4.6 Correlative-Level Coding 267 that the bit duration of the resulting time-division multiplexed signal (including a framing bit) is 7 Ty = 0.647 us Assuming the use of an ideal Nyquist channel, it follows that the minimum transmission bandwidth By of the T1 system is (for a@ = 0) 1 7: Br= We oe = 772 kite However, a more realistic value for the necessary transmission bandwidth is obtained by using a full-cosine rolloff characteristic with a = 1. In this case, we find that 1 Br= WI + a) = 2W = = = 1.544 MHz 4 Ll 4.6 Correlative-Level Coding Thus far we have treated intersymbol interference as an undesirable phenomenon that produces a degradation in system performance. Indeed, its very name connotes a nuisance effect. Nevertheless, by adding intersymbol interference to the transmitted signal in a con- trolled manner, itis possible to achieve a signaling rate equal to the Nyquist rate of 2W symbols per second in a channel of bandwidth W Hertz. Such schemes are called correl- ative-level coding or partial-response signaling schemes.> The design of these schemes is based on the following premise: Since intersymbol interference introduced into the trans. mitted signal is known, its effect can be interpreted at the receiver in a deterministic way. Thus correlative-level coding may be regarded as a practical method of achieving the theoretical maximum signaling rate of 2W symbols per second in a bandwidth of W Hertz, as postulated by Nyquist, using realizable and perturbation-tolerant filters, = DUOBINARY SIGNALING The basic idea of correlative-level coding will now be illustrated by considering the specific example of duobinary signaling, where “duo” implies doubling of the transmission ca, Pacity of a straight binary system. This particular form of correlative-level coding is also called class I partial response. Consider a binary input sequence (b,) consisting of uncorrelated binary symbols 1 and 0, each having duration T,. As before, this sequence is applied to a pulse-amplitade modulator producing a two-level sequence of short pulses (approximating a unit impulse), whose amplitude a, is defined by {nt if symbol by is 1 ay = =1 if symbol by is 0 tere When this sequence is applied to a duobinary encoder, it is converted into a three- level output, namely, ~2, 0, and +2. To produce this transformation, we may use the Scheme shown in Figure 4.11. The two-level sequence {a4} is first passed through a simple filter involving a single delay element and summer, For every unit impulse applied to the268 CHAPTER 4 ® BASEBAND PULSE TRANSMISSION Input Taeal output swoivel channe! sequence seares Higa) | Samet fed timo r= ky, iter 907) Frcune 4.1L Duobinary signaling scheme. japut of this filter, we get two unit impulses spaced T, seconds apart at the filter outpu, “We may therefore express the duobinary coder output cx as the sum of the present inpat pulse a, and its previous value a,—1, as shown by oh = a + ae (4.66) One of the effects of the transformation described by Equation (4.66) is to change the input sequence {a,) of uncorrelated two-level pulses into a sequence {c,) of correlated three level pulses. This correlation between the adjacent pulses may be viewed as introducing intersymbol interference into the transmitted signal in an artificial manner. However, the intersymbol interference so introduced is under the designe:’s control, which is the bass of correlative coding. "An ideal delay element, producing a delay of T,, seconds, has the frequency response exp(jd7fTr), 60 that the frequency response of the simple delay-line filter in Figure 4.11 js 1 + exp(—j2afT,). Hence, the overall frequency response of this filter connected in cascade with an ideal Nyquist channel is Auf) = Aryquinl f+ exp(—j2afT)] = Hyyauia(flexp(infTe) + exp(—j7fTe)] exp(—j#fTe) (4.67) = LHriyguinlf) cos fT s) exp(imfTo) where the subscript I in H,(f) indicates the pertinent class of partial response, For an ideal Nyquist channel of bandwidth W = 1/2T,, we have (ignoring the scaling factor T;) 1, [fl S127. Hysquif) = ‘ 4.68 stl” [O, hes wa Thus the overall frequency response of the dnobinary signaling scheme has the form of half-cycle cosine function, as shown by Hf) = {i cos(mfTs) exp(-j7fTs), | f| = 12Ts (4) 0, otherwise for which the magnitude response and phase response are as shown in Figures 