Sunteți pe pagina 1din 62

Naval Research Laboratory

Washington, DC 20375-5320

NRLIMR/5320--93-7181

Comparison of RADARC High-Frequency Radar Performance Prediction Model and ROTHR Amchitka Data
-

BENAmwI T. RooT JA~M M.-ADMiucK


Advanced Radar Systems Branch

RadarDivision

July 16, 1993

-,

93.

6;"

Approved for public release; distribution unlimited.

93-16508-

existing date oowce. insi*~Ntruct~ion. siina ncud'I the bins forw~w Public seatsu burden for fte cullection of information is esttn@Wd to averege I how Pa repr. m p a of tive Or any 06w I needed, and cemnpletiU and rieviewing the colection of informetion. Send cower,,u, regardia this burden 0mastem oadwko -n mdainsurq the doam bdn. to Wastngton Headquortera Services. Directorate for Inforimation Oporauens &id Rownr. 1215 Jefferson euomlomfor rdueiq bus c~latin b~rm~~n.Inddin 6 20103. Washinton, DC 40704-0188). Reduction Progect Peperwork Davis Highway. Suft 1204. Ariegton. VA 22202-4302, mid to the office of Musegemtent midlBudge. 1. AGENCY USE ONILY ILeave Blank)

12. REPORT DATE

3.REPORT TYPE AND DATES COVERED

I July 16, 19931


4. TITLE AND SUIBTITLE 5. FUNDING NUMBERS

Comparison of RADARC High-Frequency Radar Performnance Prediction Model and ROTHR Amchitka Data
6. AUTHOR(S)

PE - 204577N WU - DN 291-215 TA - 148-NRL-VARI-002

Benjamin T. Root and James M. Headrick


7. PERFORMING ORGANIZATION NAMEIS) AND ADORESSIES) B. PERFORMING ORGANIZATION REPORT NUMBER

Naval Research Laboratory Washington, DC 20375-5320

NRL/MR15320-93-7181

9. SPONSORINGJMONITORING AGENCY NAMEIS) AND ADDRESSIES)

10. SPONSORING/MONITORING AGENCY REPORT NUMBER

Space and Naval Warfare Systems Command Washington, DC 20363-5100


11. SUPPLEMENTARY NOTES

12a. DISTRIBUT1O*J/AVAILABIL17Y STATEMENT

12b. DISTRIBUTION CODE

Approved for public release; distribution unlimited.

13. ABSTRACT (Mexiwom 200 words)

Data collected with ANrrPS-71 radar has been used to validate the RADARC model. For the first ten days of January , April, and July, noon and midnight, the vertical ionograms, oblique backscatter soundings, radar ground backacatter amplitudes and noise levels have been compared with Model predictions. The differences between the Model and median ionospheric soundings are small enough to have negligible effect on predictions. The coverage predicted for oblique backscatter soundings agrees well with observed median data except for sporadic E effects which were under-estimated, Medians of observed ground backscatter agree fairly well with predictions. The actual noise that controlled radar performance was the most significant deviation from the model; all nighttime and January day noise was higher than that predicted; and this was due to "multiplicative" noise.

14. SUBJECT TERMS

15. NUMBER OF PAGES

Ionospheric Models 0TH RADAR Skywave Transmission


17. SECURITY CLASSIFICATION OF REPORT 18. SECURITY CLASSIFICATION OF THIS PAGE 119. SECURITY CLASSIFICATION OF ABSTRACT

62
CODE 16. PR~ICE

20. LIMITATION OF ABSTRACT

UNCLASSIFIED
NIN 7640-01-2604600

UNCLASSIFIED

UNCLASSIFIED

UL
standard Farm 299 Iftev. 2461 Prescr,bed by ANSI Std 238-18 296.10?

Contents

1. 2.

Introduction ..................................... Comparison of RADARC Prediction of Quasi-Vertical Ionograms (QVIs) with ROTHR Data ........................ a. Description of Analysis ............................... b. Description of Software ............................... c. Discussion ......................................

. 1 5 5 6 6 10 10 I 11 13 13 17 20 25 28 29 37 45
Ioession For

3.

Comparison of RADARC Prediction of Widesweep Backscatter Ionograms (WSBIs) with ROTHR Data .............. a. Description of Analysis .............................. b. Description of Software ............................... c. Discussion .....................................

4.

Comparison of RADARC Prediction of Clutter and Noise with ROTHR Data ................................... a. Description of Analysis .............................. b. Description of Software ............................... c. Discussion .....................................

5. 6.

Conclusions ...................................... References .......................................

Appendix A. Figures 3.1 to 3.6 ................................ Appendix B. Figures 4.1 to 4.6 ................................ Appendix C. Figures 5.1 to 6.6 ................................

D'S!: ," ,.

Dt
m ...

*... ..

"D~ i.

COMPARISON OF RADARC HIGH-FREQUENCY RADAR PERFORMANCE PREDICTION MODEL AND ROTHR AMCHITKA DATA

1.

Introduction

with both deterministic and statistical elements to solve the radar equation for the conditions that are expected to prevail in a particular location and time. This requires a correct understanding of both the radar system and the environment. We must know the various gains and losses in the system, which result from both hardware and signal processing. We must also predict the likely path of the radar beam through the ionosphere, the various path losses, the target reflectivities, and the environmental noise. Such radar performance modeling is useful not only for efficient operation of the system, but also in the design of such systems.
by a monthly median for each hour of the day. This description includes the number of electron

RADARC is NRL's high-frequency radar performance prediction program. It uses a mode!

Many of the factors entering the model are statistical. For example, the ionosphere is described

density layers, the shape, height, thickness, and density of these layers, and resulting quantities

such as the critical frequencies. In the case of sporadic E, we do not have a basically geometric model with statistically varying parameters, but use a wholly statistical description. Since the radar beam, even to refraction from a completely different layer than predicted.

description is by monthly median, the actual ionosphere on any particular day may depart significantly from the prediction, and this can lead to considerable differences in the path of the

"The environmental noise is also described by a monthly median for a given time of the day. Whether a radar is internally or externally noise limited is one of the important issues in radar performance. Not only is the noise model statistical, it is not appropriate for general application. For instance, we assume isotropic noise when in fact the noise can be direction-dependent. RADARC uses the CCIR noise curves that were first developed from an empirical database in the early 60's and have recently been revised [3,41.

It may be thought that we have better understanding and control of the radar system itself, but even here there are many unknowns. In particular the large size of a high-frequency radar (a kilometer or more in length) means that it is difficult to measure its true gain, and this gain is complicated by the environment in a way that does not occur for smaller radars. Even with a knowledge of the environment, we must resort to numerical software to estimate the gain, and these calculations are only approximate. The issue of gain can be important because with over-thehorizon radars it is often the lowest rays which reach distant targets, and the gain falls off rapidly as we graze the horizon. Furthermore, an HF radar is so large and complicated that it may be difficult to calibrate it with complete assurance, and we may underestimate or overestimate the various internal gains and losses. And it is more difficult to shield such a radar from changes in the external environment.
Manuscript approved May 13. 1993.

* It is the current version of the RADAR performance program described in NRL Memorandum Report 2226 [1). The ionospheric model used is given in NRL Report 8321 (2].

One way of correcting and improving radar prediction programs such as RADARC is to test and validate them against new radar data. Recently, a new ROTHR (Relocatable Over-the-Horizon Radar) has been installed on Amchitka in the Aleutian Islands. A large quantity of radar data has been collected and sent to several laboratories, including NRL, for analysis. This data covers all hours of the day for about ten days in each of three months: January, April, and July of 1991. These three months give a sample of three different seasons. In this report we have processed data for each of the three months. In each month two different hours have been analyzed: 0 UT, which gives an example of the daytime (noon in Amchitka), and 12 UT, which gives an example of the nighttime (midnight in Amchitka).

00 UT (noon in Alaska) and 12 UT (midnight) * January, April, and July, 1991 First 10 days of month Azimuthal sector 0' to 10" (from boresight = 270' CWN) 0 500 - 1500 nm: mostly sea with many scattered islands * 1500 - 2000 nm: half land Table 1. ROTHR Amchitka data set analyzed in this report.

Note that we have simplified the analysis by selecting from an azimuthal sector extending from 0' to 10* with respect to boresight. (Boresight is at 270' clockwise from north.) This keeps the geography, and hence the surface reflectivity, approximately constant over azimuth at a given range. This azimuthal sector is shown shaded on the following map.

/USSI,

50" N

, RADAR
1 5Ails

ERAG EA

40" N
150 miles

miles

30" N4

130"

"2000 railes70 140* E 150" E 160' E 10

180" E

Figure 1. Azimuthal sector examined in this report

We have examined three types of data. The first consists of the quasi-vertical ionograms (QVIs), which give a description of the ionosphere overhead at the radar site. If we cannot accurately predict the local ionosphere, then we cannot hope to tackle the more complicated backscatter problem involving propagation effects as well as ionospheric gradients. So it is natural to first check how well the predicted and measured QVIs match. Then we consider the widesweep backscatter ,onograms (WSBIs), which provide a synoptic view of backscattered power vs. frequency and range, but without doppler. Here, propagation effects are important, and we need to model correctly the gradients in the ionosphere over range. This is perhaps the ideal data type with which to test the RADARC model, because of its broad view of radar performance over range and frequency. We can detect patterns that might not be apparent with a narrow look at the backscatter at one operating frequency, even though the latter is far more sensitive and has doppler. We will compare the measured WSBIs with the RADARC predicted clutter-to-noise ratio map, as the two are closely related. However, one disadvantage of the WSBI is its lack of doppler, which may make it difficult to distinguish clutter from noise. Therefore, we finish our analysis with a detailed look at the actual range-doppler-amplitude data that the radar uses to find targets. The new environment at Amchitka brings new problems. Of particular concern is the issue of spread-doppler clutter. The radar's propagation path crosses regions of the ionosphere near the equator containing irregularities in the electron density after sunset. These irregularities lead to the phenomenon of spread-doppler clutter. the range-folded equatorial Bragg reflections from land and sea near 0 Hz are spread out on the doppler axis and may mask targets. The problem is especially noticeable at night, because of reduced D-layer absorption. It leads to an effective increase in the noise floor above the atmospheric noise predicted by CCIR. Reliable radar performance requires that we correctly account for this problem. Also, the new environment may lead to variations in system performance from what is expected from previous locations. In particular, the gain may be different, but so may other parts of the system.

Finally, a program as complicated as RADARC may have had unnoticed defects or deficiencies that have gone uncorrected. It has never been validated against a database as extensive as what we have recently collected in Amchitka. Analyzing this data should lead to a better understanding of the radar, not only for this particular environment, but for all environments. The overall objective of the analysis in this report is summarized in the table below. " Compare ROTHR data from Amchitka with RADARC high-frequency radar performance prediction model. "* Compare measured "clutter" and "noise" (defined on p.13) with predicted clutter and noise. "* Make corrections and improvements to model as needed. Table 2. Objective of the analysis in this report

2.

Comparison of RADARC Prediction of Quasi-Vertical Ionograms (QVIs) with ROTHR Data Description of Analysis

2a.

