60 3  Large, Single Element Transducer Models

which (compare to Eq. (3.36)) is indeed just the on-axis wave field of a virtual ro-
tated piston element of length 2b cos Φ having a velocity v0 / cos Φ on its face and
focused at z = R0.

3.3 Amplitude Weighting

In phased arrays, in addition to delaying the individual elements of the array to

generate beam steering and focusing, one can weight each element differently by
providing different gains to the elements during either sound generation, reception,
or both. This weighting is called apodization. Here, we will examine the analogous
situation where we specify continuous amplitude profiles on the face of a large,
single element transducer.
For a single element transducer the far-field behavior was given by Eq. (3.18)
in terms of the Fourier transform of the velocity field on the face of the transducer,
V (k x ), where recall +∞

 V (k x ) = ∫ v ( x ′, 0, ω ) exp(−ik x ′)dx ′
−∞
z x
(3.60)

and k x = k sin θ . Now, let the velocity field be defined in a separable form by an
amplitude term, v0 (ω ), and a spatial distribution terms, C ( x′ ), i.e.
(3.61)
vz ( x ′, 0, ω ) = v0 (ω )C ( x ′ ).

Then
(3.62)
V (k x ) = v0 (ω )C (k x ),

where C (k x ) is the corresponding spatial Fourier transform of C ( x ′ ) given by

 +∞

C (k x ) = ∫ C ( x ′) exp(−ik x ′)dx ′.
−∞
x
(3.63)

Rewriting Eq. (3.18) in terms of this transform, the far-field pressure can be ex-
pressed as
 k exp (ikr0 ) (3.64)
p (x, ω ) = ρcv0 (ω ) C (k x ) .
2π i r0

Consistent with how we defined the total directivity of the array (see Eqs. (2.38)
and (2.39)) the far-field pressure can also be expressed in terms of the far-field
­directivity, Db (θ ) of the single element transducer as
 1 exp (ikr0 )
p (x, ω ) = ρcv0 (ω ) (2kb) Db (θ ) . (3.65)
2π i kr0
3.3 Amplitude Weighting 61

Fig. 3.10   The far field

directivity for a 12 mm long
element radiating at 5 MHz
into water. Solid line—piston
transducer, Dashed line—
cosine window

Comparing Eq. (3.65) with Eq. (3.64) we see that

 1
Db (θ ) = C (k x ). (3.66)
2b

For a piston transducer of radius b, C ( x ′ ) = rect(x ′, b), where

 1 − b < x ′ < b
rect(x ′, b) =  , (3.67)
0 otherwise

the spatial Fourier transform, C (k x ),then has the form (see Eq. (2.37) with v0 = 1)

 sin(k x b) sin(kb sin θ )

C (k x ) = 2b = 2b . (3.68)
kxb kb sin θ

Figure  3.10 plots this directivity for a 12  mm long ( b = 6 mm) piston transducer
radiating into water at 5 MHz. As is typically the case for a large single element
transducer or for an array, in the far field one will see a beam structure with a large
main lobe as well as smaller side lobes. It is possible to reduce the amplitude of the
side lobes appearing in Figure 3.10 relative to the main lobe by tailoring the spatial
amplitude behavior. This is generally the purpose behind amplitude weighting in a
phased array. For example, let the amplitude profile be given by
(3.69)
C ( x ′ ) = C0 cos(π x ′ / L)rect(x ′, b),

where C0 and L are two arbitrary constants. This profile is called a cosine window.
We can write this cosine window in the alternate form

C
C ( x ′ ) = 0 (exp (iπ x ′ / L) + exp ( −iπ x ′ / L))rect (x ′, b), (3.70)
2
62 3  Large, Single Element Transducer Models

so that the Fourier transform is given directly by

C  + b   π  
+b
 π   
C (k x ) = 0  ∫ exp i  − − k x  x ′  dx ′ + ∫ exp i  − k x  x ′  dx ′ 
(3.71)
2  − b   L   −b  L   
 sin (k x b + π b / L) sin (k x b − π b / L) 
= bC0  + .
 (k x b + π b / L) (k x b − π b / L) 

