IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPAT\I11ILITY
3 DECEMBER 1967 123"
[6] L. Linderer et al., "Mileasurements oln two Nb superconductive
RF cavities," Phys. Letters, vol. 2, pp. 119-120, September 1962.
[7] C. R. Haden, J. M. Victor, and W. H. Hartwig, "R. F. residual
losses in superconductors," presented at 1966 Southwestern IEEE
Conf., Dallas, Tex., April 20-22, 1966.
[8] W. H. Hartwig and D. Grissom, "Dielectric dissipation mea-
sturements below 7.2K," in Low Teooperature Physics. New York:
Plenum Press, 1965.
[9] R. J. Allen and N. S. Nahman, "Analysis and performanice of
superconductive coaxial transmission lines," Proc. IEEE, vol. 52,
pp. 1147-1154, October 1964.
[10] T. A. Buchold, "About the nature of the surface losses in
superconductors at low frequencies," Cryogenics, vol. 3, pp. 141-149,
September 1963.
The Significance of Negative
Spectrum Analysis
ROBERT B. MAARCUS,
Abstract-In EMC measurements spectra generated by pulsed
signals are often measured. To calculate the spectra theoretically,
Fourier methods are usually used. The Fourier transform yields a
spectrum which contains positive and negative frequencies. The
exponential form of the Fourier series also yields a spectrum which
contains positive and negative frequencies. However, the trigono-
metric form of the Fourier series yields a spectrum containing only
positive frequencies. Since there seems to be some doubt about the
physical interpretation of negative frequencies, the relationship be-
tween the double sided spectrum of the exponential Fourier series
and the single sided spectrum of the trigonometric Fourier series is
shown. The reasoning is then extended to the Fourier transform to
show the relationship between the double and single sided spectrum
obtained by the Fourier transform. A method of obtaining spectral
levels for positive frequencies only using the Fourier transform is
shown.
IN E1\JC WORK it is ofteu necessary to measure pulsed
spectra. In theoretical studies associated with the
measurements, Fourier analysis has been used for many
years to predict the spectra theoretically.
In Fourier analysis one canl use Fourier tranisfoims or
Fourier series. The Fourier series gives a spectrum which
cointaiins discrete spectrunm linies. It is usually cumbersome
to perform mathematical operations OIl each spectrum
line to determine the spectral characteristics of a pulsed
signal or to determinie the output of circuits whose input
consists of pulsed signals. It has been found much more
convenient to use the spectrum of a single isolated pulse to
Mlanuscript received May 13, 1967; revised August 11, 1967. This
w ork was supported uinder Air Force Contract 19(628)-5049.
The aujthor is with the IIT Research Institute, Anniiapolis, Md.
Authorized licensed use limited to: ECOLE POLYTECHNIQUE DE MONTREAL. Downloaded on November 10, 2009 at 18:24 from IEEE Xplore. Restrictions apply.
VOL. EMC-9, NO.
[11] C. P. Bean, "M\Iagnetization of high field superconductors,"
Rev. Mod. Phys., vol. 36, pp. 31-39, January 1964.
[12] C. P. Bean et al., "A research investigation on the factors that
affect the superconducting properties of materials," General Electric
Research Laboratory, Schenectady, N. Y., Tech Rept. AFAIL-TR-
65-431, March 1966.
[13] S. Becker et al., "Interference analysis of new compon-ents anid
circu'its," Airbourne Instruments Laboratory, Melville, N. Y., Tech.
Rept. RADC-TDR-64-161, 1964; see also "A rule of thumb for
prediction of third-order intermodulation," IEEE Spectru7n, vol. 1,
p. 5, May 1964.
[14] K. Siegel, R. Domehick, and F. Arams, "Superconducting
High-Q radio-frequiency eircuit usin g niobium stannoide above
4.20K," Proc. IEEE, vol. 55, pp. 457-458, March 1967.
Frequencies in
SENIOR MEMBER, IEEE
determine the output waveshape or other signal param-
eters of a circuit whose input is a pulsed signal. The
Fourier transform gives the spectral level of a single
isolated pulse. If the actual input to the circuit is a repet-
itively pulsed signal, the output signial due to a single pulse
of the pulse train can be founid by the Fourier tranis-
form method provided that the circuit has no residual
currenit or voltage from the previous pulse.
If any standard textbook on Fourier trainsforms were
investigated, it could be seeni that the spectra found by
means of Fourier transforms conitain negative anid
positive frequencies. The question arises, what are
negative frequencies and what siginificanrce do they have
in spectrum analysis? In RF measurements negative
frequenicies cannot be defined. Are the negative frequenicies
merely discarded and looked upon as mathematical
curiosities?
The negative frequeiicies cannot be discarded math-
ematically. The following expositioni will help shed some
light on the significance and the definitionr of niegative
frequencies in spectrum analysis.
Let us first examine the Fourier selies in exponienitial
form. It will be shown that the spectrum described by the
Fourier series has negative frequeiicy componieiits in the
same sense as the spectrum described by the Fourier
transform. The relationship betweeni the niegative fre-
queincies in the two spectrums will be showni later.
The Fourier series in exponential form is
f c
f = (t)n=-coaE ico
n9t(
 124 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY DECEMBER 1967

