You are on page 1of 22

106106 ~~~~~IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. EMC-19, NO.

3, AUGUST 1977

Statistical-Physical Models of Electromagnetic Interference

DAVID MIDDLETON, FELLOW, IEEE

Abstract-Most man-made and natural electromagnetic interfer- These three tasks, in turn, are critical elements in any adequate
ence, or "noise," are highly non-Gaussian random processes, whose de- program of spectrum management (for example, [171 ).
grading effects on system performance can be severe, particularly on Our aim here, then, as earlier in this series (cf. Part I, [ 161 )
most conventional systems, which are designed for optimal or near
optimal performance against normal noise. In addition, the nature, is to provide analytical models 1) which combine the appro-
origins, measurement, and prediction of the general EM interference priate physical and statistical descriptions of general EM noise
environment are a major concern of any adequate spectral management environments; 2) which are analytically manageable; 3) which
program. Accordingly, this study is devoted to the development of ana- possess general canonical properties-i.e., are not specialized to
lytically tractable, experimentally verifiable, statistical-physical models individial noise mechanisms, source distributions, and emission
of such electromagnetic interference.
waveforms-and most important; 4) which are both experi-
Here, classification into three major types of noise is made: Class A
(narrow band vis-a-vis the receiver), Class B (broad band vis-a-vis the mentally verifiable and predictive. In addition, the basic or
receiver), and Class C (- Class A + Class B). First-order statistical "generic"! parameters of such statistical-physical models must
models are constructed for the Class A and Class B cases. In particular, be measurable quantities with specified physical structure and
the APD (a posteriori probability distribution) or exceedance probabil- interpretation. To achieve this is clearly a nontrivial problem,
ity, PD, vis; P1 (e > eo)A B, (and the associated probability densities, mainly because of the inherent, highly non-Gaussian nature of
pdf's wl(c)A B, [1]) of the envelope are obtained; (the phase is shown
to be uniformly distributed in (0, 2ir). These results are canonical, i.e.,
these random processes, a characteristic which at once predi-
their analytic forms are invariant of the particular noise source and its cates complex descriptions and resultinlg difficulties for the
quantifying parameter values, levels, etc. Class A interference is de- analysis of system performance. That these difficulties can be
scribed by a 3-parameter model, Class B noise by a 6-parameter model. effectively overcome for model-building i) and experimental
All parameters are deducible from measurement, and like the APD's verification ii) will be evident from the results and analyses in
and pdf's, are also canonical in form: their structure is based on the
general physics underlying the propagation and reception processes in- this paper (also [1], [16]). For receiver design and perform-
volved, and they, too, are invariant with respect to form and occurrence ance iii), this has already been established by recent work of
of particular interference sources. Spaulding anld Middleton [25] .
Excellent agreement between theory and experiment is demon-
strated for many types of EM noise, man-made and natural, as shown 1.] Classification ofEM nzoise
by a broad spectrum of examples. (Methods for estimating the canoni-
cal model parameters froms experimental data, essentially embodied in General EM noise or interference environments can be con-
the APD, are outlined in some detail in Section 6 of [1]1.) viently classified into three broad categories of interference
vis-a-vis any narrow-band receiver.1
PART I: INTRODUCTION, RESULTS, AND CONCLUSIONS Class A Interference: This noise is typically narrower
spectrally than the receiver in question, and as such generates
I. INTRODUCTION ignorable transients in the receiver's front-end (i.e., initial
linear stages, viz, aperture-RE-IF) when a source emission
A S in previous studies (for example, [14] -[16]) our cen- terminates.
Atral problem is to construct analytically tractable models Class B Interference: Here the bandwidth of the incoming
of man-made and natural radio noise. This is done for three noise is larger than that of the receiver's front-end stages, so
principal technical purposes: that transient effects, both in the build-up and decay, occur,
i) to provide realistic, quantitative descriptions of man- with the latter predominating. The receiver is to varying
made and natural electromagnetic (EM) interference envi- degrees "shock-excited," particularly for inputs of very short
ronments; duration, so that the receiver is said to "ring."
ii) to specify and guide experiments for measuring such Class C Interference: This is the sum of Class A and Class B
interference environments; and, interference.
iii) to determine the structure of optimal communication For Class A noise, the transient decay period is negligible
systems and to evaluate and compare their performance vis-a-vis the emission's duration, while for Class B interference
with that of specified suboptimum systems, when operating it is highly dominant.
in these general classes of EM interference. The above three categories for interference, as it impacts on
a typical (narrow-band) receiver, e.g., as (the linear, front end
Manuscript received February 15, 1977. This paper is based on a of) that reciever responds to the EM environment, provide a
portion of a Technical Report [11 prepared for the Office of Telecom- useful way of describing the different effects which these
munications, Department of Commerce. This paper was presented at
the Int. Symp. on Electromagn. Compat., Montreux, Switzerland, June different categories have on reception. This categorization is
28-30, 1977.
The author is a Contractor to the Office of Telecommunications, lThis classification can be broadened to include receivers of arbi-
U.S. Department of Commerce, Boulder, CO 80302. (212) 831-8565. trary bandwidth (cf. footnote, p. 4, [ 11]).
MIDDLETON: MODELS OF ELECTROMAGNETIC INTERFERENCE10 107

I Receiver of mathematical simplicity, but severely limited in usefulness,


. Narrow -Band by lack of generality and physical insight, and a concomitant
Signals) Mt'/R -T Processing ?DMecission dependence on local, empirical data, and circumstances. Two
important exceptions to the above are the work of Furutsu
EM ...
'Aperture x RF x IF.____ and Ishida [71 on obtaining the APD's (and associated proba-
"Environment"'
Z^ N (Class A MfN < Af (or matched) and/or bility densities (pdf's)) of atmospheric noise under rather
broad conditions, and the more recent studies of Giordano [8]
iiClass C Class A + Class 8 and Giordano and Haber [9], similarly directed to atmospheric
IInterference
Source(s) noise. Both sets of investigations, however, are (necessarily)
constrained to Class B types of interference (cf. Section 1.1
Fig. 1.1. Schema of EM interference and desired signal environment above), and neither attempts the canonical formulation, which
vis-a-vis typical narrow-band receiver. is a key feature of our current efforts [14] - [16], [1]. This
canonical formulation allows us to apply the new models
useful because receiver response is statistically different for formally by Class (A, B, etc.) to all types of (EM) interference,
each class. Here we use the so-called APD's (a posteriori unrestricted in general structure by the particular physical
probability distributions (cf. [231) such as P1(X > XO), or mechanism involved. (These latter, of course, determine the
P1 (E<E0), which are the respective probabilities that the generic properties of the model parameters, and must be
instantaneous amplitude, or instantaneous envelope, observed specifically introduced into model building if the ad hoc and
at the receiver's IF output exceed some threshold XO, or e0, as arbitrary empiricism of much of the earlier work is to be
these latter are allowed to assume values in the interval (-00, avoided.)
oo), or (0, co). The conditions "spectrally broader than," and
"spectrally narrower than," cf. F;ig. 1.1, are to be interpreted 1.3 New Results
as "sufficiently broader or narrower," etc., where in any case, The principal new results- of this study are: canonical,
care is taken to refer to the defilnitions of Class A, B, etc., in analytical, first-order statistical models of both Class A and
terms of the residual transients, versus the "on"-time of the Class B interference, specifically for the envelope (B) and
input emission which appears at the output of the IF stage of phase (4/) of the narrow-band output of the composite aper-
the receiver in question. ture-RF-IF stages of a typical receiver. These models are based
It is instructive to extend our schema of classification on a general physical mechanism (cf. [ 16] , Section 2) providing
further, in order to distinguish between man-made and natural insight into the parameter and pdf and PD structure. In addi-
interference, and between "intelligent" and "nonintelligent" tion, the general-method of approximating the governing
emissions. Accordingly, we define the following: (1st-order) characteristic functions (c.f.'s) is described, which
enables us to obtain the required canonical structures which,
i) "intelligent" noise or interference is man-made and in turn, give the resulting analytical models their broad appli-
intended to convey a message or information of some sort; cability, unrestricted by particular physical mechanisms, con-
whereas, trolled only by the underlying Poissonian postulate of
ii) "nonintelligent"} noise or interference may be attribut- independent source emissions in space and time (cf. Sec-
able to natural phenomena, e.g., atmospheric noise or tion 2.1, Part II). Included, also, are specific procedures for
receiver noise, for example, or may be man-made, but con- determining the model parameters from experimental data.
veys no intended communication, such as automobile Excellent agreement with experiment is found, and a variety
ignition, or radiation from power lines, etc. of comparisons of theory with experiment is included, involving
(We remark again (cf. [12], Section 1.3-52) that by definition, many different physical types of radio interference, demon-
"noise" or "interference" is any undesired "signal" at or in the strating the canonical character of the approach, as well.
receiver, regardless of origin.) The importance of distinguishing Finally we observe that the definition of Class A models, and
man-made from natural noise lies in the fact that the former is their quantitative identification with observed noise processes
potentially controllable, sometimes to the point of elimina- are new, although, of course, such interference has been
tion, whereas the latter cannot be eliminated at the source, physically present for many years. Class B models are
and is usually not subject to control: one can seek only to "classical," although not so designated until now, but here,
investigate its effects on the communication process. More- again, our present approach is to a large extent original,
over, the distinction between "intelligent" and "nonintelli- particularly with regard to canonical results. Finally, this paper
gent" is always significant with regard to information transfer: is divided into two principal units: Parts I and II. Part I con-
the taxonomy of the former can have greatly different tains introductory, background material (Section I), and in
implications and consequences from that of the latter (cf. Section II a summary and discussion of the main results.
Table 1.1, p. 8 of [1]). Part II, on the other hand, provides the highlights of the
analytical development of the theory.
1.2 Earlier Work
FOr the most part, earlier efforts at modeling man-made II. RESULTS AND CONCLUSIONS
and natural noise (principally atmospheric noise) have pro- Let us now summarize the principal results of the analysis
duced a wide variety of analtyical results, often with the virtue in Part II of this paper.
108108 ~~~~~IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. EMC-19, NO. 3, AUGUST 1977
2.1 Phase Statistics: The physical significance of these global parameters (AA, rA,
In the general case we may use (2.7), Part II, and the rela- Q2A) is briefly stated.
tion w1(4') = fO w1(E, 4i)dE, to obtain the pdf, and the APD 1) AA = the impulsive index (for Class A interference): this
is defined as the average number of emission "events" imping-
(= fsz w1(4')d4, (0 . 4' < 2rr)), of the instantaneous phase 4', ing on the receiver in question times the mean duration of a
which will not generally be uniform [in (0, 2rr)] . However, in
the truly narrow-band situation of Section 2.2, we obtain the typical interfering source emission [cf. (2.16), (2.18), Part II,
well-known uniform pdf [(2.13), Part II, e.g., w1(4') = 1/2ir, and associated discussion]. The smaller AA, the fewer such
(0, 2ir)] as in the simpler Gaussian examples. Because of this events and/or their duration, so that the noise properties are
first-order simplicity for the statistics of the phase, we accord- then dominated by the waveform characteristics of a typical
ingly concentrate our attention on the (first-order) statistics of event. Loosely speaking, we say that such noise is "impulsive,"
the associated envelope, which, as expected, departs radically although here the mean duration of events is sufficiently long
from a Gaussian (i.e., Rayleigh) behavior. to avoid generating noticeable transients in the receiver, i.e.,
we have Class A noise, as defined above, Section 1.1. AS AA is
made large, one approaches Gaussian (or in the case of the
2.2 Envelope Staltistics. the APD and pdf envelope here, Rayleigh) statistics.
for Class A Interference 2) rA'- aG KZ2A = the ratio of the intensity of the
independent Gaussian component UG2 of the input interfer-
Our principal analytical results here are: i) the characteristic ence including received "front-end" noise, to the intensity
function (c.f.); ii) the APD, or exceedance probability P1(e> Q2A of the "impulsive," non-Gaussian (or Rayleigh) com-
e0); and iii) the associated pdf, w1(e). These are, respectively,2 ponent. A portion, UE2, of this normal component [cf. (2.21)
(c.f,): et seq., Part II] arises from the cumulative effect of a large
number of external sources, none of which is so strong as to
Fl(iaX)A-e AA A emA 2a2X2 be considered part of the "impulsive" interference, which is
m-0 m! statistically the dominating effect (for small and moderate
indexes, AA )-
a2 = [2Q22A(l + A )I] (2.1)
3) Q2A = the intensity of the above-mentioned "4impulsive"
component.
with 2&mA 2 = (m/AA PA ')/(l PA '), and Characteristic behavior of the APD P1-A versus e0, cf.
(PD): Figs. 3.1, 3.2, is exhibited by the "Rayleigh" form [constant
slope 71 = -2, on the linear by -(1/2) Olog1 (-loge [) ]plots
Pi(e>eo) ~e AA z ee -O2I2OmA 2 of P1 -A J for the comparatively small values of threshold e0,
i.e., large values of P1(e > CO)A, followed by a very steep rise,
after which p1- A bends over and approaches some asymptote
O e< oo< (2.2) with fixed slope ?7, 0 <r1 < 2, at large CO (small P1...A) less
than that of the Rayleigh behavior for P1_A in the 0.1 -1.0
cf. (3.3), Part II; and since w =--[dP1/dCo] 6eO-e region. Thus we have
(pdf):
P1-A o e aEO
A A m ee C2I/2(J mA 2%
The limiting, finite, and bounded slope as e0 becomes very
large, after the characteristic bend-over, reflects the physical
0.e<oo. (2.3) condition of finite total average energy. (See pp. 15, 16 of [1]
for a further discussion.)
Various curves of P1 (and w1) are given in Figs. 3.1, 3.2,
Part II ([1], (4.1), (4.2)) showing typical behavior for selected 2.3 Envelope Statistics: the APD and pdf
values of the (global) parameters (AA, PA). Here c, E0 are for Class B Interference
normalized envelopes The Class B interference requires a more extensive analyt-
eEE/ 2Q2A(l+ A)
ical model. This requirement arises because two canonical
characteristic functions (c.f.'s) are now needed to approximate
Co -E0/ /22A(l + rA ) (2.4) the exact c.f. (Section 2.6, Part II), one for small and inter-
mediate values of the envelope (0 S C . EB), the other for the
where Eo is some preselected threshold value of the envelope larger values (EB . C). The principal analytic results here are,
E. (Note that our normalization introduces a third param- accordingly,
eter Q2A-)
The parameters (AA, rA^, Q22A) which appear directly in
(c.f 's):
our statistical results for P1, w1 we call "~global" parameters. Fl(iaX^)B.1- b1Aa\KAGaA/
2 See the glossary of principal symbols, at the end of the paper. O.C.CfB (2.5a)
MIDDLETON: MODELS OF ELECTROMAGNETIC INTERFERENCE 109

