Complex pulsing schemes for high frame rate imaging

Thanassis Misaridis', Mathias Fink' and Jergen A. Jensen'

'Laboratoire Ondes et Acoustique, E.S.P.C.I.,
10 rue Vauquelin, 75005 Paris, France
'Center for Fast Ultrasound Imaging, BrstedoDTU, Bldg. 348,
Technical University of Denmark, DK-2800 Lyngby, Denmark

Abstract ing the scan lines of an image in linear and phased array
High frame rate ultrasound imaging can be imaging. Such frame rate is insufficient for 3-D ultrasound
achieved by simultaneous transmission of multiple fo- imaging in real-time, a technology that can have high clini-
cused beams along different directions. However, cal value in areas such as cardiology, prenatal diagnosis [I],
image quality degrades by the interference among ultrasound-guided brachytherapy and surgery.
beams. An alternative approach is to transmit spheri- There are two main approaches for increasing the frame
cal waves of a basic short pulse with frequency coding rate. The first is parallel transmission of multiple focused
and a constant transmit delay from channel to channel. beams of coded waveforms along different directions. The
In this way, transmit diversity is provided on a time
second is sparse synthetic transmit aperture @TA) imaging,
and channel basis rather than on a beam direction ba-
sis. The non-focused transmitted acoustic waves cany where unfocused (spherical) waves are transmitted from se-
spatial informationfrom the entire imaging region. At lected elements, and receive beamforming for all directions
a given imaging point, all pulses will add up to a pulse yields images, that are summed after a small number of emis-
train. 'lie acoustically generated high time-bandwidth sions, when the whole transmit aperture is synthesized.
(TB)product waveforms can be compressed by using Simultaneous transmission of multiple focused beams
a filter bank of matched filters one for every beam di- along different directions is limited by the interference among
rection. Matched filteringcompresses the pulse vain beams. Transmitting coded waveforms can reduce the cou-
to a single pulse at the scatterer position plus a number pling by cross-correlating the received signals with the trans-
of spikeaxial sidelobes. Frequency and phase modula- mitted coded waveforms for a given beam direction. How-
tion of the transmitting pulses allows control and elim- ever, the number of coded signals with low cross-correlation
ination of the ambiguous spikes. QLFM pulse trains
properties is limited by the relatively narrow ultrasound band-
are found to give the best performance.
Simulation results and images are presented shaw- width. Instead of coding on a beam direction basis, a method
ing the feasibility of the method. The excitation can- is proposed in this paper, which combines the two methods,
sists of 32 pulses with linear frequency modulation providing transmit diversity on a time and channel hasis and
along the transducer elements, that cover the 70% heamforming a whole image for every emission.
fractional bandwidth of the 7 MHz transducer. The
resulted images (after bedonning and matched RI-
tering) show an axial resolution at the same order as 2 A novel pulsing concept
in conventional pulse excitation and axial sidelobes
down to -45 dB. With the proposed imaging strategy The new method suggests the following spatio-temporal pulse
of pulse train excitation, a whole image can be formed coding scheme: each transducer element emits a diverging
with only two emissions. making it possible to obtain
spherical wave of a short pulse centered at a given frequency.
high quality images at a frame rate of 20 to 25 times
higher than that of conventional phased may imaging. All pulses have the same duration T and, thus, roughly the

1 Introduction
+,
same energy, but they have different canier frequencies. The
frequency of each pulse can he chosen so that f. = where
0, = l , 2 , ...N represent the number of cycles in each pulse.
The transmitted Dulses are centered in N adiacent freauencies
Current ultrasound scanners have a limited frame rate of spaced 1/T apart, where no frequency is transmitted twice.
about 30 to 40 frameds, due to the sequential nature of form Fig. 1 shows an example of a transmitting sequence matrix

0-7803-7582-3/02/$17.00 02002 IEEE 2002 IEEE ULTRASONICS SYMPOSIUM - 1609

 with N=32 pulses, which excite every second element of a shows the echoes received at the individual elements from a
@-element transducer with a center frequency of 7 MHz and point scatterer located at depth 6.8 cm 26" off axis. The re-
a 70% bandwidth. sults are obtained with the simulation program Field II [3]. It
can be seen that the received train has been weighted due to
the convolution of each transmitted pulse with the two-way
impulse response of the transducer.

Figure 1: The transmit time-space matrix.