The QVIs show time delay vs. frequency for a vertically transmitted pulse. The "quasi" in quasi-vertical comes from the small separation (-20 miles) between transmitter and receiver. The vertical ionogram is one of the oldest methods for probing the ionosphere. At a frequency below the critical frequency of a given ionospheric layer, the pulse is reflected before reaching the height of maximum ionization, which results in some finite delay for the received pulse. As the transmitted frequency approaches the critical frequency, the time delay increases, becoming theoretically infinite at the critical frequency. Above the critical frequency, the pulse penetrates the layer and continues to the next layer, where this process is repeated, or else it escapes into space. A trace of time delay vs. frequency result:. Actually, there are two closely spaced traces, one for the ordinary ray and another for the extraordinary ray. At frequencies far below critical, the two curves overlap. Near the critical frequency, both curves become nearly vertical, with about 0.5 to 0.8 MHz between them. This method of probing the ionosphere measures integrated electron density along the pulse path. To some extent, the actual electron density distribution can be inferred, especially if an ionospheric model with adjustable parameters is assumed (such as parabolic layers). In general, though, it is not possible to probe a given layer above the height of maximum ionization, and the distribution below this height cannot be read directly but must be deduced by a numerically laborious procedure. The ionospheric model in RADARC is the same as that in IONCAP, which in turn traces its database back to the late 1950s when a worldwide survey of the ionosphere was undertaken as part of the IGY (International Geophysical Year). Samples of the ionosphere were taken during solar maximum and minimum activity, and modeled between these years by interpolation. The ionograms in this database were "scaled" (measured) in the following way. The critical frequencies of the various layers were identified by the characteristic vertical slope of the traces. The "height" of the layer was simply taken as the first minimum of the trace below the critical frequency, which actually corresponds to the bottom of the layer. Unfortunately, the frequency of this local minimum was not always recorded, only the height (time delay) itself. Sometimes there is no clear separation between layers (especially with the F1 layer), and the various possible layers are not always present. Also, obscuring sporadic-E can complicate the problem. RADARC predicts a vertical ionogram for each of the control points specified in the model. The separation between ROTHR transmitter and receiver is sufficiently small that we can directly use the RADARC ionogram at the transmitter for comparison with the ROTHR QVI. A QVI is taken about every 12.5 minutes, which gives us about 45 QVIs per hour per month (for the first 10 days). There is considerable variability in the measured QVIs, especially from day to day, and some kind of average must be performed. We take a median QVI by the method described in the next section.

2b.

Description of Software

We have about 45 QVIs per hour per month (first 10 days), which must somehow be reduced to one representative QVI for comparison with the RADARC prediction. Each QVI has about 230 frequency cells vs. 500 time delays. In principle, there is an amplitude in each of these frequencydelay cells, and the amplitude is quantized in 8 steps from noise to some clipping level. We will not be concerned with absolute power levels, but only with finding a representative curve of time delay vs. frequency. So we begin by outlining such a trace on each individual QVI. In principle, such a procedure requires pattern recognition and artificial intelligence, but for the moment we shall be content with the following simple algorithm. At each frequency on the QVI, and starting at 50 km of time delay to avoid ground reflection, we scan upwards in time delay for the maximum in power. This is taken as the height of the first echo at that particular frequency. We record the time delay (or slant height) of that cell When we have finished with all frequencies, we have a trace of time delay vs. frequency for one individual QVI. We repeat this procedure for all 45 or so QVIs of the month at the selected hour. We then take a median over all the resulting traces by finding the median height at each frequency. This finally gives us the one median ionogram which we compare with RADARC. An example of such a median ionogram is Figure 3.1 in Appendix A. These figures are discussed in detail in the next section. The algorithm is summarized in the following box.

" Step 1: On each QVI, at each frequency, starting at 50 km and scanning upwards in time delay, find cell with maximum power. " Step 2: Record slant height of this cell at each frequency. " Step 3: Repeat with all QVIs (about 45) for the selected hour and month. " Step 4: Take median over the resulting traces by finding median height at each frequency. Table 3. Algorithm to extract median quasi-vertical ionogram

It should be noted that near the critical frequency, we will tend to pick up the extraordinary rather than ordinary trace, since the former extends somewhat higher in frequency. Fortunately, the ionograms predicted by RADARC specify the critical frequency of the extraordinary ray.

2c.

Discussion
Table 4 summarizes our analysis of the median quasi-vertical ionograms presented in the

figures of Appendix A. We have compared both the critical frequencies and the layer heights of the
RADARC model with the ROTHR data. The critical frequency is defined as the frequency where the extraordinary ray becomes vertical. The layer height is defined as the bottom of the "valley" for the corresponding layer, assuming that such a local minimum is apparent. This is in accordance with how the database used in RADARC was originally "scaled," as discussed in Section 2a.

Critical Frequencies and Layer Heights


model
JAN 00 12

o hand

o machine
JAN 12 APR 12 JUL 12

APR 00 JUL 00
0

.. __10

-0

FF2
60 6

~....i
V 0

0
. .

...............
. .. .
44

-0

;- 0

JAN 00 600 ~20.

APR 00 JUL 00

JAN 12
............................

APR 12

JUL 12
..

S500 S400

3700

.
0 0 " O

-, 20o .: :....
Tabe .

"V

.. .. .
f equen es n

.... , ........... 4........... 4............


ae egt fqus-etcljngan

ritca

100 oo

11

0
Table 4. Critical frequencies and layer heights of quasi-vertical ionograms

If a local minimum is no- ,pparent, we choose the minimum height where the layer in question seems to start. This is etr..ially relevant for the Fl layer, which often has a monotonic increase of height vs. frequency on the ionogram. Our analysis of the data is twofold: on the one hand we read off the critical frequencies and layer heights from the median estimates generated by computer, as described in Section 2b; but on the ot',': hand, we check this by examining all the individual ionograms by eye and forming an estimate based on patterns that are apparent to us from this visual inspection. The former method is called "machine" (open diamonds) in the table, and the latter method is called "hand" (open circles). Finally, we present the RADARC predictions (called "model") as black dots. The dashed lines identify the ionospheric layers ( E, Fl, or F2). The computer-generated medians ionograms are shown in Appendix A. On these figures, the ROTHR median ionogram is represented by black dots, and the RADARC prediction by little plus signs. The month and hour are shown at the top of the plot. A numerical table of the RADARC prediction is shown in the upper right of the plot. We see the critical frequencies (FC), critical heights (HC), and semi-thicknesses (YM) for the E, Fl, and F2 layers. This model assumes parabolic layers in all cases, so that HC is the height of maximum ionization and YM is the semi-thickness of the parabolic profile of electron density. It should be stressed that HC is not the layer height presented in Table 4. The latter corresponds to the bottom of the layer rather than to the middle of the layer, where the electron density reaches maximum. Finally, the last line the RADARC numerical table, marked by ES in the left column, describes the sporadic-E statistical estimate. Traces were sketched by hand showing what the ionosphere was like on individual days, and these are shown in the inset on the figures of Appendix A. A single representative ionogram was chosen for each day, a piece of translucent paper was overlaid, and the ordinary ray was traced with pencil. The collection of these traces is presented superimposed in the inset. The number labeling each curve is the lay of the month. April day (hour 00 UT, Figure 3.3) was a special case. The individual ionograms showed such variability from day to day that it seemed that all seasons were occurring on some day or another within the month. (There was fortunately much less variability within a day, as one would expect.) The variability is quite apparent in the inset just described, and this explains why the computer-generated median performs poorly, and why we have poor agreement with the model. With such variability, it is doubtful whether any median model is meaningful as a description of the ionosphere. For other months and hours, the comparison is generally satisfactory. (Remember that the ROTHR dam is based on the first ten days of the month. One might reasonably question whether this provides an adequate statistical sample for the month as a whole.) The estimate of the critical frequencies tends to be better than estimates for the layer heights. This is expected, if we remember how the RADARC ionospheric database was originally constructed (see Section 2a). The critical frequencies are relatively unambiguous and easy to read on the ionograms. Furthermore, they are the most important parameter in predicting radar performance, as the radar tries to operate just below the critical frequency. Therefore, the RADARC database is "geared" towards the best possible estimate of critical frequencies. The layer heights, on the other hand, are often more ambiguous to identify on the ionograms, for the reasons given, and this is complicated by the fact that the original IGY analysis did not record the frequencies of the layer heights. Our analysis of the quasi-vertical ionograms is summarized by the remarks in the table below.

- In general, first 10 days of month compare fairly well with model. - April days deviate the most and are the most variable. * Simple median algorithm gives fair results and could be improved.

Critical Frequencies:

"*Model January & July midday F2 critical frequencies are high "*Good April agreement (of F2 criticals) "Note that RADARC model includes fxF2 at high frequency end. "*E critical frequencies a bit high in January. "*FI was observed in April less than half the time - a significant disagreement
with model "*Model midnight F2 critical frequencies compare with hand analysis (after adding 0.7 MHz to foF2 to estimate fxF2). Layer Heights: * Daytime hF2 observed -50 km low in January and April, but -150 km high in July. * Daytime F1 and E layers show fair agreement. * Night shows good agreement. Table 5. Summary of analysis of quasi-vertical ionograms

3.

Comparison of RADARC Prediction of Widesweep Backscatter lonograms (WSBIs) with ROTHR Data Description of Analysis

3a.

The WSBIs show backscattered power over all ranges and frequencies, but without doppler. Of course, the transmitted power is much weaker than for the range-doppler data used to find targets. The WSBI provides an overall view of propagation conditions and helps in selecting the optimum operating frequency. It closely resembles the clutter-to-noise map predicted by RADARC and is therefore the ideal data for comparison with the RADARC model. The lack of doppler, however, can sometimes make it difficult to distinguish clutter form noise, or one type of noise from another.
About 45 WSBIs per hour per month reside in the database (the same as for the QVIs). There are two types of ROTHR WSBI data: one is clipped in amplitude above a certain threshold and the other is clipped below another threshold. The first type is used to avoid regions of weak
backscattered power, and the second to scan the clutter environment and correlate with ionospheric

models in the radar's computer system. We use the second type of data, since we are trying to correlate with RADARC's ionospheric model. The clipping floor for this data can be up to 64 dB below the peak amplitude. The synoptic nature of the WSBI (the fact that it covers all ranges and frequencies) makes it easier to detect patterns in the discrepancy between the RADARC model and the radar performance. For instance, we can tell if there is a consistent bias in the leading edge of some propagation mode, or if the model and the data even agree on the modes of propagation. We can distinguish between sporadic E and normal layer propagation or between propagation from different layers with far greater ease than if we only have power vs. range at one frequency. Just by comparing the outlines of the clutter regions in the WSBIs and the RADARC CNR maps, we can obtain a good first order test of the RADARC model. If there is a large discrepancy in the outlines, we must resolve the major errors before be worry about fine-tuning the actual power levels. The present analysis will in fact concentrate on a qualitative or semi-quantitative comparison of the shape and location of the clutter in range-frequency space. As with the QVI data, we reduce the 45 WSBI to one representative median ionogram for the hour and month for comparison with the clutter-to-noise map predicted by RADARC for that hour and month. The algorithm for performing this reduction is described in the next section.

10

3b.

