acts lluvia

14
ACTS Propagation Experiment: Attenuation Distribution Observations and Prediction Model Comparisons ROBERT K. CRANE, FELLOW, IEEE, AND ASOKA W. DISSANAYAKE Empirical cumulative distribution functions for satellite- to-ground path attenuation relative to clear-sky values were compiled for 22 path years of data collected by the ACTS propagation experiment. These statistics are for two frequencies, 20.2 and 27.5 GHz, with elevation angles ranging 8–52 , latitudes ranging 28–65 , and five different rain-rate climate regions. The attenuation estimation accuracy was better than 0.3 dB. The availability of the equipment for making measurements was higher than 90% for 18 of 22 path years of observation. The empirical distribution functions were compared with predicted cumulative distribution functions generated by four different attenuation-prediction models: the model recommended by the radiocommunications sector of the International Telecommunications Union (ITU-R); the Dissanayake, Allnutt, and Haidara (DAH) rain-attenuation model; and the Crane–Global and Crane–Two Component models when combined with three different rain-rate distribution prediction models: the ITU-R model, the Crane–Global model, and the Rice–Holmberg model. On the basis of the expected differences between model predictions and experimental measurements, the only attenuation model that provided acceptable predictions was the DAH model when combined with either the Crane–Global rain-zone model or the Rice–Holmberg rain-rate model. A major problem in interpreting the results of the model-versus-measurement comparison is the unmodeled contribution of water on the surface of the ACTS propagation terminal antenna reflector. Keywords— Attenuation, attenuation measurement, microwave radio propagation meteorological factors, millimeter wave com- munication, millimeter wave radio propagation meteorological factors, rain, statistics satellite communication. I. INTRODUCTION The National Aeronautics and Space Administration (NASA) Advanced Communications Technology Satellite (ACTS) propagation experiment was designed to obtain slant-path beacon attenuation statistics at frequencies of 20.2 and 27.5 GHz [1]. Two years of observations are now available from seven ACTS propagation experiment sites in the United States and Canada. The sites are displayed Manuscript received October 30, 1996; revised January 31, 1997. R. K. Crane is with the University of Oklahoma, School of Meteorology, Norman, OK 73072 USA (e-mail: [email protected]). A. W. Dissanayake is with COMSAT Laboratories, Clarksburg, MD 20871 USA (e-mail: [email protected]). Publisher Item Identifier S 0018-9219(97)04655-0. on the map in Fig. 1. The locations and propagation- path parameters for each site are listed in Table 1. These sites were chosen to sample the rain-attenuation process in six different rain-climate regions. The map shows the climate-zone boundaries for two rain-climate models, that recommended by the radiocommunications sector of the International Telecommunications Union (ITU-R) [2] and the Crane–Global model [3]. Two of the sites are on or near the boundaries of one or both models. For example, the site in New Mexico lies on an ITU-R model boundary and the site in Oklahoma is near a boundary of the ITU-R model and in an area close to several boundaries of the Crane–Global model. Attenuation-prediction errors should be largest when the site is on or near a climate-region boundary. The climate- zone approach to rain-rate distribution prediction assumes that a single distribution can be used everywhere within a climate zone. The actual distributions for locations within a climate zone must vary in response to local changes in climate. The Rice–Holmberg rain-rate distribution model employs local climatological data to predict the distribution for a location [4]. Locally measured rain-rate distributions should provide the best information for making attenuation- distribution predictions. Unfortunately, more than five years of observations are needed to generate a stable distribution estimate [5], [6]. In practice, for applications in commu- nication system design, the required rain-rate data are not available and use must be made of either a climate-zone model or a model based on locally available climatological data. Three rain-climate prediction models were used in generating the attenuation-model predictions for compari- son to the ACTS propagation experiment observations. The attenuation statistics for each path and the comparisons with model predictions are summarized in this paper; the observed rain-rate statistics for each site and comparisons to the rain-rate distribution predictions are presented in a companion paper [6]. The attenuation statistics presented in this paper were obtained from histogram files prepared by the experimenters using the ACTS preprocessing program [1] and stored in the 0018–9219/97$10.00 1997 IEEE PROCEEDINGS OF THE IEEE, VOL. 85, NO. 6, JUNE 1997 879

Upload: diana-chavez

Post on 03-Dec-2015

5 views

Category:

Documents


0 download

DESCRIPTION

Lluvia

TRANSCRIPT

Page 1: Acts Lluvia

ACTS Propagation Experiment:Attenuation Distribution Observations andPrediction Model Comparisons

ROBERT K. CRANE,FELLOW, IEEE, AND ASOKA W. DISSANAYAKE

Empirical cumulative distribution functions for satellite-to-ground path attenuation relative to clear-sky values werecompiled for 22 path years of data collected by the ACTSpropagation experiment. These statistics are for two frequencies,20.2 and 27.5 GHz, with elevation angles ranging 8–52�, latitudesranging 28–65�, and five different rain-rate climate regions.The attenuation estimation accuracy was better than 0.3 dB.The availability of the equipment for making measurements washigher than 90% for 18 of 22 path years of observation.

