PRINCIPLES OF DIGITAL DA TRANSMISSION ercentage of communicati s hroughout most of the cee Se SCS ait aa at begat ion system anal m. However, by the en 5 , 7 dominate Te does not ‘yed to look hard to witness the See ae ont communications: from audiocassette tape to MP3 and CD, Me ae ah cc i traditional telephone to VoIP, and from VHS videotape to D Wea ne a poli digital technologies such as digital multimedia communication, digi ie ; ar renee di ing, and business transactions (e-commerce, e-business). In fact, were na refuge of by radio iy facing a strong digital competitor in the form of satellite radio. Given the dominating of digital communication systems in our lives today, it is never too early to study the basic p various aspects of digital data transmission, as we will do in this chapter. This chapter deals with the problems of transmitting digital data over a channel. Hence, the messages are assumed to be digital. We shall begin by considering the binary case, where the data of only two symbols: 1 and 0, We assign a distinct waveform (pulse) to each of these two s resulting sequence of these pulses is transmitted over a channel. At the receiver, these pulses are d are converted back to biniary data (Is and 0s). 8.1 DIGITAL COMMUNICATION SYSTEMS A digital communication system consists, of several components, as shown in Fig. 8.1. In this se conceptually outline their functionalities in the communication system. The details of their an design will be given in dedicated sections later in this chapter. 8.1.1 Source The input to a digital system takes the form of a sequence of digits. The input could be the out ata set, a computer, or a digitized audio signal (PCM, DM, or LPC), digital facsimile or HDTY, ortel data, and so on. Although most of the discussion in this chapter is confined to the binary case (cot ci Schemes using only two symbols), the more general case of M-ary communication, which uses will also be discussed in Secs, 8.7 and 8.9. 8.1.2 Line Codera oo 8.1 Digit igital Communication Systems - | Feiss eH 1011010. Other Figure 8.1 Fund lomental nal building blocks of digital communication systems. adis transmitted by no i for, = 1 is transmitted ye ae signal) as shown in Fi is v most power-effici (and 0s transmit ae fe oe ae i Cie Cee ate fla is transmitted 3 “a Another commonly ee soni) popular code in PCM i requires the least pulse —p() (Fig 820). The Pee rn ‘AMD. tere 0 is encoded by ni is bipolar, also ee power for a given noise immunity (error i 10 pulse and 1 is Guu y Ne or alternate mark a pulse p(t) or PO) dependin gon i) 3 revious 1 is encoded by ~ Ny —p(t) or p(t). In : presenting consecutive 1 Jternate in Per al min Fig. 8.2¢. This cod is code has the adv: antage that if one single error is made in the d ye detecting © nee Jse sequence will violate th ed (although ill violat i f the bipolar rule and the error can be detected (alth though not 35 500" Siig ne received pul ei ese) immediately:” i (b) a © @ ce to © Figure 8.2. Line code examples: (9) ‘on-off (RZ); (6) polar (RZ): (6) bipolar (RZ (4) on-off (NRZ}; (e) Polar (NRZ) tn cequence cou! | thei respective effects and remain ~ -rrors. Eve for single errors. we cannot tell cing the| oa Pe re Mk Ne 8.2 ine Coding 51) ap tnt ee @) en oe pick (b) L 47 NMA nnn t i—_ © igure 8.3. An ovo signa i @sum ofa random polar signa (bond a lock frequency periodic signal) 42 NE CODING sl data can be transmitted by various transmission or line codes. We have given examples of on-off, ty and bipolar. Each line code has its advantages and disadvantages. Among other desirable properties. @ we sould have the following properties: « Tonsnission bandwidth should be as small as possible. « Power efficiency. For a given bandwidth and a specified detection error rate, the transmitted power should teas low as possible. « Error detection and correction capability. Itis desirable to detect, and preferably correct, detection errors. Inabipolar case, for example, a single error will cause bipolar violation and can easily be detected. Error comecting codes will be discussed in depth in Chapter 13. + Favorable power spectral density. It is desirable to have zero power spectral density (PSD) at f = 0 (de) tecause ac coupling and transformers are often used at the repeaters.* Significant power in low-frequency components should also be avoided because it causes de wander in the pulst used + Adequate timing content. It should be possible to extract timing or clock information from the signal. * Transparency. It should be possible to correctly transmit a digital signal regardless of the pattern of 1s and (s. We saw earlier that a long string of Os could cause problems in timing extraction for the on-off and bipolar cases. A code is transparent if the data are so coded that for every possible sequence of data, the coded signal is received faithfully. 82.1 PSD of Various Line Codes ‘Example 2.30 we discussed a procedure for finding the PSD of a polar pulse train. We shalll use a similar "cedure to find a general expression for PSD of the baseband modulation (line coding) output signals as pe = Sdiioogs ing is required beeause the de paths provided by the cable pais between the repeater sites are used to transmit the power Fate the repeaters,Et a ee a ‘ “B2Line Coding 515 igaiane bere on ‘nd Width 2¢ centered at r = 0 (Fig. 8.54), This is expected aoa Fig 8d. Dat en weap between the delayed signal f(r — x) and (0); hence, na een th pulse of Sey neste frther, we find thatthe lah pulse of £(¢~ ¥) will 1 a eve another at? 88 € Approaches 7, (Fig. 8.50), Repedting the earlier argument, 0 yg Ri(t) Will have another triangular pulse of width 2e centered at ¢ Ty and of height Ri/€T Ce ve q ‘ Ri = tim — 5a, fac _ et = dim TT »; oy » ith = dea that Ry is obtained by multiplying every ene Pulse strength (a,) by the strength of its immediate neigh- igi)» Adding all these products, and then a ~ dividing by the total number of pulses, This is clearly the rege (mean) ofthe product aka4+1 andi, in our notation, arr. A similar thing happens around we tty, To, ---- Hence, R(x) consists of a sequence of triangular pulses of width 2c centered at 4Tp, #2Tb, .