910 20.

Antenna Arrays

20
Antenna Arrays

20.1 Antenna Arrays

Arrays of antennas are used to direct radiated power towards a desired angular sector.
The number, geometrical arrangement, and relative amplitudes and phases of the array
elements depend on the angular pattern that must be achieved. Fig. 20.1.1 Typical array configurations.
Once an array has been designed to focus towards a particular direction, it becomes
a simple matter to steer it towards some other direction by changing the relative phases
of the array elements—a process called steering or scanning. Figure 20.2.1 shows on the left an antenna translated by the vector d, and on the
Figure 20.1.1 shows some examples of one- and two-dimensional arrays consisting right, several antennas translated to different locations and fed with different relative
of identical linear antennas. A linear antenna element, say along the z-direction, has amplitudes.
an omnidirectional pattern with respect to the azimuthal angle φ. By replicating the
antenna element along the x- or y-directions, the azimuthal symmetry is broken. By
proper choice of the array feed coefficients an , any desired gain pattern g(φ) can be
synthesized.
If the antenna element is replicated along the z-direction, then the omnidirectionality
with respect to φ is maintained. With enough array elements, any prescribed polar angle
pattern g(θ) can be designed.
In this section we discuss array design methods and consider various design issues,
such as the tradeoff between beamwidth and sidelobe level.
For uniformly-spaced arrays, the design methods are identical to the methods for
designing FIR digital filters in DSP, such as window-based and frequency-sampling de-
signs. In fact, historically, these methods were first developed in antenna theory and
Fig. 20.2.1 Translated antennas.
only later were adopted and further developed in DSP.

The current density of the translated antenna will be Jd (r)= J(r − d). By definition,
20.2 Translational Phase Shift the radiation vector is the three-dimensional Fourier transform of the current density,
as in Eq. (15.7.5). Thus, the radiation vector of the translated current will be:
The most basic property of an array is that the relative displacements of the antenna ele-


ments with respect to each other introduce relative phase shifts in the radiation vectors, Fd = ejk·r Jd (r) d3 r = ejk·r J(r − d) d3 r = ejk·(r +d) J(r ) d3 r
which can then add constructively in some directions or destructively in others. This is

a direct consequence of the translational phase-shift property of Fourier transforms: a 

= ejk·d ejk·r J(r ) d3 r = ejk·d F
translation in space or time becomes a phase shift in the Fourier domain.
20.3. Array Pattern Multiplication 911 912 20. Antenna Arrays

where we changed variables to r = r − d. Thus, Example 20.3.1: Consider an array of two isotropic antennas at positions d0 = 0 and d1 = x̂d
(alternatively, at d0 = −(d/2)x̂ and d1 = (d/2)x̂), as shown below:
Fd (k)= ejk·d F(k) (translational phase shift) (20.2.1)

20.3 Array Pattern Multiplication

More generally, we consider a three-dimensional array of several identical antennas lo-
cated at positions d0 , d1 , d2 , . . . with relative feed coefficients a0 , a1 , a2 , . . . , as shown
in Fig. 20.2.1. (Without loss of generality, we may set d0 = 0 and a0 = 1.)
The current density of the nth antenna will be Jn (r)= an J(r − dn ) and the corre- The displacement phase factors are:
sponding radiation vector:
Fn (k)= an ejk·dn F(k) ejk·d0 = 1 , ejk·d1 = ejkx d = ejkd sin θ cos φ

The total current density of the array will be: or, in the symmetric case:

Jtot (r)= a0 J(r − d0 )+a1 J(r − d1 )+a2 J(r − d2 )+ · · · ejk·d0 = e−jkx d/2 = e−jk(d/2)sin θ cos φ , ejk·d1 = ejkx d/2 = ejk(d/2)sin θ cos φ

and the total radiation vector: Let a = [a0 , a1 ] be the array coefficients. The array factor is:

Ftot (k)= F0 + F1 + F2 + · · · = a0 ejk·d0 F(k)+a1 ejk·d1 F(k)+a2 ejk·d2 F(k)+ · · · A(θ, φ) = a0 + a1 ejkd sin θ cos φ

A(θ, φ) = a0 e−jk(d/2)sin θ cos φ + a1 ejk(d/2)sin θ cos φ , (symmetric case)

The factor F(k) due to a single antenna element at the origin is common to all terms.
Thus, we obtain the array pattern multiplication property:
The two expressions differ by a phase factor, which does not affect the power pattern. At
polar angle θ = 90o , that is, on the xy-plane, the array factor will be:
Ftot (k)= A(k)F(k) (array pattern multiplication) (20.3.1)
A(φ)= a0 + a1 ejkd cos φ
where A(k) is the array factor :
and the azimuthal power pattern:
A(k)= a0 ejk·d0 + a1 ejk·d1 + a2 ejk·d2 + · · · (array factor) (20.3.2)  2
g(φ)= |A(φ)|2 = a0 + a1 ejkd cos φ 
Since k = kr̂, we may also denote the array factor as A(r̂) or A(θ, φ). To summarize,
the net effect of an array of identical antennas is to modify the single-antenna radiation Note that kd = 2πd/λ. Figure 20.3.1 shows g(φ) for the array spacings d = 0.25λ,
vector by the array factor, which incorporates all the translational phase shifts and d = 0.50λ, d = λ, or kd = π/2, π, 2π, and the following array weights:
relative weighting coefficients of the array elements.
a = [a0 , a1 ]= [1, 1]
We may think of Eq. (20.3.1) as the input/output equation of a linear system with
A(k) as the transfer function. We note that the corresponding radiation intensities and a = [a0 , a1 ]= [1, −1] (20.3.4)
power gains will also be related in a similar fashion: a = [a0 , a1 ]= [1, −j]

Utot (θ, φ) = |A(θ, φ)|2 U(θ, φ) The first of these graphs was generated by the MATLAB code:
(20.3.3)
Gtot (θ, φ) = |A(θ, φ)|2 G(θ, φ)
d = 0.25; a = [1,1]; % d is in units of λ
[g, phi] = gain1d(d, a, 400); % 400 phi’s in [0, π]
where U(θ, φ) and G(θ, φ) are the radiation intensity and power gain of a single el- dbz(phi, g, 30, 20); % 30o grid, 20-dB scale

ement. The array factor can dramatically alter the directivity properties of the single-
antenna element. The power gain |A(θ, φ)|2 of an array can be computed with the help As the relative phase of a0 and a1 changes, the pattern rotates so that its main lobe is in
of the MATLAB function gain1d of Appendix I with typical usage: a different direction. When the coefficients are in phase, the pattern is broadside to the
array, that is, towards φ = 90o . When they are in anti-phase, the pattern is end-fire, that
[g, phi] = gain1d(d, a, Nph); % compute normalized gain of an array is, towards φ = 0o and φ = 180o .
20.3. Array Pattern Multiplication 913 914 20. Antenna Arrays

d = 0.25λ, a = [1, 1] d = 0.25λ, a = [1, −1] d = 0.25λ, a = [1, −j] d = 2λ, a = [1, 1] d = 4λ, a = [1, 1] d = 8λ, a = [1, 1]
o o o
90 90 90 90o 90o 90o
120o 60o 120o 60o 120o 60o
o o o o o o
120 60 120 60 120 60

o o o o o o o o o o o o
150 30 150 30 150 30 150 30 150 30 150 30

φ φ φ φ φ φ

o −15 −10 −5 o o −15 −10 −5 o o −15 −10 −5 o o −15 −10 −5 o o −15 −10 −5 o o −15 −10 −5 o
180 0 180 0 180 0 180 0 180 0 180 0
dB dB dB dB dB dB

o o o o o o
−150
o −30 −150
o −30 −150
o −30 −150
o −30 −150
o −30 −150
o −30

o o o o o o
−120
o −60 o
−120 −60 o
−120 −60 −120
o −60 o
−120 −60 o
−120 −60
o o o o o o
−90 −90 −90 −90 −90 −90

d = 0.50λ, a = [1, 1] d = 0.50λ, a = [1, −1] d = 0.50λ, a = [1, −j]

