IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL, VOL. 42, NO.

3, MAY 1995 429

Synthetic Aperture Imaging for Small Scale Systems

Mustafa Karaman, Member, IEEE, Pai-Chi Li, and Matthew O’Donnell, Fellow, IEEE

Abstract— Multi-element synthetic aperture imaging methods this work is that the same digital technology developed for
suitable for applications with severe cost and size limitations high channel count systems can be used to dramatically reduce
are explored. Array apertures are synthesized using an active the size and cost of low channel count systems capable of
multi-element receive subaperture and a multi-element trans-
mit subaperture defocused to emulate a single-element spatial reasonable image quality.
response with high acoustic power. Echo signals are recorded Personal computer based and pocket size hand-held imaging
independently by individual elements of the receive subaperture. devices may have widespread applications in medicine. In
Each method uses different spatial frequencies and acquisition particular, a hand-held scanner may complement a stethoscope
strategies for imaging, and therefore different sets of active trans- as a general diagnostic device. Achieving the physical size
mit/receive element combinations. Following acquisition, image
points are reconstructed using the complete data set with full and reduced power consumption of such a device requires a
dynamic focus on both transmit and receive. Various factors relatively low channel count system. In this paper we explore
affecting image quality have been evaluated and compared to synthetic aperture imaging methods producing reasonable im-
conventional imagers through measurements with a 3.5 MHz, age quality with a very small active channel count. These
128-element transducer array on different gel phantoms. Multi- methods should be directly applicable to hand-held system
element synthetic aperture methods achieve higher electronic
signal to noise ratio and better contrast resolution than conven- design.
tional synthetic aperture techniques, approaching conventional A conventional phased array (PA) imager excites all array
phased array performance. channels to form a transmit beam [9]. Similarly, every channel
contributes to receive beam forming. For an -element phased
array, the -scan signal, , and continuous wave angular
I. INTRODUCTION point spread function, , are

A POWERFUL imaging system for real-time medical ap-

plications must be agile yet maintain good image quality
at low cost. Agility includes selectable scan formats, Doppler
(1)

and color flow processing, and advanced signal and image

processing packaged in a physical size consistent with a (2)
highly mobile device. Image quality is primarily determined
by penetration (i.e., electronic signal to noise ratio) and both where and , respectively, are the echo signal com-
spatial and contrast resolution. Finally, cost is a function of ponent and beam forming delay for the receive, transmit
system complexity and hence limits both image quality and element combination, , and is the element
agility. Practical system design, therefore, always reduces to a separation. The first and second summations correspond to
trade-off between image quality and system complexity. transmit and receive beam forming. Although only the re-
The widespread availability of application specific inte- ceiver is dynamically focused in such systems, full phased
grated circuits (ASIC’s) has changed phased array front-end array imaging results in high electronic signal to noise ratio
design [1]–[3]. Digital beam formers now permit high preci- as well as good spatial and contrast resolution. The price
sion imaging with a very simple system architecture [4]–[8]. for this is many parallel transmit and receive circuits. For
As these devices increase in complexity and decrease in cost, example, today’s state of the art 128 channel scanners have 128
higher channel count systems will be possible with the same independent transmit and receive circuits. Even given modern
overall cost and agility as current commercial scanners. In electronic integration, such a system cannot meet the power
this paper we explore an alternate application of digital beam and size requirements of a hand-held system.
forming: hand-held systems. The fundamental hypothesis of An alternate approach is synthetic aperture imaging where
a large aperture is synthesized by moving, or multiplexing, a
Manuscript received August 22, 1994; revised November 4, 1994; accepted small active aperture over a large array [10]–[12]. For an -
November 16, 1994. This work was supported in part by Turkish Scientific element synthetic aperture with one active element (SAF), the
and Technical Research Council (TUBITAK), General Electric, Acuson, and
the National Institutes of Health under grant CA 54896. beam sum and continuous wave array response are
M. Karaman is with the Bioengineering Program, University of Michigan,
Ann Arbor, MI 48109-2122.
P.-C. Li is with Acuson Corporation, Moutain View, CA. (3)
M. O’Donnell is with the Electrical Engineering and Computer Science
Department and Bioengineering Program, University of Michigan, Ann Arbor,
MI 48109-2122. (4)
IEEE Log Number 9409917.

