Eﬃcient Digital Signal Processing in High-Channel-Count High-Frame-Rate 3D Ultrasound Imaging Systems
Ultrasound imaging is an important biomedical technique for analyzing soft tissues in the human body, with both diagnostic and therapeutic applications. IIS is involved in a project developping a high-performance portable 3D ultrasound platform. It will work with linear or 2-dimensional multi-excitor transducers. Portability requires low-power solutions. The goal of this Master project is an ASIC which performs digital processing of the reflected signals received by the piezzo-electric transducers. The basic idea is to delay and add the received pulses in order to collect all energy stemming from one single point. This is called receive beamforming. Per excitation pulse, this can be done for many points, which leads to enormous amounts of data. An intelligent algorithm in the ASIC shall pre-process this data in order to compact it for the transport to the imaging system. A special challenge pose the high-speed output pin drivers. In case of interest in analog design, investigations and a dedicated layout implementation are welcome.
This project is sponsored by UltrasoundToGo Nano-Tera Project
- 30% Theory and Algorithms
- 40-70% ASIC Design
- 0-30% Analog Design (pin drivers)
Recent advances in ASIC integration and low-power technologies have pushed the evolution of a new class of ultrasound imaging systems: Photo-realistic 3D pictures can be acquired from the yet unborn child; the periodic opening and closing of the heart valves can be observed in 3D and real-time; miniaturization allows the physician to carry a small ultrasound imaging device in his pocket which can be used anytime with the same naturalness as the iconic stethoscope. These new systems not only enable new diagnostic possibilities, but also push the limits of the technology and inspire eﬃcient processing solutions.
In this work, two approaches to increase the digital processing eﬃciency have been investigated: First, the compressibility of raw ultrasound sensor data in order to reduce the incidental data rates between the sensor head and the backend system, and second, new beamforming strategies to minimize the computational burden.
Our results show that the sensor data are best compressed by demodulation and decimation down to the real information content. Further, a new beamforming strategy that exploits undersampling of analytic signals has been developed. In addition, a novel beamformer architecture has partially been implemented. It is able to compute almost 270 M focal points per second with an estimated power dissipation of 48 mW per channel.