, n(z), The bandwidth of the sigmalS °° Me 139 s(T - 4) Se °.3.5 Coherent Detection nv) N alk) = E45 rik Ale, (a) = cos wt sy{t) = cos ot sgt) = £03 «9 4 H 1 k modulo-4 k modulo “1 012301 123 01 oo Shift egste ° ° 1 | contents atk 3 Filter weights matched to 54 (k) eft) = 5y(0) + alt) t= kT, Filter weights matched to 52(k} (b) 2 Figure 3.10 (a 5. : ; ; Paw 310) Sed mish ker. 6) Samed mache ise dtesn nap? 140 Bandpass Modulation and Demodulation rate J, = 2W = 1/T,; hence the sampling 7 a E ; pling interval times of 1 = ATs, the analog signal is sampled and the ecto At te clock the register of Figure 3.10a from left to right Th amples ate shifted into ir ¢ shift register with it . 00 ew-1 approximate a matched filter. Once the ee wa its coefficients pled, the continuous time notation ¢ is changed to kT, or a ae " ee . nea - 'o reflect the sampled notation HA) = sk) +n) k= 0,1,.,, where k represents a sample index. The output, 2(& fle, ata time corresponding tothe kth sample ig” SmPled matched N-1 2h) = 2 rk nen =0,1,...,moduloN 3.46) where x modulo-y is defined as the remainder of dividing x by y. For the binary demodulation application, z(k)(i = 1, 2) outputs are compared to a threshold at each value of k = N — | corresponding to the end of a symbol. The c, values are the filter weights constituting the filter impulse response that is matched to the signal, where 1 is the index of the weights and the register stages (from left to right) and k is the index of the samples as they are produced by the sampler. One can see the similarity between the convolution integral of Equation (2.56) and the summation of Equation (3.46), especially with regard to the mirror-image rotation of one of the functions prior to multiplication. Since we assume the noise to have zero mean, the expected value of a received sample for the binary case is expressed as E{r(k)} = sdk) f= 1,2 G.47) e expected matched filter outputs would be Ne E(zdk)} = LD silk — Men (3.48) n=O are matched to the correspondin} If s,(1) had been transmitted, th 1 s(k) for each Where the filter weights, Cn branch, Example 3.2. Sampled Matched Filter Consider the BPSK waveform set s(t) = CoS OF and it aD Slice A 3.100, can own in Figure 3.108 llustrate how a sampled matched filter oF comelator, Task waveform sein ta used 19 detect a received signal, say si(?, from SPsence of noise, Solution 5 rmed into the "fol Ps em amie tt 8s eat .T ive ch is Towing awe icles ae Figure +3,8b. The top bran as 8 implement Coherent Detection registers and coefficients matched to the {s(A)} sample points. The is similarly matched to the {s2(k)} sample points. The four equally go 40" bray points (k = 0, 1, 2,3) foreach ofthe (54h) areas follows (see Figured Sins © 3-105), * ak=O=1 k= 0=0 SK=D=-1, GQ so stk =) 04 The c, coefficients represent the delayed mirror-image rotation of the s the filter is matched. Therefore, c, = s(N ~ 1 — 1), where n = 0, and we can write co = s(3), ¢1 = (2), ¢2 = (Is ¢s = $40). is here thy sh reader can gain some insight as to Why the convolution operation (with ig eo" image rotation) results in the appropriate lining up of the received si with the weights (reference signal). Consider the top branch in Figure 3.10b. At the k = 0 clock time, the sample, s\(k = 0) = I, enters the leftmost stage of each register. At the nex et time, the second sample, s;(k = 1) = 0, enters the leftmost stage of each relies at this same time the first sample, s;(k = 0) = 1, has been shifted to the next ry stage in each register, and so on. At the k = 3 clock time the sample, s,(k = 3) = 0, enters the leftmost stage: by this time the first sample, si(k = 0) = 1, has bem shifted into the rightmost stage. The four signal samples are now located in ie registers in mirror-image arrangement compared to the way the prototype waveform, 5,(1), is drawn in Figure 3.10b. The task of the demodulator is to find the best mats to the incoming signal; the demodulator matches the reference coefficients of eat branch with the incoming signal samples, in the order in which the samples anv. Hence the convolution operation is an appropriate expression for describing the alignment of the incoming waveform samples with the reference coefficients, tomat- imize the correlation in the proper branch. s(k= 02-1 s(kK2N=0, kK =2=1, Signal tp hig th its mi, anal sangle, 3.5.3 Coherent Detection of Multiple Phase Shift Keying Figure 3.11 illustrates the signal space for a multiple phase shift keying (MPSK) signal set; the figure describes a four-level (4-ary) PSK or quadriphase shift keying (QPSK) example (M = 4). Binary source digits are collected two ata time, and for each symbol interval the two sequential digits instruct the modulator as (0 which of the four waveforms to produce. For typical coherent M-ary PSK (MPS) systems, st) can be expressed as _ pe 2ni) Os1sT 6.8) 0 = Vos (aot - 3) i=1....M . isthe where E is the energy content of s(t) over each symbol duration 7, and % carrier frequency. If an orthonormal signal space is assumed in Equation and (3.5), we can choose a convenient set of axes, as follows: ve gst! F608 Wot 2 | 6 T SIN Wolf cna? wD " w() 142 Bandpass Modulation and Demodulation voit) vitt) Region 4: Figure 3.11 Signal space and decision regions for a QPSK system. where the amplitude V2/T has been chosen to normalize the expected output of the detector, as was done in Section 3.5.1. Now sz) can be written in terms of these orthonormal coordinates, giving anta(t) + aaa) PST G.Sla) a LM i} VE cos (%) dn(t) + VE sin () Wo) (3.51b) Notice that Equation (3.51) describes a set of M multiple phase waveforms (intrinsically nonorthogonal) in terms of only two orthogonal carrier-wave com- ponents. The M = 4 (QPSK) case is unique among MPSK signal sets in the sense that the QPSK waveform set is represented by a combination of antipodal and orthogonal members. The decision boundaries partition the signal space into M = 4 regions; the construction is similar to the procedure outlined in Section 3.4.1 and Figure 3.6 for M = 2. The decision tule for the detector (see Figure 3.11) is to decide that s;(t) was transmitted if the received signal vector falls in