Description of Software

Each WSBI consists of about 230 frequency cells and 250 time delay (range) cells. The power is quantized into a 128-level scale of 1/2 dB steps extending down from the peak recorded amplitude. The reduction of the 45 WSBIs to one median WSBI is simpler than for the QVIs. There is no attempt at any kind of "pattern recognition" such as the simple algorithm to pick out the dominant trace in the QVI. Instead each range-frequency cell is averaged separately. That is, in each of the range-frequency cells, the algorithm determines the median power over all 45 ionogiams for the selected hour and month. Due to the great number of range-frequency cells, some smoothing of the individual ionograms is done before the median is taken. The ionogram is partitioned into little boxes containing 3 frequency cells x 5 range cells, and a simple numerical average of these 15 cells is taken and recorded as one smoothed cell. The median is then taken from the 45 smoothed ionograms as just described. The algorithm for extracting the median WSBI is summarized in the following table. " Step 1: Each individual ionograxn has 230 frequency cells x 250 range cells. These are first smoothed by taking the average power in a 3 by 5 window. The result is -45 smoothed ionograms for each hour and month. " Step 2: In each range-frequency cell of the smoothed ionograms, take the median power over all ionograms for the selected hour and month. Table 6. Algorithm to extract median widesweep backscatter ionogram

3c.

Discussion

The comparison between the ROTHR median widesweep-backscatter ionograms and the RADARC predicted clutter-to-noise ratio (CNR) maps is shown in the figures of Appendix B. The median WSBI is shown on the bottom left, and the CNR map on the bottom right. On the top are two small plots whose significance will be discussed shortly. On both the WSBI and the CNR map, a greyscale was simulated by altering the size of the little squares. The size (edge) of the square is proportional to dB below peak clutter power. (Perhaps the area of the square should have been made proportional to power for a more realistic visual effect.) As we have said, we cannot expect to correlate exact powers between model and data, but only general patterns. Nevertheless, it is a sufficiently difficult task to predict the general patterns of backscattered power, and good qualitative agreement with the data would be encouraging in itself. It should be remembered that the "noise" levels in the WSBI and in the CNR map are not necessarily the same. The former is based on the ROTHR spectrum monitor measurement during the time the data was collected, and the latter is based on the CCIR prediction. One of our tasks is 11

to find discrepancies in these noise levels, especially as relates to the spread-doppler clutter phenomenon.

The small subplots at the top of the figures are an attempt to emulate the selection of operating frequency by the radar operator. A computer algorithm examines the corresponding plot below. At each range, it searches across frequency for the maximum power level, which should usually be near the leading edge. Since the data is quantized, there may be a number of equal maximum power cells. In this case, the one with the highest frequency is chosen. The power and frequency of this cell are recorded at each range. The power vs. range is shown as the solid curve, and the frequency vs. range as the dashed curve. This algorithm is still in the initial stages of development, and the results presented in the small subplots are for illustration only. It would be inadvisable to draw conclusions from them. The comparison is necessarily qualitative and therefore subjective. The following table presents our opinion, and the corresponding figures can be consulted by the reader. As we have repeatedly emphasized, we do not expect precise agreement between the power levels, but we do hope for a good similarity in the shapes and patterns of the backscattered power. Considering how complex the ionosphere and propagation are, and how difficult it is to predict them, a good shape correlation should indicate a reasonably good model. The major error in the model is in predicted Es coverage, and it errs in having both more and less than the measurements at different times.

DAY (00 UT)


J
Good c defined by F region

NIGHT (12 UT)


Good correlation for skip zone defined by F region Mor Es than predicted

A N A

Good correlation for l-hop skip zone Data does not show I hop - 2 hop division Data has more Es than model <1200 nm: good skip zone correlation.

Good skip zone co rrelaio n

P R
J
U

Good F region skip zone correlation

L
Table 7.

Data does not show F region retuns

Model shows more Es than observed

Comparison between model and data for widesweep-backscaner ionograms

12

4.

Comparison of RADARC Prediction of Clutter and Noise with ROTHR Data Description of Analysis

4a.

In this analysis, we compare the radar performance predicted by the RADARC model with the actual performance of the ROTHR system in Amchitka during 1991. As indicated in the Introduction, we examine the radar data at 0 UT (noon local time) and 12 UT (midnight local time) for the first 10 days of January, April, and July. The radar dam consists of received power for each range-doppler cell. The range and azimuthal resolution is relatively good, but we cannot discriminate in elevation. Our analysis consists primarily of comparing the "clutter" from land and ocean and the "noise" from the environment with that predicted by RADARC. (Both these terms are defined below.) By separating out the absolute powers of clutter and noise, we understand better what is happening than if we just look at the ratio of the two (as one usually does). The present analysis will emphasize this separation of clutter and noise. We further narrow the data by selecting from an azimuthal sector extending from 0" to 10" with respect to boresight. (Boresight is at 270' clockwise from north.) This keeps the geography, and hence the surface reflectivity, approximately constant at a given range. Since the present version of RADARC assumes an average sea surface everywhere, we expect the actual reflectivity sometimes to be a significant source of error in the prediction. By examining a narrow sector, we simplify the analysis of a problem that has many sources of error, reflectivity being only one of them. The sector chosen is the one where the radar was most frequently tasked during the time of the data collection. It can be divided into two ranges intervals: one extending from 500 to 1500 nautical miles from the radar consisting mostly of sea with many scattered islands, and another "extendingfrom 1500 run to 2000 nm consisting half of land and half of sea. See Figure 1. The clutter is here defined to be the maximum power in a small window (+/- 2.0 Hz) around 0 Hz at each range. For the relatively large aircraft waveform repetition rates (WRF) selected in this analysis, the two Bragg peaks are generally not differentiated in the log-amplitude data, and so "clutter" is simply a measure of the backscattered power from sea or land against which a slow target must compete. There are various possible definitions of noise. In the RADARC program, noise is selected to be the maximum of either the ROTHR system noise curves or the galactic and atmospheric noise' from the widely used CCIR prediction tables. In the ROTHR log amp data, noise is taken (at a particular range) as the average (in dB) of the clutter at + and - WRF/4 (in a doppler spectrum where 0 Hz is at the center). If there is no spread-doppler clutter, so that there is a flat noise floor outside of the clutter, then the two definitions of noise should agree closely (at least if the WRF is chosen so that the land and sea clutter is concentrated around 0 Hz). If there is spread-doppler that extends to +/- WRF/4, then this definition of "noise" will be higher than for RADARC. The points at which the noise is measured are taken to be representative points on the doppler axis, at the centers of the unambiguous positive and negative doppler extent. See Figures 2.1 and 2.2 for examples of radar data both with and without spread-doppler clutter.

"RADARC has a spread-doppler clutter model, but this has not yet been range-folded to ambiguous ranges such as Amchitka. Our comparison will indicate the presence of spread-doppler clutter by a large discrepancy between CCIR and measured noise. 13

RADARC predicts (among other things) the clutter-to-noise ratio (CNR). The clutter power is found by adding this ratio to the CCIR noise. The CNR radar equation modeled by RADARC is as follows.

CNR = PWR + PKAVG + GAIN + LAMLOG + C1T + CAREA + OTH - R4TH - LOSS (hop 1) - LOSS (hop 2) - NOISE Table 8. RADARC clutter-to-noise radar equation
PWR GAIN
I.LAMLOG

= =
=

R4TH

= = = = = = =

transmitted power sum of transmit and receive gains 2 / (4t)3 X I /R4 loss LOSS(hop 1) + grotndloss + LOSS(hop 2) CCIR external noise (assumes IHz bandwidth) o0 times illuminated ocean area = (-29dBm / m2 )(area) average/peak power ratio = 0 dB over-the-horizon enhancement = 6 dB coherent integration time

LOSS (hop 1)
LOSS (hop 2) NOISE CAREA PKAVG OTH CiT

= non-deviative absorption + deviative absorption + obscuration (sporadic E) + auroral (where applicable)

Most of these terms are familiar from the standard radar equation. It is assumed that the signal processing in the radar is approximately equivalent to a matched filter. Details of the signal processing will be described below.

The propagation losses are peculiar to ionospheric propagation. There is no auroral loss
,n the present comparison, because we were not looking towards the auroral regions. The noise in Table 8 is the maximum of internal noise or external noise. The internal noise is

based on measured ROTHR noise figures. The external noise is taken here to be the CCIR
external noise only, but could also be set to include the spread-doppler clutter model already

incorporated in RADARC. This spread-doppler model is not yet range-folded to ambiguous ranges and would contribute only when local to the radar. Since the radar is in Alaska, spread-

doppler clutter from the equator would make no contribution for the two-hop propagation assumed in this analysis. RADARC can predict long range propagation to the equator, in which case the spread-doppler clutter would be included, but this is different from range-folding. When significant discrepancies in the comparison are found between predicted and measured noise, we expect to explain this primarily by invoking the spread-doppler phenomenon. The ocean reflectivity o0 (we use a average value of 29 dBm/m 2) is calculated for an estimated average sea-state (not fully-developed, averaged over radial ocean wave velocity, averaged over polarization). We expect that the discrepancy between this assumed average

reflectivity and the true reflectivity in the data will sometimes be a significant source of error in the predicted clutter. Not only may the sea surface vary from the assumed average, but also we will sometimes be looking at land, as is evident from the map in Figure 1. 14

3-.,IV-19U 00:00:49

I dma 201m

3-JRN-IS9J 00:00:52

Igtm

6 dma

3-Jf'991

00257

21 MtI

8 dog

-"

0, C

-21 3-Fmig

21-100 00201:00

2W

-21

21 -IW

-20

-21

0 mchi L

21-30

-20 I d M17Ws
7

17

2 dea sth

00o01025 3-Jle-r maps

17

3-iM-n 8 doa nu20er

19

__02

JI'

IO?:3"~g ~s4

Ca
Cr

21

a-Ioa 21

.2

-21

21-100

-MO

-21

21-100

-2M

7115
Figure 2. 3~M 00:02:28ane-ople

i |

for map 0028


I

8T (day

-. in

00:02tk43

itm

2aur

3,992

z is cocnrtdaonocean clte n os smle ltsoscutrpa The~~~~~~


I
-

Th

etca xsi

rnei

mi fro

th

aaad1hmoiotl

eini oe a entknoe 6bas i ncld Wv.rne h os oe


ei nme 1 setx)

xsi

opgri

th avrg ~~ f+ad--F4.Tewvfrn is~~~~


'~-15

S-..Jt-1991 12=00:39

9 de-

5-fqN-1991 12:02:06

9 dea

5-J

!-199112:03:39

9 dq

6 M.

.. _

15

S
,-s

- a
-2 0 s-, 2W I

510

-I

--" _

'

-Z

...

. ,,,,,,

W-V

. .. -

..

... _ .,9 dins S-APM-1991

.. _12:t1:12 9dea

-L
MON

5"JM-19I91

M209:39

S-J1N-!991 12:12:41

9 dm

6,

. ",i .

'.

.-.

",

6_

-is

0
DoPpLer

Is -10

_2

-is

0
Do~lp:Ler

s -0oo

-200

-Is
{=ter

IS-I00

-2

Peakino'-se

ceooknousae

peok/no~Le

Figure 2.2

Amplitude -range -doppler maps for 12 UT (night in Amchitka), January 5, 1992.

The vertical axis is range in rnmi from the radar, and the horizontal axis is doppler in Hz. The ocean clutter is concentrated around 0 Hz. A median in power has been taken over 16 beams. T'he smaller plot shows clutter peak and noise in tmscaled dBW vs. range. The noise power is the average of + and - WRF/4. The waveformcode is number 117 (see text).
16

The peak-to-average power ratio is unity because we are considering FMCW with constant amplitude. The OTH enhancement is a factor that accounts for multipath and the conductivity of the ocean surface. We have used 6 dB rather than the conventional 12 dB. Some

other terms will be explained below, after a discussion of the signal processing in the radar.

