WO2000010638A2

WO2000010638A2 - An asynchronous oversampling beamformer

Info

Publication number: WO2000010638A2
Application number: PCT/TR1998/000019
Authority: WO
Inventors: Karaman Mustafa; Kozak MÜCAHİT
Original assignee: Baskent University
Priority date: 1998-08-24
Filing date: 1998-08-24
Publication date: 2000-03-02
Also published as: WO2000010638A3

Abstract

The present invention provides a digital delay-and-sum receive beamformer with delta-sigma analog-to-digital (A/D) converters. The echoes' reflected are sensed by a transducer array. The signals received are sampled at the time instants required for the receive beamforming using the timing information stored in a memory and then digitized by ΔΣ modulators prior to summation. This requires a non-uniform sampling scheme employing different clocks at each array channel. The non-uniform sampling is achieved by performing a logical 'and' operation between a fixed master clock, which is slightly higher than the frequency at which the samples of beam sums are acquired, and a digital receive beamforming memory, which holds a '1' or a '0' for that analog sample associated with the master clock rate. The oversampled outputs, are then summed together. The coarse beam sum is further processed through a decimation filter to suppress high-frequency quantization noise.

Description

DESCRIPTION

AN ASYNCHRONOUS OVERSAMPLING BEAMFORMER

Field of the Invention: This invention relates to delay-and-sum receive beamformer of a coherent imaging system using vibratory energy, such as radar, sonar, and in particular ultrasound.

BACKGROUND OF THE INVENTION

Coherent imaging systems usually include an array of sensors being capable of receiving some form of energy radiated from the region under examination. The received echo signals by the array elements are delayed by an appropriate amount and added together, by some means, to form a receive beam. This form of beamforming is well known in the art and referred to as the delay-and-sum beamformer (sometimes called time-domain beamformer). Although being one of the earliest beamforming algorithms, delay-and-sum beamforming remains today as a powerful approach in array signal processing, because of its easy implementation. The underlying idea is that the differences in propagation delays from a single focal point to the array elements are compensated by some delay means, so that a coherent summation across the array produces the beam sum. The coherent summation, which is a discrete realization of diffraction integral across the array, reinforces the waves propagating in a particular direction and emanating from the focal point to be added constructively, while those propagating in other directions or emanating from different points are added destructively. Thus, the formed beam is only indicative of the reflected energy from a single focal point supplying information about the reflection coefficient associated with this focal point. By imparting desired time delays, the signal capturing capability of an array can be steered and focused at a particular direction and range, respectively. Hence, a region of interest can be scanned over several predetermined directions by adjusting the time delays across the array.

There are many applications employing beamforming algorithms such as radar, sonar, ultrasound imaging, sector broadcasting in satellite communications, oil exploration, high resolution space imaging, and earth crust mapping. The present invention concerns the digital delay-and- sum receive beamformer for a medical ultrasound imaging system, however, advantages is taken by the inventors of the fact that the principles of the invention can be employed in other coherent array imaging systems as well.

Medical ultrasound is a valuable non-invasive anatomical diagnostic tool, which enables real-time imaging of tissues within the body with a low cost hardware relative to other imaging methods. Ultrasound images can be reconstructed in various forms which are represented in different modes of operations. The most widely used mode of operation is called B-scan mode or pulse-echo mode. In the so-called B-scan mode, usually an array of piezoelectric transducers transmits ultrasonic waves into the body. In turn, reflected echoes are received and converted to electrical signals by the same array. The echo signals are processed by an analog or digital receive beamformer to obtain information about acoustic reflection coefficients (i.e., acoustic impedance discontinuities) of points in the region being imaged. The brightness of each pixel in the image is a function of the amplitude of the acoustic reflection coefficients.

Ultrasound pulses are transmitted and received by piezoelectric transducers. A piezoelectric transducer has the capability of producing ultrasound wave when driven by an electrical signal. Conversely, when an ultrasound wave strikes, it produces an electrical signal associated with the wave. In the transmit mode, transducer array is driven by a set of short burst of sinusoidal signals. A transducer can be viewed as a bandpass filter with a center frequency of /₀. Therefore, a set of ultrasound waves each being in the form of damped sinusoid associated with a particular frequency of /_υ are emitted into the subject being imaged. It is a well known fact that as ultrasound waves travel through the body, they are attenuated roughly linearly proportional to the frequency. Since the acoustic attenuation also increases with the propagation distance, there is a direct trade-off between operation frequency and depth of view. Typical attenuation coefficients for soft tissue and human liver are 1 dB/cm/MHz and 0.5 dB/cm/MHz, respectively, corresponding to 140 dB and 70 dB attenuation for 3.5 MHz ultrasound wave propagating round-trip trough 40 cm. In medical imaging, ultrasound center frequency lies in a frequency range of 2 to 15 MHz. Above this range, sound waves are significantly attenuated by human tissue, thus depth of view is very limited. Below this frequency range, wavelength of sound is so long that small structures cannot be resolved. The design of such a ultrasonic transducer array is beyond the scope of the invention and a number of excellent publications on this subject can be found in the literature. For example, the article by J. W. Hunt et. al., published in IEEE

Trans. Biomed. Eng., vol. BME-30, pp. 453-481, 1983, and entitled "Ultrasonic transducers for pulse-echo medical imaging" , is incorporated herein by reference.

The most powerful way of producing B-scan ultrasound images is the phased array technique, where all elements of transducer array are active in transmit and receive beamforming. Because of the resulting high electronic signal to noise ratio (SNR) as well as good spatial and contrast resolutions, phased array imaging is very favorable in medical ultrasound.

In clinical B-scan phased array systems, typically, a sector of 90° is scanned by means of transmit and receive beamforming operations to obtain a two dimensional cross sectional image of tissues. In sector scanning technique, generally, beamforming is handled electronically using variable delay lines. More specifically, ultrasound waves whether transmitted or received can be steered and focused at a direction and range, respectively, by adjusting array element delays to compensate different round-trip times. In transmit beamforming, appropriately delayed pulses are sent to the subject being imaged, whereas in receive beamforming, the echo signals are properly delayed and added together to compose a beam sum. A sector is scanned by employing transmit and receive beamforming operations for each of the scan angle comprising the desired sector.

In B-scan ultrasound sector imaging only the ultrasound pulses that are steered at a particular direction should be propagating in the field of interest at any time. Hence, once a set of pulses are transmitted, the next set of pulses should be send after a time equal to the previous pulses return back to the transducer array. Thus, in a real-time imaging, the frame rate, number of beam lines per frame and imaging depth cannot be chosen arbitrarily due to the limited sound velocity in tissue. The real-time imaging constraint can be expressed as:

F - B - 2R < c (1)

where F, B, R and c are the frame rate per second, number of beam lines per frame, imaging depth and velocity of sound, respectively. Assuming c = 1540 m/sec, R — 200 mm and F — 20 frames/sec, the number of beam lines per frame B must be equal to or less than 192. On the other hand, the spatial sampling interval (Δ sin θ) in beam space, which is defined as the difference in sinuous functions of two neighboring steering angles, is determined according to the spatial Nyquist criteria. For focused array systems, there is a Fourier transform relationship between aperture function and beam space (or angular) response of the array at the focal zone (or far-field). The aperture function of an array (spatial distribution of array elements) with N elements each separated with an equal distance of d, can be represented as:

1, (-Z + Q.5)d < x < {% - 0.b)d 0, otherwise.