region 1, that s(t) was transmitted if the received signal vector falls in region 2, and so on. In other words, the decision rule is to choose the ith waveform if z(7) is the largest of the correlator outputs (seen in Figure 3.7). The form of the correlator shown in Figure 3.7a implies that there are always M product correlators used for the demodulation of MPSK signals. The figure infers that for each of the M branches, a reference signal with the appropriate phase shift is configured. In practice, the implementation of an MPSK demod- ulator follows Figure 3.7b, requiring only N = 2 product integrators regardless of the size of the signal set M. The savings in implementation is possible because any arbitrary integrable waveform set can be expressed as a linear combination Sec.3.5 Coherent Detection 143 n 3.2.2. Figure 3.12 illustrates such in Sectio' ini own in be expressed by combining Equa. vaveforms. as sh of orthogonal wavefo an a demodulator. The received signal, r(t tions (3.50) and (3-51) as follows: sts T eee M (3.52) i in wot) + (0) 1) = Fos dicoswol + sin g:sin wo!) I o-mean white Gaussian noise process. Notice wo reference waveforms or basis functions, where @; = 2ai/M, and n(Z) is a zer correlator and a(t) = V2IT sin wot for the in Figure 3.12 that there are only t Ut) = V2T cos wot for the upper lower correlator. The upper correlator computes 7 x= f Hold) dt (3.53) and the lower correlator computes 7 y= [roan ae 3.54) 0 vt [Bees cogt 1 x= rithyy(t) dt > {* ° Jo y_, _ ¥ LB, [eompate |_| chooe Ly sy eo vattl= \/ 2 sineot area x 19-31 smallest [> Vi T = d I Y= (walt) dt ° Figure 3.12 Demodulator for MPSK signals. Figure 3.13 illustrates that the computation of the received phase angle $ can be accomplished by computing the arctan of Y/X, where X can be thought of as the phase component of the received signal, Y is the quadrature component, and is a noisy estimate of the transmitted ¢,. In other words, the upper correlator of Figure 3.12 produces an output X, the magnitude of the in-phase projection of the vector r, and the lower correlator produces an output Y, the magnitude of the quadrature projection of the vector r. The X and Y outputs of the correlators feed into the block marked arctan (¥/X). The resulting value of the angle } is compared with each of the stored prototype phase angles, ¢;. The demodulator selects the ; that is closest to the angle 4. In other words, the demodulator computes lo 52 for each of the 4; prototypes and chooses the 4; yielding the smallest output. 144 Bandpass Modulation and Demodulation Chap. 3 sin wot Y= Irising, cos wot aay Figure 3.13 In-phase and quadrature components of the received signal vector r. Noisy estimate $= arctan (¥/X) cs herent Detection of FSK oi _*“FSK modulation is characterized by the information being contained in the fre~ quency of the carrier. A typical set of FSK signal waveforms was described in Equation (3.27) as a < Sty = VFoos tor say PSUS Ty _4» where E is the energy content of (1) over each symbol duration 7, and (w,~1 ~ ©) “= is typically assumed to be an integral multiple of 7/7. The phase term, &. is an arbitrary constant and can be set equal to zero. Assuming that the basis func- tions WiC), Ya(t), - - - . Ut) form an orthonormal set, the most useful form for {W,(2}} is shown below. 2 . Wi) = Yzooset Fah. .N BS = as before, the amplitude V2/T normalizes the expected output of the de- ‘or. From Equation (3.5) we can write = 2E 2 = = cos (wit) nA COS wyt dt 3.50) w= [V2 2 cos VE fori=j 9u.= {v otherwise _ js, the ith prototype signal vector is located on the ith coordinate ent VE from the origin of the signal space. In this scheme, ral M-ary case, the distance between any two prototype signal vectors rad Sy Oeessinn boundary Region 3 a(t) Figure 3.14 Partitioning the signal space for 2 3-ary FSK signal ‘igure 3.14 illustrates the prototype signal vectors and the decision regions for a 3-ary (M = 3) coherently detected FSK system. As in the PSK case, the signal space is partitioned into M distinct regions, cach containing one prototype signal vector; here, because the decision region is three-dimensional, the decision boundaries are planes instead of lines. The optimum decision rule is to decide that the transmitted signal belongs to the class whose index corresponds to the region where the received signal is found. In Figure 3.14, a received signal vector ___ ris shown in region 2. Using the decision rule stated above, the detector classifies “> ras signal s». Since the noise is a Gaussian random vector, there is a probability ~ greater than zero that r could have been produced by some signal other than 5: % ‘of example, if the transmitter had sent s2, then r would be the sum of signal jects noise, s. + n,, and the decision to choose s2 is correct: however, if the transmitter had actually sent ss, then r would be the sum of signal plus noise. s: + ny, and the decision to select s2 is an error. The error performance of coherently | segy detected — is treated in Section 3.7.3, 3.6 NONCOHERENT DETECTION % 3.6.1 Detection of Differential PSK i The name differential PSK (DPSK) sometimes necds clarification because (WO separate aspects of the modulation/demodulation format are being referred to: the encoding procedure and the detection procedure. The term differential encoding refers to the procedure of dati ey jally; that is, the presence i modulation Chap.3 of a binary one or zero is manife: compared to the preceding sy differentially encoded PSK, scheme often classified as n in phase with the received c: herently detected. This will sted by the symbot's similarity or difference when mbol. The term differenti lly coherent detection of the usual meaning of DPSK, refers to a detection oncoherent because it does not require a reference ‘arricr. Sometimes, differentially encoded PSK is co- viieenconsien be discussed in Section 3.7.2. othe peer systems, no attempt is made to determine the actual value pl € Incoming signal. Therefore, if the transmitted waveform is RE SQ) = VF cos (wor + a(n) OS 1ST i=l... the received signal can be characterized by 08 [wot +O) tal + nr) PSSST (3.59) 2E m= \y T G=1,...,M