Chapter 4BASEBAND PULSE TRANSMISSION

• Matched filter for detecting a known signal in AWGN.

• Calculation of the BER due to the presence of channel noise.

• Intersymbol interference (ISI) arises when the channel is dispersive.

• Nyquist's criterion for distortionless baseband data transmission.

• Correlative coding (partial-response signaling) for combating ISI.

• Equalization of a dispersive baseband channel.

4.l Introduction

Baseband transmission of digital data requires the use of a low-pass channel with a bandwidth large enough to accommodate the essential frequency content of the data stream.

The channel is typically dispersive in that its frequency response

deviates from that of an ideal low-pass filter.

Each received pulse over dispersive channel is affected by adjacent pulses, thereby giving rise to intersymbol interference (ISI).

ISI is a major source of bit errors in the reconstructed data stream at the receiver.

Another bit errors source in baseband data transmission is the channel noise.

Optimum detection of data pulse involves the use of a matched filter.

4.2 Matched Filter

A receiver model shown in Figure 4.1 involves a linear time-

invariant filter of impulse response h(t).

The filter input x(t) consists of a pulse signal g(t) corrupted by

additive channel noise w(t), as shown by

x(t) = g(t) + w(t), 0 < t < T (4.1)

where T is an arbitrary observation interval.

The w(t) is the sample function of a white noise process

of zero mean and power spectral density No/2.

The function of the receiver is to detect the pulse signal g(t)

in an optimum manner, given the received signal x(t).

Since the filter is linear, the output y(t) may be expressed as

y(t) = g0(t) + n(t) (4.2)

where g0(t) and n(t) are the produced signal and noise components.

To have the output signal go(t) > the output noise n(t)

is to have the instantaneous power in go(t), measured at t = T,

be as large as possible compared with the average power of n(t).

This is equivalent to maximizing the peak pulse SNR,

h = | go(T)|2 / E[n2(t)] (4.3)

where | go(T)|2 is the instantaneous power in the output signal,

and E[n2(t)] is a measure of the average output noise power.

The requirement is to specify the impulse response h(t) of the filter

such that the output SNR in Equ. (4.3) is maximized.

Let G(f) denote the Fourier transform of the known

signal g(t), and H(f) denote the frequency response of

the filter.

The Fourier transform of the output signal go(t) is

equal to H(f)G(f), and go(t) is itself given by the inverse

Fourier transform

go(t) =∫H(f)G(f) exp(j2ft) df (4.4)

When the filter output is sampled at time t = T, we have

| go(T)|2 = |∫H(f)G(f) exp(j2fT) df |2 (4.5)

The power spectral density SN(f) of the output noise n(t) is equal to the psd of the input noise w(t) times the squared magnitude response |H(f)|2.

Since w(t) is white with power spectral No/2, it follows that

SN(f) = (No/2).|H(f)|2 (4.6)

The average power of the output noise n(t) is therefore

E[n2(t)] =∫SN(f) df = (No/2).∫|H(f)|2 df (4.7)

Substituting Eqs. (4.5) and (4.7) into (4.3), we may rewrite the peak pulse SNR as

|∫H(f)G(f) exp(j2fT) df |2 h = ------------------------------------ (4.8) (No/2).∫|H(f)|2 df

For a given G(f), the problem is to find the frequency response H(f) of the filter that makes h a maximum.

For two complex functions 1(x) and 2(x) in the real variable x,

∫|1(x)|2 dx < ∞ ∫|2(x)|2 dx < ∞

we may write

|∫1(x)2(x) dx |2 <∫|1(x)|2 dx∫|2(x)|2 dx (4.9)

The equality in (4.9) holds if, and only if, we have

1(x) = k2*(x) (4.10)

By invoking Schwarz's inequality (4.9), and setting 1(x) = H(f) and 2(x) = G(f).exp(jfT), the numerator in Equ. (4.8) may be rewritten as

|∫H(f)G(f).exp(j2fT) df |2 <∫|H(f)|2 df .∫|G(f)|2 df (4.11)