Normally one chooses L = 2b so that the amplitude is tapered to zero at x = ±b . In

that case, since sin (k x b ± π / 2) = ± cos (k x b) we find

 sin(k x b + π / 2) sin(k x b − π / 2) 
C (k x ) = bC0  + 
 (k x b + π / 2) (k x b − π / 2) 
(3.72)
4bC0 cos(k x b) 2C0 L cos(k x L / 2) .
= =
π 1 − (2k x b / π ) 
2
π 1 − (k x L / π ) 2 

If we let C0 = π / 2 from Eq. (3.66) we see that we will generate a “normalized”

directivity associated with this case that will have a maximum amplitude of one
at k x = 0 , just as in the piston case. This normalized directivity is also plotted in
Fig. 3.10 so that it can be compared to the directivity of the uniform amplitude (pis-
ton) case plot in that figure. It can be seen that the cosine amplitude weighting does
indeed reduce the amplitude of the side lobes, but Fig. 3.10 also shows that with this
side lobe reduction comes some broadening of the main lobe. Thus, other types of
weighting functions (windows) may be considered that have different effectiveness
in balancing these two effects. For example, the continuous Hanning amplitude
window function (also known as the Hann window, the raised cosine window or the
cosine squared window) is given by
C ( x ′ ) = C0 cos 2 (π x ′ / L)rect (x ′, L / 2)
(3.73)
C
= 0 [1 + cos (2π x ′ / L) ] rect (x ′, L / 2),
2

while the Hamming amplitude window is

 C ( x ′ ) = C0 [ 0.54 + 0.46 cos (2π x ′ / L) ] rect (x ′, L / 2) (3.74)

and the Blackman amplitude window is

 C ( x ′ ) = C0 [ 0.42 + 0.5cos (2π x ′ / L) + 0.08cos (4π x ′ / L) ] rect (x ′, L / 2). (3.75)

Figure  3.11 shows each of these window functions. We have written these three
windows in terms of a length, L. For the single element transducer case one general-
ly sets L = 2b. With this choice all of these windows (except the Hamming window)
tapers to zero at the ends of the element, as can be seen in Fig. 3.11. However, as
3.3 Amplitude Weighting 63

Fig. 3.11   Plots of the cosine

window ( solid line), Hanning
( cosine-squared) window
( dashed line), Hamming
window ( dashed-dotted line),
and the Blackman window
( dotted line). In all cases the
constant C0 = 1

shown in the next Chapter, for an array which uses discrete versions of these ampli-
tude weights, the tapering to zero is normally done at the centroids of the first and
last elements, which leads to a different choice for L. The spatial Fourier transforms
of all these windows can be obtained in the same fashion as done for the cosine
window (by writing the cosine terms in the form of a sum of complex exponentials
which allows us to perform the Fourier transforms directly). We find:
Hanning window:

 L sin (k x L / 2)  1 
C (k x ) = C0  2 
, (3.76)
2 (k x L / 2) 1 − (k x L / 2π ) 

Hamming window:
 L sin (k x L / 2) 1.08 − 0.16 (k x L / 2π ) 2 
C (k x ) = C0  , (3.77)
2 (k x L / 2)  1 − (k x L / 2π ) 2 

Blackman window:
 L sin (k x L / 2)  (k x L / 2π ) 2 0.16(k x L / 4π ) 2  .
C (k x ) = C0  0.84 + −  (3.78)
2 (k x L / 2)  1 − (k x L / 2π ) 2 1 − (k x L / 4π ) 2 

Another commonly discussed window is the triangular window (Fig. 3.12) where


C ( x) = C0 (1 − 2 x ′ / L)rect ( x ′, L / 2). (3.79)

To obtain the spatial Fourier transform of this window, it is convenient to use a

property of the Fourier transform involving differentiation, namely

d n C ( x ′ )
+∞ +∞
1
C (k x ) = ∫ C ( x ′ ) exp ( −ik x x ′ ) dx ′ = ∫ dx ′ n exp (−ik x x ′) dx ′. (3.80)
 −∞
(ik x ) n −∞
64 3  Large, Single Element Transducer Models

