Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
5 , MAY 1973
Abslruct-This paperis concerned with characterizing the link be- output basis with the aid of system functions as discussed in
tween an airplane and a satellite. Attention is focused on the effect of Section 11-A. Section 11-B introduces channel correlation
indirect paths scattered from the surface of the earth. Applicable
propagation-theoretic and system function-theoretic work is reviewed functions to provide the simplest second-order statistical de-
and integrated. Some new and someknown expressions for channel scription of these channels on a quasi-stationary basis. To ob-
correlation functions are presented for the “steepest descent” channel tain estimates of error rates in data transmission some assump-
model. tion must be made about probability distributions of channel
I. INTRODUCTION fluctuations. Section 11-C summarizes the various statistical
channel models that have been useful in this regard.
characteristics of the surface scatter channel. The material is It is readily seen that the time-varianttransfer function at
divided into two parts, one dealing with propagation physics the frequency f (actually f hertz away from carrier frequency
results (e.g., scattering cross sections)and the other dealing fo) is just equal to the complex modulation observed on a re-
with system function characteristics (e.g., Doppler power spec-
trum, delay power spectrum). ‘A third system function that the author has found useful is the delay
spread function U ( t , v ) [ 1 ] , which provides the input-output relation-
11. SYSTEMFUNCTION
CHARACTERIZATION
O F RANDOM ship
TIME-VARIANT CHANNELS
The propagation channels of interest here are linear and their
behavior may be described on an “instantaneous”input- This system function represents the time-variant channel as acon-
tinuum of delay and Doppler shifts with complex gain U ( < ,u ) d t du for
the delay r; and Doppler shift v . U ( t , v ) is the Fouriertransform of
Manuscript received October 26, 1972. g ( t , t ) over the t variable. It is also the double Fouriertransform of
The author is with CNR, Inc., Newton Upper Falls, Mass. 02164. the time-variant transfer function T ( f ,t ) .
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.
BELLO: CHANNEL CHARACTERIZATION 549
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.
550 IEEE TRANSACTIONS ON COMMUNICATIONS, MAY 1973
useful in predicting the onset of frequency and time-selective not besimultaneously active. Then P(v) and e([)
[or equiv-
distortionwhenthe WSSUS channel containsno discretealently p ( ~ ) q(!2)]
, aresufficient to describe the average mul-
paths.Thenboth q(S2) and p ( 7 ) drop to zero as S2, T-+ M, tipath andDopplerspreadingcharacteristics ofthechannel.
and pulses with bandwidths greater than Wcoh or time dura- In effect, for the class of waveforms considered, one mayre-
tions greater than Tcoh will suffer ever increasing amounts of place the actualscattering function s([,v) bythe simpler
distortion.Contrast this situationwiththe mixed (specular-scattering function
scatter) case (8); where q(S2) and p ( 7 ) approachnonzero
constants as S2, T + -. If the specular path is sufficiently so([, = P ( v ) Q ( t ) (16)
strong, increasing the pulsewidth beyond Tcoh or theband-
without altering the observed average time and frequency se-
width beyond Wcoh may produce only a small amount of addi-
lective distortion on the received signals.
tional distortion. Of course if the specular component is very
The reader should be reminded that to be practically mean-
strong Wcoh, Tcoh may equal 00 if they are defined as values of
S2 and T for which q(A-2) and p ( 7 ) drop to aspecific fraction of
ingful, the functions s([,v), P(v), e([)
must be regarded as
slowly varying with timeandgenerally dependent on carrier
q(0) and p(O), respectively.
frequency, in addition to being dependent upon the physical
The Fourier transform of the frequency correlation function
q(S2) is the delaypower spectrum e((),
whichcanbe ex-
location and motion of theterminals of the link.
pressed in the following forms, C. Statistics
In order to be able to evaluate analytically the performance
of various modulation techniques over a channel it is necessary
to have more statistical information than the channel correla-
tion functions defined in Section 11-B. Strictly speaking, for
This function is proportional to the intensity of the complex
an exact statistical characterizationone needs multidimen-
tap gain at delay [ in a differential tapped delay line model of sional probability distributions of the system functions. Un-
thechannel. Onemay defineamultipathor delayspread fortunately, these have not been measured and if they were,
parameter as the “width” of e([) where width is defined in they would be prohibitively complex to use. Fortunately,
some convenient fashion. however, there are useful statistical models that may be used
Two measures of width that occur frequentlyin applications in characterizingradiolinks. All these models use Gaussian
are !‘total” and “rms.” The total delay spread Ltot is meant to processes to model channel fluctuations either directlyor indi-
define the extent of e(() for values of ( where Q(g) issig- rectly. The utility of Gaussian characterization, either direct
nificantly differentfrom zero [e.g., 40 dBdownfromthe or indirect, is that the statistics can be completely specified
maximum value of e([)]. Theutility of Ltot is that it de- from the correlation functions of theGaussian processes.