where a is an arbitrary constant and is typically assumed to be a random variable uniformly distributed between zero and 2m, and n(1) is an AWGN process. For coherent detection, matched filters (or their equivalents) are used; for noncoherent detection, this is not possible because the matched filter output is function of the unknown angle a. However, if we assume thal « varies slowly relative to two period times (27), the phase difference between two successive waveforms, 8471) and @;(72) is independent of a, that is, (872) + a] ~ [6(71) + a] = O72) — O{T1) = 472) (3.60) The basis for differentially coherent detection of differentially encoded PSK (DPSK) is as follows. The carrier phase of the previous signaling interval can be used as a phase reference for demodulation. Ils use requires differential encoding of the message sequence at the transmitter since the information is carried by the difference in phase between two successive waveforms. To send the ith message (i = 1,2,..-,M), the present signal waveform must have its phase advanced by 4, = 2mi/M radians over the previous waveform. The detector, in general, calculates the coordinates of the incoming signal by correlating it with locally generated waveforms such as VIT cos wot and V2IT sin wot, The detector then measures the angle between the currently received signal vector and the previ- ously received signal vector, as illustrated in Figure 3.15. In general, DPSK signaling performs less efficiently than PSK, because the errors in DPSK tend to propagate (to adjacent symbol times) due to the correlation between signaling waveforms. One way of viewing the difference between PSK and DPSK is that the former compares the received signal with a clean reference: in the latter, however, two noisy signals are compared with each other. We might say that there is twice as much noise associated with DPSK signaling compared to PSK signaling. Consequently, as @ first guess, we might estimate that DES manifests a degradation of approximately 3 4B when compared with PSK; this degradation decreases rapidly with increasing signal-to-noise ratio. The trade-off 147 Sec. 3.6 Noncoherent Detection Walt) elayty) Stored Angle signal vector measured by detector 4) (az, by) Currently received signal vector > nil int. Vill) Figure 3.18 Signal space for DPSK for this performance loss is reduced system complexity. The error performance for the detection of DPSK is treated in Section 3.7.5. 3.6.2 Binary Differentlal PSK Example The essence of differentially coherent detection in DPSK is that the identity of the data is inferred from the changes in phase from symbol to symbol. Therefore, since the data are detected by differentially examining the waveform, the trans. mitted waveform would first be encoded in a differential fashion. Figure 3.16a illustrates a differential encoding of a binary message data stream, m(k), where kis the sample time index. The differential encoding starts (third row in the figure) with the first bit of the code bit sequence, c(k = 0), chosen arbitrarily (here taken to bea one). Then the sequence of encoded bits, c(k), can, in general, be encoded in one of two ways: e(k) = c(k — 1) @ m(k) G.61) or c(k) = c(k = 1) @ m(k) (3.62) where the symbol © represents modulo-2 addition (defined in Section 2.12.3) and the overbar denotes complement. In Figure 3.16 the differentially encoded mes- sage was obtained by using Equation (3.62). In other words, the present code bit, c(k), is a one if the message bit, m(k), and the prior coded bit, o(k — 1), are the same, otherwise, c(k) is a zero. The fourth row translates the coded bit sequence, c(K), into the phase shift sequence, @(k), Where a one is characterized by a 180° phase shift, and a zero is characterized by a 0° phase shift. Figure 3.16b illustrates the binary DPSK detection scheme in block diagram form. Notice that the basic product integrator of Figure 3.7 is the essence of this detection process; as with coherent PSK, we are still attempting to correlate 2 received signal with a reference. The interesting difference here is that the tel erence signal is simply a delayed version of the received signal. In other words. during each symbol time, we are matching a received symbol with the prior symbol and looking for a correlation or an anticorrelation (180° out of phase). Consider the received signal with phase shift sequence, 6(k), entering the 148 Bandpass Modulation and Demodulation chap. Sample index, k oily x o © 0 Information message, m(k} 1 ° 2 2 ° Ditferentialy coded message 4 Wetec Tepe f tf yt fey sys & 2 2 2 2 (a) a0 jt Decision : I pe etage Sh Coherent detector Delay 7 JReterence Detected message, i(k) Sccmec emer eecrmes a ecatee mee meres: (b) <- VI 00s wot 4 a $ T ds i Decision Bon 5 stage [ Delay a. te) bes Figure 3.16 Differential PSK (DPSK). (a) Differential encoding. (b) Differentially coherent detection, (c) Optimum differentially coherent detection. letector of Figure 3.16b, in the absence of noise. The phase, 8(k = 1). is matched with @(k = 0); they have the same value, 7; hence the first bit of the detected output is i(k = 1) = 1. Then 0(k = 2) is matched with 6(k = 1); again they have the samme: value; anit i(k = 2). 1. Then 0G = 9) is xomtchedt with 4’ = 2): they are different, so that sf(k = 3) = 0, and so on. ; It must be pointed out that the detector in Figure 3.16b is suboptimum [5} in the sense of error performance. The optimum differential detector for DPSK requires a reference carrier in frequency but not necessarily in phase with the received carrier, Hence the optimum differential detector is shown in Figure 3.16c L161. Its performance is treated in Section 3. ae. Noncoherent Detection 149 Sec, 3.6.3 Noncoherent Detection of FSK A detector for the noncohcrent detection of FSK waveforms described by Equa. tion (3.27) can be implemented with correlators similar to those shown in Figure However, the hardware must be configured as an energy detector, without measurements. For this reason, the noncoherent detector typi- ice as many channel branches as the coherent detector. Figure 1-phase (1) and quadrature (Q) channels used to detect a