IMAGE ANALYSIS AS A TOOL FOR SATELLITE-EARTHPROPAGATION STIJDIES

Riy,a Akturan, Hsin-Piao Lin and Wolfharcl J VogelElectrical Engineering Research Laboratory

The University of Texas at AustinAustin, TX, 78758

AbstractWc present a progress report on a useful new method to assess propagation problems for outdoors mobileEarth-satc]lite paths. The method, Photograrnmctric Satellite Scrvicc Prediction (PSSP), is based on thedetermination of Land Mobile Satellite Systems (LMSS) service attributes at the locations of static orrnobilc LMSS service users by evaluating fisheye images of their environment. This paper gives anoverview of the new method and its products.

IntroductionAll personal satellite communications systems (S-PCS), whether targeted for vehicular or personal usc andwhether operating at UHF, L-, S-Band or higher frequencies 19, 20, 21], are ultimately performance limitedby onc of three basic propagation states [22]. The line-of-sight signal may bc (1) clear, (2) shadowed bytrees, or (3) blocked by solid objects, i.e. buildings and mountains. Measurement campaigns to establish apropagation database for systcm simulation and performance prediction using real or simulated satellitetransmissions arc very time-consuming and expensive. Thus, a lnore cost-effective method of determiningthe probability of the three transmission states is needed.

Previous field mcasurerncnts [1, 22, 23] have demonstrated a statistical link belwcen fading and the three-state description of signal propagation. This research [2, 4] expands on that viewpoint with measurementswith a systcm that evaluates full images of the upper hemisphere and dctennincs where the sky is clear,where it is shadowed by vegetation, and where it is blocked as illustrated in Figure 1. Assuming that enoughimages are acquired in a specific environment, one can develop statistics for the three fade states as afunction of elevation and azimuth angles [3, 5, 6] and these can bc colnbined with state-dependentpropagation data to produce realistic simulated data or fade statistics [9].Sat f @ ell,azl

.4 sat. 2 @ e12,az2 Sat 363 e/3.az3

Figure 1. LMSS communication channel illustration.

There are a number of reasons why thismethod works: ( 1 ) Propagation effects for S-PCS are imposed by the physical environmentwithin a short range from a mobile userterminal. (2) I’hc propagation paths areelevated and thus permit visual inspectionalong their near-termina] extent from themol)ile’s l o c a t i o n . (3 ) The s t a t i s t i ca lproperties of the rcccived signal levels can bedescribed by a few, physically reasonabledistributions which are a function of the fadestate. (4) Suitable parameters for thedistributions at frequency bands of interesthave already been determined by manymeasurement campaigns.

Hcllce, plovidcd that the propagationparameters arc known for an LtMSS service environment, a gcncralizcd description of signal propagation ispossible when conditioned on path states, and the path states can bc readily obtained from opticalmcasurcrncnts. This idea permits the cost-effective prediction of service availability in a spccitlc areawithout the need for ncw RF measurements [9:1. The n mhod is especially effcctivc in ( I ) predicting fadiw

243

probability as a funclicm of elevation angle, (2) path (a.k.a. satellite) clivcrsity gain, or (3) good-calldurations for a specific constellation at a given latitude, all in a chosen environment.

In pho(og,rammctric analysis, still and video images of the sky arc taken through a fish-eye lens with a 180°field-of-view. Efficient image processing algorithms implement a three-state classification of the fisheyelens images. Using the products of image processing, two- and three-state environment statistics are derivedby determining the skyline and three-state fractions in the classified images. This image data is applied toderive the e]cvation angle dependence of fading in LMSS channels. In addition, by placing satelliteconstellations on the irnagcs, time series of fading and Markov ntodels of the communications channels aregcncratcd. From the time series, first order fade statistics are derived, ancl h~arkov models of thecommunication channel arc prepared. Again, for the first time. an efficient procedure to derive satellitediversity gain and good-call durations by photogrammctry is introduced. In this text, satellite diversity gainis estimated extensively in urban, suburban and rural environments at diffcl cnt latitudes for differentdiversity operations. This derivation is performed for a diversity system that is able to operate a k-foldsatellite signal combining scheme. Resulting tirnc series of the diversity and non-diversity operations arealso used in good-call duration calculations. In the following, this method and its products are presented in a]ogica] order.

Still and Video Image ProcessingOur imaging systcm includes a fisheye lens, a HI-8 camcorder, a 35-mn~ still camera, a GPS receiver, ascanner, a video frame grabber, and VCR/TV units. Images are either 35-mnl stills or 30 frame/second HI-8videos of the sky taken with a carncra through a fisheyc lens having a 180° flcld-of-view, as shown inFigrrrc 2A. Images are digitized and stored on a large capacity r nedium, typically a CD-ROM. In order toexamine propagation characteristics of different environments, pictures are takcm in urban, suburban or rurallocations. Fishcyc pictures are taken at the strwct-side of the sidewalks in the densely built areas and aretaken on the roadsides in rural areas. Sampled locations include the USA, Canada and Japan.

