The Unreasonable Effectiveness

of Number Theory in Science

and Communication (1987
Rayleigh Lecture)
Manfred R. Schroeder
Few scientists have been active in so many fields-and have
been so successful in their endeavors-as John William
Strutt, who became Lord Rayleigh on the death of his father
in 1873. Not only did Rayleigh write the monumental Theory
of Sound (still ringing loud and clear) and discover argon, the
first noble gas (for which he received-appropriately -the
Nobel Prize for Physics in 1904), the great British classical
physicist also made fundamental contributions to a broad
spectrum of problems: from the spectrum of black-body radiation (the Rayleigh-Jeanslaw, which helped to usher in the
quantum revolution) to the coloring of the (unclouded) sky,
which was elucidated by what is now known as Rayleigh
scattering. In fact, to demonstrate Rayleighs multifaceted
interdisciplinary contributions, we simply have to list the phenomena and methods that bear his name, to wit: Rayleigh
surface waves, Rayleigh radiation pressure, Rayleigh number,
Rayleigh-Benard convection, the Rayleigh disc, the Rayleigh
refractometer, the Rayleigh-Ritz method (for the analysis of
vibrational problems), the Rayleigh distribution, and so on.
To be awarded a medal named after such an exquisite mind
is a great honor, indeed. And I think Rayleigh, the polymath,
wouldnt mind -in fact, perhaps even enjoy- listening to
what I have to say about another interdisciplinary venture:
the effectiveness of discrete mathematics in physics, music,
communication (and its negation: cryptography).
INTRODUCTION

UMBER theory has been considered since time imm e m o r i a l t o b e t h e very paradigm o f p u r e
(some would say useless) mathematics. According to
Carl Friedrich Gauss, mathematics i s the queen of
sciences-and
number theory is the queen of mathe-

*Based on the authors book Number Theory in Science and Communication, With Applications in Cryptography, Physics, Digital
Information, Computing and Self-similarity (2nd Enlarged edition,
Springer, Berlin, New York 1986).

0740-7467/88/0100-0005$01.0001988 I EEE

matics. What could be more beautiful than a deep, satisfying relation between whole numbers. (One i s almost
tempted to call them wholesome numbers.) Indeed, it i s
hard to come up with a more appropriate designation than
their learned name: t h e integers-meaning
the
untouched ones. How high they rank, in the realms of
pure thought and aesthetics, above their lesser brethren:
the real and complex numbers-whose first names virtually exude unsavory involvement with the complex realities of everday life!
Yet the theory of integers can provide totally unexpected answers to real-world problems. In fact, discrete
mathematics is taking on an ever more important role. If
nothing else, the advent of the digital computer and digital
communication has seen to that. But even earlier, in physics, the emergence of quantum mechanics and discrete
elementary particles put a premium on the methods and,
indeed, the spirit of discrete mathematics.
In mathematics proper, Hermann Minkowski, in the
preface to his introductory book, on number theory,
Diophantische Approximationen, published in 1907 (the
year he gave special relativity its proper four-dimensional
clothing in preparation for its journey into general covariance and cosmology) expressed his conviction that the
deepest interrelationships in analysis are of an arithmet icaI nat u re.
Yet much of our schooling concentrates on analysis and
other branches of continuum mathematics to the virtual
exclusion of number theory, group theory, combinatorics
and graph theory. As an illustration, at a recent symposium on information theory, the author met several
young mathematicians working in the field of primality
testing, who-in all their studies up to the Ph.D.-had
not heard a single lecture on number theory!
Or, to give an earlier example, when Werner Heisenberg discovered matrix mechanics in 1925, he did not
know what a matrix was (Max Born had to tell him), and
neither Heisenberg nor Born knew what to make of the
appearance of matrices in the context of the atom. (David
Hilbert is reported to have told them to go look for a
JANUARY 1988 IEEE ASSP MAGAZINE