binary | set noncoherently. Notice that the upper two branches are ‘detect the signal with frequency ; the reference signals are 0s w;f for the I branch and V2/T sin of for the Q branch. Similarly, the ver two branches are configured to detect the signal with frequency a; the teference signals are V2/T cos w2t for the I branch and V2/T sin w2t for the Q branch. Imagine that the received signal r(1), by chance alone, is exactly of the > form cos wif + n(t); that is, the phase is exactly zero, and thus the signal com. _ ponent of the received signal exactly matches the top-branch reference signal with ‘regard to frequency and phase. In that event, the product integrator of the top branch should yield the maximum output. The second branch should yield a near- zero output (integrated zero-mean noise) since its reference signal V2/T sin wr Land Q energy Test statistic Correlation Squaring summation and decision aT] 217) 7 cE 2 fr et 0 Bee 2 aT) a efi? 7 Decision 20s y Hy suo > 21 20 23(T) Z Ha wey? _ is orthogonal to the signal Component of r(1). The third and fourth branches should also yield near-zero outputy since their «, reference signals are also orthogonal the signal component of 111), » imagine a different scenario; suppose that by chance alone, the re- analy 11). is Of the form sin of + a(t), In that event, the second branch AT should yield the maximum output, while the others should yield uts. In actual practice, the most likely scenario is that 111) is of the ©) 4 ntt); that is, the incoming signal will partially correlate and partially correlate with the sin of reference. Now s Why @ noncoherent quadrature receiver uses twice ay many a oherent one; the recciver knows nothing about the incoming sig- . The receiver essentially resolves the signal into an | component and a Q (90° out of phase) component. In Figure 3.17 the blocks following the product integrators perform a squaring operation to prevent the appearance of any negative values. Then for cach of the signal types in the set (two in this binary example) the energy from the I and Q channels is added. The final stage forms the test - statistic, 2(7), and chooses the signal with frequency w, or the signal with fre- quency wz depending on which pair of energy detectors yielded the maximum output. . Another possible implementation for noncoherent FSK detection uses band- pass filters, centered at f, = w/2n, with bandwidth, Wy = 1/T, followed by envelope detectors, as shown in Figure 3.18. An envelope detector consists of a rectifier and a low-pass filter. The detectors are matched to the signal envelopes and not to the signals themselves. The phase of the carrier is of no importance in defining the envelope; hence no phase information is used. In the case of binary FSK, the decision as to whether a one or a zero was transmitted is made on the basis of which of two envelope detectors has the largest amplitude at the moment of measurement. Similarly, for a multiple frequency shift keying (MFSK) system, the decision as to which of M signals was transmitted is made on the basis of which of the M envelope detectors has the maximum output. Even though the envelope detector block diagram of Figure 3.18 looks func- Bandpass filters centered at f, with bandwidth Wy = VT f finer |_, [Envelope [17 ui detector i 7 Filter envetape |?2""?|Decision| 2 nt) = s(t + 00) ot *| anmee el ae 7 Filter Envelope |°™ 7) fw. [| eetector Noncoherent det using envelope detectors. nt Detection st 151 oy tionally simpler than the quadrature receiver of Figure 3.17, the use of filters usually results in the envelope detector design having greater weight and cow quadrature receiver. Quadrature receivers can be implemented digitally. “thus, with the advent of large-scale integrated (LSI) circuits, they are often the Preferred choice for noncoherent detectors. The detector in Figure 3.18 can algo be implemented digitally by performing discrete Fourier transformations insteag of using analog filters, but such a design is usually more complex than a digital implementation of the quadrature receiver. 3.6.4 Minimum Required Tone Spacing for Noncoherent Orthogonal FSK Signaling Frequency shift keying is usually implemented as orthogonal signaling where each tone (sinusoid) in the signal set cannot interfere with any of the other tones. In order for the signal set to be orthogonal, any pair of adjacent tones must have a frequency separation of a multiple of I/T hertz. A tone with frequency f;, that is switched on for a symbol duration of T seconds and then switched off, such as the FSK tone described in Equation (3.27), can be analytically described by sit) = (cos 2nfit) rect (t/T) where rect (/T) = {i He ll ret ae The Fourier transform of s,(), from Table A.1, is F {s(t)} = T sine (f -faT where the sinc function is as defined in Equation (1.39). The spectra of two such adjacent tones, tone 1 with frequency f, and tone 2 with frequency f,, are plotted in Figure 3.19. In order that the two tones not interfere with each other during detection, the peak of the spectrum of tone I must coincide with one of the zero crossings of the spectrum of tone 2 and similarly, the peak of the tone 2 spectrum must coincide with one of the zero crossings of the tone | spectrum. The frequency difference between the center of the spectral main lobe and the first zero crossing represents the minimum required spacing. This corresponds to a minimum tone separation of 1/T hertz. Example 3.3 Minimum Tone Spacing for Noncoherent Orthogonal FSK Consider two waveforms cos (2nfit + 6) and cos 2mft to be used for noncoherent! FSK signaling, where f, > fz. The symbol rate is equal to U/T symbols/s, where T is the symbol duration and is constant arbitrary angle from 0 to 2m. -~ (a) Prove that the minimum tone spacing for noncoherently detected orthogonal FSK. signaling is 1/7. % (b) What is the minimum tone spacing for coherently detected orthogonal FSK signaling? 182 ©. Bandpass Modulation and Demodulation Chap. a mh te E WM Telnet = tye ' nk / , \ 4 Tone 1 \ Tone 2 ' 1 I ‘ ' ' ' i 1 t ' i i \ \ \ \ \ ‘ \ Ny Figure 3.19 Minimum tone spacing for noncoherently detected orthogon signaling. Solution constraint of Equation (3,2): ! r iL cos Qnfit +) cos Infatde = 0 trigonometric identiti (3.63) as le or cos [- cos 2nfyr cos 2nfat de ee ag T .