90
o
90
o
90
o
Fig. 20.3.2 Grating lobes of two-element isotropic array.
o o o o o o
120 60 120 60 120 60

o o o o o o
150 30 150 30 150 30
Example 20.3.2: Consider a three-element array of isotropic antennas at locations d0 = 0,
φ φ φ
d1 = dx̂, and d2 = 2dx̂, or, placed symmetrically at d0 = −dx̂, d1 = 0, and d2 = dx̂, as
−15 −10 −5 −15 −10 −5 −15 −10 −5
o
180
dB
0
o o
180
dB
0
o o
180
dB
0
o
shown below:

o o o
−150o −30 −150o −30 −150o −30

o o o
−120
o −60 o
−120 −60 o
−120 −60
o o o
−90 −90 −90

d = λ, a = [1, 1] d = λ, a = [1, −1] d = λ, a = [1, −j]

o o o
90 90 90
o o o o o o
120 60 120 60 120 60

o o o o o o
150 30 150 30 150 30
The displacement phase factors evaluated at θ = 90o are:
φ φ φ

o
180
−15 −10 −5
0
o o
180
−15 −10 −5
0
o o
180
−15 −10 −5
0
o ejk·d0 = 1 , ejk·d1 = ejkx d = ejkd cos φ ejk·d2 = ej2kx d = ej2kd cos φ
dB dB dB

o o o
Let a = [a0 , a1 , a2 ] be the array weights. The array factor is:
−150
o −30 −150
o −30 −150
o −30

o o o
A(φ)= a0 + a1 ejkd cos φ + a2 e2jkd cos φ
−120
o −60 o
−120 −60 o
−120 −60
o o o
−90 −90 −90

Figure 20.3.3 shows g(φ)= |A(φ)|2 for the array spacings d = 0.25λ, d = 0.50λ, d = λ,
Fig. 20.3.1 Azimuthal gain patterns of two-element isotropic array. or kd = π/2, π, 2π, and the following choices for the weights:

a = [a0 , a1 , a2 ]= [1, 1, 1]
The technique of rotating or steering the pattern towards some other direction by intro-
a = [a0 , a1 , a2 ]= [1, (−1), (−1)2 ]= [1, −1, 1] (20.3.5)
ducing relative phases among the elements is further discussed in Sec. 20.9. There, we
will be able to predict the steering angles of this example from the relative phases of the a = [a0 , a1 , a2 ]= [1, (−j), (−j) ]= [1, −j, −1] 2

weights.
where in the last two cases, progressive phase factors of 180o and 90o have been introduced
Another observation from these graphs is that as the array pattern is steered from broad-
between the array elements.
side to endfire, the widths of the main lobes become larger. We will discuss this effect in
Sects. 20.9 and 20.10. The MATLAB code for generating the last graph was:
When d ≥ λ, more than one main lobes appear in the pattern. Such main lobes are called
d = 1; a = [1,-j,-1];
grating lobes or fringes and are further discussed in Sec. 20.6. Fig. 20.3.2 shows some
[g, phi] = gain1d(d, a, 400);
additional examples of grating lobes for spacings d = 2λ, 4λ, and 8λ.
 dbz(phi, g, 30, 20);
20.3. Array Pattern Multiplication 915 916 20. Antenna Arrays

d = 0.25λ, a = [1, 1, 1] d = 0.25λ, a = [1, −1, 1] d = 0.25λ, a = [1, −j, −1]

o o o
90 90 90
120o 60o 120o 60o 120o 60o

o o o o o o
150 30 150 30 150 30

φ φ φ

o −15 −10 −5 o o −15 −10 −5 o o −15 −10 −5 o

180 0 180 0 180 0
dB dB dB

The relative weights are a0 , a1 , a2 . The displacement vectors are d1 = x̂d and d2 = ŷd.
o o o
o −30 o −30 o −30
−150 −150 −150 Using Eq. (17.1.4), we find the translational phase-shift factors:
o o o
ejk·d1 = ejkx d = ejkd sin θ cos φ , ejk·d2 = ejky d = ejkd sin θ sin φ
−120
o −60 o
−120 −60 o
−120 −60
o o o
−90 −90 −90

d = 0.50λ, a = [1, 1, 1] d = 0.50λ, a = [1, −1, 1] d = 0.50λ, a = [1, −j, −1]

90
o
90
o
90
o and the array factor:
o o o o o o
120 60 120 60 120 60

A(θ, φ)= a0 + a1 ejkd sin θ cos φ + a2 ejkd sin θ sin φ

o o o o o o
150 30 150 30 150 30

φ φ φ Thus, the array’s total normalized gain will be up to an overall constant:

−15 −10 −5 −15 −10 −5 −15 −10 −5
o
180 0
o o
180 0
o o
180 0
o
 
dB dB dB
 cos(0.5π cos θ) 2
gtot (θ, φ)= |A(θ, φ)|2 g(θ, φ)= |A(θ, φ)|2 



sin θ
o o o
−150o −30 −150o −30 −150o −30

o o o
The gain pattern on the xy-plane (θ = 90o ) becomes:
−120
o −60 o
−120 −60 o
−120 −60
o o o  2
gtot (φ)= a0 + a1 ejkd cos φ + a2 ejkd sin φ 
−90 −90 −90

d = λ, a = [1, 1, 1] d = λ, a = [1, −1, 1] d = λ, a = [1, −j, −1]

o o o
90 90 90
120
o
60
o
120
o
60
o
120
o
60
o
Note that because d = λ/2, we have kd = π. The omnidirectional case of a single element
is obtained by setting a1 = a2 = 0 and a0 = 1. Fig. 20.3.4 shows the gain gtot (φ) for
o o o o o o
150 30 150 30 150 30
various choices of the array weights a0 , a1 , a2 .
φ φ φ
Because of the presence of the a2 term, which depends on sin φ, the gain is not necessarily
o −15 −10 −5 o o −15 −10 −5 o o −15 −10 −5 o
180
dB
0 180
dB
0 180
dB
0 symmetric for negative φ’s. Thus, it must be evaluated over the entire azimuthal range
−π ≤ φ ≤ π. Then, it can be plotted with the help of the function dbz2 which assumes
o −30
o
o −30
o
o −30
o the gain is over the entire 2π range. For example, the last of these graphs was computed
−150 −150 −150
by:
o o o
−120
o −60 o
−120 −60 o
−120 −60
o o o
−90 −90 −90
d = 0.5; a0=1; a1=2; a2=2;
phi = (0:400) * 2*pi/400;
Fig. 20.3.3 Azimuthal gains of three-element isotropic array. psi1 = 2*pi*d*cos(phi);
psi2 = 2*pi*d*sin(phi);
g = abs(a0 + a1 * exp(j*psi1) + a2 * exp(j*psi2)).^2;
The patterns are similarly rotated as in the previous example. The main lobes are narrower, g = g/max(g);
but we note the appearance of sidelobes at the level of −10 dB. We will see later that as dbz2(phi, g, 45, 12);
the number of array elements increases, the sidelobes reach a constant level of about −13
dB for an array with uniform weights. When a2 = 0, we have effectively a two-element array along the x-axis with equal weights.
Such sidelobes can be reduced further if we use appropriate non-uniform weights, but at The resulting array pattern is broadside, that is, maximum along the perpendicular φ =
the expense of increasing the beamwidth of the main lobes. 
90o to the array. Similarly, when a1 = 0, the two-element array is along the y-axis and the
pattern is broadside to it, that is, along φ = 0. When a0 = 0, the pattern is broadside to
Example 20.3.3: As an example of a two-dimensional array, consider three z-directed half- the line joining elements 1 and 2.