As stated, the predicted clutter power is found by adding the predicted CNR to the CCIR noise. Some further adjustments are necessary when estimating absolute powers. In particular, the antenna and cable losses must be accounted for. (These cancel out in the ratio of clutter to noise and are not included in the equation above. In other words, we have replaced the customary directive gain with absolute gain. Directive gain is satisfactory when we require only CNR and the radar is externally noise-limited, so that both clutter and noise are reduced by the same factor in the antenna.) The antenna and cable losses are frequency-dependent and are shown in the following table. Frequency 5 MHz 6 MHz 7 MHz
8 MHz

Loss 21.6 dB 17.0 dB 13.9 dB


11.7 dB

Frequency 17 MHz 18 MHz 19 MHz


20 MHz

Loss 4.5 dB 4.5 dB 4.4 dB


4.3 dB

9 MHz
10 MHz I I MHz 12 MHz 13 M-Hz 14 MN-Iz 15 MHz 16 MIHz Table 9.

10.0 dB
8.7 7.4 6.3 5.6 5.2 4.9 dB dB dB dB dB dB

21 MHz
22 M-Hz 23 MHz 24 MHz 25 MHz 26 MHz 27 MIHz 28 MHz

4.3 dB
4.4 dB 4.5 dB 4.7 dB 5.2 dB 5.3 dB 5.4 dB 5.5 dB

4.6 dB

Antenna and Cable Losses as a Function of Frequency

Now turning to the recorded radar data, we must address the issue of scaling. The clutter and noise powers are provided by the "log-amp" data, which consists of received power vs. range and doppler expressed in decibels. This power is recorded after passing through the signal processing chain (range, azimuth, and doppler DFTs), but first it is offset by a so-called scaling factor. The scaling factor is chosen such that the values of power on tape are correct for a pointsource (coherent) target (which we may call a signal) at the center antenna element. However, since we have only offset the power resulting from the entire signal processing chain, the relative values of signal and clutter, signal and noise, and clutter and noise are the same as at the output of the signal processor. In other words, the radar only "sees" relative powers, which include signal processing gains and minus the losses. Naturally, the "growth" of the coherent part of the data is greater than for the incoherent part, which is how we get the signal-to-noise and signal-to-clutter gains necessary to detect targets. The signal processing growths (excluding losses) Jf zhe various data components are as follows:

17

INPUT Clutter Signal


Noise
.. . . ......... o

RANGE DFT WRF/BWpx 1


WRF/BW x
.................... oo ........... o . .... ... .o .

AZIMUTH DFT N x(O6t/OT) N1


NE.
.. ..... ..... o .. o ..... *.........

DOPPLER DFT
1

1/(WRF x CIT)
. .................. .............

CNR

NEL x(Ou/OT)

WRFxCIT

.......... ........................................
SNR BWtx/WRF NE WRF x CIT

Table 10. WRF BWx


NEL

Signal processing gains (excluding losses)


= =
= = =

Waveform Repetition Frequency Receiver Bandwidth Number of active elements


Transmit beamwidth Receive beamwidth

CIT

r Ox

Coherent Integration Time

The receive antenna gain is obtained in the azimuth DFT. Each element has its own receiver and all three transforms are performed in the digital (software) domain. Although conceptually it does not matter in which order we perform the three DFTs, the hardware is designed such that a range DFT is applied to the output of each separate receiver before they are all combined in the azimuthal DFT. Finally the doppler DFT is taken, power is extracted, and any scaling is applied. The range and doppler DFTs are normalized such that the coherent power receives unity gain and the incoherent (noise) power is reduced by the ratio of the DFT cell size to the noise bandwidth of the incoming signal. In the azimuthal DFT, on the other hand, the noise grows by the number of elements (or DFT points) and the coherent power grows by that number squared. In both cases, the gain of the coherent power with respect to the incoherent power is the same; the difference is in the normalization. Further remarks on the signal processing are provided in the following paragraphs.
In the range DFT, the clutter is treated like noise, because different patches of ocean surface at different ranges are independent so that no phase relationship between them is preserved in the Bragg scattering. The WRF gives the range DFT cell size or resolution because it is the inverse of the duration of the chirp. The bandwidth of the receiver determines the bandwidth of the noise before the DFT is applied. In the azimuth DFT, the clutter in any one azimuthal direction is treated as coherent (with

an N2 factor), because we have far-field plane waves impinging on the array. But now this

backscatter-dependent power is reduced by the ratio of receive to transmit beams. (The power is incoherent between different azimuthal cells for the same reason as in the range processing.) The noise, on the other hand, is assumed decorrelated between elements and isotropic (this last assumption may not correspond to reality) so that its gain is the customary N, where N is the number of DFT points, for a normalization such that coherent power is increased by N2 . Finally, in the doppler DFT, both signal and clutter are assumed coherent (the ocean clutter is usually coherent to first order over time at a given range), so that the SNR and CNR gains are both equal to the number of samples, which is the same as WRF times CIT (the CIT is the duration of all chirps in the waveform and we get one doppler sample at each range in each chirp). These signal processing gains must be adjusted for windowing, scalloping, and other losses, which total about 9 dB for the entire signal processing chain. 18

These signal prr--e#ssing gains indicate how we should adjust the recorded data to get agreement with the RADARC prediction. RADARC predicts the clutter and noise as observed by the radar after the signal processing chain. The data is scaled to a coherent target at the center element, but the relative powers of the different parts of the doppler spectrum are as observed by the radar. That is, the scale factor was chosen to subtract the processing gain of the coherent power, which from Table 10 is equal to the azimuthal DFT (array) gain only, as the range and doppler gains are unity for the signal. (The SNR increase in the range and doppler DFTs results from reduction of the noise power.) The array power gain is N2 , where N is the number of array elements, which varies with frequency. For the amplitude, we take N instead of N2 . Of course, we must adjust this antenna gain for the Taylor windowing losses. The compensated array gains are shown in the following table versus frequency. Frequency 5-16iM z 16-20 MHz 20-24 MHz 24-28 MHz Elements 372 300 244 202 Array Gain 25.7 dB 24.8 dB 23.9 dB 23.1 dB Windowing loss -1.2dB - 1.2 dB - 1.2 dB - 1.2 dB
--

Compensated Gain add24.5dBtodata add 23.6 dB to data add 22.7 dB to data add 21.9 dB to data

Table 11. Adjustment to data for array gain

4b.

Description of Software

We have received a very large number of ROTHR data tapes from Amchitka. Our first task was to archive the data on the 9-track radar tapes to a more convenient 8-mm format. Then we extracted the desired data from the 8-mm tapes and recorded it to a convenient database on the computer, namely, a large disk file that can be easily accessed when we want to look at the data in different ways. We refer to this operation as the "processing" of the ROTHR data tapes. The database is organized by month and hour. One hour of one day is processed at a time, as the ROTHR tapes are the actual real-time record of the radar operation. The processing involves extracting the "log-amplitude" data (i.e. received power vs. range and doppler) from all the data recorded. This log-amplitude data is recorded in units called dwells, since the radar scans over a large coverage area, illuminating one specific region at a time (for example, see Figures 2.1 and 2.2.). The data for the same hour on different days are accumulated to reduce thM variance in the clutter and noise associated with any given day. Our final goal is to obtain measures of average clutter and noise power for each geographical region of interest for certain hours of the day in each of the selected months, and compare this with the RADARC prediction based on monthly medians. When we average the data, we will use the median* rather than the linear average, so as to reduce "outliers" or other unusual circumstances which skew the data. We mean median inthe usual sense: in a given data set, half the values are greater than or equal and half are less than or equal. If the number of data points iseven, the lesser of two possible medians is chosen, although this technicality is trivial. 19

The ROTHR log amp data as recorded on tape is scaled to a reference point, namely the input to the receiver of the center element (specifically to the point where the cable from the twerp feeds the attenuator in front of the receiver). This scaling is correct only for a point source target. In order to effect the comparison with RADARC, we must undo this scaling as described in the previous section. Two computer programs are used in the analysis. This first is called Q4202. It reads the log-amplitude dam on the ROTHR tapes, one dwell at a time, averages this data over azimuth (i.e. beams) and range as described below, determines clutter and noise, and records these "data points" to a database. The second program is called QBASE. It reads the database, selects a set of ROTHR data points according to various criteria (month, hour, waveform code, range, frequency, azimuth), adds specified offsets and correction factors (if any), and plots these points along with the RADARC prediction. There are various plot formats. One is a scatter plot of ROTHR data points vs. RADARC predicted values (not shown in this report). Another plots the ROTHR data and RADARC prediction vs. range for specified frequency intervals. Figure 5.1 is an example of the latter. When program Q4202 reads the ROTHR tapes, it first reads in information from the tape headers. This includes TOA (Time of Action - both date and universal time), frequency, RWST (Range Window Start Time - which is multiplied by c/2 to give the slant range :o the first range cell), the number of range cells, the swept bandwidth (which gives the range cell resolution), the range extent of the dwell, the number of doppler cells, WRF (Waveform Repetition Frequency which gives the extent of the doppler axis), the number of beams, beam direction, beam width, and the beam spacing. The present analysis concentrates on waveform code 117 (the most common one), although it is sometimes necessary to consider waveform code 173 in order to get enough data points for a particular case. Waveform code 117 has the following parameters: "* number of range cells: 53 "* range extent: 953.2 km (514.7 nm) "range cell resolution: 17.99 km (9.71 nm) "* number of doppler cells: 128 *WRF: 42.0 Hz "* CIT: 3.1 seconds, and waveform code 173 has the following parameters: " number of range cells: 55 * range extent: 989.2 km (534.1 nm) "* range cell resolution: 17.99 km (9.71 nm) number of doppler cells: 128 " WRF: 31.0 Hz "* CIT: 4.1 seconds. To run program Q4202, a headers file must first be generated with program A4202. This headers file contains the TOA's of all the log amp dwells on a particular data tape, along with the WFC, frequency, start range (from RWST), and beam parameters. Then the tape is processed with program Q4202. A waveform code, month, and hour of the day must first be selected. A list of all available dwells satisfying these parameters is then generated by reading the headers file. To reduce the amount of data to a manageable level, only those azimuths within 20 degrees of boresight are retained. A further decimation of the data is A pair of endfur monopoles 20

achieved by selecting an arbitrarily chosen (10 in our case) maximum number of dwells from those azimuths. It is better not to have, say, 100 dwells from one azimuth and 10 from another, as we would then get a statistically biased database. Since each tape covers about half an hour, we get up to 20 dwells per azimuth per hour per day. Each dwell provides about 5 clutter and 5 noise points, after we average over a 97 nm range window (see below). This gives up to 100 clutter and 100 noise points per azimuth per hour per day. With 10 days of data per month, we get up to 1000 clutter and 1000 noise points per azimuth per hour per month. This should adequately represent the performance of the radar. The program now has a list of dwells for the data tape being processed. It begins to read and process these dwells sequentially. In each dwell, it first averages over about 80 of azimuth by which would be included if we took, say, a linear average.) Then an average (median) is taken over groups of 10 range cells (at each doppler cell) to further smooth the data. Thus the average in range is over 97 nm for the waveform codes chosen here. For any dwell, we now have a set of doppler clutter spectra that are averaged over beams and range. The program then determines noise and clutter at each of these averaged doppler spectra. This is repeated with each of the selected dwells and is recorded to the database. The whole process is repeated with each tape. The steps in this processing can be summarized as follows.

taking the median over the 16 beams at each range-doppler cell. (Taking the median eliminates "outliers"

"The raw data consists of amplitude-range-doppler maps "*Step 1: Step 2: SStep 3: * Step 4: Take median power over 16 beams at each range-doppler cell of dwell Take median power over 100 lan range at each doppler cell (-5 range cells) "Clutter" is maximum power in a small window around 0 Hz. "Noise" is dB average of power at + and - WRF/4.