The two way (transmit-receive) angular response of an array system is given by the following Fourier transform relation:

U(θ) = T{a_t(x) *a_r(x)} (3) where α_t(.) and α_r(.) are the transmit and receive aperture functions, respectively. Here, θ is the steering angle measured with respect to array normal and "_^r" and"*" denote the Fourier transform and convolution operations. Since the transmit and receive apertures are the same for a conventional phased array system, the effective aperture extend (twice the spatial frequency bandwidth), resulting from the convolution of transmit and receive arrays with N elements, is (2N - l)d/λ (the spatial frequency is usually considered in normalized units corresponding to the wavelength λ). The minimum spatial sampling frequency (_{Δg ng}) should be twice the spatial frequency bandwidth, (2N — l)d/λ. Consequently, the spatial Nyquist sampling criteria is given b

^{Λ8in 9 ≤} (_^rτμ W For such a calculation further reference is made to the book by J. W. Goodman, "Introduction to Fourier Optics", New York, McGraw-Hill, 1992, is incorporated herein by reference. For a 90° sector (i.e.,

< sin ø <

the relationship between number of beam lines and spatial sampling interval can be expressed as:

Combining equations( 4) and (5) gives a lower bound on the number of beam lines as: _{β ≥} V^(2iV -_l) ₍₆₎

A

For example, if d is chosen as equal to λ/2, then the number of beam lines per frame must be greater than or equal to /2 times the number of array elements, provided that N is large compared to 1 (i.e., _V ^> 1). For a conventional 128-element transducer array, the number of beam lines must be greater than or equal to 182. Further combination of equations (1) and (6) results in the following lower and upper bounds on the number of beam lines:

^ __ _ _ _B _ ^ CO

As a result, in a real-time 128-element B-scan phased array imaging system wherein an array element spacing of ^, frame rate of 20 Hz and imaging depth of 200 mm are chosen, a total of 182 beam lines may be acquired over a 90° sector, with each beam line being steered in increments of Δ sin θ = 0.0078.

In medical ultrasound one of the important performance criteria is the resolution of the image which refers to the ability of distinguishing closely spaced structures with clarity. The size of resolution cell is determined by lateral and axial resolutions. The lateral resolution is closely related to the beam width and sidelobe levels (i.e., array size, array element spacing and apodization), while the axial or range resolution is determined by the the ultrasound waveform (i.e., frequency and bandwidth). Typically, lateral resolution is poorer than range resolution by a factor of 4 to 10. In a real-time B-mode ultrasound imaging system, dynamic focusing is a simple approach to increase the lateral resolution. However, for real-time imaging, due to the limited sound velocity, clinical scanners employ a fixed focused beam at each scan angle during transmission, whereas the receive beamformer dynamically focuses the array to the successive focal points (image points) along each scan angle. Stated another way, reconstruction of a beam sum is accomplished by summing the received signals with proper dynamic delays corresponding to the propagation time delays from all sample focal points lying on the axis of transmitted beam. Dynamic focusing can be carried out using variable delay lines, either analog or digital, and therefore requires a large amount of dedicated hardware. Since dynamic focusing can only be employed in receive, receive beamformer is more critical than transmit beamformer and many researches in the art have been concentrated on the receive beamforming schemes. In addition to dynamic focusing, the image quality is also affected by f/number apodization or aperture apodization employed in transmit and/or receive beamforming. Such processing is used to reduce sidelobes and improve the depth of field especially in array near-field. Early ultrasound imaging systems employed analog receive beamforming where dynamic focusing is achieved using tapped L-C circuits as variable delay lines. These analog processing components are bulky and relatively expensive. Besides the system complexity, analog beam- formers also suffered from low SNR at the output and some inherent artifacts of L-C delay devices such as impedance mismatching, insertion loss and switching transients, all of which adversely degrade the image quality. Moreover, analog delay components require deep attention in manufacturing and up-keeping, because they are unstable and much effected by the environmental conditions, such as temperature drift and age. Furthermore, the use of analog delay lines limits the operation of an ultrasound imaging system in many ways and therefore is not desirable in modern clinical systems where flexibility and versatility are much important. With the advent of digital technology, digital beamforming became a promising alternative as compared to conventional analog beamforming. In addition to precision, stability and reliability, the decrease in cost of digital components encouraged state-of-the-art systems to employ digital beamformers, wherein a high-speed multi-bit analog-to-digital (A/D) converter is used on each channel to digitize the incoming echo signal. The echo signals at each array element are typically sampled synchronously at the same uniform clock rate, quantized to a digital code word and stored in a digital memory. To implement dynamically focused receive beamforming, for each of the focal points, one of the quantized samples at each channel is chosen, taking into account the steering and focusing delays, and is summed with those at the other channels. Since the received signals are delayed using discrete time increments associated with the sampling rate, the resulting time delay quantization can reduce the SNR at the beamformer output and cause some grating lobes. Therefore, the image quality is critically affected by the time delay quantization errors. For effects of the time delay quantization errors, reference is made to the article by D. K. Peterson et. al., published in IEEE Trans, on Acoust. Speech Signal Processing, vol. ASSP- 32, pp. 548-558, 1984, and entitled "Real-time digital image reconstruction: a description of imaging hardware and an analysis of quantization errors" , herein incorporated by reference. Another work by M. O'Donnell et. al., published in Proc. of IEEE Ultrasonics Sym., IEEE cat. no. 90CH2938-9, pp. 1495-1498, 1990, and entitled "Real-time phased array imaging using digital beamforming and autonomous channel control" , has reported that in a medical imaging system, generally, the time delay resolution must be on the order of 32 times the frequency at which ultrasound waves are radiated. Therefore, with regards to the implementation of a digital beamformer, one difficulty is that the sampling rate must be at least 32 times the transducer center frequency /₀, to provide a high quality beam sum. For example, if a 3.5 MHz transducer is used, then a sampling rate as high as 112 MHz is needed to provide the desired delay resolution for a good coherent summation across the array, hence for a high quality beam sum. These multi-bit (usually 8-10 bits) high speed A/D converters, associated for each of the array element, comprise a large amount of hardware consuming extremely much power, thereby increase the cost of digital beamformer as well.

The study, presented by Pridham et. al. in an article published in Proceed, of IEEE, vol. 67, No. 6, pp. 904-919, June 1979, and entitled "Digital interpolation beamforming for low-pass and band-pass signals" , describes a digital beamforming scheme which employs digital interpolation technique to ease the high sampling requirements for A/D converters. In this approach, analog echo signals are need only be sampled at a rate satisfying or slightly exceeding the Nyquist rate. To achieve required delay resolution, all digitized echo signals are processed through a digital interpolation filter prior to beamforming. Basically, the interpolation procedure involves zero padding (inserting zero) between consecutive original samples and then filtering with a low- pass filter. The price of reduced sampling rate comes here in part, where the digital processing requirements are increased. Therefore, the digital interpolation beamformer can be considered as a compromise where the hardware complexity is partitioned between A/D converters and digital processing. Many modifications concerning this method are emerged in the literature for further reducing digital beamforming hardware. For example, U.S. Pat. No. 5,345,426 issued in Sep. 6, 1994 to Lipschutz, and entitled "Delay interpolator for digital phased array ultrasound beamformers" , discloses a digital beamformer with delay interpolator. In this patent, the sampled echo signals are delayed in increments less than the sampling period, where the sampling rate is chosen to satisfy the Nyquist sampling criteria. The desired delays consist of a coarse delay and a fine delay. The coarse delay means delays the incoming echo signals at an integer multiples of the sampling period, whereas a fine delay means (delay interpolator) supplies fractional delays at a fractional multiples of sampling period through digital interpolation using programmable finite impulse response (FIR) filters. Different delay resolutions are obtained by applying different filter coefficients to FIR digital filters. The coefficients of digital filters are fit in a form of ^n , where m is an integer, so that multiplications in the FIR filters are implemented by simply shifting bits of data samples by m bits positions. This results significant reduction in interpolation circuitry.