--» The height of the pulses centered at nT, is Rq/€Ts, where im Db L R= lim — = lim — Him, 7 Li sedken = im. = aden f,isesetially the discrete autocorrelation function of the line code symbols (at). find Re(r), we let — O in Rz(r). As € — 0, the width of each triangular pulse—> 0 and the ixigh> 0 in such a way that the area is still finite. Thus, in the limit as ¢ —> 0, the triangular pulses ‘xcone impulses. For the nth pulse centered at n7, the height is R,/éT» and the area is Ry/Ts. Hence, Fig 85e) 1 Relt) = DY de =n) (8.8) % namo ‘he PSD 5.(/) is the Fourier transform of R(t). Therefore, LS pitas SN =a Rae we (8.9) “eontzing that Ry = Ry (because P(r) isan even function of r}, we have Ss : oo = [m2 spasm ax] (8.10)516 PRINCIPLES OF DIGITAL DATA TRANSMISSION The input x(¢) to the | filter with impulse response h(t) = p(t) results in the output y(t), ag TEP) e=> PY), the transfer function ofthe filer is HGP) = PC), and according re ss Sy) = PAPE * woe = newer] Pee S al ner [x 425 Ry cos rant | nat Thus, the PSD of a line code is fully characterized by its R,, and the pulse-shaping selec now use this general result to find the PSDs of various specific line codes by first det autocorrelation Ry. 8.2.2 Polar Signaling In polar signaling, 1 is transmitted by a pulse p(#) and 0 is represented by p(t). In this likely to be 1 or —1, and aj is always 1. Hence, oe moe inp ‘There are N pulses and az = 1 for each one, and the summation on the right-hand side above Ry = lim +(N)=1 dane ©) Moreover, both a; and ay are either 1 or —1. Hence, axa, 1 is either 1 or —1. Because ax is equally likely to be 1 and —I on the average, out of NV terms the product axaii 18 terms and is equal to —1 for the remaining NV/2 terms. Therefore, Possible values of axak+1 a 1, N n= jin t[40+ Fen] =0 we see that the product again is also equally likely to be 1 or 1. Asguing this way, R, 30, > nel ‘Therefore from Eq. (8.11¢) a a _ POP / SY Ts Ro r PAP — Por ae alPon accenucam: 57 comparison of various schemes, we shall consider a specific shape. Let p(#) be ® i nT Cok age aa) at p a wi? oo pw) =n () 3 n(2) a P= sine (2) (8.14) 2 2 po Ty 7m SP) = — sinc? (72% (8.15) 3) = 2 sine ( f ( ve 86 shows the spectrum S,(f). It is clear that the polar signal has most of its power concentrated in eee fequencicS Theoretically, the spectrum becomes very small as frequency increases but never becomes ma eaisipove a certain frequency. To define a meaningful measure of bandwidth, we consider its first /requency to be its essential bandwidth.* iay signal spectrum, the essential bandwidth of the signal is seen to be 2Rp Hz (where Rp is the ese From pol ¢ a ina Ryt Se! wa nt Foster ransform of both sides ofthis equation, and use the fact that &(t — nT) <=> e-P2> ana = 5 ~ nky). This yields Sgt t a a(- =)$20 PRINCIPLES OF DIGITAL DATA TRANSMISSION Substitution of this result in Eq, (8.18b) yields an Z'(-4) ib n==00 1 SO = Fy and the desired PSD of the on-off waveform (0 is [from Bq. (8.1 1a)] Poel, 1 $5 ( A SO = “an |! ay De BE zi) Note that untike the continuous PSD spectrum of polar signaling, the on-off PSD of fg, cg Additional discrete part. This diserete part may be nullified ifthe pulse shape is chosen tueh n P( —)=0, n= 0, £1,. (a) For the example case of a half-width rectangular pulse [see Eq, (8.14)], Tr 1 (Ea SN = The resulting PSD is shown in Fig, 8.8. The continuous component of the spectrum is (T),/16) sine! This is identical (except for a scaling factor) to the spectrum of the polar signal (Eq. ¢ discrete component is represented by the product of an impulse train with the continuous ¢p (Ti/16) sinc? (xx/T}/2). Hence this component appears as periodic impulses with the continuous come as the envelope. Moreover, the impulses repeat at the clock frequency Ry = 1/T), because its fu frequency is 21/T), rad/s, or 1/T» Hz. This is a logical result because as Fig. 8.3 shows, an o can be expressed as a sum of a polar and a periodic component. The polar component y(t) is the polar signal discussed earlier, Hence, the PSD of this component is one-fourth the PSD in The periodic component is of clock frequency Ry; it consists of discrete components of frequency harmoni On-off signaling has very little to brag about, For a given transmitted power, itis less imm interference than the polar scheme, which uses a positive pulse for 1 and a negative pulse because the noise immunity depends on the difference of amplitudes representing 1 and 0. same immunity, if on-off signaling uses pulses of amplitudes 2 and 0, polar signaling need use of amplitudes 1 and —1. It is simple to show that on-off signaling requires twice as much powel signaling. Ifa pulse of amplitude | or ~I has energy E, then the pulse of amplitude 2 has energy (2) Because 1/7, digits are transmitted per second, polar signal power is (E)(1/Tp) = E/T». For ‘on the other hand, each pulse energy is 4, though on average such a pulse is transmitted 0 time while nothing is transmitted over the other half. Hence, the average signal power: of on-off yj 1\_ 2B ; miata) em ce that required for the polar signal. Moreover, unlike the polar case, on-off si rent. Along sting of Os (or offs) causes the absence of a signal and can lea to erors 0 addition, all the disadvantages of polar signaling, (e.g., excessive trans! jion bandwi oan at i ‘no error detection (or correction) capability) are also present in on-off si which is twiB i f a2tine Coding 524 AR, -3Rb -2Ry Figure 8.8 PSD of an onoif signal. 32.5 Bipolar Signaling pesiling scheme used in PCM for telephone networks is called bipolar (pseudoternary or alternate mark -sei), A 0 is transmitted by no pulse, and a 1 is transmitted by a pulse p(t) or —p(t), depending on hier the previous 1 was transmitted by ~p(®) or p(t). With consecutive pulses alternating, we can avoid ixsander and thus cause a de null in the PSD. Bipolar signaling actually uses three symbols (p(t), 0, and “(and hence, it is in reality ternary rather than binary signaling. Tocalculate the PSD, we have R= in ay Nooo N Onihe average, half of the ays are 0, and the remaining half are either 1 or —1, with a? = 1. Therefore, LTE ha ee ot - —|—(41)? + =| = = Be ny [pe 30S Tocompute Ry, we c STwobits: 11, 10, O1, letthee ofthese sequ onsider the pulse strength product a,aj.41. There are four e qually likely sequences and 00. Since bit 0 is encoded by no pulse (ax = 0), the product aay, is zero for the uences. This means, on the average, that 3V/4 combinations have ajai ah = Oand only ‘ny prubnations have nonzero aya. Because of the bipolar rule, the bit sequence 11 can be encoded vain) Consecutive pulses of opposite polarities, This means the product aac4, = —1 f Role Tene pol Dp AA for the N/4 hy, Smpute Ry in a similar way, “ombinations of three a we need to observe the product ayay42. For this, we in sequence, There are eight equally likely combit © need to consider all nations: 111. 