Compact Beamformer Design with High Frame Rate for Ultrasound Imaging

Transcription

1 Sensos & Tansduces 2014 by IFSA Publishing, S. L. Compact Beamfome Design with High Fame Rate fo Ultasound Imaging Jun Luo, Qijun Huang, Sheng Chang, Xiaoying Song, Hao Wang Wuhan Univesity, Depatment of Electonic Science and Technology, School of Physics and Technology, Wuhan Univesity, 16 Luojia Mountain Road, Wuhan, , China Tel.: , fax: Received: 29 Septembe 2013 /Accepted: 7 Mach 2014 /Published: 30 Apil 2014 Abstact: In medical field, two-dimension ultasound images ae widely used in clinical diagnosis. Beamfome is citical in detemining the complexity and pefomance of an ultasound imaging system. Diffeent fom taditional means implemented with sepaated chips, a compact beamfome with 64 effective channels in a single modeate Field Pogammable Gate Aay has been pesented in this pape. The compactness is acquied by employing eceive synthetic apetue, hamonic imaging, time shaing and linea intepolation. Besides that, multi-beams method is used to impove the fame ate of the ultasound imaging system. Online dynamic configuation is employed to expand system s flexibility to two kinds of tansduces with multi-scanning modes. The design is veified on a pototype scanne boad. Simulation esults have shown that on-chip memoies can be saved and the fame ate can be impoved on the case of 64 effective channels which will meet the equiement of eal-time application. Copyight 2014 IFSA Publishing, S. L. Keywods: Ultasound, Time-of-flight, Beamfome, Synthetic apetue, FPGA. 1. Intoduction Ultasound diagnosis and teatment have been widely used in medical field. Two-dimension (2D) ultasound imaging has played an impotant ole in clinical diagnosis. Typically, ultasound imaging system is composed of the font-and back-end pocesses [1]. A high-quality beamfome is essential fo the font-end ultasound imaging system. Delayand-sum (DAS) and minimum vaiance (MV) ae two basic theoies fo beamfome design. Although MV method can give a high-quality image output, it is had to be implemented by hadwae because of its complex algoithm. DAS method is an actual way fo beamfome's cicuit implementation. Tacing the evolution of ultasound imaging histoy, many techniques have been investigated to simplify stuctue and impove image quality. Synthetic apetue (SA) [2] can obtain equivalent image quality of 64 effective channels by active 32 channels. Hamonic imaging (HI) [3] is an impotant means to impove image quality. Combining these two techniques will futhe pomote the beamfome's pefomance. Diffeent paametes ae needed fo both synthetic apetue and hamonic imaging. A lage numbes of on-chip memoies [1] wee peviously consumed to stoe the huge paametes; hence it is not flexible fo online configuation. Taditionally, beamfome is implemented by Application Specific Integated Cicuit (ASIC) o by pocessos, such as Media Pocesso (MP), Gaphics Pocessing Units (GPU) and Digital Signal Pocesso (DSP) [4]. The cost is expensive when it is implemented by ASIC and it is usually difficult to Aticle numbe P_