Using the relation of Equ. (4.11) in Equ. (4.8), we may redefine the peak pulse SNR as

h < (2/No).∫|G(f)|2 df (4.12)

This relation does not depend on the frequency response H(f) but only on the signal energy and the noise power spectral density. The pulse SNR will be a maximum when H(f) is chosen such that

hmax = (2/No).∫|G(f)|2 df (4.13)

To find the optimum value we use Equ. (4.10), which yields

Hopt(f) = kG*(f).exp(-j2fT) (4.14)

where G*(f) is the complex conjugate of the Fourier transform of the input signal g(t), and k is a scaling factor.

Except for the factor k exp(-j2fT), the frequency response of the optimum filter is the same as the complex conjugate of the Fourier transform of the input signal.

Take the inverse Fourier transform of Hopt(f) in Equ. (4.14) to obtain the impulse response of the optimum filter as

hopt(t) = k∫G*(f).exp[-j2f(T - t)] df (4.15)

Since G*(f) = G(-f), we may rewrite Equ. (4.15) as

hopt(t) = k∫G(-f).exp[-j2f(T - t)] df

= k∫G(f).exp[j2f(T - t)] df

= k g(T - t) (4.16)

Equ. (4.16) shows that the impulse response of the optimum filter is a time-reversed and delayed version of the input signal g(t), that is, it is "matched" to the input signal.

Figure 4.1Linear receiver.

PROPERTIES OF MATCHED FILTERS

A matched filter to a pulse signal g(t) of duration T is characterized by an impulse response that is a time-reversed and delayed version of the input g(t):

hopt(t) = k g(T - t)

The impulse response hopt(t) is uniquely defined, except for the delay T and the scaling factor k, by the waveform of the pulse signal g(t) to which the filter is matched.

The matched filter is also characterized by a frequency response that is the complex conjugate of the Fourier transform of the input g(t):

Hopt(f) = kG*(f).exp(-j2fT)

The peak pulse SNR of a matched filter depends only on the ratio of the signal energy to the power spectral density of the white noise at the filter input.

Consider a filter matched to a known signal g(t), the Fourier Transform of the resulting matched filter output go(t) is

G0(f) = Hopt(f) G(f) = kG*(f).G(f).exp(-j2fT) = k.|G(f)|2.exp(-j2fT) (4.17)

Using Equ. (4.17) in the inverse Fourier transform, the matched filter output at time t = T is

go(t) = ∫G0(f).exp(j2fT) df = k∫|G(f)|2 df

By Rayleigh's energy theorem, the integral of the squared magnitude spectrum of a pulse signal with respect to frequency is equal to the signal energy:

E = ∫|g(t)|2 dt = ∫|G(f)|2 df Hence

go(t) = kE (4.18)

Substituting Equ. (4.14) into (4.7), the average output noise power is

E[n2(t)] = (k2No/2)∫|G(f)|2 df

= k2NoE/2 (4.19)

The peak pulse SNR has the maximum value

hmax = (kE)2/(k2NoE/2) = 2E/No. (4.20)

From Equ. (4.20) we see that the dependence on the waveform of the input g(t) has been removed by the matched filter.

In evaluating a matched-filter receiver to combat additive white noise, we find that all signals that have the same energy are equally effective.

The signal energy E is in joules and the noise spectral density No/2 is in watts per Hertz, so that the ratio 2E/No is dimensionless.

We refer to E/No as the signal energy-to-noise spectral density ratio.

EXAMPLE 4.1 Matched Filter for Rectangular Pulse

Consider a rectangular pulse g(t) of amplitude A and duration T, as shown in Figure 4.2a. The impulse response h(t) of the matched filter will have exactly the same waveform as the signal itself.

The output signal go(t) of the matched filter produced

in response to the input signal g(t) has a triangular waveform, as shown in Figure 4.2b.

The maximum value of the output signal go(t) is equal to kA2T, which is the energy of the input signal g(t) scaled by the factor k, this maximum value occurs at t = T, as indicated in Figure 4.2b.