differential equation with the same eigenvalues, if that

ivould make them happier. They did not follow Hilberts
well-meant advice and thereby may have missed discovering the Schrodinger wave equation,.)
Integers have repeatedly played a crucial role in the
evolution of the natural sciences. Thus, in the 18th century, Lavoisier discovered that chemical compounds are
composed of fixed proportions of their constituents
which, when expressed in proper weights, correspond to
the ratios of s m a l l integers. This was one of the strongest
hints to the existence of atoms; but chemists, for a long
time, ignored the evidence and continued to treat atoms
a5 a conceptual convenience devoid of physical meaning.
(Ironically, it was from the statistical laws of large numbers, in Einsteins and Snioluchowskis analysis of Brownian rnotion at the beginning of our own century, that
the irrefutable reality of atoms and molecules finally
emerged.)
In the analysis of optical spectra, certain integer relationships between the wavelengths of spectral lines
emitted by excited atoms gave early clues to the structure
of atoms, culminating in the creation of matrix mechanics
in 1925, an important year in the growth of integerphysics.
In 1882, Rayleigh discovered that the ratio of atomic
weights of oxygen and hydrogen i s not 1 6 : l but 15.882:l.
These near-integer ratios of atomic weights suggested to
physicists that the atomic nucleus must be made up of
integer numbers of similar nucleons. The deviations from
integer ratios later led to the discovery of elemental
isotopes.
And finally, small divergencies in the atomic weight of
pure isotopes from exact integers constituted an early
confirmation of Einsteins famous equation E = mc, long
before the mass defects implied by these integer discrepancies blew up into those widely noticed, infamous
mushroom clouds.
O n a more harmonious theme, the role of integer ratios
in musical scales has been appreciated ever since Pythdgoras first pointed out their importance. The occurrence of integers in biology-from plant morphology to
the genetic code-is pervasive. It has even been hypothesized that the North American 17-year cicada selected its
life cycle because 17 i s a prime number, prime cycles offering better protection from predators than nonprime cycles. (The suggestion that the 17-year cicada knows that
17 is a Fermat prime has yet to be touted though.)
Another reason for the resurrection of the integers is
the penetration of our lives achieved by that 20th-century
descendant of the abacus, the digital computer. (Where
did all the slide rules go? Ruled out of most significant
places by the ubiquitous pocket calculator, they are sliding fast into restful oblivion.)
An equally important reason for the recent revival of the
integer i s the congruence of congruential arithmetic with
numerous modern developments in the natural sciences
and digital communication-especially
secure communication by cryptographic systems. Last not least,
the proper protection and security of computer systems

and data files rest largely on keys based on congruence

relationships.
In congruential arithmetic, what counts i s not a numerical value per se, but rather its remainder or residue after
division by a modulus. Similarly, in wave interference (be
it of ripples on a lake or electromagnetic fields on a hologram plate) it i s not path differences that determine the
resulting interference pattern, but rather residues after
dividing by the wavelength. For perfectly periodic events,
there is no difference between a path difference of half a
wavelength and one-and-a-half wavelengths: in either
case the interference will be destructive.
One of the most dramatic consequences of congruentia1 arithmetic i s the existence of the chemical elements
as we know them. In 1913, Niels Bohr postulated that certain integrals associated with electrons in orbit around
the atomic nucleus should have integer values, a requirement that 10 years later became comprehensible as a wave
interference phenomenon of the newly discovered de
Broglie matter waves: In essence, integer-valued integrals
meant that path differences are divisible by the electrons
wavelength without leaving a remainder.

MUSIC A N D NUMBERS
Ever since Pythagoras, small integers and their ratios
have played a fundamental role in the construction of
musical scales. There are good reasons for this preponderance of small integers both in the production and
perception of music. String instruments, as abundant in
antiquity as today, produce simple frequency ratios when
their strings are subdivided into equal lengths: shortening
the string by one half produces the frequency ratio 2:1,
the octave; and making it a third shorter produces the
frequency ratio 3:2, the perfect fifth.
In perception, ratio of small integers avoids unpleasant
beats between harmonics. Apart from the frequency ratio
1 : l (unison), the octave is the most easily perceived
interval. Next in importance comes the perfect fifth. Unfortunately, as a consequence of the fundamental theorem of arithmetic, musical scales exactly congruent
modulo the octave cannot be constructed from the fitth
alone because there are no positive integers k and m such
that

However, there are good approximation to (1). Writing

= 2

or
log, 3

2
m

we see that we need a rational approximations to log2 3.

The proper way of doing this is to expand log2 3 into a
continued fraction
log2 3 = [1,1,1,2,2,. . .]

which yields the close approximation m = 12, n = 19. In

other words, if we want to make a good fifth with an
equal-tempered (equal frequency ratio) scale, the basic
interval 1:2*, the semitone, recommends itself. In fact,
the semitone interval has come to dominate much of
Western music. The equal-tempered fifth comes out a5

where p i s a prime and n may be replaced by (n)mod p,

its least nonnegative residue modulo p.
It i s easy to show that the discrete Fourier Transform
(DFT) of r,, has constant magnitude. As a physical consequence, the intensities of the wavelets scattered into
different directions from a surface with reflection coefficients (3) will be equal (in the customary Kirchhoff ap2 = 1.498.. .
(2) proximation of diffraction theory [ I ] ) . The scattering
angles a k are given by the wavelength A and the step size
Another fortunate number-theoretic coincidence i s the
w (corresponding to An = 1 ) :
fact that 7, the numerator in the exponent in (21, is coprime to 12. As a consequence we can reach all 12 notes of
kh
PW
sin (Yk = -,
(4)
/kj s -,
the octave interval by repeating the fifth (modulo the
PW
A
octave). This is the famous Circle of Fifths.
yielding 2 _pw/A; t 1 different angles. ( l h e Gauss
Of course, (2) i s not an exact equality and some comprobrackets i.4 stand for rounding down to the nearest intmises have to be made in the construction of scales. Pyeger.) The different reflection coefficients are realized by
thagoras used the perfect fifth (3:2) and the perfect fourth
IIs of different depths
( 4 : 3 ) , but fudged on the minor and major thirds, which we
come out as 1 :1.185 and 1:1.265, rather that 6 : 5 and 5:4,
h
d, = -(nil
mod p,
respectively. (How Pythagoras must have wished the fun2P
damental theorem out of existence!)
More recently J.R. Pierce has tried to go Pythagoras one as illustrated in Fig. 1 for p = 17. Such wells give a roundbetter by constructing a musical scale for the tritave (the trip phase change of 2d,, * 2 r / h in accordance with the
frequency interval 3:1) based on the integer ratio 5:3. phase requirement of (3). In (4), h i 5 the longest wavePierce used trial and error, but we simply expand logc5 length to be scattered. For any integral submultiple of that
into a continued fraction. This yields the close approxi- wavelength, A/m, the reflection coefficients ( 3 ) are
0 (mod p) has the same flat
mation 5 = 313, the tritave should be subdivided into 13 changed to r;, which for m
equal intervals. (As a bonus-a
number-theoretic Fourier property as (3). For w one chooses typically half
fluke-3
also allows a very close approximation of the the smallest wavelength to be scattered over ? i ~ / 2 .
Figure 2 shows the diffraction pattern of the grating in
next prime 7. To wit: 7 = 323.)
Fig. 1 for one third the longest wavelength

