The Origins of the Sampling Theorem
Hans Dieter Liske, Aachen University of Technology
ABSTRACT
Fifty years ago the publications of Claude €, Shannon brought
the samping theorem tothe broad atention of communication
engineers. This article demonstrates how practcians, theoreticians, and mathematicians
‘scovered the impleations ofthe sampling theorem almost independent of one sates
18 1948 and 1949, Claude E. Shannon published
in which he founded
the intr
formulated ast
fo=3%,
oe
xg}
was not until these papers were published thatthe theo=
rem known as "Shannon's sampling theorem” became com-
mon property among communication engineers, although
Shanon himself writes in [2] that
This fact which is common Knowledge in the comme
A few lines further on, however, he adds
but in spite of its evident importance {it} seems not t0
fave appeared explicitly the lerature of communication
theory
‘The following analysis takes the above statement as its
starting point. I will become apparent that mathematicians,
practiians, and theoreticians in communication engineering
Exme across the implications of the sampling theorem almost
independent of one another, and that the Tinks between them
dil aot emerge unt later stages ofthis development
‘THE PRACTICIANS
1m communication engineering, the frst experiments with
time-division multiplexing (TDM) in telephony led to the
{questions of how and how often it necessary to sample a
‘Sontinuows-time signal
The aempt to transmit mor
single wire hogan shorly after the ealy commercial suc-
‘esses with telegraph in the 1840s. The fist proposals for TDM
Using synchronously rotating commutators detive from F.C
Bakewell (1848), A. V. Newton (1851), and M. B. Farmer
(1853). Technically more accomplished methods were then
developed by B, Meyer (1870), JM. E. Baudot ((0 1874), as
well a P. Lacout and P.B. Delany (1878) [3.4]. Ieis significant,
ht oly that methods were used in which complete telegraphic
Signals from different transmitters were placed in chronological
fdr (eg. Baudot, but that certain systems were ako equipped
‘sith fas rotating cooumutators which were able to transmit at
Teast ovo samples of each elementary signal (eg, Delany). This
technique mikes akliional sachroniation between tansmiter
dori Lsn 1. Rely
and sampler unnecessary, One of these
fast rotating commutators, the “distribu
tor” of the telegraph system by F.1
Patten (around 1891), was used for the
first demonstration of TDM of telephone signals. The inventor's
name was Willard M, Miner. He had his method patented in
1903 following many years of preliminary experiments [S]-Fig-
ures Land 2 from [6] show the ircuitchagram and the "Patten
Distributor” which was used. Miner determined the required
sampling rate experimentally [6
1 ill be understood, then, that the apparatus devised by
Mr. Miner, while im its general form the same as that
heretofore wsed for maitiplex telegraphs — or telephony for
‘that matter — such apparatus 1s rum ai a much greater
speed 50 as to bring the frequency of the closures of con
rection upon the several Branches or subscirulls up fo @
‘ate approximating in greater or les degree the rate of the
‘Wbrations ofthe overtones charactenzing speech, A rte of
‘closure of 1,000 or 2,000 per secon: wil not answer the
‘purpose, but as the rate increases and passes beyond
3,000, improved reslts become apparent nd are marked
1p better when a rate of 3.300 or 3,600 per second is
reached; the best results being obtained with a rate of
‘about 4,300 per second.
Miner thus assumed thatthe sampling rate would coincide
approximately with the upper frequency components of
speech. In actuality, his telephone apparatus will have had
utolf frequency of barely more than 2 KHz which fulfills the
requirements ofthe sampling theorem.
