Escolar Documentos
Profissional Documentos
Cultura Documentos
PT
Digital Transmission
:TEFI OUTLINE
()
- rduction
()
.e \4odu larion
l, lr)
\t
\l Sanrpling
-:rul-to Quantization Noise Ratio
-'iLt-\'crsus Nonlinear pCM Codcs
Channc'l Noise
-:ing l e lhods
Compandino
Vocoder\
.i
(, l+ Ditlerentiai pCM
(r l. Pulse Tritnsrnission
() i(r Si-snal Pou.er in Binarv
Digital
Si-unals
.:TIVES
i:lle d i ! i I e I I rLt t.t i.t \ i otl
.: and dcrcribe the rd\ artage\
and disiLdYantnges 0f digital transmission
jtl! de\cribe pulse N idth nroduiatirrn_ pul,,e pl,,ition
,"n.f"f,ri,,n.
::
'rc !ll),1
Ll(.(rihr
t t t
. .- rihc sir:rrrl-r,r
'
.ne
utrttlturtling
trc
d i gita
273
t
I
I
I
I
I
T
T
Dc.crihc r ocodcrs
Erplain hou ro determine PCM line speed
De..ribc' d.ltr modulation PCM
De.eribc aJaptir.- delta modulation
Erpl"in
Erl.llin
eve patterns
6.1
INTRODUCTION
-r"
discrete levcl digital pulses. The original source information may be in digital forn: -:
could be anriog signnis that have been converted to digital pulses prior to trun\mi\s1.. -a
converted back to anakrg signals in thc rcccivcr. With digital transmission systems. ii:- ical facilitl. such as a pair of wires. coaxial cable. or an optical fiber cable. is requr:.- :
intcrconnect the larioLr\ points within the system. The pulses are contained in and p :rgate doun tire cable. Digital pulses cannot be plopagated through a wireless transn..- r
slslem. such as Earth's atmosphele or tiee space (vacuum).
,AT&T de\ eloped the ti rst digital transmission system fbr the purpose of canl ir: :-itrll) encoded irnalog signals. such as the human rrrice. over metallic wire cables be:.
telephone ollices. Todal'. digital transmission systems are used to cirry not only di-i
-r"
encoded r oice and Yideo signals but also digital source infonnation directly between putcrs and corlputer networks. Digital transmission systems use both metallic and (.:flber cables tbl their transnrission medium.
The plinrarl atlr antage ol digital tron\mission over analog transmission is noise ini!. ' Digital signals are inherently less susceptible than analog signals to interference cau:::
noise because rith di-uital signals it is not necessary to evaluate the precise amplitudl
quenc]. or phase 1o Isccrtain its logic condition. Instead. pulses are evaluated during.:
cise tinle inter\al. rnd a simple determination is made whether the pulse is above or r prescribed reter-ence Ievel.
Digital signals are also better suited than analog signals tbr processing and cor::
ing using a technique calJed nultiple.ring. Digital signal processing (DSP) is the pr..,:
ing of analos signals using digital methods and includes bandlimiting the signal \\ ii
tels. amplitude equalization. and phase shifiing. It is much simpler to store digital sL-thiin analog signals. and the trunsmission rate of digital signals can be easily chang.,
ldapt to dillerent envilonments and to intedhce with dilferent types of equipmenr.
In addition. digital transmission systems are more resistant to analog systems rr
djtive noise because they use signal regetlerution rather than signal amplification. \
274
Chapter 6
6-1-2
The transmission ofdigitally encoded analog signals requires signitlcantly more bandwidth
than simply transmitting the odginal analog signal. Bandwidth is one of the nrosl inlportanl
aspects of any comrnunications system because it is costly and limitccl.
Also. analog signals rnust be converted to digital pulses prior to transmission and
converted back to their original analog form at the receiver. thus neccssitating additional
encoding and decoding circuitry. ln addition. digital transmission requires precise time synchronization between the clocks in the transnritters and receivers. Finalll. digital nansmis
sion systems are incompatible with older analcl-r tlansrnis'ion r)
'tems.
PULSE MOOULATION
tnoduluti(rt consists essentially of sampling analog inlbrmation signals and then converting those samples into discrete pulses and transporting the pulses fiom a source to a destination over a physical transmission nrediunr. The tbur predominant n]ethods ol pulse
modulation include 2ii1se vidth ntodulutiort (PWM).7lr1se positiort nrodulation (PPM\,
pulse atnpliude tnodulation IPAM), and pulse co{le nodulatiort (PCM).
PWM is sometimes caled pulse duratiotr nrodulcttiotr (PDM) or pllse length ntotlu
lution (PLM). as the width (active po ion ofthe duty cycle) ofaconstant amplitude pulse
is varied proportional to the amplitude 0fthe analog signal at the time the signal is sampled.
PWM is shown in Figure 6- lc. As the figure sho'w,s. the amplitude of sanple I is lower than
the amplitude of sample 2. Thus. pulse I is narroucl than pulse 2. The maximum analog
signal amplitude produces the widest pulse. and the mininrum analog signal iunplitude produces the naro*,est puJse. Note. however. that all pulses have the same amplitude.
With PPM. the position of a constant-width pulse within a prescribed time slot is varied according to the amplitude of the sample of the unalog signal. PPM is shou,n in Figule
6-l d. As the figure shows. the higher the amplitude of the sample. the larther to the right the
PLllse
pulse is positioned uithin the prescribed time slot. The highest amplitude sample produces a
pulse to the far right. and the lorvest amplitude sanple produces a pulse to the tar left.
ll/)lJc irrrr.-lIn
:rance
e
With PAM. the amplitude of a constant width, constant-position pulse is varied according to the amplitude of the sample of the analog signal. PAM is shou,n in Figure 6- le.
where it can be seen that the amplitude of a pulse coincides with the arnplitude of the analog signal. PAM \\'aveforms rcsemble the original analog signal more than the waveforms
a]'-<: t!
rmplilu:-=
rg s) stems
rlitlcation.
'i
mission systems. PCM is by far the rnosl prelalent lbrm of pulse modulatiol
will be discussed in nrore dctail in subscquent sections of this chapter'.
code for transmission. Each code has the same number of bjts and requires the same length
of time lbr transmission. PCM is shou n in Figure 6 l l.
PAM is used as an intermediate fom of modulation with PSK. QAM, and PCM. although it is seldom uscd bl itself. PWM and PPM are used in special-purpo\e L-omDtunications systems mainly lbr the rnilitary but are seldom used tbr commercial digital trans
tt :-
aDLl.
consequently.
PCM
..Ju,-_e an e:_:nl
'lninritted.
AIex H. Reeves is credited with inverrting PCM in 93 7 * hilc rvorkin_e fbr AT&T al its Paris
laboratories. Although the merits of PCM were recognized early in its der cloprnent. it was
not until the mid- 1960s. with the advent of solid-state electlonics. that PC\l became prevaIent. In the United States today. PCM is the preferred methocl of cturmu nications u ithin the
public switched telephone network because with PCM it is easl to cornbine digitized roice
and digital data into a single. high-speed digital signal and propagate it or er either metallic or optical fiber cables.
1
::iSnais.
n analog !t:ra
:m to anotiret lE
ied and c..:al Transmission
?75
(a)
,"
(b)
l,Jt'A
(c)
PPH
(d)
?^^
(e)
8-bit word
L-
Fctl
(f)
FIGURE 6,
[e] PAM; [f)
Pu se
modulation: [a] analog signal; [b] sample pulse; [c] PWM; [d] PPM;
CM
PCM is the only digitally encoded modulation technique shown in Figure 6-1 r-' I
rI
commonly used for digital transmission. The tetm pulse code zodalaliort is someuh;:
codins
form
of
digitally
--:.
rather
a
misnomer. as it is nc,t really a type of modulation but
log signals. With PCM, the pulses are of fixed length and fixed amplitude PCM is a cr:"'"
; ''s
sy,stem where a pulse or lack of a pulse within a prescribed time slot represents either
: -'c
as
:
binary'
seldom
digital
but
I or a logic 0 condition. PwM, PPM. and PAM are
(bit)
does not represent a single binary digit
(on'- ! '7
Figure 6 2 shows a simplified block diagram of a single-channel, simplex
sig:' r
input
of
the
analog
only) PCM system. The bandpass tilter limits the tiequency
The
sample-and-ho':
Hz
the standard voice-band frequency range of 300 Hz to 3000
276
Chapter 6
*,J
$ .v' ,!
.d,
11/
Analog
1'r
,/
31;;x,
^\
,d'
PCNI-fransmittcr
J\*
Parallel
data
"""'*
SerialPCM
code
An!log
Output
signal
PC\,1 Rcce ir cr
: 3URE 6-2 Simplified block diagram of a single-channel, simplex PCM transmrsston system
''.
cuit periodically samples the analog input signal and converts those samples to a multilevel
PAM signal. The analog-to-digital converrer (ADC) convens the PAM samples to parallel
PCM codes, which are converted to serial binary data in the p.trullel-to-se rial cotNe rter and
then outputted onto the transmission line as serial digital pulses. The transmission line /.epeaters 'are placed at prescribed distances to regenerate the digital pulses.
In the receiver, the serial-to-parallel convefier colyerls serial pulses received from
the transmission line to parallel PCM codes. The digittrl-to-analog converter (DAC) converts the parallel PCM codes to multilevei PAM signals. The hold circuit is basically a loupass filter that converts the PAM signals back to its odginal analog form.
Figure 6-2 also shows several clock signals and sample pulses that will be explained
in later sections of this chapter An integrated circuit that performs the PCM encoding and
decoding functions is called a codec (coder/decoder), which is also described in a later section of this chapter.
ord
rn
1lt
LJI
in Figure 6-l
thr
>
somewhat I:
':
digitallY coding r:
,r,le. PCM is a bir''-'
lreients either a li;
ir))? is
SAMPLING
The function of a sampling circuit in a PCM transmitter is to periodically sample the continually changing analog input voltage and convert those samples to a series of constantamplitude pulses that can more easily be converted to binary PCM code. For the ADC to accurately convefi a voltage to a binary code. the yoltage must be relatively constant so that the
ADC can complete the conversion before the voltage level changes. If not. the ADC would
be continually attempting to follow the changes and may never stabilize on any PCM code.
:m binarY, as a
ei. simplex (one-s:'
inirlog inPut sign'
sttnPle'and-hold
I ::l Transmission
(a)
Input.
tb)
'
Sample
pulse
tc)
OutpLrt
Essentially, there arc two basic techniques used to perform the sampling funcr:tr
natural sampling and flat-top sampling. Nalrral .rumpling is shown in Figure 6-3. Nar--r
sampling is when tops of the sample pulses retain their natural shape during the sample *terval, making it difficult for an ADC to conveft the sample to a PCM code. With nar-a
sampling, the ftequency spectrum of the sampled output is different from that of an ia
sample. The amplitude of the frequency conponents produced tiom narrow finite-u:g
sample pulses decreases for the higher harmonics in a (sin r)/r manner. This alters the -:formation frequency spectrum requiring the use of frequency equalizers (compensatior -1ters) before recovery by a low-pass filter.