The image processing system includes three steps [7, 8]. They arc pre-processing, state separation and stateanalyzing. lmagc processing starts with pre-processin~ of the digitized fisheye images. This involves twosteps, north direction adjusting and unwrapping. When taking inlages, a CIPS receiver is used to record thernobilc vchiclc’s heading direction in a log. In processing, the tislleye images are rotated, using the log file,by the azimuth angle from nor{h so that north always stay on the top. Later, image processing continues byunwrapping the images. Unwrapping, as shown in Figure 2B, converts from a circular image in which thezenith is at the center (o a rcctarrgular coordinate systenl in which the zenith is at the top of the frame.

Next, still images are statistically analyzed using their selected spatial features [7]. The image processingsystem extracts vegetation and blockage features using search v indows in the images and evaluates thesefeatures to perform the thre.c-state classification in the regic)ns defined by the corresponding searchwindows. To classify images, initially all image pixels are assunled as clear sky. When the total score of asearch window feature density exceeds a threshold value, a state transition occurs; otherwise the pixel staysin the same state. This way, three-state images are obtained based on clear (o blocked (Figure 2C) toshadowed (Figure 2D) sky transitions. Lastly, small objects are removed by erosion and dilation. To test thestill image analysis algorithm, three-state fractions in urban Japan arc rneasurecl using automatically andmanually classified images. The ratios of clear, shadowed, and blocked sky arc found as 58.7%, 4.4%, and36.9%, rcspcctivcly. Remarkably, the photogrammetric measurement results of the urban Japan imagesmatched with field signal strength measurements [23]. As an error analysis, automatically classified imagesare compared with those classified rnantrally using a paint prograln and mouse. The difference is assumed asthe error measure of the irnagc processing algorithm. The largest, 23%, and the lowest, 570, State

recognition errors arose in the urban and rural environments, respectively.

2 4 4

kih—. — .—

B. Unwrapped Inlay C. Two-State Image

r--—~

D. Three-State ImageA. Fishcyc lmagc

Figure 2, still image analysi~ steps

Similarly, our video image processing technique [8] uses the R(iB color palette as an object classificationset to differentiate the image pixels into onc of three state types. Since the algorithm uses spatial imagefeatures in classification and only processes images that have cl,angcd since the last scene, the processingspeed is vastly accelerated [8]. First, fisheye video images arc unwrapped, as explained previously. Second,kcy frames arc cxtractcd and t bird, using these key frames, fisheyc videos are processed.

The implementation of the video image analysis prc)cedure shows 80% clear, 1470 shadowed and 4%

blocked sky in a rural I,os Angeles area. Processing errors are 7%, 5%, 4%J in urban, suburban and rural LosAngc]cs environrllents, rcspec[ive]y. ~igr-rres 3A to 3]+ show the video image analysis steps, sample andproccssccl fishcye images. In this image set, the gray regions correspond to the shadowing objects; the whiteregions c~rrcspond to the Open sky. This specific image sequenc~ dots not contain blockage objects.

+,_J~ p=~yState Analyzer———

Optical Module image Processor

I I~

9:4?29 AM

A. lmagc analysis system B. Sample video C. Analyzed video

Figure 3. Video image analysis steps

The programming and implcrmmtation have been done in C and C++ under a PC-Windows environment ona PC-586-90 platform. The effort has been directed towards cleve]oping reliable algorithms as well asminimizing processing speed [4]. An average video image analysis takes lCSS than 5 s/image. Since therequired size of the image data to bc stored is large compared to the size of a~rai lablc disk space, a Iossless4:1 compression scheme is developed and the compressed images are stored in a binary file [4].

In the fade s(atc analysis, classified three-state images are read f] om the environment fade state tables as afunction of azimuth and elevation. Classified three-state irnagc rcsolrrtion is 360x90 pixels, which presents a1 ‘xl 0 resolution in mapping the environment.

Environment Quantification Using Two-Stnte Model: Skyline StatisticsIn this section, results of two-state image processing arc presented. T’hc flshcyc images of four rural,suburban, and urban Texas locations are analyzed to derive quantitative inforlnation about the elevation

24.5

angles at which the sky becomes visible, i.e., the skyline [3, 6]. ‘1 ‘he mean skyline angle for an urban Austinimage, Figure 4A, is shown in Figure 4B and the average histo~ram of skyline elevation angles shown inFigure 4C. Environment average skyline elevation angles in Austi n’s rural, suburban and urban areas are 5°,7°, and 26°, respectively. In the three environments, the 90’” percentile of the skyline was found to be at110, 17°, and near 55°, respectively.

