DECORRELATION TIME OF SPECKLE TARGETS OBSERVED WITH A HETERODYNE-RECEPTION OPTICAL RADAR by SUN TONG LAU B.S.E.E., State University of New York at Buffalo (1980) SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE MASTER OF SCIENCE at the MASSACHUSETTS INSTITUTE OF TECHNOLOGY June 1982 ............ Department of Electrical Engineering and Computer Science June 30, 1982 Signature of Author... Certified by...... //e'ffrey 1. Shapiro V Thesis Supervisor Accepted b . Arthur C. Sm th Chairman, Departmental Committee on Theses Archives OF TECHNOLOGY OCT 20 1982 LIBRARIES -2- DECORRELATION TIME OF SPECKLE TARGETS OBSERVED WITH A HETERODYNE-RECEPTION OPTICAL RADAR by SUN TONG LAU Submitted to the Department of Electrical Engineering & Computer Science on June 30, 1980 in partial fulfillment of the requirements for the Degree of Master of Science. ABSTRACT Coherent laser radars provide new technology for a variety of target detection and imaging scenarios. However, poor-image quality is caused by laser speckle resulting from the shortness of the laser wavelength compared to the surface roughness of typical targets. Serious signal return fluctuations are found whose time dependence is poorly understood. The purpose of this thesis is to assess the time dependence of speckle target radar returns. A data processing technique is developed to investigate the correlation property of the laser radar data. Accordingly, useful insights concerned with the causes of the return fluctuations are obtained. A mathematical model, which incorporates random radar and target tilts, is then constructed to describe the decorrelation process of the radar returns. Comparison of experimental results and theoretical results shows that atmospheric turbulence and wind are the factors which control the decorrelation process. Thesis Supervisor: Title: Jeffrey H. Shapiro Associate Professor of Electrical Engineering -3- ACKNOWLEDGMENTS I would like to thank my graduate counselor and thesis advisor Professor J. H. Shapiro for his patient guidance and invaluable advice during my studies at M.I.T. and the course of my thesis research. It has been my pleasure to work with him and learn so much from him. I also wish to acknowledge all members of the Optical Propagation and Communication research group at M.I.T. especially Dr. D. M. Papurt and T. T. Nguyen. Valuable suggestions from them have added to this work. Thanks are also due to members of the Opto-Radar Systems group at M.I.T. Lincoln Laboratory. R. J. Hull, T. M. Quist and R. J. Keyes should be mentioned for their help in providing me with the radar data and computer facilities for this research. Financial support by the U.S. Army Research Office, Contract DAAG29-80-K-0022 is gratefully appreciated. Elain Aufiero, Donna Gale and Deborah Lauricella deserve mention for their excellent typing. -4- To my mo.ZheL, 6o~'i heA etc.oLUWgQJ1Welt an'idLove. -5- TABLE OF CONTENTS Page 2 ABSTRACT.......................................................... ACKNOWLEDGEMENTS.................................................. 3 TABLE OF CONTENTS................................................. 5 LIST OF FIGURES................................................... 7 LIST OF TABLES.................................................... 9 CHAPTER I. INTRODUCTION....................................... 10 I.l. Laser Radar Configuration.................... 10 1.2. Intensity Fluctuation of Speckle TargetsProblem Statement............................ 11 Thesis Overview.............................. 16 STATISTICAL PROPERTIES OF THE INTENSITY FLUCTUATIONS....................................... 18 1.3. Chapter II. II.l. 11.2. 11.3. Scintillation-sensor/Radar Data Description............................. 18 II.1.1 Atmospheric Turbulence............... 18 11.1.2 Scintillation Measurements........... 19 11.1.3 Staring-Mode IRAR Data............... 21 Correlation Coefficient Function (CCF) Estimation................................... 21 11.2.1 CCF Estimation Procedure............. 22 11.2.2 CCFs in Various Turbulence Levels............................... 24 Chi-squared Goodness-of-fit Test............. 35 -6- CHAPTER III. MATHEMATICAL MODELING............ Page 44 Theoretical Model......... 44 III.1.1 Model Derivation. 45 111.1.2 Model Interpretat on 54 III.1 Model Verification........ 67 DISCUSSION....................... 84 111.2 CHAPTER IV. REFERENCES.... ......................... o...... 87 -7- LIST OF FIGURES Figure Page 1 Radar Block Diagram...................................... 12 2 Formation of a Speckle Pattern........................... 14 3 Autocorrelation Function of the Gate Function, g(t)...... 25 4 Estimated Autocovariance Function of x(t)................ 26 5 Estimated Correlation Coefficient Function of y(t)....... 27 6 CCF of Data Set 1........................................ 28 7 CCF of Data Set 2......... 8 CCF of Data Set 3......... ............................... 30 9 CCF of Data Set 4....... 10 CCF of Data Set 5......... 11 CCF of Data Set 6......... . ............... 33 12 Histogram of Target-Return Intensities vs. Expected Frequencies of Data Set 1.. 38 13 Expected Histogram of Target-Return . Frequencies of Data Set 2.. ........... Intensities vs. 39 14 Histogram of Target-Return Intensities vs. Expected Frequencies of Data Set 3.. Intensities ...............vs. ............... 40 15 Histogram of Target-Return ............... Frequencies of Data Set 4.. Intensities vs. Histogram of Target-Return ............... Frequencies of Data Set 5.. Intensities vs. Histogram of Target-Return ............... Frequencies of Data Set 6.. 16 17 29 31 .. I. .. . .. .. .. .. ... .. . .. . .. Expected ............. ...... . 32 41 Expected ............... 42 43 46 18 Radar Configuration........ 19 Theoretical CCFs with Only Radar Tilt Active............. 57 -8- Page Figure 20 Theoretical CCFs with Only Target Tilt Active............ 58 21 Theoretical CCFs with both Radar and Target Tilts Active,R > 1 and a; > G .. ......... ..................... 60 22 Theoretical CCFs with both Radar and Target Tilts Active,R > 1 and a0 = a . ............................... 61 23 Theoretical CCFs with both Radar and Target Tilts Active, R > l and au < a. ............................... 62 24 Theoretical CCFs with both Radar and Target Tilts Active, R < 1 and a < a .......... ............ T e....... 64 25 Theoretical CCFs with both Radar and Target Tilts Active, R < 1 and a; = a . .......... .......... o........... 65 26 Theoretical CCFs with both Radar and Target Tilts Active, R < 1 and a; > a'. 66 27 Theoretical CCF vs. Experimental CCF of Data Set 1. 71 28 Theoretical CCF vs. Experimental CCF of Data Set 2. 72 29 Theoretical CCF vs. Experimental CCF of Data Set 3. 73 30 Theoretical CCF vs. Experimental CCF of Data Set 4. 74 31 Theoretical CCF vs. Experimental CCF of Data Set 5. 75 32 Theoretical CCF vs. Experimental CCF of Data Set 6. 76 33 Best CCF Fit of Data Set 1...... 77 34 Best CCF Fit of Data Set 2...... 78 35 Best CCF Fit of Data Set 3....... 79 36 Best CCF Fit 37 Best CCF Fit of Data Set 5....... 