In the integrate-and-dump circuit shown in Figure 4.3, the integrator computes the area under the rectangular

pulse, and samples the resulting output at t = T, the integrator then is restored to its initial condition.

Figure 4.2c shows the output of the integrate-and- dump circuit for the rectangular pulse of Figure 4.2a.

For 0 < t < T, the output of this circuit has the same

waveform as that appearing at the output of the matched filter.

Figure 4.2(a) Rectangular

pulse. (b) Matched filter

output. (c) Integrator

output.

Figure 4.3Integrate-and-dump circuit.

4.3 Error Rate Due to Noise

The channel noise is modeled as AWGN w(t) of zero mean and power spectral density No/2.

In the signaling interval 0 < t < Tb, the received signal is

+A + w(t), symbol 1 was sentx(t) = { (4.21)

-A + w(t), symbol 0 was sent

where Tb is the bit duration, and A is the transmitted pulse amplitude.

Given the noisy signal x(t), the receiver is required to make a decision in each signaling interval as to whether the transmitted symbol is a 1 or a 0.

The receiver shown in Figure 4.4 consists of a matched filter followed by a sampler, and a decision device.

The filter is matched to a rectangular pulse of amplitude A and

duration Tb, exploiting the bit-riming information available to the receiver.

The resulting matched filter output is sampled at the end of each

signaling interval. The presence of channel noise w(t) adds randomness

to the matched filter output.

Let y denote the sample value at the end of a signaling interval.

The sample value y is compared to a preset threshold A in the decision device.

If threshold A is exceeded, the receiver makes a decision in favor

of symbol 1; if not, a decision is made in favor of symbol 0.

Figure 4.4Receiver for baseband transmission of binary-encoded PCM wave using

polar NRZ signaling.

Two possible kinds of error have to be considered:

1.Symbol 1 is chosen when a 0 was actually transmitted; we refer to this error as an error of the first kind.2.Symbol 0 is chosen when a 1 was actually transmitted; we refer to this error as an error of the second kind.

Suppose that symbol 0 was sent. According to Equ. (4.21), the received signal is

x(t) = -A + w(t), 0 < t < Tb (4.22)

The matched filter output, sampled at time t = Tb, is given by

y = ∫x(t) dt = -A + (1/Tb)∫w(t) dt (4.23)

which represents the sample value of a random variable Y.

By virtue of the fact that the noise w(t) is white and Gaussian, we may characterize the random variable Y as follows: ● The random variable Y is Gaussian distributed with a mean of -A. ● The variance of the random variable Y is

Y2 = E[(Y + A)2]

= (1/Tb2) E[∫∫w(t) w(u) dt du]

= (1/Tb2)∫∫E[w(t) w(u)] dt du

= (1/Tb2)∫∫Rw(t, u) dt du (4.24)

where Rw(t, u) is the autocorrelation function of the white noise w(t).

Since w(t) is white with power spectral density No/2, we have

Rw(t, u) = (No/2)(t – u) (4.25)

where (t - u) is a time-shifted delta function.

Substituting Equ. (4.25) into (4.24) yields

Y2 = (1/Tb

2)∫∫(No/2)d(t – u) dt du

= No/2Tb (4.26)

where we have used the sifting property of the delta function and the fact that its area is unity.

The conditional probability density function of the random variable Y, given that symbol 0 was sent, is therefore

fY(y|0) = [1/(No/Tb)1/2] exp[-(y+A)2/(No/Tb)] (4.27)

This function is plotted in Figure 4.5a.

Let p10 denote the conditional probability of error, given that symbol 0 was sent. This probability is defined by the shaded area under

the curve of fY(y|0) from the threshold l to infinity, which corresponds to the range of values assumed by y for a decision in favor of symbol 1.