2 suppot high data ate on eal-time ultasound imaging applications due to thei seial popeties in MP and DSP. GPUs can povide a highe fame ate, but the complex stuctue and nomally the necessay of pesonal compute (PC) limit thei applications in potable devices. Field Pogammable Gate Aay (FPGA) have povided an attactive appoach in ultasound imaging systems [1, 2, 5] fo the vitues of flexibility, paallel technique and the capacity of high thoughput. Yang [2] applied constant paametes in beamfome, but it cannot be configued online and the memoy cost is expensive. Schneide [1] intoduced a 64-channel B-mode imaging system based on a FPGA and a DSP. The expeiments showed that the system can each 120 scan lines pe fame and 30 fames pe second. A beamfome with capacity of pocessing pixels image at 40 fames pe second has been pesented by Chen [5]. He used two FPGAs to implement the beamfome. Nomally, it is a tadeoff between cost and pefomance, but thee is still a lage magin to be exploed. In this pape, a single-fpga-based compact beamfome with dynamic configuation is pesented. Fo the convenience of hadwae implementation, a simplified time-of-flight calculation method, which uses a quantized ac length fomula, is poposed. To impove the fame ate and fulfill the equiement of eal-time ultasound imaging, multi-beams (fou beams pe emission) method is adopted. Dynamic configuation povides a flexible appoach fo diffeent applications. Thus synthetic apetue and hamonic imaging can be integated to a single FPGA with efficient memoy utilization. Time shaing and intepolation ae also employed to educe the esouce cost of the beamfome. Next sections ae oganized as follows: The simplified time-of-flight calculation is deduced in section 2. Achitectue and methodology of ou design ae pesented in section 3. Implementation and esults ae shown in section 4, and the conclusion is dawn in section Method of Time-of-flight Calculation To get a focus point, dynamic time-of-flight (TOF) calculation has been pesented [7, 8]. But fom the aspect of cicuit implementation, the wok is still too complex. To ovecome this poblem, a simplified appoach of calculating the time-of-flight of a tansmit/eceive ultasound wave is pointed out. Fig. 1 is the mathematic model of the tansmit/eceive delay distance calculating diagam fo a convex tansduce aay. Thee ae N elements in the pobe aay, the pobe adius is R, and the pobe pitch is P. i is the specific element of the channels. Fom the basic sonic wave focus theoy, sonic waves ae bilateal symmety. On tansmitting, the cental line channel ( y axis) has the longest delay time, which is located in the middle of the tansduce aay. Relative delay distance ( D t ) of element i can be given by (1). D t is the absolute tansmitting delay distance and F is the focus depth of the scan lines, whee x t and y t ae expessed by (2) with angle (θ ) shown in (3). θ can be descibed by the ac length fomula ( θ = P/ R). D () i = D F = x() i + ( y() i F) F, (1) 2 2 t t t t xt () i = R*sin( θ ), (2) yt () i = R*cos( θ ) R θ = θ ( i 15.5), (3) Fig. 1. Mathematical model of time-of-flight calculation. Fou beams (solid lines in Fig. 1) ae eceived afte a tansmit event. Absolute eceiving delay distance ( D ) is given by (4). n is one of the fou beams, m is one of the eceiving elements and k is the quantized scan depth. x and y can be expessed by (5) fo a convex tansduce aay. The TOF of tansmit ( T t ) and eceive ( T ) ae given by (6). The TOFs ae nomalized by clock peiod ( T clk ) and acoustic wave speed ( c =1540 m/s). If a linea tansduce aay is used, TOF can be geneated by a simila method based on the Catesian coodinate system. D knm = x nm + y nm F k + F k, (4) 2 2 (,, ) (, ) ( (, ) ()) () x ( n, m) = R*sin(( P/ R)*( m n*0.5)), y ( n, m) = R*cos(( P/ R)*( m n*0.5)) R Tt = Dt()/ i c/ Tclk T = D ( k, n, m)/ c/ T clk (5), (6) In ou method, Pola coodinate system and Catesian coodinate system ae applied in the convex tansduce aay and linea tansduce aay 238

3 espectively. In pactical tems, the coodinato x and y ae pe-calculated by Matlab and then stoed in the two-pot andom-access memoy (RAM) of FPGA. As a esult, TOF geneation is convenient fo hadwae implementation. 3. Achitectue and Methodology Schematic diagam of the whole medical ultasound system is shown in Fig. 2. Main modules (system contol, tansmit, eceive, inteface bus, signal pocess and disp contol) ae implemented in one FPGA. Off-chip memoies (RF SRAM, DRAM and DSC SRAM) ae used fo data stoage. An ARM pocesso is used as the man-machine inteface contolle to enhance capability of the system. Communication with keyboad, teminal (univesal seial bus) and pesonal compute is handled by ARM. Fig. 2. Functional block of ultasound imaging system. DAS method is applied in the eceiving pocess. Afte a tansmit event, the switch blocks the tansmitting, RF signal coming fom convex o linea tansduces will be sampled by analog-to-digital convetes (ADC). Thee ae fou ADCs with eight channels in each ADC in ou system. The RF data is sampled at 40 MSa/s befoe being sent to FPGA via a low-voltage diffeential signal (LVDS). A digital symmetic 17-tap band-pass finite impulse esponse (FIR) filte aay is used to filte low fequency noises geneated by eceive amplifies with vaiable gain. To educe the buden of multiplies, the filte is implemented only by addition and shift opeations. A beam is focused based on DAS fo phased sub aays. The geometical TOF (epesented as eading addesses) fo each channel is geneated on the fly to get the pope data stoed in on-chip RAM. To impove fame ate fo eal-time ultasound imaging, multiple eceive beams with a single tansmit event [6] is applied in this pape. The 32-channel RF data is then summed acoss the apetue domain pio to the fomation of a B-mode image. Finally, the fomed beam is scan-conveted fo video display by the signal pocessing module and the display contol module. Since expanding the apetue size will impove the lateal esolution of an image, 32 active channels ae compounded to 64 effective channels. HI can impove the tissue-hamonic and contast-agent imaging pocesses by educing the echoes centeed on the fundamental fequency while keeping the echoes centeed on the second hamonic [3] in a medical field. The eception stuctue diagam is shown in Fig. 3. The 32-channal data is ead out to pefom intepolation at the othe pot of the RAM buffe. Reading addesses ae geneated fom the Receive Focus Geneate module. Afte fou-phase intepolation completed by Phase Geneate module and Multi-plus unit module, eceiving data of each channel is apodized accoding to lateal esolution [4] and is summed to constuct a beam. The compound module is used to compound the beams of two emissions. An additional RF SRAM is used to stoe a fame image. Pipeline technique is employed on eceiving so that beam data is efeshed by evey clock. Taditionally, 32 individual addess geneato units ae needed to fom a beam if thee ae 32 channels. An efficient and compact appoach is poposed to educe esouce cost in Fig. 4. Thee ae two addes (adde and subtacte), seveal egistes (epesented as D) and fou switches (A, B, C, S) in each channel. To accomplish fou-phase intepolation, the system is capable to handle 160MSa/s (40MSa/s 4). Each channel is efeshed by evey 128 clocks (4 clocks fo the fou-phased intepolation of a point, 4 clocks 32 channels). To explain the opeating pinciple, channel 0 is taken as a demonstation. S is switched to left fo the fist 4 clocks befoe it is switched to ight fo the est 124 clocks. B and C ae acting the same with S (down fo the fist 4 clocks, up fo the est 124 clocks). 239