where If a time function f(t) is real, then

f(t) = any periodic function of time a-n = an (3)
WT = 27rfr where
f = frequency of periodic function
E T/2 conjugate of an
atn =-
,f(t)
inw
'trdt (2) a,
T/2
O = TJf(t) e-inw rdt.
E = maximum amplitude of periodic function of time
T = period of periodic function.
If an were to be put in trigonometric form, we would ob-
An examination of (1) will show that a periodic function tain
of time consists of an infinite sum of discrete spectral lines;
E fT/2
the frequency of each line is nfT. Since n has all integer atn = T - ff(t) (cos ncw,t - j sin ncw,t) dt. (4)
values from minus infinity to plus infinity, the spectrum JT/2
mathematically exists in the negative frequency region as
well as the positive frequency region. However, when n The real part of an will be called Rn and the imaginary
equals zero, the frequency is zero which indicates the part of a,, will be called Xn; therefore
spectrum has a dc component. Some of the spectral lines
inieluding the dc component may have zero amplitude and Rn=T f(i't) cos nwo,t dt (5)
thus vanish. If the nature of each spectral line is examined ;T2

it can be seen that each line is a time function whose E T/2

amplitude is ao. The time varying part of the function
T-T12 f(t) rnw,t
Xn = -- sin dt (6)
is En T. The time function is not a sinusoidal time function
but it is a rotating vector whose amplitude is an and whose an - Rn + jXn (7)
angular frequency is nw,. Equation (1) shows that n takes *

on all integer values from minus infinity to plus infinity. n= (8) -n iX7, = at-

In the negative region the angular frequency of the

rotating vector is negative. The negative frequency
can be interpreted as clockwise rotation of the vector X --X-n
while positive frequency represents counter-clockwise
rotation of the vector. There is no other significance to When n equals zero a new quantity ao will be defined as
negative frequencies. It should be apparent now that the E JT/2
exponential form of the series yields spectral components ao -o=- f(t)dt. (9)
T JT/2
which are not sinusoids but rotating vectors of various
amplitudes and frequencies. Since the amplitude of each Equation (9) defines the amplitude of the dc component
vector is constant it is not possible to show any direct of the spectrum.
physical interpretation of the spectral lines. The double- The series can now be written as
sided spectrum containing negative frequencies is only a -1 co

mathematical and not a physical entity. f(t) == ao +n=-coE anEjnwrt + E an jncrt (10)
n=1
Most of us are familiar with the Fourier series in trigo-
nometric form. The trigonometric form contains no Substituting (3) in (10) yields
negative frequencies. Further, the trigonometric form 00 00