The most common method used for sampling voice signals in PCM systems is -:rtop sampling,which is accomplished in a sample-and-hold circrlt. The purpose ofa sarn:,e..
and-hold circuit is to periodically sample the continually changing analog input voltage::r
convert those samples to a series of constant-amplitude PAM voltage levels. With flat-:r
sampling, the input voltage is sampled with a narrow pulse and then held relativell ; ,rstant until the next sample is taken. Figure 6-4 shows flat-top sampling. As the fiS-=
shows, the sampling process alters the frequency spectrum and introduces an error cL.=
ttperture error, which is when the amplitude ofthe sampled signal changes during the :,=ple pulse time. This prevents the recoverl, circuit in the PCM receiver from exactly re5ducing the original analog signal voltage. The magnitude of error depends on how rr--.,:
the analog signal voltage changes while the sample is being taken and the width (durat:..r
of the sample pulse. Flartop sampling, however, introduces less aperture distortion ::a
natural sampling and can operate with a slower analogto-digital converter
Figure 6-5a shows the schematic diagram of a sample-and-hold circuit. The FET l-
simple analog switch. When tumed on, Q1 provides a low-impedance path to deposii
=
analog sample voltage across capacitor Cq. The time that Q1 is on is called the apefiuft
acquisition time. Essentially, C1 is the hold circuit. When Q, is off, C, does not have a c.:,:plete path to discharge through and, therefore, stores the sampled voltage. The storoge i-'E
ofthe capacitor is called the A./D conversior /lr?e because it is during this time that the -{:t:
converts the sample voltage to a PCM code. The acquisition time should be very short to :,1sure that a minimum change occurs in the analog signal while it is being deposited ac:-=i
Cr. If the input to the ADC is changing while it is performing the conversion. aprr-as a
274
Chapter 6
(u)
Input.
1g;
SamPle
pulse
(c)
OutPUi
iampling functior'
=igure 6-3.
Natur.
rcomPensation
3\I
systems
fij
lv t'r
'# [11
is/c:'
samPle'
'urpose of a
flat{o:
Inptd
nure distortion
th.
Sample AFturc
ener.
ol
aonvemion, aPerllr?
Outpst
FIGURE
6-5
279
distortion results. Thus, by having a short aperture time and keeping the input to the
-{ir:
relatively constant. the sample-and-hold circuit can reduce aperture distortion. Flat-top s;j:pling introduces less aperture distortion than natural sampling and requires a slower anal ;
to-digital converter
Figure 6-5b shows the input analog signal. the sampling pulse' and the wavef.=
developed across C r. lt is important that the output impedance of voltage follower Z, --c
the on resistance of Q1 be as small as possible. This ensures that the RC charging time ' :stant of the capacitor is kept very shon. allowing the capacitor to charge or discharge :--idly during the short acquisition time. The rapid drop in the capacitor Yoltage immedia::
following each sample pulse is due to the redistribution of the charge across C1. The in:-electrode capacitance between the gate and drain of the FET is placed in series witi: -when the FET is olf, thus acting as a capacitive voltage-divider network. Also. note =
gradual discharge across the capacitor during the conversion time. This is called droop ':ti
is caused by the capacitor discharging through its own leakage resistance and the input :pedance of voltage follower 2,. Therefore. it is important that the inpur impedance t" -and the leakage resistance of C1 be as high as possible Essentially, voltage follo*er' I
and Zr isolate the sample-and-hold circuit (Q1 and C1) from the input and output circL::'
Example 6-1
For the sample-and hold circuit shown in Figure 6 5a, determine the largest-value capacitor th;: '!
be used. Use an output impedance tbr Z 1 ofl0O.anonresistanceforQ1 of10Q'anacquisitiot a
of lO us, a maximum peak-to-peak input voltage of l0 V, a maximum output current from zr i: i
mA. and an accuracy of I 7..
Solution
i: C!
Reirranging and solving for C yields
C=,+
where
= maximum
i = maximum
dv = maximum
C
./t
capacitance (tarads)
Thercfore.
r0v
I0 nF
r:RC
where
-l
The charge time ofcapacilor Cr is also dependent on the accuracy desired from fhe device'
cent accuracy and its required RC time constant are summarized as tbllows:
Accuracy (7.)
t0
Charge Time
2.3r
I
0.1
0.0r
280
Chapter 6
6.9r
9.2r
Tl: :F
For an accuracy
of l%.
r9'{
c= .1.6(20
slower analos-
= ro87nF
To satisfi the output current limitations ofZr. a maximum cipacitance of l0 nF was rcquired. To saF
isfy the accuracy requircments. 108.7 nF was required. To satisfy borh requirements. the smallervalue capacitor must be used. Theretbre. Cr can be no larger than l0 nF.
nd the wavefofl:
3 fbllower Zr an;
harging time cor'
rrr discharge raF-
5-4-1
Sampling Rate q
The Nyquist sampling theorem establishes the minimum sanpling rate (f,) that can be used
for a given PCM system. For a sample to be reproduced accurately in a PCM receiver, each
cycle of the analog input signal (f,,) must be sampled at Ieast twice. Consequently. the minimum sampling rate is equal to tu,ice the highest audio input tiequency. Iff, is less than two
timesr,. an impairment called rrlirrs orfi dowr tlistortion occurs. Mathematically. the minimum Nyquist sampling rate is
Ltage immediatei;
'oss Cr. The inte:'
in series with C
,rk. Also, note ih:
. called draoP an:
r and the input ir-'
(6-l
A sample-and-hold circuit is a nonlinear device (mixer) with two inputs: the sampling pulse
lue capacitor that
aj
and the analog input signal. Consequently. nonlinear mixing (heterodyning) occurs between these two signals.
Figure 6-6a shows the frequency-domain representation of the output spectrum
from a sample-and-hold circuit. The output includes the two o ginal inputs (the audio and
the fundamental frequencl, of the sampling pulse). their sum and difference frequencies
f, a.l"). all the harmonics ofl" andL, (21 . 2L,, 31.. 3/,. and w on), and their associated cross
). an acquisition ti=t
;unent from Zl oi
21,-
:s
2f"- t"
3fs
f.
31" +
l"
Freqle.:y +
1b)
FIGURE 6.6
distortion
::
Transmission
2e1
I
3 kHz
FIGURE
8 kHz
6-7
Exarnple 15-2
occur. Figure 6-6b shows the results when an analog input frequency greater than ' :
modulates 1,. The side trequencies tiom one harnonic fold over into the sideband oi .:other harmonic. The tiequency that folds over is an alias of the input signal (hence '-r.
names "aliasing' or "fbldover distortion"). If an alias side frequency from the first L;monjc firlds over into the audio spectrum. it cannot be removed through filtering or -other technique.
Example 6-2
For a PCNI \vstem \r'ith a Inaxirnum audio input liequency of :l kHz. determinc the minimum '-:ple ratc and the alias tieqLrenc) produced if a 5-kHz audio siSnal were allowed to enter the sar: and hold circLrit.
Solution
TheinputbandpasstiltershowninFigure6-2iscalledanrirrriuliusirryotantifolLi,.'.-
lilte,: Its upper cutoff frequency is chosen such that no frequency greater than one-hal: sampling rate is allowed to enter the sample-and-hold circuit. thus eliminating the p.-.'bility of tbldover distortion occurring.
With PCM. the analog input signal is sanrpled. then converted to a serial binary .',:'-The binary code is transmitted to the receiver. where it is convefied back to the original ---
logsignal.ThebinarycodcsusedfbrPCMarer-bitcodes,wherenmaybeanypositi\.:teger greater than l. The codes curently used fbr PCM are sigtr-nagnintde todes, tt:.:-z
the nost sigtlific/], t Dit (MS B) is the sign bit and the remaining bits are used tbr magnii -.E
-,
Table 6 I shows an n-bit PCM code where rr equals 3. The most significant bit is use:
-=
(logic
t\\i
I = positive and Iogic 0 = negative). The
represent the sign of the sample
maining bits represent the magnitude. With two magnitude bits, there are fbur codes pc'--
Table
Sign
61
ll
t0
0
0
0
0
2A?
Chapter 6
-3
rl
0r
+i
00
00
+0
0l
t0
lt
0
I
-l
l
.]]IEE=
ble for positive numbers and four codes possible tirr negativc numbers. Consequently, there
is a total of eight possible codes (23 : ti).
With quantization. the total voltage range is subdivided into a smaller number ol
subranges. as shown in Table 6 2. The PCM code shown in Table 6-2 is a three-bit signmagnitude code with eight possible combinations (four positive and tbur negative). The
le ftmost bit is the sign bit ( I : *and0=
). and the two rightmost bits repre\ent mugnitude. This type of code is called a lbldr cl bintrv cotle because thc codes on the bottom half
of the table are a miror imagc of the codes on the top half. except tbr the sign bit. If the
negative codes were folded over on top ol the positive codes. they would match perfectly.
With a folded binary code. each r oltage level has one code assigncd to it except zero volts.
which has two codes, I 00 ( + 0) and 000 ( - 0 ). The rnagninrde diflercnce between adjacent
steps is called the qtloltia.ttion inter\\i ot clud tufir. For the code shorvn in Table 6-2. the
quantization interval is I V Therefore. fbr this code. the nraximum signal magnitude that
can be encoded is +3 V (lll) or -3 V (011), and the minimum signal magnitude is * I V
( 101) or
-l V (001). If the magnitude ofthe sample exceeds the highe't qulntizutit,n interval, overload disaol Ii./r (also called perrl //atltirtg) occurs.
