SKETCHING THE FOURIER TRANSFORM:

A GRAPHICAL PROCEDURE

Michael J. Toia

Federal Communications Commission

Office of Science and Technology

ABSTRACT forms on them. Significant radio frequencies can

be "seen" on a time-display oscilloscope.
The spectrum signature of an electronic
FOURIER SERIES
device is related to Fourier Transforms of its
time-variant waveforms. Calculation of these
The procedure to be described has been
transforms is time-consuming and often difficult.
dubbed a "Quick" transform. It differs in
several regards from the "Fast" transform. To
By applying several useful properties of
understand its development we will begin with a
integrals and the known frequency spectrum of a review of the conventional Fourier Expansion of
chain of impulses, .a method of simplifying the
repetitive waveforms.
analysis is developed. It is easily committed
to memory for rapid estimation of RFI from
Recall that cyclic waveforms can be expanded
arbitrary waveforms, and is amenable to graphical in an infinite series of sinusoids, the Fourier
solution. The method is of utility to technical
series, as follows: (Ref. l)
ly inclined personnel who lack formal mathematic
al training, for no integrals appear in its
"cook book*' format.
f(t) A +]T A cos(nwt) + ] L B sin(nwt) (l)

An algorithm applicable to microprocessors

or small computers is explored.
The D.C. term, A , is of no further interest
and will be ignored.

The coefficients An and Bn are calculated

INTRODUCTION thus:

Practicing engineers tend to shun the /t+T

application of complex mathematics. In EMC work, An = ^ j f(t) cos(nwt) dt (2)
this tendency surfaces in a reluctance to perform
Fourier Transforms when investigating complex
waveforms. As a result, the design engineer /t+T
often glosses over the generation of radio B^ = ^ J f(t) sin(nwt) dt (3)
frequencies which his device may later radiate
as RFI.
The integrals are taken over one complete period
The proper application of Fourier Transforms of f(t).
requires the solution of some often tedious,
integrals. The procedure is time consuming. As a warm-up exercise, let's find the
Constraints of scheduling to meet a market demand transform of a square wave defined by
often support the act of omitting this analysis.
/l, 0<t<T/2
The procedure given herein negates many of
these problems. Fourier Transforms are reduced f(t) = J o , T/2 < t < T (4)
to a technique so simple that no integrals need
be solved, nor is a computer required. The vf(t+kT), everywhere
process can be completed in a "doodle" mode with
nothing but pencil and paper. This waveform is shown in Figure 1.

The procedure can also be written as a 1

simple algorithm for application on a micro
computer or programmable calculator. But its
real utility remains its simplicity: it can be 0 time
committed to memory. In a short time, the
engineer becomes accustomed to viewing time- T/2 0 T/2 T 3T/2
variant waveforms and performing "mental" trans- FIGURE 1

74
U.S. Government work not protected by U.S. copyright.
Coefficients of the Fourier Series are found as A SPECIAL FUNCTION
followss
There is a function repetitive in time that
has a most curious and useful property. Consider
/T
j f(t) cos(nwt) dt an impulse, which is everywhere zero save at one
time,7 t A .' It can be defined by
/ T/2
=^ J l*cos(nwt) dij
d(t) (13)
/T j t~
+ = I O*cos(nwt) dt
'T/2 This function becomes infinite at t=t_
However, it does so in a controlled fashion, such
m f sin(T/2) - ^ B i n ( 0 ) (5) that it satisfies.

But sin( t /2) = 0, as t/2 is one-half period. A +6

Hence, lim j 0 d(t) dt = 1 (14)
f--o A -e

The special periodic function we seek is a

o

for all n (6)

11

n chain of such impulses, defined by

This reveals f(t) to be an odd function, f(t) = f(t) = d(t+kT) for all integer k (15)
-f(-t). A glance at Figure 1 confirms the fact.
so that, for any k,

Now calculate the B t +kT+C

lim j 0 f(t) dt = 1 . (16)
o s-t + k T -
*0
/-T O
B = - / f(t) sin(nwt) dt
n It / O This function has the form shown in Figure 2.
,T/2
= | sin(nwt) dt
11 Jo
T/2 time
cos(nwt) t -T t t +T t +2T t +3T t +4T
o o o o o o
nwTT v '
0
FIGUEE 2
= - i [cos(nwT/2) - | (7)
The arrows indicate a pulse height of zero width
but infinite height.
The period T and frequency w are related by
Think of f(t) as a train of rectangular
w = 2TT/t (8) pulses of constant area. As their width
decreases, their height increases without limit.
so we can write Figure 2 results in the limit of zero width.