(Urban Austin, Texas)

A. Fisheye lens image

“(l 45 9! 135 180 225 212 315 360

IMC4745L AZIAIUTH D:G

B. Skyline

,0

0

-~

I-l

‘1‘1

‘1-1

:10 !0 20 30 40 50 60 70 80 so

E!evallon degh. ,., ”,O, !,’!, 4,0,

C. Skyline elevation angles

Fi~urc 4. Fishcyc image of an urban Austin location, its skyline and skyline histogram

Above 10° elevation, on average 98’?ZO of the sky is visible in rural, 95% in suburban, 77910 in urban Austin,and 68% in urban San Antonio as shown in Figure 5A. The potential of satellite diversity to mitigate tree-shadowing or blockage depends on the variability and structure of the skyline wilh a?.imuth at any particularlocation and on the satellite constellation. The 10’h percentile, mean, and 90’h pcrccntilc of the skylineautocorrclation in all three environments decreases to l/e with a lag of about 15°, 30°, and 55°,respectively. The skyline autocorrclations in Figure 5B have a similar shape in all environments. Theautocorrelations cross zero near 70° azimuth lag. The mediarl autocorrclation falls to zero near 50°,mcaninfl that for half the locations in urban Austin there is an evetl chance that at 50° satellite separation thesecond satellite. will be clear if the first one is blocked (or the conirerse).

100

95

1

90

85

80

15

70

65

60

55 L”~”—.. ~- .—~.Jo 10 20 30 40 sO 60 70 80 90

p$v.all Elevallon Angle. deg

Skyline Aulocorre!at(on SMIISIICS, Mean Compar($on

A. Sky visible ratios above given elevations B. Means in skyline autocorrelations

F1g~lre 5. skyline visibility and autocorlc]ation statistics

Environment Quantification Using Three-State Model

10

Until now, environments have been defined verbally such as heavily or lightly shadowed. In this section, anenvironment quantification method is presented. Using ternary descriptions of environments, three-state

2 4 6

BLOCKED

/&._qe

+...89. ,

~ ‘“”” –—.____.– .-_.._._.‘)0.25 0.50 0.75

CLEAR SHADOWED

Figure 6. A ternary graph of propagation states in urban Japanas a function of the elevation angle in 5 degree increments.

images, environments were rnappcd ontoCnvironmen( triangles as illustrated inl;igure 6, so that a quantitative ternarycicscription of the environments wasobtained [12].

[Jsing the irnagcs of urban Japan, thefraction of potential satellite paths withclear, shadowe{i, or blocked line-of-sightwas calculated in 5° elevation angleincrements from 0° to 89° [10, 11]. Theresult has been Jjlotted in Figure 6 in theform of a ternary graph. It is obvious thatt,lockage as opposed to shadowing is ofI,rimary importance in the urbanenvironment and that with decreasingelevation less of the sky is clear. Forexample, in lhc low-elevation intervalfrom 10° to 14°, 17% of the sky is clear,8% is shadowed, and 75% is blocked,compared against the higher elevationinterval of 60° [o 64°, where the state

~nixture M equals (0.8, 0.03, 0.17).

Elevation Angle Dependence of LMSS FadingCombining each path state derived from images for single or multiple satellites in a specific constellationwith frequency-appropriate statistical fade models allows predicting overall performance measures such asfade dcpcndcncc with elevation angle or path diversity gain [ 12, Id]. Using the three-state method, for thefirst time, elevation dependence of fading is cicrived [10, 1 1]. In addition, the importance of includingspecular reflections under urban blockage ccmditions is estatllished with an urban three-state signaldistribution mode] [12].

Fishcyc ima:cs of urban Japan were analyzed to derive, as a function of elevation, the fraction of sky that isclear, shadowed by trees, or blocked by buildings. At 32” elevation, the path state vector matches thatderived from satellite measuren]cnts fit to a 3-state fade model [10]. Using, the same model, for the first timethe elevation angle dcpcndcncc of mobile satellite fading is predicted [ 10, 11]. ‘l-he dominance of buildingblockage fading in urban areas is expressed in the terracing of the cumulative ciistribution. It is found that inthe Rayleigh domain, the minimum elevation with 90% coverage decreases by about 5“ per dB of additionalfade margin [11 ].

Heuristically, each of the three. possible path states of a mobile satellite link can bc associated with a distinctfade distribution as expressed in this text. The clear state is usually described by a Ricean distribution,which represents the superposition of a line-of-sight (LOS) signal with multipath reflections, but neglectsspecular effects. The shadowed state was modeled by Loo [24] as a combination of a lognormallyattenuated Iinc-of-sight signal with Rayleigh distributed diffusely scattered powc.r, and in the blocked stateonly Raylcigh distributed diffuse]y scattered power is considered.

l(v) = C“ .Lice(v) + s “ .LO.(V) + B “.La,vki~k(”)( 1 )

Furthermore, the importance of including specular reflections from nearby illuminated building surfacesunder urban blockage conditions is added. This study improved on the model’s fit by including potentialspecular reftcctions in the blocked state an(i by refining tllc distribution parameter estimates. TO

acknowledge the prescncc of such specular rxftcctions, the distribution in the urban blocked state waschanged from Rayleigh to 1,00, resulting in the urban 3-state model

j(v>a) = c(~) ‘ ~~;~~(v) ‘“ ‘(a) “ ~~~~(v) + ‘(~) “ .~~{><(v) (2)

2 4 7

Including specular reflections in the blocked state reduces the rm~ modeling error (right ) from about 1.3 dBto 0.5 dB as shown in Figure 7.

tdl.32 .Jfl 02129r9G 1531 02

95 --

90 ---1.1

70 -_

50 -:

30 - -

- .