The excitation pulses are not applied simultaneously hut
with a constant delay T, from element to element. Thus, the
duty cycle of the transmitted train is DC = 100, .:
The du-
Figure 3: The received echoes in the individual channels from
ration of each transmit event in the new method is roughly
N I D C higher than in conventional imaging. For 32 trans- a single point scatterer located at a depth of 6.8 cm 26" off
axis.
mitting pulses with a duty cycle DC=90%, the emission time
shown in Fig. 1 is 80 p, which is 35 times longer than in
Using this transmission scheme, all elements are excited with
conventional pulse excitation. However, the time of flight for
imaging up to a depth of 20 cm is 260 p, and the total time short pulses and the high time-bandwidth product waveforms
are generated acoustically.
required to acquire a frame is only 30% lugher. The new
method requires that all elements are in receive mode dur-
ing the entire emission period apart from the short time in
which they transmit. Neither focusing (which is an add-only 3 Frequency-coded pulse train design
operation) nor compression are affected by the short "dead"
transmitting time.
for matched filter processing
Frsquency-cded By using inter-pulse delays and "per element'' coding in the
transmined pulses new method, it has been possible to obtain rf data, which are
staggered frequency-modulatedl trains. Processing of modu-
lated signals is generally done using a matched filter, which
compresses the signal in order to retain good axial resolu-
tion, while at the same time maximizes the signal-to-noise
ratio. The matched filter output is the auto-correlation func-
tion of the modulated signal. Tissue attenuation in ultrasound
causes a significant frequency downshift of the received sig-
nal. When the received signal that enters the matched filter
is shifted in frequency by fd, the matched filter output is het-
ter described [4] by the ambiguityfunction (AF). known from
radars [SI,a complete 2-D matched-filter response in scatterer
Figure 2 The proposed transmitting scheme results in travel range T and frequency shift fd. The processed rf signal from
ling staggered pulse trains. a single scatterer is not the auto-correlation function of the
At a given imaging point, all pulses will add up to a pulse transmitted signal, hut a cross-section of the ambiguity func-
train (Fig. 2). The propagating pulses will arrive at an imag- tion for a certain frequency downshift multiplied by a phase
ing point P a t different times, according to the travelling dis- factor [4].
tances r, of each pulse. The delays between the pulses of the The mathematical analysis of the ambiguity function of
received echoes will have slightly changed, an effect usually FSK signals can he found in [5]. It is a superposition of
referred to as staggering. Thus, each transducer element re- the cross-ambiguity functions between the individual pulses,
ceives a staggered pulse train. A duty cycle of 94% is used, translated both in delay and frequency shift away from the
which assures no pulse overlapping of the received wave- origin, centered at T = (n - m)T,, fd = fm - f. and weighted
fronts echoed from a given point for all depths [21. Fig. 3 by a phase factor. fn and fm denote the carrier frequencies
'Frequency coding from pulse 10 pulse is also refemd to as FSK modulation (frequency shift keying)

1610 -2002 IEEE ULTRASONICS SYMPOSIUM

of the n -th and m - fh pulse respectively. When one moves pulses are transmitted from every element from one emission
along the delay axis, the number of cross-ambiguity functions to the next, and therefore the effect of attenuation is the same
that superimpose decreases linearly with T,, i.e. at 7 = pT, in both images.
there are N - p combined functions. The positioning of these
cross-terms along the frequency-shiftaxis depends on the fre- 1I O /
,
~~
.,
quency firing order. The desired mainlobe is the superposi-
tion of all the ambiguity functions of the individual pulses,
Ii3,
;
I :

which allows to achieve the same axial resolution as in con-

1:: :
ventional pulse excitation. + U ,

- Frequency Shift Keying) modulated signal. In this case. all F , 5: ne

~ received
~ ~echoes afterprocessing using sue.
the cross-terms on a cross-section parallel to the frequency cessive A second train with additional PSK mod.
shift axis at f = PT, will now superimpose at ( P T , , P ~ , )where
, ulation will give matched filtered ,+data with the same main
f?is the frompulse Pulse as as the response but ambiguity spikes with opposite phase. Coherent
t a l frWencY ofthe first Pulse in the vain. The QLW-FSK sum of the rf-data from the two emissions will cancel all the
signal translates all the sidelobes away from the delay axis odd.numbered ambiguity
by p f,. The elimination of the sidelobe pedestal is an im-
portant desired property of the ambiguity function of signals Fig. 5 show the Smmation of the compressed +lines
to be used for imaging in a dense medium[4]. In ultrasound. from two ComPIementV Pulse train emissions. Since phase
the available bandwidth does not allow a frequency step large coding eliminates only the odd-numbered spikes, the trains

l:m
to all surfaces offthe 7-axis and he first sur. have to be designed in such a way that no even-numbered
faces will introduce a small number (3 to 4 depending on the spike enters the delay axis for a desired dynamic range. This
frequency stepping) ambiguous spikes in the measurement. is achievedby weighting the transmitted pulses[21. m e cross-
Fig. 4 shows the matched filter responses of the received stag- terms for rectangular Pulses =e S i n C functions weighted by a
gered QLFM-FSK pulse trains of Fig. 3. frequency-dependent phase factor, which are superimposed.
Weighting of the pulses can position the nulls of the sinc
functions at a given p position. The design parameters are
the pulse weighting functions, the frequency spacing and the
119 : duty cycle of the train. The processed rf-data shown in Fig. 5,
4: : when plotted in logarithmic scale show elimination of all
146
155 spike ambiguities below -50 dB.
€4
70 75 Bo 85 r,m$i 95 1W 105