1M1 140522 PRINCIPLES OF DIGITAL DATA TRANSMISSION 100, 011, 010, 001, and 000. The last six combinations have either the first ‘49k+2 = 0 for all these six combinations. The first two combinations are the only « Fh Ok+2. From the bipolar rule, the first and the third pulses in the combination in Yielding axaz42 = 1. But for 101, the first and the third pulse are of opposite Polarity Thus, on the average, aya; = | for N/8 terms, —1 for N/8 terms, and 0 for 3/4 te Resin —|% as Nao N LB N 3N (I+ z p+ 3 In general ie jee lim Pg Na¥00 For n > 2, the product @xae¢n can be 1, —1, or 0. Moreover, an equal number of combi and —1. This causes Ry, = 0. Thus Rn=0, n>t P(f)|2 and [see Eq. (8.1 1c)] : SN-= wore —cos 2x/Tb] 2 = mors? (fT) Note that 5\(f) = 0 for f = 0 (de), regardless of P(f). Hence, the PSD has a de null, which ac coupling. Moreover, sin? (xT) = 0 at f = 1/Tp, that is, at f = 1/T = Rp Hz. Thus, re we are assured of the first non-de null bandwidth Ry, Hz. For the half-width pulse: SN) = 7B sine? (®) sin? (fT) This is shown in Fig. 8.9. The essential bandwidth of the signal is Ry (Rp = 1/Tp), which is half using the same half-width pulse or on-off signaling and twice the theoretical minimum band Cen phase \, Polar 0 Rp 2Ry f— Figure 8.9 PSD of bipolar, polar, and splitphase signals normalized for equal power Halt oe PrB.2tine Coding 529 » obtain the bandwidth Ry, for polar (or on-off) case for full-width pulse. For the bipolar le 1 a, Hz. whether the pulse is half-width or full-width. oat ts ing, has several advantages: (1) its spectrum has a de null; (2) its bandwidtl iY as single-crror-detection capability. This is because even single detection ©7 of the alternating pulse rule, and this will be immediately detected. If & PiPO” ified, We get an on-off signal that has a discrete component at the clock fee) revantages of a bipolar signal is the requirement for twice as much power (2°) | needs. This is because bipolar detection is essentially equivalent to on-off LG ection point of view, One distinguishes between -+p(1) or —p(/) from 0 rather thar ei ion viola ism e di vantage of bipolar signaling is that it is not transparent. In practice, various substitution nt long strings of logic zeros from allowing the extracted clock signals to dri cuss two suich schemes, jomple 8.1 jer the scheme shown in the figure given below, where (x4) is a binary sequence 11101001. Show cs jean AMI coded sequence (inverted bit means its value becomes — (old bit value)). js the decoded binary sequence, genic operations begin at ¢ = 0 when the first bit xo arrives at the system input. sume that 1-1, )-1, 2-1, Y_y»2_, are of value 0. ty) be) tt) : Delay =Ty Delay = 7 Precoder ‘Binary to temary caer naa ‘Ternary to binary a poral T,: One bit interval Posooder @ Exclusive OR524 PRINCIPLES OF DIGITAL DATA TRANSMISSION Solution Ve =O Yk Ms MO yoy 100 fa ae ywety-1 ON ow y, 18D? OTT ge gas Pe ee | in (ei) marks (i.e, bit value = 1) are at xo, x1.22.%3, 45 and x7, Heth B= 1 y= m1, a= 1, y= land 27 = ‘ence alternate bits are inverted and {z4) represents an AMI coded signal, Yeeutyers; 010110004 Note that sequence (y/} and {y) are the same, w=yW Out: 0111010014 Thus (x) = (x1), implying that AMI signal is correctly decoded to yield the input bi should be noted that above operations hold for any arbitrary bit sequence (x,), a High-Density Bipolar (HDB) Signaling The HDB scheme is an ITU (formerly CCITT) standard. In this scheme the problem of in bipolar signaling is eliminated by adding pulses when the number of consecutive 0s exes eet coding is designated as high-density bipolar coding (HDBN), where N can take on . 3,.... The most important of the HDB codes is HDB3 format, which has oe ip ch has been adopted as anj The basic idea of the HDBN code is that when a run of N + 1 zeros occurs, this replaced by one of the special N + 1 binary digit sequences. To increase the timing content ofj the sequences are chosen to include some binary 1s. The 1s included deliberately viol i easy identification of the substituted sequence. In HDB3 coding, for example, the special 000V and BOOV where B=1 that conforms to the bipolar rule and V=1 that violates the b choice of sequence 000V or BOOY is made in such a way that consecutive V pulses alternate de wander and to maintain the de null in the PSD. This requires that the sequence BOOV bev are an even number of 1s following the last special sequence and the sequence 000V be used whe an odd number of 1s following the last sequence. Figure 8.10a shows an example of this coding. N the sequence BOOV, both B and V are encoded by the same pulse, The decoder has to check two ti bipolar violations and the number of 0s preceding each violation to determine if the pre substitution. Despite deliberate bipolar violations, HDB signaling retains error detecting capability. will insert a spurious bipolar violation (or will delete one of the deliberate violations), apparent when, at the next violation, the alternation of violations does not appear. This also erate violations can be detected despite single errors. Figure 8.10b shows the PSD of HDB3: of a bipolar signal to facilitate comparison.