fines the width of g ( t , [) versus [. Then, by Nyquist’s sam- Three very usefulmodels thatkeep recurring are 1) the
pling theorem, the transfer function T ( f , t versus
) f must be Gaussian WSSUS channel; 2 ) the Gaussian discrete WSSUS
sampled at least at a sampling “rate” of l/Ltot samples/Hz to channel; and 3) the Gaussian phase-modulationdiscrete sta-
allow reconstruction of T ( f ,t ) . tionary channel.
The rms delay spread L,,, is defined as twice the standard The Gaussian WSSUS channel is a WSSUS channel in which
deviation of e([) when it has beennormalized to unit area
the transmission of a carrier results in the reception of a nar-
and regarded as aprobabilitydistribution. This parameter
row-band process whose in-phase and quadrature components
may be shown to control thedegree of frequency selectivity in are stationary Gaussian processes. A special case of this chan-
a bandwidth that is of the orderof the coherence bandwidth[ 1 1 . nel, which is the one most oftenused, is the complex Gaussian
The Fourier transform of the time correlation function p ( 7 ) [2], [3] stationary scatter channel in which the in-phase and
is the Doppler power spectrum P(v), which is the power spec- quadraturecomponentsofa received carrier are ofequal
trum of a received carrier. This spectrum may be expressed as
strengthand satisfy certain symmetry conditions that force
an integration over S([, v):
the average
( fa, t + 7)= 0 .
T(f, t)T+
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.
CHANNEL
BELLO: AERONAUTICAL CHARACTERIZATION 55 1
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.
552 IEEE TRANSACTIONS ON COMMUNICATIONS, MAY 1973
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.
BELLO:AERONAUTICALCHANNELCHARACTERIZATION 553
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.
554 IEEE TRANSACTIONS ON COMMUNICATIONS, MAY 1973
values for the parameter a. Thus at a wind speed of 20 knots The values of y, 8 of interest are associated with those points
one may estimate a = 0.12 and for a wind speed of 10 knots on the surface within the glistening region. The point of mir-
a = 0.10. ror reflection between transmitter and receiver for the mean
The scatteringcrosssection(23) was calculated under the surface, called the specular point, corresponds to 0= 0 and
assumption of a perfectly conducting surface and included the y = 8 = 7 = ysp. As the scattering point departs from thespec-
combinedpowerin the horizontal and vertical polarizations, ular location y and 8 will generally differ. We have defined 7
assuming either horizontal or vertical polarization was trans- as a convenient single-angle approximation to the average value
mitted. It does not include certain effects that take place at of y, 8, which is exact for small y, 8. Columns 6 and 7in
small grazing angles, such as wave shadowing effectsand Table I present the values of 7 required to make the roughness
multiplescattering. Beckmann [6,p.981 suggests that u(r) parameter g = 1 for the various sea states. For g to any extent
can be corrected for a lossy dielectric surface through multi- greater than unity, say 3 or greater, one may assume the con-
plication by an averaged squared Fresnel reflection coefficient. dition g’ >> 1 satisfied. Also, if 7 does not differ too much
However Stogryn [ l o , eqs. (61a) and (61b)lhasmadethe from ysp,we see that g’ >> 1 implies negligible discrete specu-
more exact calculations (again for g >> 1) and they do not lar componentand g = 1 implies thatthediscrete specular
agree with Beckmann’s suggestion because of the existence of component has decreased to e-’ of its value for a calm sea. We
cross-polarized terms. For small cross-polarized power (which notefrom Table I thatexceptfor very small grazing angles, ,
appears to be valid for the sea scatter), Beckmann’s suggestion gz>> 1 is a reasonable assumption for 1.5 GHz. However, for
is reasonable. 150 MHz this assumption will beviolated for many grazing
Following this suggestion, one would use uf(r), instead of angles of interest. Thus care must be taken in applying (23) at
(r) 3 VHF frequencies.