The method, proposed in the article by J. E. Powers et. al. published in IEEE Trans. Sonics and Ultrasonics, vol. SU-29, no.6, pp. 287-295, 1980, and entitled "Ultrasound phased array delay lines based on quadrature sampling techniques" , employs a base-band beamforming scheme, where each echo signal is digitized to its in-phase and quadrature components. Using digital delay lines and phase rotators, the base-band components can be accurately delayed, rotated and summed at a very low sampling rate, which is sufficient to meet the Nyquist criteria for base-band signals. In a conventional quadrature demodulation, band-pass signal is mixed with quadrature carriers and then low-pass filtered yielding the in-phase and quadrature components. Powers et. al. has suggested an alternative quadrature sampling technique to eliminate analog mixers and low-pass filters. In this approach, two samples, that indicate in-phase and quadrature components, are taken with a time difference of one quarter of the carrier period. Hence, this quadrature sampling requires two sampling clocks which have 90° phase difference with respect to center frequency. Although the sampling rate of the A/D converters is substantially relaxed and analog mixers and low-pass filters are avoided, this method still suffers from hardware complexity. Firstly, two phase rotators are needed for each channel. Secondly, two A/D converters, two digital delay elements and two summer blocks are required on each channel to implement the beamforming operation. J. H. Kim et. al. have described a digital beamformer in an article, published in Ultrasonic

Imaging, vol. 9, pp. 75-91, 1987, and entitled "Pipelined sampled-delay focusing in ultrasound imaging systems" , in which dynamic focusing is achieved by non-uniform sampling clocks. In this method, the beamforming delays are employed not to the echo signals but to the sampling clocks. The non-uniform sampling clocks are generated by simply reading the delay patterns stored in a digital memory. The echo samples, sampled at different instances on each channel, are delivered into a first-in first-out (FIFO) buffer. The beam sum is obtained by shifting the echo samples to the output side of each FIFO buffer at the same time and then summing them. Kim et. al. have further considered to use the quadrature sampling technique as described by Powers et. al. Using both the non-uniform and quadrature sampling, the echo signals are first delayed and then digitized into its in-phase and quadrature components. Although appealing in its simplicity, this method has some deficiencies such as requiring a large number of A/D converters which is twice the number of array elements. A solution to the large number of A/D converters problem has been investigated in an article by M. Karaman et. al. published in E. Aπkan, editor, Communication, Control, and Signal Processing, pp. 1612-1618, 1990, and entitled "A front-end digital hardware architecture for real-time ultrasound imaging". Karaman et. al. have proposed a digital beamforming architecture wherein the number of A/D converters and the size of overall hardware are decreased at the expense of higher sampling frequency and increased interconnection circuitry. In this approach, it is suggested to use reasonably fast A/D converters together with an analog/digital switching unit which is capable of connecting the input of one of the A/D converter to one of the array channel. One other difficulty in the systems employing non-uniform sampling is the real-time processing of the large number of asynchronous samples to obtain signal values of the focal points. This processing consists of synchronization of samples and addition of them. Karaman et. al. have further proposed an inverse binary tree adder network with self synchronizing capability to process the large volume of irregular data, which is more feasible for implementation.

Present systems, such as those cited above, are still at a point far beyond to meet the requirements for a beamformer-on-chip, which can be used in clinical systems. The use of oversampling converters seems to be an elegant way of avoiding many difficulties encountered with conventional multi-bit A/D converters, hence to reduce the area, cost and power of a digital beamformer. Although oversampling converters are known from the half of this century, it is the last decade that oversampling methods have become popular for high-resolution conversion. Oversampling converters, more commonly termed as delta-sigma (ΔΣ) modulators, sample the analog signal at a much higher rate than the Nyquist rate to spread quantization noise over a band much wider than signal pass-band. In addition to a crude quantizer, ΔΣ modulators also incorporate closed loop filters that integrate the error between input signal and quantizer output. The closed loop filters shape quantization noise such that only a small portion of noise remains in the signal pass-band while a considerable amount of noise power is pushed to out of band. A decimation filter, consisting of a low-pass filter and a down sampler, can act as the demodulator of a ΔΣ modulator. The low-pass filter suppresses out of band quantization noise, and the down sampler further reduces the sampling rate to a desired rate. For a detailed information, the article by P. M. Aziz et. al. published in IEEE Signal Processing Magazine, pp. 61-84, January 1996, and entitled "An overview of sigma-delta converters" , and the book by J. C. Candy et. al., "Oversampling delta-sigma data converters" , New York, IEEE Press, 1992, are incorporated herein by reference. A recent attempt, wherein the use of ΔΣ modulators are in consideration for a digital beamformer, is disclosed in U.S Pat. No. 5,203,335 issued in Apr. 20, 1993 to Noujaim et. al. and entitled "Phased array ultrasound beamforming using oversampled A/D converters". In this patent, a single-bit ΔΣ modulator is incorporated on each channel, instead of a multi-bit A/D converter, to digitize the reflected echo signals. The oversampled one bit representations of analog echoes are easily delayed using shift registers and added together. Hence, a crude representation of beam sum is obtained, which is then converted by a single decimation filter to a multi-bit digital signal representing the entire dynamic range of the beam sum. Another system which also exploits oversampling principles for a beamformer has been proposed in U.S Pat. No. 5,461,389 issued in Oct. 24, 1995 to by M. Dean and entitled "Digital beamforming array" . The only difference of this method from that of proposed by Noujaim et. al. is that echo signals are converted into intermediate frequency signals to increase the noise performance of oversampling converters. However, both of these systems have some drawbacks for clinical systems, especially for those using color-flow or Doppler processing, because the image quality is severely effected by the additional noise induced by the sample repetition during dynamically changing the receive delays. The SNR of an oversampling converter is related to the oversampling ratio, which is defined as the sampling frequency divided by two times input signal bandwidth. Thus, for a beamformer based on oversampling A/D conversion, the distance between consecutive focal points should be chosen sufficiently close to achieve the desired over- sampling ratio. However, on the other hand, dynamically focused beamforming delay patterns of such closely placed focal points result sample repetitions due to time delay quantization and/or acoustical geometry. More specifically, almost in every channel (except the center element) some of the neighboring consecutive focal points may require use of the same sample for dynamically focused receive beamforming. By the way, some of the samples should be repeated whereas some of those should be withdrawn. This disrupts the synchronization between the ΔΣ modulator on every channel and the demodulator at the beamformer output. The repeated or withdrawn samples can be represented by some additional scaled impulse functions in the time domain. These unwanted impulses are also convolved with the low-pass filter in demodulator causing to pass some unexpected noise power into to signal pass-band. Under the normal operation much of the noise power at the output of a ΔΣ modulator lies well outside the signal pass-band. When a sample is repeated, however, the low-pass filter corrupts the output signal by folding back some unexpected noise energy at higher frequencies into lower frequencies where the original signal is much favor to be found. Although, any linear system can be inserted between ΔΣ modulator and demodulator, dynamically focused beamforming cannot be treated as a linear operation when sample repetitions and/or withdraws are involved. Three different approaches to handle with the sample repetition problem have been recently proposed by S. R. Freeman in an article published in Proc. of IEEE, Ultrasonics Sym., IEEE cat. no. 97CH3245-6, pp. 1706-1711, 1997, and entitled "An ultrasound beamformer using oversampling" . The first approach, termed as Insert-0 method, inserts a null sample into the bit stream where repetition occurs. In other words, if repetition of a particular sample will take place, then the sample remains unchanged at the first use, but at the second time a null sample instead of that sample is used. This null sample has a digital level between half way of ΔΣ output levels. Therefore, an extra bit is required to represent the null sample level, while all the other samples must be similarly recoded. In the second approach, termed as divide- by- 2 method, the repeated samples are divided by two and spreaded over two samples. Clearly, the first and second halves will be used at the first and second use, respectively. However, to represent the half levels, two extra bits are required. Both of these methods are effective in correcting the situation resulted from the sample repetition problem, because they can nearly preserve the frequency spectrum of original signal. On the other hand, they may negate the advantage gained by single-bit ΔΣ modulators, since extra bits are needed to recode the digital ΔΣ outputs. Freeman et. al. have further suggested an alternate solution which maintains the advantage of ΔΣ modulators by allowing the beamforming operation to be performed on one bit data. In this approach, ΔΣ modulators are forced to take into account the repeated samples in their operation by manipulating the analog feedback magnitude within the modulators. A multiplexer and a 2X analog amplifier are inserted in the feedback loop that chooses between the normal and a scaled-by-two magnitude. The samples that should be repeated in the delay structure should have a feedback magnitude of two, whereas the normal, non-repeated samples should have that of one. This modification vastly improves final image quality by keeping the synchronization between ΔΣ modulators and demodulator in the dynamically focused receive beamformer. Although being very significant, this method can only compensate the effects of the repetition of certain samples in the delay structure. In fact, the samples that will be later repeated in the delay structure should be fed twice to the modulator for an exact synchronization. However, this may defeat the purpose of simplicity, since analog buffers on each channel are required to accomplish the repeated samples being fed twice. The ΔΣ modulator including a multiplexer and 2X buffer on the feedback path can be regarded as an approximation that forces the nodes of modulator to be modified as the repeated samples are fed twice (in fact, each sample is modulated at once).