* ' Binary with N Zero Substitution (BNZS) Signaling ‘A class of line codes similar to HDBN is the binary with NV zero substitution, oF ‘Av zeros occur in succession, they are replaced, by one of the two special sequences ¢oml8,3 Pulse Shaping id 1011100001011 010000000000101101010009% 4 TOU OU i Md os (b) Figure 8.10 (a) HD83 signal and (b) its PSD. aie timing content. There are deliberate bipolar violations just as in HDBN. Binary with cight-zero vention (B875) is used in DS! signals of the digital telephone hierarchy in Chapter 5. It replaces any sf eight 210s in length with a sequence of ones and zeros containing two bipolar violations. Such a wes unlikely © be counterfeited by errors, and any such sequence received by a digital channel bank ieoed bya string of eight logic zeros prior to decoding, The sequence used as a replacement consists of pate 000VBOVB. Similarly, in B6ZS code used in DS2 signals, a string of six zeros is replaced with (WBOVB, and DS3 signal features a three-zero B3ZS code. The B3ZS code is slightly more complex than deabers in that either BOV or OOV is used, the choice being made so that the number of B pulses between cawecatve V pulses is odd. These BNZS codes with N = 3,6, or 8 involve-bipolar violations and must ‘ure be carefully replaced by their equivalent zero strings at the receiver. Tiere are many other transmission (line) codes, too numerous to list here. A list of codes and appropriate sfeences can be found in Bylanski and Ingram.? §3 PULSE SHAPING TESDS;() of digital signal y(t) can be controlled by a choice of line code or by P(f), the pulse shape. ae Wwe discussed how the PSD is controlled by a line code. In this section we examine how De PSD greed by the pulse shape p(t), and we learn how to shape a pulse p(t) to achieve a desired SN. Pot ypu 8 stongly and directly influenced by the pulse shape p(t) because S,(f) contains the ae “tn comparison to the nature of the line code, the pulse shape is a more direct tf shaping the PSD 5, (f), ain a at ROteat factor _526 PRINCIPLES OF DIGITAL DATA TRANSMISSION 8.3.1 Intersymbol Interferences and Effect + ae last section, we used a simple half-width rectangular pulse p(t) for the sake ott ten ng, in this case the bandwidth of S,(f) is infinite, since P(/) has infinite 4a Within cntial bandwidth of $f) was finite, For example, most of the power of a bipg thin the essential band 0 to Rp Hz. Note, however, that the PSD is small but is eae” 7 > Ry He. Therefore, when such a signal is transmitted over a channel of bandwidth vee of Spectrom is transmit, but a small portion us the spectru on ~15, we saw how such a spectral distortion tends to spread the pulse (dispersi beyond its allotted time fecal Ty will cause it to interfere with neighboring pu y oe interference or ISI ” SI is nor noise, ISI is caused by nonideal channels that are not distortic bandwidth. In the case of halfwidth rectangular pulse, the signal bandwidth iy anicnee TSI, as a manifestation of channel distortion, can cause errors in pulse detecti To resolve the difficulty of ISI, let us review briefly our problem. We need to Ty imterval, the Ath pulse being apt — kT)). The channel has a finite bandwidth, ang detect the pulse amplitude a correctly (.e., without ISI). In our discussion so far, we have’ limited pulses. Since such pulses cannot be band-limited, part of their spectra is su mited channel. This causes pulse distortion (spreading out) and, consequently, ISL, We this difficulty by using pulses that are band-limited to begin with so that they can be ‘renal i band-limited channel. But band-limited pulses cannot be time-limited. Obviously, variong geal and cause ISI. Thus, whether we begin with time-limited pulses or band-limited pulseg al cannot be avoided. It is inherent in the finite transmission bandwidth. Fortunately, there is an this blind alley. Pulse amplitudes can be detected correctly despite pulse spreading (or over is no ISI at the decision-making instants. This can be accomplished by a properly shaped bands} To eliminate ISI, Nyquist proposed three different criteria for pulse shaping,* where the pulses: to overlap. Yet, they are shaped to cause zero (or controlled) interference with all the other p decision-making instants. Thus, by limiting the noninterference requirement only at the dex instants, we eliminate the need for the pulse to be totally nonoverlapping. We shall consider only criteria. The third is much less useful than the first two criteria,® and hence, will not be conside 8.3.2 Nyquist’s First Criterion for Zero ISI In the first method, Nyquist achieves zero ISI by choosing a pulst shape that has a nonzero center (say t = 0) and zero amplitudes at t = +n7p (n = 1, 2, 3,...), where Tp is the successive transmitted pulses (Fig. 8.1 1a). Thus, heat (= Zi = tnt, (m=z) A pulse satisfying this criterion causes zero ISI at all the remaining pulse ee instants as shown in Fig. 8.1la, where we show several successive pulses (d 1=0, Tp, 27», 37, .-» To = 1/Ry). For the sake of convenience, we have shown all: Jt is clear from this figure that the samples at 1 = 0, Tp, 27, 37>, ... consist of the pulse (centered at the sampling instant) with no interference from the remaining, ii «Actually, 9 Put iid make the igre needlessly confsing. negative pulses would‘ 527 Agi Alaul 8.3 Pulse Shaping RR R, RR, (b) (c) 8.11 The minimum bandwidth pulse that satisfies Nyquist’ first criterion ond its spectrum: Figure ® sion of is eats theoretical minimum bandwith Ry/2 Het would be nce if 9 Now ee Ny criterion had this minimum bandwidth R,/2 Hz. Can we find such a pulse jars x oy solved this problem (Example 5.2 with B = Ry/2), where we showed that there ea fete te) pulse which meets Nyguis’s criterion (8.23) and has a bandwidth Ry /2 Hz. This pulse. eo ut, (Fig. 8115) has the property 1 1=0 a 1 (8.244) sine (Rpt) { jai, (m=z) Ry joer, the Fourier transform of this pulse is Pe pp=tnll (8.24) = mF) ubich has a bandwidth Rp/2 Hz as seen from Fig. 8.11c, We can use this pulse to transmit at a rate of Ry pues per second without ISI, over a bandwidth of only R,/2. This scheme shows that we can attain the theoretical limit of performance by using a sine pulse. Usforunately, this pulse is impractical because it starts at —oo, We will have to wait an infinite time to gerate it. Any attempt to truncate it would increase its bandwidth beyond R,/2 Hz. But even if this fis were realizable, it would have an undesirable feature: namely, it decays too slowly at a rate 1/2, fica some serious practical problems. For instance, if the nominal data rate of Ry biv/s required iets scheme deviates a litle, the