From results given in [ 6 ,p. 79, eq. (31)J one may show that
of(r) = I roI ~ o ( ~ ) , (26)
the scattering cross section given by the following integral does
where r0 is the Fresnel reflection coefficient corresponding to not require g’ >> 1
reflection from aplane lossy dielectricsurface orientedfor
mirror reflection from transmitter to receiver at the surface
location r. [6, ch. 15] discusses the results of experimental in-
vestigations of diffuse scattering from the earth’s surface. The
work of Bullington [3 11, Beard e l al. [32] ,and MacGavin and
Maloney [33] is discussed in detail. The measured u(r) are al-
ways widely scattered “about a mean value which varies be-
- exp [-g’] dl (29)
C(-)
1
tween 0.30 and 0.35. The majority of the values are included where J o ( . ) is the Bessel function of zeroth order and
between 0.2 and 0.4.” It was also found that the results were the assumed isotropic spatial correlationfunctionof wave
approximately independent of grazing angle and wavelength as heights [e.g., see (24)]. Equation (29) is based uponthe
is predicted for (23). Kirchhoff approximation (surfaceradius ofcurvature large
Returning now to (23) we note that a(r) decreases rapidly compared to h/47r sin y), perfectly conducting surface, and no
for r locations requiring tan 0 to be any significant amount shadowing ormultiplescattering. Since the Kirchhoff ap-
bigger than a. (At tan 0 = 2a, u(r) has decreased to 13percent proximation itself must break down at low grazing angles and
of the value at 0 = 0.) The region of the surface for which a(r) shadowing is neglected, (29)mustbecome invalid at suffi-
is of significant value relative to its value at 0= 0 has been ciently lowgrazing angles.
called the “glistening” region because it defines the surface ele- The other technique of surface scatter modeling, frequently
ment locations actively participating in the diffuse scattering called the “small perturbation method,” accounts for scatter-
process. ing from irregularities small compared to a wavelength. This
To explain the possible restrictive nature of g’ >> 1, we method has been found particularly effective in explaining the
have used Kerr’s table [ 16, p. 4891 for, the first 3 columns of results of radar backscatter measurement at low grazing angles
Table I, which showthe relationship betweenthe seaman’s [17] , where the scatter cross section predicted by the Kirch-
“sea state” and approximate ranges of crest to trough wave hoff method drops to zero very rapidly. The earliest work on
heights Ah. The fourth and fifth columns present 47ro/X for the small perturbation method was carried out by Rice [18].
150 MHz and 1.5 GHz, where we have assumed that the peak- Some years later two schools of investigation developed, one
to-peak fluctuations of the sea (taken as A h ) are equal to 4 in the U.S. (e.g., [19], [20]) based upon the work of Rice,
times the standard deviation u (with this assumption the prob- and one in the SovietUnion (e.g., [21], [22]). Presentlya
abilityofthe sea-surfaceheight Gaussian processexceeding “two-scale” or“composite” model is used to characterize
Ah is approximately 4 percent). radar backscatter, combining the results of the Kirchhoff and
The roughness parameter g may be expressed in the form small perturbation methods. A detailed chronology of the de-
velopment of the large-scale Kirchhoff and composite models
g = (4nu/h) sin 7 (27) and a detailed mathematical development of the latter model
where the angle 7 is defined implicitly by is given by DeRosa [23].
Surface scatter channel modeling for the aeronautical chan-
. 3
sin 7 = (sin y + sin e). (28) nel involves a forwardscatter mechanism. Forthis case,in
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.
BELLO: AERONAUTICAL CHANNEL CHARACTERIZATION 555
TABLE I
RELATIONSBETWEENELECTRICALROUGHNESS A N D SEA ROUGHNESS
[
4nu = -
- i
A :=1
Calm . 0 0 0
Smooth 0 - l f t 0 - .47 0 - 4.7 - > 12.3
Slight 1 ft -3 ft .47 - 1.41 4.7 - 14.1 > 45 12.3 - 4.1
Moderate 3 ft -5 ft 1.41 - 2.35 14.1 - 23.5
45 - 25 4.1 - 2.4
Rough 5 ft -8 ft 2.35 - 3.77 23.5 - 37.7
25 - 15.4 2.4 - 1.5
Very Rough 8 ft -12 ft 3.77 - 5.64 37.7 - 5 6 .145 . 4 - 10.2 1.5 - 1.0
High 1 2 ft - 20 ft 5.64 - 9.40 56.4 - 1904. 2 - 6.1 1.0 - .6
7 Very High > 20 ft > 9.40 z 94 e 6.1’ -= .6
contrast to backscatter, the power does not drop off rapidly sections at the specular point (p = 0, y = e , $ = 0”) yields
at low grazing angles and, moreover, a strong specular compo-
nent may appear. A quantitative comparison of scatter cross
sections/unit area may be made for the Kirchhoff and small
perturbation methods in forward scatter, but the authoris not I point
aware of such a comparison. We present such a comparison for
(34)
one case, horizontal polarization and aperfectly conducting
surface. If one may assume the slopes of the small scale ripples mod-
The small-scale surface height fluctuations will be modeled eled by SPM are comparable tothoseofthe large scale
as Gaussian with (isotropic) spatial correlation function ‘‘swells” modeled by the Kirchhoff method (KM) it becomes
evident that in the glistening region of the large-scale swells, at
C(Z)= u: exp [-(12 / L : ) ] (30) least the scattered power of the small-scale ripples is likely to
where u1 is the rms value of the surface height fluctuation and be small compared to that due to the large-scale swells. This
L , the correlation distance. In order for the small perturba- indicates that the small-scale scatter may not be important for
tion method to be applied, certain inequalities must be valid forward scatter channels over the ocean.