0885–3010/95$04.00  1995 IEEE

430 IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL, VOL. 42, NO. 3, MAY 1995

where and , respectively, are the echo signal and ods are based on aperture synthesis using an active multi-
beam forming delay for the th element. Synthetic aperture element receive subaperture and single element transmit syn-
methods are used in many applications such as radar [13], thesis using an appropriately phased subaperture. A specific
[14], sonar [15], [16], and ultrasonic imaging [17]–[25]. They multi-element approach is presented producing significantly
are very effective for special applications such as catheter higher electronic signal to noise ratio and contrast resolu-
based imaging where both system and probe simplicity are tion compared to conventional synthetic aperture imaging.
mandatory [26], [27]. SAF is attractive for a hand-held device Although not rivaling the overall performance of a full phased
because: 1) cost and power consumption are low since the array operating over the same total aperture, this system may
active channel count is low; and 2) higher spatial resolution represent a reasonable approach for hand-held applications.
is possible since full dynamic focusing is performed on both In the next section the complete multi-element system,
transmit and receive. Unfortunately, synthetic aperture imag- along with a slight variant on the basic approach, is presented
ing has some serious flaws for real-time medical applications, together with phasing of a multi-element transmitter to emulate
including: 1) low electronic signal to noise ratio due to the a single element radiation pattern with increased acoustic
small active transmit aperture; 2) poor contrast resolution power. Section III covers the experimental investigation of
due to significantly higher side lobes compared to a phased multi-element synthetic aperture methods including a compar-
array system; 3) susceptibility to motion artifacts since an ison to conventional imagers. Limitations of the experimental
aperture must be reconstructed from multiple firings; and 4) studies and further possible improvements in the performance
increased data storage since multiple rf -Scans must be of multi-element synthetic aperture methods are discussed in
buffered prior to beam forming. In the remainder of this paper Section IV.
we explore variations on traditional synthetic aperture methods
to minimize these problems without a major sacrifice in system II. MULTI-ELEMENT SYNTHETIC APERTURE
simplicity.
A major constraint on all ultrasound systems is frame rate. As an alternative to conventional SAF and phased array
High frame rates must be maintained for a wide range of imaging, we have explored multi-element synthetic aperture
medical applications. In a conventional phased array imaging methods with small active element count but with image
system there is a simple trade-off between frame rate and the quality approaching phased array systems. Two methods have
number of beam lines per frame. For each beam, the time been investigated: multi-element synthetic aperture focusing
needed to acquire all acoustic echoes to a depth is simply (M-SAF) and multi-element synthetic phased array (M-SPA).
, where is the sound velocity. If a particular frame rate Each method uses a defocused active transmit subaperture
is required, then the number of beams per frame is determined. of elements to emulate a high power, single element
That is, the number of beams per frame is simply transmitter. Echoes are recorded by individual elements of
a element receive subaperture. These subapertures are
(Frame period) (5) stepped across an -element array. To synthesize the complete
aperture, similar to SAF, image points are reconstructed with
To adequately characterize the sound field produced by an
full dynamic focus on both transmit and receive. Each method
array aperture, beams must be spaced at intervals satisfying
uses different spatial frequencies and acquisition strategies for
the spatial Nyquist criterion. For example, an array with
imaging, and therefore different sets of active transmit/receive
half wavelength elements needs independent pulse-
subapertures.
echo beams to satisfy the Nyquist sampling criterion over
a conventional 90 sector, where is the number of array
elements. Therefore, there is a direct trade-off between active A. Element Synthesis
aperture size (i.e., spatial resolution) and frame period. Using The acoustic power transmitted from a single array element
this analysis, the maximum numerical aperture for a 90 sector limits the electronic signal to noise ratio (SNR) of SAF.
imaged at 20 frames per second to a depth of 200 mm is 68 Although beam forming from multiple transmissions increases
(i.e., 136 for half wavelength element spacing), assuming SNR, it is still not acceptable for many applications, especially
no parallel beam forming. for medical imaging. To improve SNR, multiple elements can
For synthetic imaging, a similar constraint on aperture size be fired simultaneously with or without aperture apodization
exists. Again, the number of array firings is restricted by (Fig. 1). Simultaneously firing multiple elements limits the
the expression given in (5). For synthetic aperture imaging, angular response and can be efficiently used only for linear
however, a firing corresponds to excitation of a sub-array scanning and imaging with circular arrays; is not very effective
usually emulating the radiation pattern from a single element for sector scanning with linear arrays. Although aperture
and reception by a single element, or a small subset of apodization, such as with a “sinc” type window, results in
elements. Consequently, for conventional synthetic aperture a transmit beam similar to a single element response, the
imaging with a single active element, a numerical aperture of acoustic power capacity of each firing element is not fully
96 (i.e., 192 for half wavelength element spacing) can utilized and the total transmitted power, as expected, is very
be synthesized for imaging to a depth of 200 mm at 20 frames close to that of a single element.
per second. We synthesize a single element response with relatively
In this study we propose synthetic aperture imaging tech- high acoustic power by properly phasing the active transmit
niques consistent with these frame rate restrictions. All meth- elements without introducing any amplitude scaling. This
KARAMAN et al.: SYNTHETIC APERTURE IMAGING FOR SMALL SCALE SYSTEMS 431

of the 12 element defocused pattern (Fig. 2(d)) represent

small magnitude, early arriving waves of little consequence
in subsequent beam forming.
These results show that the synthesized response for simple
defocusing is very close to the single-element response for
small values of , such as 12. For large subapertures,
however, the response deviates significantly from a cylindrical
wavefront, especially in range. Although fixed defocusing de-
lays make the transmit beam range dependent, especially in the
near field of the subaperture, this dependence is insignificant
in the far field.
The defocused beam intensity, as expected, is lower than
the focused beam intensity produced by the same subaper-
ture. Fig. 3 illustrates the normalized mean intensity of the
defocused subaperture for different element counts. These
results are obtained from pulsed simulations with the same
Fig. 1. Different transmit schemes: (a) single element and five elements, (b) parameters outlined above ( and ). The
simultaneously excited, (c) simultaneously excited with aperture apodization,
and (d) defocused. mean is computed on a spatial kernel with a size of 90-
degrees and 0.5 along azimuth and depth, respectively, and
normalized to that of the single element beam. The functional
phasing forms a defocusing lens on the subaperture consisting relation between the mean intensity and subaperture size can
of the active elements (Fig. 1). A parabolic defocusing lens be approximated as for the sake of simplicity. Based
can be realized using the following delays on this approximation, SNR improvement for multi-element
firing over single-element firing is 10 dB. Note again
that all simulation results are obtained at a defocal length
(6)
of 2 with 2. The functional relation
between the mean intensity and subaperture size, as expected,
where is the distance of the th element from the subaper- approaches a linear relation as the defocal length increases.
ture center, and is the distance of the “defocal” point from On the other hand, the mean intensity will be quite low for
the subaperture. To produce a transmit beam as wide as the very wide defocused beams. In general, SNR improvement by
full sector image, the value of must be chosen properly. The a defocused active subaperture is limited by the defocusing
lateral spread of the defocused beam, as expected, is inversely angle.
proportional to the defocal length where the beam angle can be
approximated by and is the interelement
spacing. In this study, we consider a 90 sector image format, B. Multi-Element Synthetic Aperture Focusing (M-SAF)
and therefore, is chosen as . This value is used for This method uses spatial frequencies corresponding to re-
all defocused transmit beams discussed here. ceive, transmit combination , where the active receive
The overall pressure field transmitted from the active sub- elements are restricted to the range .
aperture is the sum of fields from the individual elements Subapertures are stepped across the array one element at a
with defocusing delays. The pulse response of the defocused time with coincident transmit and receive subapertures. Active
subaperture is tested through simulations using a Gaussian element combinations contributing to the beam sum and data
pulse, , where is the ultrasound frequency. Sim- acquisition are illustrated in Fig. 4 and Fig. 5(b), respectively.
ulations of different element counts are obtained assuming For an element array, the -scan and continuous wave
each element is a point radiator with . Fig. 2 array response of M-SAF are
shows the defocused beam patterns for 4 (c) and
12 (d) together with the beam patterns of a single
element (a) and simultaneously excited 12 (b) element (7)
aperture. This figure represents the point spread function of
the transmitter at a range of 200 from the array. Both
magnitude and phase over 1 1 and 193
207 are given. In this format, a perfectly
spherical radiator should produce a uniform horizontal band in (8)
magnitude and constant horizontal phase bands. As expected,
if multiple elements are fired simultaneously (Fig. 2(b)), a
spherical wave is emulated only over a limited angular extent. where is the parabolic defocusing delay of the th channel
In contrast, phase defocusing can produce radiation patterns and the spatial response of the defocused element transmit
closely approximating a spherical wave, especially over a 90 subaperture is assumed to be omnidirectional. Note that the
sector. The highly nonspherical phase fronts at the beginning beam forming delay, , is the same for all elements on the
432 IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL, VOL. 42, NO. 3, MAY 1995