Assigning PCM codes to absolute magnitudes is called quantizing. The ma-snitude of
a quantum is also called the rcsolutiotl. The resolution is equal to the \oltage of the
nlininun step si:.e. which is equal to the voltage of the /r,.r.ra.!i gnificati bit (V t,) oi the PCM
code. The resolution is the minirnum voltage other than 0 V that can be decoded by the
digital-to-analog converter in the receiver The resolution tbr the PCM code shown in
Table 6-2 is I V The smaller the magnitude of a quantunr. the better (smaller.) the resolution and the more accurateiy the quanlircd signal will resemble the original analog sample.
ln Table 6-2, each three-bit code has a rangc of input voltages that will be converted
to that code. For example, any voltage between +0.5 and + 1.5 will be converted to the
code l0l (+ I V). Each code has a quonti:utiott rtirr.gc equal to i or one-half the mag
nitude of a quantum except the codes tbr f0 and 0. The 0-V codes each have an input
range equaJ to only one-half a quantum (0.5 V).
jinq or anti|ol(b\'.'
er than one-half th'
Table
6'2
Sign
l,
I Sub
ranges <
ligital Transmission
j
Magnitude
Decimal
value
Quantizat:on range
+f
+e
+1
+0
+4.5 V to +f.5
+1,.5
+4.5
+0.5 V to +1,.5
0 V to +0.5
-0
-1
-a
-3
0 V to -0.5
-0.5 V to
V to -e.5
-a,5 V to -1.5
l,
l,
l,
283
111
+3 V
110
+2 V
101
+r
1oo 1
ooo I
@1
-1V
010
-2
011
-3
111
+3
110
+2 V
101
+1V
;s]
o"
00't
010
011
-1
-2
-3
v
Sample
time I
Sample
lime
Sample tim
ltoloorirrr
FIGURE 6-8 {al Analog input signal; [b] sample pulse; [c] PAM
signal; [d] PCM code
Figure 6-8 shows an analog input signal. the sampling pulse. the correspon:
=
quantized signal (PAM). and the PCM code for each sample. The likelihood of a sar--:'c
voltage being equal to one of the eight quantization Ievels is remote. Therefore, as sh o:
in the tigure. each sample voha,se is rounded off (quantized) to the closest available ' :
:r
and then convefted to its corresponding PCM code. The PAM signal in the transmitter i'
sentially the same PAM signal produced in the receiver. Therefbre, any round-off erro:'
the transmitted signal are reproduced when the code is converted back to analog in th: -=
ceiver. This error is called the clutnti:.ution e rror (Q,).The quantization error is equi\
'5
to additive white noise as it alten the signal amplitude. Consequendy, quantization err' r
also called 4rrrattl.:a tion rtttise (0,,). Thc maximum magnitude for the quantiTation err' : !'
equal to one-half a quantum ( a0.5 V for the code shown in Table 6-2).
The lirst sample shown in Figure 6-8 occurs at time fr. when the input Yoltage i"'
actly +2 V The PCM code that corresponds to +2 V is I10, and there is no quantiz3: f,
error Sarnple 2 occurs at time t.. when the input voltage is I V The conesponding P(l{
code is 001. and again there is no quantization error. To determine the PCM code for a :'=
ticular sample voltage, simply divide the voltage by the resolution' convert the quotie': i'.
an n-bit binary corle. and then add the sign bit. For sample 3 in Figure 6-9. the voltage ':
is approximately +2.6 V The tblded PCM code is
5
rr.,rlutron |
slmpleroltlge -
-)A-
--;
There is no PCM code fbr *2.6: therefore. the nlagnitude of the sample is rounded o::
the nearest \ alid code. which is I I I , or * 3 V The rounding-off process results in a qu':-
zation eror of
244
Chapter 6
0..+
{a)
(b)
tj t2 t3-----___--____________-_-_-_------
tN
(c)
FIGURE
signal
the correspondi:i
:Lhood of a samp .
[herefore, as sholr:
r\esl aYailable le\:
rhe transmitter is
6-9
The quantized signal shown in Figure 6-8c at best only roughly resembles the original analog input signal. This is because with a three-bit PCM code, the resolution is rather
poor and also because there are only three samples taken of the analog signal. The quality
ofthe PAM signal can be improved by using a PCM code with more bits. reducing the magnitude of a quantum and improving the resolution. The quality can also be improved by
sampling the analog signal at a laster rate. Figure 6-9 shows the same analog input signal
shown in Figure 6-8 except the signal is being sampled at a much higher rate. As the tigure
shows, the PAM signal resembles the analog input signal rather closely.
Figure 6-10 shows the input-versus output transfer function tbr a linear analog-todigital corverter (sometimes called a linear quantizer). As the figure shows for a linear analog input signal (i.e., a ramp), the quantized signal is a staircase tunction. Thus, as shown in
Figure 6-7c, the maximum quantization error is the same for any magnitude input signal.
.-
. round-off error: :
ro analog in the r:
n error is equivale::
quantization error
.luantization erlor :.
input voltage is e.
e is no quantizati..:
Example 6-3
;orresponding PC\l
PC\'[ code for a p;
n\ efi the quotienl :
6-9. the voltage at-
For the PCM coding scheme shown in Figure 6 8. determine the quantized volta8e. quunlilatjon er
ror (0.), and PCM code for the analog sample voltage of + 1.07 V
Solution
To determine thc quantized level. simply divide the sample voltage by resolution a0d then
round the answer off to the nearest quantization level:
+ I .0'7 V
lv
The quantizaiion error is the difference between the original sample voltage and the quantized le\el. or
Q.=
From Table 6-2. the PCM code for +
- lital Tnansmission
t.o1 l = 0.07
I is l0l.
285
Ouantizetion
arrq
Ouentizad rignel
Maximum
positive
n6g6tiv6 quantizing
quantizing
06-r*$B
FIGUBE
6-10
Lin
q-r-
tization; {cl G.
6-4-3
Dynamic Range
The number of PCM bits transmitted per sample is determined by several variables. :cluding maximum allowable input amplitude, resolution, and dynamic range. D_tlrd-:r:
ronge (DR) is the ratio of the largest possible magnitude to the smallest possible magnir,.r
(other than 0 V) that can be decoded by the digital-to-analog converter in the recei.=
Mathematically, dynamic range is
DR
where
=Yto
a:
V*"* =
receiver
Chapter 6
@--:-
as
v.,--
u^:
(6-l)
-------
resolutlon
1t/
DR=fi:3
A dynamic range of 3 indicates rhat the rario of the largest decoded voltage to the smallest
decoded signal voltage is 3 to 1.
Dynamic range is generally expressed as a dB value; therefore,
DR
20 tog
l"''
(6-l)
DR = 20 log 3 = 9.54 dB
1..' to
'
The number of bits used for a PCM code depends on the dynamic range. The relationship between dynamic range and the number of bits in a pCM code is
rantizrtion
2"
afiol
I >DR
(6-5a)
2".1:
DR
(6-5b)
2"=DR+1
To solve for the numberofbits (n) necessary to produce a dynamic range of3, convert to logs,
= log(DR + t)
n log 2 : log(DR + l)
log 2"
ll
togz :
lngQ.
"=
0.602:' .
o3ol
For a dynamic range of 3, a PCM code with two bits is required. Dynamic range can be ex_
pressed in decibels as
where
On,ra,
20
loSffi
DR,as,
20
log(2" l)
,1,
t6-6t
as
DR,ur, = 20 log(2')
,\
the DACS in
= 20n tog(2)
tbi
=6n
gital Transmission
t6-7
287
Table
6-3
NumberofBits
in PCM Code (n)
Number of Levels
Possible (M = 2')
18.1
.1
l6
24.t
32
30.1
36.1
6
1
l2fl
256
48.2
512
51.2
60.2
42.t
i0
102,{
11
2048
66.2
t2
.1096
'72.2
l3
8192
78.3
l4
16,384
8.1.3
15
32,768
65.536
90.3
96.3
t6
Equation 6-7 indicates that there is approximately 6 dB dynamic range for each magnir,r
a linear PCM code. Table 6-3 summarizes dynamic mnge for PCM codes with I :;:'
for values of a up to 16.
bit in
Example 6-4
For a PCM system with lhe following parameters, determine (a) minimum sample rate. (b) minir:-a
number of bits used in the PCM code. (c) resolution, and (d) quantization elror'
kHz
,1
= :t2.55 V
Solution a.
f'=2f,':2(4kHz) = 8 kHz
b.
To determine the absolute value lbr dynamic range, substitute into Equation 6-4:
16
dB = 20 los
vi+
2I : los+
,6: I :
199.5
l!:
DR
The minimum number of bits is determined by reaflanging Equation 6-5b and solving for
"-
5-:-!
'o*('oo
log
'o'
The closest whole number greater than 7.63 is 8l therefore. eight bits must be used for the magnir-I
Because the input amplitude range is a 2.55, one additional bit. the sign bit, is required. Th:fore. the total number of CM bits is nine. and the total number ol PCM codes is 2q 512. (There
255 positive codes. 255 negative codes, and 2 zero codes.)
To determine the actual dynamic range, substilute into Equation 6-6:
DRronr
20 log(2" - l)
:20(ktC256 l)
= 48.13 dB
288
Chapter 6
--
c. The resolution is determined by dividing the maximum positive or maximum ncgative !oltage by
the number of positi!e or negative nonzero PCM codes:
re.otutton
. 1.v ,,.
I
2.55
156
0.01 V
a,: resolulion
6-4-4
uuj u
u.ou,
CodingEfficiency
100
(6-8)
coding efficiency
rre.
T "
100
= 95.89%
(b) minimu=
:.5
soR
: resolution V,r,
0,,
For the PCM code shown in Figure 6 8. the worst-case (minimum) SQR occurs tbr the lou est magnitude quantization voltage (:l I V). Therefore. the ninimurn SQR is
SoR(,,,")
f,nd solving for n
:O:=2
:
or in dB
20 log(2)
= 6dB
For a maximum amplitude inpul signal of 3 V (either 1l I or 011). the na\imunr quantization noise is also equal to the resolution divided by 2. Therefore. the SQR tbr a maximum
input signal is
s0Rr.",,=
or in dB
Eital Transmission
Y2
A =orl=u
:20 log6
= 15.6 dB
289
From the preceding example. it can be seen that even though the magnitude of t:i
quantization error remains constant throughout the entire PCM code, the percentage err]'
does not: it decreases as the magnitude of the sample increases.
The preceding expression tbr SQR is fbr voltage and presumes the mlximum qu.r:tiration en1)r: therefi)r-e. it is of litlle practical use and is shown only fbr comparison puposes and to illustrate that the SQR is not constant throughout the entire runge of samF .
amplitudes. In reality and as shown in Figure 6-9. the difference between the PAM wa\.firrm and the anakrg input rvavetbrm varies in magnitude. Therelbre. the SQR is not cor'
stant. Cenerally. the quantization enor or dislortion caused by digitizing an analog samp..
is expressed as an average signal pou er- to-rr eruge noise po\ er ratio. For linear PC\l
codes (all quantization intervals have equal magnitudes). the signal porer-ro-tlthtnti:it. noise pover rzrai./ (also called.rl.grrrr1-ro-rlistortiotl rdtio or signol-to-rtoise /4lio) is deternrined hv the lbllowing tbrnrula:
Jo,t
where
R:
,.
Io los
r,r/R
r'=
resistance (ohms)
rms signal voltage (volts)
i/
(6-9a
Q-12)/R
li/)2)lR :
If the resistances
soR
6-6
t ,l
t0 1o8] ./r/ I 2
l0.E + 20 log
L'
I
l
(6-9t
u\"
Early PCM svstems used litteur rndes (i.e.. the magnitude change between any two sucessive steps is unitbrm). With linear coding, the accuracy (resolution) for the highe:
amplitude analog signals is the same as for the lower-amplitude signals, and the SQR r :
the lower-ampliiude signals is less than fbr the higher-amplitude signals. With voice trar-'
rnission. lorv-amplitude signals are more likely to occur than large-amplitude signa.