The empirical distribution functions were compared withpredicted cumulative distribution functions generated by fourdifferent attenuation-prediction models: the model recommendedby the radiocommunications sector of the InternationalTelecommunications Union (ITU-R); the Dissanayake, Allnutt, andHaidara (DAH) rain-attenuation model; and the Crane–Globaland Crane–Two Component models when combined with threedifferent rain-rate distribution prediction models: the ITU-Rmodel, the Crane–Global model, and the Rice–Holmberg model.On the basis of the expected differences between model predictionsand experimental measurements, the only attenuation model thatprovided acceptable predictions was the DAH model whencombined with either the Crane–Global rain-zone model or theRice–Holmberg rain-rate model. A major problem in interpretingthe results of the model-versus-measurement comparison is theunmodeled contribution of water on the surface of the ACTSpropagation terminal antenna reflector.

Keywords—Attenuation, attenuation measurement, microwaveradio propagation meteorological factors, millimeter wave com-munication, millimeter wave radio propagation meteorologicalfactors, rain, statistics satellite communication.

I. INTRODUCTION

The National Aeronautics and Space Administration(NASA) Advanced Communications Technology Satellite(ACTS) propagation experiment was designed to obtainslant-path beacon attenuation statistics at frequencies of20.2 and 27.5 GHz [1]. Two years of observations are nowavailable from seven ACTS propagation experiment sitesin the United States and Canada. The sites are displayed

Manuscript received October 30, 1996; revised January 31, 1997.R. K. Crane is with the University of Oklahoma, School of Meteorology,

Norman, OK 73072 USA (e-mail: [email protected]).A. W. Dissanayake is with COMSAT Laboratories, Clarksburg, MD

20871 USA (e-mail: [email protected]).Publisher Item Identifier S 0018-9219(97)04655-0.

on the map in Fig. 1. The locations and propagation-path parameters for each site are listed in Table 1. Thesesites were chosen to sample the rain-attenuation processin six different rain-climate regions. The map shows theclimate-zone boundaries for two rain-climate models, thatrecommended by the radiocommunications sector of theInternational Telecommunications Union (ITU-R) [2] andthe Crane–Global model [3]. Two of the sites are on ornear the boundaries of one or both models. For example,the site in New Mexico lies on an ITU-R model boundaryand the site in Oklahoma is near a boundary of the ITU-Rmodel and in an area close to several boundaries of theCrane–Global model.

Attenuation-prediction errors should be largest when thesite is on or near a climate-region boundary. The climate-zone approach to rain-rate distribution prediction assumesthat a single distribution can be used everywhere within aclimate zone. The actual distributions for locations withina climate zone must vary in response to local changes inclimate. The Rice–Holmberg rain-rate distribution modelemploys local climatological data to predict the distributionfor a location [4]. Locally measured rain-rate distributionsshould provide the best information for making attenuation-distribution predictions. Unfortunately, more than five yearsof observations are needed to generate a stable distributionestimate [5], [6]. In practice, for applications in commu-nication system design, the required rain-rate data are notavailable and use must be made of either a climate-zonemodel or a model based on locally available climatologicaldata. Three rain-climate prediction models were used ingenerating the attenuation-model predictions for compari-son to the ACTS propagation experiment observations. Theattenuation statistics for each path and the comparisonswith model predictions are summarized in this paper; theobserved rain-rate statistics for each site and comparisonsto the rain-rate distribution predictions are presented in acompanion paper [6].

The attenuation statistics presented in this paper wereobtained from histogram files prepared by the experimentersusing the ACTS preprocessing program [1] and stored in the

0018–9219/97$10.00 1997 IEEE

PROCEEDINGS OF THE IEEE, VOL. 85, NO. 6, JUNE 1997 879

Page 2: Acts Lluvia

Fig. 1. ACTS propagation experiment site locations and the Global and ITU-R rain-zone bound-aries.

Table 1 ACTS Propagation Experiment Site Locations

ACTS propagation experiment archives at the Universityof Texas [7]. Empirical cumulative distribution functions(EDF’s) for attenuation relative to clear-sky conditions(estimated absorption by atmospheric gases was removed)were compiled for each month of observations in the two-year December 1993–November 1995 time period. Themonthly EDF’s were combined to produce annual andworst-month attenuation EDF’s. These distributions arepresented in Tables 3 and 4 in the format recommendedby ITU-R Study Group 3 [8].