Fig. 3.12   A triangular win-

dow function

This is useful for the triangular window since two derivatives of that window gives
a form in terms of delta functions only:
 d 2 C ( x ′ ) 2C0
= δ ( x ′ + L / 2) − 2δ ( x ′ ) + δ ( x ′ − L / 2) (3.81)
dx ′ 2 L

whose spatial Fourier transform can be obtained directly to give

 −2C0
C (k x ) =
k x2 L
[exp(ik x L / 2) − 2 + exp(−ik x L / 2)]. (3.82)

But the term in brackets in Eq. (3.82) can be rewritten in terms of the square of dif-
ference of two terms, i.e.
 −2C0
C (k x ) = 2
kx L
[ exp(ik x L / 4) − exp( −ik x L / 4) ] ,
2
(3.83)

which then can be put in the even simpler form:

2

(3.84) C L  sin(k x L / 4) 
C (k x ) = 0   .
2  kx L / 4 

To compare the performance of these various windows we have plotted in Fig. 3.13

their normalized directivities, DN , obtained by choosing the constant C0 so that
the magnitude of the directivities were all equal to one at θ = 0 . Figure 3.13a com-
pares the rectangular (piston) window with the cosine and triangular windows,
while Fig. 3.13b compares the rectangular window with the Hanning, Hamming,
and Blackman windows. From the amplitude curves of Fig. 3.13 one can calculate
the − 6 dB angular width of the main lobe of the far field response and also deter-
mine how small the amplitude of the first lobe is relative to the main lobe amplitude.
Table 3.1 summarizes these results.
3.4 Multi-Gaussian Beam Model 65

Fig. 3.13   The normalized directivities for different amplitude weighting windows. a Comparison
of the rectangular window ( solid line), cosine window ( dashed line), and triangular window ( dot-
ted line). b Comparison of the rectangular window ( solid line), Hanning window ( dashed line),
Hamming window ( dash-dot line), and Blackman window ( dotted line).

Table 3.1   The − 6 dB angular width of the far field directivity main lobe and the amplitude of
the first side lobe relative to the amplitude of the main lobe for different choices of the window
function
Window function θ −6B width of main lobe First side lobe amplitude (dB)
Rectangular sin −1 (0.600λ / L) − 13.3
Cosine sin −1 (0.817 λ / L) − 23.0
Triangular sin −1 (0.885λ / L) − 26.5
Hanning sin −1 (0.997 λ / L) − 31.5
Hamming sin −1 (0.905λ / L) − 44.0
Blackman sin −1 (1.149λ / L) − 58.1

Generally, Table 3.1 shows that a window that reduces the amplitude of the side
lobes also increases the width of the main lobe. The Hamming window, however, is
an exception to that rule as it has both a smaller main lobe width and first side lobe
amplitude in comparison to the Hanning window. The windows discussed in this
section are the “classical” functions often described in the literature. However, there
are alternative windows that may be better for a particular application [5].

3.4 Multi-Gaussian Beam Model

For large, single element transducers in addition to the Fresnel integral models one
can use a superposition of a small number of Gaussian beams to represent the radi-
ated field of a piston transducer. As will be shown here, multi-Gaussian beam mod-
els also can provide an effective tool for modeling both beam steering and focusing.
If we let the pressure p ( x, z , ω ) = P ( x, z , ω ) exp (ikz ) in Helmholtz’s equation
(Eq. (2.6)) we find
66 3  Large, Single Element Transducer Models


∂2 P ∂2 P ∂P
+ 2 + 2ik = 0. (3.85)
∂x 2
∂z ∂z

If the waves are all traveling in approximately the z-direction then we can make
the paraxial approximation, which is equivalent to the mathematical condition
[Schmerr-Song]
∂2 P
(3.86) ∂P ∂ 2 P
<< 2ik , ,
∂z 2 ∂z ∂x 2

and Eq. (3.85) becomes the 2-D paraxial wave equation for P given by:
 ∂2 P ∂P
+ 2ik = 0. (3.87)
∂x 2
∂z

