Solutions For Homework #7

Problem 1:[10 pts]

Let
f(r) =
1
r
=
1
_
x
2
+ y
2
(1)
We compute the Hankel Transform of f(r) by rst computing its Abel Transform
and then calculating the 1D Fourier Transform of the result.
H(q) =
_

_
_

1
_
x
2
+ y
2
dy
_
e
i2qx
dx (2)
In the above, the inner integral can be seen as the Abel transform of f(r), while
the outer integral is the 1D Fourier transform evaluated at frequency q.
We make the following variable substitutions
x = r cos()
y = r sin()
In other words, we change coordinate systems from Cartesian to Polar. Notice
that
dxdy = rdrd (3)
Thus, the integral for the Hankel Transform above becomes
H(q) =
_
2
0
_

0
1
r
e
i2r cos()
rdrd (4)
Recalling the denition of the Bessel function
2J
0
(2rq) =
_
2
0
e
i2rq cos()
(5)
we see that,
H(q) =
_

0
2 J
0
(2rq) dr
=
2
2q
_

0
J
0
(u) du
=
1
q
1
where we have made a change of variables in the second line above, u = 2qr.
Thus, we see that the Hankel Transform of f(r) =
1
r
is
1
q
Problem 2:[10 pts]
(a) We model the aperture illumination function as a circular pillbox as follows
f(r) = rect
_
r
D
_
(6)
where D = 2.5 meters. We know that the Hankel Transform of f(r) is a
jinc function
F(q) = D
2
jinc(Dq) = D
2
J
1
(Dq)
2Dq
(7)
The Fraunhofer approximation gives the power-pattern of the antenna in the
far-eld
P() = |F
_
sin()

_
|
2
|D
2
jinc
_
D

_
|
2
where we have applied the small-angle approximation
sin()

. Figure 1
shows the 2D power-pattern, while a cut through the power-pattern is shown
in Figure 2. The cut through the power-pattern is denoted by the dotted line
in Figure 2. As we can see, the null-to-null width is about 0.1 seconds of
arc. Note: The horizontal axis of Figure 2 is plotted in terms of seconds of
arc by the following relation
q

180
o

3600 (8)
where q =
1
r
is spatial frequency. We anticipate this value of 0.1 arc seconds
for the null-to-null width of the untapered power-pattern through recogniz-
ing that the rst null of the jinc function given above occurs at the value of
1.22, corresponding to a value of , in arc seconds, of
1.22

D

180

3600 = 0.05 (9)

thereby yielding a null-to-null width of about 0.1 arc seconds.
2
angle (seconds)
a
n
g
l
e

(
s
e
c
o
n
d
s
)
Farfield power pattern (point intensity response)
0.5 0 0.5
0.5
0
0.5 160
140
120
100
80
60
40
20
Figure 1:
(b) We also note from Figure 2 that the peak-to-sidelobe ratio, in dB, is about
18.
(c) Figure 3 shows a 1D prole of the Bessel tapering function. To compute
this tapering function, we rst note that the rst minimum of the J
0
(x)
function occurs at x
0
= 3.82, with a corresponding value of 0.4028. Now,
our tapering function, which multiplies the original aperture illumination
function, is described as follows
h(r) = J
0
(ar) (0.4028) (10)
where the scale factor a is determined so that the minimum of the Bessel
function coincides with the edges of the aperture. Since the diameter D of
the aperture is 2.5 meters, we know that a =
x
0
D/2
. Thus, when r = D/2 (i.e.
at the aperture edges), the tapering function h(r) = h(x
0
) = 0 as desired.
The power pattern, in dB, of the tapered aperture illumination function is
shown in Figure 4. A cut through this power pattern in shown as the solid
line in Figure 2. As can be seen, the peak-to-sidelobe ratio is about 35 dB,
implying a signicant reduction of sidelobe level in the weighted-aperture
power-pattern. The reduction of sidelobe level in the weighted aperture case
is important for the detection of a weak start. This is because interfering
radiation from a strong star that is located off the main-lobe direction would
be supressed. However, this reduction in sidelobe level comes at the expense
of a broadening of the main lobe. The null-to-null width of the main lobe of
the tapered-aperture power pattern is about 0.18 seconds of arc, compared
with 0.1 seconds for the untapered case.
3
0.1 0.05 0 0.05 0.1 0.15
60
50
40
30
20
10
0
angle (seconds)
n
o
r
m
a
l
i
z
e
d