Sample attenuation measurements and statistical distri-butions from one of the sites are presented to provide an

introduction to the attenuation measurement problem. Thenthe statistics for all but one of the sites are presented. Theresults of predictions by four different models using threedifferent rain-rate climate models are then considered.

II. A TTENUATION MEASUREMENTS

The beacon and radiometer receiver channels in theACTS propagation terminal are operated to produce asingle estimate of slant-path attenuation for every sec-ond of equipment operation at each frequency [1]. Thereceiver-system calibration procedure requires that each

880 PROCEEDINGS OF THE IEEE, VOL. 85, NO. 6, JUNE 1997

Page 3: Acts Lluvia

Table 2 ACTS Propagation Experiment Site Performance Parameters

Fig. 2. Two-year attenuation EDF’s for 20.2-GHz observations, Norman, OK.

experimenter initially determine the calibration constantsthat ensure that the observations of attenuation have amaximum measurement error of less than 1.0 dB and atypical (root mean square or rms) error of less than 0.3dB. The experimenters then must monitor for changesin receiver-system calibration and adjust the calibrationconstants as necessary. The preprocessing program providesa summary of the calibration errors that result from thecalibration constants used in processing the data. The rmscalibration errors computed for each site for the two-yearperiod for times with clear-sky conditions are listed in Table2. With the exception of the Vancouver, British Columbia,site, the calibration errors were less than or equal to theexpected value. The preprocessing program allows for theintroduction of new or revised calibration constants at anytime and will reprocess all the data using the revisedconstants.

The attenuation observation statistics generated by thepreprocessing program and stored in the histogram filesare available with 1-s or 1-min integration times. For thestudy of attenuation due to rain, the one-minute data aremost useful. Two one-minute attenuation observations areavailable: 1) attenuation relative to free space (AFS) or totalattenuation and 2) attenuation relative to clear sky (ACS)or the attenuation value minus the gaseous absorptionestimate. Three different two-year EDF’s, 1-s AFS, 1-minAFS, and 1-min ACS for 20.2 GHz for the Oklahoma siteare presented in Fig. 2. The 1-min and 1-s integration-time EDF’s for total attenuation are nearly identical forattenuation values less than 5 dB and differ by less than2 dB for attenuation values less than 10 dB. One-minuteaverages are often used to separate signal level changescaused by rain from the shorter-term scintillation producedby atmospheric turbulence that accompanies the rain [9].

CRANE AND DISSANAYAKE: ACTS OBSERVATIONS AND COMPARISONS 881

Page 4: Acts Lluvia

Fig. 3. One-minute average total attenuation and rain-rate observations, May 29, 1994, Norman,OK.

The difference between the 1-s and 1-min integration-timeAFS EDF’s can be attributed to scintillation.

The attenuation histograms were formed using attenu-ation value ranges (bins) of unequal size. The first binspanned the 0.5 to 0.5 dB range. This wide attenua-tion range was selected to include most of the expectedobservations during clear-sky conditions (observations ofgaseous absorption alone) at elevation angles above 30.The remainder of the bins were formed with approximatelylogarithmic attenuation ranges. Times when the receiverwas not functioning were marked “bad” and binned sepa-rately. The initial distribution compilations tally the fractionof time specified attenuation values are exceeded. Therecommended format for the attenuation distribution tablesrequires estimates of the attenuation values for specifiedfractions of time that the attenuation value is exceeded.The attenuation values presented in Tables 3 and 4 wereobtained from the initial distribution compilations by log-arithmic interpolation. The errors that result from thesedistribution compilation and interpolation procedures areless than 2% in dB and are significantly smaller than thestatistical uncertainty to be associated with each empiricaldistribution estimate.

A sample one-minute integration-time time series withhigh attenuation values is presented in Fig. 3. A detailedanalysis of the observations within a minute with an averageattenuation greater than 26 dB for 20.2 GHz at the Norman,OK, site revealed intermittent periods with detected signalsinterspersed with short intervals of receiver noise. A further

investigation of high-data-rate samples (20 samples persecond) also showed signal intermittency within a second.Within the preprocessing program, a loss-of-signal con-dition (detection of receiver noise) within an attenuationevent is assigned a 35-dB attenuation value for each secondwith a loss of signal. Any value of attenuation greater than30 dB is indicative of the loss of signal condition. One-minute average values of 20–30 dB may be in error due tointermittent loss-of-signal conditions. For sites within thecontinental United States, the maximum reliably estimatedattenuation value is 20 dB. For the sites in Alaska andBritish Columbia, the nominal signal levels are lower dueto the location of the site relative to the pointing directionof the ACTS beacon antenna. For these sites, the maximumattenuation value for reliable observations is 12 dB.

