BSTJ 60: 2. February 1981: Rain Margin Improvement Using Resource Sharing in 12-GHz Satellite Downlinks. (Acampora, A.S.)

13.3 MB·English

Raln Margin Improvement Using Resource
Sharing in 12-GHz Satellite Downlinks

By A.S. ACAMPORA

‘oManuecrint received August 6, 1980)

In thin paper we consider the effectiveness of sharing a small pool
of reseroed time alot of @ Time Division-Multipte-Access (r0Ms)
Frame among a large number of ground stations fa overcome rain
feuding, With thie aaroath, a satem domamicallyaraigna tine slots
ram the reserced pool ta grow stations experiencing fade depths
‘above the buill-in margin. Powerfid error correcting cades can be
tneroduced to occuny the extra time alos, providing 10 dB or more of
txtra fade margin, Heeause Targe number of ground stations are
competing for the Donited reserved pool, blockage ean occur ifthe
fuer of simultaneous fates exceeds the maximum number that ea,
bbe accommodated. Some factors that influnce the effectiveness of
ewnuree sharing arr the mutual fade statistics the various sites,
the traffic distribution within the nencork, the number of earth
lations the size ofthe reserved pool, and the rain outage objective
Sine the mutual fade and majfe statistics are unavailable, 0
develop models that can be used ts find a conservative bound on the
required size of the resersed poot. The rain model accounts for
‘lnc, seasonal, and peagraphtcal correlation wnong attenuation
tents, Results for a. masimum resouree sharing gain of 10 dB show
that resering six percent of the ime slots ensures a realized fade
(gain in cacess nf 9 di for @ downlink outage objective of 0.005
percent if there are more than i ground stations in the netunrk, cack
Toh tun percent ar few of the trate

|. wTRODUCTION

Iman earlier parr? shared-resource concept was deseribal for
incrasing the rin fade margin of» digital etelice system by us auch
fr 10 dit whove the design fade margin. With chis approach, unused
time slots of she Time-Division-MultipleAccess (roa) frame ste
fmade avaisble io ground stations experienelng rai fading above the

RAM MARGIN: MPROVEMENT 167
bullein fade margin, and are relinguisbed when the fade event has
suded. Error-correcting ending is introduced to oosupy th extza time
loca, thereby reducing the curie-to-noee ratio (CNR) required to
‘maincain the threshold bit-rrorrate (sit). Low rain outago is there
fore achioved without radiating exceative dovertink power: Not only
‘does this conserve antllite power, but, alo, interference into the
systems of other users of the gcosynchronous erbit is minimized. In
such an application, the operating speed ofthe decoder is much lower
‘than the transponder data rave by virtue ofthe Tow TDMA duty ele
‘rsocated with each ground station

‘Because of the infrequency of simultaneous deep fading we mulkiple
siteg, a small pool of reserved time slots can often protect large
‘numberof ground stations. The degree of protection so provided ie the
subject of tha current work; we shall evaluate the reduction in rain
‘margin required to achieve a given outage objective whon all ground
tations inthe network are competing fora Tilted amber of shared.
Fewurces, We restrict our attention to the powerlimived downlink
Since uplink fading can usually be overcome By means of up link
power control A convolutional code yielding a maximum power saving,
‘of 10 dB is assumed throughout. Hesults are directly applicable to
thor a wideares coverage systom or angle scanning beam syster*
but the modeling and analytics] approsch can be extended to study
outa systems that aro fxed-boam satelite-switehed, multiple san
‘ing beam, or hybrid fied seanning beam.**

‘igure 1 shows a typical sequence for interconnecting the various
spot-boam footprints Bach intarconnection containa one or more ine
slots during which ground stations within the connected regions com-
‘municate on a sequential basis, Although all sta time slows ean be
‘ade available to accommodate normal network traffe demand, the
spproach taken here specially reserver 4 certain number of time
Sot, sown ache end ofthe frame, for use exclusively during the rain
fade events. By ao doing, we have reduced the traffic handling capa:
balicy ofthe system by a small pereentage, while guaranteeing that,

oti ade named ot of roe ease aea owes

165. THE BELL SYSTEN TECHNICAL JOURNAL, FEBRUARY 1961
some extra cme sols wil be avilable as needed to maintain reliability
teiteuita already in use when fade eventa occur.