p
o
w
e
r

(
d
B
)
Cut through center of 2D power patterns
tapered aperture
untapered aperture 18
dB
35
dB
0.1
secs
0.18
seconds
Figure 2:
(d) We inspect the plots of the power-patterns corresponding to the tapered and
untapered aperture, looking for the 3dB point (corresponding to half power)
down from the peak. We take the half-power width to be twice the width
from 0 to the location on the axis where the power-pattern is 3dB down
from the peak. We nd that for the untapered aperture, the half-power
width is about 0.04 seconds of arc, while for the tapered aperture the
half-power width is about 0.06 seconds of arc.
Problem 3:[10 pts]
(a) We wish to numerically evaluate the Hankel Transform of the following
function
f(r) =
1
(a/2)
2
rect
_
r
a
_
(11)
where, for this part, a = 32 and the spatial variable r = 0, 1, 2, ..., n 1
and n is the rst power of 2 greater than 4a. Thus, n = 256 and the spatial
variable ranges from 0 to 255. We recognize the function above as a circular
pillbox. The integral denition of the Hankel Transform is
F(q) =
_

0
f(r)J
0
(2qr) rdr (12)
4
5 0 5
1
0.5
0
0.5
1
1.5
2
Cut through aperture function and Bessel taper
Figure 3:
angle (seconds)
a
n
g
l
e

(
s
e
c
o
n
d
s
)
Farfield power pattern of tapered aperture
0.5 0 0.5
0.5
0
0.5
180
160
140
120
100
80
60
40
20
Figure 4:
We approximate this integral as a sum, as follows
F(q) =
n1

r=0
f(r)J
0
(2qr)rr (13)
where, of course, the step size r = 1. In addition, the frequency variable
q assumes the values 0,
1
2n
, ....,
n1
2n
. To implement the Bessel function, we
make use of the MATLAB command
besselj(0,r)
.
5
(b) We implement the same function f(r) given in Part (a) in two-dimensions
as follows
f(r) =
1
(a/2)
2
rect
_
_
x
2
+ y
2
a
_
(14)
where x = n, n+1, ...0, ...n1 and y = n, n+1, ...0, ...n1. The Han-
kel Transform of this 2D circularly symmetric function can be easily com-
puted using the 2D FFT (remember that for a circularly-symmetric function,
the Hankel Transform is identical to the two-dimensional Fourier Trans-
form). The Hankel Transform of the 2D function is shown in Figure 5.
We are required to take cuts through the 2D function shown in Figure 5 in
normalized frequency
n
o
r
m
a
l
i
z
e
d

f
r
e
q
u
e
n
c
y
0.2 0.1 0 0.1 0.2
0.2
0.15
0.1
0.05
0
0.05
0.1
0.15
0.2
Figure 5:
the directions dened by the lines at angles 0
o
and 45
o
from the horizontal
axis. The cut at 0
o
is easily obtained by extracting values along a horizontal
line passing through the center of the function. The code that extracts the
cut at 45
o
is given at the back. Note: when displaying the cut at 45
o
, the
1D frequency axis needs to be stretched by a factor of