In the absence of noise, the output y sampled at t = Tb is equal to -A. When noise is present, error is made if y assumes a value > . This error probability, conditional on sending symbol 0, is defined by

p10 = P(y > | symbol 0 was sent)

=∫ fY(y|0) dy

= [1/(No/Tb)1/2].∫ exp[-(y+A)2/(No/Tb)] dy (4.28)

Now introducing the so-called complementary error function

erfc(u) = (2/1/2).∫u exp(-z2) dz (4.29)

which is closely related to the Gaussian distribution. For large positive values of u, the upper bound on the erfc function is

erfc(u) < exp(-u2)/(1/2 u) (4.30)

To reformulate p10 in terms of the erfc function, we define a new variable z = (y+A)/(No/Tb)1/2 to rewrite Equ. (4.28) in the compact form

p10 = (1/1/2).∫(A+)/(No/Tb)1/2 exp(-z2) dz = (1/2).erfc[(A+)/(No/Tb)1/2] (4.31)

Figure 4.5Noise analysis of PCM system. (a)

Probability density function of random variable Y at matched filter output when 0

is transmitted. (b) Probability density function of Y when 1 is transmitted.

When symbol 1 was transmitted, the Gaussian random variable Y with sample value y of the matched filter output has mean +A and variance No/2Tb. The conditional pdf of Y, given that symbol 1 was sent, is

fY(y|1) = [1/(No/Tb)1/2].exp[-(y-A)2/(No/Tb)] (4.32)

which is plotted in Figure 4.5b.

Let p01 denote the conditional error probability, given that symbol 1 was sent. This probability is defined by the shaded area under the curve of fY(y|1) extending from -∞ to the threshold l, corresponding to the range of values assumed by y for a decision in favor of symbol 0.

In the absence of noise, the matched filter output y sampled at

t = Tb is equal to +A. When noise is present, error is made if y assumes a value < l. This error probability, conditional on sending symbol 1, is defined by

p01 = P(y < l | symbol 1 was sent)

=∫ fY(y|1) dy

= [1/(No/Tb)1/2].∫ exp[-(y-A)2/(No/Tb)] dy (4.33)

To express p01 in terms of the erfc function, we define a newvariable z = (A-y)/(No/Tb)1/2 and reformulate Equ. (4.33) in thecompact form

p01 = (1/1/2).∫(A-)/(No/Tb)1/2 exp(-z2) dz

= (1/2).erfc[(A-)/(No/Tb)1/2] (4.34)

Let p0 and p1 denote the a priori probabilities of transmitting symbols 0 and 1. The average probability of symbol error Pe in the receiver is given by

Pe = p0 p10 + p1 p01 = (p0/2).erfc[(A+)/(No/Tb)1/2] + (p1/2).erfc[(A-)/(No/Tb)1/2]. (4.35)

Note that Pe is a function of the threshold l. For equiprobablesymbols 1 and 0, we have p0 = p1 = 1/2 and the optimumthreshold is lopt = 0.

The threshold should be at the midpoint between pulseheights -A and +A representing the two symbols 0 and 1.

For this special case we also have p01 = p10, and the transmission is in a binary symmetric channel.

For binary symmetric channel (BSC), the conditional error probabilities p01 and p10 are equal. The average probability of symbol error in Equ. (4.35) reduces to

Pe = (1/2).erfc[A/(No/Tb)1/2] (4.38)

Now the transmitted signal energy per bit is defined by

Eb = A2Tb (4.39)

Accordingly, we can formulate the average probability of symbol error for the receiver in Figure 4.4 as

Pe = (1/2).erfc[(Eb/No)1/2] (4.40)

which shows that the average probability of symbol error in a BSC depends solely on Eb/No, the ratio of the transmitted signal energy per bit to the noise spectral density.

Using the upper bound of Equ. (4.30) on the erfc, the averageprobability of symbol error is bounded as

Pe < exp(-Eb/No)/2(Eb/No)1/2 (4.41)

The PCM receiver of Figure 4.4 exhibits an exponential improvement in the average probability of symbol error with increase in Eb/No.