& 4S “e ~ sin of sin 2nfyt cos 2nfat dt = 0 cos 6 f" [eos 2ntfi + fF + cos 2alfy ~ fn dr = sind [vin Intfy + fae + sin 2m(fr — fede] de sin 2n(fr + foe, sin 2m — fu i. co | Qmthi + Ja) Ini ~ 4) do cos 2m +S , costa fi 0 He + sind ercmcarae Inf, — f2) We sin 2n(fy + fT "3 sin 2m(f1 = ft) joy fae i) cos 2nifi - fT = 1] _ 9 4 90s 2m = fT = 1 2th: - fz) (a) For the two waveforms to be orthogonal, they must fulfill the orthogonality 3.63) hown in Equations (D.6) and (D.1) to Goh) G.65) (3.66) Gon 153 154 Wecan assume that fy + f2> !and can thus make the following approximation, sin In(fy + ST ,, cos 2m + LIT antfy + Ja) Inthr + $2) ining Equations (3.67) and (3.68), we can write sino [cos 20S ~ LT - =O Bay 06) Thea, com cos b sin 2n(fr — S2)T + the terms in Equation (3.69) can sum to zero only when bitrary 4, 7 Note that for simultaneously cos 2(f, — f2)T = 1. sin 2n(f, — S27 = 0 and Since sinx =0 forx = nw and cosx = 1 forx = 2ka where n and k are integers, then both sin x = 0 and cos x = | occur simul. taneously when n = 2k. From Equation (3.69), for arbitrary $, we caa therefore write: 2n(fy — f2)T = 2kew k (3.70) fi-fh=z Thus the minimum tone spacing for noncoherent FSK signaling occurs for k = 1: 1 hi-ho G.7)) (b) To find the minimum tone spacing for coherent FSK, where the angle ¢ is zero, we simply rewrite Equation (3.69) with ¢ = 0, which gives sin 2n(f, — f2)T = 0 3.72) fi-hexe T 3.73) Thus the minimum tone spacing for coherent FSK signaling occurs for n = 1 as follows: fi-fi= x G.74) Therefore, for the same symbol rate, coherently detected FSK can occupy less bandwidth than noncoherently detected FSK and still retain orthogonal signaling. We can say that coherent FSK is more bandwidth efficient. The subject of band- width efficiency is addressed in greater detait in Chapter 7. Bandpass Modulation and Demodulation Chap. 3 ( jar ERROR PERFORMANCE FOR BINARY SYSTEMS 3.7.1 Probability of Bit Error for Coherently Detected BPSK An important measure of performance used for comparing digital modulation schemes is the probability of error, Py. For the correlator or matched filter de- tector, the calculations for obtaining Py, can be viewed geometrically (see Figure _ 3.6). They involve finding the probability that given a particular transmitted signal vector, Say $1. the noise vector, n, will give rise toa received signal falling outside region 1. The probability of the detector making an incorrect decision is termed the probability of symbol error (Pz). It is often convenient to specify system performance by the probability of bit error (Pa), even when decisions are made on the basis of symbols for which M > 2. The relationship between Px and Pp is treated in Section 3.9.3 for orthogonal signaling and in Section 3.9.4 for multiple phase signaling. For convenience, this section is restricted to the coherent detection of BPSK modulation. For this case the symbol error probability is the bit error probability. Assume that the signals are equally likely, Also assume that when signal, s(¢) (i = 1, 2), is transmitted, the received signal, r(1), is equal to s(t) + a(t), where n(t) is an AWGN process. The antipodal signals, s)(1) and s2(z), can be charac- terized in a one-dimensional signal space as described in Section 3.5.1, where sD = VEW(0, O yo = 0 s(t) otherwise (3.76) Two types of errors can be made, as shown in Figure 3.9: The first type of error takes place if signal s,(1) is transmitted but the noise is such that the detector @renrss a negative value for 2(7) and chooses hypothesis Hz [the hypothesis that signal s2(1) was sent]. The second type of error takes place if signal s3(+) is, transmitted but the detector measures a positive value for z(7) and chooses hy- » pothesis H; [the hypothesis that signal s\(1) was sent]. To calculate the probability of a bit error, Pa, for this binary minimum error detector, we use the relationships developed in Section 2.9, starting with Equation (2.36b): Pp = P(Holsy)P(s1) + PCHi|s2)P(s2) (3.77) *. 37 a a for Binary Systems 155 For the case when the a priori probabilities are equal, that is, P(s}) = ° Pls). we can write | Pp = 4P(Hi|s1) + 4PCHils2) by . a 7 Because of the symmetry of the probability density functions in Figure 39 can also write My Py = PCHa|s1) = PCHils2) bn Thus the probability of a bit error, Pa, is numerically equal to the area Under 4 “tail” of either pdf, p(zlsi) or p(z|s2), that falls on the “incorrect” side of threshold. We can therefore compute Ps by integrating p(z|s1) between the limi —o and yo, or as shown below, by integrating p(z|sz) between the limits ang Poe Lo nana Pea) dz Ga where the likelihoods, p(z|s;) (i = 1, 2), are Gaussian functions with mean value a;, and the optimum threshold, yo, as shown in Section B.3.1, is equal to (a, + a3)/2. The area-related probability of bit error, Px, is seen to be the shaded area in Figure 3.9. It is shown in Section B.3.2 that Equation (3.80) reduces tp 1 we) fay — ae Pa Joeoam Vg? ( 2) du = o( 200 G8 where go is the standard deviation of the noise out of the correlator. The function, Q(x), called the complementary error function or co-error function, is defined as -tf _# 0 = sf exe ( +) du (3.82) and is described in greater detail in Sections 2.9 and B.3.2. For equal-energy antipodal signaling, such as the BPSK format in Equation G.75), the receiver output signal components are a, = VE» when s\(2) is sett and a, = —VE, when s2(1) is sent, where E, is the signal energy per binary symbol. For AWGN we can replace the noise variance, «3, out of the correlator with No/2 (see Appendix C), so that we can rewrite Equation (3.81) as follows: -{~ 1 a 3.) Pe Wea ag e%” (- a ‘ =@ (vi) (a No is result could also have been obtained by noting that the energy itt Es etween the antipodal signal vectors, s; and sz, with amplitudes of * seen in Figure 3.20a, can be Computed as the square of the distance 4 156 18 Bandpass Modulation and Demodulation eet wn) Figure 3.20 Binary signal vectors. (a) Antipodal. (b) Orthogonal. the