shows spectral lines to be of sinusoidal form. Sinusoids f(t) = ao + E a*n 6 -n- rt + E an jnc rt (11)
n=1 n=1
have physical as well as mathematical meaning. Within
mathematical restrictions, which will be shown, the si- Substituting (7) and (8) in (11) yields
nusoidal form of the series is exactly equivalent to the 00

exponential form. The transition from the exponential f(t) = ao + E (RAn jX-n) e
form to the trigonometric form will be shown shortly. n=1
To be perfectly general mathematically, a function of
time can be real or complex. While a complex function of + E (Rn + jXn) Ejnlrwt
n=1
time has no physical interpretation, it has important
mathematical uses in Fourier analyses which will not be Putting the exponential in trigonometric form yields
covered here. The foregoing statements on going from
the double-sided spectrum to the single-sided spectrum
apply only to real time functions. Therefore, the restriction f(t) ao + (Rn jXn) (cos nw,t j sin nw,t)
= - -

n=1
on going to the one-sided or positive frequency only
spectrum from the two-sided spectrum is that the time
function has to be real for the transition to be valid. + n=1 (Rn + jXn)(cos nfwTt + j sin nfwrt)

Authorized licensed use limited to: ECOLE POLYTECHNIQUE DE MONTREAL. Downloaded on November 10, 2009 at 18:24 from IEEE Xplore. Restrictions apply.
 MARLCUS: NEGXITIVE FREQUENCIES IN SPIECTP'lUM A.N ALYSIS 125

coo
where
f(t) = ao +
ii= 1
Z [Rn cOs nw,rt -
Xnsi5 nw,)rt
E T12
an = -
T 1T/2
J(t) e- c.idt. (16)
- j(RI, Sill ncOt + Xn COS nflrt)] + E [R, COS nllOrt
n = 1
If f(t) is not periodic but a single isolated pulse, the
-n Sill lCOrt + j(RI Sill ncort + X,Z COS nwcrt) I time function f(t) can be written as
co

f(t) - ao + E [2R,, cos nWrt - 2Xn Sill nflrt]. (12) f(t) = F(27f) e i2wrftd (e17)
n 1
0

Equation (12) is the familiar trigonometric form of the where

Fourier series. It is exactly equivalent to the exponenitial
form providing the time function is real. Equationi (12) F(2wrf) f (t)ej2ft.
containis no negative frequencies. What has happened to
=
f-l
OD
(18)
the uegative frequencies? The negative frequency came Equations (17) and (18) are analogous to (15) and (16).
about by considering the time varying trigoniometric It is easier to visualize the functioni of time in (17) as
fuuctioins to be made up of two vectors rotating in opposite being the integral of a continuous spectrum whose ampli-
directions. There is no physical meaniing to the negative tudes are described by (18) and whose frequency corn-
frequencies. ponents are described by the exponential part of (17).
The Fourier transform will lnow be conisidered. If onie The variable in (17) and (18) can be changed from f to
has a single isolated pulse, the spectrum resulting there- o, thus
from can be described by means of the Fourier transform.
The spectrum has uo discrete lines but is coiltilluous. The co= 27r J (19)
Fourier transforms describe a spectral level in volts per
Hertz at every frequency. dw = 2wr df. (20)
The direct Fourier transform is Multiplying and dividing (17) by 27r yields
F(w) f(t) -wdt. (13)
F(27rf) E2' 7f2wdj.
=

f(t) =22 7r _co

(21)
Equation (13) gives the speetral level of the spectrum at
any angular frequenicy co. Substituting (19) and (20) in (21) yields
If the spectrum is knowni the time funietion canl be
founid from the iniverse tranisform which is f wtdw.
_co F()We
f(t) =
2r (14)
(t)
7=JI F(w)ejictdw.
22l r -0
(14) Substituting (19) in (18) yields
MNIathematicians coinsider (13) aild (14) as a tranisformii F(co) = f(t) -ij-tdt (13)
pair; (13) gives an amplitude spectrum from a time _C