CONCERT HALLS AND QUADRATIC RESIDUES

There i s another connection between music and numbers: concert hall acoustics. Extensive physical tests and
psychophysical evaluation of the acoustic qualities of concert halls around the world have established the importance of laterally traveling sound waves. (Such waves
produce dissimilar signals of a listeners two ears, a kind of
stereophonic condition that is widely preferred for music
listening.)
In order to convert sound waves traveling longitudinally
(from the stage via the ceiling to the main listening areas)
into lateral waves, the author has recommended ceiling
structures that scatter sound waves, without absorption,
into broad lateral patterns. In the physicists language,
concert hall ceilings (and perhaps other surfaces too)
should be reflection phase-gratings with equal energies
going into the different diffraction orders (the different
lateral directions).
How should one go about designing such an ideal scatterer for sound (or light or radar waves)? Curiously, one
answer comes from a classical branch of number theory
that has exercised the great Gauss for a long time: quadratic residues. Consider a surface structure whose reflection
coefficient rn varies in equidistant steps along one axis
according to
rn =

e2~~nzp

,n=O,i1,?2

,...

QUADRATIC- RESIDUE DIFFUSOR

CEILING

-WELLS

-.cCROSS-DIMENSION

OF HALL

--c

FLOOR1

Figure 1. A reflection phase-grating, based on quadratic

residues, for effective scattering of waves. In this example residues are taken modulo p = 17. Such structures are useful in radar and sonar camouflage, diffusion
3f coherent light, and noise abatement-and concert hall
acoustics.

(3)
JANUARY 1988 IEEE ASSP MAGAZINE

WAVE DIFFRACTION A N D PRIMITIVE ROOTS

In applications, including concert hall acoustics, it may
be desirable to attenuate the specular reflection (the
zeroeth diffraction order) relative to the other diffraction orders. Are such super scattering phase gratings
possible? The answer is yes and it comes, again, from
number theory.
Instead of quadratic residues the author [21 has suggested powers of primitive roots to construct reflection
coefficients. Specifically ( 3 ) is replaced by

where g is a primitive root of the prime p (g is a primitive

root of p if its powers, g, generate all the p - 1 different
nonzero integers modulo p). The sequence rn is then
periodic with period p - 1. It i s not difficult to show that
the Discrete Fourier Transform R k of (5) has constant
magnitude except for lRoI which is p times smaller than lRkI
for k
0 (mod p - 1).
Figure 3 shows a scatter diagram of a microwave reflection phase grating (useful in radar camouflage, for example) based on p = 7 and g = 3. Note the attenuated
zeroeth diffraction order between the six strong lobes
corresponding to k = 21, 2 2 , k 3 .
The success of phase gratings based on primitive roots
raises the following mathematical question: The least
positive residues of g modulo p generate a permutation
of the integers 1 to p - 1. The 4 ( p - 1) primitive roots of
p generate d ( p - 1) different permutations, which i s typically a small fraction of all possible permutations. Not
c o u n t i n g cyclical shifts and reflections, there are

(p - 2)!/2 inequivalent permutations. For p = 17, we

have thus 653 837 184 000 such permutations, of which
d(16) = 8 (about one in 10) have the flat Fourier property.
Questions: are there other permutations that, when used
instead of g in (5), have this property?

FORBIDDING PROPERTY O F THE FERMAT PRIMES

Since Gausss great discovery of 30 March 1796 concerning the factorization of xp - 1, where p is a Fermat prime,
p = 22 + 1,
these primes have enjoyed great esteem, spreading from
the factoring factories of the experts to the sweatshops of
the amateur. The geometrical constructions of regular
p-gons rests on the fact that p - 1 has only one prime
factor, namely 2. Now, in the context of wave diffraction,
this same property- instead of permitting a particular
construction -forbids a desirable application, namely the
construction of two (or higher) dimensional phase arrays
with the flat Fourier property mentioned above.
The two-dimensional generalization of the periodic sequence ( 5 ) is the periodic array rk( with
k =

(n),d

and

t = (n),d

L,

where K and L (in best Chinese remainder fashion) are two

coprime factors of p - 1. Thus there are no n-dimensional
(n > 1) phase gratings based on the known Fermat primes
3, 5, 17, 257, 65 537. (The smallest prime for a threedimensional array i s 31, having a 2 * 3 * 5 unit cell.)

