1966

Generalization of a Technique for Binary Data

Communication where the k's are integer weighting coefficients, with the smallest
E. R. I(RETZMER, SENIOR MEMBER, IEEE b u a l to unity.
The corresponding channel frequency function is
Abstract-A technique forbinary data transmission isdes-
cribed, in which each binary symbol is chosen to be a prescribed
superposition of n impulses of form (sin Z?rFf)/&Ft, spaced at
intervals l/ZF. Such superposition leads to more than two received
levels with binaryinput, but in return permits realization of the
Nyquist rate (2F symbols/sin bandwidth F). Appropriate choice
+ k,6 t -I)'
( - "2i e-j2,ft d t (2)

of the superposition coefficients provides a variety of spectral [O < ( f j < F ] ; the band limitation permits the use of the delta
distributions to suit individualapplications. The most interesting function instead of (sin 2 a F t ) / 2 ~ F t .
classes of system functions are defined, the associated coding With these two characterizations in mind we can now tabulate
procedures are described, and the performance characteristics thosechannels which have been found of greatest interest (see
aresummari2:ed. Table I).
Each row of this table defines just one member of a class, uamely
the one having the smallest possible number n of k's and hence of
received levels. Thus, Class 1 is the polybinary class of systems [4]
INTRODUCTION having all unity-weight coefficients, while Class 2 has only linearly
A useful generalization of the so-called biternary, duobinary, or tapered distributions of coefficient values. As n is increased within
polybinary [1]-[GI techniquefor binarydata transmissionis re- each class, the number of received levels also rises and H(f) becomes
ported herewith. It is based on recognizing that one may choose increasingly "concentrated," yielding a more drastic spectral
an end-to-end system function H(f) whose transform h(t), the tapering. However, the full band F is still required to convey 2F
overall response per binary symbol, is a prescribed weighted linear bps, and performance margins, aresubstantially reduced. To
superposition of n impulses of form (sin 2 ~ F t ) / 2 ~ Fspaced
t, at inter- ensure the essential condition of continuity in H ( f ) a t f = F , certain
vals 1/2F. Use of such a superposition rather than a single impulse superpositions must be ruled out, e.g., odd values of n in Class 1.
leads to more than two received levels with binary input, but in
return it renders practical the attainment of the Nyquist rate [7] OR PRECODING
DECODING
( 2 F symbols/s in bandwidth F ) , since H(f) is zero for f 2 F and is The received multilevel signal can be interpreted with s digital
also continuous a t f = F . Resulting constraints on level transitions version of McColl's apparatus [SI, in which the contributions of
inthe received signal make possible a limited amount of error ?z - 1 preceding (transmitted) sample values-all binary in our
detection. Furthermore,appropriate choice of the superposition case-are subtracted from the present received samplevalue. A
coefficients makes availablea variety of spectral distributions to binary decision is then made on the difference, which is either 0 or
suit individualapplications. This communication will define the kl in the absence of noise. A shift register stores the resulting binary
most interesting classes of system functions (channels), describe the digit as well as the n - 1 preceding ones (see Fig. 1 ) . An error in
associated coding procedures, and summarize the performance any received sample tends to propagate until e, reaches the top or
characteristics. bottom level.
Alternatively, a precoding operationcan be undertaken a t the
DEFINITION
transmitter, such that the nth binary digit an of the original sequence
The transmission channel will be characterized by the super- is the modulo-2 summation of the nthbinary digit b, actually trans-
position relationshipmentioned: A received samplevalue cn de- mitted and the n - 1 preceding digits t,ransmitted, all weighted by
pends on n successive transmitted sample values b ~ . ..b,, as follows: their assigned coefficients, i.e.,

TABLE I

Binary 1
(ideal)
1 1 1
(n = 2)
2 1 2 1
(n = 3)

* Entry in figure is area under curve.

2- 0 -I 0 --I
Manuscript received August 6, 1965; revised October 6, 1965. This paper
is a condensation of oneentitled "Binary data communication by partial
response transmission, which was presented as CP65-419 at the 1965 IEEE
Communications Convention, Boulder, Colo.
The author is with Bell Telephone Labs.. Inc., Holmdel, N. J.

Authorized licensed use limited to: Brookhaven National Laboratory. Downloaded on December 08,2023 at 21:44:11 UTC from IEEE Xplore. Restrictions apply.
 68 FEBRUAIlY

Cn SAMPLE RECEIVED - =oic, -Sn=O) TABLE I1

~-
BINARY ~

Speed
Tolerance S/N Increase
kl k , k3 kq kg (in percent) Req. a 2t F b / s
”an 3
,.’ ZIdeal 1

1 1 \, 43 ! ,
_,

( n = 2)
1 2 1 ‘40 6.0dB I -1-
* ,
( n = 3)
2 1 -1 1.2dB . ’
7 . 2 dB (precoded)
1 0 -1 dB 2.1 15
( n = 3) <
5 -1 0 2 0 -1

Fig. 1. Decoder.
’,
r. Cousequently the ratio of the area under H(f) relative
a, (UNCODED
INPUT) area
the to for
binary
the reference system gives a measure
direct
se in signal power. The area for each m ( f ) is
able I. The “decision distance” is unity inall cases,
for Class 3 when precodine is not used, since kl = 2,

eparhres from the prescribed transfer function.