F1(iaX)EH eA eexp[A Be b2aa2A22 -_ G 2aX2], Figs. 4.3, 4.4 for the pdf (2.8a), (2.8b), for selected parameter
values.] Again, e, c0 are normalized according to (2.4) above,
eB < e <m(2.5b) with Q22A replaced by 22BE etc.
There are now six global parameters for our model: (A,,
from (3.4a), (3.4b), Part II, with a2 = [2Q2B(I + B')]-1 a, AB, rB, Q22B; NI), (cf., [1], Section 6B (Part II)). The
now, and subset (AB, rBE, Q22B) are just as for Class A interference
above, with the physical significance described in Section 2.2.
(PD): The additional parameters required here are

pl (C > eO)B pl(e > eO)B-I, 0 < 6 <` eB


4) AuC 2r-(l-a/2) (BOB a)
,-AB=
P(E >eO)B-II, EB <C 0 < (2.6) r(1 + a/2) {2Q2B(1 + rB ')}"2
Here explicitly we have an "effective" impulsive index proportional to the impulsive
index AB, cf. (2.16), (2.18), Part II, which depends on the
1 (EA> "O)B -1 =Pl ( > 60)B -I generic parameter a. Here (BOB0) is the a-moment of the basic
00 (-l)nAn envelope of the output of the composite ARI stages, cf.
- eo
n
r (1 Fig. 1 .1 above, and (2.38d), Part II.

2--p 3__
.
lF, 1+ ; 2; -CA02 5 (2.7a) 5) a=
'I surface 'I Vol
,2
|
P(IE> o)B-II. CG2
Pl
e-AB
eB 0 Am
Bm e-0 2 A B2 spatial density-propagation parameter, cf. (2.37), Part II, et
4B2 m=O M
seq. Here p, y are, respectively, the power-law exponents
(2.7b) associated with the range dependence of the denisty distribu-
tion of the possibly emitting sources, and their propagation.
with A = A0I/20GB', so = (eoNI)/2GB, where (See (2.24), et seq., Section 2.4.2; (2.26), Part II.) The param-
eter a provides an "effective" measure of the average source
0 < a < 2; (cf. (2.37), Part II, et seq.) (2.7c) density
6) N1with range.
= the scaling factor which insures that P1I-, W1-I
2- yield the correct mean-square envelope 2&22B(l + rB') (see
2&mB2 (m/AB + fB')/(l + B ), AB = ( - AB Section 3.2.1).

(2.7d) 7) EB

GB22 2(lrB the (normalized) "bend-over" point, at which the two (approx-
imate) forms of PD (and pdf) are joined, according to the
cf. (3.6b), Parrt II). (2.7e) procedures discussed in Section 3.2.1, and (3.10)-(3.12), cf.
Fig. 3.5, Part II. This is an empirically determined point,
The associated pdf's are [1, (4.3), (4.6), part II] representing the point of inflexion (for small P1-B) at which
the experimentally determined PD, or exceedance probability
W1 (C)B-I Wl (})B-I P1(e > CO)B-exp, bends, e.g., at which d2Pl_expt./deB2 = 0.
We note that, without an (experimental) EB, we cannot predict
the limiting form of the PD as e0 -* oo; we can then only
A^ z -ln nt) obtain the subset of global parameters (A, a, rB', Q2B, NI)
(cf. [1], Section 6C (Part II)]. Examples of this are the
1F1(l + no/2; 1; 42), 0 Se6 EJB (2.8a) particular cases of atmospheric and automotive ignition noise
shown in Figs. 2.5, 2.4, and interference from a fluorescent
C light, Fig. 2.3, in which last insufficient data were available to
W1(C)BE 4GB 2 m-O m
A!mE
reveal the finite slope. This is not the case, however, for the
data of Figs. 2.4, 2.5.
EB e<00 (2.8b) The six parameters (A0, at, AB, E' Q2B2, NI) are all
physically specified and measurable parameters in the analyt-
with 1F1, as usual, a confluent hypergeometric function [12, ical model (provided EB is determined). Only EB itself is an
appendix A.1.2], so that the wl(e)E = wl(e)B- Tfor 0 < e < empirical parameter, without explicit quantitative relationship
EB, while wl(EB) = Wl( )B-I1I when eC>E eBI [n Part I, [11, to the underlying physical mechanisms involved. This situation
Figs. 3.6, 3.7 show typical curves of the P 'D, (27), and exists because the simplest canonical approximation to the
110 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. EMC-19, NO. 3, AUGUST 1977

0 I I
10 6 MHz
V) -10 _ Horizontal Component E-W-
2 1.2 kHz BW
April24, 1973, 4:15p.m.
a-
a)
Shop- Office
E -20
Groce Mine
t-

E o 0 Denotes Measured
.O
0X -8 -30 Points
E
0 NI/2GB = 3.98 X 10-3
- 20
ax
EE
ax
a]
-40
.2
E A = 10-
_
a, -30I'
-50~ i= 1.85
.u -50
-v -3
I -40
a) -60
-ca,0 -70
0
cn -50 c
CR

.s

_U-6
-80 _
-60
II Il
II 01-, Io2 .1 10 20 40 60 80 90 95 98 99 0l-, Io2 10 20 40 60 80 90 95 98 99
Percent of Time Ordinate is Exceeded Percent of Time Ordinate is Exceeded

Fig. 2.1. Comparison of measured envelope distribution, P1(e> eO)A, Fig. 2.3. Comparison of measured envelope distribution, Pj(e>EO)B
with Class A model, cf. (2.2). Interference from ore-crushing machin- of man-made interference (fluorescent lights in mine shop office)
ery (data from Adams et al. [ 2 ] ). with Class B model, cf. (2.7a) (data from Adams et al. [2]).
40

20

-o

0
0

V9

m
E
ko
m

-20

-40

-60
1l4o102 10 20 40 60 80 90 95 98 99
99
Percent of Time Ordinate is Exceeded Percent of Time Ordinate is Exceeded
Fig. 2.2. Comparison of measured envelope distribution,P1(e > EO)A, Fig. 2.4. Comparison of measured envelope distribution P1(e > 60)B
with Class A model, cf. (2.2). Interference (probably) from nearby of automotive ignition noise from moving traffic with full Class B
powerline, produced by some kind of equipment fed by line (data model, cf. (2.7a,b) (data from Shepherd [22, fig. 14]).
from Bolton [4]).
MIDDLETON: MODELS OF ELECTROMAGNETIC INTERFERENCE III