(a)

(b)

(c)

(d)
Fig. 2. Simulated one-way pulsed response of the subaperture. The magnitude and phase distributions have 60 dB and 2 dynamic display ranges, where
white and black gray levels correspond to the maximum and minimum values, respectively. (a) Kt = 1, (b) Kt = 12 with simultaneous excitation,
and (c) Kt = 4 and (d) Kt = 20 with defocusing.

transmit subaperture because they synthesize a single element The first term in is due to the full element
located at the subaperture center. The summation over transmit aperture. The second term results from the element receive
subaperture (over index ) is realized in transmission, and subaperture with processing to suppress grating and side lobes
hence is not explicitly performed in synthesis. due to the first term (see Fig. 6(a)). The equivalent aperture
KARAMAN et al.: SYNTHETIC APERTURE IMAGING FOR SMALL SCALE SYSTEMS 433

C. Multi-Element Synthetic Phased Array (M-SPA)

This method synthesizes the nonredundant spatial frequen-
cies of a phased array imager. It can produce full phased
array images but with SNR loss and longer acquisition times.
For an -element phased array, there are possible trans-
mit/receive combinations. However, almost half are redundant;
the signals and are dual, and thus one of them can be
ignored yielding a total of 2 independent record-
ings. To efficiently capture all independent spatial frequencies,
receive/transmit element combinations are sampled as shown
in Fig. 4, where the number of receive channels is either
or on all firings.
Data acquisition for this method is illustrated in Fig. 5(c).
On each firing, elements are excited with a defocusing
pattern emulating a single element radiation pattern emanating
from the center of the subaperture. Similarly, elements
simultaneously receive echoes on each firing. Since is
Fig. 3. Mean intensity of simulated defocused transmit beam for different much less than the total number of elements in the array,
6
subaperture sizes. The mean is computed over beams within =4 and 6 multiple firings must be performed at the same transmitter
0.25  in azimuth and range, respectively, from the beam center. position. Over these firings, the position of the active
receivers is moved to capture all nonredundant transmitter-
receiver pairs for that transmitter position. The center of the
transmit subaperture is then stepped by one array element,
and the process repeated until all independent transmit-receive
pairs across the array are captured.
With this scheme, the total firings per frame is

(firings/frame) (10)

where indicates the smallest greater integer. At worst, one

active channel is idle for every two data acquisition steps. The
beam sum and lateral array response of M-SPA are

(11)

(12)

where
if
(13)
else.
The effective aperture size of the synthesized phased array
is . The for a particular set of
Fig. 4. Active transmit/receive element combinations contributing to the parameters is plotted in Fig. 6(a) and the equivalent aperture
beam sum. domain function is given in Fig. 6(b).
For real-time imaging, the number of active elements in
domain function, corresponding to the Fourier transform of receive must be chosen properly to maximize such that
, is symbolically illustrated in Fig. 6(b), where PA and
SAF patterns are also shown for comparison. In the limiting (14)
(frames/s) (range/firing)
case, M-SAF with 1 is identical to SAF.
For real-time operation, the synthesized aperture size and For 20 frames/s, 200 mm range, and 1540 m/s, the
the active element count of M-SAF must be chosen such that firings/frame must be less than or equal to 192, and thus the
corresponding combinations are (3, 32), ( 7, 48),
(9) (11, 64), (21, 80), and (25, 96). These particular combinations
(frames/s) (range/firing)
satisfy the above constraint and minimize the number of idle
For example, at 20 frames/s and 200 mm range, active channels. Note that to synthesize an -element phased
must be less than or equal to 192. array the actual array size must be 1.
434 IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL, VOL. 42, NO. 3, MAY 1995

(a)

(b)

(c)
Fig. 5. Data acquisition strategies: (a) and (b) active transmit and receive subapertures, and the acquisition steps of (c) M-SAF and (d) M-SPA.