"The numberof reserved alot necesnzy to provide x specified degree
ot protection a all sations is dependent upon several factors. Caz,
{he uty of rsouroe sharing is dependent upon the joint Fedo statis.
ties at multiple geographically remote sites: a small numberof reserved
slots cannot provide protection if the probability of simultaneous
‘xouss attenuation at many sts, conditioned upon the occurrence of

fexoess attenaation at any onesie, high fortunately, experimental
Jolnt-de statistics at multiple remote sites are unavailable, and we
must vote to modeling to obtain quantitative results

"The uit of resource shaving alao depends onthe numberof ground
stations in the nwcwort and the traffic niribulion sroong those ground
Stations Ifche number of ground stciona is ema, then the fraction of
‘toma tna slots which must he raerved to protect even one sive must
be large, resulting in an inefficient solution to the rain fade problem.
‘Also, ifthe trafic distribution ts highly nonuniform euch that afew
{round stations carry a deproportionately large volume of traffic, then

‘becoren impractical to reserve enough tie slots to protect
thisamall numberof large rs. Tn such an event, itmight be desirable
ta protect the Inege we by some other technique, such as Tarker
fancennas of site diversity, and empiny reauree shaving forthe exchi~
‘Sve protection of the mach lager numberof ama users.

‘Similarly, the vflectiveness of rerource sharing depends on the
geographical distribution of ground salons relative tothe profiles af
high rainattenuation regions, and upon the volume of trae caried
hy ground sation: in high rain-attonuation regions. An additional
factor i the relationship between the by hour, whan all time slots
rol reserved for resouree sharin might be expected to be in heavy
‘emand, and the tine-f-day occurrence of significant rin-atlemuation

eats
‘We shall presenc both a maltiple-site ain-attemuation model and a
population dependent (rffic model, upan whichis based the subee-
tiuent predicted performance of the resource sharing concept. All
‘eoumptiona implicit in this modeling are addveasodin detail in Section
TL Section 11 contains th mathematical analysis of outage based on
the rain and traffic models Section IV eantains numerical resalts of
{his analysis; the effoots of Use various factors axe presented paramet-
ically. A gpleal rule shows thae for 12/144GHr network of 100
jdental ground seations, share erource reservation equal to six
pereent ofthe totel transponder time alos will provide an clays af
hour per year with 9 aR less rain margin than olherwive needed;
diurnal, seasonal, snd geographical dependencies of joint rain-fade
statistics are accounted for in this prediction

RAN MARGIN MPROVEMENT 169
Figure 2 shows statistics for single site rain attenuation It plots the

tion of ime thal ain fading exceeds the level of the abscissa
averaged over une contiquous 12 month interval "This fraetion can be
interpreted athe probability that any given rain-attenuation eve i
‘xcetded. From such sngle-ite curves, we develop a model to be used
for predicting joint outages at multiple geographically remote sites.

"At fre lance, one aight assume that if two siles are widely
separated, then Tain events at the two ates oocur independently.
However, thia cannot be the case beause in tho 12-GHz satelite band,
‘ain attenuation in exons of 3 dT ix eypieallynacocitod with thander.
‘norm activity which produces intense tanfall Periods of thunder.
Storm activity are typically restricted toa four-month interval lasting
from June through September, and to an intorval of six hours each
day lating fom 1 puto 7. Thos, if we are told thet the rain
sluenuation stone of the sites ia, say, 10 dB, then the probability of
Simultaneous deep fading a. he secon] site runt be higher thac its
yearly average because. nt thas moment, we are Hey ti he in the
inerval when thunderstorm activity normally occurs, Thus, knowledge

neat enn cre svn ts sn tt oo yea

170. THE BELL SYSTEM TECHNICAL JOURNAL, FEBRUARY 1981
‘of rain attention at ane ste affecca che probability of simaltaneon
Srenustion atthe second site, and the ewo evenie ate not independent.
‘Suppose, however, chat we arsume (hat all ofthe probability of Fg
1 is attributed to the period of thunderstorm activity. Then, if we
‘earit our attention to this period only (it only during this period
‘when resource sharing in needed to combat fading), the probability
‘hat attenuetion oxceeds a given level is ubout 12 times the yearly
fveragedvelue of Fig. 2 (4 out of 2 months per year, 6 out of 4 hours
per day), Letn(A) be the yearly averaged probability that atzenuation
‘nena level A at site No.1 Thus at a given instant of time ¢

‘paAL within thunderstorm period) = amslA), a
‘pi(A\ ¢ outside chunderstorm period) = 0. e

In (1), for the illustrative example given shove, « = 12. Chocsing a
Tanger value of a enhances the probability of multiple sraultancous
Fler since the yeady.averaged probebilly is then attributed to a
arower interval, In what follots, wu conservatively assume the
thunertann peri to be 3 out of 12 mouths and 4 out of 2 hours,
Yielding «24 The factors applies at al sites inthe satelite network,