81 38 Best CCF Fit of Data Set 6...... 82 of Data Set 4...... 80 -9- LIST OF TABLES Table Page 1. Scintillation Measurement Results.................. 20 2. Decorrelation Data................................. 34 3. Parameters for Figures 21, 22, 23 .................. 63 4. Parameters for Figures 24, 25, 26 .................. 67 5. Estimated ae and - 6. Parameter Values for Best CCF Fit.................. 83 from Turbulence Theory......... 69 -10CHAPTER I INTRODUCTION Heterodyne-reception optical radars using the 10.6 -rm wavelength CO2 laser provide new technical options for a variety of target detection and imaging scenarios [1]. However, the much shorter wavelength of laser radars as compared to microwave radars implies new problems as well as enhanced capabilities [2]. One of the problems is the poor image quality which is caused by laser speckle [3], resulting from the shortness of the laser wavelength compared to the surface roughness of typical targets. Serious signal return fluctuations are found whose time dependence is poorly understood. This research will be addressed to assessing the time dependence of speckle target radar data analysis and theoretical modeling. The remainder of the introduction includes a description of the optical radar we are using, a problem statement, and an overview of the thesis organization. I.1: Laser Radar Configuration An ongoing program aimed at developing an Infrared Airborne Radar (IRAR) is underway at the M.I.T. Lincoln Laboratory [4] [5]. A radar testbed system has been constructed as part of this program which we will refer to as IRAR, although it is ground mounted. IRAR. Data for this thesis has been obtained using This laser radar uses a one-dimensional, twelve-element HqCdTe hetero- dyne detector array, a transmit/receive telescope of 13 cm aperture, and a 10 W CO2, 10.6 um laser, which is operated in pulsed mode. Presently, we are interested in the radar's performance over target ranges from one to ten -11- kilometers. More radar system descriptions can be obtained in [2] [4] [5]. In order to set up subsequent statistical system analysis, the basic structure of a heterodyne-reception optical radar is explained. represented by the block diagram of Figure 1 [5]. It can be The laser radar sends out a series of pulses and illuminates a target located a certain distance away. After passing through the exit optics in Figure 1, the laser beam propagates through the atmosphere and the illuminator beam is reflected back by the target. The reflected beam then comes back through the atmosphere and the entrance optics. Finally, the received beam is combined with the strong local oscillator beam operating at a frequency offset v IF on the surface of the photodetector. In target-detection applications, the IF signal is to be used to estimate the average target reflection strength which is compared with a threshold value to determine the presence or absence of a target. In performinq imaging, the radar first scans the target and collects arrays of echo signal returns in order to form a complete picture. Then, computer enhancement of the resulting image follows, after the estimation of the average target reflection strength is finished. 1.2: Intensity Fluctuation of Speckle Targets - Problem Statement The random intensity distribution that we call a speckle pattern is formed when fairly coherent light is either reflected from a rough surface or propagates through a medium with random refractive index fluctuations [3]. Since the wavelength of the CO2 laser is 10.6 -pm, many target surfaces are rough on the order of a wavelength. As a result, the surfaces scatter the light diffusely and form a speckle pattern. The observation at a distant point is simply the summation of the light from a large number of randomly phased TRANSMITTER BEAM TRANSMITTER LASER EXIT OPTICS BEAM COMBINER PHOTODETECTOR ILLUMINATOR BEAM ATMOSPHERIC PROPAGATION PATH RECEIVED BEAM EN T RANCE OP TICS REFLECTED BEAM A TMOSPHE RIC PROPAGA TION PA T H PHOTOCURRENT L.O. BEA M IF LOCAL PROCESSING OSCIL LA TOR TARGET IMAGE PROCESSING Figure 1: Radar block diagram. T G -13- scatterers as in Figure 2. Obviously, if the point of observation or the precise position of the target being illuminated is changed, the signal return intensity will change at the same time. Statistically, the speckle fluctuation obeys an exponential probability distribution p(I) = exp [- u(I) where I = signal returns intensity <I> = ensemble average intensity. It can be seen that speckle in infrared radar is an analogy to the Rayleigh cross-section fluctuations in a conventional radar. Intensity fluc- tuations of the magnitude associated with the exponential distribution create serious problems in imaging. However, averaging several independent image frames will result in significantly better overall image quality. The question then becomes one of obtaining independent image frames, i.e., of determining the decorrelation time for the speckle process. The exponential distribution cited above rigorously applies to the target-return intensity flucutations over an ensemble of macroscopically identical rough-surface targets. The exponential distribution has been verified experimentally by Papurt [7] via spatial sampling of the speckle fluctuations in IRAR images obtained form a large rough-surface target of uniform average reflectivity. In this spatial sampling, the target returns from non-overlapping illumination regions on the surface are independent samples from the exponential distribution. Because the use of spatial averaging to reduce speckle fluctuations in a laser radar image will necessarily entail a loss in spatial ROUGH OBSERVATION SURFACE POINT Figure 2: Formation of a speckle pattern. -15resolution, it is important to study the time averaging of intensity returns reflected from a single spot of the speckle target. To probe the time-averaging issue the radar can be operated in staring mode, that is with its scanning capability disabled. If the target, the radar,and the intervening propagation medium are perfectly rigid, then the radar will stare at one spot on the target and there will be no time dependence to the intensity returns. Prelimin- ary staring mode IRAR measurements have shown, however, serious fluctuations of the intensity returns in time. It is important to know the time correlation properties of these fluctuations since they will impact radar performance, e.g., the use of frame averaging to reduce speckle fluctuations in the radar image requires inter-frame time separations that are longer than a coherence time. Also, the contributing factors for the staring-mode fluctuations are not known yet. It is significant to see how these factors affect the decorrelation process as it may help guide future improvements in the radar system. major objectives of this thesis are two fold. The First, to investigate the time correlation properties of staring-mode speckle target intensity fluctuations. Second, to explore the causes of the decorrelation mechanism. For the simple geometry in Figure 2, the causes of the decorrelation process