10 -

J

L IIL 1 I L-1-- LL- 1.1 1 1. 1.) .l–M1-I-l. U-]–1–LL 1 L

[-””:.:.- “ _

+ Satellti!e Eleacon M e a s u r e m e n t [9)

- - - - - - - R u r a l 3. Slale Model, Ir!lerfed (C,S,B) [9]

- - - - R u r a l 3 - S t a l e Model, M e a s u r e d (C,S,B)

1Urban 3. Slale Model, Veasured (C,S,B)

+ ‘%-.

‘~. ;

\~

. . . .‘ . \ .:..

., ‘+;,, +;,

. , +;... r -

‘ ..

1--1–7 7- l–r-7--I- r ‘~–r–r r ‘qm–r T T~: –r- ,–r~r–r- ~ T T 1 11 . 1 - -LL’.

,’ ,‘.. -.. . . --

-,,/’ .’,,,,, ,.,

,, ,’, ,’

.~,.~T–r ;.,.. .—

.5 0 5 10 15 20 25.4 0 4

Fade (dB) Exceeded at Probabi l i ty P Error (dB)

Figure 7. Comparison of a CDF measured in urban Japan at 32”elevation with three 3-state models.

99--

9B -“:

95 ~.

Fade CDl~s for urban Japan atelevation angles from 7° to 82°,based on photogrammetrica]lydctcrmincd path states and the urban3-state fade probability model, arshown in l~igurc 8. Photogrammetrypermitted the prediction of fading asa function of the path elevation angle,and simp]c ternary plots are used toestimate non-diversity fading as afunction of the path state vector. Inurban Japan, it is found that therequired fade margin dccrcases byabout 0.2 dBldcgree, from 25.5 dB to15 dB at the 10% probability leveland in the elevation range from 12°to67° [13, 14].

cdl_287.grl 02/29’9t, 17:4415

1

LIJ-.J. i L M d L IL. I.

Elevation = n“ +- 2°

0 5 10 15 20 25 30 35Fade (dB) Exceeded at F’robability P

Figure 8. Fade CDFS for url,an Japan

248

Received Signal Generation‘I’he system presented in this section substitutes video image processing for propagation measurements andpermits ftexiblc fade simulation of Land Mobile Satellite Cornnmnication systems [15]. The video imageprocessing system is used to track the propagation states of the satellites in selected environments. l’hc pathstate tracking is then applied to generate a satellite to mobile unit path state sequence as a time series.Analyzing sequences of irnagcs resulted in a propagation state time series which is input to a channelsimulators shown in Figure9. By assuming that the channel attributes are quasi-static in each propagationstate [22], the statistical models are represented by Rayleigh for urban, Loo’s for shadowed and Ricean foropen direct LMSC paths. Using these state-specific propagation statistics, a time series of the receivedsignal C1lVC]OPC is generated fromthetinleseries ofthc produced state sequences. The simulated signal timeseries arc subsequently used to derive the first order propagation statistics of the communication channel

(Random number generators)

Figure 9. Received signal generation procedure

I Vm(t)

In simulation, a one-minutesection of a video clip taken insuburban Pasadena is processed.A ccmtparison of the fade seriesobtained by simulation and bymeasuring 2055 MHz. CW

transmissions from a TDRSsatellite at 210 elevation and 249°azimuth in Pasadena, CA isshown in Figure 10, right graph.The rms difference between thetwo PDFs is 0.11. This closematch S}IOWS the effectiveness ofthe ptmtogramrnetric procedure.

E ~erunantal Data Experimental ati s,mu!ated C O F—_..—m —.. –—0 --’’++==+: ‘0” : T 2“

. ————6-s..,~z> , .~ 43 .-Yi

,.. !. .

#-s ..-07-30 ~.. >

/ . -

-40 —,-... -.0

,.,

10 20 40 50 w “.s%:

. “

ci 10-’ , / ’

S!m.ktcd Data <> / ’

10

1 [

-— / ’,

/~os

,

z-lo/

t10-’

<-20.s”03 –30

_400LL—.—k.-— L__. L .—— . . . . ..— ,o-~ L_ 1 –——. - - - - - - - - - . . ..——..——

10 20 30 40 50 60 -30 -25 -20 -15 :10 - 5 0Scc Srgnal !@@ (dB) [- exr,crunenl.1, --’:stmulatd]

Figure 10. One minute time series of the measured and simulated signals and C.1).F. of one minute data

In a typical LMS scenario, while the mobile unit is in motion, the field strength of the received signal levelchanges with respect to the time and space according to the chan~ ing morphology of the environment. It hasbeen shown [25] that the switching between different attenuation levels can bc represented as a Markovstochastic process using the assumption that the channel pro~~crties are quasi-stationary in small timeperiods. Based on this propagation modeling method, our inlage processing and propagation channelsimulation algorithms are employed to obtain a statistical representation of the communication channel [4].The video image processing system is capable of tracking the propagation statr of the path to any satellite inthe selected environment, generating a satellite to MES path state sequence time series. Using random

2 4 9

number gcncra(ors and the state sequences as shown in Figure 11, the rcceivcd signal levels are prcdictcdbased on the state transition probabilities represented by Markov chains [ 16, 17, 18].