* Each of these clutter and noise samples shows up as a tiny black dot on the plot * Open circles are the medians of black dots with 100 nm intervals (e.g. 700-800 nm) Table 12. Algorithm for extracting clutter and noise from raw data.

When program QBASE reads the ROTHR database, it also makes a comparison with the prediction from RADARC. Before processing the ROTHR tapes, the necessary output must first be generated with RADARC and recorded. For a given month, hour, and azimuth, RADARC provides (among other output) a CNR (clutter-to-noise) map vs. slant range and frequency. A complete list of inputs to RADARC is shown in Table 13. A CNR map is generated for each degree of azimuth between -20 and +20 degrees with respect to boresight and is recorded. This map also contains the CCIR noise, and so we can determine clutter power by adding the CNR to the noise. Each map is a grid of CNR at discrete points vs. frequency and slant range. The frequency is in 1 MHz increments and the slant range is in 80 nm map at the same frequency and slant range as the ROTHR data point currently being examined. The new version of RADARC (2.0) has been used for all data processed in this report. It tries to correct defects in the old version (1.0), particularly with respect to the statistical databases. antenna and cable losses not accounted for in RADARC and the ROTHR tape scaling described above.

(148 kim) increments. When QBASE reads this map to make a comparison, it interpolates on the

Corrections can also be inserted in the processing done by QBASE. Examples of this include

21

When we generate a plot from the ROTHR database, we can select intervals of frequency, range, and azimuth. This simplifies the analysis by focusing on some narrow parameter region over which some of the factors affecting the data are expected to remain constant. In particular, we have geogr.phy in mind, as described in the Introduction. It is also important to consider a narrow frequenc, band, as many of the components of the CNR equation vary considerably with

frequency. "* month, hour, and azimuth "* sunspot number (taken from the Solar Indices Bulletin published by the
National Geophysical Research Center). The smoothed sunspot numbers for 1991, at the time the analysis was performed, were: 148 for January, 147 for April, and 136 for July. "* transmitter location (Amchitka is at 51.44 N and 179.16 E)

" transmitter power (200 kW) " the ionospheric layer corrective multiplicative factors (nominal values kept) "the interference noise level in Amchitka for I Hz BW (nominal value 204.0 k-ept) "range resolution (206 :.s) "* average-to-peak power ratio (nominal value of 0 dB kept) "* the OTH path enhancement (nominal 6 dB kept) " CIT (set at 3.1 s to match WFC 117 or 4.1 s for WFC 173) "* the multiplicative factor for deviative absorption (nominal 1.0 for hops I and 2 kept) "* the type of noise (selected to be the maximum of the ROTHR noise curves or of the
galactic and atmospheric noise from CCIR 322) the vertical radiation increment (nominal 1.0 degrees kept), and the antenna type (selected to be ROTHR short ground screen). Table 13. Input parameters to RADARC model

4c.

Discussion

The comparison between the RADARC predicted clutter and noise and the ROTHR data is shown in the figures of Appendix C. There are two sets of figures, which both show the same data in different ways. The first set (Figures 5.1 to 5.10) shows the ROTHR clutter and noise power vs. range as dots and the RADARC model comparison as solid curves. (The open circles are median powers over range.) The second set (Figures 6.1 to 6.6) shows the median powers of the ROTHR data and of the model over selected range, frequency, and day intervals. In the first set of figures, each figure shows the RADARC prediction and the ROTHR data for a specific hour and month, as indicated at the top. The solid curves are the RADARC predicted clutter and noise powers, and the tiny black dots are the ROTHR data. The open circles are the medians of the ROTHR data over 100 mile intervals, as discussed above. Each subplot represents the data from a I MHz band. Of course, many of the I MHz bands have little or no data,

22

indicating that these were not good operating frequencies at the time and were therefore not selected by the radar operators. Essentially all we want to determine is how well the dam points near the top (the clutter) cluster along the RADARC prediction and how well the data points at the bottom (the noise) cluster along the flat curve representing the RADARC predicted noise. The difference between individual data points and median data points should be kept in mind, as the individual data points (representing individual dwells) will tend to have much more variability than medians (unless there were few data points available in the estimate of the median). Note that the actual days of the month represented by the ROTHR data points are indicated along the tops of the subplots. The RADARC frequency is also indicated; it is always the starting frequency of the 1 MHz band (e.g. 8 M-Hz for the 8-9 MHz band). The number of individual data points contained in the median estimate points is indicated along the bottoms of the subplots underneath the corresponding open circle representing the median estimate. The second set of figures in appendix C shows median dB powers of the data and the model vs. range, frequency, and day. For example, in the case of range, all the data points for a given month and hour were sorted into 100 mile bins, the first from 500 to 600 rr 'es (slant range), the second from 600 to 700 miles, and so on up to 2000 miles. The median wtutter and noise powers were then found in each bin for both the data and the model. Note that each bin contains all points at a given range, regardless of frequency or day. A similar procedure was followed for frequency and day. In the case of frequency, the bins were 1 MHz wide, from 5 to 28 MHz. All data points of a given frequency, regardless of range or day were placed in the corresponding bin and the median power taken. In the case of day, all data points from a given day of the month, regardless of range or frequency were placed in the proper bin and the median power taken. This procedure obviously provides additional insight into the source of errors. For example, if for a given month and hour there is a large error at a certain frequency and also for a certain day, then we may suspect some connection and this will guide us in taking a detailed look at the data. Better yet would be some display that shows correlations in error between range, frequency, and day.

Due to the spread-doppler clutter problem, the noise measurements are of particular interest. Recall that the RADARC predicted noise is CCIR ambient noise (without a folded spreaddoppler clutter estimate). Therefore, when range-folded spread-doppler clutter is present, we would expect the ROTHR noise to exceed significantly the RADAR noise. We would expect this to occur especially at night (hour 12 UT). The following tables presents the dB difference between the powers of the ROTHR measured noise and the RADARC predicted noise for all the data presented in Appendix C. The columns are the operating frequencies in MHz, as indicated by the numbers along the top. The rows are the month, as is evident from the labels at the left. The top table is the daytime noise difference (00 UT), and the bottom table is the nighttime noise difference (12 UT). The bold numbers indicate differences between data and model that exceed 10 dB. Indeed, the nighttime noise differences are significantly larger than for daytime, although January day is rather high. The nighttime noise for July was especially high. It should be remembered than all of the data presented in this report was restricted to a 10" azimuthal sector, and therefore does not represent an isotropic average.

23

The spectrum monitor data do not show noise levels that depart significantly from that of CCIR or internal noise [5]. Therefore we conclude ,hat the high noise levels are due to clutter spread in Doppler.

Noise difference
Freq JAN APR JUL 5 6 7 8 9

DAY (00 UT)

10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 5

21

7 7

7
1

7
1

6 1211101115
2 3 3 2 3 5 1

0 -1 -3 -2

1 0 -3 -1 2

-3 -2 -2 -4 -3 -1 -2 -3 -3 -2 1-2 -4

Frequency in MHz.

Noise difference
Freq JAN APR JUL I 18 5 6 7 6 8 5

NIGHT (12 UT)

9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 2 117 6 6 5

10 9

105

18 18 1711 8 16 20

Table 14. Noise power difference in dB of ROTHR median and RADARC prediction

24

For the clutter, we present a table of dB power differences vs. range rather than frequency. In the case of clutter, we assume that the operating frequency is optimized for a given range, and we are more interested in the significant differences in clutter power vs. range that occur due to the changes is propagation path and surface reflectivity. The bold numbers indicate differences between data and model that exceed 10 dB. (These values are averaged over frequency.) Again, this table summarizes the data in Appendix C.

Clutter difference

DAY (00 UT)

Range 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 JAN APR JUL -4 -10 2 -12 -2 11 -12 -2 138 11 4 -6 -12 -13 -13 -9 -9 9 -1 1 19 -2 1 -1 26 1 -1 23 -4 3 35

-17 .13 -14

-15 -15

Range in nmi

Clutter difference

NIGHT (12 UT)

Range 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600 1700 1800 1900 JAN APR JUL 3 -7 -8 -4 -8 0 -5 -3 -5 -3 -3 -4 21 3 8 -4 -8 .10 -5 3 20 11 5 4 22

Table 15. Clutter power difference in dB of ROTHR median and RADARC prediction

25

The analysis of the clutter is more qualitative and therefore more subjective. The following table presents a summary of our evaluation of the data presented in Appendix C.

Median Signal Level Comparison * Generally adequate considering the limitations on our knowledge of the antenna patterns, sea-scattering coefficient, median ionospheric parameters, etc. Distribution of levels (variability) about as expected from previous analysis of one-way data

- Another source of error, as shown by the WSBIs, is that significantly more


sporadic E occurred than was predicted. * No consistent bias. Median Noise Level Comparison - Good day time agreement * Nighttime levels higher than model due in part to spread-doppler clutter. Table 16. Summary of analysis of amplitude vs. range.

26

5.

Conclusions

For three seasons, ten days, noon and midnight, we have examined the vertical ionograms, the widesweep backscatter soundings, the radar's earth backscatter levels, and the ambient noise. These data have been compared with the RADARC Model predictions. In Appendices A, B, and C the data and predictions are available for detailed examinaton. Electron Density Distributions. The median of the observed vertical ionograms has been compared with the RADARC model. The emphasis was on the critical frequencies. Heights were also compared to a lesser extent. A simple algorithm was devised to permit computer comparison in addition to the comparison from manually drawn families. The model F2 layer critical frequencies are mostly within 0.5 MHz of the observed median, and in general the model shows agreement with the data. The April day data appears to straddle both winter and spring conditions; it would be interesting to examine the Fall data. There are indications that some adjustment in model layer heights and thicknesses might improve the model. The simple algorithm for machine analysis of median ionograms can and should be improved if more detail is desired. We feel that, in general, any improvement in model electron distributions will have only a small impact on

predicted performance.

Wide Sweep Backscatter lonograms (WSBI). These oblique earth backscatter soundings should provide the best data for comparison with model performance prediction. WSBI intensity-range-frequency plots are compared with RADARC plots in a qualitative manner. The skip zones can be compared on an absolute basis, and they are very similar. Since the times used are near midday and midnight, there were probably no severe horizontal electron density gradients, and consequently the deficiencies in RADARC's treatment of "tilts" are not exposed. Further investigation will examine sunrise and sunset for the same ROTHR database as was used in this report. Radar Ground Backscatter Levels. Appendix C Figure 6, the amplitude versus range plots provide a good comparative view between prediction and observation. Some of the disagreement is certainly due to our assumption of a constant surface scattering coefficient. There appears to be a slight optimistic bias by the Model which is due in part to the existence of more sporadic E ionization than was predicted. However, considering the complexity of the ionosphere and of the propagation, and the fact that we are trying to predict these events from a statistical

database with no reliance on auxiliary real-time measurements, the comparison between signal
levels (clutter) from the RADARC model and the ROTHR data is fairly good.

familiar form. Actual noise is higher than the Model at night for all seasons and is also higher by turned off and which comes from clutter spread out in Doppler.