wave dipoles: one at the origin, one on the x-axis, and one on the y-axis, both at a distance
d = λ/2, as shown below. Example 20.3.4: The analysis of the rhombic antenna in Sec. 17.7 was carried out with the
help of the translational phase-shift theorem of Eq. (20.2.1). The theorem was applied to
antenna pairs 1, 3 and 2, 4.
20.3. Array Pattern Multiplication 917 918 20. Antenna Arrays

a0=1, a1=1, a2=0 a0=1, a1=0, a2=1 a0=0, a1=1, a2=1

o o o
90 90 90

o o o o o
135 45 135 45 135 45o

φ φ φ

o −9 −6 −3 o o −9 −6 −3 o o −9 −6 −3 o
180 0 180 0 180 0
dB dB dB

o o o o o o
−45 −45 −45
−135 −135 −135 The reflected ray may be thought of as originating from the image of the transmitting
−90
o
−90
o
−90
o
antenna at −h1 , as shown. Thus, we have an equivalent two-element transmitting array.
We assume that the currents on the actual and image antennas are I(z) and ρI(z), where
a0=1, a1=1, a2=1 a0=2, a1=1, a2=1 a0=1, a1=2, a2=2
o
90
o
90
o
90 ρ = −ρTM is the reflection coefficient of the ground for parallel polarization (the negative
o o o o o o
sign is justified in the next example), given in terms of the angle of incidence α by:
135 45 135 45 135 45

n2 cos α − n2 − sin2 α  σ η0
φ φ φ
ρ = −ρTM =  , n2 = −j = r − j σλ
n2 cos α + n2 − sin2 α 0 ω0 2π
o −9 −6 −3 o o −9 −6 −3 o o −9 −6 −3 o
180 0 180 0 180 0
dB dB dB
where n is the complex refractive index of the ground, and we replaced ω0 = 2πf 0 =
2πc0 0 /λ and c0 0 = 1/η0 . Numerically, we may set η0 /2π 60 Ω. From the geometry
of the figure, we find that the angle α is related to the polar angle θ by:
o o o o o o
−135 −45 −135 −45 −135 −45
r sin θ
−90
o
−90
o
−90
o
tan α =
h1 + r cos θ

Fig. 20.3.4 Azimuthal gain patterns of two-dimensional array. In the limit of large r , α tends to θ. For a perfectly conducting ground (σ = ∞), the
reflection coefficient becomes ρ = 1, regardless of the incidence angle.
On the other hand, for an imperfect ground and for low grazing angles (α 90o ), the
A more general version of the translation theorem involves both a translation and a rotation
reflection coefficient becomes ρ = −1, regardless of the conductivity of the ground. This
(a Euclidean transformation) of the type r = R−1 (r − d), or, r = Rr + d, where R is a
is the relevant case in mobile communications.
rotation matrix.
The array factor can be obtained as follows. The two displaced antennas are at locations
The rotated/translated current density is then defined as JR,d (r)= R−1 J(r ) and the cor-
d1 = h1 ẑ and d2 = −h1 ẑ, so that the displacement phase factors are:
responding relationship between the two radiation vectors becomes:
ejk·d1 = ejkz h1 = ejkh1 cos θ , ejk·d2 = e−jkz h1 = e−jkh1 cos θ
jk·d −1
−1

FR,d (k)= e R F R k
where we replaced kz = k cos θ. The relative feed coefficients are 1 and ρ. Therefore, the
array factor and its magnitude will be:
The rhombic as well as the vee antennas can be analyzed by applying such rotational

and translational transformations to a single traveling-wave antenna along the z-direction, A(θ) = ejkh1 cos θ + ρ e−jkh1 cos θ = ejkh1 cos θ 1 + ρ e−jΔ
which is rotated by an angle ±α and then translated.
  2 (20.3.6)
|A(θ)|2 = 1 + ρ e−jΔ  , where Δ = 2kh1 cos θ
Example 20.3.5: Ground Effects Between Two Antennas. There is a large literature on radio-
wave propagation effects [19,34,44,1367–1383]. Consider a mobile radio channel in which The gain of the transmitting antenna becomes Gtot (θ)= |A(θ)|2 G(θ), where G(θ) is the
the transmitting vertical antenna at the base station is at height h1 from the ground and gain with the ground absent. For the common case of low grazing angles, or ρ = −1, the
the receiving mobile antenna is at height h2 , as shown below. The ray reflected from the array factor becomes:
ground interferes with the direct ray and can cause substantial signal cancellation at the  
 2 Δ
receiving antenna. |A(θ)|2 = 1 − e−jΔ  = 2 − 2 cos(Δ)= 4 sin2
2

At the location of the mobile antenna which is at height h2 , the geometry of the figure
implies that cos θ = h2 /r . Thus, we have Δ = 2kh1 cos θ = 2kh1 h2 /r , and
   2
Δ 2kh1 h2
|A(θ)|2 = 4 sin2 Δ2 =
2 r
20.3. Array Pattern Multiplication 919 920 20. Antenna Arrays

where we assumed that kh1 h2 /r  1 and used the approximation sin x x. Therefore, 1 MHz 100 MHz 1 GHz
for fixed antenna heights h1 , h2 , the gain at the location of the receiving antenna drops ground type r σ r σ r σ
like 1/r 2 . This is in addition to the 1/r 2 drop arising from the power density. Thus, the very dry ground 3 10−4 3 10−4 3 1.5×10−4
presence of the ground reflection causes the overall power density at the receiving antenna medium dry ground 15 10−3 15 1.5×10−3 15 3.5×10−3
to drop like 1/r 4 instead of 1/r 2 . wet ground 30 10−2 30 1.5×10−2 30 1.5×10−1
For two antennas pointing towards the maximum gain of each other, the Friis transmission fresh water 80 3×10−3 80 5×10−3 80 1.5×10−1
formula must be modified to read: sea water 70 5 70 5 70 5
 2
P2 λ   2kh1 h2 4πh1 h2
= G1 G2 1 + ρ e−jΔ 2 , Δ= = (20.3.7) According to Eq. (17.1.6), the electric fields E1 and E2 along the direct and reflected rays
P1 4πr r λr
θ, as seen in the above figure.
will point in the direction of their respective polar unit vector θ̂
The direct and ground-reflected rays are referred to as the space wave. When both antennas According to the sign conventions of Sec. 7.2, the reflected field ρTM E2 will be pointing in
are close to the ground, one must also include a term in A(θ) due to the so-called Norton the −θ̂
θ direction, opposing E1 . The net field at the observation point will be:
surface wave [1378–1383]:
e−jkr1 e−jkr2
E = E1 − ρTM E2 = θ̂
θ jkη Fz (θ)sin θ − θ̂
θ jkη ρTM Fz (θ)sin θ
−jΔ −jΔ 4πr1 4πr2
A(θ)= 1 + ρ e + (1 − ρ)Fe
) *+ , ) *+ ,
space wave surface wave
where F(θ)= ẑ Fz (θ) is the assumed radiation vector of the linear antenna. Thus, the
where F is an attenuation coefficient that, for kr  1, can be approximated by [1370]: reflected ray appears to have originated from an image current −ρTM I(z). Using the ap-
 proximations r1 = r − h cos θ and r2 = r + h cos θ in the propagation phase factors e−jkr1
sin2 α 1 and e−jkr2 , we obtain for the net electric field at the observation point (r, θ):
F= , u= n2 − sin2 α
jkr(cos α + u)2 n2
e−jkr  
E = θ̂
θ jkη Fz (θ)sin θ ejkh cos θ − ρTM e−jkh cos θ
At grazing angles, the space-wave terms of A(θ) tend to cancel and the surface wave be- 4πr
comes the only means of propagation. A historical review of the ground-wave propagation
problem and some of its controversies can be found in [1368].
 It follows that the (unnormalized) gain will be:
 2 

2

Example 20.3.6: Vertical Dipole Antenna over Imperfect Ground. Consider a vertical linear an- g(θ)= Fz (θ)sin θ 1 − ρTM (θ)e−2jkh cos θ 
tenna at a height h over ground as shown below. When the observation point is far from
the antenna, the direct and reflected rays r1 and r2 will be almost parallel to each other, The results of the previous example are obtained if we set ρ = −ρTM . For a Hertzian dipole,
forming an angle θ with the vertical. The incidence angle α of the previous example is we may replace Fz (θ) by unity. For a half-wave dipole, we have:
then α = θ, so that the TM reflection coefficient is:
 
 cos(0.5π cos θ) 2  2
 g(θ)=    
1 − ρTM (θ)e−2jkh cos θ 
n2 − sin2 θ − n2 cos θ η0  sin θ

ρTM =  , n2 = r − j σλ
n2 − sin2 θ + n2 cos θ 2π
Fig. 20.3.5 shows the resulting gains for a half-wave dipole at heights h = λ/4 and h = λ/2
and at frequencies f = 1 MHz and f = 100 MHz. The ground parameters correspond to
the medium dry case of the above table. The dashed curves represent the gain of a single
 2
dipole, that is, G(θ)= cos(0.5π cos θ)/ sin θ .
The following MATLAB code illustrates the generation of these graphs:

sigma=1e-3; ep0=8.854e-12; er=15; f=1e6; h = 1/4;

n2 = er - j*sigma/ep0/2/pi/f;

th = linspace(0,pi/2,301); c =cos(th); s2 = sin(th).^2;

rho = (sqrt(n2-s2) - n2*c)./(sqrt(n2-s2) + n2*c);
A = 1 - rho .* exp(-j*4*pi*h*cos(th)); % array factor
G = cos(pi*cos(th)/2)./sin(th); G(1)=0; % half-wave dipole gain