2, reecting the
fact that discrete steps (in the u v plane) at 45
o
are

2 times longer than

corresponding discrete steps along the horizontal axis.
(c) We know that the analytical form of the Hankel transform of the circularly
symmetric function f(r) ==
1
(a/2)
2
rect
_
r
a
_
is
F(q) =
a
2
(a/2)
2
jinc(aq) (15)
6
Recall that the jinc function is given by jinc(x) =
J
1
(x)
2x
, and so we again
make use of MATLABs besselj routine to compute the analytical answer.
Figure 6 shows the results the numerical evaluation of the Hankel Integral
(solid line), cuts through the 2D Fourier Transform along lines at angles 0
o
and 45
o
from the horizontal axis (dotted lines and dots respectively) as well
as the analytical solution (crosses). We see that the three methods of eval-
0 0.01 0.02 0.03 0.04 0.05 0.06
0.2
0
0.2
0.4
0.6
0.8
1
normalized frequency
numerical integration
2D FFT0
o
2D FFT45
o
jinc
Figure 6:
uation produces answers that are very similar. The effect of approximating
an integral by a discrete sum is evident as the numerical solution diverges
slightly from the other solutions.
(d) Here, a = 8 and so n = 64. We perform the same summation as in Part (a)
as well as form the 2D function as in Part (b). Now, of course, the 2D grid is
smaller since we are only using 64 points. The 2D FFT of the circular pill-
box function when n = 64 is shown in Figure 7. The comparison between
the various solutions is shown in Figure 8. From Figure 8, we clearly see
a divergence between the analytical solution and the numerical one. This
is due to the fact that the approximation of the Hankel Integral by the sum
given above is poor due to the limited number of points (n = 64). Further-
more, we also notice that the cuts through the 2D FFT of circular pillbox
deviates from the analytical solution. As can be seen from Figure 7, the
fewer number of grid points used in the denition of the 2D circular pillbox
function f(r) destroys some of the circular symmetry, causing a disparity
7
normalized frequency
n
o
r
m
a
l
i
z
e
d

f
r
e
q
u
e
n
c
y
0.4 0.2 0 0.2 0.4
0.4
0.2
0
0.2
0.4
Figure 7:
in proles along the horizontal (0
o
) and 45
o
lines.
Problem 4:[10 pts]
(a) In this problem, we seek to recover the 2D, circularly symmetric function
from its Abel Transform. Recall the denition of the Abel transform:
a(x) = 2
_

x
f(r) rdr

r
2
x
2
(16)
Here, we are given the data a(x) evaluated at discrete values of x. Note that
the spacing between points in the prole a(x) is x = 0.01 and the number
of samples in le is 600. Our task is to recover the 2D, circularly symmetric
function f(r) using the Abel-Fourier-Hankel relationship. Note: to load in
the text le, we make use of MATLABs
load
command. The rst column gives the values a long the axis perpendicular
to the projection direction where the projection is evaluated. This function
is plotted in Figure 9
8
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
normalized frequency
numerical integration
2D FFT0
o
2D FFT45
o
jinc
Figure 8:
(b) Recall the Abel-Fourier-Hankel relationship:The Hankel Transformof the
2Dcircularly-symmetric function is equivalent to the 1DFourier Trans-
form of the functions Abel Transform. Mathematically, letting F(q) de-
note the Hankel Transform of the function f(r),
F(q) =
_

A(x)e
i2qx
dx (17)
where A(x) is as was given in Part (a). To recover the 2D function, we
would simply compute the Inverse Hankel Transform of F(q). Note that the
prole contained in the le hw7p4data is essentially A(x). Notice, however,
that the Abel Transform should return even functions. Thus, the data we
are given, shown in Figure 9, is probably the right half of the full, even
Abel Transform of the desired 2D function f(r). We thus form a new, even
function as follows
a

(x) = {
a(x) x > 0
a(x) x < 0
(18)
We then compute the 1D FFT of A

(x) to get the 1D prole of the de-

sired functions spectrum. Remember that the desired function is in two-
dimensions, but circularly-symmetric, implying that all cuts through the
origin of the function or its spectrum should be identical. The 1D FFT
of a