Noise. The Appendix C Figure 6 noise level versus frequency curves give a comparison in

day in January. Examination of amplitude-range-Doppler maps shows this discrepancy to generally be due to "multiplicative" noise; that is, noise that disappears when the transmitter is those due to atmospheric, galactic or internal noise. These generally occurred at night and during some January days. We feel that an improvement in performance prediction requires an improved ionospheric irregularity model since this is the source of spread doppler clutter. Equally important,
path tracing techniques must be incorporated that permit folding-in range-ambiguous and azimuthal sidelobe spread clutter. We have work underway that addresses these improvements. The major deficiency exposed in RADARC is due to radar noise levels that are higher than

27

References
[1] J.M. Headrick, J.F. Thomason, D.L. Lucas, S.R. McCammon, R.A. Hanson and J.L. Lloyd, Virtual Path Tracingfor HF RadarIncluding an Ionospheric Model, NRL Memorandum Report 2226, March 1971 J.F. Thomason, John Lloyd and Glenn Skaggs, A Global IonosphericModel, NRL Report 8321, August 1979 CCIR (International Radio Consultative Committee): World Disvibution and Characteristcs of AtmosphericRadio Noise, CCIR Report 322, International Telecommunications Union, 1964 A.D. Spaulding and J.S. Washburn, Atmospheric Radio Noise: Worldwide Levels and Other NTIA Report 85-173, National Telecommunications and Information Characteristics, Administration, April 1985 of High Frequency Signals,Noise, Availability andDuration G.D. McNeal, Characteristics of Bandwidths Based on ROTHR Amchitka Spectrum Measurements, NRL Report 9536, underpublication

[2] [3]

[4]

[5]

28

Appendix A Comparison of RADARC Predicted Quasi-Vertical lonograms and ROTHR Median Observed Ionograms

Figures 3.1

to

3.6

Hours 00 UT and 12 UT January, April, and July, 1991

29

OVI

JAN 00
:ROTHR med~on

+ : RARDRC
rC
Hc TM

E
FI

3.266 110.0

20.0

""F2
* ES

O.O0 0 0.0 0.0 11.867 31S.3 109.4


L M U

3.42 2.74 2.18

-q.

-s

RO-THR ionograms for separate days

0 7-

0H

longrmsm fo

eprtedy

ii
I.
F~requencq (MHz)
Nmm...

Figure 3.1 RADARC predicted quasi-vertical ionogram for hour 00 UT (local daytime) of January vs. the monthly median of ROTHR observed ionograms. The RADARC prediction is based on an estimated monthly median overhead ionosphere. On each individual ROTHR ionogram. the trace is taken to be the maximum power over height at each frequency. The median height of all traces is then taken at each frequency. This plot is based on about 40 to 45 ROTHR traces or about ten days of data.
30

OVI

JAN 12
ROTHR medtan

RRORRC FC HC Th 0.517 110.0 20.0 0.0 a o.0o o.o


L I1 U

E r-

'2 3.666 386.7 97.1


ES 3.29 1.23 0.91

SROTHR

ionograms for separate days

-I.C I'

iS

r* 00

impI

2....

5162

FrequencS (MHz)

Figure 3.2. RADARC predicted quasi-vertical ionogram for hour 12 UT (local nighttime) of January vs. the monthly median of ROTHR observed ionograms. The RADARC prediction is based on an estimated monthly median overhead ionosphere. On each individual ROTHR ionogram, the trace is taken to be the maxim=u power over height at each frequency. The median height of all traces is then taken at each frequency. This plot is based on about 40 to

45 ROTHR traces or about ten days of data.


31

OVI

APR 00
ROTHR medLon

+ : RRODRC
FC HC 3.966 110.0 ym 20.0
48.0 U

E "
*"F2 "

F1' 5.623 192.2 L

10.910 394.6 170.7

11 3.01

ES

3.37

2.69

C_=
4

o""
"
*
..

ROTHR ionograms for separate days


0*
I . . . I

0o

0o;

'

10

1'2 " 14

1 6

18,e

FrequencS (MHz)

Figure 3.3. RADARC predicted quasi-vertical ionogram for hour 00 LUT (local daytime) of April vs. the monthly mediRa of ROTHR observed ionograms. The RpDARC prediction is based on an estimated monthly median overhead ionosphere. On each individual ROTHR ionogram, the trace is taken to be the maximumn power over height at each freqluency. Ile median height of all traces is then taken at each frequency. This plot is based on about 40 to 45 ROTHR traces or about ten days of data.

32

CVI

RPR 12
ROTHR medLon +
C

RRDARC
IC HC
Y11

0.407 110.0

20.0

ri o.ooo
r2 CS E
SL

0.0
M

0.0
U

6.520 430.6 115.4 2.45 0.80 0.83

*
o 4

.. -'_ .-,. ROTHR lonograms for separate days

l,4

1'

r1 1, 1

1,7

`2

Frequency (MHz)}
I I I I II I I I I I

Figure ... RADARC predicted quasi-vertical ionogram for hour 12 UT (local nighttime) of April vs. the monthly median of ROTHR observed ionograms. The RADARC prediction is based on an estimated monthly median overhead ionosphere. On each individual ROTHIR ionogram, the trace is taken to be the maximum power over height at each frequency. The median height of all traces is then taken at each frequency. This plot is based on about 40 to 45 ROTHR traces or about ten days of dama

33

OVI

JUL 00
ROTHR medLon

00+

RADARC
rc

HC

TM

* F "

3.902 110.0 20.0 5.272 184.1 46.0 r2 7.835 397.7 194.6 F C


L M U

ES

4.98 3.56

2.83

-J

ROTHR ionograms for separate days

*
0%

.
00*

I
I.

tI

FrequencS ([MHzz

10

12

1" 6

1"

20

RADARC predicted quasi-vertical ionogram for hour 00 UT (local daytime) Figure 3.5. of July vs. the monthly median of ROTHR observed ionograms. The RADARC prediction is based on an estimated monthly median overhead ionosphere. On each individual ROTHR ionogram, tle trace is taken to be the maximum power over height at each frequency. The median height of all traces is then taken at each frequency. This plot is based on about 40 to 45 ROTHR traces or about ten days of data.
34

OVI

JUL 12
ROTHR medLon

7.

+ :RROARC

SFC
E F 1 0.000
* ES E L 4.71

HC

0.564 110.0 0.0

20.0 0.0

"F2 7.055 377.2 104.0


M U 1.88 0.85

"

-"

ROTHR ionograms for separate days

97,,

00

A .
2
0

ge

to

14

i'

---

IIIi

Frequencq (MHz)

Figure 3.6 . RAD)ARC predicted quasi-vertical ionogram for hour 12 UIT (local nighttime) of July vs. the monthly median of ROTHR observed ionograms. Ile RADARC prediction is based on an estimted monthly median overhead ionosphere. On each individual ROTH-R ionogram, the trace is taken to be the maximum power over height at each frequency. The median height of all traces is then taken at each frequency. T1his plot is based on about 40 to 45 ROTHR traces or about ten days of data.

35

Appendix B

Comparison of RADARC Clutter-to-Noise Maps and ROTHR Widesweep Backscatter Ionograms

Figures 4.1

to

4.6

Hours 00 UT and 12 UT January, April, and July, 1991

37

JAN 00 UT ROTHR NSBI CORR RAOARC CNR

':
O0

I, "-

: 1-

xo

U,0,
x
0 5Sa 10030 1500 2000

L. L)Q
Range (nmt*]

x!

Soo

1000O

1500

2000

Range (nm' ]

I
..

ii'ia
F . .

..........
.0

ii i il
. ,. . e.. .