One solution of Eq. (3.87) is in the form of a Gaussian given by

  x2 
P( x, z ) = P ( z ) exp  ik . (3.88)
 2q ( z ) 

Placing this solution into Eq. (3.87) we obtain the two equations

dq
=1
dz
(3.89)
dP P
2 + = 0,
dz q

which can both be solved to find

q ( z ) = z + q0
(3.90) P0
P ( z ) = ,
q( z )

where ( P0 , q0 ) are constants. Thus, the pressure is given by a Gaussian beam of the
form
(3.91) P0  ikx 2 
p ( x, z , ω ) = exp (ikz ) exp  .
q( z )  2q ( z ) 

At z = 0 we see the pressure is given in terms of the ( P0 , q0 ) constants as

P  ikx 2 
(3.92)
p = 0 exp  ,
q0  2q0 
3.4 Multi-Gaussian Beam Model 67

which can be expressed in terms of constants ( A, B) as

 p
= A exp ( − Bx 2 / b 2 ) . (3.93)
ρcv0

Comparing Eqs. (3.92) and (3.93) gives the relations between these constants:
 −ikb 2
q0 =
2B (3.94)
P0 = ρcv0 A q0 .

Using Eq. (3.94) the propagating 2-D Gaussian beam of Eq. (3.91) can be written
in terms of ( A, B) as
 q0  ikx 2 
p ( x, z , ω ) = ρcv0 A exp (ikz ) exp  . (3.95)
z + q0  2( z + q0 ) 

Another way to obtain this propagating Gaussian beam directly is to use Eq. (2.29)
for the pressure in terms of a Hankel function, i.e.
+∞
(3.96) ωρ
p (x, ω ) = ∫ vz ( x ′, 0, ω ) H 0 (kr )dx ′.
(1)

2 −∞

If we replace the Hankel function by its high frequency asymptotic value to obtain
+∞
k ρc exp ( −iπ / 4) 2
(3.97)
p (x, ω ) =
2 ∫
−∞
vz ( x ′, 0, ω )
π kr
exp (ikr ) dx ′

and assume that the normal velocity on the surface at z = 0 is given by the Gaussian
velocity profile
(3.98)
vz = v0 (ω ) A exp ( − B ( x ′ ) 2 / b 2 )

then the pressure wave field is given by

+∞
 p (x, ω ) = k ρcv0 (ω ) exp ( −iπ / 4) A exp  − B ( x ′ ) 2 / b 2  2 exp (ikr ) dx ′, (3.99)
2 ∫ 
−∞
 π kr

where A, B, and b are constants, v0 (ω ) is a velocity spectrum, and the radius

r = ( x − x ′ ) 2 + z 2 . Now assume x, x ′ << z in this radius expression. This is
equivalent to saying all the waves are propagating in the z-direction, i.e. we are
again using the paraxial approximation. Then we have, approximately,
+∞
 p (x, ω ) k  B( x ′)2   ( x − x ′)2 
= A exp (ikz ) ∫ exp  −  exp ik  dx ′. (3.100)
ρcv0 (ω ) 2π iz −∞  b2   2z 
68 3  Large, Single Element Transducer Models

When the second exponential in Eq. (3.100) is expanded out we obtain

p (x, ω ) k  x2 
= A exp(ikz ) exp ik 
ρcv0 (ω )
(3.101) 2π iz  2z 
+∞
  k iB    x x′ 
⋅ ∫ exp i  + 2  ( x ′ ) 2  exp  −ik dx ′.
−∞   2 z b    z 