Since a theoretical clarification of the sampling process was
not fortheoming, pronouncements concerning the sampling
Tate in publications as well as patent applications for TDM of
Speech signals remained similarly vague right up until the
1930s, For example, L, von Kramotin in 1923 writes ina
patent on TDM:
therefore its possible to work with a switching speed
‘which lies beyond the limits of audibiley, whereby the
‘withing noise In the individual telephones avovded and
noiseless communication is posible
Figure 1. Cicuir diagram
108 Dies4sny9sin0 0 1990 ERE
TEE Communications Magazine * Api 199In the 1930s several TDM systems for telephony were
developed. However, as Cattermole remarks (7)
The situation about 1936, hen, was that sampling and
TDM telephony were kaown empirically though the theory
was rudimentary
Some authors, suchas M. Matro in 1938, appear to give 100
{ow sampling rates for speech transmission. Marzo uses broad
sampling pulses fora duplex TDM system. Here, the effec must
be taken fato account that ifthe sampling pulses are broadened,
itis possible to reduce the sampling rate and stil achieve the
same word intligiiliy. This dependency was examined quant
tatively by G.A. Miller and J.C. R Lieder; thei result are
shown in Fig 3 [8], According to these resuls, word inteligiili-
ty with a dropping sampling rate is only reduced monotonely for
‘very short sampling impulses (up to a relative width of approx:
‘mately 6 percent ofthe sampling period). For sampling impulses
of greater relative width, by contrast, the intelligibility again
increases in the range of a sampling rate of 10 to 100 He.
‘THE THEORETICIANS
Theoretical communication engineers did not begin working on
the problem of sampling until surprisingly late. H. Nyquist and
KKipfmaller in 1924 proved tat the numberof telegraph sig-
nals which ean be transmitted over alin is proportional to the
product of transmission time and bandwidth. RV. L. Hartley in
1928 generalized this result with respect to multilevel transmis-
sion. Als in that year, Nyquist derived his famous theorem on.
Aistortonless transmission of telegraphic (digital) signals. But
the distortionlss transmission at Nyquist rate and the error-free
interpolation of sampling pulses ofan analog signal ae different
problems, even though there are some mathematical silts.
“Therefore, these works eannot be regarded nx sources for the
sampling theorem, especialy during the 192 and 1930s.
“The frst slentst to formulate the sampling theorem pre-
cisely and apply it to problems of communication engineering
Th his work “On the transmission
capacity of ether nd wie in electrical communications,” pub-
lished in 1933, he proves the sampling theorems for lowpass
‘Signals well as for bandpas signals], Hle uses these theo:
Fems in the course of his work t0 show thatthe bandwidth of
fan analog signal cannot be reduced by modulation methods.
‘The lowpass sampling theorem i formulated as follows
‘THeonem |
Any function F(0) which consists of frequencies from 010 per:
‘dss may be represented bythe following seri:
Figure 2 The "Pater Disributor.”
surement ofthe value of F\) for
Indeed obtain
2 (mis on integer), we
#(g)-20 ®
because all terms ofthe series in Eg, (1) for this value tte 10
zero, withthe exception ofthe term with = n that equals Dy,
hich may easily be established aftr calculation ofthe indefinite
point. Thus, we may recognize the next Dy according to the
progress after each 1/2; 5. If we transmit these Dy one after
‘another, following each 12s, we cam reconsirct F(\)temwise
‘according to Bq. (1) to any degree of precision.
‘Since this noteworthy work has never been published in
internationally acessbl form, the publications on the theoret-
cally exact formulation of the sampling theorem in the ltera-
ture of communication engineering came about independent of
‘one another. Thus, H. Raabe deduced the sampling theorem in
his PRD. thesis and published itn 1939 [10]. This publication
is especially relevant to the practical aplication ofthe sampling
theorem, since its here thatthe influence of sampling impulses
‘of finite duration is taken into account inthe form of "natural
Sampling.” Raabe summarizes his findings:
For the demonsirated conditions of ansmisson, the sam
pling frequency is determined by the range of signal fe
‘quencies. If these are kept below half ofthe sampling
Frequency, all of the noise frequencies remain above this
lit and can be Rept away from the receiver by a lowpass
{ter The transmission ofa signal may thus be completely
ddstortintess, ifthe sampling frequency is twice the highest
Signal frequency. The upper limitation of the signal jre-
100)
wk
h z
where k= bugger 3
ot, 2
By comin which depends on F. :
Convaiany uncon FO) which represented byte | Egg |
sere in Eg (1) ony consis of fequencles pom Doty pert. | 8 |
ote i
: oo
Tone I ° i
ete 080 Tno
“Any function F(t) which consists of the frequencies from 010; 7 Rome ae
maybe manrated continual to ay dete dere of pret. | eerste
Ton wing numbers which follow on wih 12%: Hom the mee: Figure 3. Rew of quntatve cxamnation
{BEE Comnctions Map + Api 199 1‘quencies is therefore a vital condition of distoriontess
fransmscbiy by time division muliplex transmission.