[22, pt. 11, p. 5611. B. System Function Modeling
The previous section has summarized some important radio
propagation aspects of surface scatter channel modeling. Here
we wish to relatethese propagation considerations tothe
where [ is the surface height deviation from the mean surface “blackbox” or system function modeling of the surface scatter
and d [ / d s is the height spatial derivative. If we replace I [ I and channel, which is of direct importance in modem design.
Id[/ds 1’ in (31) by their averages, these inequalities become For the purposes of such modeling, the surface scatter chan-
nel may be thought of as the parallel combination of three
u 1 2 G / x << 1 2(0l/L1)* << 1. (32) subchannels, a direct path channel, a discrete specular reflec-
tion channel, and a diffuse scatter channel. Thus the complex
Using the surface model defined above, the small perturba- envelope of the received signal may be expressed as the sum
tion method (SPM) shows that the scattering cross section per
unit area for horizontal polarization and a perfectly conduct-
ing surface (see Swift [23, eq. (14)] with plus sign corrected
w ( t ) = Wdir(t) Wsp(t) Wdif(t) ’q(t>, (35)
to minus sign or DeRosa [5, p. 1 13, eq. ( 4 3 ) ] can be placed in where q(t) is an additive noise term and the subscripts identify
the form the other terms as contributions from the three subchannels
in Fig. 3 .
I ) Direct Path Channel: As discussed previously, in the case
of the surface scatter channel as defined in this paper, we do
not include tropospheric andionosphericchannel effects.
Thus propagation through the atmosphere between two points
is characterized forthe surface scatter channelsimply by
means of a (time varying) delay, complex gains associated with
A study of (33) with careful attention to the inequalities antenna patterns, and free-space losses. In particular the re-
( 3 2 ) does not indicatea well-defined localized glistening region ceived signal from the direct path channelis given by
for small values of 0 as for the Kirchhoff model’s scattering
cross section (24). Taking the ratio of the twoscattering cross
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.
556 IEEE TRANSACTIONS ON COMMUNICATIONS, MAY 1 9 7 3
Direct
Path
Channel
Additive Noise
Specular
z(t) r\ D Reflection w(t I
\/
Channel
t
Diffuse
Scatter
Channel
where G d i is a slowly varying complex gain due to transmitter ence between the specular reflection and direct path channels
and receiver antenna patterns and free-space losses,fo the car- for the case of an airplane to synchronous satellitelink(ac-
rier frequency, and T d i ( t ) the line-of-sight path delay. As far curate down to y = 15" according to [ 151) is given by
as the modulation z ( t - Tdk) is concerned, T d i ( t ) is a slowly
varying delay that must be adequately tracked for synchroni- T , ~- T d i = ( 2 H / c ) sin y (39)
zation in data reception or measuredin the case of ranging
systems. As far as the exponential term in (36) (due to the where H is the height of the aircraft. For an aircraft at 20 km
delayed carrier) is concerned, however, T d i ( t ) produces a rap- and 20" grazing angle this time-delay difference is 45 ps.
idly time variant phase shift 2?rfoTdi(t) and a slowly varying The Doppler shift difference
Doppler shift v d k ,
d
vsp - v d i = f0 (7sp - Tdir)
(37)
fa
= - (Vsp
C
- Vdu), (40)
where c is the velocity of light and V d i the rate of change of
the path length from transmitter to receiver. Expressions for
where Vsp is therate of change of thetransmitter-specular
G d i and Tdj, are readily obtained from the geometry, velocity
point-receiver path distance, and vsp the Doppler shift on the
vectors, and antenna characteristics of the transmitter and re-
specular reflection channel.TheDopplershiftdifference is
ceiver and will not be presented here.
