PORTLAND, Ore. — Compressed sensing has been a laboratory curiosity for several years, but now the technology has been cast in an inexpensive hardware prototype that could enable ultrasound, radar and other data-acquisition applications to increase resolution while reducing costs.
In medical applications, the technique could also reduce the exposure time for patients undergoing MRI, X-ray and CT scans.
"Data acquisition is the bottleneck facing almost all digital applications today; that is the part which is often the most expensive, consumes the most power, takes the longest time and limits the resolution of your results," said professor Yonina Eldar, an electrical engineer at the Israel Institute of Technology (Technion, Haifa). "For instance, one reason ultrasound machines are so massive is the difficulty of obtaining and processing the data.
"Data acquisition is also one of the reasons it takes so long to perform an MRI scan, and it limits the resolution in a variety of military applications, such as radar."
Compressed sensing can let lower-speed hardware acquire data just as accurately as existing systems or can be used to make systems more accurate at the same sampling speed. Defense contractors are working with Technion to increase the resolution of existing radar systems; medical contractors are seeking to downsize their hardware at the same resolution.
The government of Israel, for example, is supporting a collaboration between Eldar and General Electric Israel to bring the resolution of handheld ultrasound scanners to the level achieved by larger, medical laboratory models.
Compressed sensing works like data compression, squeezing data into previously sparse encodings. But it is performed during data acquisition rather than after data is collected, thereby reducing the sampling rate so that even cheap, low-speed analog/digital converters can accurately represent high-frequency signals. Usually, high-frequency signals require A/D converters twice as fast (what's called the Nyquist frequency), but Technion's compressed sensing approach allows the use of A/Ds running up to 10 times slower than the highest frequency in a signal.
"What we wanted to do is make better use of existing A/D converters—not invent new ones—by adding some steps to the process that allow sub-Nyquist sampling rates by processing aliased data," said Eldar.