Fig. 3 indicates both times with total attenuation higherthan a few dB and times with rain detected in a collocatedrain gauge. The periods with measurable rain should co-incide with times with attenuation due to clouds and rainon the propagation path and attenuation due to water onthe antenna reflector surface [1]. From experiments with awet reflector, the contribution of the wet antenna to thetotal observed attenuation should range 2–5 dB at 27.5GHz and 1–3 dB at 20.2 GHz. The worst-case observationsattributable to a wet antenna alone that have been recordedto date at the Norman, OK, site are presented in Fig. 4. Thisattenuation event was a natural occurrence of condensationand water deposition on the antenna surface during a longperiod with thin, low-level clouds (fog) at the height of the

882 PROCEEDINGS OF THE IEEE, VOL. 85, NO. 6, JUNE 1997

Page 5: Acts Lluvia

Fig. 4. Attenuation event caused by condensed water on the antenna, November 18, 1993, Norman,OK.

antenna, which is sited on top of a 15-story building. Thepath attenuation in the fog was much smaller than the atten-uation produced by the wet antenna. Condensation and/orfog events were separately cataloged for the Oklahoma site.Dew caused by condensation can be eliminated by heatingthe antenna so the surface temperature is above the dewpoint. Water deposition by fog cannot easily be eliminated.

The observations in Fig. 3 show a slower decay inattenuation value after each rain-attenuation event thanthe rise time to maximum attenuation at the beginning ofeach event. The slow decay evident at 1240 universal time(UT) had a time constant of about 15 min. A water-sprayexperiment during clear weather conditions yielded a timeconstant of 20 min ([1, Fig. 11]). These results indicate thatthe attenuation due to rain and clouds alone should be lowerthan the measured values during times when the antenna iswet. The attenuation relative to a clear sky due to rain,clouds, and a wet antenna are exact within the expected 0.3dB measurement error. The EDF’s for total attenuation dueto rain, clouds, oxygen, and water vapor along the path anda wet antenna are also contained in the histogram files butwill not be considered further in this paper.

Wet snow on the antenna can also cause significantattenuation. An example is shown in Fig. 5. In this case, wetsnow began to accumulate on the antenna reflector at 10 hUT. By 18 h UT, the 20-GHz beacon level dropped to27dB. The snow was then removed by hand and the beaconlevels returned to clear-sky values. In this figure only thebeacon signal levels are plotted. As reference, the beacon

levels for the prior day are also plotted. The attenuation isapproximately the difference between the beacon levels forthe two days. The peak attenuation values due to the wetsnow were 24 dB at 20.2 GHz and 20 dB at 27.5 GHz. At-tenuation by wet snow on the antenna is important but canbe circumvented by the use of heaters. The melting snowwill produce attenuation because the melt water will stillcollect on the antenna reflector surface. For the ACTS prop-agation experiment, occurrences of wet snow were removedfrom the data prior to compiling the histograms. Each ex-perimenter could exclude “bad data” using the log files thatcontrol preprocessing. The bad data are maintained in thearchive records but are not included in the histogram files.

The detection of wet snow events may be difficult if anindependent record of times of snow is not maintained.Sufficient data were recorded to determine periods withattenuation when the ambient temperature was near freez-ing. Wet snow events are characterized by time intervalswhen the ratio of the recorded attenuation at 27.5 GHz tothe attenuation at 20.2 GHz is significantly lower than theratio of about 1.6 expected for rain. The rapid fluctuationsin attenuation (scintillation) resulting in the high within-a-minute standard deviation values characteristic of periodswith rain or clouds (as shown in [1, Fig. 4]) are alsomissing.

III. A TTENUATION STATISTICS

The attenuation statistics presented in Tables 3 and 4are for all attenuation events recorded as “good data” that

CRANE AND DISSANAYAKE: ACTS OBSERVATIONS AND COMPARISONS 883

Page 6: Acts Lluvia

Fig. 5. Beacon signal levels during a snow event, January 22, 1995, Norman, OK.

produced an increase in attenuation over the value expectedfor gaseous absorption. The data used to generate the tablesare attenuation relative to clear sky or total attenuationcorrected for gaseous absorption. The attenuation eventscould be caused by:

1) rain along the path;

2) nonraining clouds on the path;

3) water (rain, fog, or dew) on the antenna feed windowor on the antenna reflector surface;

4) wet snow on the antenna reflector surface or on theantenna feed window;

5) antenna-pointing error due to wind stress;

6) rapid variations in beacon output power that couldnot be tracked by the algorithm used to establish thebeacon reference levels;

7) any other unmodeled change in receiver performancethat would affect the beacon receiver output values.