Theretbre. if there were more codes for the lowcr amplitudes. it would increase the acc-'
racy $here the rccuracv is needed. As a result. there u,ould be t'ewer codes available tbr t:,:
higher amplitudes. uhich uonld incrcase the quantization etror for the larger-amplitui.
is
called
nonliteaU
nonuniJitntt enurtlin g. With nonlinear encoding. the step size increases with the amplitur
of the input signal.
Figure 6 1l shou's the step outputs from a linear and a nonlinear analog-to-digi:converter. Note. $ ith nonlinear encoding, there are more codes at the bottom ofthe sc".:
than there are at the top. thus increasing the accuracy fbr the smaller-amplitude signa ,
AIso note that the distaoce between successive codes is greater for the higher-amp.tude signals. thus increasing the quantization error and reducing the SQR. Also, becau.:
the ratio of V,,,,. to V.,',, is increased with nonlinear encoding. the dynamic range is lar-s:than with a unilbrru linear code. It is evident that nonlinear encoding is l compromr.:
SQR is sacrificed tirr the higher-amplitude signals to achieve more accuracy for i::
lower-amplitude signals and to achieve a larger dynamic range. It is difficult to fabric.--:
290
Chapter 6
nagnitude of the
percentage eITOr
maximum quan'
.ompadson puri range of samPle
n the PAM wave: SQR is not con-
rn analog samPle
. For linear PClrl
\., e
r-to- qLttuttiaitl
lnput lovel
1a)
6-11
(6-9a
^#ech6nn6lnoP-n
Unifo.m code
with midris
Unilorm code
with midtread
quantization
quantizstion
*-,r-
(6-9b
Ocodod noise
FIGURE
r the larger-amPlitud:
6-12
No decod6d noise
.7
r1
6 -
called ,lo[li,ear o:
.es with the amPlitud'
During times when there is no analog input signat, the only input to the PAM sampler is
random. thermal noise. This noise is called idle channcl toise and is conve{ed to a PANI
js quantized by the ADC'
sample just as if it were a signal. Consequently, even input noise
method
c alle,J nidtreutl quutttinoise
by
a
Figure 6,12 shows a way to reduce idle channel
is made larger in ampliturje
quite
be
lalge and still be quantized
noise
can
than the rest of the steps. consequently. input
during the encodin-:
is
suppressed
the
noise
as a positive or negative zero code. As a result.
the
plocess.
,1er-amPlitude signal:
In the PCM codes described thus tar. the loweslmagnitude positive and neertir:
codes have the same voltage range as all the other codes ( + or - one-half the resolLrlion
This is called rzirTri se quafitiaation. Figure 6 12 contrasts the idle channel noise tran:t:litteJ
with
,iing is a compromise
more accuracY for th:
r i. tlifficult to fabricat'
a midrise PCM code to the idle channel noise transmitted when midtread
quanliz,ltl.n
is used. The advantage of midtread quantization is less idle channel noise. The disad\ xnlrlie
is a larger possible magnitude for Q" in the lowest quantization interval
: lital Transmission
291
With a fblded binary PCM code, residual noise that fluctuates slightly above and ::.
- zero PCM code and, consequently, is eliminar.:
In systems that do not use the two 0-V assignments. the residual noise could cause the PC\1
encoder to alternate between the zero code and the minimum i or code. Conseque!:
the decoder rvould reproduce the encoded noise. With a folded binary code, most ol'--c
residual noise is inhelently elininated by the errcoder.
6-8
CODING METHODS
There are several coding methods used to quantize PAM signals into 2" levels. These me':ods are classified according to whether the coding operation proceeds a level at a tim. digit at a lime . or a u,ord at a time.
6-8-'l
Level-at-a-Time Coding
This type ol coding compares the PAM signal to a rarnp *avetbrm while a binary cou::j
is bcing advanced at a unifbrm rate. When the ramp waveform equals or exceeds the P: .'
sample. the counter contains the PCM code. This type of coding requires a very fast c], '-.
if the number of bits in the PCM code is large. Level-at-a-time coding also requires thl: :
sequential decisions be nrade for each PCM code generated. Therclbre. level-ara-time i .,-ing is generalll' Iimited to low-speed applications. Nonunilirrnr coding is achieved by -ing a nonlinear function as the ref'erence ramp.
6-8-2
Digir-ar-a-Time Coding
This type of coding determines each digit of thc PCM code sequentially. Digit-at-a-r:
coding is analogous to a balance where known reference weights are used to determin. -:
unknou'n rveight. Digit-at-a-time coders provide a compromise between speed and c, --plexity. One common kind of digit-at-a-time coder. called a Jeeclback coder, uses a suc,.sive approximation register (SAR). with this type of coder. the entire PCM code uor: :
detemined simultaneously.
5-8-3
Word-at-a-Time Coding
Word-at-a{ime coders are flash encoders and are more complex; however, they are r: ,.:
suitable tbr high-speed applications. One common tvpe of word-at-a-time coder uses rtiple threshold ci[cuits. Logic circuits sense the highest threshold circuit sensed by the P:-\'
input signal and produce the approximate PCM code. This method is again impracric!. ; r
large values of n.
6-9
COMPANOING
Cotrpandiry is the process of colrpresslng and then e.rpairrling. With companded sysr::-the higher-amplitude analog signals are compressed (amplified less than the louer-ampli:-signals) prior to transmission and then expanded (amplified more than the lo$er-ampli:-.=
signrJs) in the receiver. Companding is a means of improving the dynamic mnge of a .. munications system.
Figure 6- l3 illustrates the process of companding. An analog input signal with 3 :
nanric range of 50 dB is compressed to 25 dB prior to transmission and then, in the rece:.
=
expanded back to its original dynanic range of 50 dB. With PCM. companding may be -complished using analog or digital techniques. Early PCM systems used ana)og comp--:ing. uhereas more modern systems use digital companding.
6-9-1
AnalogCompanding
Historically. analog compression was implemented using specially designed diode. :serted in the ana]og signal path in a PCM transmitter pdor to the sample-and-hold cir. - ,
Chapter 6
{rEa
r\ eliminate.
rJuse the
PC\l
25 dB
Compressed
dynamic
range
Clxsequentli
le. most of th.
+10 dB
+10 dB
+10 dB
0dB
o
binary coun:-
0dB
0dB
0
f
3
6'
l-xt-a-time co:b1 u:
'chieved
Digit-at-a-ti--r
determina i
.peed and cc:-
-30 dB
tLr
7: Llses
\l
lnput
FIGURE
impracticai :
Outpul
Analog expansion was also implemented with diodes that were placed just after the low-
namic range of the analog signal is compressed, sampled, and then convened to a linear
PCM code. In the receiver, the PCM code is converted to a PAM signal, filrered. and then
expanded back to it. r'riginal dynamic range.
Different signal distributions require different companding characteristics. For instance, voice-quality telephone signals require a relatively constant SQR pedormance over
a wide dynamic range, which means that the distortion must be proportional to signal amplitude for all input signal levels. This requires a logarithmic compression ratio. which requires an infinite dynamic range and an infinite number of PCM codes. Of course. this is
impossible to achieve. However, there are two methods of analog companding cunent)r being used that closely approximate a loga thmic function and are often called log-PCM
codes. The two methods are p-/aw and the A-lax companding.
prnded syster''
1.,\\
6-13
Transmission
media
are mi-
alrder uses m,
r.ed by the PtY
1
a succ3-
code word
r. Ihey
-30 dB
er-amplitur
er amplitu::
range of a c0:-
1..$
Y.", ln(l +
%,,
!al Transmission
ttvrlv*.)
ln(l +
[)
(6-l0l
2S3
r;;;** I fu r--;t'V
a*r"s-.*l nrri, I
fh
PAM
\,
FIGUBE
npu'i
614
. I ".-,.;- l--*l'i*P,f;:lf
l.l
lAl
'
l-*
I
uL_
PCM transrniner
FIGURE
6-15
p]aw compression
characteristics
where
v',,.,,.
trr
y",, =
r
Figure 6- l5 shows the compression curves for several values of p Note that the hiShe:
(no
compressioi
is
linear
g.1h. ,nr. .o-p..ssion. Also note that for p = 0, the curve
The parameter p determines the range of signal power in which the SQR is relatire"
constant. Voice tlansmission requires a minimum dynamic range of 40 dB and a sever-:r
PCM code. For a relatively constant SQR and a40-dB dynamic range, a p > 10O is requ:-I
The early Bell System PCM systems used a seven-bit code with a p = 100 However.
:
255'
most recent PCM systems use an eighGbit code and a U
2 9,1
Chapter 6
Example 6-5
For a compressor with a p
255, determine
a. The voltage gain for the following relative values of yin: y,,,",. 0.75 4,.,. 0.5 y,,,"..
Serial
PCM
b.
_'l
Solution r.
Substituting into Equation 6-10. rhe tbllowing voltage gains are achieved fbrthe given
input magnitudes:
Compressed
VrtaSe Gaio
I
l(r)
0.75 v,,,,,,
0.5 v-"_
0.25 v,.,,
--*r,",
PCM
1.26
L75
3.00
Y,"
(.
Voltase
Gain
Y.,,,
1.00
.1.00
0.75(",,:3v
t)6
_1.78
0.50Y,,,.,=2V
t.75
3.50
0.25Y-,-=lV
3.00
300v
lopl
- I: dB
20
:0 l,rF1 -
compression
2.5 dB
:12dB-2.5dB=9.5dB
To restore the signals to their original proportions in the receiver, the compressed
voltages are expanded by passing them through an amplifier with gain characteristics that
are the complement of those in the compressor For the values given in Example 6-5, the
voltage gains in the receiver are as follows:
Expanded
V{)ltage Gain
. comprcs5ion
1.00
0.75 v.,".
0.5
0.25 v,,".
t;",
:ime (volts)
unitless)
v,,,
:0.75 v-,"
V,"
0.5
V-,-
1.26x0.19=t
1.75 x 0.57 = I
3x0.33=l
is require:
100. However.
t-
6-9-l-2 A-lau conrpanding. In Europe, the ITU-T has established A /rlr com,
panding to be used to approximate true logarithmic companding. For an intended dYnamic
:al Transmission
0.33
txl=l
> I 00
o.5'7
The overall circuit gain is simply the product of the compression and expansion factors, which equals one lbr all input voltage levels. For the values given in Example 6-5.
o'79
295
Compressed
parallel
J\ft
Compressed
serial
lth
Jrni_
t
corpiessed
Linear
parallel
PCM
FIGURE
6-16
parallel
PCM
A-la'f companding has a slightly flatter SQR than U-law AJ'n' companding'
noise) The
eve;. is int'edor t; !-law in tcrms of small-signal quality (idle channel
is
pression characteristic tbr A-lan companding
range.
l6-1
1+
1n(A
+ lnA
6-9-2
vi"/v,,",
Yr.