D. Outline of System Parameters where SNR is the SNR of a single element synthetic aperture
Table I outlines some of the factors affecting overall per- recording. An SNR improvement of 10 dB is expected
formance for the different techniques. The SNR is computed for a multi-element transmitter, but the defocusing reduces
assuming uncorrelated additive electronic noise on receive, this gain to about 10 dB (estimated experimentally).
KARAMAN et al.: SYNTHETIC APERTURE IMAGING FOR SMALL SCALE SYSTEMS 435

TABLE I
SYSTEM PARAMETERS COMPARING METHODS WHERE Ne = N 0 Kr +1 AND B = v=f(frames)2
g
2 (range=firing ) WITH A 90 SECTOR ANGLE AND A MINIMUM f/NUMBER OF 2
PA M-SPA M-SAF SAF
Transmit Channel Complexity N Kt Kt 1
Receive Channel Complexity N Kr Kr 1
SNR/SNR0 (dB) 10 log N3 10 log (Kt Ne (Ne + 1)=2) 10 log (Kt Kr Ne ) 10 log (N)
Transmit Focusing fixed dynamic dynamic dynamic
Storage __ __ Ne (Ne + 1)=2 Kr N e N
Ne (Ne + 1)=2 Kr N e
Beam Sums
p2N  B
N
Ne  B
N
Maximum Aperature Size Ne j N2K+2 j  B N≤B
Motion Compensation —— required required required

flexible beam forming through digital processing. In particular,

adaptive beam forming employing aberration correction and
adaptive spatial filtering for image enhancement can be im-
plemented efficiently using the stored records. The proposed
techniques involve relatively more storage and coherent ad-
ditions. Also, synthetic aperture imaging necessitates motion
compensation for real-time applications.

III. EXPERIMENTAL INVESTIGATION

The proposed synthetic aperture methods are investigated
experimentally and compared to conventional approaches us-
ing rf data acquired from three different phantoms. A phantom
containing only six wire targets is used to measure beam
patterns and SNR. Another phantom containing diffuse scat-
terers and four cyst regions is used to test contrast resolution
by computing the average contrast to noise ratio (CNR) for
the cysts. The contrast to noise ratio is an appropriate way
to quantitate contrast resolution in speckle generating objects
[28]. Additionally, a standard ultrasound resolution phantom,
(a) which contains wire and cyst objects as well as diffuse
scatterers, is used to test image quality in general. Throughout
this paper, these phantoms are referred to respectively as: Wire
Phantom, Cyst Phantom and Standard Phantom.
For our experimental studies, phased array imaging with
both dynamic transmit and dynamic receive focusing is con-
sidered the “gold-standard” (GS-PA) achieving the best image
quality with a finite array size. -scan images of each phantom
for all imaging methods including GS-PA are reconstructed for
qualitative comparison. Mean and standard deviation of SNR
and CNR results over the wire and cyst targets, respectively,
for different aperture sizes and/or active element counts are
presented to illustrate general trends including spatial variation
of the results. All SNR and CNR results are normalized by
GS-PA counterparts.

A. Data Acquisition
(b) Complete rf data sets from each phantom are acquired
Fig. 6. (a) Simulated point spread functions for continuous wave excitation with an experimental system using a 3.5 MHz, 128-element
with d = =2 at a steering angle of =4. (b) Equivalent aperture functions transducer array with 0.22 mm inter-element spacing (Acu-
corresponding to the point spread functions shown in (a). son, Model #: V328). All possible rf -scans from every
transmit/receive element combination, a total of 16 K records,
System complexity, defined as the number of active channels, are collected, where only a single element is active both
for multi-element imaging is reduced by a factor of in transmit and receive. Each -scan is sampled at 13.89
and , respectively in transmit and receive, compared MHz with 12-bits, and has 2 K samples corresponding to a
to phased array imaging. Synthetic aperture imaging enables spatial range of about 120 mm. To reduce noise imposed by
436 IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL, VOL. 42, NO. 3, MAY 1995

Fig. 7. B -scan images of wire phantom: (a) GS-PA, (b) PA, (c) M-SPA, (d) Fig. 8. B -scan images of cyst phantom: (a) GS-PA, (b) PA, (c) M-SPA, (d)
SAF and (e) M-SAF. SAF and (e) M-SAF.

C. -Scan Images
the data acquisition system, signal averaging over successive
Wire Phantom: Fig. 7 shows -scan sector images of the
acquisitions was employed with the final data stored as 16-
Wire Phantom displayed over a 50 dB display dynamic range
bit words. All beam forming and image reconstruction are
where each image corresponds to a different technique. The
carried out digitally to emulate a digital imaging system. Also,
sampled rf data were filtered with a digital bandpass filter for image of each point target represents the point spread func-
additional noise removal. tion of the imaging system at that location. All side lobes
above 50 dB are visible. A comparison of GS-PA and PA
images illustrates the effects of fixed/dynamic transmit focus,
B. Image Reconstruction especially in the near field. Although the aperture size of
Although any arbitrary scan format can be used for syn- -SPA is approximately one half that of PA, its image
thetic aperture processing, all images presented here were quality approaches that of PA because of dynamic transmit
reconstructed on a standard sector scan grid with 200 beam focusing. -SAF and SAF have higher side lobes and finer
lines positioned at constant increments of over a 90 main lobes and phased array imagers. Large grating lobe
sector. Each rf -scan is digitally mixed with a complex artifacts are present in SAF images whereas grating lobes are
carrier followed by low pass filtering to obtain the complex virtually eliminated in -SAF. -SAF also exhibits reduced
baseband representation of the signal. Beams are formed by side lobes compared to conventional synthetic aperture. In
time delaying, phase rotating and summing these complex general, measured beam patterns closely match theoretical and
baseband signals using a standard dynamic aperture approach simulated predictions in previous sections.
with a minimum /number of 2. The output of the beam former Cyst Phantom: Fig. 8 shows -scan images of the Cyst
is a complex valued image in - format. -scan images Phantom over a 50 dB display dynamic range. The phantom
are produced from these data through envelope detection, gain was constructed from gelatin with nylon particles used as
compensation and scan conversion, followed by logarithmic ultrasonic scatterers at a concentration sufficient to produce
compression. fully developed speckle. Each cyst region does not contain
KARAMAN et al.: SYNTHETIC APERTURE IMAGING FOR SMALL SCALE SYSTEMS 437

Fig. 10. Mean intensity of six wire echoes as a function of transmit sub-
aperture size.