‘Nom, given that (is within the thunderstorm-actvity period, the
‘eventa of attenuation exceeding levsl A te ite 1 and level B st site 2
may be ascumed (oe independent ifthe (wo sites are widely sepa
faved. For example, knowledge that attonuation excess level A in
‘New York during che thunderstorm-activity period provides no infor.
ration concerning the event that attemiation exceeds level B in

o
“

where p:2(A, 0 the probabiley that attenuation at sie 1 exeveds
level and attenuation a site 2 exceods B simultaneously a Lime £
‘auntion (is readily generalized forthe ense ofan apbiteary aumber
of widely separated sites,

For wn closely spaced sites, unether degree of attenuation event
comtelation iv tssinisl beyond the eennal ad diurnal earelations
jut edressed, Here, attenuacion atthe lwo alts may be produced by
the same sors, We assume tht, for closely spaced site, all fados in
facet of some thunderstorm characteristic level Ay always occur
‘rithin an ZP-hour interval ofeach other, where He much larger than.
the typical several minute duration of doep fades. Suppose tht a fade
tf level A > Ay aceute at some se. Then, at « neighboring site, he
Drobbility that a fade of level A> A, i simultaneously occurring is
ven by CU}, a fade level dapendent vonstant over the Fehour

RAIN MARGIN IMPROVEMENT. 171
Wwindosr. At this cecond site, the yearly everaged probabiliy of «fade
of level B > Ay is ateibuced exclusively to che H-hour iovervals
‘urrounding the events A > A, atthe original site. Then,

a
PAB) = e018), o

whore Tis the number of hour in 2 year and «isthe average number
Urevents per year that che atzanuation level exceeds A, Now,

piss) «
Pit, B|t) = falptAdpAB. (@a)
tard else see

Pant An 209, AD) = Aap one pA), (8)

1h the fllowing, we will allow the “geographical correlation fate
‘Alas vary between 1 and 6, The axtemne value = 1 implies tat Ff

7. and that, within the thunderstorm period, fadee occur indepen.
dently. Recognizing that ais equal to the numberof houra ina Year
fdividad by the annual number of hour inthe thunderstorm-activty
prlad, we see that for = 1, the average nuraber of fades per year
multiplied by the uncertainty window H equals the number of hours
in due dnunderstora-actvity peiod. Hence, knowledge that a fade of
level A > Ap is occurring at some site does not restrict the interval
within the thonderstarmactivity peviod when a fade of level 3 > Ay
‘ay occur at @ neighboring site, Similarly, tho extreme value f — 6
implies for example, that knowledge of deep fading at a given ste
pinpoints the twe out of four thunderstorm hours per day and the
average of one out of three days during the thunderstorm period when
Intense rainfall might oceur in the region surrounding that site. Con-
Aitoned upon an aftenuacion event at ome site, the probability of an
stenaation event at a eecond site within the surrounding region i
then ai times higher than would be expected from diumal and sem
tonal-orrelation conaiderations alone,

‘Thieconservatism ofthe values = 24nd B= Bean be demonstrated
by anplying oa, (Ba to experimental aeration data obtained with
site diversity euch as appears in Ret 7, Averaging (8a) over onc full
year, we obtain

prs) =} th, B))= Aopen,
For attenuation grastr than 5 4B, the sive diversity meusurements
‘are somewhat more optimistic then predicted by eq. (2). Tis obser-
‘ation, coupled with ground ration separation much wider than used
Tor the site diversity measurements and the physiel interpretations
iven tow and above, confirms the conservatism of the approach.

"To apply the above rain-attenuation event model, we need to know
the yearly:averaged atenvetion statistics al he Tocaion of each ate
in the nclwork. The flloming simplification a invoked: We divide the
‘continental United States into tee regions uch that che yearly
‘averaged attenuation statisties that apply ata representative site in
‘ove region aze (ypical forall etes in that region (eee Fig. 3)- Los
‘Angeles is chosen ns representative of the western region (Region 1),
‘where ain atleastion occurs infrequently. New York is chonen as
representative of the northeast contr region (Region 2), throughout
tthich rain attenuation bs moderate. Finally, Atlanta is chosen as
Teprecentative of Ue southeast central region (Region 8), where rain
‘ttenuation oocurs frequently. When applying the model, emake the
pesimistie wsunption that the “gwgraphieal correlation facto.” &,
fpplics al all sites throughout an cole region; for sites located in
‘ferent regions this factor ia neglected. Thus, for this model, "close
‘ness of tro sla i defined by whether both sites are within the sume
region, The enrors incurred hecauso of sites lected neur euch other
but on opposite sidee of regional buwniry nme more than offeet by
the large “cormelationdistancee™ nani

22 Tha ratte model

‘Th trate nel used in tis wri based on a vank ordering
ofthe 100 rast populous continental Vnited States cits, ax shown i