In summary, it should be evident that there is a need in the art for a dynamically focused beamformer based on oversampling A/D conversion which completely avoids the aforementioned sample repetitions, but still preserves the simplicity gained by single-bit ΔΣ modulators.

SUMMARY OF THE INVENTION

A general object of the present invention is to reduce hardware complexity involved in the digitizing and delaying means of a dynamically focused digital delay-and-sum receive beamformer in a clinical ultrasound imager without making any concession in image quality. It is another object of the present invention to provide a dynamically focused, digital delay- and-sum receive beamformer using a set of oversampling A/D converters in a medical ultrasound imaging system, that employs phased array technology, with an image quality as compared to that of conventional systems achieved by using multi-bit A/D converters.

Another object of the present invention is to provide a dynamically focused digital delay-and- sum receive beamformer using a set of ΔΣ modulators to digitize the analog echo signals at an oversampled rate, where the bit resolution (preferably single-bit) of internal quantizers within the modulators are far less than the number of bits required to represent the entire dynamic range of echo signals.

A further object of the present invention is to provide a dynamically focused digital delay-and- sum receive beamformer, wherein the beamforming operation is performed on digital signals of the number of bits determined by the internal quantizers within ΔΣ modulators, to produce a coarse representation of beam sum.

A still further object of the invention is to provide a dynamically focused digital delay-and-sum receive beamformer, wherein a single decimation filter is used in order to suppress high frequency quantization noise on the coarse representation of beam sum, so that hereby an accurate beam sum with the bit resolution required to represent the desired dynamic range at the output, is recovered.

Yet another object of the present invention is to provide a dynamically focused digital delay- and-sum receive beamformer wherein sample repetitions in delay structure are completely avoided, and thereby produce high quality beam sums in which the only existing noise is due to ΔΣ modulators and therefore can be controlled by theoretical calculations.

These and other objects are realizable with the medical ultrasound imaging system of the present invention, in one preferred embodiment, including a linear array of transducers with N elements for transmitting and receiving acoustical waves. The medical ultrasound imager of the present invention further includes a transmitter means for steering and focusing ultrasound waves at the desired direction and range, respectively. The transmit beamformer imparts appropriate delays to the respective array elements for supplying a fixed focused beam along each scan angle, where fixed focusing range is chosen, in general, as the mid-range of imaging depth.

In another preferred embodiment of the present invention, echoes reflected from the human or animal tissue are sensed by the same transducer array, and converted to electrical signals at the electrodes of each array element. In receive mode, array is steered at the same direction as it is in the transmission mode, but in contrast to fixed focusing dynamic focusing is employed for successive focal points along that scan angle. Therefore, the receive beamformer circuit applies a different delay on each channel, and varies that delay with depth (or time). Moreover, the receive beamformer also employs an f/number apodization by excluding the contribution of a number of the array end elements to beam sum at the focal points near to array. More specifically, when the array size is greater than range of receive focal point divided by /_num, where f_num is a predetermined value and usually taken as 2, then array length is shortened until the above constraint is satisfied. Thus, only some predetermined elements of array contribute to beam sum for some focal points at the near field. On the other hand, when array size is smaller than range of the focal point divided by f_num i then all elements are allowed for contribution in beam sum for that focal point. As stated before, acoustic waves are attenuated as they travel through tissue of interest. In still another preferred embodiment of the present invention, a Time-Gain-Control (TGC) amplifier array is included prior to receive beamforming. Each echo signal is applied to the respective element of Time-Gain-Control (TGC) amplifier array to compensate for the attenuation of ultrasound waves as they propagate, and hence to supply uniform contrast for image points with the same acoustical property. A plurality of pre-amplifiers, comprising Pre-Amp array, each associated for each of the transducer array element, follow the TGC array for further amplifying the incoming echo signals. Such a Pre-Amp array is usually used to control the maximum input dynamic range of A/D converters.

In yet a further preferred embodiment of the present invention, following the Pre-Amp array, a ΔΣ array, associated one for each of the array channel, is included to digitize the incoming echo signals. According to one principle of the present invention, only the echo samples, required for the receive beamforming are forwarded to ΔΣ modulators. This would appear to require a non-uniform sampling scheme employing different clocks for each of the array channel. The non-uniform clocks are generated by performing a logical "and" operation between a uniform master clock, which is slightly higher than the frequency at which the samples of beam sums are acquired, and content of a receive beamforming memory consisting of timing information. One of the another critical principle of the present invention is the choice of the master clock frequency. By using a minimum appropriate master clock rate, the aforementioned sample repetitions are avoided during dynamically focused receive beamforming, so that any unexpected noise is completely eliminated and a substantial good image quality is obtained. As it will be shown later, a master clock rate /_m , which is greater than or equal to _R J "^11Tn . X BF, provides the desired time resolution required to being non of the samples repeated on each channel for each of the focal point and for each of the scan angle. Here, fβF is the beamforming frequency indicating the frequency at which the samples of beam sums are acquired. It is a significant advantage of the fact that, for a given /BJΓ, the choice f_m is dependent on only the f/number apodization constant /„_am , which is readily in use in all commercial clinical scanners. In the case of exemplary f/number apodization of 2 (i.e., f_num = 2), f_m must be selected at least 1.0323 X /FB _ι which corresponds to a slight increase in operation frequency of ΔΣ modulators, and thereby does not involve significant hardware overhead. In general, the noise performance of oversampled converters can be defined by the order of the noise shaping function (i.e., order of the modulator) , oversampling factor and bit count of the conversion. These factors should be traded in such a way to minimize the beamforming hardware complexity. The beamforming hardware complexity decreases as the bit count of A/D conversion and/or order of the modulator is reduced while the oversampling factor should be raised to maintain the desired SNR performance. Usually, modulators of order higher than 2 are not desirable, since they are suffering from the loop stability problem. Preferably, the digitizing means incorporated in receive beamformer on each channel, is a single-bit 2^nd order ΔΣ modulator. Clearly, the oversampling factor is greatest for one bit conversion whereas the beam- forming hardware is simplest. However, in some applications where the resulted oversampling ratio cannot be easily accommodated, conversion to more than one bit may not be avoidable. In such cases, ΔΣ modulator per channel may include internal quantizer with a bit resolution more than one bit, but still far less than the number of bits required to represent the entire dynamic range of echo signals.

ΔΣ modulators have a relatively simple hardware and are more robust against circuits im- perfections, thus easily implementable even in low cost monolithic Very Large Scale Integration (VLSI) technologies. Another advantage of the present invention is that a considerable reduction in digitizing hardware of a dynamically focused digital receive beamformer is provided by taking the advantage of oversampled converters to be easily manufactured. A yet another advantage of the present invention is that dynamically focused beamforming operation is performed on the digital signals with the number of bits determined by the internal quantizers within ΔΣ modulators, so that the delaying and summing circuits are simplified compared to traditional systems.

The receive beamforming memory holds a "1" or "0" associated for that analog sample with the master clock rate, for each of the array channel, where "1" or "0" represents that the analog echo signal should be sampled at that time for the receive beam formation or not. By performing a logical "and" operation between the uniform master clock and the content of the beamforming memory, a non-uniform clock different for each of the array element, can be easily generated, so that only the samples required for receive beamforming are chosen as the input of the ΔΣ array. Each ΔΣ modulator, associated for each of the array element, is allowed to operate at the respective non-uniform clock rate. The oversampled outputs, each consists of the associated steered and focused original echo signal plus some high-frequency quantization noise, are then summed together in order to obtain a coarse representation of beam sum. The coarse beam sum is further processed by a decimation filter to attenuate high-frequency quantization noise, so that hereby an accurate beam sum with the bit resolution required to represent the desired dynamic range at the output, is recovered. The beamforming process according to the present invention is split into two parts: dynamic delaying and combining. Since dynamic delays are employed prior to digitization, only summing operation, which is a linear operation, is incorporated between ΔΣ modulators and demodulator. Therefore, an advantage of the present invention is that resultant beam sums have a SNR performance identical to the theoretical SNR expectations.