that of a synthetic phased array (M-SPA) because the spatial

frequency bandwidth of synthetic apertures is approximately
two times that of a synthesized phased array.
Standard Phantom: -scan images of the Standard Phan-
tom for each imaging technique are presented in Fig. 9.
The phantom is a tissue mimicking ultrasound resolution
phantom (RMI Model #412) containing wire targets, cysts and
distributed scatterers. A portion of the sector at the right side
is black because one wall of the phantom is within the sector.
This configuration is useful to test acoustic noise, where that
portion of the image will be completely black if the CNR is
infinite.
Fig. 9. B -scan images of standard phantom: (a) GS-PA, (b) PA, (c) M-SPA,
(d) SAF and (e) M-SAF. Phased array images (GS-PA, PA and M-SPA) illustrate
the effect of fixed/dynamic transmit focus on image quality.
Artifacts in the black portion arise both from acoustic and
any scatterers. The ringlike structure around each cyst has
electronic noise. PA has more artifacts in the dark region
slightly different scatterer density than the rest of the phantom.
compared to GS-PA. The fixed transmit focus decreases the
This slight concentration around the border was purposely
CNR, and also slightly degrades SNR. Hence these artifacts on
produced to highlight potential artifacts near the cyst border.
the PA image are due to both reduced CNR and SNR relative to
The diameter of each cyst is about 20 mm.
GS-PA. Comparison of M-SAF and SAF indicates that M-SAF
In general, M-SPA cyst detectability is close to those of GS-
almost completely eliminates grating lobe problems whereas
PA and PA. Because the transmit focus is fixed at midrange
SAF has serious artifacts due to grating lobes.
(95 mm) on the PA image, its contrast resolution away from
this position is slightly degraded compared to those of GS-PA
and M-SPA. Range dependent contrast resolution is clearly D. Intensity of Defocused Transmit Beam
observable in the cysts and the ringlike structures around the To test simulations presented in Section II-A, the mean
cysts. Low contrast detectability of -SAF is much better intensity from the six wires of the Wire Phantom is measured
than that of SAF, but it is not as good as those of phased for different transmit subapertures using a single receive
array imagers. Comparing M-SAF to conventional phased element. Since the wires are distributed in azimuth, the angular
array and synthetic aperture images, contrast resolution is response of the receiver slightly influences this measurement.
closer to phased array than to synthetic aperture. Clearly, the Such variation has little influence on the final results, though,
contrast resolution of SAF is the worst of all since the low and hence is ignored. The mean is computed over a 5 (20
contrast, ringlike structures cannot be differentiated from the samples) temporal kernel for each wire echo and normalized
cysts. Moreover, the cysts, especially the ones nearest to and to the mean echo value with a single transmitting element.
farthest from the array (at the top and bottom of the image), Fig. 10 illustrates the normalized mean intensity for differ-
are not easy to discriminate from speckle. ent defocused transmit subapertures. The functional relation
These images clearly illustrate that the average speckle between the mean intensity of wire echoes and subaperture
size of synthetic aperture (M-SAF and SAF) is smaller than size can be approximated as 1.25 . Note that a similar
438 IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL, VOL. 42, NO. 3, MAY 1995

(a) (b)

(c) (d)
Fig. 11. Expected and measured relative SNR: a) GS-PA and PA, b) SAF, c) M-SPA, and d) M-SAF.

relation between mean intensity and size, , is obtained 10 10 , 10 and 10

for the simulations given in Fig. 3. Therefore, the experiments , respectively, where focusing and defocusing effects
(Fig. 10) are in close agreement with simulations (Fig. 3), are ignored.
indicating that the improvement in transmitted beam intensity Experimental SNR results are obtained from the Wire Phan-
from a -element defocused subaperture is on the order of tom. To do this, -scan images without logarithmic compres-
, where the defocal length is one half of the subaperture sion are used. The signal power at each target is obtained from
size . This particular result depends on the the rms image brightness at pixels within a small kernel over
defocusing scheme and array parameters. that target. Similarly, the noise power is estimated from the
rms signal within a kernel including no reflectors and no side
lobe contributions. The SNR for each method is calculated at
E. SNR Results each target location, and normalized by the SNR of GS-PA
Theoretical SNR relations of multi-element transmit systems with 128. Then, the mean and standard deviation of the
(see Table I) cannot be tested directly using rf data acquired normalized SNR over all 6 targets are calculated. These com-
with a single firing element. Therefore functional SNR rela- putations are performed for different array and/or subaperture
tions for the emulated systems in this study are obtained and sizes. Fig. 11 depicts SNR results on a dB scale. Each point
tested experimentally. For our emulated system, the expected represents 20 , where and are
SNR for PA (same for GS-PA), M-SPA, M-SAF and SAF are the mean and standard deviation of the normalized SNR. Each
KARAMAN et al.: SYNTHETIC APERTURE IMAGING FOR SMALL SCALE SYSTEMS 439