usually quite small. We quote some results from [15] for the
2) Specular Reflection Channel: The specular reflection
airplane-satellite channel. For horizontal flight inthe great
channel is used to characterize wSp(t)the received signal due
circle plane containing specular point, airplane, and satellite,
tothe average (over the surfaceheights) field discussed in
the maximum Dopplershifthasa broad maximum between
Section 111-A. Assuming a very large scatteringsurface,this
10" and 30" for airplaneheights of 10-20 km andairplane
field is focused entirely in the specular direction and behaves,
velocities of 600-1600 knots. For 1600 knots and 20 km the
apart from a complex gain, similar to a mirror reflection from
Doppler shift difference is 5 Hz. Motion at right angles to the
the surface. Thus
great circle plane produces no Doppler shift. Vertical motion
w,p(t) = r Gsp z ( t - Tsp) exp [ - i 2 ~ f o T s p ( t ) l (38)
oftheaircraft will produce aDoppler shiftdifference. At
L band (1.6 GHz) and 20-degree grazing angle this Doppler
where I' is a complex gain (given by (21) by Kirchhoff model difference is around 3.5 V Hz where V is the vertical velocity
surface) that accounts for the electrical and roughness proper- in meters/second.
ties of the surface, G,, a slowly varying complex gain due to The importance of the specular component depends on the
complex antenna gains of the transmitter and receiver in the size of the reflection coefficient r and antenna discrimination
specular directionand free-space path losses, and ~,,(t)the against energy fromthe specular direction. Over theNorth
total path delay from transmitter to specular reflection point Atlantic it appears that at L band the surface is rough enough
to receiver. From a mathematical point of view Wsp(t) has the to make the specular reflection channel negligible most of the
same structure as Wd&(t), (36). The difference lies in the rela- time. However, atVHFthe specularreflection channel can
tive complex gains, delays, and Doppler shifts. A detailed set produce asignificant output relative tothedirectpath. As
of delays and Doppler shifts for the direct path and specular discussed in Section 111-A, the simple surface scattering model
channels is given in [15] for an airplane to synchronous satel- provided by the Kirchhoff approximation will be valid much
lite link. less in-land than over the sea. At any rate, as the terrain
Auseful approximate expression for the timedelay differ- changes in the vicinity of the Fresnel zone, one may expect
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.
BELLO: CHARACTERIZATION 551
that I r 1 willvary from negligible values to values close to by Mallinckrodt [24], [28], Staras [25], Durrani and Staras
unity (over a lake perhaps). [26], and DeRosa [27]. It is implicit in such calculations that
3 ) Diffitse Scatter Channel: The diffuse scatter channel ac- the channel may be characterized as a WSSUS channel. The
counts for all received signals other than those due to the di- other approach is more basic, starting with the formulation of
rect path and the discretespecular path. Over the ocean the expressions (usually integrals) for system functions (e.g., time-
diffuse scatterpathsproduceacontinuum of infinitesimal variant transfer functions) and then proceeding to the averag-
contributions to the received signal, and the diffuse channel is ing required to determine channel correlation functions. The
generally representable as arandom time-varying dispersive latter approach provides the means for answering basic ques-
filter with a time variant impulse response containing no dis- tionsnotobtainable by the scatteringcross-section formula-
crete components. (The only exception is in the limiting case tion, such as the degree of statistical dependence between sig-
of very small rms slope a,when the diffuse paths concentrate nals received on different polarizations and the validity of the
so closely in the specular direction that the entire diffuse scat- quasi-WSSUS channel model. The more basic formulation has
ter channel may be approximated by a single fluctuating dis- been carried out by DeRosa [ S I for a composite surface con-
crete path having the same path delay as the discrete specular sisting of the combination of large-scale fluctuations (satisfy-
component.)’ In addition, en’ough statisticalregularityexists ing the requirements of the Kirchhoff method) and small-scale
in the surface fluctuations so thatquasi-stationary WSSUS fluctuations (satisfying the requirements of the small perturba-
models with associated channel correlation functions (e.g., fre- tion method).
quencycorrelationfunction, Doppler power spectra,delay- We first present the simplest nontrivial results derived from
Doppler scatter functions) represent useful means of character- the use of scattering cross sections. These involve the compu-
ization. Moreover, withthe aid ofthe Kirchhoffand small tationofthefrequencycorrelationfunction q(Q), delay
perturbation methods, one may obtain meaningfulestimates power spectra e({),Doppler power spectrum P(u), time cor-
of channel correlation functions forsome cases of interest. relation function p ( ~ ) time-frequency
, correlationfunction
Over land, the characterization problemis considerably more R(Q, T ) , scatter functionS ( { , u), and cross-power spectral den-
difficult. The scattering surface exhibits far greater statistical sity functionP(L-2, u). Approximate expressions for these func-
irregularity (desert, foliage, mountains, cities, etc.) and the tions will be derived for the channel between an aircraft and a
Kirchhoff and small perturbation methods will only be useful synchronous satellite. The resultspresented here are derived
for a fraction of the scattering surfaces. While’a continuum of underthe same conditions and withthe same “steepest de-
return paths is likely, there will be situations where strong lo- scent”approximations as those of Mallinckrodt [24], [28]
calized reflectionsmay occur, such as reflections from man- who derives expressions for Q ( t ) , P ( u ) , S ( { ,v) and Staras [25]
mademetallic structures (e.g., buildings) and natural objects who derives expressions for q(Q) and ~ ( 7 ) .The results of these
(e.g., mountains). When such localized reflections occur, one authors are compared and generalized to include an arbitrary
must add discrete multipath components to the diffuse channel direction of the airplane velocity vector and analytic expres-
as defined here. It is still likely that a quasi-stationary WSSUS sions for R(L-~,T) and P(Q, v).