4) the governing, physically structured parameters of these


PD's and pdf's which are likewise also canonical, can be
obtained from approximate experimental data (usually ex-
pressed as an APD).
The importance of the canonical character of these models
cannot be over-stressed: with such models we avoid the very
limited and nonpredictive quality of all ad hoc models, whose
structure must be verified and whose parameters provide little
or no physical insight into the underlying process itself. Fur-
ther, because these models are derived from physical principles,
their parameters are physically defined, are consequently
canonical, and are quantifiable in specific instances from
empirical data. Their structure, however, is independent of
any particular measurement.
X ~~~~Aa=1.0
a=i
Figs. 2.1 and 2.2 show APD's, e.g., P1(e > eO)A versus the
-20
normalized envelope threshold co, for Class A interference,
respectively from ore-crushing machining in a mine (data from
-30-
-0 i [2] ), and from a powerline (from [4] ). Observe the character-
Receiver No istic very-steep rise following the Rayleigh region (constant
slope), followed in turn by the expected bending over of the
-40-
APD for the rarer "events," in each case. [Similar examples of
Class A interference, but from man-made intelligent sources,
have also been observed.]
Other, typical APD's of Class B interference [for example,
-4 -2
10 10 10 20 40 60 80 90 95 98 99
Percent of Time Ordinate is Exceeded
Figs. 2.3, 2.4, [11, respectively, for i) primarily urban auto-
Fig. Comparison of measured envelope distribution P1(e > eO)B
2.5.
motive ignition noise [24] ; ii) atmospheric noise [6];
of atmospheric noise with full Class B model, cf. (2.7a,b) (data from
Esperland and Spaulding [6, p. 42] ). Spaulding, 1970]] are available. Fig. 2.3 shows an APD, for
flourescent lights, in a mine shop office [2]. Observe the more
exact c.f. [(2.38), Part II] requires a two-part c.f., approxi- gradual departure from the straight-line Rayleigh region, and
mate for one to the small and intermediate values of the the continuing rise, with constantly increasing slope, in the
envelope, and for the other, large to the envelope. values of the figures (which is equivalent to r7 -- 0 for exp (--2e30l), as
This second c.f., [and PD = P1(e > CO)B-II, provides the cx0).
-+ [In Fig. 2.3 the inevitable "bend-over" points EB, lie
needed "bending" of the APD curves for the rare events, as outside the range of data taken, e.g., for P1-B < 10-6, SO
sketched in Fig. 3.5, Part II, for instance, and shown in some that we are able to obtain all the global parameters, except for
of the experimental examples of Section 2.4. A fixed asymp- AB, cf. [1], Section 6C, Part II.] This is not the case, however
totic slope (r7 > 0) is required to ensure the existence of all for the Class B examples of Figs. 2.4, 2.5, e.g., respectively, for
moments, which in turn is demanded by the condition of i) ignition noise from vehicles moving on a freeway [22];
finite total average energy. Again, increasing the impulsive ii) atmospheric noise [6]. Here, the required bend-over of the
index and/or increasing the independent Gaussian component APD's is exhibited, along with the inflection points, EB. [In
(UG 2) eventually yields a wholly Guassian process (Rayleigh, these cases, we can obtain numerical estimates of all the six
of course, in the envelope), as expected. global (and hence all the generic) parameters characteristic of
each example of interference,3 man-made or natural, by the
2.4 Compatisons with Experiment methods briefly cited below in Section 2.5].
A variety of comparisons of our theoretical models with Figs. 2.1 -2.5 are typical of Class A and Class B interfer-
experiment, for both Class A and Class B interference, are now ence, man-made and natural. They are not intended to be
given, cf. Figs. 2.1 -2.5. Four significant features are at once
evident: 3Note that rB UG2/Q2B, where UG2 is the independent Gaus-
1) the agreement between theory and experiment is excel- sian component, which is different from the total Gaussian component
lent, i.e., the approximating forms are effective, analytical
'AG2 =UG 2 + b2aAB, cf. (2.39a). Thus, in Fig. 2.4, we must calculate
relations for predicting the desired first-order statistics;
522B from the data curve and then obtain UG2 from r. From the
other figures, all the remaining generic parameters are then readily
2) the canonical nature of our models is demonstrated: the found. On the other hand, for Fig. 2.5, Q2B occurs at 0 dB, by normal-
ization. Since P1 = 0.36 determines the total Gaussian component
form of the results [here APD's: P1(e > ce),], is invariant of (AaG2 for Class B, UG2 for Class A noise), from the data of Fig. 2.5
the specific source mechanism; we get aG2 = -17 dB (=2.10-2) and therefore rG = rB'E2B -
3) Class A and Class B interference are observably and 10 loglo (8 10 3) -21 dB, which gives, in turn bZaAB - 0.012
(in units of Q2B). Again, all remaining generic parameters are now
quantitatively different noise types (vis-a-vis the narrow-band
obtainable, from this and the other parameter data on Fig. 2.5 (also in
receiverused); units of Q2B here).
12 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. EMC-19, NO. 3, AUGUST 1977

exhaustive. Extensive additional APD data (mostly Class B) TABLE I


are available, for example, in Espeland and Spaulding [6], and Class A Class B
Bensema, Kanda, and Adams [3 ].
Finally, we distinguish two sets of model parameters: 1. New Models and Results; "Classical" (20 Yrs. Old), But New
Approach; New Results
a), the so-called global parameters, which appear explicitly in 2. 3 Global and 3 Generic Parameters 6 Global and Generic Parameters
the analytical forms for the APD's, etc., and b) generic param- (A
(AA..C2A): (A
(AA,' a
o(2BA)) B`5~(A,,A.' 2A`N
AB a2 IC, ), o Nj)
(An. "GI'BoB
eters, which are defined directly in terms of the underlying B
I

CB: empirical
imation
paramater of approx-
canonical, statistical-physical model. Once the global param- 3. All Moments X/R6\, 0<6 exist All moments t), 0<0 exist
eters have been estimated from the data, which usually 4. Insensitive to Source Distribu- Sensitive to Source Distribution and
tion in Space and Propagation Law; Propagation Law (a); Canonical
requires the calculation of the (first-order) APD, the generic Canonical Forms; Forms;
parameters can be calculated from them. The method for 5. Waveform in IF Output: "Gaps" in
Time [P, (e=)>)
Waveform in IF Output: no "Gaps" in
Time [e.g. P(e=0)=0];
obtaining the global parameters is described in detail in Sec- 6. No Gaps in Time if Gaussian No Gaps in Time (aa.0);
tion 6, Part II of [1]. We note that the Class B parameters Background
fX Gauss P.D. as Ag*-; [X Gauss P.D. as (A AB)-
(ct, (BOB a)), provide additional information about the emitting 'Y6 Rayleigh P.D.) and/or G a . Rayleigh P.). and/or o
sources, e.g., source density, basic waveshape, etc.Accordingly, 7. No Hall Models Exist; Hall Models for Special Values of c;
(Gauss Component Absent)
it is suggested that, to assess the interference environment more
fully, in addition to Class A measurements, when possible,
ARI receiver bandwidths also be selected to produce Class B
interference at the output of the ARI-stage, so as to obtain In the preceeding sections, we have summarized the princi-
oa and (BOBa), in addition to AB, UG2, (BOB2)), which are pal results of our present study of the (first-order) envelope
analogous to the corresponding Class A set. The important and phase statistics of man-made and natural electromagnetic
new parameter, a, gives us an estimate of an effective mean interference, whatever its physical origins and characteristics.
source density with range, and the acutal one (with range), These analytical models, of Class A and Class B interference,
if the governing propagation law (?y) is also known, or meas- are mathematically tractible and canonical in application: the
ured. Further information about source distributions may be forms of the results, and the number, type, and general struc-
obtained with the help of steerable directional beam patterns. ture of the associated parameters, are invariant of the particular
source. Of course, particular parameter values do depend on
2.5 Some Additional Results; General Comments the specific properties of the particular source involved. These
Additional results obtained in this study or in [1] are as are estimated in turn, by general procedures outlined here, cf.
follows. Section 2.4 above, and [1], Section 6, Part II, from experi-
1) First-OrderMoments, (er): These are obtained analytically mental data, principally the APD [= exceedance probability
for both Class A and B noise in [1] , Section 5 (Part II). They P1(e > eo)]. The canonical character of these models and their
exist for all (real, finite),, although the (approximate) expres- parameters is derived from the general underlying physical
sions for the Class B cases are necessarily more complex then structure upon which the models are based. This, in turn, is
for Class A. itself a general space-time model of propagation, source dis-
2) Conditions for Class A, B, C Noise: More precise, analy- tribution, and emission ([16], and Sections 2, and principally
tical conditions are derived in Section 7 (Part II) of [1], Sections 2.1, 2.4, Part II, here).
mutually to distinguish Class A, B, and C interference, than As expected, the resulting statistics of amplitude and en-
those qualitatively discussed in Section 1.1 above. velope are highly non-Gaussian (or non-Rayleigh), as the
3) Approach to Rayleigh Statistics: This occurs when either, analysis and examples in Part II and the experimental results
or both, the impulsive index or the independent Gaussian com- of Section 2.4 indicate. This fact necessarily has a critical
ponent becomes very large, cf. Section 2.2, Part II. effect on conventional receiver and system operation, which
4) Hall Models: A primary empirical model, constructed may be in conventional usage, (approximately) optimized, e.g.
earlier by Hall [10], is frequently used for ad hoc repre- "matched," to desired signals in Gaussian noise (so-called
sentations of the interference environment. Our Class B results, correlation receivers and their extensions), but which is
upon deletion of the additive Gaussian component (both from radically suboptimum for this kind of electromagnetic environ-
the impulsive and independent sources), can be shown to ment [25].
exhibit a Hall form, with Hall parameter (OH all = 2). (See [1] Here, we are concerned with first-order interference statis-
Section 3.2B, Part II; also [25], chapter 2.) Such models, tics themselves, not only for purposes of system design,
however, have a of
variety drawbacks, among them being their optimization, and comparison, but also for the tasks of meas-
ad hoc character, with the parameter(s) entirely empirical, and uring and assessing the properties of EM interference fields.
the nonexistence of the second moment, in many instances, as Excellent agreement between model and observation is found,
well as the nonexistence of all moments (e3), where 3 > as the examples of Section 2.4 above demonstrate. In addition,
OHall - 1. No Hall models are derivable from Class A models, explicit numerical results are also obtained for the global and
however. generic parameters of the interference phenomenon in ques-
5) Class A Versus Class B Interference: Some Summary tion, e.g., automotive ignition noise, communications,
Remarks: A concise comparison of some of the salient prop- atmospherics, machinery, power-line emissions, and the like.
erties of Class A and Class B interference is presented in These parameter values, along with the basic physical structure,
Table I. permit us to deduce general properties of the interference
MIDDLETON: MODELS OF ELECTROMAGNETIC INTERFERENCE 113

field, such as average source distribution in space (a), emission that, while the basic waveforms emitted all have the same
density in time (impulsive index, A), mean intensity (Q2), the form, their scale, durations, frequencies, etc., may be randomly
amount of external Gaussian noise (UG 2), etc., and of course, distributed. Our fundamental postulate of this basic interfer-
the associated APD, or exceedance probability P1(e > co), as ence model is that: i) the locations of the various possible
well as various moments ((eg)) of the interference process. emitting sources are Poisson-dsitributed in space; ii) the emis-
First-order statistics of these highly non-Gaussian EM noise sion times of the possible sources are similarly Poisson-
environments as embodied in the APD, P1 (c > co), P1 (X > distributed in time. Physically, this means that the sources are
XO) for example, are, however, minimal for the proper treat- statistically independent, both in location and emission. Thus,
ment of the general class of communication systems operating by a slight generalization of earlier results [13], [14], [16],
in such environments. In may situations, the performance we can write for the first-order characteristic function of the
bounds established from these first-order statistics are quite instantaneous amplitude, X, of the received interference
adequate [the independent sample cases of Spaulding and process
Middleton [25], for example], where higher order time struc-
tures are not significant. However, when they are, one clearly
needs appropriate extensions of the present models. In addi- F, (i;)x = expj ( | p(X, e) ei U(t e, 1) dX de| .

tion, the joint statistics of signals and noise are also required,
0
of first- and higher orders as well. Such extensions are part of
our continuing effort to develop an applicable analytic des- (2.1)
cription of the real-world EM non-Gaussian interference
Here e is an epoch, indicating vis-a-vis the receiver's (i.e.,
environment.
observer's) time t when a source may emit. The X = (X, 0, 0)
are coordinates, or a vector magnitude, appropriate to the
PART II. ANALYSIS geometry of the source field, located in the region A, and of
I. INTRODUCTION TO THE ANALYSIS the receiver, with dX (=dXdq) for a surface element; (=dXdOd4)
for a volume element. The quantity p(X, e) is the "process
Part II of this paper is devoted to the analytical highlights density" of this joint space-time poisson interference process,
of our statistical-physical interference models, where we are and is nonnegative, and can be regarded as proportional to a
concerned principally with the exceedance propabilities (PD's) probability density (cf. (2.28), [1] ). The ( O denotes a statis-
P1(E > EO), P1(e > co) of the envelope of the input noise tical average, e.g., f1 [w]w1(O)dO, over various random
process following the combined (linear) aperture-RF-IF (or parameters (0) which may be pertinent to our source model,
ARI) filtering stages of a typical narrow-band receiver. Three such as Doppler, source amplitude, and duration, etc. The U
principal classes of interference process are defined: Class A are the typical waveforms of the emitting sources, after recep-
noise, where (in qualitative language) the input noise is spec- tion by the (assumed linear) aperture-RF-IF stages of our'
trally narrower than the ARI-filter at the receiver's front end; "narrow-band" receiver. The received process X is gfven by
Class B interference, where the reverse is true-this input proc-
ess is spectrally broad vis-a-vis the ARI filter; and a general X(t)
Class C noise, which consists of the sum of Class A and Class B
= J U1(tIZ) dN(Z) (2.2)
Z(=AX O)
components. The present analytical presentation is a highly
condensed version of that given in Part II of the original where the {dIV}are a Poisson point process4 (in X and e)
Report, [1] . For details omitted, see [1]. Since we are interested here in the envelope of the received
process X, which is always narrow band, in as much as the
receiver is itself narrow-band, we have to consider the new
II. FORMULATION: THE CHARACTERISTIC FUNCTION random variables Xc, X5, representing the slowly varying "in-
In this section, we obtain the general forms of the first- phase" and "out-of-phase" components of X, viz.
order characteristic functions (c.f.'s) and probability densities
(pd's) and distributions (PD's) for the "impulsive" interference X(t) = Xj(t) cos wot + X8(t) sin coot
of the various man-made and natural sources described in
Part I above. - Re {(Xc - iXs)eiwot} = Re {Xoeiwot} (2.3)
Our first step is to derive the desired general forms of the
characteristic functions for the envelope of the received wave.
= Re {Eei(wOt PV)} (2.3a)
The next step is to take advantage of the various physical con-
ditions of the model, further to reduce our results to the where now wO(=27rfO) is the central (angular) frequency of
particular expressions appropriate to the Class A and Class B the final (=IF) stage of the receiver, and
interference, which can then be put in forms suitable for
evaluation. E-= NfXc2+Xs2; =tan- 1(X8/Xc) (2.3b)
2.1 The Basic Statisical Model 'For a general development of process statistics, not necessarily
We assume as before [16] for our basic model that there is limited to the Poisson case of independent sources, see, for example,
recent work [16], [18], in the development of generalized scattering
an infinite number of potential sources in a source domain and models.
114 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. EMC-19, NO. 3, AUGUST 1977

