Application of the Fourier Transform in Antenna Pattern Measurements

The Fourier transform of an antenna radiation pattern is the aperture space distribution across the antenna. The transform from a single pattern results in a collapsed distribution and is therefor most useful for antennas that are physically long and arrayed in one dimension (typical characteristics of cellular and PCS antennas). Investigation of the transformed distribution is helpful in the identification and location of aperture (amplitude and phase) errors. In addition the aperture distribution may be used in the synthesis of an error correction for finite range length patterns. A description and limitations of the processes using both synthesized and measured data is presented. Ronald A. Marino Celwave Division of Radio Frequency Systems Inc. Marlboro, NJ

I. Introduction Many times in the development of antenna array, the engineer finds himself investigating the pattern performance of his design. He may find himself asking numerous questions. Why doesnt it perform like it was designed? Why are the sidelobes so high? Where did the beam squint come from? What is the error mechanism? When there arent any obvious definitive solutions one may have to rely on analysis tools to evaluate the measured performance. One such tool I have found very useful in the evaluation of antenna array pattern performance is the Fourier transform. The Fourier transform of an antenna radiation pattern results in the collapsed aperture distribution along the baseline of the array. It is therefor most useful in the evaluation of antenna arrays that are physically long in one dimension such as the panel type antennas used in cellular/PCS systems. The aperture data is useful in both the identification and location of amplitude and phase errors of the antenna. In addition the aperture data may be used in a modified inverse transform that corrects for finite range effects and known amplitude tapers.

A finite range length produces a spherical phase front that induces a quadratic phase error across the aperture of an antenna. A 2D2/ range length (where D is the longest dimension across the antenna) results in a 22.5 phase error at the edge of the antenna. The primary effects caused by this phase error are to fill nulls1 and to change the level of the inner sidelobes. The error is a function of the projection of the aperture and is therefor most prevalent at 0 and diminishes to zero at the edges of the pattern (90). This null filling effect is a major hindrance to the cellular/PCS antennas where the level of null fill is an important design parameter. Even on a 2D2/ length range, the error introduces a significant change in the level of the null fill. In most cases this error enhances the null fill resulting in a pattern that appears to have better performance than actual. A method that corrects the pattern data, by synthesizing the aperture distribution and reforming the far field pattern with the phase error removed will be presented. The method requires both the amplitude and phase of the antenna radiation pattern to be recorded over a 90 scan. The process of correcting the finite range effects of the radiation pattern involves the following 3 procedures: 1. Interpolation of pattern data with equal interval to equal Sine Space intervals 2. Fourier Transform to Aperture Space 3. Reforming Far Field pattern with phase and amplitude corrections

II. Details of the Transform The details and assumptions of the discrete Fourier transform as it applies to antenna radiation patterns and antenna aperture distribution shall first be presented. If the antenna radiation pattern, a function of (radians), is represented as: f (sin( ) ) and the collapsed aperture distribution, a function of x (wavelengths), represented as:
f (x )

(1)

(2)

the discrete Fourier transform (F) of the radiation pattern is the collapsed aperture distribution and is represented as: F {f (sin( ) )}= f (x ) =

sin = 1

f (sin ( ))e ( )

j 2x sin ( )

(3)

where the summation is calculated over N equally spaced (sine space) points. The inverse Fourier transform of the collapsed aperture distribution is the radiation pattern and is represented as:
n

F 1 {f (x )} = f (sin ( )) =

1 2 n f (x )e j 2x sin ( ) n x =
2

(4)

where for an alias free response: n

(N 1)
2

(5)

An assumption in the derivation of the discrete Fourier transform is that the sequence is a single period of a periodically repeating waveform (Figure 1). The discontinuities at the boundaries between the pseudo periods typically result in very high frequency components (ripple) in the transformed response. This effect has been investigated on arrays with narrow beams with the peak of beam in the proximity of 0. If the transformed response contains significant ripple, windowing techniques2 may be used to minimize the effect. Using the measured radiation pattern of a 48 array (Figure 2), Figures 3 and 4 show how windowing reduces the ripple in the transformed amplitude and phase responses. Windowing simply reduces the level at the boundaries of the pattern data to near zero permitting the boundary levels to be equal. Windowing
3