"C"
..

~~~C
,I 0

. .

. .. .

. .

. . .

.25
,

0 0
, ec.Fr..(

.o * is o
aa(.z

25o

10 2

5 25

15

20

Fiur 4. nos (CR

1 ma

ROH fo 0 UT

wiewe (no

bakcte inAcik)ooanay

iooga

iWBI 91 h

vs

RAACcuteSssonhr r

Frequency

(M1Hz]

Freouen~cy (M'Hz]

Figure 4.__ _.

ROTHR widesweep backscatter ionogram (WSBI) vs. RADARC clutter-to-

noise (CNR) map for 0 UT (noon in Amchitka) of January, 1991. The WSBIs shown here are

each the average in power over about 45 individual WSBIs recorded during the first 10 days of the month. The values in each frequency-range cell are averaged separately. The CNR map is not expected to be identical to the WSBI, but should at least have a similar shape.

38

JAN 12 UT ROTHR WSBI CORR RADARC CNR

.%n
II12 "" o0l

~
wi
S.

Ln0
2 -

3+
E~q

z
0

x
0

.......

50o

1000

1500

2000

5001

000

1500

2000

Range (nm'l)

Range

[nmt.)

t i I
o
i

. . ... .

I
g

fl'
i

...

IC
.

i'
20 (MHz) 2S

In

__

_ _ _ _ __

_ _ _ _ _ __

_ _ _ _

10

is

10

is

20
(MHz)

25

Frequency

Fr'equenc.

Figure 4.2 ROTHR ;videsweep backscatter ionogram (WSBI) vs. RADARC clutter-tonoise (CNR) map for 12 UT (midnight in Amchitka) of January.

39

APR 00 UT ROTHR WSBI CORR RAOARC CNR

ZI

f)

x
0 5600 10,00.... 15,00, 2000 0 500 1000 1500 2000

Range (nmL.A

Range (nmil)

. ...

. .

. .

Gaega . ..

..

2C C C

. . .*

*lit

rrequenc. (MHz)
.ROTHR

rrequency (M1Hz)

Figure 4.3 widesweep backscatter ionogram (WSBI) vs. RADARC clutter-tonoise (CNR) map for 0 UT (noon in Amchitka) of April.

40

APR 12 UT ROTHR NSBI CORR


-

RADARC CNR
0

"o0-

"p-N

"

,' I

,r
II C

D' Z Up

~I10)
1500 2000

"-_r,

al

CD

500

1000

Soo

1000

1500

2000

Range (nm,

Range (nmL,]

""tip "
.... ....

:"~~~~

In

i 0 a"

~~

g.

a -C
I
,

a
Se.s. IO~

."

itti i

;1
, .

.I
I . ..

.";I*
. I

10

is

20

25

10

is

20

25

r'requencS (MHz)

Frequenct

(MHzj

Figure 4.4. ROTHR widesweep backscatter ionogram (WSBI) vs. RADARC clutter-tonoise (CNR) map for 12 UT (midnight in Amchitka) of April.

41

JUL 00 UT
ROTHR WSBI CORR RAOARC CNR
.,
4

-a,
0 Wi ,

Ji
-riw,,,.

,,
I

~N0

U,

"

%Zt

Range
-CD

(n

Rang

('--'u

,,j 1
.. . .I . . I
.1

..-

S.
00

S.....'

Fiue45
0

RTRne
10 is

[nmewee
2 0 2 0 0 10

s RDRCcuteWBI Rongeate [nmra


is 2 0 25

Foigue (C

r JWulys.R bAmchitka)e ionf a fr0U (noon inesee

cute-o

42

JUL 12 UT ROTHR NSBI CORR RAOARC CNR

.~.9

0~

500

lo00

1500

2000

500

1000

1500

2000

Range

[nm'6)

Range

(nmL]

II

..-...

........

.. . .. . . . .. . .

... . . . . ... .
1 0 i t . . . . . . .

10

is

20

25

10

is
4

20
(MHz)

2S

FrequncS M~z)Frequenct
.ROTHR

widesweep backscatter ionogram (WSBI) vs. RADARC clutter-toFigure 4.6 noise (CNR) map for 12 UT (midnight in Amchitka) of July.

43

Appendix C

Comparison of RADARC Predicted Clutter and Noise and ROTHR Data from Amchitka

Figures

5.1

to

6.6

The following plots compare RADARC predictions with measured data from Amchitka. The data consists of peaks (clutter maximum near 0 Hz) and noise (average in dB of power at + and WRFI4). The latest version (2.0) of RADARC is shown. Data is shown for hours 00 UT (noon in Amchitka) and 12 UT (midnight in Amchitka) for January, April. and July of 1991. All plots are restricted to an angular sector of 0 to 10 degrees with respect to boresight, so that the geography (and hence reflection) at a given range is relatively constant over azimuth. This geography consists of mixed land and sea from 500 to 1500 nm and mostly land from 1500 to 2000nm. The data has been scaled in dBW to approximate the absolute level of power at the output of the array. The data has been smoothed by taking the median over all sixteen beams and over 100 nm range intervals.

45

Amch*,tko CLutter & Notse


SoLd Line
-

JAN
Doys;

0 UT
8 910

5-17 MHz

RROIRC

Open circLee - ROTHR medans over 100 rno

M I"
0

5 MHz
0

o500

1000

1500

2000

500

1000

1 500

0
..

2M

I.........

500

10150
...

HZ2.....1

6 MHz

..

2000

"-'

7.......
So00 100 1500

.'12

MHz

Co

/_135

7 MHz

MHz

0 C4

500 500
M8MHz

1 3000

~ 150o 000 1500

Dogs:
2000 0

3 54 7 100 500
500

981

1000

,500 1500
14:MHz

-0

.. .
*

5.56._ ...

500

10R00.

. 1500 .s

2000

50s

lo00

a p

noise) from Amchitka. Circles are medians of data points over 100 rnm intervals. Solid lines are RADARC predictions (version 2.0). Data is for January 1991, 0 UT (local daytime), 517 MHz, 0 to 10 degrees wit~h respect to boresight. Each subplot is a 1 MHZ band.

46

AmchLtko CLutter Z Notse


SoL.d LLne - RROWC

JAN

0 UT

17-29 MHz

Ooen cirtcLos - ROTHR maoLons over 100 rm

-o Days;
L
CD

1 2 3

6 8

9 10 11 7a

Dy:346

di02 -1

i 21

CD 7

P 0 Days 1000 1500

0 u. Da0s 2
1000

1500

2000

8~ Dogs:

2 3

456

8011C

Dogs: ID 4~

0 -

50

1000

. .00

L.

2000

500

1000

1500

DOogs: CL 0

3 ,L 5 = .... s.... 00 500 500 1000

""20 "As0 *I. 789101


1500 2000

as469 0

0
o00

i
1000 1500

0In
Li

0o

0
g*0egseoN

s00

1000 Ranqe

15

oo00

0oI0 0000 Ronqe [m

fri)

Figure 5.2.
QI, .al

Radar backscatter power vs. range. Dots are ROTHR data points (peaks and

noise) from Amchitka. Circles are medians of data points over 100 rim intervals. Solid lines (local daytime), 17are RADARC predictions (version 2.0). Data is for January 1991, is0a UT 1 MHZ band. 28 MHz, 0 to 10 degrees with respect to boresight. Each subplot

47

AmchLtko CLutter & Notse


SoLLd u.n.
-"

JAN 12 UT
-

5-17 MHz

RROARC

Open circLea

ROTHR moduons over 100 nO

0ogs; 0
O

4" 5

9 10

0.:.5 MHz

11 MHz

W lie
CUI 0 0 500
r2 112,

40

______________________

1000

1500

2000

500

1000

1500

S.
Li1 0

.6

,Hz....

12 MHz

Soo

1000

1L 1500

5000 S00

1000

1500

.1H

o
CL

oo*o
50 60
1000 1500

1o oo 2000

"so
500o

'0"
1000

"so
1500

00

D-= ogs;

6 500 1000 1500 2000 0


0O

"
L

~.
Day0os:

.... .... II..................... ......

I..

I S0
.

1000

100

M9
Ln -, Figure 3 L 7a)

M1Hz
0

15 M1Hz

r.ge
:0

Dot
o

.. .aa .akce .oe d.t .s .r .OH .ons(ekn


00 10 T lcaMihttmz

50o

1000

1500

ar

AAC rdcios(esin20.

Dt

o Janar
0

1911

10 P1-a-0
500 1000 2500 2000 0 500 1000

1611H.z..
1500

Radar backscatter power vs. range. Dots are ROTHR data points (peaks and Figure 5.3 . noise) from Amchitka. Circles are medians of data points over 100 run intervals. Solid lines are RADARC predictions (version 2.0). Data is for January 1991, 12 UT (local nighttime), 5-17 MHz, 0 to 10 degrees with respect to boresight. Each subplot is a 1 MHZ band.

48

AmchLtko CLutter & Nouse


SoLLd LLne " RRORRC

APR

0 UT

5-17 M1Hz

Goen ctrcLev - ROTHR medLons over 100 n

SDous;

M-

5MHz

io

11MHz

0 N.

M "L 7

500

1000

1500

/ MHz 2000

7]

500

1000 1 5 6810

1500

SDogs;-1

.
1 iHz

M 2L -01000
a)

MHz
I 2C00
21

Soo
-A

1000
o

1500

8 MHz

Doys: g

14 MHz

O__ 0

500

1000

1500

2000

Soo

1000

1500
. 15 MHz

S Dogs:
Liin

1 2 6 7 8

9 MHz

7-

500

1000

1500

200

500

1000

1500

;~Dags:
M

452Dogs;

2 3 5 6 7 8

o10

MHz

16 MHz

.. 0 0-0

"

So0

isT

low0
tnm)

1500

26000

S0o

1000
Ronqe (nm}

1500

Si',G

Ronqe

Radar backscatter power vs. range. Dots are ROTHR data points (peaks and Figure 5.4. noise) from Amchitka. Circles are medians of data points over 100 nm intervals. Solid lines are RADARC predictions (version 2.0). Data is for April 1991, 0 UT (local daytime), 5-17 MHz, 0 to 10 degrees with respect to boresight. Each subplot is a 1 MHZ band.

49

Amchttko CLutter & Notse


SoLa.d Line
-CO -

APR

0 UT

17-29 MHz

RAOARC

Open cLrcLse

Dogs:

1 2

- ROTHR meOa.nw over 100 no Oys: -1 1 2 5 7 8 9

.M7Hz
L~
03

=
0.

CD
o500

0 0

1000

1 500

20 00

0 O
0

S o00
3 6 7 98 to

i "-

ogs: -0.

4
-~

6 o.a.

~48M11.

Dogs: )Hz

30

-~Dogs:

5 8

9 1012

ocus:

5 6

7 8

o I
0
a

0T

,oo
1000 1500 10 3459

MHZ 20
2000

0o
0 Dogs:

500
500 -1

1000
1000 1500
O

zso

Dogs;

MOtHz
to
La
-

QR

(500So Dogs: 5 6

1000

1500

2000

500

10'00*

1500

CD

Dogs:

-1 1

00
0.. So00

0
1500 00 0 5Soo 1000 1500

Dogs:
L

1 2

3 4 6 7
7:

8 9

0 11

__

____311,___,_N____T_

CL0
&-f*o-W

5oo

1000

1500

2000

500

Ronqe (nm)

1000 Rarqe nmm1

1S00

Figure 5.5. Radar backscatter power vs. range. Dots are ROTHR data points (peaks and noise) from Amnchitka. Circles are medians of data points over 100 rnm intervals. Solid lines are RADARC predictions (version 2.0). Data is for April 1991, 0 UT (local daytime), 17- 28 MHz, 0 to 10 degrees with respect to boresight. Each suuplot is a 1 MHZ band. 50

AmchLtko CLutter & NoLse


SoLLd LLne
-

APR 12 UT 2

8-20 MHz

ROFIRC

Open ctrcLes - ROTHR me.uons over 100 n

Dogs:

00 CL 0

500

1000

15,00

2000

S00

1000

1500

8
CD
LA

_
-0

500
. . . .

10oo00
. .

1500
. ..

2000

0
4!1,1
.. Dous:

Soo 1 2

1o0o

1500

2C3 3

3.

5 ..6

Lf

0'"i

560

10 00

1500

20'00

Soo 0 1 3

1003

100

?71
Li-

8 Days:
7]

6 7

*1i

0
Or.

500
. . . .

1000
. .

1500
. .

2000

500 QM ous:

1000

1500

2
0
.I

7 9
,~,. . . I .@. .

38t j2 0 M 10 to
n

lolls,

50 98

10 e
( m

10s00

0
o,,nac Dogs:

5;0

1000
m

15M00

7 10o a

-0 "'1
04fU

ane(,)Rao

rm

FiLir

5-.Rdrbcsatrpwe

iag.Dt

OTRdt

os(ek

are RADAC ~

prditonsq(ersinm

2.0).

Daai

pi

912U

loclnihtme,8

a-

III

20 MHz, 0 to 10 degrees with respect to boresight. Each subplot is a I MHZ band.


51

200

56

00

10

Rmchttko CLutter 4 Noise


SoL~d Ltne
CD

JUL
0

0 UT

5-17 MHz

" RROARC

Open cLrcLes -

ROTHR meOmons over 100 nro

5 MHz

11 MHz

0 0 500 1000 1500 2000 0 500 1000 . . 1500

"-o
L-I

S6
__ __ _---_--__

MHz
U0

12 MHz

ii7
__ _

___

a-

500

1000

1500

2000 0

500

1000

1500

30 M7

DOus:

MHz

13 MHz

0-500

1000

,500

2000

Soo

1000

5Soo

Doss:

1 6

8 9

Li7
0
. . . .. a "

M -~
-00

M0

500

1000

80 1500

MHz 20'00 9 MI

0 Dous; 7...,.1J4 0

0
p

1 500

2 31004 7 8 10 15009 MHz 1.5l MH:1

nie0

5oo

1600

15a00

o-Too d

1000

1500

i , 1.MHz
0

.0.z

Drusy

1 2

10
01I

4 to

8 9 10

0- 0

S0o

1000o

1500

2000

500

1000

1500

S-fu*W io0@0143 Ronqe

(nm]

Ronqe (rim]

Radar backscatter power vs. range. Dots are RQTHR data points (peaks and Figure 5.7. Solid lines I I are Imedians of data points over 100 nm intervals. Circles noise) from Amchitka. 5-17 (local daytime), 0 UT July 1991, is for 2.0). Data (version predictions RADARC are MHz, 0 to 10 degrees with respect to boresight. Each subplot is a 1 MHZ band.
II

52

9mch'Ltko CLutter I Nouse


SoLLd LL.n
-

JUL
-

0 UT
5
oi

17-29 MHz

RR0ARC

0Opn -LrcLes

ROTHR medans over 100 rn

Dous: -2
Ma

910

Qous: 0.0 0

17 MHz
30

23 MHz

0
-~Dogs:

500

1000

1500

2000 8
i

500

1000

1500

-2 -

1 2 67 o,18 ,MHz

Days:

5 24 MHz

b.

,"0

o
U-

5oo
500

1000
2

15oo0
MHz 15000

2000

CD -I
0

0
0

Soo
500

1000" ' io00 "


51H
1000 1500 MHz

M -19 a 0

SDays: -2 4

-25

10600jo

a~

500

1000

So*MHz 2191H
1500

2000

500

1000

15I0

27 MHz
MHz

M
L.

20 MHz

7-26

Figur

5
C0 0 M

l...._._. 3aa bakcte


10 -00
So

wt

rM

.21

pwrv.rag.DtRaeRT MHz
5 insor 0So 1000

daap0t

(ek oi

n e

1e0e

1500
intras

fo Acita 500 CD Dogs; 5 6


0

1000le
1000

150me0ian ofdtap
0

0n
IOI

0 0

500

1500

200

M~~~~zM~ to1Mere
-0

05 ih

epc

obrsih.Ec
53

upotiz

ad

flmchi~tkc CLutter A Not'se


SoLLd L~ne RROARC

JUL 12 UT
R0THR meoauori

5-17 MHz
over- 100 nm

Open c~rcLes-

L -0)E '0

500

1000

I1500

2000

So0

1000

1500

1
-0

2= 6MMz
7
0l

Days;

4 7

's4

a-0

5600

ioa

15600

2000

Soo Days. 7 7 8

1000

15oo

Mz

5-00So

1000

1500

2000

2o

l0001500

Days:

25~3 1

5~~ 7os

100

1;-il

M 0t10 adegee wit respec to0 borsiht Eah


45

6ublo0' 10MHZ and

AmchLtko CLutter & Notse


SoLiLd LLie R-MOARC

JUL 12 UT
-

17-29 MHz

Open curcLes 0

ROTHR medions over 100 no

a ys:

2 7

~0
0 0 S00 1000 1500 2000 0

500

1000

1500

7 ,o

Dao so

o oo
.so

24 MoHz
ioo o2oo3M i

t.o_
0 e0

0A0AR
-040

50

1000

1500

2000

5D0

10o00.

1 5

M2 L

0-

So

1000

1500

2000

7. 0

500

1000

1500

MHz

0os)fo

scht.

1000le

at

onsoe

10n

nevls 1500aso

271MHz

oi

ie

0.J0

I0n

0.So140

150

CA

M22
-0

28 MHz

CL0

500

1000

1500

2000

500

1000

1500

IO *II

Ronqe (nmi

RonQe Irim)

Figure 5.-10. Radar backscatter power vs. range. Dots are ROTHR data points (peaks and noise) from Amchitka. Circles are medians of data points over 100 rnn intervals. Solid lines are RADARC predictions (version 2.0). Data is for July 1991, 12 UJT (local nighttime), 1728 MHz, 0 to 10 degrees with respect to boresight. Each subplot is a 1 MHZ band.
55

0- RRORRC cLutter 0- RRORRC noLse


G- ROTHR not.SO

ROTHR cLuLter

JNUARY

00 U

I04'

10.17.04

"'

...............................................................

.............................................................................

C ................................................
C]. S00 I

............ e.. . . ...... ...... ....... ....... ........ ............ ...... .... -... .
e3...3A31 0 00 0
I15 7 142 135 20 151 131 I 67 46

PeS.I

18 24

1000 I0

15s0

2000

Range
_ ................................................................ ....... 0 ............... . . ................ 0 ............ ..........

.07 0 .0 =................................ ............ o. .. .... .. .. .. . .. . 0 .. ... . ... .,' . . . .. .0 0 .. ... ... "-o

.. Q

". .......... .. . . O ............ . .. . . ... . . ... .. ... ... 0 .. . .


',IUYfI.
M

.................

465 0

3SUU32 N14

04US174

54 3Sl

230 12

*07 0-

C5 .......................................................................................................................................... Fruquorcmg
S ........... "............ ............. _.............. .....................

... .... ... . .......... .... ................................................ O.......".. .....


0

M,

0
N MIS 74

0
304 I

0
32 I

0
140

0
7 12 T i

0
94tM 7.

0
3230 8

0
12 304

0
4Q

10

Figure 6.1

Radar backscatter power medians with respect to range, frequency, and

.. . ... . .. . . . . . . . . . . . . . . . .

.. . .. .

.. . . . .

.. . . . . . . . . . .. . . . . . . . . . .. .. .. . .. .. .

day for 00 UT of January. The data points are the same as in the corresponding month and hour of the preceding set of figures. For example, the median over range at 500 nm includes all data points between 500 and 600 nim, regardless of day or frequency. The number of data poits in the median is shown along the bottom.

56

ROTHR cLuttor

JANUARY

12 UT

o- RADRRC cLutLer

a- ROTHR no ,so
0- RRDRRC nose
10,@WuaO106=0104

. . ................... . . . .. . . . . .. . . . . .. . .2 ..
i0

. . ..... . . .
o .....

9 -"C ... .. . . . .. . .... .. . ... " .. W.........................o.....g......... . . . .. .. .. .................................. "0