therefore Similarly, the PD, and exceedance probability, or APD (a


posterior probability here, that E exceeds a level E0(>O)) are
XO = Xc - iXs = Ee "O'. (2.3c) defined as usual by

Here E, ; are, respectively, the envelope and phase of the DEo


narrow-band received process X. The characteristic function D(Eo)- J W, (E) dE
which we need now is for the random variables Xc, Xs, namely,
P1(E>EO)n W1(E)dE= 1-D(Eo) (2.9)
F, (it, i7)X C' XsX 0o

and are :found to be


exp p(X, e)[ei'Uc+i7us 11 dX de (2.4)
.~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
which is the two-dimensional generalization of (2.1) required D1(Eo) =Eo f J1(rEo)dr f F(ir, 0)dq5/27r,
here. Using a number of results from our earlier development
of the physical model (Section 2.2, [16] ) to write (cf. (2.10)- Eo>O (2.1Oa)
(2. 1Oa), [1])

U, = Bo cos [1' +dWOO(X + ') -


EdO t] P1(E > Eo) = 1-Eo JJ (rEo) dr
Us = Bo sin [dIs' + MdC.O(X) +) EdWOt] (2.5)
J F (ir, 0) do/27r. (2. I Ob)
we find the c.f. F1 to be explicitly

F,(rS=
IKf,pxe Our results (2.6)-(2.10b) are generalizations of earlier
results [7], [14], [8], where our basic assumptions, so far,
postulate only Poisson distributions of source location and
eirBo cos [<5s'+AldW0(X+ )-CdWOt
( ] -I dXde
emissions, e.g., essentially independent sources. No restrictions
on the specific character of the statistics of the source param-
eters are as yet introduced. It is for this reason that the
characteristic function F1 depends on ;b, as well as on r.
(2.6)
where Bo (>O) is an envelope (cf. Section 2.2 ff.); JD' = D+ 2.2 First Reduction of the c.f F1:
'kT + OR [respectively, a typical source phase and phase angles The Narrow-Band Receiver Condition
of source (T) + receiver (R)]; 1d = 1 + Ed, Ed a normalized At this point, we invoke certain properties of the basic
relative doppler between T and R. The first-order pd for the waveform, Ufb = U, cos wot + Us sin wot = Bo cos [4)' +
envelope and phase (E, 7), is then found to be PdWO(X +) - Oett -] , which appears in the exponent in
the integrand of (2.6). We use the facts that i) Bo, k5, are both
slowly varying functions of X; and ii) the process density
0
2 ird p(X, e) is likewise slowly varying, vis-a-vis cos COOdX.
W1(E, k)EJ r dr f; 2 sin COOPdX- Employing the familiar expansion in Bessel func-
tions, exp (ia cos 4') = loimemJm(a) cos mq5, in (2.6), we see
*F (r,q5)e-iEr cos ( -) that only for m = 0 does the integrand (containing the expo-
E>O, 0<0<2ir nent) contribute, as all the other terms are highly oscillatory in
(2.7) regions where Bo, (D, and p are slowly changing with X. The
result is
given by5
F1(ir, 4)

WI(E) -
I2rr
wl (E, ;) d; exp {J p(X, )[JO(rBo[t, I, 0] -1])
X dxdI
f f ~~~~~~~~~~~~~~~~~Al = F1 (ir) (2.11)
-- EJ rJo(rE) dr FJF(ir, 0) d0lla.
Io which is valid, provided that U, b is truly narrow band, e.g.,
AfA RI < fo. The important analytic feature of (2.11) is that
I
As usual, functions of different arguments are different functions, now, because of the narrow-band receiver condition AfARI <
e.g., W1(E) W100 * W1(XC, Xs), etc., unless it is otherwise stated. fo, the c.f F1 is independent of 4'. Accordingly, we see that
MIDDLETON: MODELS OF ELECTROMAGNETIC INTERFERENCE 115

(2.7), (2.8), (2.10) reduce to the simpler forms this becomes [uniform pd of e]:

E r r
Wj(E, 0) = rJo (rE)Fl(ir) dr

= WJE)Wj(;)
(2.12)
with specifically here with

AAe T/T T2T: average no. of emissions per second,


W1(E) = EJ rJo(rE)Fl(ir) dr in the observation period T,

TTs T: "density" of the process [cf. Section


W1(O) = 1/27r 0 < 0 < 27r 11.2;(11.74),Middleton [12]:
(2.13) (average no. of emissions per second)
X (mean duration of an emission).
and
(2.1.6b)
and
D(Eo) = Eo J1 (rEO)Fl (ir) dr Bo = AR(X,fO)AT(X, fo) IAOT(TSZ I 0)g(X) (2.17a)
0 (2.1 4a)
the received envelope Bo of a typical emission, where
P(E >Eo) = 1 -Eo JJ (rEO)Fl (ir) dr. AR, AT (complex) beam patterns of receiver and typical
(2.14b) interfering source;
AOT (real) envelope of the source emission
(=AOeOyuO(z));
The results (2.12)-(2.14b) are formally identical to those g(X) a geometric factor, which describes the propaga-
derived by Furutsu and Ishida ([7], (2.9)-(2.11)), Giordano tion law, from source to receiver (which are
[8], and Giordano and Haber [9], for example. Furthermore, assumed to be in each other's far field);
Fl(ir) is clearly a Hankel transform of W1(E), from (2.13) and uo(z) normalized envelope waveform at IF output:
the fact that the inverse of (2.13) is uo = 0, z > Tl/T8, z < 0;
eo,y a limiting voltage setting, at which receiver re-

00
sponds to test signal, above receiver noise.
Jo (rE) Wi(E) dE-(Jo (rE))E (2.1 7b)
F, (ir)=
(For this receiver, although the aperture may be comparatively
eirEcoskWi(4i)diz1) broad-band, as may be that of the source, it is the narrowest
filter of the combination (aperture X RF X IF) which is con-
trolling. By assumption, one or more of these filters is very
=(eirE cos 4)E,X. (2.15) narrow vis-a-vis fo, so that the effective aperture response here
is determined essentially by the response at (and about) fo,
Next, by analyzing the process density p(X, e) in terms of cf. (2.17).) This limiting form (2.16) is the expression which
the physical density of emissions in space and in time, accord- we shall exploit in the remainder of the study.
ing to the steps indicated in (2.25)-(2.35), [1] and restricting,
The quantity A appearing in (2.16) is
our attention to the most common situation of "local station-
arity," whereby it is assumed that there are no changes in A. (=,y.): impulsive index (of the present analysis).
average source numbers and emission properties during the
observation period T, and that the emission probability w1 (e) (2.18)
is uniform, we obtain finally the basic form for the integrand As we have already noted in our earlier studies [14] [16] the
- ,

of (2.1 1). For the idealized steady-state situation where T-* oo, impulsive index is a measure of the temporal "overlap" or
"density," at any instant, of the superposed interference wave-
T-i oA (A
VT V0 lim TIT)/ forms at the receiver's IF output. It is one of the key param-
lim
Tlioo
=
=AA T-T eters of the interference model, in that it critically influences
the character of the pd's and PD's of the interference, as
T-o
lim 'YT = ?Yoo; observed at the output of the initial (linear) stages of a typical
116 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. EMC-19, NO. 3, AUGUST 1977

narrow-band receiver. With small values of A, the statistics of - Class C -


Uoc(z) Tr
the resultant output waveform are dominated by the overlap-
ping of comparatively few, deterministic waveforms, of
different levels and shapes, so that the interference has an "im-
pulsive," somewhat structured appearance. For increasingly
large values of A. the resultant approaches a normal, or Gaus-
sian process, as one would expect from the Central Limit
Theorem ([12], Section 7.7), as we shall see in more detail
later (cf. Section 2.4). 0 (z t z' TSB
TSA 'SB
2.3 Interference Classes A, B, and C-
Fig. 2.1.(II) Typical output envelope of ARI filter output in (narrow-
the Role of Input and Receiver Bandwidths band) receiver, showing Class A and Class B envelope components
We are now ready to examine the basic form, (2.16), of (positive halves only); Class C = Class A + Class B.
IO(r) [=logF1(ir)]. The role of the duration T of a typical
emission (as perceived at the output of the ARI (=aperture X (Tin = TSA fixed for the moment), e.g.,
RF X IF) stages of the narrow-band receiver) is critical in A A _A
determining the form of I. (r). I (r) = I(r)A + I (B c
Let us consider first the important special case when the
emission duration Ts is fixed. From (2.63a, b), (2.70), (2.72a) In terms of the characteristic function (2.11), we see at once
of [1] we may write for the envelope Bo, cf. (2.17a), Bo = that
A OeO, AR A(, fo) g(X)uO(z). F1 (ir) = Fl (ir)A * F B(ir)s- F (ir)c
With TS fixed, we have in general the situation shown in
Fig. 2.1(11) for the envelope of the narrow-band output of the with the important result that Class C interference consists
ARI filter, produced by a typical interference emission of of the independent sum of Class A and Class B components,
finite duration, Tin. The output envelope (-uo(z)) produced as defined above. Specifically, when we can ignore the Class B
by a typical input interference envelope (shown as a rectangu- component [TSB(<) TS,A e.g., sufficiently narrow-band
lar pulse in Fig. 2.1(11), always consists of two parts: a part input vis-a-vis the receiver6] , we have
which we shall call Class A with normalized envelope UOA(z),
which is produced by the input emission (E0o-in), which is
"on" during the interval 0 < z < Tin (=TSA = TSA); and a F1(ir)c -* F1 (ir)A
part we shall term Class B, with normalized envelope uO(z -
1) (#UOA(Z)) which represents the transient decay of the = exp {A 00rA f lo)-
)', x dzj
output of the ARI filter, following the termination of the
input emission ['E0(z)in]. The sum of Class A and Class B =e -,A -
exp {AA(Jo(rBOA))XA}
envelopes is called Class C, e.g., uoc(z) = UOA(Z) + UOB(Z -1)
(cf. Fig. 2.1(11), where, of course, UOB = 0, Z < 1, UOA = 0; (2.1 9a)
z < 0, z > 1 in our definition. Thus all receiver outputs are
typically Class C, with variable amounts of Class A and Class B, (V-. T, =)A,A >A,B(=-VTSB) (2.19b)
depending on the duration of the typical input interference
waveform vis-a-vis the response time of the ARI filter at the where the averages ( z are explicitly
front end of our receiver. Equivalently, the relative extents of
the Class A and B components depend generally on the ratio ( )Z, x, =| dzj [ ] dXdf'. (2.19c)
of the bandwidth of the input (Afij) to the bandwidth AfARI o 14oAA
0
of the linear "front end" of the receiver. With Afin > 4fAARI,
for example, the usual case of very wide band interference Similarly, when the Class A component is ignorable [TSA =
(automobile ignition, flourescent lights, atmospheric noise), T.i. < TSB, e.g., very broad-band inputs vis-a-vis the receiver's
TSA is very small vis-a-vis TSB: the build-up time (=Tin) is very ARI stages7] we get
brief for Class A, while the duration of Class B depends on the
decay time ("AfARI)1 of the ARI stage, which is much A
Fi(ir)c F1 (ir)B
A