(x) is exactly one such radial cut of the 2D spectrum the desired 2D
9
0 1 2 3 4 5 6
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
spatial variable
noisy projection
Figure 9:
function. Let this spectrum be denoted A(q). To recover a radial cut of
the desired 2D circularly-symmetric function f(r), we compute the Inverse
Hankel Transform, in one dimension, of the 1D spectrum A(q). The Inverse
Hankel Transform is dened as
f
1D
(r) =
_

0
A(q)J
0
(2qr) qdq (19)
We approximate this integral by a sum, as we did in Problem 3.
f
1D
(r)
N1

i=0
A(q
i
)J
0
(2q
i
r)q
i
q (20)
Here, q refers to the frequency spacing in the FFT domain. We know that
the spatial separation of points in the data le hw7p4data is x = 0.01 and
the number of samples N = 2600 = 1200 (remember, the length of a

(x)
is twice the length of a(x)). Therefore, the frequency spacing q =
1
Nx
and the FFT frequencies q = 0,
1
Nq
, ...,
N1
Nq
. Plugging these values in the
discrete sum above, we recover the 1D cut, shown in Figure 10, through the
desired radially-symmetric 2D function f(r).
(c) The MATLAB code that we use to form the image from the 1D radial cut
f
1D
(r) shown in Figure 10 is given at the back. Essentially, we create a grid
(x, y) and copy the i-th value in the 1D prole f
1D
(r
i
) to all locations where
_
x
2
+ y
2
r
i
. The recovered 2D, circularly-symmetric function f(r) is
shown in Figure 11 f(r).
10
0 1 2 3 4 5 6
10
5
0
5
10
15
20
25
spatial variable
recovered 1D profile of
circularlysymmetric function
Figure 10:
horizontal distance
v
e
r
t
i
c
a
l

d
i
s
t
a
n
c
e
recovered 2D circularly symmetric function
5 0 5
6
4
2
0
2
4
6
5
0
5
10
15
20
Figure 11:
Problem 5:[10 pts]
The integral denition of the Bessel function of First Kind, order 0, J
0
(2qr)
is given by
J
0
(2qr) =
1
2
_
2
0
cos(2qr cos()) d (21)
In the above, q denotes frequency. For this problem, we are asked to produce three
Bessel functions of frequencies 4,7 and 10. This equation arises by adding up 2D
11
cosine functions rotated from the x-axis at various angles . That is,
J
0
(2 qr)
1
2