b
n
= -i-
niTL
Fi - cos(mr)l
' J
(9) A straightforward Fourier Transform of f(t)
can be performed. Coefficients A are found as
usual;
For even (odd) values of n, cos(mr) is
clearly equal to +1 (-1)j so
rT
A
n
= ^
Tv
j f(t) cos(nwt) dt (17)
2/nir n odd (10)
Bn =
But f(t) is zero everywhere save near t .
B = 0 n even (11) Therefore we can break (17) into three integrals.
n

Thus we obtain a familiar result, viz.

t - f
A = if I 0*cos(nwt) dt
n Jo
/t +
f(t) = sin( wt) + -j sin(3wt)
+ J 0 f(t) cos(nwt) dt
''t
t -f
+ i sin(5wt) + i sin(7wt) (12)
I 0 cos(nwt) dt (1 8 )
ir A

75
The first and third of these vanish, leaving only
the second integral to he evaluated.
f(t) = +J A 'n cos(nw*:) ^
How, if epsilon is chosen sufficiently small, co
oos(nwt) will remain (essentially) constant over / B*n sin(nwt) dt (2 4 )
the range of integration. We may consider it a
constant, an approximation that improves, without
limit as epsilon shrinks. We obtain
from which it follows that
t +
An = ^f oos(nwto ) j 0 f(t) dt
f(t) = A' t + ^ T a *
/
fcos(nwt) dt
o ]_ n J
= ^ cos(nwtQ ) (19) CO s
(25)
+ J" B (sin(nwt)dt
1 n J
By choice of the origin of time we may
co B*
set t = 0 without loss of generality. Then
o f(t) = A ' t + - cos(nwt)
1
co A *
( 20) + 4_v.Tr
/ ? si n(vnw t /
) (26)

We have found a frequency comb! Each harmonic Thus the series for f(t) has coefficients
has exactly the same amplitude. This well-known given by
result surprises some people.

The reader may perform an entirely similar (27)

analysis with sine integrals to obtain the other
result,
for all values of n.
The spectrum of the impulse is simple.
Memorize it. The Quick Transform is rooted in
this relationship.
A USSjFOL THICK
The results derived above will be needed
shortly. We diverge to explore another property
of Fourier.Series.
Suppose we have a periodic function f(t)
whose transform is -unknown. Suppose further that
we know the transform of f'(t), its derivative.
The two transforms are related.
The relationship is obtained by expanding
f'(t) in a Fourier Series. From equation (l),
(28)
(21)
From equations (27) and (28) we observe the
following facts on integration of a time-
periodic waveform:
1) Every harmonic in the spectrum has its
amplitude modified by the factor l/w.
2) Every harmonic further has its amplitude
modified by the factor l/n. That is, its
amplitude is reduced by its harmonic number.
3) Each cosine and sine term of f'(t) is
retarded in phase by 9 0 .
ADBITIOHAL PROPERTIES OF IUTEGRALS
Two remaining properties of integrals that
are then properties of Fourier Transforms are
required for our continuing analysis. These are:

* (*) = A V + I A'n cos(nwt) f

J s.
f(x) dx = f
Ja.
f(x) dx + f
'o
f(x) dx (29)
00
+ ] T b * sin(nwt) (22) where a < cf~b, and
1 n

B y definition,
j{ fx (x) + f2 (x> ) dx =

f(t) =Jf'(t) dt (23)

Jf^(x) dx + J f 2 (x) dx (30)

76
 These relations are easily verified. Recall
the fundamental theorem of integral calculus
which states that
b b
f(x) dx = lim
i-*cd a and their timing varies by one-half cycle from
An integral is simply the sum of many infinites-
imally small terms, and the value of a sum is
independent of the order in which terms are
added.
A GRAPHICAL APPROACH TO FOURIER ANALYSIS
The focus of this paper is demonstrating
that the integrals of the proceeding pages are
unnecessary. One may establish the spectrum of
periodic waveforms "by graphical constructs alone.
the remaining residue, This is permitted by
equation (30).
The residue is itself a chain of delta
functions. However, these differ from the first
(3 D chain in two respects: they are negative spikes,
the first chain. Their frequency spectrum is
nonetheless the same: a fundamental and all
harmonics of the same amplitude. Note that the
step giving rise to the negative chain is the
same in amplitude as that giving rise to the
positive spikes.. That is, the integral across
either a positive or negative spike has the same
magnitude. Therefore, relative amplitudes of the
two spectra are identical.
Now expand Table I to include the negative
Table II results.

Reconsider now the square wave of Figure 1.