pulse amplitudes will not vanish at the other pulse centers, Because pulses decay only as 1/t, the cumulative interference at any pulse center from all the remaining pulses Sol the form S>(1/n). It is well known that the infinite series of this form does not converge and can : “to a very large value. A similar result occurs if everything is perfect at the transmitter but the sam- ts Tale atthe receiver deviates from the rate of R, Hz. Again, the same thing happens if the sampling Nats deviate alittle because of pulse time jitter, which is inevitable even in the most sophisticated sys- eee therefore fails unless everything is perfect, which is a practical impossibility. And all Win sine (77Ryt) decays too slowly (% pulse p(é) that satisfies s (as 1/1). The solution is to find a “decays faster than 1/, Nyquist has shown that such a pulse requires a bandwidth AR /2, withRy Rb I (b) Figure 8.12 Derivation of the zero ISI Nyquist criterion pulse, This can be proved as follows. Let p(t) <=> P(/), where the bandwidth of P(f) is in the ran (Fig. 8.12a). The desired pulse p(t) satisfies Eq. (8.23). If we sample p(1) every 7, seconds P© by 57, (an impulse train), then because of the property (8.23), all the samples, ex a origin, are zero. Thus, the sampled signal p(t) is f PB) = (sr, () = 5) Following the analysis of Eq. (5.4) in Chapter 5, we know that the spectrum of a sampled signal times) the spectrum of p(#) repeating periodically at intervals of the sampling frequency Ry, Ty Fourier transform of both sides of Eq. (8.25) yields : 1< 1 = P(f = ni where Ry = — i, f — np) oe DY Pe =n) = 7 ina, Thus, the sum of the spectra formed by repeating P(f) spaced Rp apart is a constant Fig. 8.12b.* Consider the spectrum in Fig. 8.12b over the range 0 < f < Rs. Over this range only two ts P(f — R,) in the summation in Eq. (8.27) are involved. Hence ‘ 3 Pf) + POF — Ro) = Toy O 2B, where B is the bandwidth (in hert) of P(/), the repetitions of P(f are non cannot be satistied. For Ry = 2B, the condition is satisfied only forthe ideal lowpass PLAC) = sine Hence, we must have B > Ry/2.29 8,9 Pulse Shopind e beeen be fy fyi Figure 8.13 Vesligial (cisedcosine) spectrum vel) te) «il! Ry R : o(s+ 5) +0( -%) = be |x| < 0.52, re! vac conivgte symmetry property (Eq, 2.799) on Eq, (8:28) yields eof P é a) +P (@ - ) =, Ix] < 0.5K), (8.29) aos Uf) 10 be real-valued and positive, then only [P()] needs to satisfy Eg. (8.29). Because POD vg. (8.29) implies Rp Ry Re V4 [p(B bG *) | G jee, [P(P)| should be of the form shown in Fig. 8.13. This curve has an odd symmetry about the set of sxsintesecting at point o (the point on |P(f)| curve at f = Ry/2}. Note that this requires that =T, |x| < O.5Rp (8.30) |P(O.SRp)| = 0.5|P(0)| The bandwidth, in hertz, of P(f) is O.5Rp + fe, where f, is the bandwidth in excess of the minimum Saividth R,/2. Let r be the ratio of the excess bandwidth f, to the theoretical minimum bandwidth R,/2: Excess bandwidth pee ae Theoretical minimum bandwidth ae 0.5K, = Ts Sbierve that because f; cannot be Jarger than Ry/2, (8.31) Osrsl a (8.32)Figure 8.14 Pulses satistying Nyquist’ firs! criterion: solid curve, ideal fy = 0 (r = 0); light dos fe = Rye = 0.5); heavy dashed curve, fe = Rj /2(r = 1). ie In terms of frequency f, the theoretical minimum bandwidth is Rp/2 Hz, and the excess Ry /2 Hz. Therefore, the bandwidth of P(f) is rR, 1+0Rp fig ew ‘The constant r is called the roll-off factor and is also expressed in terms of percent. For is a Nyquist first criterion spectrum with a bandwidth that is $0% higher than the theoretical roll-off factor r = 0.5 or 50%. A filter having an amplitude response with the same characteristics is required in the band modulation discussed in Sec. 3.5 [Eq. (3.26)]. For this reason, we shall refer to the Eqs. (8.29) and (8.30) as a vestigial spectrum. The pulse p(t) in Eq. (8.23) has zero IST at all other pulses transmitted at a rate of Rp pulses per second. A pulse p(¢) that causes zero, ters of all the remaining pulses (or signaling instants) is the Nyquist first criterion pulse. We that a pulse with a vestigial spectrum [Eq. (8.29) or Eq. (8.30)] satisfies the Nyquist’s fist c zero ISI. Because 0 ) Figure 8.16 (o) The minimum bandwidth pulse that satisfies the duobinory pulse criterion and (b) its spectrum. ceToot 547 ple value is 8.6 Eye Diagrams: An Important zg: that is, if the pulse sample value is positive, the digit is detected as 1; if the sa™ ysis itis detected a5 O. ‘ en ne ite 2 decision of whether to declare 1 or 0 could be made readily from the pul tect Ope tert nis random, meaning that its exact value is unpredictable. It may have @ ing instant ise Je negative as well as positive, It is possible that 1 is transmitted but n at the samp! ther hand, tea ve value. This will make the sample value A, +n small or even negative. On ge ae ie med and be eee eb ae a onli ae fate Nena ; ora jit will be detected wrongly as 1. This is clear from Fig. 8.24b. r ol es cee of digital communication systems is typically specified by the average raed the Fe Pee or example, if two cellphones (receivers) in the same spot are attempting (0 etter 3 een from acellular tower, the cellphone with the lower number of detection eco he ee Hans ly to have fewer dropped calls and less trouble receiving clear speech. How hate cover sso sometimes one cellphone may be better while other times the other Came of errors eis rindi real measure of receiver performance is therefore the average ratio of the Leen Maat wot a punber of transmitted data, ca the meaningful performance comparison is the like! o the detection error probability, gesion om ecise analysis and evaluation of ths error likelthood require the knowledge and tools from cai neory, We Will postpone error analysis until after the introduction of Probability it Cie Mater 9, we will discus fully the error probability analysis of different digital communication Lats for different noise models as well as system designs against different noises. For example, aa sem can generally characterize the random channel noise from thermal effects and intersystem cr : st m0 yum detectors can be designed to minimize the error likelihood against Gaussian noise. However, wk OPeMamsients, lightning