‘This work also contains a special sampling theorem for
bandpass signals. Raabe's work is cited in a related publica-
tion by W. R. Bennett from the year 1941 (11), and Bennct's
work is in turn cited by Shannos in [2] as one of the sources
Of the sampling theorem
Lastly, i should be mentioned thet the sampling theorem
is also treated in 1949 inthe Japanese book Hake Denso (Sig-
nal Transmission) by 1. Someya. Hence, the term “Someya's
Theorem’ may be found in some Japanese literature.
‘THE MATHEMATICIANS
For mathematicians, the sampling theorem is a special theo-
rem from the field of approximation theory. Approximation
theory asks, for example, what functions can be represented
by a linear jum of given basic functions such as algebraic or
trigonometric polynomials and with what approximation error
(One possibie approach is to determine these linear sums such
that they take on at definite points the same values as the
function wich sto be approximated
Ta this sense, sampling theorems make it possible to estab-
lish how this task of interpolation can be solved, especialy for
functions bounded in the frequeney domain with a vanishing
approximation error.
"A fist approach in this ditection was described as early as,
1765 by J. Lagrange. Lagrange determines a linear sum in
hharmonical sinus funetions in such a way that it coincides with
the function which i to be approximated atm equidistant points.
Generalizing this approach, it may be said that knowledge of 2
+ Tequidvtant functional values of one period is sufficient in
order to earesent a periodic function which may be described
by a trigonometric series, each with sine and cosine terms as
well as one constant. This well known theorem may be viewed
4s sampling theorem for bandimited periodic functions,
“The firs. proposal forthe interpolation of equidistant func-
tional values using the single function was published in 1908
by Cl. de la Vallée Poussin in the Bullen Academie Royale
de Belgique. However, the special significance of ths interpola-
tion for bandlimited fonctions isnot yet explored in this work
'E.T. Whittaker’s paper “On the functions which are repre
sented by the expansions of the interpolation theory” in 1915
‘must therefore be regarded as the fist work ever to address the
sampling theorem forall bandlimited functions (12). Whittaker
ddtesses the problem of achieving the smoothest possible intr
polation without singularities nd without “rapid oscillations
for given tbular values of «function fo). in order to full the
latter condition, be shows that under certain conditions itis pos-
sible to interpolate given sampling values at intervals of w such
thatthe Fourier transform of this interpolation function does
‘ot contain any terms with periods less chan 2v, This interpola-
‘ion named “cardinal fanetion” Cla) by Whitaker has the form
= Kernan (e~2-r
cw
Mn E(y-a-rw)
(is an arbitrary siting term) withthe folowing characteris:
We defined originally os that unique function of the
cotabular set, which has no singularities inthe finite part
‘of the plane and no constiuents whose period i les than
twice the tabular interval
Here the “cotabular st” refers tothe set ofall possible func
tions with the same tabular or sampling values, whereby the
‘tabular inverva” then corresponds tothe sampling interval
Furthermore, Whittaker demonstrates that the cardinal
function is the only interpolation function with these character-
istics. This also implicitly postulates that every function whose
Fourier transform is limited to frequencies < 1/2 may be
described by sampling values at intervals of, and can unique
ly be Interpolated again inthe form of the cardinal funetion.
'W. L Ferrar pointed out in 1925 that sampling and interpola-
tion ofthe cardinal function itself again lead to the same func-
tion, respective of how the sampling values ate shifted in time
“re refers this important invariance propery ofthe sampling the-
‘rem as “consistency.” Subsequent publications onthe cardinal
function ae cited ty J, M, Whitaker (not tobe confused with E.
‘T. Whittaker) [3], Shannon then refers to this book in [2]
By way of conclusion it should be mentioned that an exten-
sive tutorial review of the development of the sampling theo-
‘rem after Shannon was published by A. J, Jers in 1977 (14),
CONCLUSION
‘The sampling theorem for lowpass functions plays an important
role in communication engineering 3s a connecting ink between
‘continuous-time and discrete-time signals. The numerous dif-
{erent names to which the sampling theorem i attributed inthe
literature — Shannon, Nyquist, Kotelnikov, Whittaker, to
Someya — gave rise to the above discussion ofits origins. How-
ever, tis history also reveals a process which is often apparent
in theoretical problems in technology ox physis: ist the practi-