Tm mae ne ep ne te
sgn eR ORES Sct pete

RAIN MARGIN MIPROVEMENT. 173
“Table L! We assume thatthe trafic between any two af Unese ites
inversely proportional to the product af thelr indices Trafic between
te cities closer than 500 miles excluded from the network. Under
‘hese mssunptions, Region L offre 24 percent of the total etelite
‘afi, while Hegions 2 and J offer 63 percent and 13 pereent, respec:
tively

"The number of ground stations werving a given region ix asurned to
be proportional to the trafic fered hy that region. For example, if
‘the network contains 100 ground stations, chen 24 are located in
[Region 1, 6 are in Region 2, and 18 are in Region 3. AT! ground
‘tations are assued lo carry identical tai ero sections, Ts for
‘the purposes of analysis, che caf differences among regians.or among,
‘lies within etch region are accounted for by sesening more o fewer
[dential capacity ground stationa, a the caso may b, to accommodate
the traffic, The modeling und analysis ean be extended to networks
comaining a mixture of lage and soll tafe yroural nations If the
lange users are protected by site diversity and if resource sharing is
‘applied to provoe only the larger number of amall users. This extension
Thae not heen cirved out, however, and the numerical ral to he
prevented are valid only forthe former case.

"Two additional aumprione are alzo made, Pret, all ground etationa
Ina given region are assumed tobe identical, bave the same built
in rain ce rargi the bili narging af ground saline dierent
gins nee ot The the same, however. Second, we conservatively
tsesume that dhe leafic-buey period, when all ime late not reserved
for resoutee sharing are in fll-cime ue, coincides with the thunder.
stormeactivty period. Thus, the veulta are intentionally made to he
Pedsimiatic since i is precisely when recource shaving ia needed that
‘xtra time slote not contained in the Teeerve pool are unavailable
during trafic off-peak hours, whce additional slots are available,
‘eoturee sharing iv nat needed beenuse rin fading doen not cour

IIL RAIN OUTAGE ANALYS'S
3.1 General sperosch
Consider the thrse-rogional map shown in Fig. 3. Let there be
ground stations or sites lated in Region 1, cach of which has a bult-
fn fade margin of Ay 4B. Similarly, let there be N and Ny ies located
n Rogions 2 and 8, respectively, with fade margin of Ay andl An a.
‘The extra fade margin proved by resource sharing if xtra tee lots
sre available is M dB, that is, dB ig the maximum extra margin
provided by the cading approach employed. All wtes camry the axme
‘olume of taffic, and encugh time slots are reserved to accommodate
-K simultaneous fades. We wish to find the yously-averaged probability
that a given sive is operational when all sites in the network are
‘competing, 9 needed, forthe reserved resource-sharing slots, With mo

174 THE BELL SYSTEM TECHNICAL JOURNAL, FEBRUARY 198%
jndod Aq S0qF> "S'T YBABLE| 09 | Gt yO BuLeRIO-NUEN—I MEL

AN MARGIN PROVEMENT. 175,
lose in generality, we find his probability for site Tocated in Region
1a simple permutation of indies enables this result to be applied in
cither of the ewe remaining 20%

At uny ste oithin Region 1, Uhre disjoint events may oocur at any
‘point in time (1) the face depth F raay be les han Ay, (2) may be
hotwoen Ay and Ay + Bf, ond (3) F may exceed A; + M. Call these
event, Ey, and Ey, Then

Ploperational|t) = 5 Ploperational|E,0PUE¢). 0)

Clearly,
P(operational| ££) = 1, ay
Properational |, #) = 0, @
Pale
[Lets Csithin thunderstorm period, 1
1 otherwise, i
Pu|O
JaptAs) —ap(As + 3), twithin thanderstorm period, 4
ny otherwise.

‘Thus for ithin the thunderlarm petiod,
operational 1) =~ apts)
+ aP operational | Es, [PA:) ~ P(A + 7, (15)
‘nd for ¢ outside the thunderstorm period,
‘(operational |#) = 1. 8)
From (15) and (16), we obtain the rsult tht, averaged over an entire
year
Pinot operational) = p(s)
= Ploperational| i ALA) ~ pid +A}, (17)
where é= (¢ within thunderstorm-acivity period)

3:2 Borivaton of Plaperational| Et}
for € within the thandenorm period and conditioned on the
event Fz a partiular site wil he operational if che number of other
‘et which eel to use the reserve pool sles than KC~ 1. Also, i the
number of other sites which need to use the pool is equal to j = K,
‘hen the probability chat a particular sit is one ofthe K ites permitted
tore the poo ie equal to K/\j + 0

{76 THE BELL SYSTEM TECHNICAL JOURNAL, FEBRUARY 1051