i
cos(2q(xcos(
i
) + y sin(
i
))) (22)
Here, q =
1
wavelength
is the spatial frequency, and we dene wavelength as
N
m
, m =
4, 7, 10, where N is the number of points along one dimension in our grid, is
the spacing between points and m indicates number of cycles over the span of the
grid. The MATLAB code implementing this is given at the back.
Figure 12 gives the three plots. The dots indicate the sum approximation,
while the solid line is the theoretical curve J
0
(2qr) where, of course, r =
_
x
2
+ y
2
.
0 10 20 30 40 50 60
0.5
0
0.5
1
approximation and theoretical, freq = 4
sum approx
theoretical
0 10 20 30 40 50 60
0.5
0
0.5
1
approximation and theoretical, freq = 7
sum approx
theoretical
0 10 20 30 40 50 60
0.5
0
0.5
1
approximation and theoretical, freq = 10
sum approx
theoretical
Figure 12:
MATLAB code for Problem 4
12
% load in data
im = load(hw7p4data);
data2 = im(:,2); x = im(:,1); data2 = data2;
Delta = x(2)-x(1);
% plot noisy projection
figure(1); plot(x,data2);grid on;
h=gca;set(h,FontSize,20); xlabel(spatial variable);
h1=title(noisy projection);set(h1,FontSize,20);
% make data an even function
data = [fliplr(data2(2:end-1)) data2];
N = length(data);
% Abel to Hankel - take FFT of data,
% but keep only positive Fourier coefficients since function
% real and even
F = real( fft(fftshift(data)) ); F = F(1:N/2);
q = [0:N/2-1]/N/(Delta);
% Inverse Hankel Transform, approximated by
% sum_{i=1}N/2 F(q_i) q_i J0(2 pi q_i r) Delta q
r = [0:N/2-1]
*
(Delta
*
sqrt(2));
f = zeros(size(r));
Delta_q = q(2) - q(1);
for k=1:length(r)
f(k) = sum(F(:).
*
Delta_q.
*
q(:).
*
besselj(0,2
*
pi
*
r(k).
*
q(:)));
end
% plot recovered 1D function
figure(2); plot(r, f); grid on;
h=gca;set(h,FontSize,20); xlabel(spatial variable);
h1=title({recovered 1D profile of,circularly-symmetric function});
set(h1,FontSize,20);
13
% create 2D function from 1D profile
rmax = max(r); NN = 4
*
ceil(rmax./Delta/sqrt(2));
horiz = [-NN/2:NN/2-1]
*
Delta; vert = [-NN/2:NN/2-1]
*
Delta;
theta = linspace(0,2
*
pi,2
*
length(f));
f2D = zeros(NN,NN); counts = zeros(NN,NN);
for k = 1:length(f)
if(r(k) = 0)
Rows = round( r(k)
*
sin(theta)./Delta/1 ); Rows = Rows + NN/2 + 1;
Cols = round( r(k)
*
cos(theta)./Delta/1 ); Cols = Cols + NN/2 + 1;
ind = find((Rows <= NN) & (Cols <= NN) );
Rows = Rows(ind); Cols = Cols(ind);
f2D(sub2ind(size(f2D),Rows,Cols)) =
f2D(sub2ind(size(f2D),Rows,Cols)) + f(k);
counts(sub2ind(size(f2D),Rows,Cols)) =
counts(sub2ind(size(f2D),Rows,Cols)) + 1;
else
f2D(ceil(NN/2),ceil(NN/2)) = f(k);
end
end
f2D(find(counts=0)) = f2D(find(counts=0))./counts(find(counts=0));
figure(3); imagesc(horiz,vert,f2D); grid on;
h=gca;set(h,FontSize,20); xlabel(horizontal distance);
ylabel(vertical distance); axis image;
h1=title(recovered 2D circularly symmetric function);
set(h1,FontSize,20); colorbar\\
MATLAB code for Problem 5
14
% parameters
N = 1024;
Delta = 0.1;
a = [N/4 N/7 N/10]
*
Delta; % wavelength
% create spatial array
horiz = [-N/2:N/2-1]
*
Delta; vert = [-N/2:N/2-1]
*
Delta;
[Cols,Rows] = meshgrid(horiz,vert);
rr = sqrt(Rows.2 + Cols.2);
% create angles between 0 and 2 pi radians
% Angles are measured from x-axis;
theta = linspace(0,2
*
pi,20);
% create cosines of varying wavelengths
for k = 1:length(a)
% create horizontal and vertical spatial frequencies
% from theta definition and wavelengths
u = a(k)
*
cos(theta); v = a(k)
*
sin(theta);
q = 1./sqrt( u.2 + v.2 );
% perform sum of 2pi radians as
% (1/2/pi) sum_{i=1}N cos( q r cos(theta) )Delta_theta
map2 = zeros(size(Cols));
for m = 1:length(theta)
disp(m);
map2 = map2 +
cos( 2
*
pi
*
(q(m)
*
cos(theta(m))
*
Cols + q(m)
*
sin(theta(m))
*
Rows) );
end
map2 = map2./length(m);
% extract horizontal cut and store
num_bessel2(k,:) = map2(N/2+1,N/2+1:N)./max(map2(N/2+1,N/2+1:N));
theo_bessel(k,:) = besselj(0,2
*
pi
*
q(1)
*
[0:N/2-1]
*
Delta);
end
15