*NOTEIFOR kl E V E N , bn-1 REPLACES bn. ETC

irtues of the channels described are mainly that they afford
Fig. 2. Precoder.
simple means of attainiug 2F bps in bandwidth F with
utoffs, andthat one can choose from a Q‘
tapered spectral distributions which may be
Comparing this with (I), we find that dvantageous
in specific applications. Instead of the basic (sin?
a, cn mod 2,
i.e., an even-numbered value of the received sample implies a zero
from F to 2F.
- i u the original sequence, an odd-numbered value a one. The
coding operation thus simplifies the interpretatio~l of the received
signal aud eliminates error propagation; on the other hand, it& l m&s-
u

ACKNOWLEDGMENT

- the decision distance” (differerlceiII signal values to be distin uished

by detector) to ullity instead of klPa matter
F. K.
consequeIlb
0.
The author acknowledges the contributious of
15.
Becker aud
Burton in clarifying some of the preceding concepts.
- -
where kl exceeds unity, as in ClZi3. The precoding implementation
appears in Fig. 2. REFERENCES
[ l ] A. A . Cloakey, “Automatic printing equipment for long loaded submarine
PERFORMANCE
. telephone cable::” Bell Sys. Tech. J . , vol. 6, p. 402, July 1927.
[2] A. P. Brogle, Anewtransmissionmethodforpulse-codemodulation
Performance is appraised by two criteria: communicationsystems, IRETrans.onCommunicationsSystems, vol.
CS-8, pp. 155-160:,September 1960.
1) the speed tolerance, i.e., the percentage increase Over the [3] 0. E. Ringelhaan,Systemfortransmission of binaryinformation a t twice
the normal ‘;?te,” U. S. Patent 3 162 724, December 1964.
Nyquist rate which will just cause overlap between adjacent [41 A. Lender,Theduobinarytechniqueforhigh-speeddatatransmission,”
IEEETrans.onCommunicatio:andElectronics, vol.82, pp. 214-218.
levels (“eye closure”); and M a y 1963.Seealso A. Lender,Correlativedigitalcommunicationtech-
2) the required increase in signal-to-noise ratio over a reference niques,” I B E E Trans.onCommunicationTechnology, vol.COM-12, pp.
128-135, Decemier 1964.
binarysystem (operating a t like bit rate) for a fixed error 151 R: D. Hpwson, An analysis of thecapabilities of polybinarydatatrans-
mlsslon, IEEE Trans.on
Communication Technology, vol. COhI-13,
rate in the presence of white Gaussian noise (ignoring both pp. 312-319, September 1965.
+.[GI E. R. Kretsmer,“Anefficientbinarydatatransmissionsystem,” IEEE
error propagation and error detection capability). Trans. o n Communications Systems (Correspondence), vol. CS-12, pp. 250-
251, June 199$.
Criterion 1 has been evaluated [9] by computer
---- simulation using [71 H. Nyquist,Certaintopicsintelegraphtransmissiontheory,”
A I E E , vol. 47,,,pp. 617-644, April 1928.
Trans.
“worst possible” data sequences (see Table 11). Criterion 2 is [8] L. A. McColl,Signallingmethodandapparatus,” s. Patent 2056 284,u.
October 1936.
readily formulatedinterms of the “decision distance” and the [yl w. R. Bennett and J . R, Da,.ey, DataTransmission, New York: McCraw-
“noise bandwidth” of [H(f)]1’2.The exponent 112 arises from Z 9 Hill,1965,ch. 7. S y also, S. Habib,internalcommunication.
101 R. L. Townsend,Aclass of partialresponsesystems,”internal com-
geometrically equal apportionment of H(f) between transmitter munication.
..
, .. .
1 , . ~

. ,: , . hk
r-
’ - , . \ ~ \ r \, ((-; c , - 11-1, ;
ii II >
1
I ,
-. ,; ,I , , ,
. ’ .I.. <-,\. ’. - J
i
r-- 1 c A’,
’
- I .
--
I L+\’ i
A S

,
1

,.,>
i 1 b’

<j
,

,I
’1 ,
’ LC L C

I
’, ._ I , I ’”(

Authorized licensed use limited to: Brookhaven National Laboratory. Downloaded on December 08,2023 at 21:44:11 UTC from IEEE Xplore. Restrictions apply.
15 ,b?, 5 LJL&
Y

!kLjjcLL ,c .e
k:-c 3