**3.7. Performance**

A NVIDIA Quadro 4000 graphics card was used to test the beamforming time achieved with the system proposed here. This card has 256 cores and 1GB global memory. It was installed in a computer with a four-core 2.66GHz Intel Q9450 processor and 4GB RAM. GPU-based implementation of the beamformer was done and tested for all acquisition strategies exposed along this chapter. In Figure 17 computing times considering image sizes starting from 200×200 to big size 800×800 for both TFM and minimum redundancy solutions are presented where it is evident than despite using the great power of GPU's the TFM solution is a very intensive procedure.

In Figure 18, the frame rate obtained for different image sizes when 2R-SAFT and kA-SAFT are employed is presented. In particular, attending to the case of an image with 500×500 24 Breakthroughs in Ultrasound Imaging

Strategies for Hardware Reduction on the Design of Portable Ultrasound Imaging Systems 25

Strategies for Hardware Reduction on the Design of Portable Ultrasound Imaging Systems

http://dx.doi.org/10.5772/55910

267

frame rate point of view. So by using a simple graphics card equipped with NVIDIA CUDA technology, rates that go up to 200 images per second were obtained depending on the image size chosen. Therefore, this solution allows the development of high quality imaging systems

This work has been supported by the Spanish Ministry of Science and Competitiveness under

[1] Caroline Holmes, Bruce W. Drinkwater, and Paul D. Wilcox. Post-processing of the full matrix of ultrasonic transmit–receive array data for non-destructive evaluation. *NDT &*

[2] Caroline Holmes, Bruce W. Drinkwater, and Paul D. Wilcox. Advanced post-processing for scanned ultrasonic arrays: application to defect detection and classification in

[3] Alan J. Hunter, Bruce W. Drinkwater, and Paul D. Wilcox. The wavenumber algorithm for full-matrix imaging using an ultrasonic array. *IEEE transactions on ultrasonics,*

[4] Mustafa Karaman, Pai-Chi Li, and Matthew O'Donnell. Synthetic aperture imaging for small scale systems. *IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency Control*,

[5] Jørgen Arendt Jensen, Svetoslav Ivanov Nikolov, Kim Lø kke Gammelmark, and Morten Hø gholm Pedersen. Synthetic aperture ultrasound imaging. *Ultrasonics*, 44:5–15,

[6] Carlos J. Martín-Arguedas, O. Martínez-Graullera, G. Godoy, and L. Gómez-Ullate. Coarray synthesis based on polynomial decomposition. *IEEE transactions on image*

[7] Carlos J. Martín-Arguedas, D. Romero-Laorden, O. Martínez-Graullera, M. Pérez-Lopez, and L. Gómez-Ullate. An Ultrasonic Imaging System Based on a New SAFT Approach and a GPU Beamformer. *IEEE Transactions on Ultrasonics, Ferroelectrics and Frequency*

with low requirements and excellent capabilities in a compact architecture.

Centro de Acústica Aplicada y Evaluación No Destructiva (CSIC), Madrid, Spain

non-destructive evaluation. *Ultrasonics*, 48(6-7):636–42, November 2008.

*ferroelectrics, and frequency control*, 55(11):2450–62, November 2008.

**Acknowledgments**

**Author details**

**References**

the project DPI2010-19376.

D. Romero-Laorden, J. Villazón-Terrazas, O. Martínez-Graullera and A. Ibáñez

42(3):429–442, May 1995.

*processing*, 19(4):1102–1107, 2010.

*Control*, 59(7):1402–1412, 2012.

December 2006.

*E International*, 38(8):701–711, December 2005.

**Figure 18.** Images per second achieved using GPU for different image sizes for 2R-SAFT and kA-SAFT

pixels, the GPU is able to get 135 images per second, which is in nearly to the acquisition system's rate. The evidence here is that we are using a smaller dataset than that obtained with TFM method but preserving the image quality with all GPU cores completely dedicated to fast computation.