The beamforming frequency BF I^S a precalculated value, and is determined according to the desired dynamic range at the output of ΔΣ modulators. However, whatever the case, fgp is preferably greater than 32 times the frequency at which the ultrasound waves are radiated, so that the desired time delay quantization is achieved for the fine delaying. The medical ultrasound imager of the present invention can be easily adapted for extracting the flow information of moving objects by including a signal processor, for example a pulsed Doppler processor. The output of receive beamformer is forwarded to the signal processor, and usually, time or frequency domain Doppler techniques or correlation based motion imaging approaches are used to obtain the flow information which can be displayed by means of a spectrum.

The foregoing and other advantages, features and objects of the present invention will be apparent from the following detailed description of the preferred embodiments. For a complete understanding of the present invention, reference is made to accompanying drawings which form a part hereof, and in which the preferred embodiments are shown by way of illustration. However, such embodiments do not necessarily reflect the full scope of the invention, thus further reference is made to the claims herein for covering the full scope.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram of a preferred embodiment of the medical ultrasound imager of the present invention;

FIG. 2 illustrates the sector scanning format of the medical ultrasound imager of FIG. 1, and the differences in paths from any two consecutive receive focal points to an arbitrary array element, which is used to determine a minimum master clock accuracy to avoid sample repetitions; FIG. 3 is a plot illustrating minimum desired master clock frequency f_m normalized by beam- forming frequency fβ_F versus the range of the first focal point ro, for different values of f/number apodization constant /_num;

FIG. 4 is a block diagram of TGC & PREAMP array which forms a part of the medical ultrasound imager of FIG. 1;

FIG. 5 is a block digram of the embodiment of the receive beamformer used in the medical ultrasound imager of FIG. 1, which employs the principles of the present invention;

FIG 6 is functional block diagram of Sampling Clock Generator (SCG) used in the receive beamformer of FIG. 5; FIG 7A and 7B illustrate the functional block diagram of the preferred ΔΣ modulator topology and block diagram of the decimation filter, respectively, used in the receive beamformer of FIG. 5; and

FIG.s 8A-8C are the emulated B-scan images using different type of beamformers illustrative the effectiveness of the present invention.

DESCRIPTION OF THE PREFERRED EMBODIMENT

Referring to FIG. 1, a preferred embodiment of the medical ultrasound imager of the present invention includes an array of N piezoelectric transducer elements, which is hereinafter referred to as transducer array 101, for use in converting a plurality of electrical signals into ultrasonic signals that can be applied to tissues, and for use in converting reflected ultrasonic signals into electrical signals. The elements of transducer array 101 are positioned with respect to each other so as to form a linear array, as illustrated. The number of array elements N is an even number and can be 64, 96, 128 and even as high as 256 depending upon the application. However, for purposes of example only, in the preferred embodiment of the present invention, there are 128 array element hosted in the transducer array 101. The imager further includes a digital controller, referred to as controller 102, for directing and coordinating the operations of the pulse generator 103, transmit beamformer 104, transmit-receive switch 105, TGC & PREAMP array 107, receive beamformer 109 and signal processor 111.

The basic operation of system is initiated by the controller 102 causing the pulse generator 103 to produce N pulses. These set of pulses are delayed by the transmit beamformer 104 to steer and focus the transducer array 101 at an angle θ and range Rj, respectively. The delayed electrical signals at output of the transmit beamformer 104 are driven to transducer array 101 through the transmit-receive switch 105, which is readily placed to its transmit mode by the controller 102, and converted to ultrasound waves at electrodes of transducer array elements 101(1)-101(N). As ultrasound waves propagate, they are interacted with the tissue of interest by means of reflection, scattering or absorption causing some echo signals return back to the transducer array 101 whereas some portions of the transmitted waves continue traveling toward o the tissue. Therefore, following transmission of ultrasound waves into the tissue, the controller 102 switches the transmit-receive switch 105 after a short offset time to its receive mode to drive received electrical signals at the transducer array 101 to the TGC & PREAMP array 107. The N received electrical signals 106, which are representatives of N received echo signals, are time-gain-compensated and then further amplified by N respective time-gain amplifiers and pre- amplifiers, which comprise TGC & PREAMP array 107. The amplified signals 108 are then applied to the receive beamformer 109 to be appropriately delayed and summed together, hence to produce the steered and focused beam sum 110. Since a 90° sector scanning is composed of series of the beam sums 110 for successive steering angles θ, the procedure outlined above is repeated for different steering angles until all the angles are scanned. The signal processor 111 typically envelope detects, scan converts and compresses the beam sums 110 to produce a B-scan image of the tissue of interest, which is displayed on the CRT unit 112. As mentioned before, the medical ultrasound imaging system of the present invention can also be used for generation of color flow images. In this mode of operation, commonly referred to as color flow mode, the signal processor 111 may employ some flow processing techniques prior to the envelope detection. It should be apparent to those skilled in the art that the controller 102 can be in many forms depending on the flexibility and versatility of the particular medical ultrasound imager. In the preferred embodiment described above, the controller 102 includes a programmable digital micro-controller and an interface unit to supply interaction between operator and imager.

Since a B-scan image is the envelope of the focused and steered radio frequency (RF) beam sums, detection of envelope should be carried out by the signal processor 111. Furthermore, as described before, the beam sums 110 are acquired in its natural polar (r, θ) form, which is not convenient for display, therefore, a scan conversion is employed to transform the data from polar

(r, θ) format to rectangular (x, y) format. This is accomplished by simply interpolating each point on the (x, y) grid from its four neighbors on the (r, θ) array. The rectangular (x, y) repre- sentation of image is further logarithmically compressed to a desired dynamic range depending upon the application. Such a polar coordinate to cartesian coordinate conversion is described in an article by S. C. Leavitt, published in Hewlett-Packard Journal, pp. 30-33, October 1983, an entitled "A scan conversion algorithm for displaying ultrasound images" , herein incorporated by reference. In the color flow mode, to extract flow information the signal processor 111 uses data prior to envelope detection employing an algorithm such as known, for example, from an article described by Halberg et. al. published in Hewlett-Packard Journal, pp. 35-40, June 1986, and entitled "Extraction of blood flow information using doppler-shifted ultrasound" . Design and implementation of such a signal processor is possible by using special-purpose micro-chips and/or commercially available programmable microprocessors manufactured for processing large volume of data, and therefore, found beyond the scope of the invention.

Referring to FIG. 2, in conjunction with FIG. 1, the transducer array elements 101(1)- 101(N) are separated between each other with an equal distance of d = -^ . The number of beam lines per frame, B, scanned over the sector should be consistent with the equation (7). In another preferred embodiment of the present invention, there are B = 182 total beam lines acquired for a 90° sector with each beam line being steered in increments of Δ sin ø = 0.0078, where θ is the steering angle measured with respect to array normal 203. Under directions of the controller 102, transmit beamformer 104 imparts steering and focusing delays to each of the respective pulses applied to successive transducer array elements 101(1)-101(N) according to the equation: x_n ² x_n sin θ __, .„. _{Tn =} __^_ ₊ ^___ _{+ ro (8})

where x_n = (n — y - 0.5)d. Here, n = 1, 2, , N enumerates the individual array elements and RT is the fixed focusing range, which is usually taken as the mid-range of imaging depth (i.e Rγ = §) . The constant time o is a sufficient magnitude included to compensate negative delays across the array. The design of such a transmitter is beyond the scope of the present invention, and for a detailed description, reference is made to U.S. Pat. No. 4,870,971 issued in Oct. 3, 1989 to R. H. Russell, and entitled "Transmit focus generator for ultrasound imaging" , incorporated herein by reference.

In accordance to the principles of the present invention, the receive beamformer 109 employs dynamic focusing along each scan angle, by imparting the variable time delay to each of the respective time-gain compensated and further amplified echo signals 108(1)-108(N) received by the transducer array elements 101(1)-101(N) according to the equation:

^{Tn {r)} - - 2Tc ⁺ ~ ~c~ ^{+ To (9)} where r is the range of receive focal point. During receive beamforming, r is changed dynamically to focus the array to the desired image points. Therefore, for a predetermined scan angle θ, receive beamformer circuit applies a different delay on each channel, and varies that delay with depth (or time) . This is accomplished by changing the range of focal point r from the desired first focal point range ro to the image depth R (i.e. ΓQ < r < R ) with increments of Δr. The difference in range of consecutive two focal points, Δr, can be determined by Δr = _jf — where fβF indicates the beamforming frequency corresponding to the time of flight difference for propagation between two consecutive focal points.