4.124 and 4.12b, respectively. An advantage of this frequency response is that it can be easily F proximated, in practice, by virtue of the fact that there is continuity at the band edges: From the first line in Equation (4.67) and the definition of Hyyquin(f) in Equal (4.68), we find that the impulse response corresponding to the frequency response Al4.6 Correlative-Level Coding 269 baal am 20 Rm @ Fiure 4.12 Frequency response of the duobinary conversion filter. (a) Magnitude response. (b) Phase response. consists of two sine (Nyquist) pulses that are time-displaced by T,, seconds with respect to each other, as shown by (except for a scaling factor) sin(at/T,,) a sin{(t — T,)/T;] wT, ae TT, sin(mt/T,) __ sin(at/T,) atlT, a(t — T,\/T, _ Th sin(art/T) © atl, = 2) The impulse response /(f) is plotted in Figure 4.13, where we see that it has only nwo distinguishable values at the sampling instants. The form of b,(t) shown here explains why we also refer to this type of correlative coding as partial-response signaling, The response to an input pulse is spread over more than one signaling interval; stated in another way, the response in any signaling interval is “partial.” Note also that the tails of hy(t) decay as 1|t|?, which is a faster rate of decay than the 1/|¢| encountered in the ideal Nyquist channel. The original two-level sequence {a,) may be detected from the duobinary-coded sequence {c4} by invoking the use of Equation (4.66). Specifically, let 4, represent the estimate of the original pulse a, as conceived by the receiver at time t = kt,. Then, sub- tracting the previous estimate dy, from c,, we get dg = ey — dey (4.71) Ait) = (4.70) Itis apparent that if cx is received without error and if also the previous estimate 4, at time t= (k ~ 1)T, corresponds to a correct decision, then the current estimate d, will be Aw 1.09, — aT, TF 2h, “Ti, 0 th Ficune 4.13 Impulse response of the duobinary conversion filter.270 CHarTer 4 © BASEBAND PULSE TRANSMISSION correct too. The technique of using a stored estimate of the previous symbol is called decision feedback. ‘We observe that the detection procedure just described is essentially an inverse of the operation of the simple delay-line filter at the transmitter. However, a major drawback of this detection procedure is that once errors are made, they tend to propagate through the output because a decision on the current input a, depends on the correctness of the decision made on the previous input 441. A practical means of avoiding the error-propagation phenomenon is to use precoding before the duobinary coding, as shown in Figure 4.14. The precoding operation performed on the binary data sequence {B,) converts it into another binary sequence {dy} defined by d= bh Ody (4.72, where the symbol © denotes modulo-two addition of the binary digits b, and dy_1. This addition is equivalent to a two-input EXCLUSIVE OR operation, which is performed as follows: He Rae 1 if either symbol by or symbol d,_1 (but not both) is 1 = symbol 0 otherwise (4.73) The precoded binary sequence {d,} is applied to a pulse-amplitude modulator, producing a corresponding two-level sequence of short pulses {ay}, where a, = £1 as before. This sequence of short pulses is next applied to the duobinary coder, thereby producing the sequence {c,} that is related to {a,} as follows: Ce = ag + Aye (4.74) Note that unlike the linear operation of duobinary coding, the precoding, described by Equation (4.72) is a nonlinear operation. The combined use of Equations (4.72) and (4.74) yields pee f° if data symbol by is 1 a +2 if data symbol b, is 0 (425) which is illustrated in Example 4.3. From Equation (4.75) we deduce the following decision. rule for detecting the original binary sequence (b,} from {cx}: If|c| <1, say symbol by is 1 (4.76) If |e) > 1, say symbol by is 0 Input an outa | “tear” [ooo tee sempieat tp = KT Delay Precoder Fire 4.14 A precoded duobinary scheme; details of the duobinary coder are given in Fige 4.11.46 Correlative-Level Coding 271 teh thee! > Say by Lif lel <1 sl etter Decison [> Sayd=0it Joy > Threshold =1 FIGURE 4.15 Detector for recovering original binary sequerice from the precoded duobinary coder output. When |¢,| = 1, the receiver simply makes a random guess in favor of symbol 1 or 0. According to this decision rule, the detector