heads of the antipodal vectors, or in terms of the waveforms r Eg = f [sD — soe) P de (3.85) r r r Fs if ao dt + f si(1) dt — 2f s(t)sa(t) de (3.86) Assuming equal energy signals, 7 : Fe= [i sitoae= [st ar 687 q Eq = 2Ey — 2Eop = 2Es(1 — p) (3.88) where a paz [soso at (3.89) the time cross-correlation coefficient and E, is the average energy of the binary als, s,(t) and s2(t). The correlation coefficient, p, is a measure of similarity ween the two signals, s;(1) and s2(t), such that -lsps! (3.90) rms of signal vectors, the cross-correlation coefficient can be written p = cos 0 GB.91) re 0 is the angle between the two signal vectors s, and s2 (see Figure 3.6). In tion (2.62), we developed an.gypensog for the probability of bit error in , ry Systems) 187 terms of the energy difference between the two binary signals, as follows. Pu =@Q (\3 No By, Substituting Equation (3.88) into Equation (3.92), we get EWl — p) fae TN) 8.9 For p = 1 (or @ = 0), the signals are perfectly correlated (identical), Forp = (or 8 = 7m), the signals are anticorrelated (antipodal). Since the binary PSK are antipodal. we can set p = ~ 1, and Equation (3.93) is then identical to E, (3.84). Note that the bit error probability, Ps, for the coherent detection of bandpas, antipodal signals, as seen in Equation (3.84), is the same as the Pg for the ‘Matched filter detection of baseband antipodal (bipolar) signals in Equation (2.67). als QUiation 3.7.1.1 The Basic SNR Parameter for Digital Communication Systems average signal power to average noise power, S/N (or SNR). By introducing the signal bandwidth W, we can write the following identities, showing the relationship between E,/No and SNR for binary signals. ae Ey _ ST s SW Ss ® Ny The parameter E4/No in Equation (3.84) can be expressed as the ratio of No No RNo RNW N R (3.94) where S = average modulating signal power T = bit time duration R = WT = bit rate N= NW Analysis similar to that used for developing Ps in Equations (3.84) and (3.93) is used in finding the Pg expressions for other types of modulation. Figure 32! illustrates the “‘waterfall-like” shape of most probability of error curves in the field of digital communications. The curve describes a system's error probability performance in terms of available E,/No. For Ex/No = xo, Pe S Po. The diet sionless ratio E,/No is a standard quality measure for digital communications system performance. Note that optimum digital signal detection implies 4 © relator (or matched filter) implementation, in which case the signal bandwidth equal to the noise bandwidth. Often we are faced with a system model for which this is not the case; in practice, we include a factor in the required Ey/No ti accounts for such suboptimal detec! 2. Required E»/No can be CO sidered a metric that characterizes the perfec ne system versus another 3 158 Bandpass Modulat Cate, Far Ey/Ng > xg. Pr < Po E QIN, Figure 3.21 General shape of the Pe x0 cee versus Es/No curve. the smaller the required E,/No, the more efficient is the system modulation and detection process for a given probability of error. Figure 3.22 is a plot comparing the bit error probability, Pz, of several binary modulation/demodulation types. The Pz for coherent detection of PSK, as shown in Equation (3.84), is plotted as the leftmost Px curve. Example 3.4 Bit Error Probability for BPSK Signaling Find the bit error probability for a BPSK system with a bit rate of 1 Mbit/s. The received waveforms, 5;(1) = A COS wof and 52(1) = —A COS wof, are coherently de- tected with a matched filter. The value of A is 10 mV. Assume that the single-sided noise power spectral density is No = 10-'' W/Hz and that signal power and energy per bit are normalized relative to a 1-9 load. Solution 10°V. oF = 107s Thus 2E% By=ZT=5x 10'S and Po = o( vi] =9G19 & Using Table B.1 or Equation (2.43), we obtain 3.16 Py = 8 x 107* Sec. 3.7 Error Performance for Binary Systems soncoherent detection of FSY Coherent _- detection PSK Coherent detection of differentially encoded PSK Differentially conerent etection of differentially encoded PSK (PSK) Shannon fimit 7 (-16 08) -8-6 -4-2 0 2 4 6 & 10 12 14 16 Figure 3.22 Bit error probabil fr Ey No (48) several types of binary systems 3.7.2 Probability of Bit Error for Coherently Detected Differentially Encoded PSK . Channel waveforms sometimes experience inversion; for example, when usirg2 coherent reference generated by a phase-locked loop (see Chapter 8). one Mu! "have phase ambiguity. If the carrier phase were reversed in a DPSK modulati» application, what would be the effect on the message? The only effect woul ® an error in the bit during which inversion occurred o the bit just after inversi™ since the message information is encoded in the similarity or difference bel adjacent symbols. The similarity or difference quality remains unchanged i carrier is inverted. Sometimes, systems are differentially encoded and coheret! detected. simply to avoid these phase ambiguities. 7 The probability of bit error for coherently detected, differentially enc! © PSK is given by [7] $ This relationship is plotted in Figure 3.22. Notice that there is a slight deg- jon of error performance compared to the coherent detection of PSK. This the differential encoding since any single detection error results in two . Error performance for the more popular differentially coherent ) is covered in Section 3.7.5. Bit Error ly Detected FSK = $2 Equations (3.83) and (3.84) describe the probability of bit error for coherent an- tipodal signals. A more general treatment for binary coherent signals (not limited > to antipodal signals) yields the following equation for Pa [8]: Lop 2 ° Vim a 6. 