are probably the atmospheric turbulence effects along the laser propagation path, and the wind induced vibrations of the IRAR equipment and the target. Staring-mode IRAR data will be used to study the decorrelation process. Simultaneous scintillation-sensor measurements will be used to estimate atmospheric turbulence levels. Thus, we shall be able to compare the time correlation properties of radar data taken in various turbulence strengths. To properly account for the speckle target intensity return fluctuations, the probability density function and the correlation coefficient function in -16- time should be known. The former furnishes information concerning the properties of the intensity fluctuations in the amplitude domain, whereas the latter describes the degree of correlation of the data in the time domain. In the latter case, we will determine the decorrelation time, i.e., the time it takes for the data to become uncorrelated, from IRAR Using this decorrelation time, a collection of uncorrelated measurements. samples will be extracted from the IRAR data and compared with the exponential ensemble statistics predicted for the former case. In support of the data examination, a mathematical model is developed to describe the decorrelation process. This model assumes free space propagation with random radar aiming errors and random target tilts. These statistical quantities may represent turbulence-induced phase tilts whose strengths can be estimated from turbulence theory using the scintillation measurements. With these estimated values the predictions of the decorrelation model will be compared with experimental results from the radar data. 1.3: Thesis Overview In Chapter 2, we begin with a complete description of the radar data format. Then the procedure for estimating the correlation coefficient function is explained. Correlation results based on IRAR data taken in various atmospheric turbulence levels are presented. Finally, a chi-squared goodness-of-fit test for the exponential distribution is performed on the radar data. Overall, the results of the data manipulation give us -17- some insight into the nature of the intensity fluctuations. In Chapter 3, our model for the decorrelation process is postulated and analyzed. We shall exhibit the behavior of the model as its parameters are varied. The correlation coefficient predictions of the theoretical model are then compared with the experimental radar results of Chapter 2. Chapter 4 contains a discussion of the target return time dependence as understood from our experimental and theoretical results. -18CHAPTER II STATISTICAL PROPERTIES OF THE INTENSITY FLUCTUATIONS This chapter is devoted to our experimental efforts aimed at understanding the decorrelation process. We begin with a brief discussion of atmospheric turbulence and a summary of the scintillation sensor data. format of staring-mode IRAR data is then described. The Next, we shall explain the detailed procedure for estimating the correlation coefficient function (CCF) of the IRAR data. Subsequently, CCFs turbulence levels are presented. for six data sets taken in various Finally, chi-squared goodness-of-fit tests to the exponential distribution are performed on the six data sets. II.1: Scintillation-Sensor/Radar Data Description Radar data taken from the IRAR has been investigated in order to understand its basic statistical properties. It was taken in various turbulence levels with scintillation measurements made simultaneously. Theory for wave propagation in the turbulent atmosphere has been well established during the past decade [8] [9]. In order to provide pertinent information relevant to our research, turbulence effects on laser propagation in atmosphere is introduced first. Second, we shall explain the scintillation measurement and its implications. Then, the radar data description is given at the end of this section. II.1.1: Atmospheric Turbulence Atmospheric turbulence refers to the refractive index fluctuations -19- which are due to turbulent mixing of air parcels of nonuniform temperatures in clear weather conditions. These air blobs will dephase an optical wave, hence causing transmitter beam divergence and receiver angle-of-arrival fluctuations. Also, the random lensing of the wave by the turbulence leads to constructive and destructive interference, i.e., amplitude fluctuations, called scintillation. These effects on laser propagation were described for a time independent medium. In the atmosphere, since the array of turbulent eddies tend to drift with the nominal wind velocity. Consequently, the turbulence has a typical coherence time tc of 10-3 to 10-2 seconds [101. We strongly suspect that turbulence effects are the prime factor controlling decorrelation time in staring-mode speckle-target measurements. As we go along, the intuition from the data manipulation and the statistical modelling should help justify this statement. 11.1.2: Scintillation Measurements The turbulence strength along the atmospheric path between IRAR and the speckle at a particular time can be estimated from the amplitude fluctuations (scintillation) of laser pulses that have propagated over this path. To perform this scintillation measurement, two lasers, CO2 and GaAs, were located next to IRAR with their receivers located one kilometer away adjacent to the speckle target. Data acquisition equipment and data processing programs have been developed at Lincoln Laboratory to produce good estimates of the turbulence strength parameter, Cn2 [5] and the atmospheric coherence time tc from the received CO2 and GaAs laser pulse streams. Six sets of scintillation data were taken while staring-mode IRAR measurements were being made. The resulting -20- Cn and tc are summarized in Table 1, where they have been ordered according to their turbulence strength. TABLE 1 Scintillation Measurement Results Data Set Number Cn2 m-2/3) tc (ms) 1 0.95 x 10l 4 39 2 0.87 x 10-13 23 3 0.107 x 10-12 45 4 0.13 x 10-12 16 5 0.2 x 10-12 26 6 0.34 x 10-12 52 -2111.1.3: Staring-mode IRAR Data The IRAR data was taken in staring mode using the Lincoln Laboratory flame-sprayed aluminum speckle-target calibration plate at one kilometer range. The pulse-repetition frequency of the radar is 18.9 KHz, ing to an inter-pulse time interval of approximately 52 vi staring mode IRAR data is recorded in frames of pictures. sec. correspondEven in Each frame has 128 by 60 picture elements (pixels) which are linearly proportional to the return strengths of the associated laser pulses. Unfortunately, the data is not taken in a completely continuous manner, thus giving rise to some difficulty in computing the statistical properties such as the CCF. details were as follows. The Each frame has 60 active lines of 128 data points each plus 82 missing data points because of hardware mechanics. Because of the periodic missing information, we had to formulate a procedure to estimate the CCF from such an intermittent structure. Discussion of the CCF estimation procedure forms the essence of this chapter. It should be noted that the data which we deal with is the square of the IF signal envelope, because the squared envelope is proportional to the return light intensity. 