strikes, power line load switching, and other singular events cause very high oS pals of short duration to contaminate the cable pairs that carry digital signals. These pulses, joe my called impulse noise, cannot conveniently be engineered away, and they constitute the most sent source of errors from the environment outside the digital systems, Errors are virtually never, there- fi Bin insolation, but occur in bursts of up to several hundred at a time. To correct error burst, we Use ‘Suburst error correcting codes described in Chapter 13, ple, except a small stil get, 10 8.6 EYE DIAGRAMS: AN IMPORTANT TOOL Inte last section, we studied the effect of noise and channel ISI on the detection of digital transmissions. \Weaso described the design of equalizers to compensate the channel distortion and explained the timing- extaction process. We now present a practical engineering tool known as the eye diagram. The eye diagram iscasy to generate and is often applied by engineers on received signals because it makes possible the visual examination of severity of the ISI, the accuracy of timing extraction, the noise immunity, and other important ftors. We need only a basic oscilloscope to generate the eye diagram, Given a baseband signal at the channel output 90 = D> ayp(t — kT) ‘Lean be applied to the vertical input of the oscilloscoy ‘ame rate 1/T, as that of the incoming pulses, ‘aismited data symbol ay. The oscilloscope ‘e channel output y(¢). What appears on the 0: 9 T, and then superimposed on top of one human eye, hence the Iervals, or mT), pe. The time base of the scope is triggered at the and it yields a sweep lasting exactly 7, the interval of one shows the superposition of many traces of length T from scilloscope is simply the input signal (vertical input) cut up another. The resulting pattern on the oscilloscope looks like name eye diagram. More generally, we can also apply atime sweep that lasts m symbol The oscilloscope pattern is simply the input signal (Vertical input) cut up ever, po eeof Yuk Zero crossing then superimposed on top of and has the shape of m eyes in a horizontal row. We now present an example. Consider the transmission of a binary signal by ola (Fig. 8.25a). Its eye diagrams are shown in Fig. 8.25b for the time base of 7} and 27), reg this example, the channel has infinite bandwidth to pass the NRZ pulse and there is no channel di Hence, we obtain eye diagrams with totally open eye(s). We can also consider a channel output us same polar line code and a different (RZ) pulse shape, as shown in Fig. 8.25c. The resulting eve; shown in Fig. 8.25d. In this case, the eye is wide open only at the midpoint of the pulse duration. Wi timing extraction, the receiver should sample the received signal right at the midpoint where the munity at the decision point (Sec. 8.5.3). This is because the mid ‘open, to achieve the best noi the eye represents the best sampling instant of each pulse, where the pulse amplitude is interference from any other neighboring pulse (zero ISI). ‘We now consider a channel that is distortive or has finite bandwidth, or both. After passing nonideal channel, the NRZ polar signal of Fig. 8.25a becomes the waveform of Fig, 8.25e. The1 signal pulses are no longer rectangular but are rounded, distorted, and spread out, The eye a not fully open anymore, as shown in Fig. 8.25f. In this case, the ISI is not zero. Hence, pulse respective sampling instants will deviate from the full-scale values by a varying amount in each blurs, resulting in a partially closed eye pattern. Tn the presence of channel noise, the eye will tend to close in all cases. Weaker noise portionately less closing. The decision threshold with respect to which symbol (1 or 0) was the midpoint of the eye." Observe that for zero ISI, the system can tolerate noise of up to opening of the eye. Any noise value larger than this amount can cause a decision error if its sig to the sign of the data symbol. Because ISI reduces the eye opening, it clearly reduces noise | eye diagram is also used to determine optimum tap settings of the equalizer. Taps are adjust maximum vertical and horizontal eye opening. Jao When p(?) and —p(@) are used for 1 ‘and @, respectively). For: two-level decision fe. = Thi for a This is ume fr a wor lewe dor bipolar signaling).Slope shows sensitivity to ‘sampling time Level crossing Distortion at Level-crossing, sampling point ambiguity shows timing jitter Figure 8.26 Reading an eye diagram. onl i i iments. It is not jagram is a very effective tool for signal analysis during real-time experi ae but it also provides very rich and important information about the quality and sesceP* eceived digital signal. From the typical eye diagram given in Pig. 8.26, we can extract Sever regarding the signal quality. opening point. The eye opening amount at the sampling and decision instant et fr f noise the detector can tolerate without making an error. The quantity is known 2s ia he instant of maximum eye opening indicates the optimum sampling or decision-making * y to timing jitter. The width of the eye indicates the time interval over which correct decisicm © ade, and it is desirable to have an eye with the maximum horizontal opening. If the dectss0 nstant deviates from the instant when the eye has a maximum vertical opening. the mares nce is reduced. This causes higher error probability in pulse detection. The slope of the w fast the noise tolerance is reduced and, hence, the sensitivity of the decision noise tolerance of the sampling instant. It demonstrates the effects of timing jitter. ssing (timing) jitter. Typically, practical receivers extract timing information about the pulse Fr ampling clock from the (zero) level crossing of the received signal waveform. The variation sing can be seen from the width of the eye comers. This measure provides information about ' ter such a receiver is expected to experience. we provide a practical eye diagram example for a polar signaling waveform. In this case. ne roll-off pulse that satisfies Nyquist’s first criterion of zero ISI. The roll-off factor is choser The eye diagram is shown in Fig. 8.27 for a time base of 27;. In fact, even for the same signal, s may be somewhat