1 Pie aeons ( >) du 3.96) From Equation (3.91), p = cos @ is the time cross-correlation coefficient between 4 signal s,(1) and s2(1), where 6 is the angle between signal vectors s; and s2 (see Figure 3.6). For antipodal signals such as BPSK, 6 = =, thus p = ~I. For orthogonal signals such as binary FSK (BFSK). @ = 7/2, since the 5; and s2 vectors are perpendicular to each other; thus p = 0, as can be verified ic Pa = | eam exp ( G.97) where the co-error function, Q(x), is defined in Equation (3.82). The result could also have been obtained by noting that the energy difference between the or- thogonal signal vectors, s; and s2, with amplitudes of VE,, as shown in Figure 3.20b, can be computed as the square of the distance between the heads of the orthogonal vectors, to be Ey = 2Ep. Using this result in Equation (3.92) yields 4 the same result as in Equation (3.97). Equation (3.97) is plotted in Figure 3.22 (coherent detection of FSK). If we compare Equation (3.97) with Equation (3.84). we can see that 3 dB (a factor of 2) more E,/No is required for BFSK to provide the same performance as BPSK. It should not be surprising that the performance ‘© of BFSK signaling is worse than BPSK signaling, since for a given signal power. -€, orthogonal vectors are spaced closer to one another than antipodal vectors. The bit error probability, Px, for the coherent detection of orthogonal band- oj Pass signals as seen in Equation (3.97) is the same as the Pz for the matched filter +" detection of baseband unipolar signals in Equation (2.64). As mentioned earlier. “the details of on-off keying (OOK) are not treated in this book. However, it is worth noting that the Pg, described in Equation (3.97), is also identical to the error performance for the coherent detecti OOK signaling (matched filter reception). 7" ? 161 Sec. 3.7 Error Performance for Binary Sys 3.7.4 Probability of Bit Error for Noncoherently Detected FSK Consider the equally likely binary FSK signal set, (s1(1)}, defined in p G.27) as follows: sl) = 2 costa + #) Os1ST f= 1,2 The phase term, 6, is unknown and assumed constant. The detector jg ch: terized by M = 2 channels of bandpass filters and envelope detectors, 88 sh in Figure 3.18, The input to the detector consists of the received signa) s(t) + n(t), Where n(1) is a white Gaussian noise process with two-sideq ~) * spectral density, No/2. Assume that s(t) and s2(t) are separated in f; sufficiently that they have negligible overlap. We start the probability of ¢™? Pp, computation the same way that we did for coherently detected PS oh Equation (3.78). ation Ps = $P(Hs)) + $P(Ails2) le Ler 8% = 5 Jy reelsr) de + cf P(z|S2) dz For the binary case, the test statistic, z(T), is defined by z(1) — z,( that the bandwidth of the filter, W,, is 1/7, so that the envelope of hehe is (approximately) preserved at the filter output. If there was no noise atte receiver, the value of 2(T) = V2E/T when s\(1) is sent, and 2(7) = ~ViEq when s(t) is sent. Because of this symmetry, the optimum threshold is =0. The pdf p(z|s,) is similar to p(z|s2); that is, wee P(ds1) = p(—z|s2) 3.99) Therefore, we can write 7 Po = fi" plelsa) de 3.100 or Pa = P(z, > zaJs2) G.100) where 2, ai shown in Pe sere ne Outputs z1(7) and z2(7) from the envelope detectors that (2) = 53) + n(0), the cua nate tHE tONE Sa(¢) = cos ag is sent, th only; it has no signal com © output, 2:(7), is a Gaussian noise random variable velope detector veld Roonent. A Gaussian distribution into the nonlinear er felds a Rayleigh distribution at the output [8), so that J exp ( a Plzils2) = ¢ % x) z2=0 ui) 0 162 a<0 cap.$ Bandpass Modulation and Demodulation 2 where o@ is the noise at the filt " ler out distribution, since the input cet pul. On the other hand, z2(T) has a Rician he lower envelope detecto {8}. The pdf. p(zalsa), is written ae ris 4 sinusoid plus noise & (3 + A) A exp | G4 42] |, (aa Plzals2) = 4% [ 208 |a( a aed ™ : Lg 3.103) 0 a<0 eee pertzend a before, vB is the noise atthe filter ouput. The function tts). ko fied zero-order Bessel function of the first kind {9], is 1 opin To(x) = seit exp (x cos 6) d8 (3.104) When s2(1) is transmitted, the receiver makes an error whenever the en- velope sample z,(7) obtained from the upper channel (due to noise alone) exceeds the envelope sample zx(7) obtained from the lower channel (due to signal plus noise). Thus the probability of this error can be obtained by integrating p(zi|s2) with respect to z; from z2 to infinity, and then averaging over all possible values of zz. That is, a Pa = Plz > 22Is2) =f otals)| {- oes ds] dee (3.105) /. oe 2 ie 2 exp( 54) de] dzx (3.106) Fo. (ia _@+ A?) 2A) % ei; Sex| =e }(3) where A = V2E/T and where the inner integral is the conditional probability of an error for a fixed value of zz, given that s(t) was sent, and the outer integral averages this conditional probability over all possible values of zz. This integral can be evaluated [10], to yield ‘ ee aw 3.107 Bot oP, 408, (3.107) Using Equation (1.19), we can express the filter output noise, 06, as a-2(%)m 2 where G,(f) = No/2and W;is the filter bandwidth. Thus Equation (3.107) becomes (3.108) Equation (3.109) indicates that the error performance depe filter bandwidth, and that Pz becomes smaller as Wyis d Sec. 3.7 Error Performance for Binary Systems ht _— ! valid only when the intersymbol interference OSI) is neslivible, ‘ihe allowed (efor an ISI iy ablanned rem Fquation (2.77) vith the iti y lore Thus Wy Ke bitsy W/L, and we ean write bquation Cogn , a te Lexp ( inc) : 2 4Ni, By, ~ exp (- ) = 2No, Bun, where F), ~ (1/2)A77 is the energy per bit. When comparing the error Perform, of noncoherent PSK with coherent FSK (see Figure 3.22), it is seen tha same Py, noncoberent FSK requires approximately 1 dS more EyINg the for coherent FSK (for Py = 107%), The noncoherent receiver is easier nim ment, since coherent reference signals need not be generated. Therefore, 4 all FSK receivers use noncoherent detection. It can be seen in the fol section that when comparing noncoherent FSK to noncoherent DPSK, the 3-dB difference accurs as for the comparison between coherent FSK and cohereg PSK. As mentioned earlier, the details of on-off keying (OOK) are not treated a this book. However, it is worth noting that the bit error probability, Pe, described in Equation (3.111) is identical to the Py for the noncoherent detection of OOK signaling. 