11.2: Correlation Coefficient Function Estimation The CCF for a wide-sense stationary random process y(t) with auto- covariance function Kyy(v) is CCF(v) = Kyy(v) / Kyy (o) It is well known that CCF(v)I < 1 with ICCF(v)| = 1 when y(t) and y(t + v) are completely correlated, and CCF(v) = o when y(t) and y(t + v) are -22uncorrelated. The decorrelation (or coherence) time of the process y(t) can therefore be defined as the time it takes for CCF(v) to drop from one to zero. 11.2.1: CCF Estimation Procedure If the data were continuously spaced, a direct method could be employed to compute the correlation coefficient function (CCF). Unfortunately, because of the regularly missing observations in the radar data, a special formulation had to be developed. Let y(t), t = integer, be the discrete time stationary process representing the data that would be gotten were there is no regularly missing observations. Let g(t) be the periodic gate function with period a + 3 g(t) ={ 0 t = a + 1 + + where a = 128 and B = 82. If x(t) denotes the actual data, we can write x(t) = g(t) y (t) We are interested in the CCF of the random process y(t), which is CCF(v) = Kyy(v) / Kyy(o) where Kyy(v) is the autocovariance function of the process y(t) and the latter is assumed to be wide-sense stationary. We shall use as our estimate -23of CCF(v) the function CCF(v) = Kyy(v ; T ; N) / Kyy (0 ; T ; N) where Kyy (v ; T ; N) is a covariance function estimate based N data streams of length T obtained as described below. Consider the following estimation equation, =1 A T-vV vF x(t) - my(Ti)g(t) t=l L aV-- Kxx(v ; t) = 'x(t+v) - my(T)g(t+v) where Kxx(v ; T) = estimated autocovariance function of x(t) at lag v based on a T-length data stream, T T and m (T) = E x(t) t=1 / E g(t) t=l is the sample mean of all the non-zero data points. It is well known that my(T) is an unbiased consistent estimator of m the mean of the process yy y(t) [11]. Thus, if T is large we can use m (T) ~ m in Kxx(v then easily shown that T-v E[K xx (v ; T)] E T~~~yt 1 y(t) - mY )g(t)g(t T-V T = g(t)g(t + v) E = R (v)K gg (v), yy + v)[y(t + V) - m ++v-my]j y(t) - my (y(t + v) - m T). It is -24T-v where Rgg(v) Z q(t) g(t+v) is the autocorrelation function of the gate t= 1 g(t). It can be shown that Rgg(v) is given by [61 ,9for v=o, ... for v=, ... ,-for v=a, .. , ,a6 which is Dlotted in Figure 3. It follows from the above that Kxx(v;T) / Rqg(v) is an approximately unbiased estimater for Kyy(v) for any T-length data stream that is long enough to ensure my(T) ~ my. The stability of this estimator depends on the stability of Kxx(v;T), which will be good for v<<T and poor forvZ~T [12]. Improved stability can be obtained by taking Kyy (v;T;N) to be the sample mean of N Kxx (v;T) / Rgg(v) estimators obtained from N different T-length data streams. Our CCF estimation algorithm generates my(T) and Kxx(v;T) / Rgg(v) for N = 10 pictures each with T = 128x60 pixels. These were averaged together to yield Kyy (v;T;N) and CCF(v) = Kyy (v;T;N) / Kyy(O;T;N). For the lag values of interest it was found that averaging the 10 pictures together gave satisfactory stability. Typical examples of Kxx (v;T) and CCF(v) are given in Figures 4 and 5, respectively. 11.2.2: CCFs in Various Turbulence Levels The estimated CCFs for our six data sets are shown in Figures 6-11. -25- 0. 60 Rgg ,. 0.360 S. 3" 0.150 0. 1*. 0*. 3". 400. s*. 6*. 7". a". 9*. PULSES Figure 3: Autocorrelation function of the gate function, g(t). 10*. -26- I II Ii tII III lI t I i i I lillIll lilii II 5111191 I I 7.5 E+7 Kxx 5.ees*O7 a.seE+r7 0.O S. Ie. ae. 30. II 1191111111111111111 IIlI 1111111 I||i 4"0. 55. a"0. 750. 11111 Illill 350. PULSES Figure 4: Estimated autocovariance function of x(t). I 950. 1SM. -27- [[ii Iii ii i 511i 115 11i11 II ilil i ll Iii I i i ii III 1111511 I.00 0.90e CCF 0.40 0.30 *.ao 0.10 0.e 4.10 0. 100. a**. 300. 400. 500. 600. 700. 900. N00. PULSES Figure 5: Estimated correlation coefficient function of y(t). 1000. -28- CCF 0.O5. I I I I I I I I I I I I I I I I I I I Ii i aese. Is"*. 1"s. see. . PULSES Figure 6: CCF of Data Set 1. I I t. 2500. 30*. -29- -- -- r I I i I I I i i I T i 0.75 CEF 0.S 6. See. 1290. 10. 2OO. PULSES Figure 7: CCF of Data Set 2. 25e0. 3000. -30- ~ I~~T ~ I o .50 I I T I I ~~ I II I I I I I ' T .. CCF * .as 0.*5. I I I I I I I .Se. I I I i*. I I I I ise.. as PULSES Figure 8:. CEF of Data Set 3. s25*. 3000. -31- a.-IS e.so . CCF 0.S L . L I I. I I S. I I see. I I I iee. I I I I isee. ee20. PULSES Figure 9: CCF of Data Set 4. asee. 3900. -32- 0.75.- CCF as. 0.8 L .L I L I S.ao0. L I I I I I I I I I...I IS". 50. 2000. PULSES Figure 10: CCF of Data Set 5. a50e. . . 30". -33- 1.00 0.7S O.Se ........... CCF *.25 0.9 - .l11. 0. L1.I 5". LL IIII lose. 1 111 ise. PULSES Figure 11 : CCF of Data Set 6. I 111 11111 ass. 3000. -34- For each figure we have computed the decorrelation time using the 52 vsec pulse spacing. The results are given in Table 2 along with the Cn2 values from Table 1, and the weather description recorded by the IRAR operators. TABLE 2 Decorrelation Data Data set no. 1 Decorrelation Time (ms) 156 Cn2 -2/3) Weather 0.95 x 10-~14 Haze, Overcast 2 65 0.087 x 10-12 Hiqh Solid Cloud Cover 3 78 0.107 x 10-12 Partly Sunny 4 39 0.13 x 10-12 Clear, Sunny 5 39 0.2 x 10-12 Clear, Sunny 6 52 0.34 x 10-12 Clear, Sunny Two interesting points we can easily observe are: 1) Data Set 1 was taken in the weakest turbulence conditions, i.e., haze and overcast. It took 156 ms for the CCF to drop from one to zero which implied that the data were highly correlated. In other words, the intensity return did not fluctuate very much in this data set. 2) The remaining data was taken in more or less the -35same turbulence level, since the Cn2 values differed only slightly. On the other hand, the decorrelation times for Data Sets 2-6 varied from 39 to 78 ms. An immediate implication of the first observation is that the intensity return fluctuations depend upon the atmospheric turbulence strength very much. In weak turbulence, the atmosphere is just like a "frozen" medium. Therefore, the intensity returns stay constant for relatively long time periods. Conversely, the intensity returns start to fluctuate more as the turbulence strength gets stronger. The second observation leads us to suspect the other contributing factor, which is wind speed. As we shall see in the next chapter, wind speed in fact has an effect on the intensity return fluctuations. 