consists of a rectifier, the ‘output of which is compared in a decision device to a threshold of 1. A block diagram of the detector is shown in Figure 4.15. A useful feature of this detector is that no knowledge of any input sample other than the present one is required, Hence, error propagation cannot occur in the detector of Figure 4.15, > Exampte 4.3 Duobinary Coding with Precoding Consider the binary data sequence 0010110, To proceed with the precoding of this sequence, which involves feeding the precoder output back to the input, we add an extra bit to the precoder output. This extra bit is chosen arbitrarily to be 1. Hence, using Equation (4.73), we find that che sequence (4,} at the precoder output is as shown in row 2 of Table 4.1. The polar representation of the precoded sequence {d,) is shown in row 3 of Table 4.1. Finally, using Equation (4.74), we find that the duobinary coder output has the amplitude levels given in row 4 of Table 4.1. To detect the original binary sequence, we apply the decision rule of Equation (4.76), and so obtain the binary sequence given in row S of Table 4.1. This latter result shows that, in the absence of noise, the original binary sequence is detected correctly. < @ Mopirtep Duoginary SIGNALING In the duobinary signaling technique the frequency response Hf), and consequently the Power spectral density of the transmitted pulse, is nonzero at the origin. This is considered to be an undesirable feature in some applications, since many communications channels cannot transmit a DC component. We may correct for this deficiency by using the class IV partial response ox modified duobinary technique, which involves a correlation span of two binary digits, This special form of correlation is achieved by subtracting amplitude- modulated pulses spaced 27, seconds apart, as indicated in the block diagram of Figure [Taste 4.1 Illustrating Example 4.3 on duobinary coding Binary sequence {b,} o 0 1 0 14 14 9 Precoded sequence {d;} t+ 1 41 0 0 4 0 Tworlevel sequence {a,} LH Ft tad Duobinary coder output (ce) 420042000 -2 0 Oo -2 Binary sequence obtained by o o 1 O09 1 1 0 applying decision rule of Eq. (4.76)272 «9 Cuarren 4 & Basepanp Putse TRANSMISSION Input binary sequent te Precedee Mocitied duobinary conversion iter Hf) FIGURE 4.16 Modified duobinary signaling scheme. 4.16. The precoder involves a delay of 2T; seconds. The output of the modified duobinary conversion filter is related to the input two-level sequence {ay} at the pulse-amplitude mod. ulator output as follows: Ck = dk ~ Gea (4.77, Here, again, we find that a three-level signal is generated. With a, = 1, we find that q takes on one of three values: +2, 0, and —2. The overall frequency response of the delay-line filter connected in cascade with an ideal Nyquist channel, as in Figure 4.16, is given by Fiv(f) = Hoyyquise(f)[1 — exp(-i4a Ts) Hysyqusilf )sin(27 fT) exp(—j27fTo) where the subscript IV in Hiy(f) indicates the pertinent class of partial response and Hyyguis(f) is as defined in Equation (4.68). We therefore have an overall frequency re sponse in the form of a half-cycle sine function, as shown by Hyif) = ‘2j sin(2nfT,) exp(—j2afTs)s | f| S$ V2T, 7 0, elsewhere ‘The corresponding magnitude response and phase response of the modified duobinary coder are shown in Figures 4.17a and 4.17b, respectively. A useful feature of the modifiel duobinary coder is the fact that its output has no DC component. Note also that this (4.78) (4.79) ang Lyy(P0 2 @ o FIGURE 4.17 Frequency response of the modified duobinary conversion filter. (a) Magnitude response. (b) Phase response.4.6 Correlative-Level Coding 273 second form of correlative-level coding exhibits the same continuity at the band edges as in duobinary signaling. From the first line of Equation (4.78) and the definition of Hxyyeuin(f) in Equation (4.68), we find that the impulse response of the modified duobinary coder consists of two sinc (Nyquist) pulses that are time-displaced by 2T,, seconds with respect to each other, as shown by (except for a scaling factor) sin(nt/T,) _ sin{r(t ~ 2T,VT,)] tT, a(t — 2T,VT, sin(at/T,) ___sin(t/Ts) Ayy(t) = mT, = mit — 27, NTs, Maing _ 27} sin(at/T,) wt2T, —t) This impulse response is plotted in Figure 4.18, which shows that it has three distinguish- able levels at the sampling instants. Note