channel is useful. However, much more rapid changes in the Theearth is assumed to have a flat mean surface with
channel correlation functions are to be expected. In any case Gaussian-distributed surface heights relative to the mean and
the received signal due to the diffuse scatter channelcan be sufficiently smooth fluctuations to satisfy the requirements of
represented by the Kirchhoff method. In addition the roughness criterion
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.
558 IEEE TRANSACTIONS ON COMMUNICATIONS. MAY 1 9 7 3
To Sotellile
Fig. 4. Geometrical relationships for scattering in airplane-synchronous satellite channel assuming flat mean earth surface.
1
W ( X , Y ) = W ( X )W ( Y )=
2 a H csc2 8 6
(44)
10
It is readily shown that in the plane Y = 0, p = (e - $/2. Thus when
where R ~R ,, R~ are the lengths of the direct path, the pathp = e/2, there can be energy scattered to the receiver by areas at in-
finity on aplanesurface. Of course on a spherical earth y = 0 cor-
from airplane t o scattering area element, and the path from responds to the points at which planesthrough the airplaneare tangent
scatterer to satellite, respectively; A (X, Y ) is the surface il-
to the earth. In the former case the glistening zone extends to infinity
and in the latter case it is still so large as to violate the conditions re-
lumination pattern of the combined transmitter and receiver quired for the use of (44). If 01 << e-one may argue that the power re-
antenna gains normalized to thedirectpath gains;and ps, turnedat values of p that violate (44) will be negligible. However
the author is not aware of any criticalanalysis to determine simple
PD are the total scattered and direct path powers. The total criteria for use of (44).
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.
BELLO:AERONAUTICALCHANNELCHARACTERIZATION 559
x c
R H H S
We considernow the case wherein (48) appliesand deter-
mineexpressions forchannelcorrelationfunctions. To de-
termine the Doppler power spectrum P(v), delay power spectra
e@), or delay-Doppler scatter function S(t;, v), one may asso-
where we have used the notation *
ciate a single delay and Doppler shift and a particular power
density with each differential areaof the surface." By express-
ing the Doppler shift v as a function of X and Y , and regard-
ing X , Y as random variables, the probability density function
of v will beidentical tothe desired Dopplerpowerspec- Use of ( 5 1 ) in (50) then shows that
trum, normalized to unit area. The factor I roI2 A (H/tan 0 , O )
may then be applied to correct the height of P(v). An exactly v = -f[oV x . ; S 3 t V z C S ~ ]
analogous procedure will be used to derive Q(t;)and S(t;, v) as
probability density functions of the random variable t; and the
C (G z)
t-VS-
H
Y
where R isvector
a
ing element.
The
extending
Doppler
shift
from the
relative
airplane to
that to
the
scatter-
produced
an by
P(v) =
Brms
fi
6
exp [-$1,
Bms
(56)
v = -fo
Vx
C
(," )
--cos
c
V y - +y- V zfo
t - fo
R c
(sin--
. ); .
standard deviation,
Following the procedure described above, v is to be regarded The time correlation function p ( ~ is) the Fourier transform
as a random variable related to the random variables X and Y of (561,
by (50). The probability density function of v is the normal-
ized Doppler power spectrum. Before determining the density
~ ( 7=) ~ X P[- 3 (TTBrrns)' I * (58)
function of v we approximate v by leading terms in a Taylor Mallinckrodt computed P(v) for airplane motion in the X di-
series expansion of the right-hand side of (50). The approxi- rection (i.e., V = ( V x , 0, 0) and inthe Y direction ( V =
mations used areidentical withthose used in deriving the (0, V y , 0)). For these special cases P(v) in (56) agrees with
Gaussian approximation to W ( X , Y ) , (47). There is no point Mallinckrodt's results (note that his s = a/L = a f i and fix
in seeking any moIeaccuracy in representing v. All the ex- typos which left out fi on [28, p. 271). Staras [25] com-
pansions that are used depend upon the expansion of either R puted p ( 7 ) for the cases of motion in the direction of each co-
ordinate axis, i.e., (V,, 0 , o), ( 0 , Vy,o), ( 0 , 0, V,). Our result
I ' Since the sea surface is moving this is clearly an approximation (to (58) when specialized to these cases agrees with his calcula-
be included among all the others). However one may show that little tions for the two cases V = ( V x ,0,O)and V = (O,O,V,). For
errorcan be introduced in the estimate of P ( u ) , Q ( t ) ,and S([, u ) by
this approximation in airplane communication channels. airplane motion in the Y direction, however, his results differ
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.