The experimenters were instructed to mark as “bad data”any attenuation event that could have been caused by wetsnow or by receiver or satellite malfunction, i.e., items4)–7). The inclusion of condensation events as “good data”was at the discretion of the experimenter. For a comparisonbetween statistics for attenuation produced by rain andmodel predictions, only the attenuation events caused byrain and clouds should be included in the EDF’s. Unfor-tunately, during periods with rain at an experiment site,the antenna reflector will get wet and the resulting EDF’swill include wet antenna effects. For model development

and verification, the occurrences of attenuation by a wetantenna must be included in the model. Wet snow andfog/condensation events should be removed from the datain the histogram files because both are dominated by theattenuation produced by water on the antenna and are notaffected by attenuation along the path from the antenna tothe satellite.

The EDF’s in Tables 3 and 4 were compiled from thehistograms as prepared by the experimenters. They havebeen subsequently edited to remove observations beyondthe dynamic range of the receiver system. The data fromthe British Columbia site were not included in the tablesbecause of high calibration errors.

Data from fog/condensation events that did not coincidewith rain were removed from the histograms for the Okla-homa data. Figs. 6 and 7 display the EDF’s for rain (andclouds and wet antenna) and for fog/condensation (wet an-tenna) events alone. The contributions of fog/condensationevents are more than an order of magnitude smaller thanthe contributions of rain in the 2–5 dB range. The expectedstatistical uncertainty in the annual EDF (at the 0.1 signif-icance level, i.e., the parent distribution matches the EDF90% of the time for such a comparison) is indicated bythe separation between the upper and lower model boundcurves. The uncertainty is much larger than the possibleerror in the EDF caused by the inclusion or exclusion offog/condensation events.

Three of the four observed EDF’s lie within the expectedbounds for the Crane–Two Component attenuation modelwhen used with the Crane–Global rain-rate model [8]. At

884 PROCEEDINGS OF THE IEEE, VOL. 85, NO. 6, JUNE 1997

Page 7: Acts Lluvia

Table 3 Annual EDF Attenuation (in dB) Exceeded for Specified Percentage of a Year

Table 4 Worst-Month EDF Attenuation (in dB) Exceeded for Specified Percentage of a Month

attenuation values less than 4 dB, three of the four observedattenuation EDF’s exceeded the predicted values by morethan the expected statistical uncertainty (exceeded the upperbound). At these attenuation levels, clouds of liquid waterare a probable cause. The Crane models are based onthe physics of attenuation by rain and predicted rain-rateprobability distributions [10]. Clouds in the absence ofrain were not considered in the modeling. Clouds alonecan produce attenuation events with attenuation values ashigh as 6 dB at 27.5 GHz at low elevation angles. At thehigher attenuation levels produced by rain on the path, themeasured attenuation values may be as much as 3 dB toohigh at 20.2 GHz and 5 dB too high at 27.5 GHz when therain is over the site. The effect of a wet antenna reflector is

to produce attenuation values that are higher than predictedover the entire range of attenuation values. The effect ofclouds without rain is to increase the observed attenuationvalues relative to the predicted values at attenuation levelsbelow 2–4 dB.

The availability of the receivers for attenuation obser-vations is also important to the statistical quality of theEDF’s. The goal for the ACTS propagation experiment isan availability of 90% of the time or higher and statisticalindependence between times when the receivers fail andtimes with rain. The site availabilities for the first two yearsof the propagation experiment are listed in Table 2. Exceptfor one year of data at each of two sites, the 90% goalwas met. Bad data times were counted as times when the

CRANE AND DISSANAYAKE: ACTS OBSERVATIONS AND COMPARISONS 885

Page 8: Acts Lluvia

Fig. 6. Measured and modeled attenuation EDF’s, 20.2 GHz, Norman, OK.

Fig. 7. Measured and modeled attenuation EDF’s, 27.5 GHz, Norman, OK.

receiver was not available for measurement. EDF’s weregenerated by either including or not including the bad datain counting the times when the receiver was not available

for measurement. The differences between the EDF’s weresmall and could be neglected. If the availability figures for ayear of data are the same for the two frequencies, the EDF’s

886 PROCEEDINGS OF THE IEEE, VOL. 85, NO. 6, JUNE 1997

Page 9: Acts Lluvia

Fig. 8. ACTS propagation experiment annual EDF’s, 20.2 GHz, start date in 1993.

were generated using the normal histogram compilationmode and the EDF’s can be used for frequency scalingstudies [1].