11.3: Chi-squared goodness-of-fit test The ensemble and spatial-sampling statistics of speckle-target radar returns obey the exponential distribution. In this section we shall use our six data sets to examine whether exponential statistics apply to staring-mode target returns from a speckle target. To make a quantitative assessment we will use a special type of hypothesis test called the chi-squared goodnessof-fit test, which is widely employed to test the equivalence of a probability density function of sampled data to some theoretical density function. Since the decorrelation time for each set of data is known from the previous section, independent samples can be obtained. We first provide a brief description of the test and then give the test result in the sequel. -36Consider N independent observations from a random variable x whose probability density function is p(x). Let the N observations be divided into K intervals to form a frequency histogram, where f. denotes the observed frequency in the ith interval. The number of observations which could be expected to fall within the ith interval if the true probability density function of x were p0 (x) is called the expected frequency, F . To measure the discrepancy for all intervals, a chi-squared value is computed via 2 K' (f. - F.)2 X = F. where K' is the number of intervals in which the expected frequency is higher than or equal to five. In other words, intervals in which F. is smaller than five are combined to form one interval. The number of degrees of freedom n is equal to K' - r - 1 where r is the number of parameters estimated from the data for the hypothesized distribution. Having obtained x2 and n standard statistical tables will provide a corresponding level of significance a which indicates how good the fit is. Generally a value of at greater than or equal to 0.05 is regarded as verifying the theoretical distribution. Further details about the test can be found in [12]. For our case we use the exponential distribution -1 -x/x x > 0 p (x) = 0 otherwise where the mean x is set equal to the sample mean of the data. A sample of N = 210 independent observations was used for each of the six data sets. -37The return values range from 0 to 65025, which is divided into K = 17 intervals and r = 1 because x has been matched to the sample mean. A com- puter program was written to perform the test, with the results given in Figures 12 - 17. In each figure, the bar chart is the histogram of observed frequencies, and the curve is the exponential density fit to the sample mean. It is not an easy task to explain our results, however some useful comments can be made. Data Set 4 has the best fit to the exponential distribution while the others do not fit as well. The fact that Data Set 1 has the worst fit enhances our CCF estimation result ; there is very little randomness in this data set, which was taken in the weakest turbulence. It further convinces us that atmospheric turbulence indeed is an important contributing factor to the return fluctuations, because the atmosphere acts like a "frozen" medium in weak turbulence. -38- , n = 6 x2 = 43.59 100. I I I II II I Ii II ef=.0 ill I II I I II 90. B0. 70. w C-, C 60. S.U 5. 4C S.- 45. 30. 2s. 19. 5. .1.. -o a ~ -~- E.5 5.0 7.5 10.0 12.5 IS., Target-return intensities Figure 12: Histogram of target-return intensities vs. expected frequencies of Data Set 1. -39- x i oe. I 2 I 5 18.11 = I I I , I I n = 6 I I I , I a _=. 0.005 I ~ I I I I I I I I I 90. 80. 70. V.) S.- C) 50. 4- E 39. 29. Is. 0. inL..L w.w LL 2.S 7.5 is.. 12.5 5.e Target-return intensity Figure 13: :Histogram of target-return intensities vs. expected frequencies of Data Set 2. -40- x2 = 24.8 , n = 12 '. 0.02 , 1le. -r 90. 80. 70. w U a, 60. S.- U U So. C S.- a, E 40. 38. 80. L I 9. a C.sm -M b. I 7.S 10.1 12.5 is. Target-return intensities Figure 14: Histogram of target-return intensities vs. expected frequencies of Data Set 3. -41- , n = 8 x2 = 10.91 S I I I I I~ I I I I a , I I I . 2 I I I I I I I 90. 80. 70. a?) U 60. C--) So. E 40. .P 3,. . .A 29. 1. ............ 6. .0 2.5 5.0 7.5 1.@ 12.5 Is.0 Target-return intensities Figure 15 : Histogram of target-return intensities vs. expected frequencies of Data Set 4 -42- x2 = 15.25 , n = 6 , a '=.0.02 100. popI I I ;I I I I I I I I i 90. 80. 70. (j~ w U w 6e. U U se. 0 w -o E 4e. 30. in.L L .. a.5 2S. je. S. S.. 7.5 10.6 12.5 IS.0 Target-return intensities Figure 16 :Histogram of target-return intensities vs. expected frequencies of Data Set 5 -43- x 100. iI I 2 , n = 10 =12.21 I I I I I I I I , a '=. 0.03 I I I I I I I J I I II 90. BO. 70. 60. U U 0: So. 40 E .. 40. 39. 20. .. to. K 9., 2.s S., 7.5 19.0 12.5 is., Target-return intensities Figure 17: Histogram of target-return intensities vs. expected frequencies of Data Set 6 -44CHAPTER III MATHEMATICAL MODELING In this chapter we will report on our model for the time dependence of the staring-mode intensity return fluctuations. The model ascribes the time dependence to random tilts in the radar and target planes. Our first step is to derive a theoretical CCF for staring-mode measurements from the model. Next, because our experimental results have convinced us that atmospheric turbulence is the major cause of the intensity return fluctuationswe use turbulence-induced tilt quantify the.CCF model. experimental CCF results. standard deviations to The model predictions are then compared with the As we shall see, very interesting and significant result is found. III.1: Theoretical Model We will model the random radar motion by a random aiming angle error e(t), and the random target motion by a random tilting angle (t). As a result, in the analysis that follows, the transmitted beam and received beam complex envelopes will include the phase term exp j I(t) - T and the target reflection process will include the phase term exp j .LT(t) - P' -45- as shown in Figure 18. The remaining pieces of our radar model parallels that employed in [5], with continuous-wave laser operation and far-field free space propagation assumed. III.1.1: Model Derivation 1. Let u1 (p,t), the complex envelope of the transmitted laser beam, be given by, u(Pt) = (PT) exPp (t) circ2id where PT is the laser power, d is the diameter of the exit pupil, p is the displacement vector on the radar plane, and li(t) is the random aiming error. 2. Let u (P',t) be the transmitted beam complex envelope as it arrives at the target L meters away from the radar. Fraunhofer diffraction theory gives us the result, exp t7L 1!2 (Pt)= + 1P 12J jxL dp u pt - exp- j L'-p where p' is the displacement vector on the target plane. 3. Let u3 (p,t), the reflected beam complex envelope at the target plane, be given by, T(p', t) = UI(, texpp (')exp (t) p' -1 Speckle Target Radar N L.O. Beam Figure 18: Radar Configuration -47- u3(P' t) = + angle, (t) is the target tilt where F ()exp t g2(W' (t) and T (p') is a rough surface complex-field reflection function. 4. The complex envelope, u4(p, t), resulting from propagation back from the target to the radar plane obeys d} d 5. 