Q@LLt:Markov Chain Generated signs/State Generator—- —..

‘-’””’”h ‘~iii&,&!l y)——. ,., .,.

P(t,)— — — - .

J j--rE_.]Filter

(Random number generators)

Figure 11. Signal generation using Markov chains

data, there is no one-to-one correspondence with the measurements.

The sinmlation procedure wasapplied to the above two-minutesection of simu]tancous satellitefade da~a and video clip taken in

suburban Pasadena, CA. Using thepreviously dcscribcd procedures,the path state sequences, transitionmatrices and modeling parameterswere founcl in the sclectcdenvironntcnts for thresholds T1=-2.5dB and T2=-20dB.

Comparisons of the simulated fadetime series with measured data areshow]l in Figure 12. Because thesetirnc series arc generated using theMarkov method with state andtransition probabilities derivedfrom the entire two minutes of

5r — – -—-—— ———.-— -- ----- -.–—.-. --–——-0 - - - - -

- 5 - .,,,

-10 - ‘ .~-

.S-15 - .~ .

‘ .~ .20.p ‘ \U? ‘ .

-25 ‘ .

‘ .‘ .

-30 . .- .

~:

-..

-35 ‘ .. . .

-40 ——— .—-. ——— ——— .. —... —.2479 89,76 9917 99,94 9999

C F.D [-,:expcrlmcntaf, ,--’:simulakd]

Figure 12. CDF of the suburban Pasadena da(a

The signal level firsl and second order statistics,i.e., th~ cumulative fade distribution (CDF) andlevel crossing rate (LCR) agree, however. The rmsnormalized difference between measured andsimulated cumulative ICVC1 probabilities is 2.5%and the J ms normalized 1.CR error in Hz is 3.7 forwooded Pasadena. I“he best match of the CDFS isobtained in the suburban environment, where thestate transitions occurred with regularitythroughout the mcasurerncnt and at a moderate ratethat could easily bc followed in the video images.Related state duration and transition Markovmatrices are given in Figure 13 with a diagram ofthe Markov process.

As descl ibed, fishcyc image sequences are used togenerate the received signal levels. A sample of the

software is shown in Figure 14 to illustrate this process [4]. A Windows-based program processes thefishcye irnagcs and extracts states. Then, using the statistical signal distributions in respective states, signallevels arc generated as a function of time.

WccState. duration Matrix: State Transition Matrix:

W := (0.7522 0.2142 0.0336)

[

0.9815 0.0166 0.0018’

J’:= 0.0597 0.9170 0.0233

0.0413 0.1488 0.8099,Pbc, PC’CI

~l~ure I ~. State durations and tr-artsitions in suburban Pasadena, C A

2 5 0

Figure 14. Signal generation software

Satellite Diversity GainFor (I1c tirs( time [ 12, 14], diversity gain statistics are estimated for LEO satellite PCS. An image processingprocedure to track the visible satellite path states is developed. The diversity gain estimation involves a 4-fold path diversity scheme with k from one to four, where the bes~ k satellites are used in the link [12]. Thecalculation of k-fold diversity gain involves 4 steps: (a) deter] nining the CDI’s according to coherentcombining: v c v] + V2+..+Vk and switching combining: v = lnax(vl, \~2,.., vk )’ (b) placing the satellite

constellation into the 3-state images and determining the p] obability of occurrence for each statecombination, (c) summing all CDFS from step (a) weighted by the probabilities of step (b), and (d)subtracting the resulting CDF from the CDF of the highest satellite at the same city at probabilities from 1 %to 99%.

4

3

1

0315 330 345

Time (rein)

Figure ] 5. ThC image-average number of sate] ]itcs

Stacked bar graphs for 30 minutes of the in~age-averagcd number of satellites visible with (frombottom to top) clear, shadowed, and blockedpath states arc depicted in Figure 15 for aGlobalstar-like constellation, 48 satellites on 8planes, in Tokyo. The number of clear satellitesvaries periodically between <1 and <2 and mostof the time three or four satellites are available.The minimum elevation angle was 10°.