The relative permittivity r = /0 and conductivity σ (in units of S/m) are given below g = abs(G.*A).^2; g = g/max(g); % normalized gain

for some typical grounds and typical frequencies:†

dbp(th, g, 30, 12); % polar plot in dB
† ITU Recommendation ITU-R P.527-3 on the “Electrical Characteristics of the Surface of the Earth,” 1992.
20.4. One-Dimensional Arrays 921 922 20. Antenna Arrays

h = λ /4, f = 1 MHz h = λ /4, f = 100 MHz comes:

θ 0o θ θ 0o θ

30 o
30o
30 o
30o A(θ, φ)= an ejnkd sin θ cos φ (20.4.1)
n

60o 60o 60o 60o Because the angular dependence comes through the factor kx d = kd sin θ cos φ, we are
led to define the variable:
−9 −6 −3 −9 −6 −3
90o
dB
90o 90o
dB
90o ψ = kx d = kd sin θ cos φ (digital wavenumber) (20.4.2)

Then, the array factor may be thought of as a function of ψ:

120o 120o 120o 120o

A(ψ)= an ejψn (array factor in digital wavenumber space) (20.4.3)
150
o 150o 150
o 150o
o o n
180 180

h = λ /2, f = 1 MHz h = λ /2, f = 100 MHz The variable ψ is a normalized version of the wavenumber kx and is measured in
θ 0o θ θ 0o θ units of radians per (space) sample. It may be called a normalized digital wavenumber, in
30o 30o 30o 30o
analogy with the time-domain normalized digital frequency ω = ΩT = 2πf /fs , which
is in units of radians per (time) sample.† The array factor A(ψ) is the wavenumber
60o 60o 60o 60o
version of the frequency response of a digital filter defined by

−9 −6 −3 −9 −6 −3
A(ω)= an e−jωn (20.4.4)
o o o o
90 90 90 90 n
dB dB

We note the difference in the sign of the exponent in the definitions (20.4.3) and
o 120o o 120o (20.4.4). This arises from the difference in defining time-domain and space-domain
120 120
Fourier transforms, or from the difference in the sign for a plane wave, that is,
150o 150o 150o 150o
180o 180o ejωt−jk·r

The wavenumber ψ is defined similarly for arrays along the y- or z-directions. In

Fig. 20.3.5 Vertical dipole over imperfect ground
summary, we have the definitions:

ψ = kx d = kd sin θ cos φ (array along x-axis)

Thus, the presence of the ground significantly alters the angular gain of the dipole. For
ψ = ky d = kd sin θ sin φ (array along y-axis) (20.4.5)
the case h = λ/2, we observe the presence of grating lobes, arising because the effective
ψ = kz d = kd cos θ (array along z-axis)
separation between the dipole and its image is 2h > λ/2.
The number of grating lobes increases with the height h. These can be observed by running The array factors for the y- and z-axis arrays shown in Fig. 20.1.1 will be:
the above example code with f = 1 GHz (i.e., λ = 30 cm) for a cell phone held vertically at  
A(θ, φ) = an ejky yn = an ejkyn sin θ sin φ
a height of h = 6λ = 1.8 meters.

n n
 
A(θ, φ) = an ejkz zn = an ejkzn cos θ
20.4 One-Dimensional Arrays n n

where yn = nd and zn = nd. More generally, for an array along some arbitrary direction,
Next, we consider uniformly-spaced one-dimensional arrays. An array along the x-axis
we have ψ = kd cos γ, where γ is the angle measured from the direction of the array.
(see Fig. 20.3.4) with elements positioned at locations xn , n = 0, 1, 2, . . . , will have dis-
The two most commonly used conventions are to assume either an array along the z-
placement vectors dn = xn x̂ and array factor:
axis, or an array along the x-axis and measure its array factor only on the xy-plane, that
   is, at polar angle θ = 90o . In these cases, we have:
A(θ, φ)= an ejk·dn = an ejkx xn = an ejkxn sin θ cos φ
n n n ψ = kx d = kd cos φ (array along x-axis, with θ = 90o )
(20.4.6)
where we set kx = k sin θ cos φ. For equally-spaced arrays, the element locations are ψ = kz d = kd cos θ (array along z-axis)
xn = nd, where d is the distance between elements. In this case, the array factor be- † Here, Ω denotes the physical frequency in radians/sec.
20.5. Visible Region 923 924 20. Antenna Arrays

For the x-array, the azimuthal angle varies over −π ≤ φ ≤ π, but the array response
is symmetric in φ and can be evaluated only for 0 ≤ φ ≤ π. For the z-array, the polar discrete-time signal processing discrete-space array processing
angle varies over 0 ≤ θ ≤ π. time-domain sampling tn = nT space-domain sampling xn = nd
In analogy with time-domain DSP, we may also define the spatial analog of the z-plane sampling time interval T sampling space interval d
by defining the variable z = ejψ and the corresponding z-transform: sampling rate 1/T [samples/sec] sampling rate 1/d [samples/meter]
frequency Ω wavenumber kx
 digital frequency ω = ΩT digital wavenumber ψ = kx d
A(z)= a n zn (array factor in spatial z-domain) (20.4.7) Nyquist interval −π ≤ ω ≤ π Nyquist interval −π ≤ ψ ≤ π
n
sampling theorem Ω ≤ π/T sampling theorem kx ≤ π/d
The difference in sign between the space-domain and time-domain definitions is also spectral images grating lobes or fringes
evident here, where the expansion is in powers of zn instead of z−n . The array factor frequency response A(ω) array factor A(ψ)
A(ψ) may be called the discrete-space Fourier transform (DSFT) of the array weighting z-domain z = ejω z-domain z = ejψ
sequence an , just like the discrete-time Fourier transform (DTFT) of the time-domain transfer function A(z) transfer function A(z)
case. The corresponding inverse DSFT is obtained by DTFT and inverse DTFT DSFT and inverse DSFT
pure sinusoid ejω0 n narrow beam e−jψ0 n

π windowed sinusoid w(n)ejω0 n windowed narrow beam w(n)e−jψ0 n
1
an = A(ψ)e−jψn dψ (inverse DSFT) (20.4.8) resolution of multiple sinusoids resolution of multiple beams
2π −π
frequency shifting by AM modulation phased array scanning
This inverse transform forms the basis of most design methods for the array coeffi- filter design by window method array design by window method
cients. As we mentioned earlier, such methods are identical to the methods of designing bandpass FIR filter design angular sector array design
FIR filters in DSP. Various correspondences between the fields of array processing and frequency-sampling design Woodward-Lawson design
time-domain digital signal processing are shown in Table 20.4.1. DFT Blass matrix
FFT Butler matrix
Example 20.4.1: The array factors and z-transforms for Example 20.3.1 are for the three choices
for the coefficients:
A(ψ) = 1 + ejψ , A(z) = 1 + z Table 20.4.1 Duality between time-domain and space-domain signal processing.
jψ
A(ψ) = 1 − e , A(z) = 1 − z

A(ψ) = 1 − je jψ
, A(z) = 1 − jz The total width of this region is ψvis = 2kd. Depending on the value of kd, the
visible region can be less, equal, or more than one Nyquist interval:
where z = ejψ and ψ = kd cos φ.

d < λ/2 ⇒ kd < π ⇒ ψvis < 2π (less than Nyquist)
d = λ/2 ⇒ kd = π ⇒ ψvis = 2π (full Nyquist) (20.5.2)
d > λ/2 ⇒ kd > π ⇒ ψvis > 2π (more than Nyquist)
20.5 Visible Region
The visible region can also be viewed as that part of the unit circle covered by the
Because the correspondence from the physical angle-domain to the wavenumber ψ- angle range (20.5.1), as shown in Fig. 20.5.1. If kd < π, the visible region is the arc
domain is through the mapping (20.4.5) or (20.4.6), there are some additional subtleties za zzb with the point z = ejψ moving clockwise from za to zb as φ varies from 0 to π.
that arise in the array processing case that do not arise in time-domain DSP. We note In the case kd = π, the starting and ending points, za and zb , coincide with the ψ = π
first that the array factor A(ψ) is periodic in ψ with period 2π, and therefore, it is point on the circle and the visible region becomes the entire circle. If kd > π, the visible
enough to know it within one Nyquist interval, that is, −π ≤ ψ ≤ π. region is one complete circle starting and ending at za and then continuing on to zb .
However, the actual range of variation of ψ depends on the value of the quantity In all cases, the inverse transform (20.4.8) requires that we know A(ψ) over one
kd = 2πd/λ. As the azimuthal angle φ varies from 0o to 180o , the quantity ψ = complete Nyquist interval. Therefore, in the case kd < π, we must specify appropriate
kd cos φ, defined in Eq. (20.4.6), varies from ψ = kd to ψ = −kd. Thus, the overall values of the array factor A(ψ) over the invisible region.
range of variation of ψ—called the visible region—will be:

− kd ≤ ψ ≤ kd (visible region) (20.5.1)

20.6. Grating Lobes 925 926 20. Antenna Arrays

Fig. 20.5.1 Visible regions on the unit circle.