<I
ls
(6'l1n
DigitalComPanding
has .--:=
Digital comlanding iniolves compressior in the transmitter after the input sample
decG: +
coirerted to a linear PCM code and then expansion in the receiver prior to PCM
system'
PCM
companded
fbr
a
digitally
diagram
block
the
Figure
- 6-16 sho\\,s
N\I
wi,h digi,ul.nrnpanding, the analog signal is first sampled andconverted to a linear
compressedPC\l: 'code and thenlhe linear code is digitally compressed ln thereceivelthe
rccent digitall) ' 'trThe
most
analog)
to
(i.e
back
convened
decoded
is expanded and then
'
; 'r
\\::
I
p-/r'tl curves
PCN{
pressed PCM systems use a 12-bit linear PCM ctde and an eight-bit compressed
--(
'
designared Aan X are truncated during compression and subsequently lost Bits
296
Chapter 6
'ai
]V
Segment
Sgmont
Sggment
Segment
+4
+5
+6
*7
64:1
comprBsion ratio
16:1 comprgssion
letio
S6gment
l.,trr
_l
Segmenl12
+3
et
Segmgnt
Segment
nrpanding.
r.-,ise). The
+1
+0
1:1 no comprossion
1:1 no comprsssion
h!-;'
1.0
c.':'
FIGUBE
(6- l
lr
sisntt |
1.+ I
0-- I
(6-1lt'
rr sample has
6-l7
3-Bit |
I
quantiration
000to'll1 I
0000to1111
6ogmant
idontifier
+Bir
ifi.rv.|
A BCD
tr=
.:tem.
r..d to
linear PC\i
rpressed PCM
il=
ct'=
-/rirr'curves lr't: i
:nents (segmen--' i
Encoded
Dacoded
PCtrt
PCM
12-Bit
&Bit
&Bit
mmprcss6d
compressod
code
12-Bit
r6covered
codo
Sgm6nt
previous segnEa
S.gmont
linearcod6
80(X)0000ABCD
s000ABCO
SoO0ABCD
soolABCD
s00lABCD
s(x)OmoIABCD
s0O0(Al ABCDX
s0o001A8CO)O(
s0l0ABCO
s011A8CO
slmABCD
s0l0ABCD
s01lABCD
s1(DABCD
s10lABCD
s101ABCD
sO001ABCD100
s001ABCD1 000
3110ABCD
.IlOABCD
s01ABC010000
slt IABCD
s11'IABCD
s l ABCD
100000
1e
t
)J l|-seqnent c -'*
rhough there are
Lght
line with
,l
c-'r-
r-
s0o0lABCOx)(x
s001 ABCOXn<X
so tABCoXXXXX
slABCO)(xXxxX
rpressed code i:
ions designated tt
.rgnated A, B, C.
code
S0000000ABCO
s0O00001ABCO
s000001ABCD1
ABCD10
3
4
sO0O0'l
{b)
iURE 6-18 12,bit-to-8-bit digital companding: [a] 8-bit p255 compressed code format
-255 encoding table; {cl p255 decoding table
297
D are transmitted as is. The sign bit is also transmitted as is. Note that for segments [r -,;
I , the encoded I2-bit PCM code i s duplicated exactly at the output of the decoder (com; --:
Figures 6-l8b and c), whereas fbr segment 7, only the most significant six bits are du:,cated. With I I magnitude bits, there are 2048 possible codes, but they are not equall) - tributed among the eight segments. There are l6 codes in segment 0 and l6 codes in .:;'
ment l. [n each subsequent segment. the number of codes doubles (i.e.. segment 2 ha. .-:
codes: segment 3 has 6.1 codes. and so on). However. in each of the eight segments. i
16 l2-bit codes can be produced. Consequently. in scgments 0 and 1. there is no comp.:sion (of the l6 possible codes. all l6 can be decoded). In segment 2. there is a compre.. t
ratioof2:l (of the 32 possible codes. on ly l6canbedecoded). In segment 3, there is " compression ratio (64 codes to l6 codes). The compression ratio doublcs with each -cessive segment. The compression ratio in segment 7 is 1024l)6, or 64 1.
The compression process is as tbllows. The analog signal is sampled and conr;:--:l
to a linear l2-bit sign-magnitude code. The sign bit is transferred directly to an eigh:-z
compressed code. The segment numbcr in the eight-bit code is determined by counting number of leading 0s in the I I -bit nragnitude poftion of the linear code beginning u it: most significant bit. Subtract the number of leading 0s (not b excced 7) from 7. The r.-..r
is the segment number. which is converted to a three-bit binary number and insened .:ir
the eight-bit compressed code as the segment identifier. The lbur magnitude bits (A. B :
and D) represent the quantization interval (i.e., subsegments) and are substituted int, least significlnt fbur bits of the 8-bit compressed code.
Essentially, segments 2 through 7 are subdivided into srnaller subsegments. Each qment consists of l6 subsegments, which correspond b the l6 conditions possible tb: r*
A. B. C, and D (0000 to I I 1 1). In segment 2, there are two codes per subsegment. In --:ment 3. there are fbur. The number of codes per subsegment doubles with each subse!,5
segment. Consequently. in segment 7, each subsegment has 6,1codes.
Figure 6-19 shows the breakdown of segments versus subsegments for segmer:. I
5, and 7. Note that in each subsegment, all 12-bit codes. once compressed and expa:]:a
yield a single 12 bit code. In the decoder, the most siSnificant of the truncated bits is :.rserted as a logic 1. The remaining truncated bits are reinserted as 0s. This ensures th.:
=
maximum magnitude of error introduced by the compression and expansion process i. -:imized. Essentially. the decoder guesses what the truncated bits were prior to encoding -l
most logical guess is halfway between the minimum- and maximum-magnitude code. :r
example, in segment 6, the five least signiticant bits are truncated during conrpre.. 'r
therefore. in the receiver, the decoder must try to determine what those bits were. The :'sibilities include any code between 00000 and 1 I I I l. The logical guess is t0000, app- '.mately half the maximum magnitude. Consequently. the maximum compression er: - r
slightly more than one-half the maximum magnitude tbr that segment.
Example 6-6
Determine the l2 bit linear code. the eight bit conpressed code. the decoded 12-bit code. the :-5
tization eror. and the compression error fbr a resolution o10.01 V and analog sample voltage: !
+0.053 V. (b) -0.318 V and (c) + 10.2,1,1 V
Solution r. To determine the l2-bit linear code. simply divide the sample voltage by the
tion. round ott'the quotient. and then convert thlr result rc a l2-bit sign-magnilude code:
+0051v
ffi
0000000
ABC
0101
quantization error
I
sign bit
(l= +)
298
Chapter 6
r:r *
Sogment
rgments 0 anc
.rder (comPar.
si::cr9t
64:
s111'11100000
15
64:
s11110'100000
14
64:
s11101100000
54:
s11100100000
12
sl1011100000
1l
64:
s11010100000
10
64:
s11001100000
64 |
s11000100000
s'l1111mOOOO
sl1110111111
'"iii1;iffiin
r codes in
:ment 2 has ll
.egments, onl,'
i\ oo comPle!l comPressit':
L there is a -1: .
\\ ith each su;se-s-
s11101111111
s11101000000
s1l'100111111
s1r100000000
s11011111111
54
s1101t000000
s11010111111
I und converte:
s11010000000
s1100'l111111
to an eight-b::
;iiil1ilil;
hi counting th:
:inning with th.
sll000?11111
rm 7. The resu'.
s1
nd inserted in::
le bits (A. B. C
.ror i i iri:iririro
s101101111'!1
1000000000
s10111111111
64:1
s10111100000
64:
s10110100000
64:
s10101100000
64:
s10100100000
64:
sl00'11100000
64:
s1001010000O
64:
s10001100000
64:
s10000100000
s:0110000000
s10101111111
01000000
s!010O111111
s l01
segment. ln se-i-
elch subseque:.'
s10100000000
sl0O111l1:11
. fbr segments
s1001 1 000000
s100101t 1111
:d and exPandec
s10010m0000
sl0001111111
s10001000000
ing comprcssit'
iu rvere. The Pt':
s100001111:1
............
1
0000000000
sIABCO--.-(a)
10000, aPPror.:-
rpression error
l'
rion error
10000000
:tal Transmission
=O)
sign
bir
identifier
unit
(+)
(segment 0)
000
000
(000
= segment 0)
segment 0
sign
bir
has
(+)
seven leading 0s
000000
-'7
t1
= +5(0.0t
0101
ABCD
quantization
interval
(s)
0101
0101
ABCD
quantiTation
intenal
(0101
0010
5t
1=
+5
+{).05
2Sg
Segm6nt
12-Bk linosr
cod.
s00111111111
Subagment
16:
30011't 111000
15
'16:
s00111101000
14
s00111110q)0
s00111101111
s{0111100000
s0011101',1111
101 1000
'13
0010000
12
16:1
s0O110111000
11
16 r1
10
16:1
16:1
s001100010o0
16:1
s0010'1111000
'|
16:l
soo10'1101000
'16:
s001 01 01 1000
16:
s00101001000
s001
s00111010000
s00111@1111
16:
s001
s0011r000000
s00110':11111
s001 101 10000
5{0110101111
;ini
1; iii,i,oo
s00110011111
soo110010000
s00110001111
s00110000000
s00101111111
600101110fi)0
s00101101111
;oiri;ii;i,i,oo
s001010'111'11
s001010 10OO0
soo'!01001111
<m101000000
s0010o111111
16:1
s001001
16:1
s00'! 00101000
16r
s00100011000
16:
s00100001000
1000
sml00110000
s00100101111
soo1 00100000
s0010001I1II
sool mO1O,r0O
s0o1{XD01111
ioo10O000O{X)
soolAaCD-.'{b}
FIGURE 5-1
12-b:: ---As Example 6-6 shows. the recovcred l2-bit code ( +5) is exactly the same as the original
r i -9
encoded
the
original
same
as
(+0.05
V)
is
the
voltage
(
ear code +5). Therefore, the decoded
error in se!=:'
( + 0.5). This is true for all codes in segments 0 and L Thus. there is no compression
qur:' :.