By employing this procedure, satellite diversitygain of any satellite communication system isestimated without actually using any satellites.Measut ements have shown that satellite diversityincreases availability of the communicationchannel by introducing, more satellites in clearpaths. However, satellite diversity gain greatlydcpencls o n t h e c o n s t e l l a t i o n ’ s orbitingcharacteristics. Path diversity gain for combiningand hand-off diversity is found for up to 4-folddiversi[y a t a high-, mid- and low-latitude

location for the sample constellation. With 2-fo/d or higher ordc~ diversity, it is near 10 dB and 6 dB atprobabilities below 20% when using two or rnorc satellites in the non-equatorial and equatorial locations,rcspcclively.

251

Figure 16 shows cohcrcnt combiningpath diversity for the Globalstar-likeconstellation in Tokyo that reducesfading relative to using the highestsatellite. Fading at 10% probability,for instance, can bc rcduccd fromabove 20 d13 to below 1() dB bycombining signals from as many asfour satellites.

Figure 17 shows a comparison of pathdiversity gain achieved by combining(thick lines) and hand-off (thin lines)diversity with up to four satellites ofthe Globalstar constellation in Tokyo.Gains for both methods are equalwhen using the best satellite, but withmore satellites combining diversityachieves greater gain.

Figure 18 shows a scrvicc comparisonhctween the four environments for theservice coverage of 989i0, 95Y0, 90%by rccciving signal from only thehighest satellite or the two bestsatellites using hand-off diversity.Using this setup, at 95% coverage,

98

95

90

80

70

60

50

40

30

20

10

5

Com TO KC D.grl 03/04/96 15:?4:53

~1 1 L~J-LJ.l LJd-L~J. I! I .I I l . I.J_L

.

—.

—

—

.

—

.

.

—

:/

y::, -...~’. , s.

<:. . . ‘.,. . \ I.

. . ‘ \. . . ‘, ‘ \

L-1ti.1 ILL-i

-1 ‘w. - . .

Combining DiversityGlobalstar, Tokyo

-—--— Highest Satel l i te

- - - - - - - Bes t Sa te l l i t e

------- 2 Best Satel l i tes

--------- 3 Best Satellites

-------- 4 Best Satellites

\

1. ‘:S ‘.. -’\x.. \..— \\ ., .\.

- 77 T rrl-r~ :-T-~rTl-ri I \1 r-, , ?l-l-}, T 1 l-h] I T-7 ~

-5 0 5 10 15 20 25 30 35Fade (dB) Exceeded at Probability P

Figure 16. Coherent combining path diversity

L–

the improvement obtained by using the two best satellites instead of the highest is 9 dB, 9 dB, 6 dB and 3dB in urban Austin, urban San Antonio, suburban Austin and rural Austin. ‘Ilc two urban environmentsshow similar fade characteristics. As it can be seen clearly, non-diversity fading in a suburban environmentis lower than diversity fading in an urban environment. This shows that satellite diversity is mostly valuablein urban environments although itdecreases fading in suburban anclrural environrncnts. The path diversitygain in mid- and low- probabilitiesare almost half of the diversity gain inurban Austin.

As a result of the photogramrnetricmethod, it is observed that satellitediversity is useful in providing anadequate link margin [4]. It increasesavailability of the communicationchannel and lowers the need for highsatellite elevation angles. Diversitygain s t rongly depends on theconstellation’s orbit parameters.Large azimuth separations betweenthe visible satellites and a highstandard deviation in the averageelevations of the visible satellitesresult in a high diversity gain, recalledfrom above. Diversity gain is high inheavily shadowed or blockedenvironments. According to thecnvironmcn( dcpcndcnce plots [4], a

C&ti TokOG,grf 031041961555:4:

, 5 +-h..] L-l . . . . . . I I ,. L..J 1 . . . . --L---

i 4 Best Satellites ~;::

/,..,, ./

‘ /“A ..:::.0,/

3 Be I Satel l i tes .-. ~-~;.” .

/ 1\--z.’.. ?.... .- :-----------

/

\:r--::::: L“;.”” ,.# .- ‘<’;=,.,+ --- - - - -

,(’/’”-.\.\.’<\.; - - - -, /./;:,.

dJ’f ‘“, ‘==:; ““’””

!;Y ‘\.\. 2 Best Satellites

Lf\

$?‘ \

‘ \‘\.

@t\.\.

- - - - - - - - - - -/’#*f Best Satellite

‘;2 - - - - - - C o m b i n i n g

Hand-off

o- 1111-]1,1[]111]1111]1111 II I II

50 40 30 20 10 5 2 1Probabihty (%)

Figure 1 T comparison of combining, and hand-off diversity

252

3 dB/environment gain increase is estimated from the combining diversity gain vtilues that arc 10 dB, 7 dBand 4 dB in urban, suburban and rural Austin, respectively.

I 1. Urban 1! I 2. Urban

Coverage

1=8~1l-lE! 95%

k<.?: g. ❑ 90%f’@

, % 1 ,.lz%kr= , Eki-E% &_,

L -L

\ I 3. Suburban B I 4. Rural B

&“’”J j k- [-Figure 18 Comparison of path diversity gain using combinin~ diversity with four satellites in Austin, TX.