20.6 Grating Lobes

In the case kd > π, the values of A(ψ) are over-specified and repeat over the visible
Fig. 20.6.1 Two-element interferometer and typical angular pattern.
region. This can give rise to grating lobes or fringes, which are mainbeam lobes in
directions other than the desired one. We saw some examples in Figs. 20.3.1 and 20.3.2.
Grating lobes are essentially the spectral images generated by the sampling process In either case, the output is essentially cos(kd cos φ), and thus, exhibits the grating-
(in this case, sampling in space.) In ψ-space, these images fall in Nyquist intervals other lobe behavior. Cross-correlating interferometers are more widely used because they are
than the central one. more broadband.
The number of grating lobes in an array pattern is the number of complete Nyquist The Very Large Array (VLA) radio telescope in New Mexico consists of 27 dish an-
intervals fitting within the width of the visible region, that is, m = ψvis /2π = kd/π = tennas with 25-m diameters. The antennas are on rails extending in three different
2d/λ. For example in Fig. 20.3.2, the number of grating lobes are m = 4, 8, 16 for directions to distances of up to 21 km. For each configuration, the number of possible
d = 2λ, 4λ, 8λ (the two endfire lobes count as one.) interferometer pairs of antennas is 27(27 − 1)/2 = 351. These 351 outputs can be used
In most array applications grating lobes are undesirable and can be avoided by re- to make a “radio” picture of the source. The achievable resolution is comparable to that
quiring that kd < 2π, or d < λ. It should be noted, however, that this condition does of optical telescopes (about 1 arc second.)
not necessarily avoid aliasing—it only avoids grating lobes. Indeed, if d is in the range The Very Long Baseline Array (VLBA) consists of ten 25-m antennas located through-
λ/2 < d < λ, or, π < kd < 2π, part of the Nyquist interval repeats as shown in out the continental US, Puerto Rico, and Hawaii. The antennas are not physically con-
Fig. 20.5.1. To completely avoid repetitions, we must have d ≤ λ/2, which is equivalent nected to each other. Rather, the received signals at each antenna are digitally recorded,
to the sampling theorem condition 1/d ≥ 2/λ. with the antennas being synchronized with atomic frequency standards, and then the
Grating lobes are desirable and useful in interferometry applications, such as radio recorded signals are digitally cross-correlated and processed off-line. The achievable
interferometry used in radio astronomy. A simple interferometer is shown in Fig. 20.6.1. resolution is about one milli-arc-second.
It consists of an array of two antennas separated by d  λ, so that hundreds or even We note finally that in an interferometer, the angular pattern of each antenna element
thousands of grating lobes appear. must also be taken into account because it multiplies the array pattern.
These lobes are extremely narrow allowing very small angular resolution of radio
sources in the sky. The receiver is either an adder or a cross-correlator of the two Example 20.6.1: In Fig. 20.3.2, we assumed isotropic antennas. Here, we look at the effect of
the element patterns. Consider an array of two identical z-directed half-wavelength dipole
antenna outputs. For an adder and identical antennas with equal weights, the output
antennas positioned along the z-axis at locations z0 = 0 and z1 = d. The total polar gain
will be proportional to the array gain:
pattern will be the product of the array gain factor and the gain of each dipole:
 2  
g(φ)= 1 + ejkd cos φ  = 2 + 2 cos(kd cos φ)  2  cos(0.5π cos θ) 2
gtot (θ)= |A(θ)|2 gdipole (θ)= a0 + a1 ejkd cos θ  


sin θ
For a cross-correlator, the output will be proportional to cos(Ωτ), where τ is the
time delay between the received signals. This delay is the time it takes the wavefront to Fig. 20.6.2 shows the effect of the element pattern for the case d = 8λ and uniform weights
travel the distance d cos φ, as shown in Fig. 20.6.1, that is, τ = (d cos φ)/c. Therefore, a = [a0 , a1 ]= [1, 1]. The figure on the left represents the array factor, with the element
pattern superimposed (dashed gain). On the right is the total gain.
 
2πf d cos φ The MATLAB code used to generate the right graph was as follows:
cos(Ωτ)= cos = cos(kd cos φ)
c
d=8; a=[1,1];
20.7. Uniform Arrays 927 928 20. Antenna Arrays

array gain factor total gain rectangular window. From this point of view, Eq. (20.7.3) is the DSFT of the rectangular
o
θ 0 θ θ 0o θ
30 o o
30 30 o
30o window.
The array factor has been normalized to have unity gain at dc, that is, at zero
60o 60o 60o 60o wavenumber ψ = 0, or at the broadside azimuthal angle φ = 90o . The normalized
power gain of the array will be:
   
−9 −6 −3 −9 −6 −3  sin(Nψ/2) 2  sin (Nkd/2)cos φ 2
90o 90o 90o 90o    
dB dB
g(φ)= |A(ψ)|2 =   =  (20.7.4)
 N sin(ψ/2)   N sin (kd/2)cos φ 

120o 120o 120o 120o Although (20.7.2) defines the array factor for all ψ over one Nyquist interval, the
actual visible region depends on the value of kd.
o 150o o 150o
150
180o
150
180
o Fig. 20.7.1 shows A(ψ) evaluated only over its visible region for an 8-element (N = 8)
array, for the following three choices of the element spacing: d = 0.25λ, d = 0.5λ, and
Fig. 20.6.2 Grating lobes of two half-wavelength dipoles separated by d = 8λ. d = λ. The following MATLAB code generates the last two graphs:
d=1; N=8;
a = uniform(d, 90, N);
[g, th] = gain1d(d, a, 400); [g, phi] = gain1d(d, a, 400);
gdip = dipole(0.5, 400); A = sqrt(g);
gtot = g .* gdip; psi = 2*pi*d*cos(phi);
dbp(th, gtot, 30, 12); plot(psi/pi, A);
dbadd(1, ’--’, th, gdip, 30, 12);
 figure(2);
dbz(phi, g, 45, 20);

20.7 Uniform Arrays

The simplest one-dimensional array is the uniform array having equal weights. For an
array of N isotropic elements at locations xn = nd, n = 0, 1, . . . , N − 1, we define:

1
a = [a0 , a1 , . . . , aN−1 ]= [1, 1, . . . , 1] (20.7.1)
N
so that the sum of the weights is unity. The corresponding array polynomial and array
factor are:

1  1 zN − 1
A(z) = 1 + z + z2 + · · · + zN−1 =
N N z−1
(20.7.2)
1  1 ejNψ − 1
A(ψ) = 1 + ejψ + e2jψ + · · · + e(N−1)jψ =
N N ejψ − 1
where z = ejψ and ψ = kd cos φ for an array along the x-axis and look direction on the
xy-plane. We may also write A(ψ) in the form:
 
Nψ
sin
A(ψ)= 2  ej(N−1)ψ/2 (uniform array) (20.7.3)
ψ
N sin
2
Fig. 20.7.1 Array factor and angular pattern of 8-element uniform array.
The array factor (20.7.2) is the spatial analog of a lowpass FIR averaging filter in
discrete-time DSP. It may also be viewed as a window-based narrow-beam design using a
20.7. Uniform Arrays 929 930 20. Antenna Arrays