In Figure 4.6, the probability Pe is plotted versus the ratio Eb/No. We see that Pe decreases very rapidly as the ratio Eb/No is increased, so that eventually a very "small increase" in transmitted signal energy will make the reception of binary pulses almost error free.

Figure 4.6Probability of error in a PCM receiver.

4.4 Intersymbol interference (ISI)

Consider a baseband binary PAM system shown in Figure 4.7. The incoming {bk} consists of symbols 1 and 0, each of duration Tb. The PAM modulator modifies {bk} into sequence of short pulses, whose amplitude ak is represented in the polar form

+l if symbol bk is 1 ak = { (4.42)

-1 if symbol bk is 0

The sequence {ak} is applied to a transmit filter of impulse response g(t), producing the transmitted signal

s(t) = k ak g(t - kTb) (4.43)

The signal s(t) is modified as a result of transmission through the channel of impulse response h(t).

The channel adds random noise n(t) to the signal at thereceiver input.

The noisy signal x(t) passed through a receive filter of impulse response c(t).

The filter output y(t) is sampled synchronously with the transmitter. The amplitude of each sample is compared to a threshold l. If the threshold l is exceeded, a decision is made in favor ofsymbol 1. If the threshold l is not exceeded, a decision is made in favor of symbol 0.

The receive filter output is written as

y(t) = k ak p(t – kTb) + n(t) (4.44)

where m is a scaling factor, and the pulse p(t) is to be defined.

The scaled pulse p(t) is obtained by a double convolution involving the impulse response g(t) of the transmit filter, the impulse response h(t) of the channel, and the impulse response c(t) of the receive filter:

mp(t) = g(t) * h(t) * c(t) (4.45)

We assume that the pulse p(t) is normalized by setting p(0) = 1.

Figure 4.7Baseband binary data transmission s

ystem.

Since convolution in the t-domain is transformed into multiplication in the f-domain, we may change Equ. (4.45) into the equivalent form

P(f) = G(f)H(f)C(f) (4.47)

where P(f), G(f), H(f), and C(f) are the Fourier transforms of p(t), g(t),h(t), and c(t), respectively.

The term n(t) in Equ. (4.44) is the noise produced at the output of the receive filter due to the channel noise w(t). It is customary to model w(t) as a white Gaussian noise of zero mean.

The receive filter output y(t) is sampled at time ti = iTb,

y(ti) = k ak p[(i – k)Tb] + n(ti) = ai + m k≠i ak p[(i – k)Tb] + n(ti) (4.48)

The 1st term mai represents the contribution of the i-th transmitted bit. The 2nd term is the residual ISI effect of all othertransmitted bits on the decoding of the i-th bit. The last term n(ti) is the noise sample at time ti.

In the absence of both ISI and noise, we observe from Equ. (4.48) that

y(ti) = ai .

which shows that the i-th transmitted bit is decoded correctly.

In the design of transmit and receive filters, the objective is to minimize the effects of noise and ISI and to deliver the digital data totheir destination with the smallest error rate possible.

When the SNR is high, the system is largely limited by ISI rather than noise; in other words, we may ignore n(ti). We wish to determinethe pulse waveform p(t) for which the ISI is completely eliminated.

4.5 Nyquist's Criterion for Distortionless Baseband Binary Transmission

The data extraction involves sampling the output y(t) at time

t = iTb. The decoding requires the weighted pulse contribution

ak p[(i-k)Tb] for k = i be free from ISI due to the overlapping tails of all other weighted pulse contributions represented by k ≠ i.

This requires that we control the overall pulse p(t), as shown by

1, i = k

p(iTb - kTb) = { (4.49) 0, i ≠ k

where p(0) = 1, by normalization.

If p(t) satisfies the conditions of Equ. (4.49), the receiver output y(ti) given in Equ. (4.48) simplifies to

y(ti) = ai, for all i

which implies zero ISI.

It is informative to transform Equ. (4.49) into the frequency domain. Consider the sequence {p(nTb)}, where n = 0, +l, +2, …. Recall that sampling in the t-domain produces periodicity in the f-domain:

P(f) = Rb P(f - nRb) (4.50)

where Rb = 1/Tb is the bit rate in bits per second (b/s).