EULER TOTIENTS A N D CRYPTOGRAPHY

One of the most spectacular applications of number
theory in recent times is public-key cryptography in which
each potential recipient of a secret message publishes his

encryption key, thereby avoiding the (often substantial)

problems of secure secret-key distribution. But how can a
key be public and yet produce secret messages?The answer i s based on Euler's totient function d ( r ) and the role
it plays in inverting modular exponentiation. The public
key consists of a modulus rand an exponent s, coprime to
4 ( r ) . The message i s represented by an integer M,
1 < M < r, and the encrypted message E is given by a
number in the same range, calculated as follows

SOME OF THE MANY USES OF FINITE FIELDS

Certain periodicsequenceswith elements from the finite

(Galois) field GF(p), formed with the help of irreducible
polynomials over GF(p"), have unique and much soughtafter correlation and Fourier transform properties [71.
These Galois sequences, as I have called them, have found
ingenious applications in error-correcting codes (compact
discs and picture transmissions from inter-planetary satellites) and in precision measurements from physiology to
E = M'mod r.
(6) general relativity. Other applications are in radar and
sonar camouflage, because Galois sequences, like quadDecrypting E i s accomplished by calculating Et mod r, ratic residues (see above), permit the design of surfaces
where the decrypting exponent t i s given by
that scatter incoming waves very broadly, thereby making
ts = 1 mod 4 ( r ) ,
( 7 ) the reflected energy "invisible" or "inaudible." A similar
application occurs in work with coherent light, where a
i.e., ts = k 4 ( r ) 1 for some k. With such a t,
"roughening" of wavefronts (phase randomization) i s
M k d ( r ) + l mod r
often desired (for example, to avoid "speckles" in holoEt = M S t
grams). Light diffusors whose design is based o n Galois
which, according to Euler's theorem, give the message arrays are in a sense the ultimate in frosted (milk) glass.
M back.
Finally, Galois sequences allow the design of loudspeaker
So far, so good and-theoretically-trivial.
The trick in and antenna arrays with very broad radiation characterpublic-key encryption is to choose r as the product of two istics [2].
very large primes, each say 200 digits long. (There is no
paucity of such primes, and enough for all foreseeable
ERROR CORRECTION CODES FROM GALOIS FIELDS
purposes can be easily ferreted out from the jungle of
Galois sequences, with periods n = pm - 1, are concomposites in the IOzoo neighborhood.)
Now. with a composite r, prescription ( 7 ) ,so easily writ- structed with the help of an irreducible polynomial g(x) of
ten down, becomes practically impossible to apply be- degree m with coefficients from GF(p), such that g(x) is a
1 but not a factor of xr - 1 for r < n.
cause $(r) can be calculated only if the factors of r are factor of x"
Binary Galois sequences (p = 2) with elements 0 and 1
known-and this knowledge i s not published. In modern
(or, in certain other applications, 1 and -1) are the most
parlance, the mapping (6) is a trap-door function.
A trap-door function is, as the name implies, a (mathe- important practical case. For p = 2 and m = 4, an irreducmatical) function that is easy to calculate in one direction ible polynomial with the stated property is
but very hard to calculate in the opposite direction. For
g(x) = 1 x
x",
example, it takes a modern computer only microseconds
to multiply two 100-digit numbers. By contrast, to decom- from which the recursion
pose a 200-digit number, having two 100-digit factors, into
(8)
akt4 = ak+l + ak
its factors can take "forever," even on the fastest computers available in 1987 and using the most efficient factoring is obtained. Beginning with the initial condition 1000 (or
almost any other tuple, except the all-zero tuple) (8) genalgorithms known today (31.
O n the other hand, knowing the factors of r, as the erates the binary Galois sequence of periodic length
legitimate recipient of the encrypted message E does, 4 ( r ) 24 - 1 = 15:
can be easily calculated and decrypting becomes possible.
1000,10011010111 ; etc. (repeated periodically) .
The decrypting exponent t is obtained by Euclid's algoThe error correcting properties of codes based on such
rithm [7] or by solving (7) directly:
sequences result from the fact that the 2" = 16 different
t
S ~ ( & d r ) ) - l mod + ( r ) .
initial conditions generate 16 different code words of
Not so long ago, the most efficient factoring algorithms length 15, that form a "linear code" (the sum of two code
on a very fast computer were estimated to take trillions of words is another code word). These code words define
years. But algorithms get more efficient by the month and t h e r e f o r e a 4-dimensional linear subspace o f t h e
computers become faster and faster every year, and there 15-dimensional space with coordinates 0 and 1. In fact, the
is no guarantee that one day a so-called "polynomial-time" 16 code words describe a simplex (in 3 dimensions a simalgorithm will not emerge that will allow fast factoring of plex is a tetrahedron) in that space and the resulting code
even 1000-digit numbers. Few mathematicians believe that is therefore called a Simplex Code [41.
a true polynomial-time algorithm is just around the corner,
Its outstanding property is that every pair of code words
but there also seems to be little prospect of proving that has the same Hamming distance (the number of 0,l disthis will never occur. This is the Achilles heel of public- parities), namely 2m-' = 8. Thus, the code can recognize
key cryptography.
up to 2"-' = 4 errors and correct up to 2"-' - 1 errors.