longer than Tin. On the other hand, with narrow-band inputs


of long duration [AfARI > Afinj, the transient at the ter-
mination of the typical input is of negligible effect vis-A-vis -=eXP fAOBJ (JO)rBOB) )X.odz}
the Class A component. For comparable bandwidths (AfARI
Afin) both Class A and Class B make comparable contribu- (2.20)
tions, e.g., neither can be ignored vis-'a-vis the other, so that we
have then generally the Class C waveform in the receiver. [In 6 The precise conditions for effectively Class A or Class B inter-

all cases,f'
lo u B(t)kdt, k> 0, are finite.]
-
ference alone are developed in Section 7, Part II, of [ 1 ] .
7Techiica1ly, Giordano [81 and Giordano and Haber [91 express
From (2.16), we see accordingly that I.(r) can now be their results in a Class A format whenever sample size T is finite (cf.
written as the sum of the Class A and Class B components, viz: remarks in Section 5.3, [1]).
MIDDLETON: MODELS OF ELECTROMAGNETIC INTERFERENCE 117

A- >> A A,,with2
A,B
the
wihouavhthe 2.4 Second Reduction of the c.f F1:
with the averages given by (2.19c), without
average over z. Note that, when r -+ oo, F1-A -* exp (-A.,A), The Role of Source Distribution and Propagation Law
while F1_B,C -+ 0. This fact means that for Class A interfer- Our major problem, as stated earlier in Part I, is to obtain
ence, there will be a nonzero probability of "gaps-in-time," analytically tractable results, as well as a pertinent physical
i.e., finite (nonzero) intervals in the receiver's output when foundation for our models of man-made (and natural) elec-
there is no waveform present, while for Class B and C inter- tromagnetic interference. Technically, our central problem
ference, there is always a nonvanishing waveform level and now is to evaluate the probability densities and cumulative
hence no "gaps-in-time". probabilities (2.13), (2.14), when the interference is Class A or
Of course, physically there is always some inherent system Class B, accompanied by Gaussian noise, with the respective
noise, which makes it strictly impossible to have a true "gaps- characteristic functions (2.22), (2.23). (The detailed analysis
in-time" situation. of Class C interference is reserved to a subsequent study.)
Because there is thus an accompanying Gaussian background The desired evaluation may now be achieved by recalling
noise, which may arise in a number of ways: (as in Section 3 of Part I, [16]) that the general character of
the pd (and hence of the PD) of a random variable at large
i) as system noise in the receiver; power level AR 2 values is controlled principally by the behavior of the associ-
ii) as e-xtemal interference, which is the resultant of many ated characteristic function at, and near, zero values of its
independent sources, none of which is exceptionally domi- argument. Thus the behavior of F1 (ir) at, and in the vicinity
nant with respect to the others (so that the Central Limit of, r = 0, is determined by the largest r-dependent contribu-
Theorem applies); power level GE 2; tion which establishes the large-amplitude structure of W1(E),
iii) as a mixture of receiver noise and (independent Gaussian P1(E), etc., i.e., as E -* oo. In fact, for these general classes of
interference: non-Gaussian noise, this corresponds to the expected slower
fall-off of W1(E), as E - 00, than the Rayleigh pd, for example,
U02 = GR2 E2 (2.21) here. (See also the discussion in Section 2.6.1 following.)
Our preliminary procedure for obtaining the required
we must extend the preceeding analysis to include it. This development of the c.f. F1 in the neighborhood of r = 0 con-
extension is readily done (cf. (2.46), (2.47), [11). where we sists of i) expressing JO - 1 as an integral; ii) using an explicit
also extend the treatment to include the more general situa- class of propagation law and source distribution; iii), reversing
tion of interference inputs of random duration, e.g., Tin = the order of integration in i), ii), and observing the bounds
TSA (#TsA) generally. Only the Class A portion of Ih(r) is imposed by the fact that UoA is of finite duration, while
modified. Letting zo -Ti/ TSA, we have directly, including the uOB(z) # 0, 0 < z < oo, cf. Fig. 2.1(II) above.8 As we shall see
(additive) Gaussian components, the desired characteristic below, it is this latter condition (on UoA VS. UOB) which
functions: critically affects the explicit form of the needed development
ofFl.

Class A Interference and Gauss Noise: 2.4.1 Propagation Law:


We next introduce the somewhat restrictive condition that
(ir)A +G =eUG2r2 /2-A oo,A the source distribution and propagation law are expressible in
the factored form: a(X)[b(o) or b(G, f)]. The beam patterns
exp iA-OOAKJ [JO(rBOA) dz
O zo,A. o
are always independent of distance (cX), cf. Fig. 2.2(II). Thus,
for the propagation law, g(X) in (2.1 7a), we write
(2.22) g(X) = [gs(0P), gv (0s )] /(4rrcX)y, (2.24)

Class B Interference and Gauss Noise: where gs, v are angular factors, usually taken to be unity in
Fl(ir)B+G = e-G2r2/2 the common propagation models. In general, y > 0, and, in
fact, y > 2: y = corresponds to the "wave guide" modes
A often encountered in long-distance propagation in the atmos-
exp AooB f ([JO(rBOB)X-
lx, e' dz phere, while -y = 1 applies for the usual spherical spreading of
less distant sources. For practical application, sources -and
(2.23) receiver in a common plane, Fig. 2.2(a)(II) is typical of most
mobile land-transport communication environments, while the
4'volume" situation of Fig. 2.2(b)(II) ground/air, or gound/
We remark, again, that Class A [and consequently, Class C] satellite, or air/air environments. Also, for practical purposes,
interference models are new. The earlier "classical" analyses atmospheric noise may often be regarded as essentially co-
by Rice [211, Middleton [11], Furutsu and Ishida [7],
Giordano [8], Giordano and Haber [9],7 for example, all 8Our procedure here is a generalization of that used by Giordano
dealt with Class B interference, and, for the most part, in much [81, who, however, considered what in the limit (T -
-) is ultimately
only Class B interference, and only special choices of source distribu-
less general terms and by different modes of approximation. tions and propagation law.
118 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. EMC-19, NO. 3, AUGUST 1977

Class B:
Class A:
x

TI
I4/

(a) (b)
max.
Fig. 2.2(11) Geometries of source and receiver beam patterns: (a) in
a plane; (b) in a volume. Fig. 2.3(11) Schema of (x, X)-domains of intergration for Io, (2.28),
for Class A,B interference.
planar with the surface (and '1y), unless the principal dis-
crete sources are comparatively near to the receiver, i.e., us to write
OR (=6T) is large, e.g., [> 0(5 - 100)1
BOA,B = GOA B(Z, A0, eoy, AR T; 0, or 0, )IIQ/
2.4.2 Source Distributions:
(2.27a)
For the moment, we continue to assume that the source
distributions are also factorable into the form a -a(X)b(O, with
q5). Then, the density wl(X) required in the averages in (2.22), GOA B -e 07( B)A OUOA,B ART I gS v(4irc)Y (2.27b)
(2.23) is
containing the (possibly) random parameters AO, eoy, ART.,
w1(X) = F U(X)c2XUs (Eb)S As 1 (2.25) for both surface and volume regimes and Class A and Class B
LUV(X)C~3X2 Sin OaV(O, q5) interference. We then use (2.25), (2.26) to write
for the surface and volume regimes, where the normalizing
factors As, v are given by A dzo,
IOO(r)A,B = -A- IA,B aJ d
As=| dO C2Xas(X) dX (2.25a)
[B] -eff ] -eff
| A S, V- 1as. vd(O, 0 )
Avf
J(O, q)-eff
cA
av(O, q) sinO dO do JE As][CI.r221 Ji(x) dx
'[XI dXc3X42/xIifxroK
c3X2aV(X) dX. (2.2 5b) (2.28)
XI -eff
The Aeff the effective domains of the possible interfering
are with 0 zo, Ao, etc., where the upper term in [ ] applies for
=

sources, namely those capable of registering at our receiver surface sources and the lower for those distributed in the
[OR, cf. Fig. 2.2(11)], i.e., observeable in the receiver back- volume.
ground noise. The receiver, accordingly, has a limiting range Next, we implement the key step iii), Section 2.4, in order
cxmax' which depends on eoy, cf. (2.17b), e.g., Xmax = to interchange the order of integrations over X and x in (2.28).
Xmax(eo,y). In our present applications, however, we shall This permits us to develop I.o explicitly as a function in r,
assume Case I, e.g., Xeff = Xmax < XA, which is by far the to which we can then apply the approach indicated at the
more prevalent situation in practice: the potential source beginning of Section 2.4, to obtain controlling term(s) at and
domain always exceeds that of the receiver's acceptance near r =0 for the characteristic function. Since
region. Finally, we shall, where necessary, postulate the
following range dependence of source density:
xrGanXd;
= .'. X = (rGO)l x-1 Ja
and
as, v (X) =
I/X", 0< (2.26)
xo = rGOINMax (2.29)
for both the volume and surface situations.
is the value of x corresponding to Xm which establishes the
ax,

2.4.3 Role of Input Signal Duration, Ts domain of sources perceivable by the receiver [in the present
We begin by developing in fuller detail the structure of the Case I, cf. Fig. 2.3(11)]. Now we use the fact that UoA is non-
basic received envelope BOA,B, cf. (2.17a). Using (2.24) allows zero only for (0 S z < zo), while UoB * 0, (0 < z < 0). Since
MIDDLETON: MODELS OF ELECTROMAGNETIC INTERFERENCE 119

GOA,B UOA,B, Cf. (2.27b), we see at once that for


'
values 0 (<100) of the impulsive index A,A, (2.18), which
are typical of this class of interference, is now the following
Class A: direct expansion of the terms containing A-,A in (2.34):
0 x .XO, sinceUOA = 0, z >Zo