0 and l. and rhe only error produced is from the quantizing process (for this example' the
oil-ll v
- -ll.x.$hichi'roundedofl to -J2.proclucinF -a
'001 v
-0'002 v
quJnli/ation error O, - 0'2;u0l vr
12-bit linear code
ABC D
00000100000
(- I l -bit magnitudc bils -------+
0
I
sign bit
(0:
300
Chapter 6
Sogmgnt
sdxm111111
sooooo111110
s00000111101
s00o0o1111OO
s0000011? :11
I z,t
s000001]l101
s00000111011 I
2
I 2: t
s000001110t1
....,....,-, } 2:1
s00000111001
s00000111010 l
s0000o111001 I
s00000111000
. ...
s0m00110111
lz.,
s00000110t11
............ lI z:t
s00000110101
Subogmant
............ iz,t
............
codo
l2-git oxpencled
14
11
s00000110110 l
s00000110101 'l
SOO0OO1lOlOO
s000O0110011 l
L,.i
s00000110011
5O0OOO110010 .l
s000OO11OO01 1
,,.,,,,,'.,, Iz.r
s00o0o110000
s000001 '10001
s00000101111
''''
'''' 12:l
s0000010'1111
s0@0010 1 'l l0
s00000101101
s00000101101
s00000101100
s00000101 01
s00000101011
s00000101010
s00000101001 r
lr,r
s00000101001
--,,----,... l z,t
s00000100',111
s00000101000 l
s000000100111 )
2
SOOO0O100110
s0000010O101
'',''''',,,,
Iz.t
s00000100100
s000001001 01
s000000100011
. ..,.-,,.,. l
so00oo1ooo10 l
s0o000100001 I
s(aoo01o00o0
z,,l
s00000100011
s00000t00001
s000001ABCD
(c)
FIGUHE 6-1
9 [Continued)lc) segment 2
he
!: rrn
error in segmarl
\rmple. the
quanl:-_-_
NI]V
000000'l
(7 5=2)
0
sign
unit
bit
identifier
()
(seSment 2)
010
0
has
()
five
leading 0s
00000
1
0000
0000
ABCD
quantizati(nl
interval
(0000
0)
:
AB C D
1 0 0 0 0 1=
in."l"a
33(0.1 )
= -0.13 V
truncated
interval
(0)
segment 5
bir
quantization
(7-2:5.)
sign
decoded voitage
00000
ABCDX
tn.l
13
"u
301
Note the two inserted ones in the recovered 12-bit code. The least significant bit is determinec
signifi'
f.rom the decocling table shown in Figure 6-18c. As the figure shows. in the receiver the most
(0s)
segment:
For
are
cleared
bits
cant ofthe truncated bits is always set(1),andall other truncated
wa!
is only one truncated bitl thus. it is set in the receiver' The inserted I in bit position 6
codes. there
droppedduringthel2bit.to-8-bitconve$ionprocess'astransmissionofthisbitisredundantbecauif it were nol a l. the sample would not be in that segment. Conscquently' for all segments except seg'
ments 0 and l. a I is automatically inserted between the reinsened 0s afld the ABCD bits'
For this example. there are two errors: the quantization error and the compression elro' Th'
quantization e.ror ii du. to rounding off the sample voltage in the encoder t{) the closest PC}i
ciode. and the compression error is caused by forcing the truncated bit to be a I in the receire:
Keep in mind that the two errors are not a]ways additive. as they could cause errors in the opPisite Jirection and actually cancel each other The worst-case scenario would be when the two eno::
wefe in the same direction ancl at their maximum values. For this example, the combined erri.
was0.33V-0.318V=0.012VTheworstpossibleerorinsegments0andlisthemaximu::
quantization eror. or halfthe magnjtude of the resolution. [n segments 2 through 7' the worst pos:-
c.
rounde,l olf
lo
<-
I02.1. producinP r
0.410.01 v' = -0 u04
ABCD
11111
1111
sign bit
+)
111',1
---)
xxxx
11111
ABCDX
I
I
truncated
8-bit compressed code
To determine the 12-bit
I
I
l0
s
decoded voltage
't't
011',|
110
segmenr 6 A B C
1111110
I rgc
0 0 0
+1r:.r
I
insefied
inse(ed
+1008(0.01) = +10.08V
The dilterence between the oriSinal 12-bit code and the decoded l2-bit code is
10.23
10.08:0.15
1011ll1l1l11
l0llt1ll0000
-
ir l -
Is(o.ot)
o.l5 v
For this example. there ate again two errors: a quantization error of0.004 V and o comprersion
10.08 V : 0. l5'l V
of 0.15 V The combined enor is 10.234 v
:-v
As seen in Eiample 6-6. the magnitude of the compression error is not the same for all s'=
ples. However, the maximum percentage error is the same in each segment (other thax q
ments 0 and 1, where there is no compression enor). For comparison purposes, the lb-- :sing formula is used lbr computing the Percentage eror introduced by digital compre::::E
7o error =
3C2
Chapter 6
x 100
r'-
bit is determine:
Example 6-7
The maximum percentage error will occur for tlre smallesl number in the louesl subsegment within
any given segment. Because there is no compression crror in segments 0 and l. tilr seg ent 3 the
maximum percentage error is computed as lbllows:
s . For segmen:
bil position 6 $e
.dundant becau:t
ments except
transmit l2-b;tcode
receive l2-bitcode
se=:-
) birs.
error. I:r
rhe closest PC\l
_ession
1000000 lo(xml0
I in the receir:
rlrrs in the opp Ien the two
s 0 0 0 01 0 0 0 0 0
s 0 0 0 01 0 0 0 0'1
00000000010
1000010
6+
ert!-l
r .ombined er::.
is ihe maxim=
'. lhe worst po.i_
'rrrst significan: J
and
66
0
0
100
100:3.03%
tbr segment 7
code s 1 0 0 0 0 0 0 0 0 0
12-bitcode s 1 0 0 0 0 1 0 0 0 0
transmit l2-bit
rcceive
0
O
00000100000
t0000000000
01v
102,1
111
111
naated
+ll'rr:
-O
aomptesslor
I00
100
l.o l u.,
The PCM coding and decoding processes described in the preceding secrions \\cre concemed primarily with reproducing wavefbrms as accurately as possible. The precise nature
of the waveform was unimportant as long as it occupied the voice-band t-equency range.
When digitizing speech signa)s only. special voice encoders/decoders called yocoders are
often used. To achieve acceptable speech communications, the short-term power spectrunt
of the speech information is all that rnust be preserved. The human ear is relatiYelf insensitive to the phase relationship between individual fiequency components rvithin a r oice
waveform. Therefore, vocoders are designed to reproduce only the short-tern] po!\ er spectrum, and the decoded time waveforms otien only vaguely resemble the original input signal. Vocoders cannot be used in applications where trnalog signals other thln r oice are pre,.ent, such as output signals from voice-band data modenls. Vocoders rl,picallv produce
unnatural sounding speech and. therefore. are generally used for recorded information.
such as "wrong number" messages. encrypted voice fbr transmission over analog telephone
circuits, computer output signals, and educational games.
el-r
tal compressi!-E
100
1056
VOCODERS
<
10000100000
As Example 6-7 shows. the maximum magnitude of error is higher frrr segment 7;
however, the maximum percentage error is the same lbr segments 2 through 7. Consequently. the maximum SQR degradation is the same for each segment.
Although there are several ways in which the l2-bir-to-8-bit compression and 8-bir
to-12-bit expansion can be accomplished with hardware. the simplest and most economical method is with a lookup table in ROM (read-only memory).
Essentially every function pedormed by a PCM encoder and decoder is now accomplished with a single integrated-circuit chip called a corlec. Most of the more recently developed codecs are called.o,rDo chips. as they include an antitliasing (bandpass) filter, a
sample-and-hold circuit, and an analog-to-digital conyener in the tmnsmit section and a
digital-to-analog converter, a hold circuit, and a bandpass filter in the receire secrion.
\xx
o o
It,t)f)
0000 t 00000
16-::
:al Transmission
303
6-1O-1 ChannelVocoders
The first channel vocoder was developed by Homer Dudley in 1928. Dudley's vocc'.t
compressed conventional speech waveforms into an analog signal with a total bandu i;
of approximately 300 Hz. Present-day digital vocoders operate at less than 2 kbps. Di-r:-.
channel vocoders use bandpass filters to separate the speech waveform into narrower si:bands. Each subband is full-wave rectified, filtered, and then digitally encoded. The 3:coded signal is transmitted to the destination receiver, where it is decoded. Generally spe.r'
ing, the quality of the signal at the output of a vocoder is quite poor However, some of ',I
more advanced channel vocoders operate at 2400 bps and can produce a highly intelligib,'-
6-1O-2 FormantVocodens
A tbrmant vocoder takes advantage of the fact that the shofi{erm spectral density of t)i
cal speech signals seldom dist butes uniformly across the entire voice-band spectr-l
(300 Hz to 3000 Hz). Instead, the spectral power of most speech energy concentrate: i
three or four peak frequencies called/ormanrs. A formant vocoder simply determines -r
location of these peaks and encodes and transmits only the information with the most !.:nificant short-term components. Therefore, formant vocoders can operate at lower bit ra=
and, thus. require narrower bandwidths. Formant vocoders sometimes have trouble tra;r
ing changes in the formants. However, once the formants have been identified, a form=
vocoder can transler intelligible speech at less than 1000 bps.
6-10-3
A linear predictive coder extracts the most significant portions of speech informat: -,r
directly liom the time waveform rather than from the frequency spectrum as with'jc
channel and lbrmant vocoders. A linear predictive coder produces a time-Ya4:::
model of the yocal tract excitation ald transfer function directly from the speech u a. :form. At the receive end, a s)'nthesizer reproduces the speech by passing the specit.::
excitation through a mathematical model of the vocal tract. Linear predictive cod-,
provide more natural sounding speech than either the channel or the fbrmant vococLinear predictive coders typically encode and transmit speech at between 1.2 kbps ::2.,1 kbps.
6.1
where
line speed
samples/second
bits/sample
Chapter 6
second
=+
+
sample
16-1.:
.EiE
':
Example 6-8
rh intbrmatir.:
col.
3ded by a
h application.
er, lhe fontu.:
Solution
linc soee,l =
Jiel s vocoj<
rtal band*
i,L
kbps. Digi-;
N ITOWCI JLJ:oded. The e:-
.2
sample is larger or smaller than the previous sample. The algorithm fbr a delta modulation
system is quite sinple. Il the current sample is snaller than the prcvious sample. a logic 0 is
transmitted. lf the current sanple is larger than the previous sarnple. a lo_uic I is transmittecl.
concentrates,r
Figure 6-20 shows a block diagram of a delta modulation transmitter. The input analog is
sampled and con\,erted to a PAM signal. which is compared with the output of the DAC.