+ +

JANUARY 1988 IEEE ASSP MAGAZINE

The price tor this error correcting property is a reduced

signaling efficiency, nameiy m j n = 4/15, where n =
-3"' 1 i s the length of the code word.
~~

Several other codes can be derived from the simple

Simplex Code. For example, the famous (and historically
earlv) Hamming Codes. The code words of a Hamming
Code are given by the orthogonal subspace of the Simplex
Code of the same length. Hamming codes carry n
m
intormation bits and m check bits, and can correct prer-isely one error. The functioning of the Hamming Code
for rn = 3 , n = 7, i s illustrated in Fig. 4.
-

Using this notation, the constant Hamming distance

between code words of a Simplex Code (whose members
are generated by cyclic shifts) translates immediately into
the following circular autocorrelation property

"

c,: =

SkSk+r

-I

for r

+ 0 mod n ,

k=l

and, of course, c, = n tor r 5 0 mod n. As a result of this

two-valuedness of c,, the Fourier transtorm ot s k :
s,

sk
k

exp(-i2vkm/n)

has constant magnitude for m 9 0 mod n. In the lingo ot

the physicist and computer scientist: the sequence s k has
a flat (or "white") power spectrum.
If we identify the index k with (discrete) physical time,
then we can say that the "energy" IskIL = 1, of the sequence sk i s equally distributed over all time epochs. And
because of IS,1' i s constant, we can make the same claim
with respect to the distribution of energy over all (nonzero) frequency componen t s. This equal "energy
spreading" of the Galois sequences s k with period length
n z . 2 " - 1, obtained with the help of polynomials over
GF(p"), has many impressive applications, some of the
more astounding ones occurring in the interplanetary
distance measurements.
GALOIS SEQUENCES A N D THE FOURTH EFFECT OF
GENERAL RELATIVITY

The n - m = 4 information bits, say 1001, are entered

into the 4 inner areas of the Venn diagram (indicated by fat
characters in Fig. 4). The m = 3 check bits (thin characters)
are entered into the 3 outer areas such that the parity in
each circle i s even (the sum modulo 2 equals 0).
The receiver of a code word, which may have been
contaminated in transmission, checks the parity in each
circle and marks all circles with odd parity. The intersection of these circles then specifies uniquely a single bit
error (including in the check bits themselves). These 3
parity checks allow the receiver to distinguish between
p r e ~ i s e l y 2=~8 different possibilities: a single error in any
of the 7 transmitted bits or n o error. No wonder the Hamming Code is called a perfect code.

CORRELATION A N D FOURIER PROPERTIES OF

GALOIS SEQUENCES
For many purposes it i s advantageous to use the elements 5k = 1 or -1 instead of ak = 0 or 1. The mapping is
sk

10

= 2ak

1.

IEEE ASSP MAGAZINE JANUARY 1988

General Relativity, the theory of gravitation propounded

by Einstein in November 1915 in Berlin (and 5 days earlier
by Hilbert in Coettingen) passed three important experimental tests during Einstein's lifetime:
1) The perihelion motion of the orbit of the planet
Mercury, which was already known from earlier
astronomical observations.
2) The bending of light waves near the sun, first observed during the total eclipse of 1919 by the
Eddington expedition.
3 ) The gravitational red shift, first seen in the light from
massive stars, but now measurable even on Earth herself using the ultrasensitive Moessbauer effect.
A fourth effect inherent in Einstein's theory was not
confirmed until fairly recently: the slowing of electromagnetic radiation in a gravitational field.' This effect was
'In the long struggle t o put his principle of general equivalence of
different references frames into proper mathematical clothing, Einstein discovered-as earlyas 1909-that the speed of light could not
be constant (as in special relativity) but must depend on the gravitational potential 4 . Althogh he had no general theory then, Einstein
found that, t o first order, c(4) c + 4/c, where c is the usual vacuum velocity of light i n field-free space (Note: 4 < 0). Ironically, the
slowing of radiation in gravitational fields, although appreciated very
early, was not considered a testable proposition until the perfection
of radar technology, using Galois sequences, in the second half of
this century. The reason for this delay in testing the extra delay was,
of course, that n o one could picture himself (or anyone else, for that
matter) floating next t o the sun, stopwatch in hand, clocking the
passing photons.

observed by means of radar echoes from the planets

Venus and Mercury as they disappeared behind the sun as
seen from the earth (superior conjunction). In that position, both the outgoing and returning radar waves have to
travel near (indeed around) the sun. Even after taking
plasma effects near the suns surface and other factors into
account, physicists found an extra delay of 200 p - v e r y
close t o the prediction of general relativity [5].
Why was this measurement not done long ago? The
reason i s t h a t t h e e c h o e n e r g y f r o m M e r c u r y exceedingly weak even when visible-drops to IO- of
the outgoing energy as the planet slips behind the sun.
The astounding fact that reliable results have been obtained in spite of these miniscule reflected energies is due
mainly to the proper choice of the transmitted sequence
of radar pulses, based o n primitive polynomials over finite
number fields.