functions take the shape of many forms but differ in the way they transition from the low level on the ends to the high level in the middle. The windowing technique results in a distortion of the transformed response that is directly related to window shape. A modified Bessel function window has proven to work well with the types of patterns studied and causes minor distortion with a significant reduction of ripple. Array radiating elements whose patterns taper to zero have the advantage of their pattern shape to act as a built in window. Radiating elements such as dipoles when arrayed in the e-plane have this shape and have been found to have minimal ripple in the transformed response without the use of a window. Arraying the dipoles in the h-plane results in a finite response (dependent upon the array factor) which would typically require windowing. The Fourier transform requires N points of equally spaced (sine space from 1 to 1) pattern data. Pattern data, which is normally recorded in equal angle space increments, must first be translated to sine space in equal increments. Since the translation is not linear, interpolation of the measured data is required. The interpolation is made using the linear voltage (real and imaginary) components of the measured data (as opposed to using the amplitude and phase terms). Significant loss of null and peak values may occur if interpolation is used on the amplitude and phase data. The real and imaginary sine space arrays are both separately ran through a Fourier Transform. One of the Fourier transform algorithms investigated was the FFT (Fast Fourier Transform). The FFT is a very efficient algorithm for computing the DFT of a sequence in which N is required to be a power of two and there are N outputs. The FFT requires N*log2(N) complex operations compared to N2 complex operations for the DFT. The length can be set to a power of two by filling the shorter array with zeros. Truncation effects of the filling technique have been shown to result in high frequency components all throughout the transformed data. A preferred method is to interpolate the sine space data to a length of a power of two when using the FFT algorithm. The unaliased span of the Fourier transform is (N-1)/2 wavelengths which results in an aperture space resolution of just under half a wavelength when using the FFT. The above DFT algorithm is useful if finer resolution is required. Any aperture space increment is achievable by adjusting the span and number of points. This algorithm is a more versatile brute force approach, which requires much less overhead than the FFT. On a 200 MHz Pentium Pro the time to compute a DFT of 1024 pattern points to 1200 aperture points is less than 1.5 seconds. Figure 5 is a FORTRAN 77 program listing the DFT algorithm subroutine. The transformed amplitude and phase data has been used in practice to identify differences from the design values. If the error mechanism is simple, corrections typically require the path length of the feed network to change to account for phase errors and/or power divider splits to be modified for amplitude differences. The transform technique is being used extensively in the product development of new PCS antennas and has increased productivity significantly.

For antenna feeds that incorporate non-isolated reactive type dividers, VSWR effects can cause significant aperture errors and are extremely difficult to account for. This is typically identified when a change is made in a path length of an array element that results in a significant change in the amplitude distribution. Another sign of a VSWR effect is if there is a significant variation in amplitude/phase error response over a narrow frequency span. If the error is the result of the VSWR, it is a much wiser approach to retune the feed network and antenna match than to try to adjust for aperture errors. I have had the unfortunate experience of trying to correct aperture phase errors of a multimodule slot array in which the modules were not matched well. It became a cut and try process in which nothing was learned. The transform has also had limited success in the study of the cross polarization component of the radiation pattern. The cross polarization pattern of a linear polarized antenna array when ran through the transform was able to reveal the location and level of the cross polarization components. The technique helped identify a radiating feed line, which when shielded, significantly reduced the cross polarization of the pattern. I would like to share an example in which the transform was extremely useful. When employed at MDM Inc, a company that manufactures and designs high performance waveguide slot array antennas there was a pattern performance problem with a group of production antenna arrays. The array was approximately 24 inches long and had previsions for adding shims to adjust the phase in numerous locations. The technician would systematically adjust the patterns by choosing the size and location of the shim. There was a group of 14 arrays, which had accumulated over a period of time, that could not be tuned and were ready for the scrap heap. These were costly, high performance dip brazed arrays. Scott Horton, the Manager of the Test department at that time, requested the antennas to be measured again on an automated range to see if any could be salvaged. The antennas were measured and the data was run through the transform algorithms. After changing the shims to what the program predicted a total of 8 achieved performance specifications.