However, this integral can be evaluated exactly [Schmerr-Song]. After combining

terms and some algebra one finds

 p (x, ω ) −ikb 2 / 2 B  x2 
= A exp (ikz ) exp  ik  , (3.102)
ρcv0 (ω ) ( z − ikb 2 / 2 B)  2( z − ikb / 2 B ) 
2

which is of exactly the same form as Eq. (3.95) in terms of the constants A and B.
This result shows that if we generate a Gaussian velocity profile on z = 0 in the par-
axial approximation this Gaussian profile simply launches a Gaussian beam travel-
ing in the fluid for z > 0 . In the paraxial approximation the pressure and velocity,
vz , satisfy the plane wave relationship p = ρcvz [Schmerr-Song] so the velocity is
also in this same form:
 vz (x, ω ) −ikb 2 / 2 B  x2 .
= A exp (ikz ) exp ik (3.103)
 
v0 (ω ) ( z − ikb / 2 B )
2
 2( z − ikb / 2 B ) 
2

In 1988 Wen and Breazeale [6] showed that one could model the wave field of a
circular piston transducer radiating waves in three dimensions by superimposing ten
Gaussian beams having different complex coefficients ( An , Bn ) .These coefficients
were obtained with a non-linear least squares optimization procedure to find the
( An , Bn ) that best matched a constant piston profile on the transducer face given by

 vz 1 ρ / b ≤ 1
= circ(ρ 2 / b 2 ) =  , (3.104)
v0 0 otherwise

where ρ was a radial coordinate and b was the radius of the transducer. In 1990,
Wen and Breazeale improved the accuracy of their model with a slightly larger set
of 15 Gaussian beams [7]. If we simply set ρ = x we can use this same set of coef-
ficients for our 2-D problem for a 1-D transducer of length 2b.
Using Eq. (3.102) and the 15 Wen and Breazeale coefficients the pressure wave
field in our 2-D problem can be written compactly as

 15
(q0 ) n  ikx 2 
p ( x, z , ω ) = ∑ ρcv0 An exp (ikz ) exp  , (3.105)
n =1 z + (q0 ) n  2( z + (q0 ) n ) 

where
3.4 Multi-Gaussian Beam Model 69

Fig. 3.14   Comparison of the

multi-Gaussian beam model
with the Fresnel integral
model for the transducer
parameters; b = 6  mm,
f = 5  MHz, c = 1500  m/s.
Shown is the cross-axis nor-
malized pressure at z = 60  mm
versus the distance, x. Solid
line—Fresnel integral model,
Circles—multi-Gaussian
beam model (see Fig. 3.2b
for a similar comparison)

 −ikb 2
(q0 ) n = . (3.106)
2 Bn

The MATLAB® function Gauss_2D(b, f, c, x, z) in Code Listing C.10 uses

Eq. (3.106) and the 15 Wen and Breazeale coefficients obtained from the MAT-
LAB® function gauss_c15 (Code Listing C.11) to implement a 2-D multi-Gaussian
beam model for a piston transducer of length 2b. The calling sequence for this
function is
>> p = Gauss_2D(b, f, c, x, z);
where 2b is the element length (in mm), f is the frequency (in MHz), c is the wave
speed (in m/s), and (x, z) is a point in the fluid (in mm) at which the n­ ormalized
pressure, p / ρv0, is calculated. Figure 3.14 shows a comparison of a beam profile
calculated with the multi-Gaussian beam model and the Fresnel integral model.
Both beam models are based on the paraxial approximation but recall in Fig. 3.2b
a similar comparison was also made with a multiple line source model. From these
figures it can be seen that both paraxial models accurately predict the wave field of
the large (b = 6 mm) element. Similar results can be found with beam models based
on the paraxial approximation for points about a diameter away from the face of the
element [Schmerr-Song].
Now, consider the case where a linear phase variation of the form exp(ikx′ sin Φ )
is applied to a Gaussian velocity profile Av0 exp( − B( x ′ ) 2 / b 2 ) on the surface of
a 1-D transducer. Then a Gaussian beam of exactly the same form as found in
Eq. (3.102) is generated traveling in the z -direction with amplitude Av0 / cos Φ and
where b → b cos Φ. This result can be obtained directly by using this velocity pro-
file in Eq. (3.97):
k ρcv0 (ω ) exp( −iπ / 4) A
p (x, ω ) =
2
(3.107) +∞
2
· ∫ exp (ikx ′ sin Φ ) exp  − B( x ′ ) 2 / b 2  exp (ikr )dx ′.
−∞
π kr