IV. COMPARISON TO ATTENUATION PREDICTIONS

The annual EDF’s for the first two years of the ACTSpropagation experiment are presented in Figs. 8–11. As aguide to their interpretation, Fig. 12 presents the expectedcumulative distribution functions (CDF’s) for each sitefor 20.2 GHz calculated using the Dissanayake, Allnutt,and Haidara (DAH) rain-attenuation model [11] and theRice–Holmberg rain-rate distribution [4]. The DAH modelwas prepared by regression analyses using the data storedin the ITU-R data base [8]. It summarizes all the data in thedata base. The Rice–Holmberg rain-rate distribution modelwas constructed from hourly precipitation data and exces-sive five-minute precipitation data collected in the UnitedStates. The DAH model predictions display a smooth set ofcurves with Tampa, FL, having the highest probability for agiven value of attenuation and Las Cruces, NM, having thelowest probability. The shapes of the distributions proceedin an orderly manner from one site to the next. The modeltakes into account rain climate and the path parameters,such as elevation angle and polarization. Except for thedetails, each of the prediction models considered in thepaper produces a similar pattern of CDF’s.

The Rice–Holmberg model [4] uses the mean annualrain accumulation, the average number of thunderstormdays, and the maximum monthly accumulation in a 30-year

period to determine the two parameters needed to calculatethe probability distribution for rain rate. The two parametersfor each site are listed in Table 2. The DAH model usesonly the rain rate to be exceeded for 0.01% of a year in thecalculation of an expected attenuation CDF. The shape ofthe distribution and the variation of shape with latitude areobtained from the regression analysis using data from over120 path years of observations from different locations,frequencies, rain-climate regions, and elevation angles. Thepredictions in Fig. 12 present a picture of the results tobe expected based on more than 30 years of attenuationmeasurements.

A comparison between the ACTS propagation experi-ment observations in Figs. 8 and 10 with the predictionsin Fig. 12 shows that the measurements from several sitesdo not agree with the predictions. On the basis of dis-tribution shape, Norman, OK, and Reston, VA, produceEDF’s similar to the predicted CDF’s. The distributionsfor Norman and Reston are nearly identical for attenuationvalues greater than 3 dB. The Crane–Global rain-climatemodel puts these two sites in the same rain-climate zoneand, because of the similarity in path parameters, thedistributions should be the same. The EDF’s consistentlyproducing the lowest attenuation values for a fixed proba-bility level are from Greeley, CO, while the model predictsLas Cruces, NM.

To quantify model performance and to look for reasonsfor the apparent inconsistencies between observations andmodel predictions, the performance of each prediction

CRANE AND DISSANAYAKE: ACTS OBSERVATIONS AND COMPARISONS 887

Page 10: Acts Lluvia

Fig. 9. ACTS propagation experiment annual EDF’s, 27.5 GHz, start date in 1993.

Fig. 10. ACTS propagation experiment annual EDF’s, 20.2 GHz, start date in 1994.

model is measured by the rms deviation (RMSD) betweenthe predicted attenuation value and the observed attenuation

value at fixed probability levels [12]. Statistically, theexpected year-to-year deviations in the annual EDF’s from

888 PROCEEDINGS OF THE IEEE, VOL. 85, NO. 6, JUNE 1997

Page 11: Acts Lluvia

Fig. 11. ACTS propagation experiment annual EDF’s, 27.5 GHz, start date in 1994.

Fig. 12. Predicted annual attenuation probability distributions, DAH model, 20.2 GHz.

their underlying CDF is 0.23 (RMSD value computed fromthe average and variance of the natural logarithm of the

ratio of measured to modeled attenuation [each in dB] atseveral values of probability) [5], [13]. The expected RMSD

CRANE AND DISSANAYAKE: ACTS OBSERVATIONS AND COMPARISONS 889

Page 12: Acts Lluvia

Table 5 Composite RMSD Values for the Different Models,20.2-GHz Attenuation Data

value for the deviations between a model prediction andmeasurements for a single year of measurements is 0.29 (or34% in dB). Based on a log-normal model for the expecteddeviations, the RMSD values between measurements andpredictions should be less than 0.40 for 90% of the pathyears of data collected in the ACTS propagation experiment(the bounds displayed in Figs. 6 and 7).

By using the log-normal distribution assumption andcombining the data from all the path years of observations,a single composite RMSD value can be computed fora model. This reduces the measure of performance ofa model to a single number for each model. Assumingthat the frequency scaling for each model is perfect, theobservations for the same year and path but differentfrequencies should be highly correlated, yielding similarRMSD values. On the basis of a simple hypothesis test at a0.1 significance level, any model combination that producesa composite RMSD value greater than 0.33 can be rejectedas not consistent with the ACTS observations.

Four models for the calculation of an attenuation CDFgiven a calculation of the rain-rate distribution and threerain-rate distribution models were compared with the EDF’sfrom 11 path years of ACTS propagation experiment ob-servations at each frequency. The results are given in Table5. The RMSD values for observations at 20.2 GHz showthat the DAH model is consistent with the measurements ifeither the Crane–Global or Rice–Holmberg rain-rate modelsare used. Agreement is best when the Crane–Global rain-rate model is used. Similar results were obtained at 27.5GHz. For this frequency, the ITU-R attenuation modelis acceptable when used with the Crane–Global rain-ratemodel. The Crane attenuation models could be rejected onthe basis of this hypothesis test.