CHINESE REMAINDERS FEED FAST ALGORITHM

In the computation of discrete circular convolutions
M

f(k)g(n

h(n) =

k)modM,n = 1 , 2 . . . . M ,

(9)

k=l

a pervasive task in numerous numerical applications [6],

the n u mbe r of ar it h met ical ope ration s ( m u Iti p I icat ion s
and additions) is M2,where M i s the period of the involved
sequences. It is easy to show [71 that if M has r > 1 coprime factors

M = mlm2..

. m,,

then the one-dimensional convolution (9) can be converted into an r-dimensional convolution by expressing
the summating index n in the Sino notation:
n,N,M/m, mod M I

n =
I

where n, is least positive remainders of n modulo m,, and

N, is given by the congruence
N,M/m, = 1 mod m,
The necessary summations over the n, require a total of
MZm, operations, which can be considerably smaller than
M2.For example, for M = 1 007 760 = 13 * 15 . 16 17 19,
the number of operations drops by a factor M/Cm, =
12 597-a very substantial saving, comparable t o the
economy offered by the Fast Fourier Transform (FFT),
which also converts a one-dimensional operation into a
multi-dimensional one. In fact, the fast Chinese convolution [71 described here complements nicely the FFT which
works most efficiently if all factors m, of M equal 2 rather
than being coprime.

EPILOGUE
It i s clear that only a sprinkling of the numerous applications of number theory outside mathematics proper could
be mentioned here. Among the many topics that had to be

counted out from this brief account are

1. The application of continued fractions to electrical
network problems which, incredible, led to the construction of the squared square [7]-long considered impossible [8I. (The squared square i s a square
with integer sides, completely covered, without
overlap, by smaller incongruent integer squares.
2. Heat conduction in thin tours: the solution is based
on the representation of integers by the sum of two
squares [91.
3. The eigenvalue distribution of normal modes in
cubical (and near cubical) resonators, which depends
on the representation of integers by the sum of three
squares [71. (The authors first encounter with
number theory, in his Ph.D. thesis on concert hall
acoustics.)
4. Search algorithms, game strategies [IO] and countless
other applications based on Fibonacci numbers.
5 . Certain unexpected properties of the zeroes of
Riemanns zeta-function, found by A. Odlyzko [Ill,
and their possible relation to the Wigner distribution
function (which governs the distribution of energy
levels in the atomic nucleus and eigenfrequencies in
complex vibrational systems).
6. And, most recently, the elucidation of the structure
of quasi-crystals, a new state of matter combining
forbidden 5-fold rotational symmetry and sharp,
crystal-like, diffraction patterns [7].
What riddle will be solved next by number theory? Is
this effectiveness of the higher arithmetic completely unreasonable? O r are we witnessing here a pre-established
harmony 3 la Leibniz between mathematics and the real
world?

REFERENCES

M. R . Schroeder, Diffuse sound reflection by

maximum-length sequences, J. Acoust. Am. vol. 57,
140-150 (1975). See also: M . R. Schroeder Toward
better acoustics for concert halls, Physics Today,
24-30 Oct. 1980.
M. R. Schroeder, Constant-amplitude antenna arrays
with beam patterns whose lobes have equal magnitudes, Archiv f u r Electronik u n d Uebertragungstechnik (Electronics and Communication)
vol. 34, 165-168 1980.
C. Pomerance, Recent developments in primality
testing, The Mathematical Intelligencer, vol. 34,

97-105 (1981).
F. J. MacWilliams and N. J. Sloane, The TheoryofErrorCorrecting Codes, North-Holland, Amsterdam 1978.
I. I. Shapiro et al., Fourth test of general relativity,
Phys. Rev. Lett., vol. 20, 1265-1269 1968.
J.H. McClelland and C. M. Rader, Number Theory in
Digital Signal Processing, Prentice-Hall, Englewood
Cliffs 1979.
M.R. Schroeder, Number Theory in Science and
Communication -With Applications in CryptograJANUARY 1988 IEEE ASSP MAGAZINE

11

phy. Physics, Digital Information, Computing and

Self-similarity, Second Enlarged Edition, SpringerVerlag, Berlin 1986.
[8] C. J. Bouwkamp, A. J. Duijvestijn and P. Medema,
Tables Relating to Simples Squared Rectangles, Dept.
of Mathematics and Mechanics, Technische Hogeschool, Eindhoven 1960.