Xmax > X>0 cf. Section 2.4.2 (2.30a) F1 (ir)A +G

Class B: -e l-OO,A
_0

2.
r mRe2l|,22v
AoA_ Lm OA-\OztuG-J r-/
m=o m!
0.x < 00, sinceUOB # 0, z< 00
(U0B -+
0, z -
oo) {1 + 0(r4)}. (2.35)
x max .,x>0, cf. Section 2.4.2. (2.30b)
Fig. 2.3(11) shows the allowed domains of x and X for these (Equation (2.35) is formally the same as our earlier result
two classes of interference. (3.16), [16], which is not surprising, since the c.f.'s have the
same form, cf. (2.33). The parameters differ somewhat in
2.5 The C.F. for Class A Interference detail, of course.)
With Case I conditions [cf. (2.26), (0 < X < Xmax)] on the However, in approximating the exact (cf. (2.31)) by the
source distribution vis-a-vis the receiver range (max) and approximate forms (2.33), we obtain ultimately (cf. Section 3
(2.30a) applicable here, we see that x -- xo for the upper to follow) approximate PD's (and pdf's Section 4 of [1])],
limit on the integrand (for x) in (2.28). Accordingly (2.28) which may not be properly normalized, in the sense of yield-
becomes ing a mean square value, (E2)A, of the envelope, different
from the exact relation (E2)A = 2Q2A(O + A). Therefore,
A 0 x
we choose the c.f., here and subsequently, the leading term of
L- (rA =-AcOA J dzj J1(x) dx (2.35), and from it determine the associated PD and pdf.
Then, from the pdf we determine (e2)A . If Wl ()A is properly
xo = rGo/Xmaxy = rBOA . (2.3 1) normalized, (C2)A should be unity, which explicit calculation
reveals to be the case (cf. p. 73, [1] ). Accordingly, the Class A
Next we reexpress (2.31) as cases discussed subsequently need no scale or other adjust-
ments (unlike the Class B model, where the approximating
A [1 / Izo\ forms do require scale and level adjustments.)
h,(rA=-AA [ \]Jo(xo) dzj (2.32)
2.6 The C.F. for Class B Interference
Here we use (2.30b) for the exponent (2.28). The result
with zo = TS/TS <0), and then employ the steepest descent is a term like (2.31), plus an additional term for xo <x < oo,
method of Report I [Middleton [16], Section 3, (3.10) with Xmax . X > 0. Thus using (2.29), reversing the order of
et seq.; as mentioned earlier (cf. Section 2.4), it is the largest integration [according to the regime of iii), Section 2.4, and
contribution to the exponent in the neighborhood of r = 0, Fig. 2.3(II) above], and noting, in particular the order of the
which controls the behavior forE> 0)], to give the desired limits on the X-integration, which correspond to the variation
development in r in the neighborhood of r = 0. (The details in x, from x = xo to x -oo, we finally get (cf. p. 75, []) the
are given p. 70, [1]). We obtain finally (exact) relation
h4r)A _-AOO A +A"
Aer2(BoA2)/4{I + 0(r4)} (2.33)

where the correction terms [O(r4 in I.)] are given by (2.73),


(2.74a), (2.75) et seq., [11.
Combining (2.33) with the practical situation involving an
additive background Gaussian component, we get the desired
approximating form of the c.f. (2.22) for Class A interference

F1 (ir)A + G with
-exp {-uG2r2/2 -A, A+A Aer2(Be A2/4 xo =
rGo/XmaxY rBO,B =

[
1 O(r4)] }. (2.34) as before, and the new parameter
J,
The final step in the reduction of these c.f.'s to the desired 3-,
or (2.37)
technically manageable form, particularly for the smaller 7Y surface 7Y vol
120 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. EMC-19, NO. 3, AUGUST 1977

This parameter, a, we call the spatial density-propagation 2.6.1 The Approximating CF. 's for (O < a < 2):
parameter, since it depends on the interacting spatial effects of Unlike Class A interference [Section 2.5] where the c.f. is
source density and source propagation law. The lower limit on solely a function of r (cf. (2.34), (2.35)), and where a single
a is established by the present condition,9 i.e., upon the upper "steepest-descent" approximation provides a good fit for both
limit on p (=2, or 3),vy > 0. large, small, and intermediate values of r2 (and hence for
The first integral in (2.36) is readily evaluated by expanding Eo, E), Class B noise requires a pair of approximating c.f.'s,
the Bessel function, followed by termwise integration. For the one of which will at least ensure suitably bounded behavior
second integration, over x in the second integral of (2.36), of the exceedance probability P1(E > EO) as Eo -* 00, in-
we use a Barnes integral representation of J1(x) (cf. [12], cluding the existence of all finite moments of the envelope
(13.10)]. Evaluating the integrands at the various (single) (Ef)(O < < o), and the other of which will provide a satis-
poles, and with the additive, accompanying Gaussian back- factory account of P1 for small and intermediate values of
ground we obtain E (>EO). It is the presence of the term O(r`) in the (exponent
of the) c.f. (2.38) for Class B interference, in addition to the
typical development in powers of r2 (analogous to that for the
Class A noise), which forces this double approximation for our
F1 (ir)B + G = exp {-b 1 aA -,Bra (u2 b2aA B)r2/2 canonical c.f.'s, and PD's, P1(E >EO), etc., here.
At this point we define the Gaussian variance

z (-1)lb(2l+2)0,A ,Br21+2 } AGG -aG +b2A,B


= b2fAoo,Bg(l + UG2/b2aAo,B)
0<ca<2 (2.38) Q2B(G)(I + rB(G)) (2.39a)

which, like (2.36), is also exact. Here we have explicitly with

F(1 a- /2) B _a,y Q2B(G)b2oz,B


2a-'r(l+a/2)
rB (G )-(
/ G > _independent Gaussian intensity
1(1 -a/2)
/ BoB V'\
a "impulsive" Gauss Intensity
(2.38a)
2c12-lr(i+a/2)\ T (>0 4-oa
I\2,2B (2.39b)
b2 (4 ) (G0, B2ma2 e )mBx
where Q22B (G) is the "impulsive" contribution to the Gaussian
(>0) (2 38b) component arising from the Class B noise alone, and where
UG2 (=UE2 CR2) are the (independent) inherently Gaussian
(41+ 4 -) (BoB2l+2) contributions from potential external (Gaussian) sources and
b (21+ 2)
l!(l 1)!(21 + 2- a)(21 + 2) 221+1

from the receiver noise (essentially all arising in the initial


(2i 38c)
linear input stages). (Note that Q22B(G), FB(G) are also func-
(>0); (BO,B = GO,B/Xm ax) tions of a here.)
For the c.f. which is appropriate to the intermediate range
since 0 < a < 2, and from (2.28) we write of envelope values, including the very small (E, Eo -+ 0), the
controlling term in the exponent of the (exact) c.f., (2.38) is
(GO,B ) = (eo(B)a) (AO')(I ART I A (gs, v) the smallest power of r with negative coefficient, e.g.,
00 -biaA-, Brc here, so that this approximate form remains a
*(47w)-cl I UO(Z)BOT dZ (>) (2.-38d) proper c.f., e.g.,

and formally (GOB 21+2) is given by (2.38d) on replacing by a lim F1 = 1, lim F1 -+ 0.


r- 0 r- oo
21 + 2, etc.
The form of the associated pdf and PD for small and moderate
'
Analytically, for the integral over x in (2.36) to be convergent, we
require that > -1/2. There is, however, no (finite) upper limit on a,
a
values of F, Eo is governed principally by the behavior of the
so that we can write -1/2 < < -. For the purposes of the present
a c.f. as r becomes large. Moreover, practically there is always at
study, we shall, however, further restrict to the range (O < oa < 2), least an observable Gaussian system noise component, and
which covers many of the practical cases encountered in applications, at
least down to quite small values of the exceedance probabilities P1 (c < here, as well, an additional Gaussian term (-b2aA-,B) con-
60) - tributed by the "impulsive" Class B noise, so that the realistic
MIDDLETON: MODELS OF ELECTROMAGNETIC INTERFERENCE 121

intermediate c.f. here is now Comparison with (2.34) shows at once that this approxi-
mate c.f. for Class B interference has the same (approximate)
form as that for Class A noise, and thus will yield the same
blUA-Bro'rttGr/
AUG2r2I2 type of pdf's and PD's, etc., cf. Section 3.1. This has the
l(r)(B+G)-l-~ba
FI)(+Ie
important consequence of ensuring that all (finite) moments
of the envelope, (ET), exist, as required by the physics of the
0<a<2 (2.40) situation in all cases (cf. Section 4, [1]).
We remark that both approximating c.f.'s (2.40), (2.42) for
where the subscript (-I) indicates the c.f. for the range (0 < E, the true Class B c.f. are such as to give pdf's which are not
EO S EB) of envelope values. (The precise definition of EB properly normalized; each pdf, W1(E)B-I, W(C)B-11 ((4.3),
will be given presently, cf. Section 3.2). (4.4), [1]) does not yield (E2)B = 1. The former gives an
For values of E, EO > EB we require a c.f. approximating infinite value, while the latter, although Class A type, cf.
the exact relation (2.38) where the largest (r-dependent) con- (2.35), yields (62)B_II = 4GB2(#l), where GB is given by
tribution to the exponent about r = 0 and in the finite (non- (3.6b). Thus, W1(E)B-II, = (4GB2Y)W1(e)B-II, while
norm

zero) neighborhood of r = 0 is the controlling term. For this, the normalization of W1()B-1 requires, instead, a change of
we seek again a "steepest descent" form for the exponent of scale for the argument e (and, therefore, e0 in the associated
(2.38) exclusive of the term in ra, which, as we shall see PD). How this is done is described in Section 3.2.1. Finally, it
below, is always here smaller than the former (for 0 < r < e) is important to observe that unlike the Class A interference
and thus does not control the character of the c.f. at small r discussed in Section 2.5 above, the (first-order) statistics of
(and hence for large E, EO). Accordingly, as in the Class A Class B noise are obviously sensitive to the combined effects
cases, (cf. Section 2.5 above), we wish to represent the Class B of source distribution (,) and propagation law (-y), through
terms (exclusive of ra) in (2.38) by a series of the form the density propagation parameter ag, cf. (2.37). (Physically,
this may generally be explained by the argument on p. 82,
2 0
[1] .)
2 1=1
r22
III. PROBABILITY DISTRIBUTIONS
P1(e > 60)A,B, 0< < 2. a

-Ae- ar2[ + Bkr2k] (2.41a)


k=1
We are now ready to obtain the first-order exceedance
probabilities P1(E > EO), cf. (2.14b), when an independent
where the "steepest-descent" nature of the approximation is Gaussian component is present, so that (2.35), and (2.40),
exhibited, not only by the exponential factor, but by requiring (2.42) apply, respectively, for the characteristic functions
the vanishing of the B1-term in the right-hand series, where for Class A and Class B interference, to be used in (2.14b).
the nearest "correction" term (k = 2) is 0(r4) and quite ignor- We begin with the normalizations (2.4)I, for both Class A
able vis-a-vis unity. This condition and a term-by-term com- and Class B noise, and writing
parison of (2.41a) determine all the parameters A, a, Bk
(k > 2), which are readily found to be aA orB -{2Q2(0 rl)1-1/2 A orB. +

A-A-,B; a=b2a/2; (B1 =


0) We see therefore that e = aE, eo = aEo in each case, and
therefore r = aX in (2.35), (2.40), and (2.42), so that the
B2 =b4 -b22/8 desired exceedance probabilities now have the generic form
(2.41b)
b4ab2 + b20,3
B3 = b60 +r 2 48
etc.
P1(6 > eO)A,B = 1 -e Jl(XCO)Fl(iaX)A,B dX (3.1)

Clearly, Ao-,Be-ar2 dominates -A-,Bbljer, A- ,Be-ar2,r4 cf. (2.14b), where, of course, the specific parameter values in
etc., at and in the neighborhood of r = 0, and this is the deter- the normalization factor a have different forms for the Class A
mining element for this approximation to the exact c.f. (2.38). and Class B interference.
Accordingly, we have finally for the c.f. appropriate at least to
the large values of E, Eo, i.e., for the "rare events," 3.1 Class A Interference
Applying the above normalization and changes of variable
to (2.35) allows us to write the Class A c.f. in the following
Fl(ir)(B+G )-II desired approximate form:

_ {e A,B exp [A ,Be b2er/2/2 2r/2] } Fl(iaX)A -~-e AA e-A


AAm mA 2a2X2/2
m=0 m!
* [l+O(rt,r4)], 0< a<2. (2.42)
2amA 2-(m/AA + rA )/(l + rA') (3.2)
122 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. EMC-19, NO. 3, AUGUST 1977
40 40

20 20

0
u)
E E
q4;.
Ato
A
OD
a:> CIO

o4
0

-20 -20
= 10-

-40 -40

F~=io-

-60 1 1 1 1 1 1
10 I110-30.01 0] 0.2 0.4 0.6 00 0.9 .95 98 .99 1061110-3.01 0.1 0.2 04 0.6 0.8 09 95 98 99
P1 (e > 6.) Pi(e> o)
Fig. 3.1 .(II) Envelope distribution [prob(E > co)I calculated for Class Fig. 3.2.(II) Envelope distribution [prob(e > e0)] calculated for Class
A interference for AA = 0.1 and various rA from (3.3). A interference for FA' = 10-4 and various AA from (3.3).