As φ varies from 0o to 180o , the visible regions for the three cases are:

d = 0.25λ, ψ = (π/2)cos φ ⇒ −π/2 ≤ ψ ≤ π/2

d = 0.5λ, ψ = π cos φ ⇒ −π ≤ ψ ≤ π
d = λ, ψ = 2π cos φ ⇒ −2π ≤ ψ ≤ 2π
Thus, in the first case the visible region is only half of the Nyquist interval; in the
second case, it is the full interval; and in the third case, the Nyquist interval is covered
twice, and therefore, grating lobes will appear. Because ψ = 2π cos φ, the grating lobes
at ψ = ±2π correspond to the endfire angles of φ = 0o and 180o (the larger width of
the endfire lobes is explained in Sec. 20.10.)
The N − 1 zeros of the array polynomial A(z) are the N-th roots of unity, except Fig. 20.7.3 Mainlobe width and sidelobe level of uniform array.
for the root at z = 1, that is,
2πi
zi = ejψi , ψi = , i = 1, 2, . . . , N − 1 Solving for Δφ3dB , we obtain Δφ3dB = Δψ3dB /(kd)= 0.886(2π/N)/(2πd/λ), or
N
Because these zeros lie on the unit circle, they will correspond to nulls in the angular λ
pattern, as long as they lie in the visible region. For d = 0.25λ, and in general for any Δφ3dB = 0.886 (3-dB width at broadside) (20.7.6)
Nd
d < λ/2, only a subset of these zeros will fall in the visible region. The zeros of the
8-element array patterns of Fig. 20.7.1 are shown in Fig. 20.7.2. The mainlobe beamwidth gets narrower with increasing N, while the relative sidelobe
level remains the same. To achieve better (lower) sidelobe levels, one must use non-
uniform weights obtained from non-rectangular windows.
The quantity D = Nd is the effective aperture of the array. Thus, we recognize
Eq. (20.7.6) as the classical Rayleigh limit on the resolving power of an optical system,
which states that the angular resolution achieved by an aperture of length D is essentially
λ/D.
The beamwidth expression (20.7.5) and the 13-dB sidelobe level can be justified as
follows. The peak of the first sidelobe occurs approximately half-way between the first
two nulls, that is, at ψ = 3π/N. More precisely, it occurs at ψ = 2.8606π/N. Thus,
the sidelobe level in dB will be:
Fig. 20.7.2 Zero locations and visible regions of 8-element uniform array.
   
 A(ψ)   sin(1.4303π) 
R = −20 log10  
 A(0)  = −20 log 
10 

The two most important features of the uniform array are its 3-dB beamwidth Δψ3dB , ψ=2.8606π/N N sin(1.4303π/N) 
or Δφ3dB in angle-space, and its sidelobe level R. These parameters are shown in    
 sin(1.4303π)   
Fig. 20.7.3, for an 8-element uniform array with d = 0.5λ. −20 log10   = −20 log  sin(1.4303π)  = 13.26 dB
 N(1.4303π/N)  10 
1.4303π

For N larger than about 5–6, the sidelobe level becomes independent of N and has
the limiting value of R = 13 dB. Similarly, the beamwidth in ψ-space—defined as the where we used the small-x approximation, sin x x, in the denominator, which is justi-
full width of the mainlobe at the half-power level—takes the simple form: fied when N is large. Setting x = Nψ/2, the sidelobe peak corresponds to the secondary
2π maximum of the approximate array factor sin x/x, which by differentiation leads to the
Δψ3dB = 0.886 (3-dB width in ψ-space) (20.7.5) equation x = tan x, having solution x = 1.4303π, or ψ = 2x/N = 2.8606π/N.
N
The 3-dB width Δψ3dB is twice the 3-dB or half-power frequency ψ3 , defined to be
The first nulls in the array factor about the mainlobe are at ±ψ1 = ±2π/N, and the solution of the equation:
therefore, 2π/N represents half of the base of the mainlobe.
 
The 3-dB width Δφ3dB in angle space can be obtained by differentiating the equation  sin(Nψ /2) 2
 3  1
ψ = kd cos φ, that is, dψ = (∂ψ/∂φ)dφ = (−kd sin φ)dφ. Evaluating the derivative |A(ψ3 )|2 =   =
 N sin(ψ3 /2)  2
at broadside (φ = 90o ) and assuming a narrow mainlobe, we have:
  Because ψ3 is always smaller than 2π/N, it will be small for large N, and therefore,
 ∂ψ 
 
Δψ3dB =   Δφ3dB = kd Δφ3dB we may make the same approximation in the denominator as above, giving the simplified
 ∂φ 
20.8. Array Directivity 931 932 20. Antenna Arrays

equation: The maximum of this quantity is reached when all the coefficients are equal to each
 
 sin(Nψ /2) 2  
 sin x3 2 other. The common value may be adjusted so that their sum is unity, that is:
   = 1
 =
3
 
 Nψ3 /2  x3  2
1
where x3 = Nψ3 /2. The quantity x3 is determined to be the constant x3 = 0.443π.
an = , n = 0, 1, . . . , N − 1
N
Thus, ψ3 = 2x3 /N = 0.443(2π/N), and Δψ3dB = 2ψ3 = 0.886(2π/N).
The maximized value of D becomes:

20.8 Array Directivity Dmax = N (20.8.3)

Thus, the uniform array with half-wavelength spacing achieves maximum directivity
The value of kd has an impact also on the directivity of an array. In the array processing
equal to the number of array elements. This result is analogous to finding the optimum
literature, the directivity of an array is usually defined with reference to a z-directed
N-tap lowpass FIR filter that minimizes the noise reduction ratio, that is, the sum of the
array consisting of isotropic radiators. The wavenumber is ψ = kd cos θ and the max-
squares of its coefficients.
imum of the array factor is assumed to occur at broadside θ = 90o , or ψ = 0. This
For arbitrary spacing d, it is shown in Problem 20.6 that the optimum array vec-
basically means that the array factor will have a lowpass shape as a function of ψ, with
tor a = [a0 , a1 , . . . , aN−1 ]T that maximizes (20.8.1), and the corresponding maximum
a maximum value at dc given by
  directivity, are given by:
N− 
 1 

|A(0)| =  an  a = A−1 u , Dmax = uT A−1 u
 (20.8.4)
 n=0 
where u = [1, 1, . . . , 1]T is a vector of N ones and A is the so-called prolate matrix
It follows that the normalized power gain of the array will be:
[1274] with matrix elements:
g(θ)= c|A(θ)|2 
sin kd(n − m)
Anm = , 0 ≤ n, m ≤ N − 1 (20.8.5)
where c = 1/|A(0)| . The corresponding beam solid angle will be:
2
kd(n − m)

π
π
The coefficients a may be renormalized such that their sum is unity. When d is an
ΔΩ = 2π g(θ)sin θ dθ = 2π c|A(θ)|2 sin θ dθ
0 0 integer multiple of λ/2, the prolate matrix reduces to the N×N identity matrix, resulting
Changing variables of integration from θ to ψ, which varies over the visible region into (20.8.3).
(20.5.1), we obtain:

kd
kd 
2π 2πc 20.9 Array Steering
ΔΩ = c|A(ψ)|2 dψ = an a ∗
me
j(n−m)ψ
dψ
kd −kd kd −kd n,m
An array is typically designed to have maximum directive gain at broadside, that is,
Performing the integration, we get at φ = 90o (for an array along the x-axis.) The maximum of the array factor A(ψ)
 corresponds to ψ = kd cos φ = 0, so that |A|max = |A(0)|.
 sin kd(n − m)
ΔΩ = 4πc an a∗
m We wish to “electronically” rotate, or steer, the array pattern towards some other
n,m kd(n − m)
direction, say φ0 , without physically rotating it. The corresponding wavenumber at the
Therefore, the directivity of the array becomes: desired look-direction will be:
 2
 an  ψ0 = kd cos φ0 (steering phase) (20.9.1)
4π n
D= =   (20.8.1)
ΔΩ sin kd(n − m) Such steering operation can be achieved by wavenumber translation in ψ-space, that
an a ∗
n,m
m
kd(n − m) is, replacing the broadside pattern A(ψ) by the translated pattern A(ψ − ψ0 ). Thus,
we define:
In the particular case of half-wavelength spacing d = λ/2 or kd = π, the sinc function A (ψ)= A(ψ − ψ0 ) (steered array factor) (20.9.2)
acts as a delta function δ(n − m), and the sum simplifies into:
:N−1 2 and the translated wavenumber variable,
 
n=0 an
D = :N− (20.8.2)
1 ψ = ψ − ψ0 = kd(cos φ − cos φ0 ) (steered wavenumber) (20.9.3)
n=0 |an |
2
20.9. Array Steering 933 934 20. Antenna Arrays