The beamforming frequency fβ_F i a, precalculated value, and is chosen according to the desired SNR at the output of the ΔΣ modulators. It is a well known fact that, there is a direct relationship between the bit resolution of an A/D converter and its SNR. According to P. M. Aziz et. al., recited before, for a conventional multi-bit A/D converter, the SNR can be expressed by the following equation:

SNR = 10 log

- 10 log v + 4.77 + 6.02 (dB) (10) where σ². and W stand for input signal power and number of bits of resolution, respectively. Furthermore, V_m is the maximum amplitude of the input signal; i.e. the input signal amplitude is always in the range of V_m and —V_m. On the other hand, as explained in an article by R. M. Gray, published in IEEE Trans, on. Communication, vol. COM-35, pp. 481-489, May, 1987, and entitled "Oversampled sigma-delta modulation" , the in-band SNR for an L^th order single-bit ΔΣ modulator can be written as:

SNR = 10 log σ_x ² - 10 log (dB) (11)

where f_s is the sampling frequency and fβ is the bandwidth of the input signal. As it will be explained in more detail, by using Pre-Amp array, associated for each channel, it is ensured that all incoming echo signals are always limited by a predetermined voltage value of V_m. Each ΔΣ modulator, comprising the ΔΣ array, is set to operate according to this maximum input voltage range, V_m. In medical ultrasound, usually 8 or 10 bits of conversion resolution is required. Assuming 10 bits of resolution (i.e. W = 10), the desired SNR at the output of conventional multi-bit A/D converter corresponds to (64.97 -I- 10 log

- 10 log V ) (dB) according to (10). For example, if the use of 2^nd order single-bit ΔΣ modulator is in consideration instead of a conventional multi-bit A/D converter, then the SNR at the output of modulator is (-2.1 + 10 log

- 10 log V² + 50 log fz;)) (dB) according to (11). Equating and solving above equations for - f-, yields an oversampling ratio approximately equal to 22 (i.e. OSR « 22). The beamforming frequency fβ_F can be determined according to this OSR, for example, assuming 3.5 MHz transducer with a fractional bandwidth of 50 %, which corresponds to a bandwidth of fβ equal to 4.38 MHz, results a beamforming frequency of fβ_F = 192.5 MHz. For this pilot case, the distance between adjacent focal points is determined approximately as 4 μm (i.e., Δr « 4 μm). The design methodology developed here for such 2^nd order modulator, 3.5 MHz transducer center frequency and 10 bits of conversion resolution can be easily extended without loss of generality.

FIG. 2 also illustrates the distances from any two consecutive focal points, at ranges r and r + r respectively, to n^th array element 101 (n). As a matter of convenience, distances from the focal point r and r + Δr to n^th array element are called pi and _{2 ι} respectively. As it is clear that, ultrasonic waves reflected from r and r + Δr reach at different time instances to the transducer element 101 (n), where time difference, denoted as Δt, is given by At = 2^{P ~Pl} . By using simple geometrical tools, p_\ and p can be written as follows:

/>₂ = sj(r + Δr)² + χl - 2(r + Δr)x_n cos ( - θ) (13) where x_n is defined as the distance from the center of n^th array element 101(n) to the array center and given by x_n = (n — y — 0.5)d. For the sake of simplicity, using the paraxial approximation, also used in calculating transmit and receive beamforming delays, more simpler expressions for p_\ and />₂ can be obtained such that: x² />ι « r + -^ - x_n sin 0 (14) λr

^ r + Δr + _2(r ^_Δr) - x_n sin fl (15)

Thus, time difference At can be expressed as:

Δr

Δr - ^- (16) c 2 r(r + Δr) Solving for At = 0 gives r(r + Δr) = \ For the receive beamformer based on oversampling A/D conversion, Δr is chosen sufficiently small compared to r to meet the desired SNR condition and therefore can be reasonably omitted in the last equation. Consequently, it is obtained that Δt = 0 for r R. £fe, which means that from consecutive focal points r « ^a. and r+Δr « ^ +Δr reflected echoes may arrive exactly at the same time to the n^th array element 101(n), which means further that these two focal points require the use of same sample on the n^th channel during dynamically focused beamforming due to acoustic geometry. However, fortunately, by using f/number apodization, which is readily used to improve image quality, the above mentioned condition is never met. The f/number apodization can be expressed as:

where f_num is a predetermined integer constant (i.e. f_num = 1, 2, 3....) and D represents the length of array. It is obvious from equation (17) that for focal points close to the array, not the whole array, but only a portion of it is allowed to contribute to the receive beam formation. On the other hand, beyond a certain focal range all the array elements are activated in receive. In other words, the range of receive focal point is always larger than the array length by a factor of a constant (i.e. r > f_numD ), therefore the condition r « never occurs which requires the use of the same sample for some focal points due to acoustic geometry. On the other hand, the aforementioned sample repetitions during receive beamforming due to time quantization can be eliminated, if a time resolution, referred to as T_m hereinafter, smaller than the minimum time difference (At)_mi_n is selected. That is T_m < (Δt)_mi„, where

2 A _ l^n. mαx &^r

(At)min = - (18) c 2 r_Q(r₀ + Δr) can be easily found by minimizing equation (16) with respect to parameters n and r. Here (ι-)_mα- is the distance from the most far array element to array center, and ro is the desired first focal point range. Substituting Δr = f— in to the equation (18) gives:

Thus, the minimum master clock frequency f_m = ψ- to resolve repeated samples required for the beam formation can be expressed as:

FIG. 3 demonstrates a plot of the ratio between minimum desired master clock frequency f_m and the beamforming frequency fsF versus the range of first focal point o- The plot is obtained for different values of f_num by using equation (20) where ro is swept from ^ to 5D. The system parameters are chosen as described previously such that N = 128, d = _j and o = 3.5 MHz. As it is apparent from the plot of FIG. 3, in the worst case which is when the depth of the first focal point is less than the array length multiplied by f/number apodization constant (i.e., ro < fn_umD ) and usually the case, there is a constant ratio between f_m and feF, whereas for bigger r₀ the constraint on f_m is relaxed. Clearly, for a fixed f_num there is a worst case in which the ratio of f_m and }BF is only depends on f_nUm for any choice of ro- To find out this constant ratio, since — § < x_n < § for all used array elements, irp — can be substituted instead of (^n)_mox in to the equation (20) . Moreover, Δr is omitted when compared to ro, so that hereby a more reduced expression for the desired minimum clock frequency f_m is obtained, which is independent from ro and given by:

8 f_

Consequently, a master clock frequency f_m at least ..j^. times the beamforming frequency fBF h s to be chosen in order to avoid sample repetitions during receive beam formation for any choice of ro when an f/number apodization constant of /_num is used.

With reference to FIG. 4, the TGC & PREAMP array 107 include N Time-Gain-Control 0 (TGC) amplifiers 401 and N pre-amplifiers 402 for amplifying N electrical signals 106(1)-

106(N) received by the transducer array 101 prior to beamforming. The gains 404 of TGC amplifiers 401 are adjusted by a set of variable potentiometers 403 which can be controlled manually by an operator and/or automatically by the controller 102. As stated before, the attenuation in tissue that ultrasound waves are employed, is related to the total flight path, therefore amplitude of ultrasonic wave vanishes as it propagates to deeper depth. The TGC amplifiers 401 are used for compensating this attenuation, hence a relatively uniform contrast over different range of points which have the same acoustical characteristic is readily obtained. The output of each TGC amplifier 401(1)-401(N) is forwarded to respective pre-amplifier 402(1 )-402(N). The pre-amplifier circuitry 402 is used for further amplifying and clamping the time-gain-compensated echo signals at a voltage level suitable for subsequent processing. The incoming echo signals are clamped at an appropriate predetermined voltage level V_m to control the input of oversampled A/D converters (ΔΣ modulators) to have the maximum input dynamic range of the converters. Stated another way, the time-gain-compensated echo signals are not only further amplified by the pre-amplifiers 402, but also guaranteed to be in the range of -V_m and V_m. FIG. 5 illustrates receive beamforming section, referred to as receive beamformer 109, which forms a part of the medical ultrasound imager shown in FIG. 1. Referring to this figure, the receive beamformer 109 consists of N sample and hold devices (S/H), N ΔΣ modulators, N one-bit wide buffers, a summing node, a digital decimator filter and, finally, a sampling clock generator, which are referred to as S/H array 501, ΔΣ array 502, buffer array 503, summing node 504, decimation filter 508 and sampling clock generator 505 hereinafter, respectively. The receive beamformer 109 is responsible for imparting the appropriate differential beamforming delays into each of the received echo signal for dynamic focusing and steering.