figure also shows the expected SNR based on the functional TABLE II
relations given above. TYPICAL VALUES OF ARRAY SIZE AND ACTIVE ELEMENT
COUNTS, AND CORRESPONDING SNR AND CNR RESULTS
In general, measured SNR results match expected values
PA M-SPA M-SAF SAF
within the standard deviation for all imagers. The measured
Ne
Effective Aperture Size, 128 64 117 128
SNR of GS-PA matches expectations where the relatively K
Active Transmit Elements, t 128 11 12 1
small standard deviation indicates less spatial variation com- K
Active Receive Elements, r 128 11 12 1
pared to other schemes. The SNR of PA shows a different Expected Relative SNR (dB) 0 -20 -21 -42
characteristic with the standard deviation increasing with in- Measured Relative CNR 1 0.9 0.8 0.3
creasing aperture size. This variation is primarily due to the
fixed transmit focus and the fixed minimum f/number of 2.0.
of 3 elements from even to odd values. For example, CNR
The SNR of M-SPA increases with array size and subaperture
values for 8, 11, and 12 element subaperture are 0.708,
size up to 64 with 11, and stays almost
0.711, and 0.755, respectively. Even valued results in
constant thereafter. SNR improvement for M-SAF with
better suppression of grating and side lobes compared to odd
12 is not as significant as with 3 12. The SNR of
valued subapertures, since the cw radiation pattern from an
SAF has an almost constant standard deviation for different
even valued is exactly 0 at grating lobe positions. The
array sizes.
CNR of SAF is the worst of all schemes in terms of both
mean and standard deviation: the mean varies between 0.23
F. CNR Results
and 0.29 for array sizes between 16 and 128 elements while
To quantitatively test detectability of a low contrast lesion the standard deviation is within 23 to 35 percent of the
within a speckle generating object, the CNR of each method mean.
is investigated using the experimental data acquired from
the Cyst Phantom. Sufficient signal averaging was used for
IV. DISCUSSION AND CONCLUSION
these studies to insure that contrast measurements were not
influenced by electronic noise. CNR is calculated by dividing Multi-element synthetic aperture methods were investigated
the image contrast by the speckle standard deviation. All com- both theoretically and experimentally. The overall performance
putations are performed on a logarithmic scale, as discussed in of each method was evaluated quantitatively in terms of SNR
[28], where the contrast is defined as the logarithmic difference and CNR and qualitatively with -scan phantom images.
between mean values of image kernels fully within and outside Table I and Figs. 7–9 show that reasonable image quality can
the cysts. That is, be obtained for a system with about 10 active channels in both
transmit and receive. Such a system, integrated with an LCD
(15) display, can be produced in a small package appropriate for
personal computer based and hand-held scanners.
where is the mean intensity in dB of the cyst, and and Theoretical SNR and measured CNR for typical combi-
are the mean and standard deviation of the intensity in nations of array size and active element count are given in
dB within the speckle. The CNR is computed for each of the Table II, where the frame rates of M-SPA and M-SAF are
four cysts within the phantom using a 60 dB display dynamic about 20 and 30, respectively, for a 200 mm imaging depth.
range. The computed CNR for each method is normalized CNR results are normalized to the CNR of the 128-element
to that of the 128-element GS-PA. The mean and standard PA. Multi-element synthetic aperture imaging methods show
deviation of CNR over the four cysts are computed to include better performance than SAF in terms of SNR and CNR,
any spatial dependence. Fig. 12 presents the CNR for different and produce better images than standard synthetic aperture
array and/or subaperture sizes. imaging, approaching phased array quality.
Normalized CNR values of GS-PA and PA vary within 0.66 Another variable in synthetic aperture imaging not explicitly
to 1.0 with array size between 16 and 128, where considered in this manuscript is element spacing. Element
the dependence on array size is almost logarithmic for this spacing can be manipulated to change both spatial and contrast
particular cyst size. The standard deviation of CNR for GS- resolution. For example, the grating lobes present in a classic
PA increases with array size whereas that of PA increases with SAF image can be completely removed if spacing is used
array size up to 64 and then slightly decreases. The mean CNR instead of the spacing assumed here. Unfortunately, such
of M-SPA, as expected, is less than that of PA, especially for a system would have twice the number of elements. Generally,
small array sizes. For example, the CNR’s of M-SPA and PA element dimension must also be considered in optimizing
for an array size of 64 elements are 0.89 and 0.91, respectively. overall performance.
Spatial variation of CNR for M-SPA increases significantly All synthetic aperture imaging techniques, including the
with array/subaperture size. multi-element aperture methods discussed in this paper, move
The mean CNR of M-SAF varies between 0.40 to 0.88 system complexity from channel electronics to data processing.
for a subaperture size from 3 to 24. M-SAF shows a near Digital beam forming ASIC’s can be used effectively for
logarithmic CNR behavior with subaperture size. This figure real-time processing of synthetic aperture data. Multi-element
illustrates that differential CNR improvements achieved by synthetic aperture methods can also be implemented in various
a one element increase in the subaperture size, from odd alternate forms with different tradeoffs between complexity
to even values, is more significant compared to an increase and quality. To decrease complexity further, the active receive
440 IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL, VOL. 42, NO. 3, MAY 1995

(a) (b)

(c) (d)
Fig. 12. Measured relative CNR: (a) GS-PA and PA, (b) SAF, (c) M-SPA, and (d) M-SAF.

subaperture and/or synthesized array can be undersampled defocused composite pulse can be accurately measured for
[14]. This results in reduced active channel count and data image reconstruction points [30], and thus deviation from a
size while maintaining subaperture and/or synthesized array cylindrical phase front can be compensated by choosing the
sizes. However, image quality will be reduced due to grating beam forming delays accordingly. Recent studies on synthetic
lobes produced by undersampling. Furthermore, multi-element aperture imaging [26], [31] have demonstrated the efficiency of
synthetic aperture methods can also be realized with random such techniques. Consequently, spatial domain inverse filtering
array techniques [29]. In particular, active receive channels of multi-element synthetic aperture baseband data can be
can be distributed randomly over the array by reorganizing the effective to further suppress side lobes [26], [32].
active transmit/receive element combinations shown in Fig. 4. In general, contrast resolution is a function of main lobe
In this case, the point spread function for the multi-element to side lobe energies. Thus, detectability of a low contrast
synthetic focusing technique will be different than that given lesion depends on its size relative to the main lobe width.
in (8) while (12) still holds for the multi-element synthetic Measurements on the Cyst Phantom were based on fixed size
phased array. cysts and therefore do not include CNR variations due to cyst
The quality of multi-element synthetic aperture images can size. Measurements on the Wire Phantom produce the point
be improved further by minimizing the effects of imper- spread function of each system and therefore can be used to
fect transmit defocusing through beam pattern optimization predict the CNR for different sized cysts. Future studies will
and/or optimal filtering. In particular, the flight time of the quantitate CNR variations due to cyst size.
KARAMAN et al.: SYNTHETIC APERTURE IMAGING FOR SMALL SCALE SYSTEMS 441