P(f) is the Fourier transform of an infinite periodic sequence of delta function of period Tb :

P(f) =∫ m [p(mTb) (t-mTb)] exp(-j2ft) dt (4.51) = ∫ p(0) (t) exp(-j2ft) dt = p(0) (4.52)

Since p(0) = 1, it follows from Eqs. (4.50) and (4.52) that the condition for zero ISI is satisfied if

n P(f - nRb) = Tb (4.53)

The Nyquist criterion for distortionless baseband transmission

in the absence of noise is therefore as follows:

The frequency function P(f) eliminates ISI for samples taken at intervals Tb provided that it satisfies Equ. (4.53).

Note that P(f) refers to the overall system, incorporating the

transmit filter, the channel, and the receive filter in accordance with

Equ. (4.47).

IDEAL NYQUIST CHANNEL

One way to satisfy Equ. (4.53) is to specify the frequency

function P(f) to be a rectangular function:

1/2W, -W < f < W

P(f) = { (4.54)

0, |f| > W

= (1/2W) rect(f/2W)

where rect(f) stands for a rectangular function of unit amplitude and centered on f = 0, and the overall system bandwidth W is defined by

W = Rb/2 = 1/2Tb (4.55)

According to the solution described by Eqs. (4.54) and (4.55),no frequencies of absolute value exceeding half the bit rate areneeded.

A signal waveform that produces zero ISI is defined by thesine function:

p(t) = sin(2Wt) / 2Wt = sinc(2Wt) (4.56)

The special bit rate Rb = 2W is called the Nyquist rate, and Wis the Nyquist bandwidth. The ideal baseband pulse transmissionsystem described by Equ. (4.54) in the f-domain, or Equ. (4.56) inthe t-domain, is the ideal Nyquist channel.

Figures 4.8a and 4.8b show plots of P(f) and p(t), respectively. In Figure 4.8a, the normalized form of the frequency function P(f)is plotted for positive and negative frequencies.

In Figure 4.8b, signaling intervals and the corresponding centered sampling instants are included.

The function p(t) can be regarded as the impulse response of an ideal LPF with passband magnitude response 1/2W andbandwidth W.

The function p(t) has its peak value at the origin and goesthrough zero at integer multiples of the bit duration Tb.

If the received waveform y(t) is sampled at the instants of t = 0, +Tb, +2Tb, …, then the pulses defined by p(t - iTb) with arbitraryamplitude m and index i = 0, +l, +2, …, will not interfere with eachother.

This condition is illustrated in Figure 4.9 for the binary sequence1011010.

Difficulties that make it an undesirable objective for system design:

1.It requires that the magnitude characteristic of P(f) be flat from -W to W, and zero elsewhere. This is physically unrealizable because of the abrupt transitions at the band edges +W. 2.The function p(t) decreases as l/|t| for large |t|, resulting in a slow rate of decay. This is also caused by the discontinuity of P(f) at +W. Accordingly, there is practically no margin of error in sampling times in the receiver.

To evaluate the effect of the timing error, consider the sample of

y(t) at time t = t, where t is the timing error. In the absence of noise, we have

y(ti) = k ak p(t – kTb)

= k ak sinc[2W(t – kTb)] (4.57)

Since 2WTb = 1, we may rewrite Equ. (4.57) as

y(t) = a0 sinc(2Wt) +

(/).sin(2Wt).k [(-1)kak/(2Wt – k)] (4.58)

The first term defines the desired symbol, whereas the remaining series

represents the ISI caused by the timing error t in sampling the output y(t). It is possible for this series to diverge, thereby causing erroneous decisions in the receiver.

Figure 4.8(a) Ideal magnitude response. (b) Ideal

basic pulse shape.

Figure 4.9A series of sinc pulses corresponding to the

sequence 1011010.