The time-gain-compensated and amplified echo signals 108(1)-108(N) are sampled by the S/H array 501 at the time instants required for the receive beam formation. To achieve this, the sampling clock generator produces N different non-uniform clocks 506 and 507 under directions of the controller 102, and each S/H device 501(1)-501(N) is clocked according to the respective generated non-uniform clocks 507(1 )-507(N). The different non-uniform clocks 506(1)-506(N) and 507(1)-507(N) generated by the sampling clock generator 505 can be expressed as: c_n(t) = ∑ m(t)δ(t - Q[^{P ' TBF ~ Tn(r)}] m) (22) p=-0 ^{l m} where P is the total number of focal points on a beam line ranging from o to R with Δr increments, TBF — _j^~ is the time period between consecutive focal points, δ(.) is the Dirac delta or impulse function, and τ_n(r) is the receive beamforming delay defined by the equation (9) wherein r = p ^F + r₀. T_m = 4- is the period of master clock rate chosen according to the equation (21). The function Q[x] represents the smallest integer greater than x and the function m(.) is given by:

Steering and focusing are accomplished by sampling the incoming echo signals at the instants required by the receive beamforming as indicated previously. N steered and focused analog samples are then digitized using N one bit ΔΣ modulators 502(1)-502(N), comprising the ΔΣ array 502, prior to coherent summation. Each ΔΣ modulator 502(1)-502(N) is also clocked with one of the respective non-uniform clocks 506(1)-506(N) for providing synchronization with the corresponding S/H device 501(l)-501(N). The output of each ΔΣ modulator is a stream of one bit digital data including the original steered and focused respective echo signal plus some high frequency quantization noise. Since, the arrival times of the echo signals corresponding to different focal points is non-uniform in time, one bit representations of echo samples at the output of ΔΣ modulators appear at different time instants. If the time period between two consecutive focal points TBF is grater than the time corresponding to the maximum receive delay between elements for all focal points and for all steering angles, max[τ_n(r)], then for any focal point all the samples required for receive beam formation appear before any beamforming sample for the consecutive focal point. Here, τ_n (r) is the formula described in (9) with omitting the constant time o which is incorporated to avoid negative delays. However in general, TBF < max[τ_n(r)] and therefore all samples occurring over a period of max[τ_n(r)] at each channel must be dynamically stored. The digital buffer array 503 is used for this purpose where the minimum length of each buffer 503(1)-503(N) can be determined by:

L_b = ^mα*^[τ"^(r)l (24)

TBF

With regards to the buffer length, further reference is made to the article by T. K. Song et. al. published in IEEE Trans, on Ultrason., Ferroelec, Freq. Contr., vol 41, No. 3, pp. 326-332, May 1994, and entitled "Ultrasonic dynamic focusing using an analog FIFO and asynchronous sampling" , herein incorporated for reference.

The delayed one bit coded samples at all array channels are piped out to summer node 504 simultaneously to compose a digital signal, which is referred to as the coarse beam sum 509 hereinafter, wherein each sample is represented by preferably 8 bits of resolution. As stated before, since beamforming delays are accomplished with the non-uniform sampling clocks 507 generated by the sampling clock generator 505, echo samples digitized by one bit appearing at different time instances are synchronized through buffer array 503 and directly summed at the summer node 504. The coarse beam sum 509 is further processed by the decimation filter 508 to eliminate high frequency quantization noise injected by the ΔΣ modulators 502 and also to reduce the sampling rate to preferably near to the Nyquist rate, so that an accurate representation of beam sum 110 is obtained wherein each sample is represented by a number of bits to represent the required dynamic range at the output. Although 17 bits of resolution is required to represent the entire dynamic range of beam sum (assuming each echo signal is quantized with W = 10 bits of resolution) , this high resolution may exceed the dynamic range of the display device. Therefore, depending upon the application and dynamic range of the CRT unit 112, the accurate beam sum 110 is usually truncated down to a number bits representing the desired dynamic range. In the preferred embodiment of the present invention, the samples of the accurate beam sum 110 is represented with, for example, 12 bits of resolution rather than 17 bits, however, this should not be always the case and according to the principles of the present invention the entire dynamic range of beam sum can also be provided if necessary.

With reference of FIG. 6, the sampling clock generator 505 incorporates a digital beamforming memory 601 and N "and-gate" array 602. The beamforming memory 601 is a 2D matrix whose columns and rows represent beamforming sampling times according to the beamforming frequency fβ_F associated with the master clock rate f_m and the index of the array element n, respectively. Each column consists of a stream of "1" and "0"s where "l"s represent the receive beamforming samples, and "0" 's indicate that the corresponding samples is unnecessary for the beamforming, and hence, that samples should be dismissed. The location of "l"s is determined according to following relation:

whereas the remaining locations are incorporated with "0"s. The length of each column is defined by: R - ro)} column ⁼ ty (26)

C - Tm J and hence, total size of the digital beamforming memory 601 can be given as L_coι_umn • N bits. For the typical values preferred in the embodiments of the invention, a digital RAM or ROM with a total size of 51.5 A^" x 128 bits is required as the beamforming memory 701.

The generation of the non-uniform clocks 506 and 507 defined by equation (22) is accomplished by simply performing a logical "and" operation between the content of the beamforming memory 601 and the master clock f_m using the "and-gate" array 602. At each clock cycle of the master clock f_m , the associated bit is read from each row of the beamforming memory 601 and a logical "and" operation is executed with the master clock through "and- gate" array 602(1)-602(N) to produce the associated non-uniform clocks 506(1)-506(N) and 507(1)-507(N), therefore a maximum access time of 4- is required for the beamforming mem- ory 601.

It is appreciated by those skilled in the art that there are many types of ΔΣ modulator structure which can be applied successfully to the present invention. FIG. 7A illustrates the preferred topology for ΔΣ modulator 502(n) which is an element of the ΔΣ array 502. The preferred topology is a standard 2^nd order modulator which is widely used in the art of communication. Since the input of modulator 701 is the associated non-uniform sampled analog echo signal, the analog delay elements 704(n) and 707(n) in integrators are also clocked with the non-uniform clock 506(n) generated by the sampling clock generator 505. The quantized signal is an integrated version of the difference between the analog input 701 and an analog representation of the digital output 710. The quantizer 708 is preferably a single-bit quantizer (or simply a comparator) with output voltage levels V_m and — V_m. The levels V_m and — V_m are digitally coded as "1" or "0" , thus the output of the modulator 710 is a stream of one bit digital data. The digital to analog converter 709 used in the feedback loop supplies V_m or —V_m volts if its input is "1" or "0", respectively. The ΔΣ modulator 701 can be implemented using several VLSI techniques such as based on switched-capacitor, charge coupled device (CCD) or even super conductor design methods. Some examples of switched-capacitor implementation of oversampling converters are disclosed in U.S. Pat. No. 4,999,634 issued in Mar. 12, 1991 to Brazdrum et. al, and entitled "Integratable switched-capacitor sigma-delta modulator" and U.S. Pat. No. 5,140,325 issued in Aug. 18, 1992 to Yu et. al., and entitled "Sigma-delta analog- to-digital converters based on switched-capacitor differentiators and delays". For ultrasound frequencies, the desired oversampling ratio may not be achievable by analog circuits that can be attainable with switched-capacitor techniques or other technologies, and therefore super conducting methods may be used, such as known from U.S. Pat. No. 5,198,815 issued in Mar. 30, 1993 to Przybysz et. al., and entitled "Two loop superconduncting sigma-delta analog-to-digital converter" .