Multi-element synthetic aperture methods necessitate an [13] R. P. Perry and L. W. Martinson, “Radar matched filtering,” in Radar
increased frame rate with limited aperture and imaging depth Technology, E. Brookner, Ed. Boston: Artech House, 1978, chap. 11,
pp. 163–169.
and/or motion compensation for real-time imaging. For imag- [14] A. Morcira, “Real-time synthetic aperture radar (SAR) processing with
ing to a 200 mm depth, M-SAF with 128 and M-SPA a new subaperture approach,” IEEE Trans. Geosci. and Remote Sensing,
with 64 and 11 achieve 30 and 20 frame/s, vol. 30, pp. 714–722, July 1992.
[15] N.-C. Yen and W. Carey, “Appliction of synthetic aperture processing
respectively. Should faster imaging be required, an M-SAF to a towed-array data,” J. Acoust. Soc. Amer., vol. 86, pp. 754–765,
imager with 64 can image to a 150 mm depth at 80 Aug. 1989.
frames/s, corresponding to a 12.5 ms acquisition time. Note [16] S. Stergiopoulos and E. J. Sullivan “Extended towed array processing
by an overlap correlator,” J. Acoust. Soc. Amer., vol. 86, pp. 158–171,
that the array size for M-SPA and M-SAF is different for July 1989.
real-time application. Additionally, motion and phase com- [17] C. B. Burckhardt, P.-A Grandchamp, and H. Hoffmann, “An experimen-
tal 2 MHz synthetic aperture sonar system intended for medical use,”
pensation techniques can effectively be applied to successive
IEEE Trans. Son. Ultrason., vol. 21, pp. 1–6, Jan. 1974.
firings of M-SAF [33]–[38]. These techniques can also be used [18] S. Ganapathy, W. S. Wu, and B. Schmult, “Analysis and design
for M-SPA with partially overlapped receive subapertures or considerations for a real-time system for nondestructive evaluation in
the nuclear industry,” Ultrason., pp. 9–15, Nov. 1982.
[19] K. Nagai, “A new synthetic-aperture focusing method for ultrasonic B -
randomly selected receive elements in successive acquisition.
Two-dimensional motion can be estimated by correlation- scan imaging by the Fourier transform,” IEEE Trans. Son. Ultrason.,
based block matching [39]–[42] using “low resolution” images vol. 32, pp. 531–536, July 1985.
[20] Y. Ozaki, H. Sumitani, T. Tomoda, and M. Tanaka, “A new system for
reconstructed from each receive subaperture. Note that the real-time synthetic aperture ultrasonic imaging,” IEEE Trans. Ultrason.,
final image of the multi-element synthetic aperture is the Ferroelect., Freq. Cont., vol. 35, pp. 828–838, Nov. 1988.
sum of these low resolution images. Therefore, a motion [21] Al Elgarem, “Multidepth synthetic aperture processing of ultrasonic
data,” IEEE Trans. Ultrason., Ferroelect., Freq. Cont., vol. 36, pp.
compensated final image can be obtained by correcting low 384–385, May 1989.
resolution images prior to summation. Preliminary studies [22] Forrest Anderson, “3D ellipsoidal backprojection images from large
have shown that the correlation coefficient between images arrays,” in Proc. 1992 IEEE Ultrason. Symp., 1992, pp. 1223–1226.
[23] L. F. Nock and G. E. Trahey, “Synthetic receive aperture imaging with
of two successive 12-element subapertures (11 overlapping phase correction for motion and for tissue inhomogeneities—Part I:
elements) is about 0.75 and 0.96 for speckle and wire targets, Basic principles,” IEEE Trans. Ultrason., Ferroelect., Freq. Cont., vol.
respectively. They also indicate that the accuracy of motion 39, pp. 489–495, July 1992.
[24] J. T. Ylitalo and H. Ermert, “Ultrasound synthetic aperture imaging:
estimation in range, as expected, is much higher than in Monostatic approach,” IEEE Trans. Ultrason., Ferroelect., Freq. Cont.,
azimuth. Future studies will focus on increasing the correlation vol. 41, pp. 333–339, May 1994.
coefficient between images used for correlation analysis and [25] S. D. Silverstein and L. J. Thomas, “Analytical comparison of sensor
signal processing enhancements for NDT synthetic aperture imaging,”
experimentally demonstrating the accuracy of motion com- IEEE Trans. Image Processing, vol. 2, pp. 60–67, 1993.
pensation. Proper motion compensation is the key to real-time [26] M. O’Donnell and L. J. Thomas, “Efficient synthetic aperture imaging
application of the multi-element synthetic aperture imaging from a circular aperture with possible application to catheter-based
imaging,” IEEE Trans. Ultrason., Ferroelect., Freq. Cont., vol. 39, pp.
methods presented in this paper. 366–380, May 1992.
[27] A. P. Proudian et al., “Aparatus and method for imaging small cavities,”
U.S. Patent 4,917,097, Apr. 1990.
REFERENCES [28] P.-C. Li and M. O’Donnell, “Improved detectibility with blocked
element compensation,” Ultrason. Imaging, vol. 16, pp. 1–18, 1994.
[29] B. D. Steinberg, Principals of Aperture and Array System Design. New
[1] M. E. Schafer and P. A. Lewin, “The influence of front-end hardware
York: J. Wiley, 1976.
on digital ultrasonic imaging,” IEEE Trans. Son. Ultrason., vol. 31, pp.
[30] M. Karaman, A. Atalar, and H. Köymen, “Optimization of dynamic
295–306, Jul. 1984.
receive focusing in ultrasound imaging,” in Acoustical Imaging, H.
[2] M. O’Donnell, “Applications of VLSI circuits to medical imaging,”
Proc. IEEE, vol. 76, pp. 1106–1114, Sept. 1988. Ermert and H. P. Harjes, Eds. New York: Plenum Press, 1992, vol. 19.
[3] A. L. Robinson and J. H. Mo, “Application of microelectronics and [31] P.-C. Li and M. O’Donnell, “Synthetic aperture imaging using a
microfabrication technology to ultrasound imaging systems,” in Proc. Lagrange based filtering technique,” Ultrason. Imaging, vol. 14, pp.
1992 IEEE Ultrason. Symp., 1992, pp. 681–691. 354–366, 1992.
[4] R. M. Lutolf, A. Vieli, and S. Basler, “Ultrasonic phased-array scanner [32] S. Freeman, P.-C. Li, and M. O’Donnell, “Real-time compensation
with digital echo synthesis for Doppler echocardiography,” IEEE Trans. of transmit beam defocusing using filtering techniques,” to appear in
Ultrason. Ferroelect. Freq. Cont, vol. 36, pp. 494–506, Sept. 1989. Ultrason. Imaging.
[5] M. O’Donnell et al., “Real-time phased-array imaging using digital beam [33] G. S. Edelson and E. J. Sullivan, “On the performance of the overlapped-
forming and autonomous channel control,” in Proc. 1990 IEEE Ultrason. correlator synthetic-aperture technique,” J. Acoust. Soc. Amer., vol. 90,
Symp., 1990, pp. 1499–1502. pp. 2000–2003, Oct. 1991.
[6] M. O’Donnell and W. E. Engeler, “Correlation-based aberration cor- [34] M. Soumekh and H. Yang, “Complex phase error and motion estimation
rection in the presence of inoperable elements,” IEEE Trans. Ultrason. in synthetic aperture radar imaging,” in Proc. SPIE, Image Process and
Ferroelect. Freq. Cont., vol. 39, pp. 700–707, Nov. 1992. Algo. Tech. II, vol. 1452, pp. 104–113, 1991.
[7] B. D. Steinberg, “Digital beamforming in ultrasound,” IEEE Trans. [35] G. E. Trahey and L. F. Nock, “Synthetic receive aperture imaging with
Ultrason. Ferroelect. Freq. Cont., vol. 39, pp. 716–721, Nov. 1992. phase correction for motion and for tissue inhomogeneities—Part II: Ef-
[8] M. Karaman, A. Atalar, and H. Köymen, “VLSI circuits for adaptive fects of and correction for motion,” IEEE Trans. Ultrason., Ferroelect.,
digital beamforming in ultrasound imaging,” IEEE Trans. Med. Imaging, Freq. Cont., vol. 39, pp. 496–501, July 1992.
vol. 12, pp. 711–720, Dec. 1993. [36] J. A. Fawcett, “Synthetic aperture processing for a towed array and a
[9] T. A. Shoup and J. Hart, “Ultrasonic imaging systems,” in Proc. 1988 moving source,” J. Acoust. Soc. Amer., vol. 94, pp. 2832–2837, Nov.
IEEE Ultrason. Symp., 1988, pp. 863–871. 1993.
[10] C. W. Sherwin, J. P. Ruina, and R. D. Rawcliffe, “Some early de- [37] S. W. Flax and M. O’Donnell, “Phase-aberration correction using signals
velopments in synthetic aperture radar systems,” IEEE Trans. Military from point reflectors and diffuse scatterers: Basic principles,” IEEE
Electron., vol. 6, pp. 111–115, Apr. 1962. Trans. Ultrason. Ferroelect. Freq. Cont., vol. 35, pp. 758–767, Nov.
[11] W. C. Curtis, “Synthetic aperture fundamentals,” in Radar Technology, 1988.
E. Brookner, Ed. Boston: Artech House, 1978, chap. 16, pp. 231–237. [38] D. Zhao and G. E. Trahey, “Comparisons of image quality factors for
[12] R. N. Thomson, “Transverse and longitudinal resolution of the synthetic phase abberation correction with diffuse and point targets,” IEEE Trans.
aperture focusing technique,” Ultrason., pp. 9–15, Jan. 1984. Ultrason. Ferroelect. Freq. Cont., vol. 38, pp. 125–132, Mar. 1991.
442 IEEE TRANSACTIONS ON ULTRASONICS, FERROELECTRICS, AND FREQUENCY CONTROL, VOL. 42, NO. 3, MAY 1995