~~~~~~~~~~~ ..... ..... S.............................
00 123 0t0 123

. .. . . .= .. . " .. . .. . . ... ......... '


0 0n

" '

. . . . .......... .... '".... .... =.........................


00 W0

.
0

0 0 0 0 12 so so 4

Range

..i.... ........................................................................................ - ... ....... ........


0

........ ......

.. . . . ... . . ... . . ... . ... . . . ... . . ... . . .. . . . ... . . ... . . .

07
*

........................................................................................... r ...................
10
510 IS 2 S

Frequenc

! .................................... j ........... a .......... It.......... a - .".......... 2........... 6l .......... 0 ........... 0 ........... N.

8 .. . ................................

0.......................................

............ ..........
0
I IhI4S

..........
0
60

..........
0
3~

.......... ........... .................................... ........... ...........


0:
60

0
60

0)

*0I
V 8

0 9

0
I

14

IQ 1

Figure 6.2_. Radar backscatter power medians with respect to range, frequency, and day for 12 UT of January. The data points are the same as in the corresponding month and hour of the preceding set of figures. For example, the median over range at 500 rn includes all data points between 500 and 600 rnm, regardless of day or frequency. The number of data points in the median is shown along the bottom.
57

O- RRODRC cLutter

o ROTHR cLut.ter

APRIL

00 UT

1- ROTHR nose n- RAODRC notse


IO-II 00 0 a 10121141

. ................................

g 8 000 0 W ..................

0 0. ... ...................

C.9 .............................................................. 0-0

a....................................... a..................................
6 91 104 308104914 93111 it 41

'eli W"
0.. .. .

11

111II 55

01

460

""w0

106015 Range

2M0

. ......................................................
.............................................
wa , "a

o...'....o........ ........... 0. .......... ........

... ....................... .

. -4 . ......... W0 ...... . J.... .. ........................ . . .......... ..... ...... ..... . .S


0T

...
0 0

.......... ... . .......... ............. ......................................................................


0 0

"o ...........
0 0

10

*8.
0 0

cM

10

[)as

Figure 6.3. Radar backscatter power medians with respect to range, frequency, and day for 00 UT of April. The data points are the same as in the corresponding month and hour of the preceding set of figures. For example, the median over range at 500 rum includes all data points between 500 and 600 rim, regardless of day or frequency. The number of data points in the median is shown along the bottom.
58

0- ROTHR cLuLter
RRORRC cLutter a- ROTHR noLse 0- RRDARC no.se
0-

APRIL

12 UT

.........

...

s21

C,
. ,

............... . . ... . . ... . . -........ ......................................................... ..................> ...... 0 0


...............................
WT1 1 1 low100

................................................
54

o ... .... ... 3 .., 13 ..... 3o


54 110 Ito 1500 417 47

2000

. . I ................................................... "t"

....................... 0 0 0 ............ 0........... ...... ................... * * .* 4p 0 ............................................................... ...........

........

C.........................................................................................................
0 .J 0

..................................
..........................
a

MU

................................
22 i4 n72n 33 2
24 A I0 2

NMI=

Frequency

7
0

0
0 ...... 0

C51 ..............., ........... ..............., ............................. ................... ........... ........... ............... ......... ~0
Wn'a 20
2 2

0-0

am
3 1

to S

40 6

34 6

20 10 11

Do

Figure 6.4.

Radar backscatter power medians with respect to range, frequency, and

day for 12 UT of April. The data points are the same as in the corresponding month and hour of the preceding set of figures. For example, the median over range at 500 rum includes all data points between 500 and 600 rim, regardless of day or frequency. The number of data points in the median is slown along the bottom.
59

- ROTHR cLuLt.ar 0- RAOARC cLut.ter


a- ROTHR noa.se C- RAORRC na.se

JULY

00 UT

10--

10M24032

"_
"" -

.................................

.. .

.............................................

............. ............ O .................................................................... 0 m0 1 0 0 0 0 0 8 0 ....... 1-...... 0 ............ 0'" .... " I...*..***...... 9 .... o 0

".J 0. ............................................................................................................................. R 1 a a = R 6 9
Po4 V 23 62 114 72 137 155 23

0 9
22

Q.,.... 0
4 v

. ..

Ronge

L4._ . ........................................ .......... .... 0 o..................... ......... ............. .....


............... ...... a....... A. ... ......
0

.............
0

,.......................................................................................................................................... NM 1f0 0 Si 108 33 1"4 71MIN 1 D1

I'
Frequonce

26

is

iD

............ ........... ..........................................................................


C0

........... ...........

...........

.................................... 0. . . . .. ............. .=. _............................................................... o


00

0 ..............

' ............ ......... ... ........... -............ i.. . . . . . ............. . . . . . . . .......... ........... ........... -.......... .
0-0

*0o

Figure 6..55 Radar backscatter power medians with respect to range, frequency, and day for 00 UT of July. 'Me data points are the same as in the corresponding month and hour of the preceding set of figures. For example, the median over range at 500 rnm includes all data points between 500 and 600 tnm, regardless of day or frequency. The number of data points in the median is shown along the bottom. 60

- ROTHR cLuLter
0- RARDRC cLuLter

JULY

12 UT

a- ROTHR noLSe 0- RRDSRC noise


10"4WO- I0M2940

. . a
07 .

...............................

...................................

.........................
.............

. .......................................................................................................................... .N

0
...... ,0O 12 12 1NIle 2 33 V 24 .................. . . . . . . ..... . . . k........ . .( . .................................................... 0 S? 000 103 24 22 ............. 00 22

560

1000

1500-

20600

Ronge

-v
"C

......................................... a....
S
.

..............

7- . . .. . . . .. . . .. . . .
. .................................................

e 0S 20 .......................... ................. 1..


O ................................................................

.........................

',NMI

24

al I0 138 a3 34 to

S ..... .....................

.... a................................................ Frequency

0
00 U

8
UO

C .9 ..............................................................
.............
0
WMII 30

. B............ ............................ ...................................


....
0 0
40

..........................
0 a
40 52

0
U

0
4

0
3

34

Dog

Figure 6..._. Radar backscatter power medians with respect to range, frequency, and day for 12 UT of July. The data points are the same as in the corresponding month and hour of the preceding set of figures. For example, the median over range at 500 nm includes all data points between 500 and 600 run, regardless of day or frequency. The number of data points in the median is shown along the bottom.

61

S-ar putea să vă placă și