Communication Channel Stability: Good Call DurationsSimilar to [26], a good-call duration analyzer is derived from the diversity analysis procedure [4]. Theprocedure places the satellite constellations on the classified images [ 12]. Later, for each time sample, thebest satellite and the highest satellites arc chosen, and their states are dete.rminccl as a function of time. Thestates for (hcse two satellites are written into a state-log file. One log file for the best satellite (non-diversityoperation) and onc log file for the highest satellite (diversity operation). Then, using a math package, statedurations are examined.

A good state or channel occurs when the line of sight (LOS) signal can be transmitted for the duration ofthe- call. Conversely,a bad state or channc]occurs when the LOSsignal can not bctransmitted for 18Scc . A bad statecauses interruption insignal transmission,and the current call isdropped. Toimplement thismethod, it is assumedthat the good stateoccurs when the LOSis clear, when therearc no obstructions inthe direct path from amobile or fixed userto the satellite.Therefore, the clearstate durations aretracked and call-completion success is

L.—. .—-— ‘[O (Diversity), O (No diversity) for a 3-Min call Ism~~~~. U99% Lrl 00%

C)87% 099%

Time(Min.)

❑ 0.946❑ 1.693❑ 3.16

❑ 4.103

■ 5.056

❑ 6.(D4

Figure 19. Good-call durations in different environments

2 5 3

calculated using the above procedure. The results are derived using a Global s{ar-like constellation withoutconsidering specific system parameters and implementations.

As Figure. 19 suggests, 60% and 359. of the three-minute calls ale successful] y completed without an 18 secchannel interruption using diversity and ncm-diversity opc[ ations, respectively, in the three urbantmvironmcnts. These numbers go up to 99% and 100!10 in suburban and rural environments, respectively,with diversity operation. The rural environment without diversi(y supports a call completion probability ofaround 99%, whereas in the suburban environment without dive] sity operation it drops 1270 down to 87910.

ConclusionAs a result of this work, derivation of a ternary description for clear, shadowed and blocked path statesallowed one to combine satellite orbit design with propagation statistics to predict performance such as fadeprobability or diversity gain in a specified environment. Photogi ammctry is a ncw tool with which onc canderive service predictions for personal and mobile satellite communications systems in any environment. 1[is inexpensive and simple compared to other techniques. The data collection process does not requirerccciving a signal from a satellite or other platform, and results can be applied to any frequency band forwhich statistical fade pararnctcrs are available. The products of ]Ihotogrammetric satellite service predictionshould be of immediate use to satellite system designers. The method opens new capabilities of propagationprediction for Land Mobile Satellite Communication Systems. A milestone table is given in Figure 20.

@EPEz:I::d—. .——

3-Fade State Classification by Image Processing(ClearNmdowed/bIocked) (@cc., 1994)

-— —- ——— ___ _—— —_ .._Statistical AnaIysis & “J’ime Scrks Simulation

. ,--... —— _____ -.— —.—e,F::5

2.State environment statistics (June, 1994)3-State environment statistics (Dec., 1994)

Fade probability ies (April, 1995)Satellite diversit~ gain (Oci., 1995)

Recewed slgna) smmlatlons (Sep., 199S)State duration & transition iMarkov models (Oct., 1995)

Good call durations (Dec., 1995)

—-

Figurc 20. Photogrammetric measurement sys!em architecture with thecompletion times of its steps.

Bibliography[1]

[2]

[3]

[4]

[5]

Birdwell, B., “Design of an optical sensor system to reldicate the effects of signal blockage andshodowing on satellite signs 1 J]ropagation ,“ M.S. Thesis, The University of Texas-Austin, Austin,Texas, May 1993.Akturan, R. and Vogel, W.J., “Photogrammctric Mobile Satellite Service Prediction,” Proc. of theEighteenth NASA Propagation Experiments Meeting, NAPEX XVIII, Vancouver, BC Canada, June 16-17, 1994.Akturan, R. and Vogel, W. J., “Photogrammetric h40bile Satellite Service Prediction,” Project Report,Submitted to Jet Propulsion Laboratory of NASA, Pasadena, CA, Nov. 1, 1994.Akkrran, R., “Photogrammetric Mobile Satellite Service Prediction,” Ph.D. Dissertation proposal,Submitted to the Graduate School of the University of Tex~s-Austin, Nov. 17, 1994.Akturan, R. and Vogel, W. J., “Photog,rammetric Mobile Satellite Service Prediction,” NationalAcademy of Sciences atld Engineering Meeting, [JRSI’95, Iloulder, CO, Jail, 3, 1995, p 210.

2 5 4

[6] Akturan, R. and Vogel, W. J., “Photogrammetric Mobile Satellite. Service Prediction,” ElectronicLet/crs, Feb. 2, 1995, vol. 31, No. 3, p 165.

[7] Akturan, IL and Vogel , W.J., “classification of Atmmation Sources In Satellite PCS Using ImageProcessing,” Proceedings of the Signal a)ui lrnage Processing Corfcrence, I-as Vegas, NV, Nov 20,1995.