3.7.5 Probability of Bit Error for DPSK Let us define a BPSK signal set 2E al) = F cos (oot + b) OstsT 2E xt) = “F 508 (wot + b = mr) OstsT A characteristic of DPSK is that there are no fixed decision regions in the sip space. Instead, the decision is based on the phase difference between successive received signals. Then for DPSK signaling we are really transmitting each bit the binary signal pair B11) Si) = Cm) or (2,02) OS 1 5 2T Gd) 8) = (nm) or (a,x) 050527 i a where (x1, 4) (i,j = 1, 2) denotes x41) followed by x(t) defined in Equi (3.112). The first T seconds of each waveform are actually the last 7 secon the previous waveform. Note that s\(1) and s2(r) can each have either 4 cheb 16e Bandpass Modulation and Demodulation Titan =» 7 Hthat 42) and 91) are antipedal signals. Ehus the ¢ ehatian, S20) fOF any combination of forms can be written as "3 ” 2Q) = I sss dt Jo Chtddy 1 ; = if bacaP dt = [, [nor dr = 0 Therefore, paits of DPSK signals can be represented as orthogonal signals 27 seconds long, Detection could correspond to noncoherent envelope detection with four channels matched to cach of the possible envelope outputs, as shown in Figure 3.23a. Since the two envelope detectors representing cach symbol are negatives of each other, the envelope sample of each will be the same. Hence we can implement the detector as a single channe! for s4(1) matched to either (44. 6) or (vz, .t2), and a single channel for s:() matched to cither (xy, x2) OF a, ti), as shown in Figure 3.23b, The DPSK detector is therefore reduced to a standard two-channel noncoherent detector. In reality, the filter ean be matched to the difference signal so that only one channel is necessary, For orthogonal signa this operates with the bit error probability in Equation (3,111), Since the DPSK signals have a bit interval of 27, the s4¢) signals defined in Equation (3.113) have Fitters matched to signal envelopes fan tof Decision 2 i ape fae ea 5, Lag. (ab Filters matched to signal envelopes. -———— fab] Decision din q a de ' ie Figure 3.23) DPSK detection. (a) Four: ee a channel differentially coherent Lf} . Sereton af inary BPSK. Equivalent two-channel detector for co) 4 165 Sec. 3.7 Error Performance for Bint ae 1 Po = 56x eS ®) t Ne 3.115) herent detection of differentially encoded PSK, or simply DPSK. This expression is valid for the optimum DPSK detector shown in Figure 3.16c. For the detector shown in Figure 3.16b, the error probability will be slightly inferior to that given in Equation (3.115) [5]. When comparing the error performance of Equation (3.115) ith that of coherent PSK (see Figure 3.22), it is seen that for the same Pg. DPSK requires approximately | dB more Es/No than does BPSK (for Pa = 10°‘). Iris easier to implement a DPSK system than a PSK system, since the DPSK receiver does not need phase synchronization. For this reason, DPSK, although less ef- ficient than PSK, is sometimes the preferred choice between the two. = 3.7.6 Comparison of Bit Error Performance for Various Modulation Types 4 The P, expressions for the best known of the binary modulation schemes dis- m cussed above are listed in Table 3.1 and are illustrated in Figure 3.22. For Ps 10~*, it can be seen that there is approximately a 4-dB difference between the best (coherent PSK) and the worst (noncoherent FSK) that were discussed here. In some cases, 4 dB is a small price to pay for the implementation simplicity gained in going from coherent PSK to noncoherent FSK; however, for other cases. even a 1-dB saving is worthwhile. There are other considerations besides Pg and » system complexity; for example, in some cases (such as a randomly fading chan- %» nel), a noncoherent system is more desirable because there may be difficulty in establishing and maintaining a coherent reference. Signals that can withstand significant degradation before their ability to be detected is affected are clearly desirable in military and space applications. | TABLE 3.1. Probability of Error for Selected Binary Modulation Schemes Modulation Ps a Coherent PSK Q am, we Noncoherent DPSK. 3.8 M-ARY SIGNALING AND PERFORMANCE 3.8.1 Ideal Probability of Bit Error Performance The typical probability of error versus £,/Na curve way shown to have a waterfall: like shape in Figure 3.21. The probability of bit error (/’,) characteristics of various binary modulation schemes in AWGN also display this shape. as shown in Figare any should an ideal Py versus Ex/Ny curve look like? Figure 3.24 displays - the ideal characteristic as the Shannon fimit. The limit represents the threshold E,/No below Which reliable communication cannot he Maintained. Shannon's work is described in greater detail in Chapter 7. We can describe the ideal curve in Figure 3.24 as follows. For all values of Ex/No above the Shannon limit, Pa is zero. Once Ey/No is reduced below the Shannon limit, Py degrades to the worst-case value of 1. (Note that Py = 1 is not __ the worst case for binary signaling, since that value is just as good as Py = 9; if the probability of making a bit error is 100%, the bit stream could simply be _. inverted to retrieve the correct data.) 1t should be clear, by comparing the typical Ps curve with the ideal one in Figure 3.24 that the large arrow in the figure describes the desired direction of movement to achieve improved Px performance. Typical Pg versus Ey/No curve Direction of movement for Pg improvement Ep /No (dB) ‘Shannon Himit Figure 3,24 Ideal Py versus Es/No 1.648 curve. 3.8.2 M-ary Signaling Let us review M-ary signaling. The processor considers & bits at a time. It instructs the modulator to produce one of M = 2* waveforms: binary signaling is the special case where k = 1. Does M-ary signaling improve or degrade performance? Be careful with your answer—the question is a loaded one. Figure 3.25 illustrates the probability of bit error, Pale. versus E,/No for coherently detected orthog- Sec.38 Mary Signaling and ee . 167 Bit error probability, Pg(M) k=e—+ Ep/No (48) 3.25 Bit error probability for coherently detected M-ary orthogonal sig: (Reprinted from W. C. Lindsey and M. K. Simon, Telecommunication Systems Engineering, Prentice-Hall, Inc., Englewood Cliffs, N.J., 1973, courtesy ae C. Lindsey and Marvin K. Simon.) A ni 7 _ onal M-aty signaling over a Gaussian channel. Figure 3.26 similarly illu’ © Pa(3) versus Es/No for coherently detected multiple phase M-ary signalite © Gaussian

You might also like