RAISED COSINE SPECTRUM

We may overcome the practical difficulties encountered with the deal Nyquist channel by extending the bandwidth from the minimum

W = Rb/2 to an adjustable value between W and 2W.

We specify the overall frequency response P(f) by retaining three terms of Equ. (4.53) and restrict the frequency band to [-W, W]:

P(f) + P(f-2W) + P(f+2W) = 1/2W, -W < f < W (4.59)

A particular form of P(f) that embodies many desirable features is provided by a raised cosine spectrum.

This frequency response consists of a flat portion and a rolloffportion that has a sinusoidal form:

1/2W, 0 < |f| < f1 P(f) = { (1/4W){1- sin[(|f|-W)/(2W-2f1)]} , f1 < |f| < 2W - f1 (4.60)

0, |f| > 2W - f1

The frequency parameter f1 and bandwidth W are related by

= 1 - f1/W (4.61)

The parameter a is the rolloff factor, it indicates the excessbandwidth over the ideal W. Specifically, the transmission bandwidth BT is

BT = 2W - f1 = W(l + )

The frequency response P(f), normalized by multiplying it by 2W, is plotted in Figure 4.10a for three values of a, namely, 0, 0.5, and 1.

For a = 0.5 or 1, the function P(f) cuts off gradually as compared with the ideal Nyquist channel (i.e., a = 0) and is easier to implement.

The function P(f) exhibits odd symmetry with respect to Nyquist bandwidth W, making it possible to satisfy the condition of Equ. (4.59).

Using the P(f) defined in Equ. (4.60), we obtain

p(t) = sinc(2Wt).[cos(2Wt)/(1-(4Wt)2)] (4.62)

which is plotted in Figure 4.10b for a = 0, 0.5, and 1.

The time response p(t) consists of the product of two factors: the factor sinc(2Wt) characterizing the ideal Nyquist channel and asecond factor that decreases as 1/|t|2 for large |t|.

The first factor ensures zero crossings of p(t) at the desired sampling instants of time t = iT with i an integer.

The second factor reduces the tails of the pulse considerably below that obtained from the ideal Nyquist channel, so that the transmission of binary pulse waves is relatively insensitive to sampling time errors.

In fact, for = 1 we have the most gradual rolloff in that theamplitudes of the oscillatory tails of p(t) are smallest.

Thus the amount of ISI resulting from timing error decreases as the rolloff factor a is increased from zero to unity.

The special case with = 1 (i.e., f1 = 0) is the full-cosine rolloff, for which the frequency response of Equ. (4.60) simplifies to

(1/4W)[1+ cos(f/2W)] , 0 < |f| < 2W P(f) = { (4.63)

0, |f| > 2W

Correspondingly, the time response p(t) simplifies to

p(t) = sinc(4Wt)/[1-(4Wt)2] (4.64)

This time response exhibits two interesting properties:

1.At t = +Tb/2 = +1/4W, we have p(t) = 0.5; that is, the pulse width

measured at half amplitude is exactly equal to the bit duration Tb.

2.There are zero crossings at t = +3Tb/2, +5Tb/2, … in addition to

the usual zero crossings at the sampling times t = +Tb, +2Tb, ….

These properties are useful in extracting a timing signal from the received signal for the synchronization.

The price paid for is a channel bandwidth double that required for the ideal Nyquist channel corresponding to = 0.

Figure 4.10Responses for differe

nt rolloff factors. (a) Frequency response. (b) Time respons

e.

Example 4.2 Bandwidth Requirements of T1 System

The signal format for the T1 carrier system is used to multiplex 24 independent voice inputs, based on an 8-bit PCM word. The bit duration of the resulting time-division multiplexed signal (including a framing bit) is

Tb = 0.647 ms

For an ideal Nyquist channel, the minimum transmission bandwidthBT of the Tl system is (for = 0)

BT = W = 1/2Tb = 772 kHz

A more realistic value for the necessary transmission bandwidth is obtained by using a full-cosine rolloff characteristic with = 1. In this case, we find that

BT = W(1+) = 2W = 1/Tb = 1.544 MHz