The output ofthe DAC is a voltage equal to the regenerated magnitude ofthe previous sarn_
ple. which was stored in the up down countcr as a binary numbcr. The up_down counter is
'.-
e trouble tra.-r-
form:r
ch informati:'r
um as with 'jE
time-varyii:
speech u ar:-
-12.000 bps
t'ied. a
_ 7 bits
" sample
I)elta nrotlulatiort
er. some of LT
secr,nd
:nerally speor-
determines
6000 slmnler
'
l, =
g the specitlx
:dictive
code-.
1 , Analog
'
Oeta rcM
.r
of the PCM
te and the nu=-
{6- 1.:
FIGURE
:el Transmission
6-20
305
FIGURE
6-21
:::is*
signal
Delta PCM
FIGURE
6-22
logiclcondition(+V).indicatingthatthecurentsampleislargerinamplitudethanijE
previous sample. On the next clock pulse. the up-down counter is incremented to a c' -r
of L The DAi now outputs a voltage equal to the magnitude of the minimum step size ::olution). The steps change value at a rate equal to the clock frequency (sample rate) C'':sequently. with the input signal shown, the up-down counter follows the input analog ' ;nal up until the output of the DAC exceeds the analog sample: then the up-down cou: =
wi]lbegincountingdownuntiltheoutputoftheDACdropsbelowthesampleamplitur
In the idealized situation (shown in Figure 6-21)' the DAC output follows the inpur sig-Each time the updown counter is incremented, a logic I is transmitted. and each time up ilown counter is decremented. a logic 0 is transmitted'
6-'1 2-2 Belta Modulation Receiver
Figure 6-22 shows the block diagram of a delta modulation receiYer. As you can see. lh. =
+
ceiver is almost identical to the transmitter except for the comparator As the logic 1s ar:
arc received. the up-down counteris incremented ordecremented accordingly. Conseque::'
the ourput of the DAC in the decoder is identical to the output ofthe DAC in the transmr:=
With delta modulation, each sanrple requires the transmission of only one bit: tL:-fbre. the bit mtes associated with delta modulation are lower than conventional PCI\I . tems. Howevcr. there are two problems associated with delta modulation that do not L\ --E
with conventional PCNI: slope overload and granular noise'
irr
greater than the delta modulator can maintain and is called slope overload. Increasins
clockfrequencyreducestheprobabilityofslopeoverloadoccurring.Anotherwayto:':.
vent slope overload is to increase the magnitude of the minimum step size'
3OE
Chapter 6
_:
Anslog input
DAC output
Ste6p slop,
rapid chang
FIGUBE
6-23
Original signal
FIGURE
Grcnular noise
6-12-2-2 Granular noise. Figure 6-24 contrasts the original and reconstructed signals associated with a delta modulation system. lt can be seen that when the original analog input signal has a relatively constant amplitude, the reconstructed signal has variations
that were not present in the original signal. This is called granular noise. Granular noise in
delta modulation is analogous to quantization noise in conventional PCM.
Granular noise can be reduced by decreasing the step size. Therefore. to reduce the
granular noise, a small resolution is needed, and to reduce the possibility of slope overload
occurring, a large resolution is required. Obviously, a compromise is necessary.
Granular noise is more preyalent in analog signals that have gradual slopes and whose
amplitudes vary only a small amount. Slope overload is more prevalent in analog signals
that have steep slopes or whose amplitudes vary rapidly.
6-24
(r'-
.I3
re up-down coun:=
e samPle amPlitu';:
Adaptive delta moduLdtion rs a delta modulation system where the step size of the DAC is
automatically varied, depending on the amplitude characteristics ofthe analog input signal.
Figure 6-25 shows how an adaptive delta modulator works. When the output of the transmitter is a string of consecutive I s or 0s, this i ndicates that the slope of the DAC ou tput is
rf only
Origjnal
enalog
FIGURE
:p size.
ial Transmission
6-25
347
less than the slope of the analog signal in either the positive or the negative dire(tion
Essentially. the DAC has lost track of exaclly where the analog samples are' and the
possibility ol slope ovcrload occurring is high. With an adaptive delta modulator. after
a predeternincd nurnber of consccutive 1s or 0s. the step siTe is automatically increased. Aftel the next sample. if the DAC output amplitude is still below the sample
amplitude. ihe next step is increased even further until elentuall) the DAC catches up
u,ith the analog signal. When an alternativc sequence of 1s and 0s is occurring. this in-
dicates that the possibilit) of granular noise clccurring is high. Consequently. the DAC
will automaticall), rcr,ert to its mininrun step size and. thus. reduce the magnitude oi
the noise error.
A contmon algorithnr tbr In adaptive deltil modulttor is when three consecutjvc ls or
0s occur.. thc step sizc of the DAC is incrcased or decreased b), a lactor of I .5. Various othe:
algorithms may be useci tbr adapti\e delta modulators. depending on particular system re'
quirements
6-14
DIFFEHENTIAL PCM
In a typical PCM-encoded speech var elbrn. there are oflen successive sanrples taken i:
uhich there is little ditterence between the amplitudes ofthe t*tr samples. This necessitate'
pulse code moctr ansmitting ser eral identical PCM codes. rvhich is redundant. Ditferential
redursample
to-sample
of
the
to
lake
advantage
(DPCM)
specifically
is
designed
ulation
of t$
amplitude
in
the
ditterence
With
DPCM.
the
speech
wavetirrms.
dancies in typical
successive sarnples is Iransntitted rather than the actual sample. Because the range of san:'
ple dittclences is tl pically lcss than the range of individual samples. l'erver bits are require:
.5
PL
Figure 6-26 shows a simpJiticd block diagram ofa DPCM transmitler. The analog i:put signal is bandlimited to one-half thc sample rlte. then compared with the preceding 3'cunlulated signal !e\el in the difttrentiato|.. The output olthe dilfcrentiation is the ditGren..
between the t\\'o signals. The ditlcrence is PCM encoded and transmitted. The ADC ope:'
ates the san'ie as in a conventional PCM system. except that it typically uses fiwel bits p':
sample.
Figure 6-27 shovs a simplificd btock diagram of a DPCM receiver' Each receir;:
sample is converted back to lnalog. stored. and then sumn]ed u'ith the next sample r''
ccived. In the receivel shorvn in Figure 6-27. the integration is pertbrmed on the analog !i:nals. although it could also be perlirrmed digitally.
r1I_
F GURE
6-26
DPCM fiansmtte.
Chapter 6
k:e]
Tna
tl ll
.\ e direction
,!raticallY in-
the samPle
\C catches
uF
IU-LTL
Parallel
DPCM
Serial
DPCM
rn
T-----------I-+
I Seflal L__ | Dig'tal to
>| to parallel L _| anatog l +
I converter
I converter
l
,n.e!-utive
Is
Adder
+
(inlegrato0
+
t\-r-r-
(':
:. \ arious oth.:
Sum
signal
out
l"
.Lmples taken
::
FIGUBE
6-27
DPCM receiver
Ihis necessitat<'
,1
lequiri:
PULSE TRANSMISSION
All digital cader systems involve the transmission of pulses through a medium with a finite bandwidth. A highly selective system would require a large number of filter sections,
which is impractical. Therefore, practical digital systems generally utilize filters with
bandwidths that are approximately 307o or more in excess of the ideal Nyquist bandwidth.
Figure 6-28a shows the typical output waveform from a bandlit ited communications
channel when a narrow pulse is applied to its input. The ligure shows rhat bandlimiting a
pulse causes the energy from the pulse to be spread over a significantly longer time in the
form of secondar-v- lobes. The secondary lobes are called ringing toils. The output frequency spectrum corresponding to a rectangular pulse is referred to as a (sin J)/r response
and is given as
The analog i:
ihe preceding r'
er'.
differen:'
.1. The ADC oP':'
,n is the
f\@)
sin(orl2)
: \r\-;;-:
(6- l.l
where o = 2rl(radians)
Figure 6-28b shows the distribution of the total spectrum power. It can be seen that
approximately 907o of the signal power is contained within rhe first .?eclla I null (i.e.. f =
l/I). Therefore, the signal can be confined to a bandwidth,B = 1lT and still pass most of
the energy from the original waveform. In theory, only the amplitude at the middle of each
pulse interval needs to be preserved. Therefore, if the bandwidrh is confined to B = 1/27.
the maximum signaling rate achievable through a Jow-pass filter with a specified bandwidth without causing excessive distortion is given as the Nyquist rate and is equal to twice
the bandwidth. Mathematically, the Nyquist rate is
R:28
where
tal Transmission
(6-15)
309
(t)
,/
Sacond pul36
lO
Socondary lobes
t
I
I
I
t
t
I
I
cutoff
frequency
Sgcond6ry lobes
i\
-37
-27
3T
21
-T
.{
trequoncy
{b)
of a
bandlimited filteri
Chapten 6
figital Trans
lO
4T
Sampling intarv.ls
- ted filter;
FIGUBE
one source are multiplexed together, the amplitude, frequency, and phase responses become
even more critical. ISI causes crossrall between channels that occupy adjacent time slots
in a time-division-multiplexed carrier system. Special filters called, equalizers are inserted
in the transmission path to "equalize" the distortion for all frequencies, creating a uniform
tuansmission medium and reducing transmission impairments. The four primary causes of
plitude distortior'
!e at Precisely thi
rrect (which in re, n in Figure 6-29.
rr always attain th:
ISI are
as
follows:
1. Timing inaccuracieJ. In digital transmission systems, transmitter timing inaccuracies cause intersymbol interference if the rate of transmission does not conform to the
ringing frequency designed into the communications channel. Generally. timins inaccura_
cies of this type are inrignificant. Because receiver clocking inlormarion i. deived from
the received signals, which are contaminated with noise, inaccurate sample timing is more
likely to occur in receivers than in transmitters.
2. lnsufficient bandwidth, Timing errors are less Iikely to occur if the transmission
rate is well below the channel bandwidth (i.e., the Nyquist bandwidth is significantly be_
low the channel bandwidth). As the bandwidth ofa communications channel is reducetl, the
ringing frequency is reduced, and intersymbol interference is more likely to occur
u: interfering \lir
rrg) in the form (':
]ppears during thr
.ommonly calle'
,.n in the transmi:'
rf,ie response. sir'
igital Transmission
311
channel to bandl:-'r
3. Attqlitu(le distotrior. Filters are placed in a communications
are also used to : Filters
interf'erence
r';Jo.. or eliminate predicted noise and
.lgnrf,
a
channel canno: 'of
"n;
response
ar'aa ,p"aina pulse response. HoweYer, the frequency
"i"-pr"J"[a absolutely When the frequency characteristics of a communi':at :'
palse distorlion results Pulse di:: ''
"ryll
c#nnel aepan f.om the normal or expectetl values'
ringing frequencie. :
ti* o".o,, *tl.n tt'e peaks of pulses are reduced, causing improper
equali:atktr:
amplitude
is
called
it. ,irn" ao-uin. compensation for such impairments
of harmonic-'
series
a
of
1, Phase distttriion A pulse is simply the superposition
if the r:relationships'.Therefore'
specific amplitude and phase
lhir:-
5-2
EYe Patterns
of: -:.
of a digital transmission system depends' in part' on the ability
p"rfnr^on..
ih.
pr'r-'
.,.",".,o ,.n.n.rr,. the original pul'es. similarl)' the quxlit) o[ lhe regeneration ir:rir
the
at
signal
quality
ofthe
the
J..ision ciriuit within the repeater and
i.o."a.
ca: .*
"rii.