[91 1. Rohlfs, Mathematische Miniaturen, in F. Hirzebruch

(ed.), Birkhauser, Basel, 75-91 1983.
[IO] E. R. Berlekamp, H. H. Conway, and R. K. Guy, Winning Ways, Academi 7 Press, London 1981.
[ I l l A . M . Odlyzko, " O n the Distribution of Spacing
Between Zeroes of the Zeta Function," Math. Comp.,
vol. 48, 273-308 Jan. 1987.
See page 30 for M. R. Schroeder's technical biography.

IASTED International Symposium On Signal Processing and its Applications [ISSPA 871
Brisbane, Australia, 24-28 August, 1987
Conference Report
This first Symposium was held under the auspices of
IASTED and co-sponsored by IEEE, IREE and IE (Aust) and
was attended by 250 delegates, representing eighteen
nations and all five continents. It saw many professional
bodies from Universities, Industry, Defence Organisations and Research Establishments, co-operating to enable
participants to update themselves on the latest developments in Signal Processing, and to provide an opportunity for the interchange of ideas.
The Symposium ran for 5 days, with a large number of
delegates (100) attending the Tutorial presentations on
Days 1 and 2, and (250) attending the Conference proper
on Days 3-5. The success of the Tutorials demonstrated
the growing interest in the field of Signal Processing
worldwide, and more especially in Australia. Eminent
researchers- Harper Whitehouse, USA, Prof. J . Lim,
USA, Prof. A. Constantinides, UK, Drs. Gray, Steele and
Cantoni, Australia, Prof. S. Mitra, USA, provided stimulating presentations in such areas as:
Two Dimensional Signal Processing
Adaptive Array Processing
Digital Filter Design
Advanced Signal Processing Algorithms, and Parallel
Architectures for Signal Processing.
The Conference was officially opened at the end of
Day 2 with an Opening Ceremony and Welcome by the
General Chairman Dr. B. Boashash. An opening address
was also presented by M r . H . d'Assumpcao, Chief
Defence Scientist, Department of Defence, Australia. In
the address Mr. d'Assumpcao gave an outline of the various applications of Signal Processing in Defence, and previewed the ISSPA presentations of the large number of
Australian Defence Department representatives participating in the conference. In his address he also alluded
to the fact that a new approach to research might be forthcoming, one in which inherent usefulness of the research,
rather than simply mathematical eloquence, must be
given priority.
The final three days of the conference were comprised
of plenary sessions, oral and poster presentations, a Tech-

12

IEEE ASSP MAGAZINE JANUARY 1988

nical Exhibition, and many opportunities for delegates to

interact and exchange ideas. The plenary sessions at the
beginning of Days 3 and 4 were addressed respectively by
Prof. B. D. 0. Anderson, who spoke on "Decision Feedback Equalisers: Concepts Towards Design Guidelines,"
and by Prof. J . Imberger, who spoke o n "Signal Processing
Applications and Requirements in Oceanography." The
speech by Prof. J . Imberger, a well-known researcher in
Oceanography, illustrated the interest that Signal Processing is generating in many fields of Engineering.
The oral and poster presentations, which extended over
32 sessions, covered a wide variety of topics from 14 areas:
Digital Signal Processing
Digital Filters
Spectral Estimation
Time-Frequency Analysis
Detection, Estimation and Adaptive Signal Processing
Seismic and Array Signal Processing Applications
Radar and Sonar Signal Processing
Speech Processing and Pattern Recognition
Image Processing, Recognition, Synthesis
VLSl Signal Processing
Optical Signal Processing
Biomedical Signal Processing
Digital Communications
Applications of Signal Processing Techniques
The attendance at sessions was good, and discussions
were in many cases stimulating. The social activities of the
conference were also well attended, congenial in atmosphere, and gave further opportunity to delegates for interacting with colleagues on an informal basis.
The high standard and success of this first symposium
has received great support and acceptance within Australia's fast growing Signal Processing Industry. It was therefore decided to organise a second symposium at the Gold
Coast, Australia, in August 1990.
For further information please contact Dr. B. Boashash,
c/o UniQuest Limited, University of Queensland, Brisbane, Qld, 4067, AUSTRALIA.

IEEE ASSP SOCIETY DISTINGUISHED LECTURERS 1986-7

Russell M. M e r s e r e a u (S69-M73SM78-F83) received the S.B. and S.M.

degrees in 1969 and the Sc.D. degree in
1973 from the Massachusetts Institute of
Technology, Cambridge.
From 1971 to 1973 he was an Instructor
in the Department of Electrical Engineering at M.I.T., and from 1973 to 1975
he was with the Research Laboratory of
Electronics and Department of Electrical
Engineering as a Research Associate.
Currently he is a Professor in the School
of Electrical Engineering, Georgia Institute of Technology, Atlanta, where his
p r i m a r y research interest is m u l t i dimensional digital signal processing.
Dr. Mersereau was formerly the Associate Editor for Signal Processing of the /Transactions o n Acoustics, Speech, and Signal Processing and Technical Program Chairman of ICASSP-81. He is a former
member of the ASSP AdCom and the Steering Committee for the /E
Transactions on Medical Imaging. He is currently a member of the
MDSP Technical Committee and the Editorial Board of the Proceedings of the IEEE. He was the co-recipient of the 1976 Browder J .
Thompson Memorial Prize from the IEEE for the best paper by an
author under the age of 30 and a recipient of the 1976 Research Unit
Award from the Southeastern Section of the ASEE. He was the recipient of the IEEE Key to the Future Award (Education) in 1984, and
received a teaching award from Georgia Tech in 1978.