4 Receiver processing
Figure 4 The received echoes of Fig. 3 after matched filter
processing. The inter-pulse delays are spatial-variant, consisting of the
sum of the propagation delays with the applied transmit inter-
Cancellation of the ambiguity spikes is possible for two re- element delays. However, all elements will receive a pulse
sponses which have the same mainlobe. and sidelobes equal train with the same set of delays from pulse to pulse. This
in amplitude but with opposite phase. This requires that the can be found from simple delay calculations. Therefore, the
second transmitting train is a QLFM-FSK signal with the same compression filter can be used for all channels. How-
same frequency coding as the first one, with additional phase ever a different compression filter has to be applied for evely
coding from pulse to pulse (such trains are referred to as line in the image, i.e. the trains are decoded on a line in-
FSUPSK signals). The build up of the first spike, for in- stead of a channel basis. Ideally, a different filter should be
stance, is due to the superposition of the cross-terms of all constructed for every depth. However, compression of the
pulses with n - m = l . i.e. for adjacent pulses. Setting every QLFM-FSK is rather insensitive to staggering, and one fil-
second pulse in the train to he 90" out of phase will add a ter for every beam direction centered at half the depth of the
minus sign to all cross-terms. With this simple phase coding image is sufficient. The presence of the ridge in the ambi-
all pulses with n - m=3 will also be out of phase, which will guity function of the QLFM-FSK signals makes compression
yield cross-tem with minus sign that contribute to the third insensitive to frequency shifts of the received signal due to at-
ambiguity spike. Thus, all odd-numbered spikes will have the tenuation. This is a general property of linear FM signals that
same response, but inverted sign compared to the response of yields robust performance to attenuation effects[l], justifying
the first train. This approach has the advantage that the same the use of one matched filter for all depths.

2002 IEEE ULTRASONICS SYMPOSIUM - 1611

 For constructing each filter, the delays are calculated ge- higher than those in current ultrasonic imaging with only a
ometrically. they are then translated into number of samples small loss in image quality.
which are added to the constant transmit inter-uulse delays.

Figure I : Simulated images of a phantom consisting of sev-

eral p i n t scatterers (a) for conventional phased array imag-
ing with dynamic receive focusing using 51 emissions, and
@) for QLFM-FSWPSK pulse train imaging after two emis-
Figure 6:The receiver architecture. sions. The dynamic range of the images is 55 dB.
The received echoes are first beamformed for all beam
directions. Subsequently, the beamformed data are cross-
correlated with a bank of matched filters, one for each beam Acknowledgments
direction. The number of filters (and therefore the number of
the required correlators) is equal to the number of the scan This work was supported by Marie Curie grant HPMF-CT-
lines. Thus, the received data from every emission are fed in 2001-01 165 from the European Community.
parallel to L beamformers (where L is the number of lines)
followed by L correlators, that store the L reference wave-
forms. In an actual implementation, the d-data (after beam-
References
forming and compression) from emission n - 1 are stored in
[I] A. Fenster and D.B. Downey. 3-D ultrasound imagmg:
memory, and are added to the rfdata of the current emission
a review. IEEE Eng. Med. Bioi. Mag., 15:4149, 1996.
n. The new ghost-free image is displayed, and the rf-data
of emission n are pushed to the RAM. Using this recursive [2] T. Misaridis and I. A. lensen. A new coding concept for
scheme as first suggested by Nikolov et.al.[b], a new image fast ultrasound imaging using pulse trains. In Medico1
can be formed after every emission. Imaging 2002: Ultrasonic Imaging and Signal Process-
ing, volume 4681, pages 68-78. Proceedings of SPIE,
2002.
5 Pulse train imaging
131 J. A. Jensen. Field A program for simulating ultrasound
Simulated images using the proposed method are shown in systems. Med. Bioi. Eng. Comp., 10th Nordic-Baltic
Fig. 7. Axial resolution is preserved, axial sidelobes are down Conference on Biomedical Imaging, Vol. 4, Supplement
to -45 to -50 dB, lateral resolution is, however, degraded. This I, Part 1:351-353, 1996b.
is because most transmitted energy is at the center of the ar-
ray. effectively reducing the transmit aperture. This can be [41 T. X. Misaridis, M. H. Pedersen. and I. A. lensen. Clini-
improved in a variation of the method, where two trains cov- cal use and evaluation of coded excitation in B-mode im-
ering half the aperture are combined in four emissions. The ages. In Pmc. IEEE Ulrrason. Symp.. volume 2. pages
lugh time-bandwidth products of the long pulse trains also 1689-1693,2000.
preserve the SNR. When white Gaussian noise was added
to all the received channels before heamforming, to simulate [5] I. P. Costas. A study of a class of detection waveforms
electronic receiver noise, it was found that the SNR loss for having nearly ideal range-Doppler ambiguity properties.
train imaging is only at the order of ahout 8 dB compared Proc. IEEE, 72(8):996-1009, 1984.
to phased-array imaging. For comparison, synthetic transmit
[6] S. 1. Nikolov, K. Gammelmark, and I. A. Jensen. Recur-
aperture imaging with two emissions yields images of 25 dB
sive ultrasound imaging. In Proc. IEEE Ulrrason. Symp.,
less SNR. In conclusion, in applications where speed is a pri-
volume 2. pages 1621-1625, 1999.
ority, the new approach can achieve frame rates 25 to 50 times

1612 -2002 IEEE ULTRASONICS SYMPOSlUM