Therefore, the performance of a digital tansmission system
.i..uit.
d..i.ion
io ih.
5-1
b's
superimposed over adjacent sig:-:at the data iate. ihu.. ull *ur.fo.- .,r-binations are
pattem is 3 ;
ete
irg-i;,"tt;s. S*ll display is called an e-r'e pattern or dicrgr"lr? An.eye into the pu ':"
introduced
degradations
'3
u."ni"ni,".t niqr. for determining the effects ofthe
eye patteJn is shown in Fi:--:
,fr"y ,r-"f ," ,fr" ,.gan"ruto. Th" t"'t setuP to display an
",
input ofthe oscilloscope' and the s1::O-:0. it. r"."i""a pu[e stream is fed to the vertical
rate is set aPproximatel)' e' -r
hol clock is f-etl to the extemal trigger inPut, while the sweep
;;;;;i;
rt" .yJopening lthe areain the middle ofthe eye pattem) defines a bounc--'
code-pattern condition The : ':
"qouriiutiirn.
w'ithin which no wav eform tmiectories can exist un<ier any
r\
G.:
-
input
Osailloscopo
FIGUBE 6
30
EYe
surement setuP
J l<
Chapter 6
diagram mea-
rl to bandlimil
!o used to pronel cannot al-
mmunication\
-L\
. Pulse distor-
irequencies in
Zo?o
ruali:ation.
i harmonicalll
rre. if the rela-
crossings
I
.
l/
'il
I 0ccurs. Phase
cross hairs
are Placed in
90% opening
Oocision time
of time delal
rr
Decision level
distoradjust them'
re phase
li
ability of
a re'
Sampllng instant
0
eration Proces!
:nal at the inpu:
1 \) stem can tE
'rS
FIGURE
6-31
Eye diagram
rdjacent signal-
opening is a function of the number of code levels and thc intersymbol interlerence caused
by the ringing tails ofany preceding or succeeding pulses. To regenerate the pulse sequence
without error, the eye must be open (i.e., a decision area must exist). and the decision
cnrsshairs must be within the open area. The el'fect of pulse degladation is a reduction in the
size of the ideal eye. In Figure 6-31. it can be seen that at the center of the eye (i.e., the sampling instant) the opening is about 907r. indicating only minor ISI degradation due to filtering imperlections. The small degradation is due to the nonideal Nyquist amplitude and
phase characteristics of the transmission systen. MatheInatically. the ISI degradation is
.hown in Figur'=
pe. and the sym-
ISI
where
Le
:
ft :
11
20 log
(6-l6r
l0 loe
q0
,O"O
In Figure G31. it can also be seen that the overlapping signal pattern does not crrx. the
horizontal zero line atexact integer muitiples ofthe symbolclock. This is an impaimrenr kn()\\'n
asdotatft,:nsitiotljiterThisjitterhasaneffectonthesymboltiming(clock)recorerr.Lreuir
and. ifexcessive. may significantly degrade the pedbmance ofcascaded regeneratir e .ectionr.
j.15
::arn mea-
rital Tnansmission
313
a>-<
(a)
IL--L
"j[L
l-< T >l
ru,
T
FIGURE
lal
power can be averaged over an entire message duration, and the signal can be modei:: l'
a continuous sequence of alternating 1s and 0s as shown in Figure 6-32. Figure G'1
shows a stream of rectan-qularly shaped pulses with a pulse width-to-pulse duration r:r
t/Iless than 0.5, and Figure 6-32b shows a stream ol square wave pulses with a t/f -::r
of 0.5.
The normalized (R - t) average power is derived for signalflt) from
F=
where f is the period of integration.
tion 6- l7 reduces to
[' 7i,.11',i,
li.+
Ifft)
: +[''t,1r;1',/r
j\tJ
r,:
h.
V with a
t/f
6'-{r
0, then
(v o<r=r
. :10
t<t<T
'1I)
Thus, from Equation 6-1li,
|
n' ru f'(vf dr: +vri,
),,
I ( lv
;'
and
Lv'
\r/R
= (y*,)2/R.
E
v.". 1r(v)
xFrTvl
- rry-) (rR).
With the square wave shown in Figure 6 -32. xlT : 0.5. therefore, P :
Becau,'e P
= {y,,. ):
R.
]a,
f'.:
the rms voltage for Jhe square wave is the same as for sine waves, V"..,
Chapter 6
the
V)
l2R.T:
= Vl\/2.
:,
]UESTIONS
lhe aJr:rnlrpes and di\aJ\ anrage. oi digitrl lrunsrnt)\ion
\!{Contrart
, o-:r.'U hat arc the lour mo\l common methnd\ ol pulse modulalton l
7r-1. Which method listed in quesrion 6-2 is the only fbrm of pulse modulation that is used in a diSittl transmission system'l Explain.
What is the purpose of the sample-and hold circuit?
Dcfine aperturc auJ acquisitiott tine.
argital signals:
= o.5
G6. What
( !jl!)
I can be modeled a.
Whal
i.
rhe
\)quisl
.amplrng rate',
l/Iratii
6- l
l.
Explain quantizing.
tiom
09
(6'l-
Explain the relationship between dynamic range, resolution. and the numberolbits ill a pCM
code.
6-11. What is SQRI What is the reladonship between SQR, resolution. dynamic rangc, and the num
- ber of bitr in a PCM codel
Equa-
X@raont,,',
\lyp.tExpltin
(6-lt
ratt=0,then
@.]hu,
16-le
6-1.1. What is the cffect of digital compression on SQR, resolution, quantization interval. and quantization noise l
6-:5.
({fi
oa,n",tn1,
\-6i7.
Whar
i\
ott
rt,rtJ
modulationl
']OBLEMS
=
(6-21
re.F = y2/2R.Thu!
-.:
t,
6- l
2$D 6-f.
Z!, \
6-.1, For a sample rate of 20 kHz, derermine the maximum analog input frequency.
a$1, 6-1.
vl\/2.
2S[
- : tal Transmission
Determine rhe Nyquist sample rate tbr a maximum analog input frequency of
a. 4 kHz.
.^ -r " '14i1{r1.
h. l0 kHz.
For the sample-and-hold circuit shown in Pigure 6,5a, determine the largest-value ca,. ..
that can be used. Use the lbllowing paranreters: an oulpur impedance for Zr : 20 O. at. , sistance of Q, of 20 0, an acquisition time of l0 Us. a maximum outpnt current fror 1
20 mA, and an accuracy ol l%.
Determine rhe alias tiequency for a l4,kHz sample rate and an analog input frequenc) of
kHr
315
6-6. Determine the minimum numher of bits required in a PCM code for
dl namic range :
'
;-
a.0ll0l0l
h. 00{)00I
c. 1000001
d.0lllrrr
e. I000000
6-ll. Determine
of0-2v.
6-9. Determine the resolution and quantiTation eror tbr an eight-bit linear sign-magnitude P:|t
code for a maxinlum decoded voltage of 1.27 V
6-l{). A l2-bit linear PCM code is digilall} compressed into eight bits. The resolution : 0.03 \ ,r
termine the following for an analog input voltage
ol
1..165
V:
a. l2
c. Decoded l2-bit
d.
e.
6-l
l.
code
Decoded !oltagc
Percentage errof
a. 100000000001
b.000000000000
c. 110000000000
d.010000000000
e. 100100000001
6-ll.
I: l0l010l0l0t0
For each ot' the fbllowing 1 2 bit linear PCM codes. determine lhe eighl-bit compressed ci
\\'hich they would be converted:
:j
a. 100000001000
h. 100000001001
c. 100000010000
d. 000000100000
e. 010000000000
f: 010000100000
6-1.1. Determine thc Nyquist sampling rate for the following maximum analog input frequen.::
kHz.5 kHz. l2 kHz- and 20 kHz.
/r :
6-15. Determine the maximum analog input frequency fbr the fbllowing Nyquist sample ratet --:
kHz- .1 kHz- 9 kHz. and I I kHz.
6-16. Dctcrmine the alias frequency for the following sample rates and analog input tiequenci..
.4,
(kHz)
/, (kHz)
.l
n:
6- I
3.1
ll.
7- 8- 12. and
Dctermine lhe nrinimum number of bits required ibr PCM codes with the fbllowing
ranges and determine the coding etTiciencies: DR : 24 dB.48 dB. and 72 dB.
Chapter 6
.i.-.,
1,1.
dy
r. ;
ri.
range of
ym\
!
.eIen-bit si-s:'
(v)
r/,,,
(v)
255
0.75
100
0.75
0.5
255
6-20. For the iollowing resolutions. determine the range of the eight-bit sign magnitude PCM
Code
t0l
I I 000
0.1
]1.ignitude PC\l
{nl
I1000
0.1
.r =
00011100
001 l0 r0l
Illulll
0.03 V. D:-
codes:
Re\olution (V)
0.05
0.01
0.01
0.01
0.02
1l 100000
000001I l
6-21. Determine the SQR for the folbwing input signal and quantitarion noise magnitudes:
v,
t;, (vl
001
0.02
3 r,,,,.
0.01
.1r
0.:
6-22. Detemine the resolution and quantization noise for an eight-bit linear sign-magnitude PCM code
for the following maximum decoded voltages: y,,,". = 3.06 Vn. 3.57 Vp, ,1.08 Vp. and 4.59 Vp.
6-2-1.
.rnpressed coda
6.24. For the l2-bit linear PCM codes given. determine the voltage mnge that would be converted k)
them:
l:-Bil
Linear Code
l000l
i:u! tiequencia:
l00 t0
0.12
0.10
00000 r000001)
{x)0 ]l
l u
r:l
\ illue caPi._art
,). ln on resiliil:
l t {l{10
I0000
0.l,t
0 t?
6-25. For the following l2-bit linear PCM codes. determine the eight bit compressed code to which
thev would be converted:
rrZr of 10m\ t
l2 Bii Linear
Code
10001111{nl0
00000 I 000000
rtul t'requencle!
0001r1l11000
llllll|0010
000000 I 00000
6-26. For the tbllowing eight-bit compressed codes. determine the expanded l2 bit code.
Eight-Bit Code
I
tiilude PC\l
l00l0l0
00010010
i"=
l0l0 t0l0
0i
0lt)
101
I I I I 0000
ll0il0ll
:31
Transmission
317