III. Finite Range Length Correction An algorithm, that removes the finite range errors from radiation patterns, has been investigated. The algorithm, which uses the transformed pattern data taken on a finite range, incorporates a modified inverse DFT, which accounts for the path length differences along the aperture at varying scan angles. The modified inverse DFT: F 1 {f (x )}= f (sin ( )) =

sin = 1

f (x )e ( )

j 2r ( , Range, x )

(6)

where r is a function of the range length (R), and X is represented by the following: r ( , R, x ) = R 2 2 Rx sin ( ) + x 2 R (7)

The expression under the square root is simply the law of cosines for the triangle formed by the transmitter, origin (center of rotation of antenna) and the x coordinate on the antenna (Figure 8). Sin( ) replaces cos( ) since the geometry defines as the complimentary angle. The algorithm was tested with pattern data synthesized at 1D2/ and 2D2/ range lengths. ESP43 (Electromagnetic Surface Patch Code Version 4) was used as the program to perform the synthesis. ESP4 employs the method of moments for the analysis of radiation patterns and has the option of setting a variable range length. A ten element dipole array spaced /4 above a rectangular reflector was used as the model in the synthesis. Figure 6 shows the overlay of the synthesized pattern performance taken on both a 1D2/ range and infinite (far field) range. Notice the finite range effects are most prevalent in the inner angles and that the peak level is also affected. The 1D2/ synthesized pattern data was ran through the DFT and modified inverse algorithms to remove the effect of the finite range and has been plotted against the far field pattern for comparison (Figure 7). Excellent correlation exists all through out the radiation pattern with the minor exception of the null depths below 30 dB. In addition the peak level was also corrected to its original value. The 2D2/ range corrected patterns had similar performance. For a final verification the corrected pattern was run again through the DFT in order to compare the transformed excitations to that of the model. As shown in Figure 9 and 10 excellent correlation exists for both the amplitude and phase responses. It is important to note that when the algorithm is used, that the windowing technique is to be negated. Windowing techniques though useful in the reduction of ripple in aperture space result in a loss of information that is not recoverable resulting in large distortions at the edges of the patterns.

The algorithm may also be modified to account for an amplitude taper (t (x)) across the aperture due to a narrow beam transmitting antenna. The value of t (x) is typically acquired from the field probe data and/or known patterns of the transmitting antenna. Again like the finite range error the amplitude error is a function of the projection of the aperture and is therefor dependent upon both x and . The equation modified for an aperture taper then becomes: f (x )

F {f (x )}= f (sin ( )) =

sin = 1

t (x, cos( )) e ( )

j 2r ( , Range, x )

(8)

IV. Conclusions The use of a Fourier transform technique that transforms the antenna radiation pattern to the amplitude and phase aperture distribution across the antenna has been proven to be a significant tool for the identification of aperture errors. It has been used successfully in development stages of antenna arrays and in the tuning of production units. The use of a modified inverse DFT which corrects for both the finite range length effects and a known aperture taper has been proven instrumental in the correction of the radiation pattern response. Both of the algorithms have been tested and proven in both theory and test. The techniques have significantly improved the productivity of the development of PCS/Cellular type antenna arrays. I would like to express my thanks for the privilege of having been a student of Vincent Benevento, Charles Donohue and Maxine Vogel of Valley Stream South HS and Dr. Anthony Ferraro and Dr. Lynn Carpenter of the Pennsylvania State University. In addition I would like to thank Frank Liebert for his direction, both on and off the field of engineering.

References Evans, G.,"Antenna Measurement Techniques", Artech House, Inc., 1990

Embree, P. and Bruce, K.,"C Language Algorithms for Digital Signal Processing", Prentice-Hall, Inc., 1991

E. H. Newman, A Users Manual for the Electromagnetic Surface Patch Code: ESP Version IV, Report 716199-11, August 1988, The Ohio State University ElectroScience Lab, Dept. of Electrical Engineering, prepared under grant number NSG 1498 for the National Aeronautics and Space Administration Center, Langley Research Center, Hampton, VA 23665.

Bibliography Ronald A. Marino was born in Brooklyn, New York. He received his B.S.E.E. from the Pennsylvania State University in 1981, where he specialized in antennas and microwaves. He has been an antenna design engineer for 17 years and currently works for Celwave RF as a Sr. Antenna Design Engineer developing new printed circuit antennas. As a Senior Member of the Technical Staff at Litton Amecom (82-87) he designed antennas for EW applications and received a patent for a conformal interferometer array which incorporated printed circuit spiral antennas. When employed with MDM (89-94), where he also acted as a consultant, he designed a modular 44 GHz slot array comprised of 5184 radiators. With Dorne & Margolin in 1994 he received an Engineering Achievement Award for technical excellence in proposal preparation. He has also been responsible for developing automated turn-key antenna test facilities and analysis tools for the following companies; Litton Amecom, Northrop DSD, MDM Inc., Dorne & Margolin, Datron Transco and Celwave RF. E-mail: [email protected].