4 Switching of A is simila to C except that they ae not switching at the same time (4 clocks ahead of C). Diffeence of B and C is shifted to ight by 5 (divide by 32) befoe it is accumulated. As a esult, fouphase intepolation is accomplished in 4 clocks and the est 124 clocks ae used fo linea intepolation. The timing elationship of geneated addesses is illustated in Fig. 5. Geneated addesses (RD0,..., RD31) ae coming fom coesponding channels (channel 0 to channel 31 in Fig. 4). The 31 data (labeled by the fist two chaactes CH) between RD0 and RD32 is the esult of intepolated data. Since time shaing and linea intepolation techniques ae employed in this pape, only one squae oot is needed. Compaed with diect calculation of TOF, ou stuctue is compact and esouce is saved (appoximately 124 folds with 31 times of esouces ae saved). Fig. 3. Stuctue of eception. Fig. 4. Stuctue of eading addesses geneato. Fig. 5. Timing elationship of the geneated addesses. 240

5 4. Implementation and Results 4.1. Implementation The font-end design has been tested on a pototype boad with a single Cyclone IV FPGA and an ARM copocesso in Fig. 6. In ou system, 128 elements of tansduce with cente fequency of 3.5 MHz is used as the detect pobe. The tansmit focus can be set to one of seveal possible points fo any given emission. Receive focusing is dynamic. On eception, a aste-type scanning method is employed. The point by point scanning is fistly along the hoizontal line and then along the vetical line until the whole fame image is scanned. The maximum imaging depth is moe than 310 mm to fom an image of pixels. The whole system (tansmit, eceive, system contol and inteface bus modules) has been veified by Modelsim SE 6.2b and synthesized by Quatus II The synthesized esult is shown in Table 1. Specially, the esouce cost of tansmitting and eceiving blocks ae also listed, because they ae the key modules of beamfome design. tansmit event. T is the eceiving peiod epesented by 2 h/ c with h standing fo imaging depth. T d is the constant delay time between tansmitting and eceiving. F = N T = N ( T + T ) (7) ate T ate T d Relationship between maximum fame ate and imaging depth is shown in Fig. 7. The maximum fame ate is deceased with the incease of the imaging depth fo the Nomal, SA and HI scan modes. When the imaging depth is 310 mm, unde the nomal scanning mode, the HI scanning mode and the SA scanning mode, the maximum fame ate pe second is 37.77, and espectively. The elationship between maximum fame ate and scan lines pe fame fo these modes is illustated in Fig. 8 (the imaging depth is 150 mm). In all cases, fame ate is invese popotion to scan lines pe fame. It can be seen fom Fig. 7 and Fig. 8 that a high fame ate is achieved (38.64 f/s with 150 mm imaging depth and 120 scan lines) which can meets the equiement of eal-time application. Fig. 6. Pototype boad of medical ultasound scanne. Fig. 7. Maximum fame ate unde 120 scanlines. Table 1. Resouce usage based on device EP4CE75F29C6. Module Logic elements Memoy Bits Multiplie 9-bits Fequency (MHz) Beamfome Tansmit Receive Result and Analysis To evaluate diffeent scanning modes in influence of fame ate, the elationship of maximum fame ate and imaging depth is compaed. Fame ate of a medical imaging system can be calculated by (7), whee N T is the numbe of tansmitting times of a fame image and T ate is the scanning peiod of a Fig. 8. Maximum fame ate unde the imaging depth of 150 mm. 241