different for different time offset (or initial point) values. Figure 8.27. ill e diagram of this polar signaling waveform for a display time offset of T,/2, whereas Fig. 8. ormal eye diagram when the display time offset value is zero. It is clear from comparison 1 jagrams have a simple horizontal circular shift relationship. By observing the maximum eye or see that this baseband signal has zero ISI, confirming the basic feature of the Taised-cosine pu * hand, because Nyquist’ first criterion places no requirement on the zero crossing of the pu sram indicates that timing jitter would be likely. KA--AA ADV DACEM AIR IAL rain RR$50 PRINCIPLES OF DIGITAL DATA TRANSMISSION ‘Sampling and decision instant @ ) ‘of @ polar signaling system using @ raised cosine pulte with roll of factor @ 4 ‘symbol periods 274, with a time shift T/2; (b) without time shif. 0. Ifthe transmitter would like to send bits at a much higher rate, 7, may be shortened, For ey the bit rate by M, J, must be reduced by the same factor of M; however, there is a heavy in bandwidth, As we demonstrated in Fig. 8.9, the bandwidth of baseband modulation is pee pulse rate 1/7}. Shortening Tj, by a factor of M will certainly increase the required M- Fortunately, reducing 7;, is not the only way to increase data rate. A very effective practical wy allow each pulse to carry multiple bits. We explain this concept here. For each symbol transmission within the time interval of 7) to carry more bits, there must *wo symbols to choose from. By increasing the number of symbols to M, we ensure that the transmitted by each symbol will also increase with M. For example, when M = 4 (4-ary, 9 we have four basic symbols, or pulses, available for communication (Fig. 8.28a). A sequence of four possible sequences (viz., 11, 10, 01, and 00). Because we have four distinct symbols assign one of the four symbols to each of these combinations (Fig. 8.28a). Fach symbol now oce duration of T;. A signaling example for a short sequence is given in Fig. 8.28b and the 4-ary shown in Fig. 8.28c. ‘This signaling allows us to transmit each pair of bits by one 4-ary pulse (Fig. 8.28b).. n bits, we need only (n/2) 4-ary pulses. This means one 4-ary symbol can transmit the binary digits. Also, because three bits can form 2 x 2 x 2 = 8 combinations, a group of t transmitted by one 8-ary symbol. Similarly, a group of four bits can be transmitted by one I general, the information Iyy transmitted by an M-ary symbol is I = logy M bits Dhis means we can increase the rate of information transmission by increasing M. ‘This special M-ary signaling is known as the pulse amplitude modulation (PAM) | information is conveyed by the varying pulse amplitude, We should note here that PAM is & ible choices of M-ary signaling. There are an infinite number of such choices. Sti truly effective in combating noise and efficient in saving bandwidth and power, A’ in Sec, 8.9. ‘of other M-ary signaling schemes will be presented a little later, 'As in most system designs, there are always prices to pay for it to increase data rate is power. As M increases, the transmitted power also it to have the same noise immunity, the minimum separation between pulse p every possible gain. 7“ rn) « . “dey PAM signaling: [6] four RZ sync; fb) bawaband wanimision; fe} the Ary RZ ave dlegror” 6 that the Therefore, pulse amplitudes increase with M (Fig. 8.28), It can be shown a a tary PaPascs as M? (Prob, 87-6). Thus, t increase the rte of commeanication lbpapete foes PNT required increases as M?, Because the transmission bandwidth depends only sot oe Fo olse amplitudes, the bandwidth is independent of Mf. We will use the folowing 08 0 iustate this point ais i a ste age pomple 8:5 = RI pipiens oat oR o se the PSD of the quaternary (4-ary) baseband signaling in Fig. 8.28 when the message bits # andl ne el sion ay line code has four distinct symbols corresponding to the four different combinations of (wo Sie ‘its. One such mapping is —3 message bits 00 —1 message bits 01 +1 message bits 10 +3. message bits 11 (3.56) ‘Therefore, all four values of ay are equally likely, each with a chance of | in 4. Recall that a Pe fas Nido ‘Wahis the summation, 1/4 of the ay will be +1, and £3. Thus, ay pa Nm Nea P+ FCW? + 7+? + F4sF | = s nite other hand, for n > 0, we need to determine 1 R= WL atte insy b “sti sscrage value, we build & table With ll the possible Vals ofthe enone552 PRINCIPLES OF DIGITAL DATA TRANSMISSION From the foregoing table listing all the possible products of apdi-+n. We see that eacty tion auax 49 can take on any of the following six values 41, £3, and +9. First, 1, 4gy (1 in 8). On the other hand, +3 are equally likely (1 in 4). Thus, we can show that " sce IN N N N N N Ry = en Ne KO j on alin, [4c N+ G4 + GC N+ gent FC +765] 5 5 = sin=Zune Asa result, Sf) = qT, ___ Thus, the M-ary line code generates the same PSD shape as binary polar signaling. is that it utilizes five times the original signal power. Although most terrestrial digital telephone network uses binary encoding, the subscriber ip the ISDN network uses the quaternary code, 2B1Q, similar to Fig. 8.28a. It uses NRRZ pray 160 kbit/s of data at a baud rate (pulse rate) of 80 kbivs. Of the various line codes standards committee, 2B1Q provided the greatest baud rate reduction in the noisy and: cable plant environment. Pulse Shaping and Eye Diagrams in PAM: In this case, we can use the Nyquist these pulses have zero ISI at the sample points, and, therefore, their amplitudes can be by sampling at the pulse centers. We can also use the controlled ISI (partial-response. signaling.” Eye diagrams can also be generated for M-ary PAM by using the same method used. ulations. Because of multilevel signaling, the cye diagram should have M levels at the optim instants even when ISI is zero. Here we generate the practical eye diagram example for a1 signal that uses the same cosine roll-off pulse with roll-off factor.r = 0.5 that was used in th of Fig. 8.27. The corresponding eye diagrams with time offsets of T+/2 and 0 are given in] : respectively. Once again, no ISI is observed at the sampling instants. The eye diagrams ¢ equally separated signal values without ISI at the optimum sampling points. 