Then, A (ψ)= A(ψ ). The maximum of A (ψ) will coincide with the maximum of The MATLAB functions steer.m and scan.m of Appendix I can be used to implement
A(ψ ), which occurs at ψ = 0, or equivalently at ψ = ψ0 , or at angle φ = φ0 . Eq. (20.9.4). Their usage for even or odd number of array elements is discussed in
Fig. 20.9.1 illustrates this wavenumber translation process and the corresponding ro- Sec. 21.1.
tation of the angular pattern, for an 11-element uniform array with d = λ/2, steered
Example 20.9.1: In Examples 20.3.1 and 20.3.2, we considered the three cases having progres-
from broadside to φ0 = 60o . The MATLAB code for the last two graphs was:
sive phases ψ0 = 0, π, π/2. These may or may not correspond to a physical steering angle
d=0.5; N=11; ph0=60; φ0 , depending on whether or not ψ0 lies in the visible region.
a = uniform(d, ph0, N); % steered uniform weights
[g, phi] = gain1d(d, a, 400); % calculate normalized gain g(φ) In the case ψ0 = π and d = 0.25λ, we have ψ = 0.5π cos φ, and therefore it is not possible
psi = 2*pi*d*cos(phi); % φ to ψ transformation to find a solution for 0.5π cos φ0 = ψ0 = π. However, the array factor does correspond
figure; plot(psi/pi, sqrt(g)); % plot in ψ space to a pattern rotated towards endfire. This can be seen from the expression,
figure; dbz(phi, g, 30, 20); % azimuthal gain plot in dB    
|A(ψ)| = |1 − ejψ | = 2sin(ψ/2) = 2sin(0.25π cos φ)

which is maximum towards endfire and minimum towards broadside. In the case ψ0 =
π/2 and d = 0.25λ, there is a solution to 0.5π cos φ0 = ψ0 = 0.5π, that is, φ0 = 0o ,
which corresponds to the maximum of the steered array.
In the case ψ0 = π and d = 0.5λ, we have ψ = π cos φ, and the solution to the equation
π cos φ0 = π is φ0 = 0o . However, because the phase ψ0 = π is indistinguishable
from the phase ψ0 = −π (both lead to e−jψ0 = −1), we will also have the solution to
π cos φ0 = −π, which is φ0 = 180o .
In the case ψ0 = π/2 and d = 0.5λ, the solution to π cos φ0 = π/2 is φ0 = 60o , which
corresponds to the maximum, as can be seen in Fig. 20.3.1.
In the case ψ0 = ±π and d = λ, we have ψ = 2π cos φ, and the solutions to 2π cos φ0 =
±π are φ0 = 60o and 120o .
Finally, in the case ψ0 = π/2 and d = λ, the solution to 2π cos φ0 = π/2 is φ0 = 75.5o .
However, there is another grating lobe maximum towards φ0 = 138.6o , which corresponds
to the solution of 2π cos φ0 = −3π/2. This is so because ψ0 = π/2 and ψ0 = −3π/2 are
indistinguishable phases, both leading to e−jψ0 = −j.


The concepts of visible region, beamwidth, and the condition for absence of grating
Fig. 20.9.1 Array steering or scanning by translation in wavenumber space. lobes, translate with minor modifications to the case of a steered array. As the angle φ
varies over 0o ≤ φ ≤ 180o , the translated wavenumber ψ of Eq. (20.9.3) varies over the
It follows from the translation theorem of Fourier transforms that the weight coef- shifted visible region:
ficients an of the translated pattern A (ψ) will be given by:
−kd(1 + cos φ0 )≤ ψ ≤ kd(1 − cos φ0 ) (shifted visible region) (20.9.5)
an = an e −jψ0 n
(steered array weights) (20.9.4)
where its total width is again 2kd. The condition for absence of grating lobes is obtained
so that we have: with the help of the inequality:
    
|ψ | ≤ kd| cos φ − cos φ0 | ≤ kd | cos φ| + | cos φ0 | ≤ kd 1 + | cos φ0 |

A (ψ)= an ejψn = an ej(ψ−ψ0 )n = an ejψ n = A(ψ )
n n n
To ensure no grating lobes, ψ must remain strictly less than 2π, which results in
Because of the progressive phase factors e −jψ0 n
in the weights an , the steered or 
the sufficient condition: kd 1 + | cos φ0 | < 2π, or replacing kd = 2πd/λ,
scanned array is sometimes called a phased or scanning array.
The time-domain version of array steering is AM modulation, in which a baseband λ
signal is translated up in frequency by modulating with it a sinusoidal carrier, much like d< (no grating lobes) (20.9.6)
1 + | cos φ0 |
Eq. (20.9.4). Frequency translation is also used in DSP for mapping a lowpass filter into
a bandpass one and for designing filter banks. We will use it in Sec. 21.4 to design arrays At broadside, φ0 = 90o , this reduces to the earlier condition d < λ. At endfire,
with angular sector patterns. φ0 = 0o or 180o , it reduces to d < λ/2.
20.10. Array Beamwidth 935 936 20. Antenna Arrays

20.10 Array Beamwidth To summarize, the angular 3-dB width of the steered array can be computed in terms
of the broadside 3-dB width in wavenumber space by:
Because the steered array has a mainlobe towards the direction φ0 , the beamwidth must ⎧
be calculated by linearizing the map ψ = kd cos φ about φ0 , that is, ⎪
⎪ 1
⎪
⎪ Δψ3dB , for 0o < φ0 < 180o
⎪
⎨ kd sin φ0
 
 ∂ψ  Δφ3dB =  (20.10.6)
  ⎪
Δψ =   Δφ = | − kd sin φ0 | Δφ ⎪
⎪
 ∂φ  ⎪ 2 Δψ3dB ,
⎪ for φ0 = 0o , 180o
φ0 ⎩
kd
which leads to the 3-dB beamwidth in angle-space:
In particular, if Eq. (20.10.2) is used:
1 ⎧
Δφ3dB = Δψ3dB , (3-dB width of steered array) (20.10.1) ⎪
⎪ 0.886 λ
kd sin φ0 ⎪
⎪ b, for 0o < φ0 < 180o
⎪
⎨ sin φ0 Nd
For window-based narrow-beam design methods, the beamwidth Δψ3dB is approxi- Δφ3dB =  (20.10.7)
⎪
⎪
⎪
⎪ λ
mately equal to the product of the beamwidth of the uniform array, Eq. (20.7.5), and a ⎪
⎩ 2 0.886 b, for φ0 = 0 , 180
o o

so-called broadening factor b, whose value depends on the choice of the window. Thus, Nd
we have: In degrees, Eq. (20.10.7) reads as:
⎧
2πb ⎪
⎪ 50.76o λ
Δψ3dB = b Δψ3-dB, uniform = 0.886 (3-dB width in ψ-space) (20.10.2) ⎪
⎪ b, for 0o < φ0 < 180o
N ⎪
⎨ sin φ0 Nd
Δφ3dB =  (20.10.8)
⎪
⎪
Combining Eqs. (20.10.1) and (20.10.2) and replacing kd by 2πd/λ, we get: ⎪
⎪ λ
⎪ 107.86o
⎩ b, for φ0 = 0 , 180
o o
Nd
0.886 λ
Δφ3dB = b, (3-dB width in angle-space) (20.10.3) In some designs such as binomial arrays, it is easier to determine Δψ3dB directly
sin φ0 Nd
from the array factor A(ψ). In other designs, it is more convenient to estimate Δψ3dB
The 3-dB angles will be approximately φ0 ± Δφ3dB /2. Because of the presence of using Eq. (20.10.2).
sin φ0 in the denominator, the beamwidth Δφ3dB will broaden as the array is steered The broadening factor b depends on the choice of the window and its sidelobe level.
from broadside to endfire. The larger the sidelobe attenuation, the larger the broadening factor. Some examples of
Exactly at endfire, φ0 = 0o or 180o , Eq. (20.10.3) fails and the beamwidth must be broadening factors for different windows are given as follows:
calculated by a different procedure. At φ0 = 0o , the translated wavenumber ψ =
Rectangular: b = 1, (R = 13 dB)
ψ − ψ0 becomes ψ = kd(cos φ − 1). Using the approximation cos x = 1 − x2 /2, we
may relate the 3-dB angle φ3 to the corresponding 3-dB wavenumber by: Hamming: b = 2, (R = 40 dB)
6(R + 12)
 1 Taylor-Kaiser [1265]: b=
ψ3 = kd(cos φ3 − 1)= kd (1 − φ23 /2)−1 = − kdφ23 155   2
2 2
Dolph-Chebyshev [1263]: b = 1 + 0.636 cosh acosh2 (Ra )−π2
Ra
It follows that the 3-dB width in ψ-space will be Δψ3dB = 2|ψ3 | = kdφ23 . Solving for