(cf. (2.17)10 [16] ,and we henceforth abbreviate A0,,A =AA, explicitly'3


etc. Applying (3.2) to (3.1) and integrating with the help of
the relation jF (1; 2; -x) = (1 -e-x)Ix ((A.1-19b), [12] ), F(ia)B II- e bloABaaXa-AUG2a2X2/2 (3.4a)
yields"I Fl(ia)B-II eAB exp [ABeb2ea2X2/2 0G 2a2X/2] .

i ~~~00 X m
(3.4b)
Pl(e>6o)A -eAA z e-eoAmAm
A
m=0 m
These, in turn are applied to (3.1) to give us P1I -,II, respec-
tively. We get
C0 >0
(3.3)
for the desired approximation of P1-A. We observe at once A1.(A > 40) _ 1-enA
n=O
,n
that the Class A exceedanee probability P1-A is (primarily)
a weighted sum of Rayleigh probability distributions (PD's),
each with a variance which increases with order (m). Figs. F I a-) IF,
(+ + 2a; 2; -eAO2) (3-5 a)
3.1(II) and 3.2(11) based on (3.3) show some typical distribu-
tions Pi -A versus threshold e0, with rA' and AA, respec- with
tively, as parameters. As expected, these PD's are highly o=-(eoN.)12GB; A. =Aa/2ceGB0 (3.5b)
non-Rayleigh'2 for the rarer "events," e.g., those which
exceed the larger thresholds e0, while the Rayleigh forms
appear for the less rare events (e0 small), also as expected, Au 2&bluaOIAB = 2GbloUABI[2&2B(l
=
+ IB')]G/2
cf. (3.9) in [1] (For further comments, see p. 86, [1].)
2l7X-a1-
2r'(1 -ot12) A{BO,B BOB
3.2 Class B Interference (O < a < 2) (3.6a)
r(1 + a/2) -2Q22B(2 B) /J /
In a similar way, from (2.40), (2.42), for the two c.f.'s
which approximate the Class B interference, we obtain and
Note that 2 mA2 here is equal to amA2 (cf. (5.7), of [16]).
10
This PD is properly normalized for ( e2 )A -1 (cf. remarks fol-
1 1 GB22 = -
1 (+r
(1 rB')1 ar
+FIB'J (4 (3.6b)
lowing (2.35)).
4 \2 -a ,
12 envelopes E here, e.g., equivalently nongaussian for the
corresponding instantaneous amplitudes [Middleton [1611. 1 3 For compactness, we set AOO,B -AB, henceforth, cf. (3.2) et seq.
MIDDLETON: MODELS OF ELECTROMAGNETIC INTERFERENCE 123

8o 80

60 60

40 40

a;

z
44J stAU0
20 20

20 - 1 0 Q 1 0 01 . o . . . -20
o'" 010IT .01 .05 01 0.2 0.4 0.60.0 0.8 85 0.9 .95 .98 .99 lo- 10i103.01 0.1 0.2 0.4 0.6 0.8 0.9 .95 .98 .99

>(& g0) p .
Fig. 3.3.(II) Envelope distribution [prob(t > e0)] calculated for Class Fig. 3.4.(II) Envelope distribution [prob(C > 6o0)] calculated for Class
B interference for a = 1.0 for various Aa from (3.5a), (3.7). B interference forlA.
= 1.0 for various a from (3.5a), (3.7).

cf. (2.39a, b), where NI is a scaling factor which scales as discussed earlier (cf. Section 2.6.1), to obtain14, finally,
P1-(B-I), to ensure that (e2)B = 1, with aB2AUG -2GB2.
The quantity A, is the effective Class B impulsive index,
which is proportional to the impulsive index AB, for this
Class B interference. In addition, it depends spatially on the
eA1(
OB114B2 AEm
2/2mm B'
spatially sensitive parameter, a, and on the relative Gaussian 4GB2 m=O
component rB'. For large e0 we obtain formally, the follow- 60 > FB -
ing expressing for P1:

P16 e o)B-I (Figs. 3.1(11), 3.2(11) for Class A interference illustrate the
character of (3.8), which, of course, is only applicable here,
Class B, for the larger values of eo (>eB)-)
3.2.1 The Composite Approximation: The problem with
Aa n(-l)n+l r 1+2) the approximating results for P1-B in the Case B model, cf.
n=1 n! a(lfn (3.5a) and (3.8), is that these forms, stemming as they do from
approximate c.f.'s [cf. 3.4a, b)], are not properly scaled, or
2
"normalized," in the sense that each approximating form,
P1-I, P1I11, does not yield the correct mean square value of
(1 n/2)(&n)
(62CB = 1, or (E2)B = 2E22B(O + rB'). The approximation
A -n z

+ 2^A + ],EO2
2 >

(3.7) P1_I, in fact does not possess a finite mean square on (0, oo),
. Ona[
while P1 -II, the "Type A" form, yields (O2)B-II 1. Accord-
ingly, since the precise mean square is finite and is known to
This shows that fOlim 00 P1I -> (o-0) -< 0. However, as
be (C2>B = 1, by calculation from the exact c.f. (cf. (5.10a),
explained in Section 2.6.1, for EO greater than some (large) and Section 5.2-B, [11) we must suitably scale (or "normal-
value EB, which is determined from vi), (3.10) below, we must ize") the (pdf)1,11 so that (C2)B exists and is equal to unity.
use the second form of c.f., (3.4b). Figs. 3.3(II), 3.4(11) here This is done as follows.
are based on (3.5a), (3.7), and are valid representations, pro- i) Let us consider first the pdf wl II, [1], and calculate
vided e0 is not too large, e.g., co < eB.
For the "rare events," or large EC, we apply (3.4b) to (3.1), 1
4This PD is now properly normalized.
124 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. EMC-19, NO. 3, AUGUST 1977

(e2)11 on (O < e < oo). The result is easily seen to be

(E )B-II A IB
m ( ABm

4-(1
-+ r 4
=
1
= 4GB2 (*I1) (3.9a)

where GB2 is given by (3.6b), so that here we require the


normalization factor N0I2 - (4GB2)- 1, e.g.,
1
W1(C)B-I1-norm - 4GB2 W1(E)B- II= NII 2W(c)B-IF
(3.9b)
ii) The case of w1 I ((4.3), [1]) requires a different ap-
proach, since (E2)B-I on (O < e < oc) becomes infinite (O <
a < 2), cf. Section 5.3, [1]. ((2%B_II on (O < e < EB < 0), iS -401 W.I
of course finite.) Here we need to scale c0 according to (3.5a) 10 10I0 001 II012 014 06 08 09 095 098 099
above: eo - eoNI (and therefore E6 = (eoNo)/2GB). The
rationale for this is the observation that P11I and wl I) must Fig. 3.5.(II). Schema of Pl-B, (3.12), obtained by joining two
have the same values in the Rayleigh region (co2 < 1), where approximating forms (3.5a), (3.8).
P1-I - 0.9, or 0.99, etc., as does the precise distribution,
P1-B, based on the (intractable but) exact c.f. (2.38), hence EB: this insures a common tangent, i.e., a smooth fit; more-
(3.5a). The scaling factor, NI, is to be determined by fitting over, we have the following.
the two approximate forms P1 I, P1-II together by the pro-
cedure outlined below, which is based on the canonical proper- / d2P1-I d2P1-II\
ties of the Class B model generally. Note, finally, that the
(#0)
v) K de2 de2 /
"Class A" form (II) is coupled to the Class B form (I) through
the Class B parameter a, and vice versa through the "Class A" this follows as a consequence of iv), and the continuity of the
parameter FB', appearing in GB, common to both approxi- derivative at EB, ensuring that the associated pdfs are continu-
mations I, II. ous at the joining point EB. However, note that EB is not
To combine the suitably scaled P1 I and normalized P1 II usually a point of inflexion of P1j_jI.
to form the composite approximation for Class B interference vi) EB: This is the point of inflexion (d2Pl/deB2 = 0) of
which is valid for all ec > 0, we now use the following desired the actual P1, and is determined as such (by inspection,
properties of Pi.composite, which is sketched in Fig. 3.5(11). usually), of the empirical exceedance probability P1_ exp,
cf. Fig. 3.5(11). (3.10)
i) P1-I P1-II Accordingly, we have explicitly for i)-v) above as follows.
in the Rayleigh region, e.g., 0 <. c0 small. Equality of the two
approximations in the Rayleigh region is required, since both _~~~~~~~A
(- l)n ( Aa)
n
an
must represent the same (small) amplitude behavior, character- 1-B2 On r I +-
istic of all these PD's. 4GB2 n= n! 2
dP1_ dP1 II
ii) - in the Rayleigh region. 1 2 02e AB ABm
A
de0 deo ~~(2UmB2)-l1 (3.1 Ia)
4GB2 4GB2 m MO

iii) Pi-, =PI-II ii) Same as (3.11 a), without the e02 factors common to
both members of the equation, and "1" in left member.
at the "bendover" or junction point of the two approxima- (3.1 lb)
tions, cf. Fig. 3.5(11). This point, EB, is empirically determined
from the data, e.g., from the experimental APD or exceedance AOr(1 + a/2) (EBNIA\1
probability curve P1 (e > CO)exp., as described below, cf. vi). 2oGBar(I - a/2) 2GB /
iv) dP1_I - dP1_II *[1 0(uO)([CBNI] CB2NI2)]

de de EBi AB
AB m-6B2/2a 2
the (finite) slopes of the approximating P1_111 are equal at 4G 2 m! EE22mB. (3.1 1c)
MIDDLETON: MODELS OF ELECTROMAGNETIC INTERFERENCE 125

2GB Ac,ar(l + a/2) (BANI A--1 GLOSSARY OF PRINCIPAL SYMBOLS


iv) NI 2AGBa(l -a/2) 2GB / Equation numbering in Part II does not contain a "II"
suffix; when equations from Part II are referred to in Part I,
*1 + 0(iv)([CBNI] -, [eBNI] ) they are so designated, e.g., (3.5), Part II, etc.
EBe AB Am e-EB2/2mB Figure numbering in Part II is similarly indicated by a "II',
following the number.
4GB2 M=O M! amB2 (3.1ld)
A
v), vi): EB cannot be determined analytically from either AO peak amplitude of typical input signal
approximating form P1 I II. It must be established as a point
of inflexion from the empirical PD, as noted above. AA, AB, AO,
impulsive indexes (Class A, B interference)
(In using iii), iv), we may need at least the next set of "cor- A,A, A-,B
rection" terms in the asymptotic developments of P1_j, Au,
effective impulsive index
aB, a normalizing factors
dPi-I/deBE) We note that (given EB) these three relations aA, APD a posteriori probability; here, Distribu-
[i), iii), iv)] are sufficient to determine in principle, any three tion = P1
of the six parameters NI, a, Ac,, AB, Q2B, rB', when the other ARI combined aperture-RF-IF receiver input
three are specified. [In Section 6, [1] , it is shown how (3.10), stages
(3.1 1) may be extended to permit us to obtain, from the experi- AT, AR source, receiver beam patterns
mental exceedance probability Pj(e > eo)xpt, the six param- 0e spatial density-propagation parameter
eters NI, a, A,, AB, Q2B, rBE (or, more fundamentally,
[a, AB, BO,Ba ,2BE, rB'], cf. (3.6).) Various illustrative B
calculations are provided in Figs. 3.6(II), 3.7(11), [1]. It is BO, BOA
convenient to preset CEB; NI, at, Ac,, and determine AB, rB', BOB generic or typical envelope of waveform
Q72B from i), iii), iv). Other possibilities are: fix (EB; NI, a, from ARI receiver stage
rB'), determine (AR,, Q2B, AB); fix (CEB; Ac,, a, FB'), deter- blot, b2ce,
mine (NI, AB, Q22B); fix (EB; AB, rB, Q2B), determine b21+2 1a weighted moments of the generic envelope
(NI, a, Ac,); fix (CEB; NI, AB, rB'), determine (A.,, ,2B), BOB
etc. In any case, we have now f3 exponent of moment
Pl-B = Pl-I, 0 < CO < B; = Pl-II, CO > (B C