6 Fou beams with a single tansmit event is used to decease the times of tansmitting so as to impove fame ate accoding to (7). A pefomance compaison is summaized in Table 2. With 64 effective channels, the fame ate is impoved about 28.8 % compaed to [1]. As a esult, ou beamfome system achieved an enlaged function with impoved fame ate in a single modeate FPGA. Table 2. Pefomance compaison (imaging depth: 150 mm, scanning lines: 120). Channels Fame ate pe second Scanning mode Device In [1] Nomal A EP2C35 FPGA and a TMS320C GHz DSP Nomal Poposed SA A EP4C75 FPGA HI 5. Conclusions A flexible, univesal and compact beam-fome has been implemented in a single modeate FPGA. A compact stuctue is designed by combining synthetic apetue and hamonic imaging. To impove fame ate, a simplified TOF calculation method is investigated to achieve fou eceive beams with a single tansmit fiing. The online configuation povides much moe flexibility and wide ange of applications fo the system. Time shaing and linea intepolation ae applied to save esouce cost. Wide ange of imaging depth and multiple scanning modes ae achieved in B-mode eal-time imaging with a high fame ate. The beam-fome not only shows a bight potential in colo Dopple imaging system with less modification, and also povides necessay fame ate fo thee dimensions eal-time ultasound imaging. Acknowledgements The wok is suppoted by Natual Science Foundation of Hubei Povincial, China unde Gant No. 2011CDB272 and the Fundamental Reseach Funds fo the Cental Univesities unde Gant No Refeences [1]. F. K. Schneide, A. Agawal, Y. M. Yoo, T. Fukuoka, Y. M. Kim, A fully pogammable computing achitectue fo medical ultasound machines, IEEE Tansactions on Infomation Technology in Biomedicine, Vol. 14, Issue 2, 2010, pp [2]. S. W. Yang, H. C. Yoon, J. Cho, S. B. Kye, T. K. Song, A mobile medical device fo point-ofcae applications, in Poceedings of the IEEE Intenational Ultasonics Symposium, Beijing, China, 2-5 Novembe 2008, pp [3]. A. Tucco, F. Betoa, Hamonic beamfoming: pefomance analysis and imaging esults, IEEE Tansactions on Instumentation and Measuement, Vol. 55, Issue 6, 2006, pp [4]. P. Chen, M. Butts, B. Budlong, Medical ultasound digital beamfoming on a massively paallel pocesso aay platfom, in Poceedings of the Intenational Confeence on Pogess in Biomedical Optics and Imaging, San Diego, USA, Febuay 2008, pp [5]. L. J. Chen, R. Y. Xu, J. Yuan, An efficient Bscansample-based ΣΔ beamfome fo medical ultasound imaging, in Poceedings of the IEEE Intenational Confeence on Biomedical Cicuits and Systems, Beijing, China, Novembe 2009, pp [6]. O. Dandeka, C. R. Casto-Paeja, R. Shekha, A scalable beamfoming achitectue fo eal-time 3D ultasonic imaging using nonunifom sampling, in Poceedings of the Intenational Confeence on Pogess in Biomedical Optics and Imaging, San Diego, USA, Febuay 2006, pp [7]. S. I. Nokolov, J. A. Jensen, B. Tomov, Recusive delay calculation unit fo paametic beamfome, in Poceedings of the Intenational Confeence on Pogess in Biomedical Optics and Imaging, San Diego, USA, Febuay 2006, pp. D1470- D1470. [8]. M. J. S. Feeia, D. Santos, J. Batista, FPGA-based contol system of an ultasonic phased aay, Jounal of Mechanical Engineeing, Vol. 57, Issue 2, 2011, pp Copyight, Intenational Fequency Senso Association (IFSA) Publishing, S. L. All ights eseved. ( 242