Could one guess the spectral content of this TABLE II
waveform at a glance? Probably not - at least
not yet. True, the fundamental frequency is HARMONIC ANALYSIS
obvious, and one supposes that many harmonics
exist. Harmonic
1 2 3 4 5 6 etc
number
The Quick Transform procedure requires that
the waveform be converted to impulse functions. Amplitude 1 1 1 1 1 1 1 etc
Let us sketch the time derivative of the square
wave. The dual set of delta functions as in Amplitude 2 1 1 1 1 1 1 etc
Figure 4 result. position fi 180 360 180 360 180 360 etc
time fi 180 180 180 180 180 180 etc
total fi 0 180 0 180 0 180 etc

Grand Total
time Amplitude 2 0 2 0 2 0 etc
-T Phase 0 - 0 0 - etc
FIGURE 4
What is the significance of these new
entries? Let's take them one at a time.
Equation (30) permits us to divide this waveform,
into two separate functions, the positive spikes The negative chain of spikes produces
and the negative spikes. Analysis is performed harmonics whose amplitudes, relative to the first
on each set separately. Resulting spectra are chain's harmonics, are listed as "Amplitude 2."
then summed to obtain the spectrum of Figure 4* But this chain is positioned in time onehalf
cycle late compared to the first chain. Thus,
Analysis of the positive-going chain is each harmonic is delayed by one-half period at
simple. We have already done so, and memorized the fundamental frequency, a constant time.
the result that all frequencies are present and Therefore, the fundamental suffers a l8 o degree
of the same amplitude. Further, each cosine wave phase shift, the second harmonic suffers a 360
is phased to have a maximum at the position of degree shift (the delay time is constant) and so
the spike. This result is tabulated in Table I. on. These shifts are shown as "position f i . "

The second chain of spikes is negative.

TABLE I This causes each cosine-wave in its spectrum to
be negative, too - they must all add to produce
HARMONIC ANALYSIS the spike. Thus each harmonic in the spectrum
is a negative cosine wave, and is shifted 180
Harmonic degrees as a result. These shifts are recorded
number 1 2 3 4 5 6 etc.
in the row marked "time f i . " The shift is exactly
180 degrees for each harmonic, and is not
Amplitude 1 1 1 1 1 1 etc. dependent on harmonic number.

The total phase of each harmonic is the sum

The first chain of delta functions has now of the position and time phase shifts. These
been considered. Remove it from the picture - results are recorded in the "total f i " row.
literally, erase it from Figure 4 - and analyze

77
 We are now in a position to generate the f ' ( t ):
complete spectrum of Figure 4* We add veetor-
ially the "Amplitude 1" and "Amplitude 2"
components. The phase of Amplitude 1 is always
zero, for we chose time to he initiated at the time in
position of the first chain of spikes, which we cycles
chose as a positive chain. The vector sum of
()J3 (d)
spectral components is tabulated in the rows 1/4 1/2 3/4
"Grand Total Amplitude" and "Grand Total Phase."
FIGURE 5B
The last two rows of Table II give us the
spectrum of Figure 4 's waveform. But we want the
spectrum of Figure 1, a square wave. Recall now f"(t): m = 2
equations (27) and (28). What is required is the
9 0 degree retardation of each component of the h
(a)
spectrum, and a l/n reduction of the amplitude time in
of the n-th harmonic. The spectrum of a square
wave then follows. It contains: |(h) 8
cycles

* A fundamental of 2 relative units amplitude. 0 li/4 1/2 3/4 1

* A third harmonic of 2/3 relative units.
FIGURE 5C (Proceed to Table V)
* A fifth harmonic of 2/5 relative units.
etc.
FURTHER OBSERVATIONS
Compare this result with equation (12).
One may note that the Quick Transform does
Qualitatively, they are identical.
not always work. In fact, a large class of
functions exist such that no derivative produces
THE QUICK TRANSFORM
spikes. Consider the sine wave, for example.
Does the Quick Transform, then, have application
We now have a handy method of doing Fourier
only in certain situations? Seemingly so.
Analysis without resorting to integrals! The
procedure can be condensed to a "cook-book"
Consider, though, a "smooth" function - one
recipe. It is:
whose derivatives of all order exist, A type as
in Figure 6 might satisfy this constraint.
THE QUICK TRANSFORM Figure 6 also demonstrates the result of a "step
and hold" sampling circuit at work on the
1) Sketch the time-variant waveform. function. This waveform is an ideal candidate
2) If it contains no spikes, sketch its for the Quick Transform.
time derivative.
3) Continue sketching derivatives until
spikes appear. Record the highest order
of derivative sketched as "m."
4) In tabular form record the relative
amplitude and phase of each harmonic,
using the known spectral property of
delta function chains.
5) Add spectral components of the table
vectorially.
6) Divide the n-th harmonic's amplitude
by nm *2Tt/ t for each harmonic.
7) Retard the n-th harmonic's phase by
n* 9 0 for each harmonic. FIGURE 6
8) Remove the spikes from the sketch.
If a residue remains, continue this
F'(t)
procedure.