3( 9 ', t - jL L + 4(p, 't) = exp j exp - i . - Finally, the intermediate frequency (IF) signal has complex envelope = I.t) d - ci rc(2. p 1/2 exp where e(t) is the aiming error incurred on reception. Combining the above equations, and using Gaussian beams instead of circular beams to simplify the integrals, we get y(t) =- exp j L 2 -48- - dp' T (W')exp X (t - ) 7 2 2 -2 exp -d -exp - t) 2( 2 21TXL 2j'~t 12] ' exp Obviously, the first exponential term in the integration is the moving part of the target reflection model. The second and the third terms are the randomly displaced transmitter beam pattern and the back-propagated local-oscillator pattern, respectively. Our task is to calculate the correlation coefficient function with this IF signal model by assuming statistical properties for T (p') , 6(t) and (t). From this calculation we will be able to see how e(t) and I(t) affect the decorrelation time of the signal intensity fluctuation. Let CCF(T) be the correlation coefficient function, that is CCF(t) = < y_(t + T)I2 <|y(t) |y_(t) 1> - < y(t)j2,2 4 2 2 >- <ly_(t)I > Since there is a great deal of tedious algebra involved in deriving CCF, we shall only present the key results here. -49The expected value of the IF envelope intensity with respect to the target ensemble is ~ 2 ( ')d <y(t)2T -- where T (p') ~P sT Sd P 0 P2L d ,t2 JL t - -e(t)|2 2 X has been taken to be a pure speckle target model, i.e., T is a zero-mean circulo-complex Gaussian process with correlation function [5] <T (P)T*(p2)> = 2Ts 6(P1 - P2 with Is being the average intensity reflection coefficient of the surface. Assume e(t) is a zero-mean stationary vector Gaussian process with independent identically distributed components whose: autocovariance function is z(T) = K <ly(t) 2> 2) =><< yS (T) = K y 2 TT (p' -0 (k). 2 From this result, we can see that if decorrelation time of z(T), then We then find that, sT 2L _ + 2T 2d 2 ~z(O) - z _ . is short compared with the -50- P Tsd2 2 <1y(t)K> 2L2 In other words, if F(t) stays constant over times comparable to C or longer, the average signal intensity return will be unaffected by the Note that the average signal intensity is random aiming error ~(t). always independent of 6(t), the random tilting angle of the target. The next step is to calculate the quantity, <_y(t + T)1 2 jy(t)j 2>, to be followed by the correlation coefficient function. Averaging with respect to the target ensemble we get 2 -t) 4~dP T2~6t- <ly(t+-) 2 1iy(t) 2 > T s -exp -27r2d)2 4L 4 2 + c 2 + - exp {r2d2 eF 2X Td) Ts 4 t+d2T- -2 L C 4 - 4L t+- t ; 4L 2 _L t+--- <t + - + +TI + T)|2 } - MWWAMQ41 "_ .11, -51- (t) ensemble, assuming f(t) to be a zero-mean Next, we average over the stationary vector Gaussian process that is statistically independent of i(t) and has independent identically distributed components whose autocovariance function is z'(T) = K (T) = K The result we (T). obtain is t - S exp>272dd 2L <<«y(t+)I!y_(t) 2> T 04L~ d 4p 2LI 2 2 SdP T s L - 0 t e - ) 2 [z'(o) - z'(T)] + 1 d2j To find 2 2t) 2 (t) +6(t+T) X + + -C 2 ++ t+T- -6(t+T) 2X 2X 2L + t+T + 2d22 T - L 16 L 2 + A4 F1 t -exp{ d2 Tit - X 2 -52- <Iy(t + T)I2 Iy_(t) I2> = <<<y_(t + T) I 2 yt)2 let Fx(t) ex (t + T) tx t + T denote a zero-mean Gaussian random vector with covariance matrix z(0) (2 L~ zw zt z(0) z 2L z(T) z f IC - + z(T) z(0) z(2L z [?qL z(0) A z(T) Z - TJ z r + z(T) We then have that <jy(t + T)J 2 y(t) 1 2 > = <<<y(t + _) 12 y(t)1 2>T>_>_ T) -53- T2 d 4 p T I2 4L4 l+ I+22AEl I+2A C - 162L [z' (o) - z'(T)]+1 where 3 -1 0 0 07 3 0 0 3 0 (rd~ 2 XJ) 1 -2 1 4? -1 0 0 1 0 0 0 o 1 -1 0 o -1 1 1 -1 and E = 'w 2 2~ 3 Finally, combining the preceding results for <Iy(t)1 2 > and <1y(t + T)| 2 [y(t)jz> we get our predicted CCF + I+2Aw(T)EI I 1 ()+ + 2Tr 2 d 2z I+2AW(T)Cl[ d2(z'1(o)-z ' (T))+1] CCF(T) = (o)E )-z 1 + II+2 I 2 JI+ 2Aw(o)CI} 1 2 [z(o)-z( L) + 2Tr[d c , 2 2 J ]c -54- A substantial simplification results if decorrelation time of z(t). c is much smaller than the As our experimental results show decorrelation times of many msec and 2L/c is about 6 ipsec for our data sets, we will set 2L/c = 0 and use the simplified form CCF(T) = L(z(o)+z([)) (z(o)-z(-)) + 1 a 16d 2L2 1z (O) 111.1.2: + 4(z(o) - Z'(T)) + 1 Model Interpretation To examine the implications of our model we shall assume that the tilt autocovariance functions have the following forms: Z(T) and = T2 e e -55- -T 2 /TL z'(T) = a where Y and ay e are the standard deviations of the random radar aiming error (radar tilt) and the random target motion (target tilt), respectively and Te and -c are the decorrelation times of the radar tilt and target tilt, respectively. We would like to see how the predicted CCF behaves as a function of the dimensionless parameters Trda 4La a T/Tr 6 and T/T . Intuitively, if measurements more than T d N > 1), radar ac 6 e is larger than Trd ( sec apart are likely to illuminate essentially independent portions of the target surface. % > 1 (ay is larger than By the same token, if ), the radar will be likely observed statistically independent target speckle patterns at time separated by more than T sec. Let us first consider the behavior of CCF when only one tilt mechanism is active, i.e., we shall plot CCF vs. T/Te when z'(T) = 0, and CCF vs. T/T4 when z(T) = 0. With these curves, we can compare the CCF decorrelation time with the decorrelation time of each random tilt. -56- z'(T) = 0 Case i. In this case, we have that CCF(T/T ) = L which has been plotted in Figure 19 for a = These curves show 1,3,5. that CCF(T/T ) decorrelates faster as a2e' increases, and for the range e of a 2 , shown the GCF decorrelation time is appreciably faster than T. e e Case ii. z(T) = 0 In this case, we have CCF(t/ = /E 4 - e }+ which has been plotted in Figure 20 for a ] = 1,3,5. CCF decorrelation time decreases as the normalized tilt Once again the angle variance increases, but compared with the previous figure we see that CCF decorrelation time is larger for the same tilt variance and decorrelation time. -57- 0.80 0.70 e.Ge 0.60 CCFs 0.50 0.40 a 0.30 - 3 e~ae . 0.0 0*. 0.10 6.20 0.30 0.40 0.s 0.60 0.70 0.80 T/Th Figure 19: Theoretical CCFs with only radar tilt active. 0.90 1.60 -58- *so 0.70 0.60 CCFs 0.40 0.30 0.*20 6.10 0.0 0.16 8.20 8.39 0.40 0.50 0.60 0.70 0.80 Figure 20 :Theoretical CCFs with only target tilt 0.9, active. 1.00 -59- Now let us examine how CCF behaves when both tilt mechanisms Here it is worthwhile to define R = T /1e and to are present. distinguish between R > 1 (radar tilt decorrelates more rapidly than target tilt) and R < 1 (vice versa). In the former case, we will plot CCF vs. T/T6; in the latter case we will plot CCF vs. T/T . Case iii. R > 1 In this case, we have 71 CCF(T/Te) = 4(L '' 1- 1+e e{ L - +4(a )2 I {e _ J Je + 1 1 0 (a 127) e ( 2 R +1 Figures 21, 22 and 23 give CCF vs. T/Te for the parameter values shown in Table 3. -60- 1 .00 0.90 0.80 R= 1 0.70 a 0.60 = 0.1 0.50 R= 3 CCFs = 3 a '=0.3 0.30 0.20 R= 5 a ' 0.5 0.6 0..0 0.10 *.29 0.36 0.40 *.S* T/T 0.60 0.70 0.80 0.90 6 Figure 21: Theoretical CCFs with both radar and target tilts active, R > 1 and ae' > 1.00 -61- o.90 0.80 Rl R =1 0.60, e.so CCFs R 0.40 6.30 * 3 = a' 3 a 3 .ae. R= 5 a 0.1. 0.a' 5 =5 0.0 0.10 *.80 0.30 0.40 *.S@ 0.60 0.70 0.80 . T/T 6 Figure 22: Theoretical CCFs with both radar and target tilts active, R > 1 and a6 ' = a('. 1.0 -62- 1.00 0.90 0.80 R=l a ' e e. 70 a '=5 o .60 0.50 CCFs 0.40 R= 3 a '=3 0.30 a '-15 R= 5 1 *~ * 0. 