The composite average differences between measure-ments and model predictions can be calculated from theaverage of the natural logarithm of the ratio of measuredto modeled attenuation (each in dB) at several values ofprobability. The composite average differences are listedin Table 6. A second hypothesis for consistency betweenmodel predictions and measurements can be made usingthe values in this table. A model is consistent with theobservations at a 0.1 significance level if the magnitude ofthe average difference is less than 0.07. With this additionaltest, only the DAH model with the Rice–Holmberg rain-rate

Table 6 Composite Average Difference Values for the DifferentModels, 20.2-GHz Attenuation Data

Table 7 Number of Favorable Comparisons Between ModelCDF’s and Measured EDF’s, 20.2-GHz Attenuation Data

model is consistent with the 20.2-GHz measurements, andthe DAH and ITU-R models are consistent at 27.5 GHzonly if the Crane–Global rain-rate model is used.

Table 7 presents the results of hypothesis testing for eachpath year of observations. A value of 11 indicates thatthe model is consistent with each set of data at a singlefrequency. Because the number of observations (degrees offreedom) in each test is much smaller than for the testssummarized in Tables 5 and 6, the test is less powerful.

The DAH [11] and ITU-R [14] attenuation models aresimilar in design and differ only in the complexity of themodel used in the regression analyses. They both representdifferent ways to summarize the data in the ITU-R databank [8] and use those data to predict the CDF’s for differ-ent locations and path parameters. The models first generatean attenuation prediction at a 0.01% probability level, thenextrapolate that prediction to other probability levels usinguniversal (but different) curves. The two Crane models, theTwo Component [10] and Global [15], are based on a differ-ent design philosophy. They depend on 1) a model for theentire rain-rate distribution, not just the expected rain rate at0.01% of a year, 2) different models for the spatial structureof rain, and 3) the same model for the dependence ofspecific attenuation on rain rate as a function of frequencyas the DAH and ITU-R models. In contrast to the lattertwo models, the parameters in the physical Crane modelswere obtained from analyses of rain-gauge and weather-radar observations but without any reference to attenuationmeasurements. The physical models have been tested usingthe attenuation data in the data bases but the parametershave not been adjusted to fit the data in the data bases.

890 PROCEEDINGS OF THE IEEE, VOL. 85, NO. 6, JUNE 1997

Page 13: Acts Lluvia

The rain-rate models also differ in design. TheRice–Holmberg model uses locally available climatologicaldata to estimate the model parameters [4]. The originaldevelopment of the model was based on five-minuteaverage extreme rain-rate observations and hourly rain-rateEDF’s. They did not have one-minute average rain-rateEDF’s for statistical analysis but engineered a plausibledistribution shape from the data then available. Theresult was a smooth distribution between 0.001% ofa year and percentages above 0.1%. The extrapolationacross the 0.01% probability range has been notedto produce prediction errors at the probability valueneeded for the regression model predictions [15]. TheCrane–Global rain-zone model was developed to identifyrain-climate regions where sparsely available one-minuteaverage rain-rate distributions could be combined toproduce a single distribution estimate. The distributionsin the model are the median distributions for all theobservations in a climate region that were available priorto 1985. The ITU-R climate-region model differs from theCrane–Global model in the procedure used to establishclimate-region boundaries. The major difference betweenthe Rice–Holmberg model and the other two models is inthe use of local climatological data.

The RMSD values in Table 5 and the composite averagevalues in Table 6 show that with the regression modelsfor attenuation prediction, using the Crane–Global modelis better statistically than using the Rice–Holmberg model,and both are better than using the ITU-R model. Inpart, the success of the Crane–Global rain-rate modelarises from using measured values at 0.01% of a year.Using either the Crane–Global rain-climate model or theRice–Holmberg model, the DAH attenuation-predictionmodel performed better than the other three attenuationmodels.

The large average differences between the measuredEDF’s and Crane attenuation-model CDF’s provide thereason for the large RMSD values. A review of Figs. 6and 7 suggests that the average differences are due tothe unmodeled effects of clouds and wet-antenna reflectorsurfaces. The regression procedures include cloud effectsbecause they were not removed from the data prior toperforming the regression analyses. Wet-antenna effectswere included only to the extent that the problems fordifferent antenna designs are similar to the problems forthe ACTS propagation experiment antennas [1]. For theACTS experiment antenna design, the losses producedby water on the antenna reflector surface are larger thanthose reported for any of the other Ka-band attenuation-measurement programs. The average prediction errors re-ported in Table 6 for the DAH and ITU-R attenuation-prediction models are puzzling because the predictions aretoo high (a negative average number results from predictedattenuation values that are larger than the measured valuesat the same probability value). If the added attenuationdue to a wet antenna is larger than for prior experiments,why are the predicted attenuation values larger than wereobserved?