[8] Akturan, R., Hudson, F. and Vogel, W. J., “A Fast Video IInage ProccssinS, I’echniquc For TrackingAttenuating Objects in Sa(ellitc Communication Paths,” Pr<)ceedings of the Second Joint Conferenceon lnforrncrtion Sciences, Wrightsville, NC, Sep.2tl, 1995.

[9] Akturan, R, Lin, H.P. and Vogel, W.J., “Elevation Angle I )cpendence of I~ading for Satellite PCS inUrban, Suburban and Rural Environments,” National Academy of Sciences and Engineering Meeting,Boulder, CO., Jan. 9, 1996.

[10] Akturan, R. and Vogel, W. J., “Elevation Angle Dependency of Fading for Satellite PCS in an UrbanArea,” Electronic I.etters, July 1995, vol. 31, No. 14, pp. 1125-1127.

[11 ] Akturan, R. and Vogel, W. J., “Optically Derived Eleva(ion Angle Dcpenclence of Fading For SatellitePCS,” Proc. of the 19’{’ NASA Propagatioti Experiments Meeting, NAPEX XIX, Fort Collins, CO, June14, 1995.

[12] Akturan, R. and Vogel, W. J., “Path Diversity for LEO Satellite-PCS in the Urban Environment,”Project Report, Submitted to Jet Propulsion laboratory of A’ASA, Pasadena, CA, Dec. 13, 1995.

[ 13] Akturan, R. and Vogel, W. J., “A Method to Derive Satellite Diversity, ” Nafional Academy of Sciencesand Engineering Meeting, URSI’96, Boulder, CO, Jan. 9, 1996.

[14] Akturan, R. and Vogel, W. J., “Path Diversity for LEO Satellite-PCS in the Urban Environments”Submitted to IEEE Trans. on Antennas and Propagation.

[15] Akhrran, R., Lin, H. I’. and V08C1, W.J., “Propagation Modeling In I.and Mobile Satellite SytemsUsing Photogrammc[ry,” Proc. of the IEEE Vehicular Technology Conference ’96, Atlanta, GA, April28, 1996.

[ 16] Lin, H. P., Akhrran, R. and Vogel, W.J., “Propagation Channel Simulaticm for LMSS and SatellitePCS,” National Academy of Sciences and Engineering Meeting, URS1’96, Bou]dcr, CO, Jan. 9, 1996.

[ 17] Lin, H. P., Akturan, R. and Vogel, W. J., “Photogrammetri( Satellite PCS Channel Modeling UsingMarkov Chain Approach,” Proc. of the lnternationa[ Confe) ence on Cor?lt?llttticotiorzs’96, Dallas, TX,JLInc 23, 1996.

[ 18] I.in, H. P., Akturan, R. and Vogel, W. J., “Satellite-PCS Channel Simulation in Mobile UserI;nvironrncnts Using Photogramrnetry and Markov Chains,” Submitted to Special Selected PapersISSUC in Communications, ACM/t{altzer Wireless ltforrnatiorl Networks (WIAIET).

[ 19] Schindall, J., “Concept and implementation of the Globals[ar Mobile Satellite System”, Proceedingsof the Fourth International Mobile Satellite Conference, Ottawa, Canada, June 6-8, 1995.

[20] l’adano, H., “IRIDIUM- a Lockheed Transition to Cornmcrcial Space”, Proceedings of the Fourthinternational Mobile Satellite Conference, Ottawa, Canada, June 6-8, 1995.

[21 ] Sturza, M., “Architecture of The Teledesic Sa(cllite System,” Proceedings of the Fourth InternationalMobile Satellite Conference, Ottawa, Canada, June 6-8, 1995.

[22] Cioldhirsh, J. and Vogel, W. J., “Propagation Effects for Lalld Mobile Satellite Systems: Overview ofJ3xpcrimental and Modeling Results:’ NASA Reference Publication 1274, February 1992.

[23] Karasawa, Y., Minamisono, K. and Matsudo, ‘1’., “A PI opagation Channel Model For PersonalMobile-Satclli@ Services,” Proc. ]994 Progress in llectrornagnets R e s e a r c h Symp., ESA,(Noordwijk, The Netherlands). 11-15 July, 1994.

[24] LOO, C., “A Statistical Model for a Land Mobile Satellite Link,” IEEE Trans. on VehicularTechnology, vol. 34, no. 3, pp. 122-127, Aug. 1985.

[25] Vucctic, B., and EM, J., “Channel Modeling and Simulation in Satclli{c Mobile CommunicationSystems,” IEEE Journal on Selected Areas Itl Conir)zutzicati<,tzs, Vol. 10, No. 8, Oct. 1992.

[26] Penwardcn, K., “Propagation Model for LEO Satellites,” Uh’S1 ’96, A1ational Academy of Sciences andEng. Meeting, Boulder, CO., Jan. 9, 1996.

255