fuinctioni anid (14) gives a time functioni from ain amplitude It cail IlOW be said that
spectrum.
However, an amplitude spectrum has ino physical I
meaning to the enigineer. Furthermore, it seems very f(t) 2r _ coF(w)Uw't dco = F (14)
strange that one can obtaini a time fuiletioin by iintegratinig
an amplitude funetion. In (14) the amplitude function wheire
F(w) is multiplied by a time funietioni EjXt; therefore, the
illtegral yields a time funietioni. The poillt is that the F(co) f(t) -i@tdt = E (13)
mathematicians have separated the amplitudes of the _co

spectral componients from the time funetioni and have and

pgotteii a transform pair. Enginieers would like to coinsider
spectral compoinents as time funictionis with associated F(co) = R(w) + jX(w) (22)
amplitudes and not just as disassociated amplitudes.
Let us reconsider the expoinential series R (c) = f(t) cos wt dt (23)
00

f(t) = E a1nE (1)

f f(t) sin cot dl.
n = -co
X(co) -
_co
(24)
If 27rf, is substituted for Cor, then
The spectral components are F(CO) E i'. These com-
poneints are not discrete lines but are continuous. The
co

f(t) = a,, e j27rfrt (15)

n =-o spectral components are time functions and mathemati-

Authorized licensed use limited to: ECOLE POLYTECHNIQUE DE MONTREAL. Downloaded on November 10, 2009 at 18:24 from IEEE Xplore. Restrictions apply.
 126 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY DECEMBER 1967

cally exist at negative as well as positive frequencies. How- Equation (26) is the Fourier transform for the spectrum
ever, just as in the exponential series, the time functions in trigonometric form. The frequency limits of the spec-
are rotating vectors which do not change amplitude with trum range from minus infinity to plus infinity and there-
time. Again if only real time functions are considered, then fore the spectrum is double-sided. But since R(w) =
R(-w) and X(co) = -X(-w), then
F(co) = F(-c,,)
R(w) =R(-) f(t) = I [2R(w) cos wt - 2X(o) sin tl]dc. (27)
X(W) = -X(-Cw). Equation (27) is the Fourier transform for the single-
Substituting (22) in (14) yields sided spectrum. It contains no negative frequencies.
It can be seen that the amplitudes of the spectral com-
1I co
f(t) =- [R (X) + jX (w)]Ie Jtdw. (25) ponents of the one-sided spectrum are exactly double
2r co -
those of the two-sided spectrum. It must still be remem-
Putting f-Xt in trigonometric form yields bered that the foregoing applies only to real time func-
tions.
f(t) = I-
2 7r _co
[R(w) + j'X(G)](cos wt + j sin wt)dw Negative frequencies are associated with time functions
representing clockwise rotating vectors and positive
I
frequencies are associated with time functions representing
f(t) = {R(w)cos wt -X(w) sin cut counter-clockwise rotating vectors. These are math-
2 r oo
ematical representations of the frequency components of
+ j[LXo) cos wt + R(w) sin wt]}dw the spectrum and do not have direct physical interpreta-
tion. The mathematical combination of the two oppositely
since
rotating vectors produces sinewaves. The sinewave
X(W) = w) components of the spectrum have physical meaning and
it is these sinewaves that constitute the actual physical
anid spectrum. It has been shown how to combine the rotating
R(w) (-cw) vectors mathematically to produce a spectrum of sine-
it follows that waves which contain no negative frequencies. Therefore
negative frequencies are mathematical tools which have
[X(w) cos cot + R(w) sin wt]dw O. =
no direct physical interpretation.

Therefore REFERENCES
[1] A. Papoulis, The Fourier Integral and Its Applications. New
York: McGraw-Hill, 1962.
f(t) = 2 J [R(w) cos wt -
X(w) sinl wt]dw. (26) [2] M. Javid and E. Brenner, Analysis, Transmission, and Filtering
2 r co0 of Sign,als. New York: McGraw-Hill, 1963.

Authorized licensed use limited to: ECOLE POLYTECHNIQUE DE MONTREAL. Downloaded on November 10, 2009 at 18:24 from IEEE Xplore. Restrictions apply.