2s 0.S 5 a =20 0.10 0.20 I .0 0.30 0.40 *.S 0.66 0.70 0.86 e.g. T/T e Figure 23 : Theoretical CCFs with both radar and target tilts active, R > anda < 1.00 -63- Table 3: Parameters for Figures 21, 22, 23 Ratios a6 21 1, 3, 5 1, 3, 5 22 1, 3, 5 1, 3, 5 1, 3, 5 23 1, 3, 5 1, 3, 5 5, 15, 20 Figure Number 0.1, 0.3, 0.5 R < I Case iv. In this case, we have CCF(T/T ) F {fl2] 4(a { 1 )4 1+ V N) {1 ei~2 1 R2 +4(a')2 { - e $ 2 +7 2 Figures 24, 25 and 26 give CCF vs. in Table 4. e +1 T/T for the parameter values shown -64- 1.00 R = 0.1 a= 0.1 0.80 0.70 R = 0.2 ' = 0.3 0.60 .a CCFs 0.40 6.30 R = 0.3 a' - 0.10 - 0.0 , 0 0. = 0.5 Q'= 5 l I 0.10 I I 0.20 I I 0.30 I I 0.40 6.S0 6.60 0.70 0.80 0.90 Figure 24: Theoretical CCFs with both radar and target tilts active, R < 1 and a ' < U' 1.00 -65- 1.00 R 0.89 0.1 R = 0.1 0.70 0.60 CCFs 0.se 0.40 R = 0.2 ' = 3 0.30 *.20 a' = 3 0.10 R =0.3 -- . 0. 0 0.10 0.20 ae' = 5, (a' I 0.30 0.40 5 I 0.50 6.60 I I -I 0.70 I 0.86 A- I 0.90 1.00 Figure 25: Theoretical CCFs with both radar and target tilts active, R < 1 anda ' = -66- 1.00 I I I I I I I I I I I 0.90 0.80 0.70 0.60 CCFs { 0.50 0.40 1 a' = 5, R = 0.2, a = 15, a ' = 3 R =0.1, R = 0.3, ' = 20, a = 5 0.10 1i 6.6 6. a 0.10 i 0.as I i 0.30 i I I 6.46 6.50 6.60 I I 0.70 6.86 6.96 1.06 T/ T Fiqure 26 : Theoretical CCFs with both radar and tarqet tilts active, R < 1 and a' > ' -67- Table 4: Parameters for Figures 24, 25, 26 Ratios Figure Number a' 0.1, 0.3, 0.5 1, 3, 5 24 0.1, 0.2, 0.3 25 0.1, 0.2, 0.3 1, 3, 5 1, 3, 5 26 0.1, 0.2, 0.3 5, 15, 20 1, 3, 5 Figures 21-26 reinforce the conclusion drawn earlier from cases (i) and (ii), i.e. the random tilt has a more significant effect than does the target tilt in causing the radar return to decorrelate more rapidly than the tilt itself. Also, as found in cases (i) and (ii), G has to be significantly larger than a these two random effects have comparable impact on CCF. to make Note that the target is only a calibration plate, which is not as heavy as IRAR. and so the former is more vulnerable to external vibration caused by the wind. 111.2. In fact, that is what we will infer in the next section. Model Verification To compare our model with the CCF data from Chapter 2, we need to quantify Ge, C, and $(t) and T- T0 It is reasonable to suppose that 8(t) are turbulence induced tilt angles. take a, = a and c, = T1. This implies we should The values for these parameters will be obtained from the scintillation measurements reported in Chapter 2 and -68- substituted into our CCF model for comparison with CCF data. By trial and error, however, we have found that to best fit the CCF model to should be made linearly proportional to the CCF data the ratio y/a TeV which is assumed equal to r . As we shall see, this result is interesting and it allows us to actually predict the decorrelation time of a set of staring data by using the knowledge of the turbulent conditions. For d < p0 , where d is the diameter of the radar optics exit pupil and p0 is the atmospheric turbulence coherence length, ae can be computed from [13], a0 2 p5/6 d1/6 0 On the other hand, the formula to calculate T needs several steps to develop. It can be shown that in weak turbulence the log-amplitude coherence distance is about equal to /X[ and T ~ //vT ' gives the coherence time of the scintillation in terms of IvTI the magnitude of the tranverse wind velocity which blows perpendicular to the propagation path. Similarly, we have that -69- PO/IVTI As a result, IT5 Finally, p0 is given by, p0 = (1.09 Cn2 k2 L)-3/5 for a spherical wave [5]. Using the scintillation data from Table 1 and the preceding equations we have obtained the a6 and Table 5: Data Set No. Te values shown in Table 5. Estimated a and u6 from Turbulence Theory T (ms) & (pul ses) 1 a (rad) 4.3 x 10-6 2 1.3 x 10- 5 27.25 524 3 1.4 x 10- 5 46.8 900 4 1.6 x 10- 5 14.7 282 5 2.1 x 10'5 18.67 359 6 2.7 x 10 -5 27.2 522 174.3 3352 -70- Figures 27-32 show the theoretical and experimental CCF curves for our six data sets assuming a, = (3, Te = T and the values from Table 5. Obviously, the smooth curve is the theoretical CCF in each figure. At high turbulence levels, Figures 30-32, the theoretical CCFs are very close to the experimental CCFs before the former reaches its asymptotic value. On the other hand, a serious discrepancy occurs in the weak This suggests our model is good turbulence cases, Figures 27-29. Also, it leads us to believe turbulence only at high turbulence levels. is not the sole factor that causes the tilt effects in our model. To force the theoretical CCFs to fit the experimental CCFs better, we have tried to vary various parameter values. that if a6 is kept constant and a It was found is obtained from jT G6 where T = -- 110 (measured in pulses) the discrepancy between theoretical CCFs and experimental CCFs is minimized, as shown in Figures 33-38. The parameters for these figures are given in Table 6. -71- 1.00 Theoretical CCF *.75 0.50 CCFs e.as Experimental CCF 0.0 I 11 I 0. I s5. I I I I I I I"*. 1I I I I is"0. I I I I 20". I I I I I I I I as". PULSES Figure 27 : Theoretical CCF vs. experimental CCF of Data Set 1. 3000. -72- 1.00 e. Theoretical CCF ,. 50 CCFs L S.s5 Experimental CCF 0.6 0. 500. Is". IS". 20". 2as. PULSES Figure 28: Theoretical CCF vs. experimental CCF of Data Set 2. 3000. -73- 1.00 0.7s Theoretical CCF 0.50 CCFs ,.25 Experimental CCF 0.O 0. 50. 1000. IS10. 2090. as*0. PULSES Figure 29: Theoretical CCF vs. experimental CCF of Data Set 3. 3000. -74- 1 .00 0.75 Theoretical CCF CCFs .. as. Experimental CCF I I 0. I I I See. I I I I I in*. I I I I I IS". I I I I I 20". I I I as**. PULSES Figure 30 :Theoretical CCF vs. experimental CCF of Data Set 4. 3". -75- I I I I I I I I I I I I I I I I I I 0.7S Theoretical CCF CCFs Experimental CCF .5as 0.. I I 0. i i i See. I i i 1*. II I I I is". 1I I i a29. i I II I I I asse. PULSES Figure 31 : Theoretical CCF vs. experimental CCF of Data Set 5. 3"0. -76- 1.00 0.7S Theoretical CCF S. *.s CCFs Experimental CCF 0.ZS S".. IS0. as*** PULSES Figure 32: Theoretical CCF vs. experimental CCF of Data Set 6. 3s". -77- 0.7S Theoretical CCF Experimental CCF CCFs *.as 6.S5 I 0. I I I I S". I I II I I is". I I I I I 15". I I I I 20W. I I PULSES Figure 33 : Best CCF fit of Data Set 1. I I I 25. 3"0. -78- 1.00 0.7s 8.58 CCFs Theoretical CCF Experimental CCF 8.25 0.0 ll 0. 1 111 1i ii Is". I I IS"0. as*$. PULSES Figure 34: Best CCF fit of Data Set 2. ii Ji as". 30". -79- 1 .00 I I II I I I I I I I I I I I I 0.75 Theoretical CCF 8.se Experimental CCF CCFs *.25. e.g I I I II I I I I I I S."0. 10. I I I I 1ISO. I I I I I 1 I 20"6. PULSES Figure 35 : Best CCF fit of Data Set 3. I I I I as"0. 30"0. -80- 1 .00 0. I lf i lI I ~I Pi I II |7 I Jl i I I 75 Theoretical CCF O.se Experimental CCF CCFs *.25. 6.S I I II ;.e0. I I I I i0. I I I I I I ISO$. I I I I I I I M"e. PULSES Figure 36: Best CCF fit of Data Set 4. III as".. I 30"e. -81- 1.00 0.7 ,*se CCFs Theoretical CCF . ~Experimental CCF 0.s o. 5". I I .I 5I. 20II. PULSES Figure 37 :Best CCF fit of Data Set 5. as". 39 . -82- 1.00 0.75 Theoretical CCF @.S CCFs Experimental CCF 0.2 I 0. I I I i S".. I I I I I I#". I I I I I IS". I I I I I I I 20S. * PULSES Figure 38 : Best CCF fit of Data Set 6. I I I as".. 