(3 -12) c.f. characteristic function


D
with P11 I, P1-I, given, respectively, by (3.5a) and (3.8). [The
curves of Figs. 3.6(11), 3.7(11), [1] are equivalent to Figs. Db probability distribution
3.1.1, 3.1.2 of Furutsu and Ishida [7], with (V/O)F+I -* AB, delta (singular function)
(aOIu)F+I -+ (rB ')-1, and (R/u)F+I -o Co and exhibit the same E
kind of "elbow" in the transition region from the Rayleigh
behavior (for e02 < 1), with a bend over to a constant slope E, EO instantaneous envelope
eo,y limiting receiver voltage
(P1 -
e-a eO 7 17>0) C, C0, C0 normalized (instantaneous) envelopes
eo , C0 envelope threshold
as for Class A noise, when e0 -+ oo, cf. Figs. 3.1(II), 3.2(II), CB "bend-over" point (Class B), empiript. of
3.5(II), [1]. inflexion in P1-B
77 exponent
impulse epoch
3.3 Remarks on Omissions and Additional Results normalized doppler
C60, 'Ed
Considerable technical material has been omitted here,
which is available in [ 1 ] In particular, we mention
.
F
F1, F1 characteristic functions
i) Section 4, on the pdf s of Class A, B envelopes; lF1 confluent hypergeometric function
ii) Section 5, on the calculation of moments; AfN, AfARI noise, receiver bandwidths
iii) Section 6, on the experimental determination of the f, fo frequency
basic (3 and 6) first-order parameters for Class A and
B models; G
iv) Section 7, analytical conditions for Class A or Class B, GO basic waveform
versus general Class C models.
g(X) geometrical factor of received waveform
Comparison of the present text with [1] also clearly indicates 1A F1E ratio of (intensity of) Gaussian component
the very considerable condensation and omission of analytic to that of the "impulsive," or non-Gaussian
detail, which the present format has required. component
126 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. EMC-19, NO. 3, AUGUST 1977

F(x) gamma function U


-y exponent of propagation law, with range U, Un b basic waveforms out of ARI receiver stage
Uo, UOA,B normalized envelope waveform at output of
H ARI stages
I
A A
V
IT,I.o exponent of characteristic function
IC incomplete r-function w
iR unit vector WI, Wl probability density function
J x
Jo,J' Bessel function, 1st-kind (0, 1 order) x instantaneous amplitude
JA Jacobian xO c.f. variable
K
y
KA,B conversion factor, for arbitrary normaliza-
tion z
L
zo normalized time.
A domain of integration
X argument of the c.f. REFERENCES
X (X, 0, 4), coordinates
[1] D. Middleton, "Statistical-physical models of man-made and
natural radio noise, Part II: First-order probability models of the
M
envelope and phase," Office of Telecommunications, Technical
exponent of source density law with range Report OT-76-86, April, 1976 (U.S. Gov. Printing Office, Wash.
D.C. 20402).
1d normalized doppler [2] J. W. Adams, W. D. Bensema, and M. Kanda "Electromagnetic
noise, Grace mine," National Bureau of Standards, Report
N NBSIR 74-388, 1974.
[3] W. D. Bensema, M. Kanda, and J. W. Adams "Electromagnetic
n.b. narrow band noise in Robena no. 4 coal mine," National Bureau of Standards
Technical Note 654, 1974.
0 [41 E. C. Bolton, "Simulating LF atmospheric radio noise at 60,
76, 200 kHz," Office of Telecommunications, Technical Memo-
Q22A Q2B mean intensity of the non-Gaussian compo- randum OT-TM-97, 1972 (available only from author).
nent [5]* R. Courant, Differential and Integral Calculus. New York:
CO, c0 angular frequencies (coo = carrier angular fr.) Nordemann, vol. II, 1935.
[6] L. R. Espeland and A. D. Spaulding "Amplitude and time sta-
p tistics for atmospheric radio noise," ESSA Technical Memo-
randum ERLTM-ITS 253, 1970.
P1 APD orexceedance probability [7] K. Furutsu and T. Ishida, "On the theory of amplitude distri-
bution of impulsive random noise," J. Appl. Phys. (Japan),
pdf probability density function vol. 32, pp. 1206-1221, July 1960.
T, 0 phase of narrow-band wave [8] A. A. Giordano, "Modeling of atmospheric noise," Ph.D. disser-
OT, OR aperture phase tation, Dept. of Elect. Eng., Univ. of Pennsylvania, PA, 1970.
[9] A. A. Giordano and F. Haber, "Modeling of atmospheric noise",
Radio Science, vol. 7, no. 11, pp. 101 1-1023, 1972.
Q [10] H. M. Hall, "A new model for impulsive phenomena: application
to atmospheric-noise communication channels," Stanford Uni-
R versity, Electronics Laboratories Tech. Report Nos. 3412-8 and
7050-7, SU-SEL-66-052, 1966.
r c.f. variable [11] D. Middleton, "On the theory of random noise. Phenomenologi-
p poisson "density" cal models I, II,"J. App. Physics, vol. 22, pp. 1143-1152; 1153-
1163, Sept. 1953.
S
[12] -, An Introduction to Statistical Communication Theory.
New York: McGraw-Hill, 1960.
U, UG, U,
[13] -, "A statistical theory of reverberation and similar first-order
scattered fields-Part I: Waveforms and the general process,"
UmA,B, I.JGG2 IEEE Trans. Inform. Theory, vol. IT-13, pp. 372-392, July
UA, UR2, UE2 variances 1967.
Us, v source density [14] "Probability models of received, scattered, and ambient
,

fields," in Proc. IEEE 1972 Int'l Conf: Engineering in the


Ocean Environment, Newport, RI, Sept. 13-16, 1972.
T [15] -' "Man-made noise in urban environments and transportation
emission duration systems: Models and measurements," IEEE Trans. Communi-
Ts, Ts;A,B cations, vol. COM-21, no. 11, pp. 1232-1241, Nov. 1973.
t, tl, t2 times [16] -, "Statistical-physical models of man-made radio noise, Part
0,0' sets of waveform parameters I: First-order probability models of the instantaneous ampli-
IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. EMC-19, NO. 3, AUGUST 1977 127

tude," Office of Telecommunications, Technical Report OT- IEEE Trans. Vehicular Techn., vol. VT-23, no. 3, pp. 72-83,
74-36, April, 1974 (U.S. Gov't. Printing Office, Washington, Aug. 1974.
D.C. 20402). [23] A. D. Spaulding, "Amplitude and time statistics of urban man-
171 -, "Performance of telecommunication systems in the spec- made noise," in Proc. Int'l. Commun. Conf., Montreal, Canada,
tral-use environment: I., Objectives and approaches for the June 1971, paper No. 37, pp. (37-8)-(37-13); also, IEEE NTC
general data base," Executive Office of the President, Office of '72 Conf. Record, vol. 72 CHO-601-5-NTC, Dec. 4-6, 1972.
Telecommunications Policy, Jan. 31, 1975 (Research Paper). [241 A. D. Spaulding and L. R. Espeland, "Man-made noise charac-
[18] ', "A new approach to scattering problems in random media," terisitcs on and in the vicinity of military and other government
(Invited Paper) 4th Int. Symposium on Multivariate Analysis, installations," Office of Telecommunications, Technical Memo-
June, 1975, Wright State University, Dayton Ohio; also Report randum OT-TM-48 (available only from authors), 1971.
on Contract N00014-70-C-0198 with Office of Naval Research. [25] A. D. Spaulding and D. Middleton, "Optimum Reception in an
See DDC, Arlington, VA. impulsive interference environment," Office of Telecommunica-
191 * P. M. Morse and H. Feshbach, Methods of Theoretical Physics. tions, Technical Report OT-75-67, June, 1975 (U.S. Govt. Print-
New York: McGraw-Hill, 1953. ing Office, Washington, D.C. 20402.) To appear as two papers in
(20] * K. Pearson, Tables of the Incomplete Gamma Function. Cam- IEEE Trans. Commun., vol. COM-25, Sept. 1977.
bridge, MA: Cambridge University Press (Biometrika), 1951. [261 * G. N. Watson, Theory of Bessel Functions. New York: Mac-
(21] S. 0. Rice, "Mathematical analysis of random noise," Bell Sys- Millan, 1944.
tem Tech. J., vol. 23, pp. 282, 1944, vol. 24, pp. 46, 1945.
[22] R. A. Shepherd, "Measurements of amplitude probability dis-
tributions and power of automobile ignition noise at HF," *Indicates a reference used in [1], but not explicitly here.

Current and Charge Densities Induced on the Surface of a


Prolate Spheroid Illuminated by a Plane Wave
Electromagnetic Field
CLAYBORNE D. TAYLOR, MEMBER, IEEE,AND CHARLES W. HARRISON, JR., FELLOW, IEEE

Abstract-The current and charge densities induced on the surface [3]. In addition, the induced surface charge density is also
of a highly conducting prolate spheroid in the presence of a plane wave considered.
electromagnetic field are obtained. The derived formula for the charge
density is approximate except at the tips of the spheroid, where it Both analytical and numerical solution techniques are
yields exact results regardless of the eccentricity. One application of the presented for determining the charge density induced on the
theory is the determination of the total permissible current in a missile, surface of a prolate spheroid by an incident plane wave with
before dielectric breakdown of the ambient medium occurs at the tip the electric field directed parallel to the axis of the spheroid.
of the missile. Although the circumferential variation is ignored, the ana-
lytical formula becomes exact at the tips of the spheroid
INTRODUCTION where the charge density generally reaches a maximum. The
A FEW years ago an exact formula was obtained for the analytical formula is obtained via a boundary value problem
J total axial current in a prolate spheroidal receiving and solution and, in most cases, is less convenient to evaluate than
scattering antenna when the incident field is directed parallel the numerical solution. However, the analytical solution is the
to the major axis of the spheroid [1] . In the theoretical devel- more accurate. A comparison of the two results indicates that
opment, the circumferential variation in surface current the numerical solution may be -sufficiently accurate in many
density was not considered. However, this circumferential cases.
variation is significant, particularly at low frequency [2]. Penetration of electromagnetic energy through electrically
Therefore, an approximation is introduced in this paper to small openings in otherwise shielded enclosures has been
predict the circumferential variation of the current density studied recently [4] - [7] . In the sequence of steps to solve
the problem, it is necessary to determine the charge and cur-
Manuscript received July 22, 1975; revised September 24, 1976.
rent densities established on the structure by the total electric
This work was supported in part by the U.S. Air Force Office of Scien- field and magnetic field at the position of the opening when
tific Research under Grant AFOSR-76-296 1. metallically closed [4]. Since a prolate spheroid is geometri-
C. D. Taylor is with Mississippi State University, Mississippi State, cally similar to a missile, the charge density expression obtained
MS 39762. (601) 325-6064.
C. W. Harrison, Jr. is with General Electro-magnetics, Albuquerque, in this paper is directly applicable to the missile shielding
NM 87110. problem.

You might also like