[39] P. Maragos, “Morphological correlation and mean absolute error crite- Pai-Chi Li received the B.S. degree in electri-
ria,” in Proc. of ICASSP ’89, pp. 1568–1571, 1989. cal engineering from National Taiwan University,
[40] L. N. Bohs and G. E. Trahey, “A novel method for angle independent Taipei, Taiwan, R.O.C., in 1987 and the M.S. and
ultrasonic imaging of blood flow and tissue motion,” IEEE Trans. Ph.D. degrees from the University of Michigan,
Biomed. Eng., vol. 38, pp. 280–286, Mar. 1991. Ann Arbor, in 1990 and 1994, respectively, both
[41] J. Meunier, “Ultrasonic texture motion analysis: Theory and simulation,” in electrical engineering systems.
in Proc. of SPIE, Medical Imaging, vol. 1896, pp. 381–396, 1993. From 1987 to 1989 he served as a Signal Officer
[42] L. K. Ryan, G. R. Lockwood, T. S. Bloomfield, and F. S. Foster, with the R.O.C. Army. He was a Research Assistant
“Speckle tracking in high frequency ultrasound images with application with the Department of Electrical Engineering and
to intervascular imaging,” in Proc. IEEE Ultrason. Symp., 1993, vol. 2, Computer Science from 1990 to 1994. He joined
pp. 889–892. Acuson Corporation, Moutain View, CA, as a Mem-
ber of the Technical Staff in June 1994. His current research interests include
ultrasonic imaging and signal processing.
Dr. Li was the recipient of the Distinguished Achievement Award in
Mustafa Karaman (M’89) received the B.S. degree Electrical Engineering Systems for his outstanding academic achievement at
from Middle East Technical University, Ankara, the University of Michigan.
Turkey, in 1986, the M.S. and Ph.D. degrees from
Bilkent University, Ankara, in 1988 and 1992, re-
spectively, all in electrical and electronics engineer-
ing. Matthew O’Donnell (M’79–SM’84–F’93), for photograph and biography, see
He is currently a Post-Doctoral Fellow in the p. 82 of the January issue of this TRANSACTIONS.
Bioengineering Department of the University of
Michigan, Ann Arbor. His research interests are
in VLSI signal processing and medical ultrasound
imaging.