It is always ensured that the analog input 701 of each modulator 502 (n) is limited by ^V_m so that input of the quantizer 708 can have a magnitude of at most 2V_m. In other words, the quantizer is never overloaded. Stated before, all incoming echoes are clamped at the voltage level V_m by the pre-amplifiers 402 for proper operation of the ΔΣ array 502. Referring to FIG. 7B, the decimation filter 505 consists of a low-pass filter and a down sampler.

The low-pass filter, which is used to filter out high frequency quantization noise injected by the ΔΣ modulators 502, is a multi-bit digital filter whose output is represented by a number of bits required to represent the desired dynamic range at the output, and whose input is the coarse beam sum 509 represented by 8 bit resolution. The low-pass filter performs an energy averaging process and is preferably an FIR filter such that described by E. Dijkstra et. al. in the article published in ISC AS 87, pp.479-482, and entitled "A design methodology for decimation filters in sigma-delta A/D converters" . This paper is incorporated herein by reference. Moreover, a frequency decimation by a factor of L = 22 is performed after filtering to reduce the sampling rate from fβ_F = 192.5 MHz to the Nyquist rate 2 β = 8.75 MHz. Finally, the output of the decimation filter 505 termed as the accurate beam sum 110, which is a digital signal with preferably 12 bits of resolution representing steered and focused beam data at a rate of 8.75 MHz, is provided to the signal processor 111.

FIG.s 8A-8C are the emulated B-scan sector images using different type of beamformers employing prior art methods and the principles of the present invention. The images are reconstructed using the RF data acquired from a wire phantom with a commercial 128-element 3.5 MHz transducer array connected to a multi-bit A/D converter. These kind of data is usually used to test beamforming algorithms. The complete data set consists of 128 X 128 records obtained from all possible transmit-receive combination of the array elements. Each record is sampled at 13.89 MHz and quantized to 10 bits of resolution. The sampling rate of RF data is increased through digital interpolation by a suitable factor. The phantom contains six wires in a water tank at different angles and ranges, and is used to assess the point spread function at different spatial positions. FIG. 8A, 8B, and 8C illustrate B-scan images of wire phantom obtained by using different types of dynamically focused beamformers wherein 10 bit A/D converters, single-bit ΔΣ modulators as proposed by Noujaim et. al. (recited before), and non-uniform sampling scheme with single-bit ΔΣ modulators as described within the principles of the present invention, are employed, respectively. All B-scan sector images, obtained through envelope detection, scan conversion and logarithmic compression, are presented over 70 dB dynamic range. A fixed focused transmit beamforming is employed over all records to compose the incoming echo signals. After that, dynamically focused receive beamforming running at 222 MHz (i.e., fs_F = 222 MHz) is emulated on these echo signals wherein an f/number apodization constant of 2 is also used. It should be also noted that in the latter two beamformer an ideal software simulation of 2^nd order single-bit ΔΣ modulator is incorporated on each channel to digitize the samples of echo signals. The up-sampling factor for the first two and for the last beamformers is chosen as 16 and 17, respectively.

FIG. 8A shows the B-scan sector image generated by ideal 10-bit dynamically focused beam- former. The strongest reflector is the third wire (when counted from the top) as expected, since fixed transmit focusing range is selected as the mid-range of imaging depth. As moving away from the mid-range, point spread functions begin to being distorted due to the fixed transmit focusing. Acoustic artifacts near the third and fourth wires, probably due to air bubbles in the water tank, are also visible. FIG. 8B illustrate B-scan sector image using an architecture similar to that of proposed by Noujaim et. al. (recited before). Although the wires are visible over a snowy background noise, the noise floor is worsened approximately by 20 dB compared to conventional beamformer of FIG. 8 A. These unexpected noise is primarily due to sample repetition during dynamically focused beamforming and limits the ultimate use of this kind of beamformer in clinical settings. FIG. 8C presents the B-scan sector image generated by a dynamically focused beamformer employing the principles of the present invention. A significant improvement in image quality is evident where the noise floor is only approximately 1.5 dB worser than the ideal beamformer of FIG. 8A. The point spread functions at six wires are nearly equal to that of ideal 10-bit beamformer and the acoustic artifacts can also be visible.

There has been shown only certain preferred features of the invention which satisfies all the objects and advantages sought, however, many changes, modifications and other applications of the present invention may occur to those skilled in the art without departing from the true spirit and scope of the invention as defined by the appended claims.

Claims

What is claimed is: 1. A medical ultrasound imaging system comprising; an array of piezo-electric transducers with N elements for transmitting N separate ultrasonic signal into an object being imaged, and in response thereto, producing N analog echo signals; a pulse generating means producing N separate pulses for driving the transducer array in transmission mode; a transmit beamforming means for separately delaying the said pulses to produce a steered and fixed focused beam at each scan angle in transmission mode; a set of N amplifying means for compensating, further amplifying and clamping said N separate analog echo signals produced by the said transducer array; a receive beamforming means for appropriately delaying and summing said N separate analog echo signals to produce a beam sum; and a digital processing means for envelope detection, scan conversion and logarithmically compressing of the said beam sums to produce an image of object being displayed in a displaying means.

2. The medical ultrasound imaging system as recited in 1 further includes a digital controller means for directing and coordinating the operation of the said medical ultrasound imaging sys- tem.

3. The medical ultrasound imaging systems as recited in 1 wherein the receive beamforming means comprises a clock generating means for generating N separate non-uniform clocks according to the receive beamforming delays.

4. The receive beamforming means as recited in 3 comprises N separate sample and hold means for sampling and holding one of the said respective N analog echo signals.

5. The set of _V sample and hold means as recited in 4 is adapted to operate according to one of the said respective N non-uniform clocks for providing only the echo samples required by the receive beam formation.

6. The receive beamforming means as recited in 3 includes a set of N oversampled analog to digital converters to digitize one of the said respective N delayed echo signals.

7. The set of N oversampled analog to digital converter of claim 6 is adapted to operate according to one of the said respective N non-uniform clocks for providing synchronization with one of the said respective sample and hold means.

8. The set of N oversampled analog to digital converter of claim 6 contains internal quantiz- ers with a bit resolution far less than the number of bits required to represent the entire dynamic range of the said echo signals.

9. The set of N oversampled analog to digital converter of claim 6 are all set to have the maximum input dynamic range at which the said echo signals are clamped by the said amplifying means.

10. The receive beamforming means as recited in 4 employs f/number apodization with a minimum f/number apodization constant of f_num by avoiding contributions of some elements of said transducer array for some focal points during receive beam formation.

11. The controller means as recited in 2 is responsible for generating a master clock whose rate is greater than or equal to _╬æ╬¢"^u l, times the frequency at which the samples of said beam sums are acquired.

12. The frequency at which the samples of said beam sums are acquired, is dominantly determined by the desired dynamic range at the output of said oversampled analog to digital converters.

13. The frequency at which the samples of said beam sums are acquired as recited in 12, is preferably more than 32 times the frequency at which ultrasound waves are radiated for easily achieving fine delaying.

14. The clock generating means as recited in 3 comprises; a digital storage memory whose cells are one bit wide, wherein "l"s and "0"s are stored according to the beamforming delays; and a set of N "and-gate" means to perform a logical "and" operation between the said master clock and the content of said digital storage memory to produce the set of said N non-uniform clocks each corresponding to one of the respective N sample and hold means and oversampled analog to digital converter means.

15. The receive beamforming means as recited in 3 includes a set of N buffer means for avoiding sample disappearing that may be occur because of non-uniform sampling employed during receive beamformation.

16. The set of N buffer means as recited in 15 include a plurality of cells each storing a number of bits determined by the said internal quantizers of said oversampled analog to digital converters.

17. The number of said cells in one of the said respective N buffer means is determined by the time, corresponding to the maximum receive beamforming delay, multiplied with the said frequency at which the samples of beam sums are acquired.

18. The receive beamforming means as recited in 3 comprises an adder means for arithmetically summing the outputs of said N oversampling converters to compose a coarse representation of said beam sum.

19. The receive beamforming means as recited in 3 comprises a decimation means for filtering out high frequency quantization noise on the said coarse beam sum introduced by the said oversampling converters and, moreover, for reduction of sampling rate to Nyquist rate.

20. The medical ultrasound imaging system as recited in 1 wherein the said digital processing means is adapted to extract the flow information of moving objects through some suitable signal processing techniques.