Manfred Robert Schroeder was born in

Ahlen, Germany on 12 July 1926. He
holds the degrees of Vordiplom Mathematik (B5), Diplom-Physiker (M5), and
Dr rer.nat in Physics (PhD), gained in
1949, 1951, and 1954 from the University
o f C o t t i n g e n , Federal Republic o f
Germany HIS thesis was on precision
measurements with electromagnetic microwaves, which he later used to investigate the distribution of resonances
i n rooms-especially concert halls
Both his PhD thesis and a subsequent
theoretical paper on sound transmission
in larger rooms corrected widely held
opinions about room resonances, and laid the foundation of a new
discipline. Statistical Wave Acoustics, characterized by a critical frequency, now called Schroeder-frequency
After graduation, Professor Schroeder went to the USA to join the
research staff of Bell Laboratories, Murray Hill, New Jersey, where his
main goal was to synthesize intelligible speech signals from minimal
descriptions, such as written text He was also one of the leaders in
the emerging art of computer simulation As early as 1959 he studied
sound transmission in concert halls and complex speech coding systems by digital simulation This work led to one of the most important
innovations in digital speech synthesis linear predictive coding
(LPC), done in collaboration with B. 5 Atal Professor Schroeders
work on hearing resulted in the discovery of new monaural phase
effects and a formula for what has become known as Schroederphases
In 1958 Professor Schroeder was appointed Head of the Acoustics
Research Department at Bell, in 1963 he became Director of the

Acoustics & Speech Research Laboratory; and in the following year

assumed responsibility for all areas of acoustics and ultrasonic research.
In 1969 he was appointed Professor of Physics and Director of the
Drittes Physikalisches lnstitut of the University of Gottingen, and in
1972 served as Chairman of the Physics Department. His work on
concert hall acoustics, undertaken in collaboration with his students,
resulted in the recognition of the importance of lateral sound waves
for good acoustic quality, and the invention of Schroeder-diffusors
for optimally distributing sound waves. This invention is characteristic of many of Professor Schroeders proposals, in that it is based on
an appreciation of deep theoretical connections- number theory in
this case.
Professor Schroeder is one of the pioneers in computer graphics.
In 1969 he was awarded the First Prize at the International Computer
Art Competition for his application of concepts from mathematics
and physics to the creation of artistic works. In 1972 he received the
Gold Medal of the Audio Engineering Society for the successful
merging of theory and practice in the analysis of sound transmission
and for pioneering efforts to apply computer technology to architectural acoustics. In 1977 the Institute of Electrical and Electronics
Engineers awarded the Baker Prize to Professor Schroeder for his
tutorial paper Models of Hearing. In 1987 he was awarded the
Rayleigh Gold Medal by the British Institute of Acoustics.
Schroeder is also noted for new measurement methods in acoustics and other disciplines. He holds over 40 States Patents for inventions in these areas. His book Number Theory in Science and
Communication (Springer, Berlin, 1984,2nd Enlarged Ed. 1986) treats
applications in cryptography, physics, digital information processing,
self-similarity and computing. He also edited the book Speech and
Speaker Kecognhon (Karger, Basel 1985). He is a former Associate
Editor of the journal of the Acoustical Society of America, and currently Editor for Information Sciences for Springer-Verlag.
Professor Schroeder has been active in a number of assignments
for the US Government, and in 1963 was a guest to the Soviet Government to consult on the acoustics of the Palace of Congresses in the
Kremlin. He is a member of the National Academy of Engineering
(Washington, D.C.) and an honorary member of the Max-PlanckSociety (Federal Republic of Germany). In 1986, he was elected a
Fellow of the American Academy of Arts and Sciences.
He is married to Anny Menschik of New York, and has three children, Marion, Julian, and Alexander.

Donald W. Tufts received the B.A degree

in Mathematics from Williams College,
Williamstown, MA, in 1955 He received
the S.B , S M I and Sc.D degrees in
Electrical Engineering f r o m Massac h u s e t t s I n s t i t u t e o f Technology,
Cambridge, in 1957, 1958, and 1960, respectively. He was a Bell Telephone
Laboratories Fellow at M.I.T.
From 1960 to 1967 he was at Harvard
University, Cambridge, MA, first as a Research Fellow and Lecturer and later as
an Assistant Professor of Applied Mathe, matics Since 1967 he has been a Professor of Electrical Engineering at the
University ot Rhode Island, Kingston He has been a Consultant
to Bell Telephone Laboratories, Sanders Associates Inc., and other
companies