5 60 IEEE TRANSACTIONS ON COMMUNICATIONS, MAY 1973
r ; =H
-c- (sin
s2m0 2 ~.( X - c t n ~ ) 2 t $ ) 0.3
-
+ . . . higher order terms. (60)
0.2 -
Neglecting the higher order terms we see that the path delay
E , regarded as a random variable, may be expressed as the sum
of squares oftwo
independent zero-mean Gaussian random 0.1-
8'30"
variables. The density function and characteristic function of
such a random variable are readily computed in closed form.
0.0
The results are 2.0 4.0 6.0 8.0 10.0 12.0
NORMALIZEDDELAY+ 8 50
x = sin' e (E-
X ctn e)
1 Y
which provide the normalized delay power spectrum and fre-
Y = %'E.
quency correlation function of the diffuse channel (referenced
to the specular path delay). Aside from a factor of lln in The variables x, y arezero-mean Gaussian with unit variances.
front and a scale factor on E, the Q(E) computed by Mal- In terms of normalized variables (55) and ( 6 0 ) show that
linckrodt [28, eq. ( 2 6 ) ] is the same as (61). Also Staras
[25] has obtained the result (62), (complete the integration in
-(L
6 = sin 0 ty.)
(68)
2 sin2 e
his eq. (23a)). Plots of Q ( t ) and 9(a)as a function of the
normalized variable l / 4 a 2(H/c) and Q4a2(H/c) presented
are w = vextV,sinOy
(69)
in Figs. 5 and 6 for several elevation angles.
The time-frequency correlationfunction of thechannel is where
just the joint characteristic function of the random variables,
E, v. Thus v p = (fo/c) vp (70)
R(C2,T ) = exp (-i27~nE)exp (i27r~v),(63) is the Dopplershift as measured along an axis in thedirec-
tion ( V g is the direct path Doppler shift v, sin 6 t v, cos e).
where E, v are given in terms of the X , Y variables. It is con-Consistentwiththenormalizations (64), (65) we used the
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.
BELLO: AERONAUTICAL CHANNEL CHARACTERIZATION 561
F=(FREOUENCY DIFFERENCE) X (H/c)(Za)' The Fourier transform of R (a,r ) along the r axis P(R,v) is
H = AIRCRAFT HEIGHT the cross power spectral density between two received carriers
c = VELOCITY OF LIGHT
= RMS SEA SLOPE separated byhertz in frequency. This function is of direct
0 = ELEVATIONANGLE interest in the evaluation of tone ranging systems where range
estimates depend upon the phase difference between filtered
sidebandsspacedin frequency.TheFouriertransform is
readily carried out to yield (in normalized variables)
[ (
. exp - 2 0 2
[ I + i2nF/sin e] [ I + i2nF sin e]
vi + v; sin' 0 + i 2 n F sin 8 (vi + v;)
I--
To obtain the scattering function one may Fourier transform
0.5 1.0
Norrnolized Frequency F +
Po(F,o)along the F axis or, more simply, regard 6 , w as new
random variables related to the variables x , y , by (68) and
(a) (69). Note that for each pair 6 , w there can be as many as two
Normalized Frequency F - r pairs ofx,yvalues(x1(6,w),y1(6,w)),(x2(6,o),y2(6,w)),
which satisfy these equations. Since the Jacobian of this trans-
formation is given by
F=(FREPUENCY DIFFERENCE) x(H/c)(2af
H = AIRCRAFT HEIGHT
c = VELOCITY OF LIGHT
a = RMS SEA SLOPE (77)
36 0 = ELEVATIONANGLE
F = R4a2Hlc
T = r2a.
Then (63) becomes
Ro(F, T ) = exp(-i2nF6)exp(i2nTw). (73) For the velocity vector along the y axis So(&,w ) has the same
form as (79) with Vg replaced by v,,. For the general velocity
If (68) is used in (73) it is trivial to show that
vector it is rather more involved, as shown in the Appendix.