8.8 DIGITAL CARRIER SYSTEMS ‘Thus far, we have discussed baseband digital systems, where signals are transmitted d in frequency. Because baseband signals have sizable power at low frequencies, tf mission over a pair of wires and coaxial cables, Much of the modern comm However, baseband signals cannot be transmitted over a radio link or satellites tate impractically Jarge antennas to efficiently radiate the low-frequency sp these applications, the signal spectrum must be shifted to a high-freq541 8.5 Digital Receivers and Regenerative Repeaters qualizers fi b 45 i _renvated and distorted by the transmission medium, The attenuation can be compensate of et a ereas the distortion is compensated by the equalizer. Channel distortion is in Oe train. life caused by an attenuation of certain critical frequency components ofthe data pulse te info i ulizer should havea frequency characteristic thats the inverse of that of the transmiss4 16 il restore the critical frequency components and eliminate pulse dispersion. ua tie iit vpces the received channel noise by boosting its components at these critical frequencies: Me enh renomenon is known as noise amplification. only needs E gsinble Pi jgnals, however, complete equalization is really not necessary, because a detector ae oxi ely simple deisions—such as whether the pulse is positive or negative (or whether the Pil nade sit ent), Therefore, considerable pulse dispersion can be tolerated. Pulse dispersion results fp 5 sequent increas in eror detection, Noise increase resulting from the equalizer (which boos iiaod he Or ncies) also increases the detection error probability. For this reason, design of an opti f . r eke etch ave an inevitable compromise between reducing ISI and reducing the channel noise. A iui ‘ aie cof the equalization characteristics is a central feature in all well-designed digital communica sic ee ose corcing Equalizer . F te nat necessary to eliminate or minimize ISI (interference) with neighboring pulses for all t. All jisreally ‘needed isto eliminate or minimize interference with neighboring pulses at their respective serie a only. This is because the receiver decision is based on sample values only. This kind of (relaxed) ‘evjation can be accomplished by equalizers using the transversal filter structure encountered earlier. trie watonal filters, transversal filter equalizers are easily adjustable to compensate against different jumels or even slowly time-varying channels. The design goal is to force the equalizer output pulse to have jul values at the sampling (decision-making) instants. In other words, the equalizer output pulses satisfy the Nyquist or the controlled ISI criterion. The time delay T between successive taps is chosen to be Tj, the ineral between pulses Th begin, set the tap gains co = 1 and cx = 0 for all other values of k in the transversal filter in a. Thus the output of the filter will be the same as the input delayed by NT. For a single pulse p,(¢) 2b) at the input of the transversal filter with the tap setting just given, the filter output po(r) will be ly pi! — NT), that is, p-(t) delayed by NT). This delay has no practical effect on our communication system and is not relevant to our discussion. Hence, for convenience, we shall ignore this delay. This means tap) in Fig. 8.22b also represents the filter output po(t) for this tap setting (co = 1 and cy = 0, kK #0). We require that the output pulse p(t) satisfy the Nyquist’s criterion or the controlled ISI criterion, as the cnpr(t nT) nN (8.50) The samp. =m Samples of.po(t) at t= kT are$42 Princ; IPLES OF DIGITAL DATA TRANSMISSION Figure 8.22. Zeroforcing equalizer analysis. Nv Po(kT) = Yo cnbr(kTo —nTo), k= 0, £1, £2; t3ynu a) to denote p,(KTb) and polk] to denote po(kTp), Eq, (851 By using a more convenient notation p(k] expressed as N polkl = >> cnpelk—nh n=-N , 1, +2, 23,005 jres the samples po{k] = 0 for k # 0, and pofk] = 1 for k = 0. Up faneous equations in terms of if we specify the values of pol Nyquist’s first criterion requ these values in Eq, (8.51b), we obtain a set of infinite simult Clearly, itis not possible to solve all the equations. However, points as 1 k=O POM = Vo pact], ct QeoubN 3e at SE then a unique solution exists. This assures that a pulse will have zero interference and WV succeeding pulses. Because the pulse amplitude decays rapidly, i pulse is not significant for N > 2, in general, Substitution of the condition (8.52) into F4: set of 2N+1 ‘simultaneous equations for 2N + | variables, These 2N + 1 equations | _—ateiw farm of543 Pp 8.5 Digital Receivers and Regenerative Repeater? 3 con , c-N+1 0] ppt) = Pk) PA-2N +1) pl -2N1 prlll prl0} + Pel-2N +2} pl-2N + 1) eat (6.53) 0 * = i co = Ci o| | p2N- 1 pA2N = 21 *. p,f0) pAl-1) ; PON} P2N=1) pl) p/10) : Py diagonal expression, the (2N ++ 1) x (2N+ 1) matrix P, has identical entries along all the oder ; a 1 ibing convo! known as the Toeplitz matrix and is commonly encountered in describing co" ice proper- s fully determined by its first row and first column. It has some nice PF oa -ompact exPrS ostinato 5 Sg, A Toeplitz mat 7 : fen iP: < simpler algorithms for computing its inverse (see, e.g., the method by Trench’). ‘The tap iam ane .d by solving this set of equations by taking the inverse of the matrix P, pe obta af c=P;'po gomple 8.4 geneived pulse p+(®) in Fig, 8.226, let Fe prl0)=1 Pll = -03 p,[2] =0.1 pil = 02 — pl-2) = 0.05 paigna three-tap (N = 1) equalizer. sulin sasituting the foregoing values in Eq. (8.53), we obtain 0 1 -0.2 005 cnt 1 }=] -03 1 -02 co. (8.54) 0 01-03 1 ce Sslaion of this set yields c_; =0.210, co = 1.13, and cy =0,318. This tap setting assures us that po[0] = 1 sndpo(-1] = pol] =. The ideal output po(2) is sketched in Fig. 8.22c. $$$ ett li ints es Kote that the equalizer determined from Eq. (8.53) can guarantee only the zero ISI condition of 44852) In other words, ISLis zero only for k = 0, -t1, ..., -EN. In fact, for k outside this range, it is quite mare” tt the samples pp(KTs) # 0, indicating some residual ISI. For instance, consider the equalizer ‘slm in Example 8.4. The samples of the equalized pulse have zero ISI for k = 1. 0. 1, However, from Nv Polk] = SO enprlk =n)