Figure 5 FORTRAN 77 DFT Algorithm C ********************************************************************* C C SUBROUTINE DFT C C PURPOSE C PERFORM A DISCRETE FOURIER TRANSFORM ON A COMPLEX ARRAY OF C RADIATION PATTERN DATA SAMPLED IN EQUAL SIN SPACE INCREMENTS C C USAGE C CALL DFT(I,D,NUMPTS,FREQUENCY,XSTART,XSTOP,XINC,F,X) C C DESCRIPTION OF PARAMETERS C NUMPTS: Number of sin space points C FREQUENCY: Frequency in GHz C XSTART: Aperture space start position C XSTOP: Aperture space stop position C XINC: Aperture space increment C I(): Input complex pattern array C D(): Input sine space position array C F(): Output complex aperture space array C X(): Output aperture position C C SUBROUTINES AND FUNCTION SUBPROGRAMS REQUIRED C NONE C C METHOD C F(X) = SUM { I(N)e^(j2XSIN()} C C AUTHOR C RON MARINO C C ********************************************************************* SUBROUTINE DFT(I,D,NUMPTS,FREQUENCY,XSTART,XSTOP,XINC,F,X) COMPLEX I(2000),F(2000),JBETAX,J REAL D(2000),X(2000),LAMDA INTEGER XSTART,XSTOP J=CMPLX(0.0,1.0) PI=4.0*ATAN(1.0) RAD=PI/180.0 LAMDA=11.802/FREQUENCY BETA=2.0*PI/LAMDA NUMT=FLOAT((XSTOP-XSTART))/XINC+1 DO 20 K=1,NUMT X(K)=FLOAT(K-1)*XINC+FLOAT(XSTART) JBETAX=J*BETA*X(K) F(K)=CMPLX(0.0,0.0) DO 30 M=1,NUMPTS F(K)=F(K)+(I(M)*CEXP(JBETAX*D(M))) CONTINUE CONTINUE RETURN END

30 20

12

Ron Marino

Waveform Pseudo Waveform Pseudo Waveform

FIGURE 1 Periodic Assumption of Fourier Waveform

f(Theta)

Theta

Ron Marino

Measured Radiation Pattern

FIGURE 2 Measured Data

Frequency: 1.92 GHz

-10

-20

AMPLITUDE (dB)

-30

-40

-50

-60 -90

-60

-30

30

60

90

SCAN ANGLE (DEG)

Ron Marino

No Window Modified Bessel Function Window

FIGURE 3 Window vs No Window Transformed from Pattern Data

Frequency: 1.92 GHz DFT Algorithm

-5

-10

-15

AMPLITUDE (dB)

-20

-25

-30

-35

-40 -40

-30

-20

-10

10

20

30

40

Aperture Space (Inches)

Ron Marino

No Window Modified Bessel Function Window

FIGURE 4 Window vs No Window Transformed from Pattern Data

Frequency: 1.92 GHz DFT Algorithm

180

140

100

60

PHASE (DEG)

20

-20

-60

-100

-140

-180 -40

-30

-20

-10

10

20

30

40

Aperture Space (Inches)

Ron Marino

ESP4 Synthesized Far Field ESP4 Synthesized 1D^2/Lamda Range

FIGURE 6 Far Field vs Finite Range

Frequency: 1.92 GHz

-10

-20

AMPLITUDE (dB)

-30

-40

-50

-60 -90

-60

-30

30

60

90

SCAN ANGLE (DEG)

Ron Marino

ESP4 Synthesized Far Field Corrected 1D^2/Lamda Range

FIGURE 7 Ideal vs Range Corrected

Frequency: 1.92 GHz

-10

-20

AMPLITUDE (dB)

-30

-40

-50

-60 -90

-60

-30

30

60

90

SCAN ANGLE (DEG)

Ron Marino

Corrected Pattern Transform Model Values

FIGURE 9 Corrected Transform vs Model Excitation

Frequency: 1.92 GHz

1.0

0.0

-1.0

-2.0

AMPLITUDE (dB)

-3.0

-4.0

-5.0

-6.0

-7.0 -24

-18

-12

-6

12

18

24

Aperture Space (Inches)

Ron Marino

Corrected Pattern transform Model Values

Figure 10 Corrected Transform vs Model Excitation

Frequency: 1.92 GHz

10

-5

PHASE (DEG)

-10

-15

-20

-25

-30 -24

-18

-12

-6

12

18

24

Aperture Space (Inches)