A TACIT EXAMPLE

f(t) time
jL - l
4 t | r
t t +T
0 o
2'

0
\s time in
0 1 /4 1 /2 3 /4 1 cycles 1
FIGURE 5A FIGURE 7

78
 TABLE V The transform of Figure 7 is oBvious. Each
HARMONIC ANALYSIS OP FIGURE 5 spike may Be treated separately, and terms added
to a taBle of harmonic analysis. When all spikes
Harmonic have Been accomodated, it is necessary only to
number 1 2 3 4 5 6 etc. sum vectorially the terms in the table, divide
each harmonic amplitude by its harmonic number,
m = 1 components of f (t) and retard each phase shift By 1 / 4 cycle. One
derivative is all that is needed, as no residue
Amplitude (c) 3 3 3 3 3 3 etc. exists after the above analysis.
position fi 3/4 6 /4 9/4 1 2 /4 15/4 1 8 /4
time fi l/ 2 1 /2 1 /2 i/ 2 EXPLORING A N ALGORITHM
!/2 1 /2
total fi 1 /4 0 3/4 '2/4 1/4 0
The discussion above suggests a simple and
Amplitude (d) 1 1 1 1 1 etc. general algorithm. A Fourier Series expansion,
position fi 4/4 8 /4 12/4 16/4 20/4 24/4 itself an approximation to a Fourier Transform,
time fi 1 /2 1 /2 i/ 2 1 /2 !/2 !/2 could be computed quickly as follows.
total fi 1 /2 1 /2 1 /2 1 /2 1 /2 1 /2

m = 2 components of f" (t ^
GENERAL ALGORITHM
THE QUICK TRANSFORM
Amplitude (a) 8 8 8 8 8 8 etc.
position fi 0 0 0 0 0 0 1) Approximate the function in a "step and hold"
time fi 0 0 0 0 0 0 format.
total fi 0 0 0 0 0 0 2) Differentiate this approximation. Label each
spike with its integral's amplitude.
Amplitude (b) 8 8 8 8 8 8 etc. 3) Construct a table of harmonic analysis. Enter
position fi 2/4 4/4 6 /4 8 /4 IO/ 4 12/4 therein the contributions of each spike.
time fi 1 /2 1 /2 1 /2 !/2 1 /2 !/2 4 ) Vectorially add all harmonic contributions.
total fi 0 1 /2 0 1 /2 0 1 /2 5) Divide each amplitude by the harmonic number.
6) Shift each harmonic's phase by -90.
m = 1 components of f"( t l
divide by w = 27]/T = 2 TT At this writing such algorithms are being
divide by n explored for application on various small systems
delay fi by 1 / 4 cycle including the MC6800 microprocessor.

Amplitude (a) 1 .2 7 .637 .424 .318 .255 .2 1 2 CONCLUDING REMARKS

total fi 1/4 1/4 1/4 1/4 1/4 1/4
It is hoped tha t.the non-integral method of-
Amplitude (b) 1.27 .637 .424 .318 .255 .2 1 2 sketching Fourier Transforms, or more accurately,
total fi 1/4 3/4 1/4 3/4 1/4 3/4 Fourier Expansions, will be of assistance. Math
has been reduced to a matter of addition.
From above,
The method, while possessing shortcomings, allows
rapid pencil-and-paper analysis differing little
Amplitude (c) 3 3 3 3 3 3
total fi from "doodling." If we must doodle, it may as
1/4 0 3/4 1 /2 !/4 0
well be useful.
Amplitude (d) 1 1 1 1 1 1
total fi 1 /2 1 /2 ACKNOWLEDGEMENTS
1 /2 1 /2 1 /2 1 /2

VECTOR SUM I thank various professors whose style and

5.63 2 2.37 4 3.65 2
teachings remain vivid, but whose names have
evaporated from memory in the intervening years.
m = 0 divide By 2 V ; divide By n They set the foundation on which this analysis
slowly rose.
VECTOR SUM .159 -126 .159 .216 .053
In particular I thank my colleague, Hector
Davis, for his many patient hours spent in
discussing this approach with me. Without such
exchange ideas do not take root.

We may sketch the stepwise approximation to REFERENCES

obtain its derivative, Figure 7. Although each
spike is infinite, they have been drawn to a scale 1. I.S. Sokolnikoff and E.S. Sokolnikoff,
"Higher Mathematics for Engineers and
indicative of the magnitude of their integral.
Physicists," McGraw-Hill, 1941. pp 63-72

79