V. CONCLUSION

The four attenuation-prediction models coupled withthree different rain-rate prediction models did not performequally well in predicting the attenuation distributionscollected by the ACTS propagation experiment. Of themodels used, the Crane–Global rain-climate model whencombined with the DAH attenuation-prediction modelwas best. The physically based models did not considernonraining cloud attenuation or the effects of a wet antenna.They predicted attenuation values that were too low. Someof the modeling prediction errors can be attributed to thewet-antenna problem. The solution to this problem is tobuild a new model for attenuation by a wet antenna. Betteryet, for communication system design, the solution is animproved antenna design.

REFERENCES

[1] R. K. Crane, X. Wang, D. B. Westenhaver, and W. J. Vo-gel, “ACTS propagation experiment: Experiment design, cal-ibration, and data preparation and archival,” this issue, pp.863–878.

[2] “Characteristics of precipitation for propagation modeling,”Recommendation PN.837-1,ITU-R Recommendations,PN Se-ries, ITU, Geneva, 1994.

[3] R. K. Crane, “Evaluation of global model and CCIR modelsfor estimation of rain rate statistics,”Radio Sci., vol. 20, no. 3,pp. 865–879, 1985.

[4] P. L. Rice and N. R. Holmberg, “Cumulative time statisticsof surface-point rainfall rates,”IEEE Trans. Commun.,vol.COM-21, pp. 1131–1136, Oct. 1973.

[5] R. K. Crane, “Estimating risk for earth-satellite attenuationprediction,” Proc. IEEE, vol. 81, pp. 905–912, June 1993.

[6] R. K. Crane and P. Robinson, “ACTS propagation experi-ment: Rain-rate distribution observations and prediction modelcomparisons,” this issue, pp. 946–958.

[7] W. J. Vogel, ACTS Propagation Experiment Data Archives,Electrical Engineering Research Laboratory, University ofTexas, Austin, 1996.

[8] “Acquisition, presentation and analysis of data in studies oftropospheric propagation,” Recommendation ITU-R PN.311-7,ITU-R Recommendations,PN Series, ITU, Geneva, 1994.

[9] C. E. Mayer, B. E. Jaeger, R. K. Crane, and X. Wang, “Ka-bandscintillations: Measurements and model predictions,” this issue,pp. 936–945.

[10] R. K. Crane and H.-C. Shieh, “A two-component rain modelfor the prediction of site diversity improvement performance,”Radio Sci.,vol. 24, no. 6, pp. 641–655, 1989.

[11] A. W. Dissanayake, J. E. Allnutt, and F. Haidara, “A predictionmodel that combines rain attenuation and other impairmentsalong earth-space paths,”IEEE Trans. Antennas Propagat.,submitted for publication.

[12] R. K. Crane, “Comparative evaluation of several rain attenua-tion prediction models,”Radio Sci.,vol. 20, no. 4, pp. 843–863,1985.

[13] , Electromagnetic Wave Propagation Through Rain.NewYork: Wiley, Interscience, p. 245, 1996.

[14] “Propagation data and prediction methods required for earth-space telecommunication systems,” Recommendation ITU-RP.618–4, ITU-R Recommendations,P-Series Fascicle, ITU,Geneva, 1995.

[15] R. K. Crane, “Prediction of attenuation by rain,”IEEE Trans.Commun.,vol. COM-28, pp. 1717–1733, Sept. 1980.

Robert K. Crane (Fellow, IEEE), for a photo and biography, see thisissue, p. 877.

CRANE AND DISSANAYAKE: ACTS OBSERVATIONS AND COMPARISONS 891

Page 14: Acts Lluvia

Asoka W. Dissanayake received the B.Sc.degree in electronic engineering from theUniversity of Sri Lanka, Moratuwa, in 1972,the M.Sc. degree in digital electronics fromLoughborough University, U.K., in 1975,and the Ph.D. degree from the University ofBradford, U.K., in 1978.

During 1981–1987, he was a SystemsEngineer at the European Space and TechnologyCenter in Noordwijk, The Netherlands. In 1989,he joined INTELSAT in Washington, D.C.,

where he spent two years as a radio-frequency engineer. In 1991, hejoined COMSAT Laboratories, where he is currently a Senior Scientist inthe Antennas and Propagation Department. His areas of interest includeradiowave propagation and wireless communication.

892 PROCEEDINGS OF THE IEEE, VOL. 85, NO. 6, JUNE 1997