φ3 , we have φ3 = Δψ3dB /kd. Thus, the 3-dB width in angle space will be Δφ3dB = 2φ3 , where R and Ra represent the sidelobe level in dB and absolute units, respectively,

Δψ3dB R = 20 log10 (Ra )  Ra = 10R/20 (sidelobe level) (20.10.9)
Δφ3dB = 2 , (3-dB width at endfire) (20.10.4)
kd
Here, R and Ra represent the attenuation of the sidelobe and, therefore, R > 0 and
The same expression also holds for endfire towards φ0 = 180o . Replacing Δψ3dB Ra > 1. The corresponding gain of the sidelobe relative to the mainlobe peak will be
from Eq. (20.10.2), we find the width in angle space: R−
a = 10
1 −R/20
, which is less than one.
 The MATLAB function bwidth.m of Appendix I implements Eq. (20.10.6). Its inputs
λ are the quantities d, φ0 , Δψ3dB and its output is the 3-dB width in degrees Δφ3dB . Its
Δφ3dB = 2 0.886 b, (3-dB width in angle-space) (20.10.5) usage is:
Nd
Dphi = bwidth(d, phi0, Dpsi); % map Δψ beamwidth to Δφ beamwidth
20.11. Problems 937 938 20. Antenna Arrays

20.11 Problems 20.4 Four identical isotropic antennas are positioned on the xy-plane at the four corners of a
square of sides a, as shown below. Determine the array factor A(φ) of this arrangement as
20.1 Show that the modified Friis formula (20.3.7) for two antennas over imperfect ground takes a function of the azimuthal angle φ. (Assume the look direction is on the xy-plane.)
the following frequency-independent form in the limit of low grazing angles and h1 h2  λr :
 2
P2 h 1 h2
= G1 G2
P1 r2

20.2 Consider two horizontal dipoles I over imperfect ground, oriented along the x and y direc-
tions, as shown below. Show that the effect of the direct and ground-reflected rays can be
obtained by considering an image dipole ρI.
20.5 The array factor of a two-element array is given by:
 2 1 + sin ψ π
g(φ)= a0 + a1 ejψ  = , ψ= cos φ
2 2

where φ is the azimuthal angle (assume θ = 90o ) and ψ, the digital wavenumber. The array
elements are along the x-axis at locations x0 = 0 and x1 = d.

a. What is the spacing d in units of λ? Determine the values of the array weights, a =
[a0 , a1 ], assuming that a0 is real-valued and positive.
b. Determine the visible region and display it on the unit circle. Plot |A(ψ)|2 versus ψ
By considering the relative directions of the electric field along the direct and reflected rays, over the visible region. Based on this plot, make a rough sketch of the radiation pattern
show that the resulting in array factor has the form: of the array (i.e., the polar plot of g(φ) versus 0 ≤ φ ≤ 2π).
c. Determine the exact 3-dB width of this array in angle space.
A(θ)= ejkh cos θ + ρ e−jkh cos θ
20.6 Defining the array vector a and the prolate matrix A via Eqs. (20.8.4) and (20.8.5), show that
with ρ = ρTM for the x-directed case and ρ = ρTE for the y-directed one, where ρTM , ρTE are
the directivity defined in Eq. (20.8.1) can be written in the compact form, where the dagger
given by Eq. (7.4.4) with n2 = r − j60σλ.
† indicates the conjugate transposed operation:
20.3 A z-directed half-wave dipole is positioned in front of a 90o corner reflector at a distance  † 2
d from the corner, as shown below. The reflecting conducting sheets can be removed and u a
D= (20.11.1)
replaced by three image dipoles of alternating signs, as shown. a† A a

a. Show that the maximum of D is attained for a = A−1 u and that the maximized D is
Dmax = u† A−1 u. Show that the value of Dmax is not affected if a is defined with an
arbitrary normalization factor μ, that is, a = μA−1 u.
b. Show that an equivalent problem is the minimization problem:

a† Aa = min , subject to u† a = 1

c. Show that (20.11.1) is a special case of the more general problem of the maximization
of the Rayleigh quotient :
a† Q a
D= † = max
a. Thinking of the equivalent image problem as an array, determine an analytical expres- a Aa
sion for the array factor A(θ, φ) as a function of the polar and azimuthal angles θ, φ. where A, Q are positive-definite Hermitian matrices. Show that the solution of this
b. For the values d = 0.5λ, d = λ, and d = 1.5λ, plot the azimuthal pattern A(90o , φ) problem is the eigenvector corresponding to the maximum eigenvalue λ = λmax of the
at polar angle θ = 90o and for −45o ≤ φ ≤ 45o . generalized eigenvalue problem Q a = λAa. Explain how this formulation leads to the
same solution in the case of (20.11.1).
c. For the cases d = 0.5λ and d = 1.5λ, calculate the directivity D (in dB and in absolute
units) and compare it with the directivity of a single half-wave dipole in the absence d. Show that the directivity (20.11.1) of a uniform array (a = u) is given by the two equiv-
of the reflector. alent forms:
|u† u|2 N2
d. Suppose that the corner reflector is flattened into a conducting sheet lying on the yz Dunif = † =
u Au 1
N−
 sin(kdn)
plane, i.e., the 90o angle between the sheets is replaced by a 180o angle. Repeat parts N+2 N − |n|
kdn
(a–c) in this case. n=1
20.11. Problems 939

20.7 Computer Experiment—Optimum Directivity. Using the matrix formulation of the previous
problem, calculate the optimum directivity for an N-element array over the range of spacing
values: 0.1 ≤ d/λ ≤ 2 and plot it versus d. Carry this out for the values N = 5, 10, 15 and
place the results on the same graph.
21
The directivity D of (20.11.1) can be evaluated for any given vector of array weights. Evaluate
it for the uniform array a = u and plot the results on the same graph as above. You should Array Design Methods
observe that directivity of the uniform array comes close to that of the optimum one for
most (but not all) of the spacings d.
For each d and for the case N = 15, calculate the directivities of the array weights a designed
with the MATLAB function taylor1p of the next chapter, with sidelobe attenuations of R =
20 R = 30 dB, and place them on the same graph.

21.1 Array Design Methods

As we mentioned in Sec. 20.4, the array design problem is essentially equivalent to the
problem of designing FIR digital filters in DSP. Following this equivalence, we discuss
several array design methods, such as:

1. Schelkunoff’s zero placement method

2. Fourier series method with windowing
3. Woodward-Lawson frequency-sampling design
4. Narrow-beam low-sidelobe design methods
5. Multi-beam array design

Next, we establish some common notation. One-dimensional equally-spaced arrays

are usually considered symmetrically with respect to the origin of the array axis. This
requires a slight redefinition of the array factor in the case of even number of array
elements. Consider an array of N elements at locations xm along the x-axis with element
spacing d. The array factor will be:
 
A(φ)= am ejkx xm = am ejkxm cos φ
m m

where kx = k cos φ (for polar angle θ = π/2.) If N is odd, say N = 2M + 1, we can

define the element locations xm symmetrically as:

xm = md, m = 0, ±1, ±2, . . . , ±M

This was the definition we used in Sec. 20.4. The array factor can be written then as
a discrete-space Fourier transform or as a spatial z-transform:


M M %
 &
A(ψ) = am ejmψ = a0 + am ejmψ + a−m e−jmψ
m=−M m=1
(21.1.1)

M 
M
 
m m −m
A(z) = am z = a0 + am z + a−m z
m=−M m=1