also that, as with duobinary signaling, the tails of by(t) for the modified duobinary signaling decay as 1/|t|?. To eliminate the possibility of error propagation in the modified duobinary system, we use a precoding procedure similar to that used for the duobinary case. Specifically, prior to the generation of the modified duobinary signal, a modulo-two logical addition is used on signals 27), seconds apart, as shown by (see the front end of Figure 4.16) dy = be Oden symbol 1 if either symbol by or symbol dz (but not both) is 1 (4.81) symbol 0 otherwise where {b4] is the incoming binary data sequence and {d,} is the sequence at the precoder output. The precoded sequence {d,} thus produced is then applied to a pulse-amplitude modulator and then to the modified duobinary conversion filter. In Figure 4.16, the output digit ¢, equals —2, 0, or +2, assuming that the pulse- amplitude modulator uses a polar representation for the precoded sequence {d,}. Also we find that the detected digit 6, at the receiver output may be extracted from ¢, by disre- garding the polarity of cx. Specifically, we may formulate the following decision rule: Klee] > 1, say symbol by is 1 If|e| <1, say symbol by is 0 (4.82) yl FIGURE 4,18 Impulse response of the modified duobinary conversion filter.274 © Cuarren4 ® Bastpanp PuLse TRANSMISSION ‘When | ci | = 1, the receiver makes a random guess in favor of symbol 1 or 0. As with the duobinary signaling, we may note the following: > In the absence of channel noise, the detected binary sequence {6,} is exactly the same as the original binary sequence (b,) at the transmitter input. > The use of Equation (4.81) requires the addition of two extra bits to the precoded sequence (a;}. The composition of the decoded sequence {b,) using Equation (4,82) is invariant to the selection made for these two bits. § GENERALIZED ForM OF CORRELATIVE-LEVEL CODING (PARTIAL-RESPONSE SIGNALING) ‘The duobinary and modified duobinary techniques have correlation spans of 1 binary digit and 2 binary digits, respectively. It is a straightforward matter to generalize these two techniques to other schemes, which are known collectively as correlative-level coding o: partial-response signaling schemes. This generalization is shown in Figure 4.19, where Hyyauin(f) is defined in Equation (4.68). It involves the use of a tapped-delay-line filter with tap-weights wy, wa, +, wy-1. Specifically, different classes of partial-response sig, Input deal Outpt tno channel, seg Spe Malte sequence Hryuis(f) [Sample at Squeret tad rae, oe Ficure 4.19 Gencralized correlative coding scheme.4.7 Baseband M-ary PAM Transmission 275. TABLE 4.2. Different classes of partial-response signaling schemes referring to Figure 4.19 Type of Class = N wo ty agg Comments I 2 1 1 Duobinary coding i 3 nee 1 ml 3 2 1 -1 Vv 3 1 00-1 Modified duobinary coding v 5-1 0 0 -1 naling schemes may be achieved by using a weighted linear combination of N ideal Nyquist (sinc) pulses, as shown by os : we) = > we sine 2 *) (4.83) 2 t ‘An appropriate choice of the tap-weights in Equation (4.83) results in a variety of spectral shapes designed to suit individual applications. Table 4.2 presents the specific details of five different classes of partial-response signaling schemes. For example, in the duobinary case (class I partial response), we have Wy = +1 w= +1 and w, = 0 for # = 2. In the modified duobinary case (class IV partial response), we have +1 0 and w, = 0 for n 2 3. The useful characteristics of partial-response signaling schemes may now be sum- marized as follows: » Binary data transmission over a physical baseband channel can be accomplished at a rate close to the Nyquist rate, using realizable filters with gradual cutoff characteristics. > Different spectral shapes can be produced, appropriate for the application at hand. However, these desirable characteristics are achieved at a price: A larger signal-to-noise ratio is required to yield the same average probability of symbol error in the presence of noise as in the corresponding binary PAM systems because of an increase in the number of signal levels used. L4.7_ Baseband M-ary PAM Transmission In the baseband binary PAM system of Figure 4.7, the pulse-amplitude modulator pro- duces binary pulses, that is, pulses with one of two possible amplitude levels. On the other