Note that S o @ , w ) in (79) vanishes for w' > 2 6 4 sin 8 and
Ro(F, T ) =qo(F) exp becomes infinite at 0' = 2 6 4 sin 8 . More generally, the Ap--
pendix shows that S o @ , w ) vanishes for w2 > 26(vf + v;)
sin 0 and becomes infinite for w' = 26(vf + v;) sin 8. Expres-
sions differing from (70) only by multiplicative constants and
scale factors have been derived by Mallinckrodt [24], [28] for
where the velocity vector in the X or Y direction.
qo(F) = 1/d(1+ i2nF/sin e) (1 + i 2 n F sin e) (75) Channel correlationfunctions are particularlyuseful when
they lead to a morecomplete statistical description of the
is thefrequencycorrelation'functionforthe normalized channel as the case of the complex Gaussian WSSUS channel.
variable. To explain simply the conditions leading to Gaussian statistics
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.
5 62 IEEE TRANSACTIONS ON COMMUNICATIONS, MAY 1973
(the discussion may be made rigorous) we note heuristically control systems. Thusthe need forchannelcharacterization
that a transmitted signal of bandwidth W cannot resolve path experiments is clearly indicated in order to gather necessary
delays less than l/W apart.Thelocus of surface scatterers data for systemdesign.
causing channel path delays lying between E and E + l/W is an
elliptical ring. If the surface area of the ring extends over a ACKNOWLEDGMENT
large enough surfacearea to encompassmanycorrelation The author would like to acknowledge the benefit of helpful
lengths of the surface fluctuations,then, via central-limit- discussions with his colleague Dr. J. K. DeRosa andthank
theorem-type arguments (and representation of the channel as Dr. Karp and L. Frasco of the Department of Transportation,
a tapped delay line for a bandlimited signal [ l ] , [34]), one Transportation Systems Center for reading and commenting
may argue that the channel will act upon the input signal as if on the manuscript.
it were a Gaussian WSSUS channel. Moreover, as DeRosa has
shown [5] the roughness conditiong' >> 1 leads to the satis- APPENDIX
faction of the symmetry conditions required for the complex SCATTERING FUNCTION FOR GENERAL VELOCITY VECTOR
Gaussian property.Thebandwidthatwhichthe Gaussian The scattering function for a general velocity vector is most
WSSUS property breaks down appears to be much larger than easily derived by defining new variables
neededin aeronauticalcommunications,at least whenthe
roughness condition is valid.
There exist only a few measurements for airplane-satellite
channels, and essentially only at UHF-VHF [35] -[41] . Gen- X
r=--sin@++cos@, (A.2)
erally speaking the measurements are not sufficiently detailed S
to provide the basis for checking the multipath and Doppler where S = sin 6 and
spread theories described above. Moreover, very little sea state
measurements were taken in conjunction with the tests. How- vy/ve = tan @. (A.3)
ever the measurements seem to check the theory in a gross
In terms of these new variables the normalizeddelay and
way with regard to strength of the ground return and Doppler Doppler shift variables are
spread. Recent measurements at L band are reported in this
[43].issue [42], 6 = ( s / 2 ) [ t 2+ r2 1 (A .4)
0= s d - t . (A.5)
IV. CONCLUSION
conditions:
1 1
1) oneterminal so farfromtheearththat rays totheother w l ( t , r, = 2 7 1 0 , 0 , 4 ~ - 2(1- p 2 )
terminal and glistening zone appear parallel;
2) elevation angle to far-terminal B >> rms surface slope a ;
3) roughness criterion (4710 sin >> 1 ;
. (-4 + -)}
t2 r2 2prt
7-
or 0 r 0 t
(A.6)
But
complexity
theto
due involved a computer is needed to -COS' e V y V e
obtain channel correlation
functions.
The simplest
general P = d(Uf + s' v;)(u; -.+ s' uf (A.9)
results may be obtained for the scattering function because it
involves only the computation of the Jacobian of the trans- The Jacobian of the transformation between 6 , w and t , r is
formationfrom surface coordinates to delay-Doppler coordi- quickly found to be
Such
nates. general results have been programmed by 1DeRosa
[5] for a composite surface
model for general terminal posi- J(6, w)= . (A.lO)
d spherical
tions a earth.
Restrictions 4) and 5) cannot be readily relaxed by existing
s4- w2 jy-
s'(u; + v : )
approaches. at low enough grazing and Note that foreach a,6 therearetwo pairs of values t, r:
vation angles 4) and 5) will become violated. Unfortunately
low grazing angles are of considerable interest because of the
reduced multipath
discrimination providedtheby nominally
hemispherical coverage antennas proposed for use on air-traffic
t=
S"-
w
r=rt d y .
s 2 @ ; + u;)
(A.ll)
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.
BELLO: AERONAUTICAL CHANNEL CHARACTERIZATION 563
Authorized licensed use limited to: CINVESTAV. Downloaded on June 17,2020 at 16:08:28 UTC from IEEE Xplore. Restrictions apply.