30". -83- Parameter Values for Best CCF Fit Table 6: Data Set No. 1 a (rad) a6 (rad) 1.31 x 10~4 4.3 x 10-6 3352 6.2 x 10-5 1.3 x 10 -5 524 1.15 x 10~4 1.4 x 10- 5 900 4.1 x 10- 5 1.6 x 10- 5 282 6.85 x 10- 5 2.1 x 10- 5 359 1.3 x 10~4 2.7 x 10- 5 522 Table 6 shows that a T (pulses) varies from 41 pirad to 131 prad. Intuitively this could represent the turbulence effects augmented by mechanical vibration of the target. such vibrations is the wind. We believe the main source of Because the calibration plate is much lighter than IRAR, it is more sensitive to the wind. relationship between aV, a In short, the and T is significant in the sense that we are able to predict the decorrelation time beforehand by utilizing the scintillation data. -84CHAPTER IV Discussion The time dependence of staring-mode speckle target radar returns has been studied through a combination of laser radar data analysis and In Chapter II, a special formulation was employed mathematical modeling. to estimate the correlation coefficient function (CCF) of the radar return fluctuations in various atmospheric turbulence levels. The experimental CCF results showed that the radar return statistics depend upon the The least degree of data randomness was found in the turbulence strength. weakest turbulence. There was decorrelation time variation, however, between a number of data sets collected in similar turbulence strengths. This observation led us to believe that wind speed is also a contributing factor to the decorrelation process. The experimental data was also compared to the exponential probability density function expected for ensemble or spatial sampling of laser speckle, using a chi-squared goodnessof-fit test. A fair agreement was found in most of the high turbulence data; the weak turbulence data set definitely did not fit the exponential distribution. In chapter III, a theoretical model was developed for the decorrelation process, which modeled the time dependence as being due to random radar and target tilts. The CCF predicted by this model was evaluated for a variety of parameter values. It was found that the random radar tilt has a stronger effect than does the random target tilt drop from one to zero. in forcing CCF to In other words, the target tilt has to be significantl.y -85larger than the radar tilt in order that both effects have comparable impact on CCF. By assuming both tilts were due to turbulence, the theoretical model was quantified using parameter values estimated from scintillation measurements that were taken concurrently with the radar data. The theoretical CCF results were than compared with the experimental results. Fairly good agreement was found in the high turbulence data sets. The discrepancy that was found in the weak turbulence data sets convinced us that in addition to atmospheric turbulence wind-induced target vibration plays a role in the decorrelation process. In trying to force the theoretical CCF model to better fit the experimental CCF results, an interesting result was found. If the radar tilt standard deviation ae is kept constant and the target tilt standard deviation a is obtained from a6 where T = =T 110 gives the tilt decorrelation times measured in pulses, the discrepancy between theoretical CCFs and experimental CCFs is minimized. Needless to say, this is an ad hoc procedure for estimating a and T . Nevertheless, this method did work very well in all six of our data sets. Though a definite conclusion cannot be drawn, we have shown that atmospheric turbulence and wind are indeed the prime contributing factors to staring-mode radar return fluctuations. Papurt [7] and Robertson [14] have worked on related aspects of the return fluctuation problem, so it is instructive to compare their work with ours. Papurt has shown that reduced-scan mode radar return data from a retroreflector had a significant fluctuation component due to turbulence -86- induced radar tilts in addition to the previously predicted scintillation component [5]. A retroreflector is a corner cube made of glass surfaces which ideally will reflect the light pulse back onto itself, so that no target plane tilts can be inferred from retroreflector data. lends credence to our assumption that the radar tilt by turbulence. Papurt's data is caused primarily Robertson did computer simulation of staring-mode speckle target histograms using a l- dimensional version of the random radar tilt and target tilt model that we have proposed. His results showed that as the tilts get larger, the return intensity statistics will approach the exponential probability density function when (1 + 7 ') (1 + G ') >> 1. This result is qualitatively consistent with our experimental return-intensity histograms. In the future, several possible topics may be investigated to extend this work. First, in order to make the CCF predictions more precise, better estimates for a , the target tilt standard deviation, and T , the target tilt decorrelation time, are necessary. Moreover, the ad hoc relation- ship between a., a , T,, and c should be explored by testing it against additional data sets. One might also try to extend Robertson's simulation to the 2 - dimensional problem. -87REFERENCES 1. J. H. Shapiro, "Imaging and Target Detection with a Heterodyne-Reception Optical Radar," Project Report TST-24, Lincoln Laboratory, M.I.T., October 1978. 2. R. C. Harney "Infrared Airborne Radar," Proceedings of the IEEE 1980 Electronic and Aerospace Systems Conference (EASCON), pp. 462-471. 3. J. C. Dainty, ed., Laser Speckle and Related Phenomena (Springer-Verlag, Berlin, 1975). 4. R. C. Harney and R. J. Hull, "Compact Infrared Radar Technology," Proc. SPIE, Vol. 227, pp. 162-170, 1980 5. J. H. Shapiro, B. A. Capron, and R. C. Harney, "Imaging and Target Detection with a Heterodyne-Reception Optical Radar," Appl. Opt., Vol. 20, pp. 3292-3313, 1981. 6. E. Parzen, "On Spectral Analysis with Missing Observations and Amplitude Modulation," The Indian Journal of Statistics, Series A, Vol. 25, Part 4, 1963. 7. D. M. Papurt, "Atmospheric Propagation Effects on Heterodyne-Reception Optical Radars," Doctoral Thesis, E.E.C.S. Dept., M.I.T. Cambridge, MA, 1982. 8. A. Ishimaru, Wave Propagation and Scattering in Random Media, Vol. 1, Academic, New York, 1978. 9. A. Ishimaru, Wave Propagation and Scattering in Random Media, Vol. 2, Academic, New York, 1978. 10. J. H. Shapiro, "Imaging and Optical Communication through Atmospheric Turbulence," J.W..Strohbehn (Ed.), Laser Beam Propagation in the Atmosphere (Springer-Verlag, Berlin, 1978). 11. J. S. Bendat and A. G. Piersol, Engineering Applications of Correlation and Spectral Analysis, John Wiley and Sons, New York, 1980. 12.r J. S. Bendat and A. G. Piersol, Random Data: Analysis and Measurement Procedures., John Wiley and Sons, New York, 1971. 13. D. L. Fried, "Optical Resolution Through a Radnomly Inhomogeneous Medium for Very long and Very short Exposures," Journal of the Optical Society of America, Vol. 56, Number 10, October 1966. 14. R. R. Robertson, "Target-return Statistics from Optical Radar Systems in Staring Mode," Bachelor Thesis, E.E.C.S. Dept., M.I.T, Cambridge, MA, 1982