Advertisement

Blog

Data Converters in Massively Parallel Analog Systems

In an earlier blog, The Race From Power Management to Data Converters , I wrote how power management is slowly becoming commoditized, and as a consequence, multiple high-performance analog semiconductor manufacturers are moving to other areas with higher profit margins. One of those key areas is data converters, and more specifically, integrated circuit (IC) systems with analog-to-digital converters (ADCs). Medical imaging, especially ultrasound imaging, is a huge consumer of IC data converter technology.

For a few years now, a pregnant mother has been able to take her family to a local shopping mall and “meet” the new baby long before birth. The quality of imagery possible in that non-medical setting is astounding to me (see Figure 1). This has all become possible given the increasing levels of analog integration now found in these systems.

Figure 1

Ultrasound works on the basic premise of picking up en masse reflected ultrasonic signals that are corrupted by multiple artifacts and then summing them together through complex mathematical processing to construct an image. It is not unreasonable to see system requirements of 160dB dynamic range at multi-megabit sampling rates to affect these imaging results.

In modern ultrasound systems, the number of transducer channels can exceed 1,000. These transducer channels, the sources and sinks of the ultrasonic waves, are usually multiplexed down to tens or hundreds of electronic signal-processing channels. Still a very large electronic design problem custom tailored for analog integration. Probably the most sophisticated consumer use of analog signal processing bar none. Military-class radar in a box, if you will.

To put 160dB into perspective, in a 5V system this would be roughly equivalent to a noise floor of 50nV. While this by itself would be a heroic design challenge, when the required system bandwidth of multi-megahertz is factored in, the heroics become impractical. This is not the domain of audio-quality delta sigma converters with 100,000 samples per second at 110 dB SNR.

While there are multiple modes of operation in an ultrasound system, one mode problem is resolved using a massively parallel analog signal-processing system broken down into stages of low noise and variable-gain amplification followed by filtering and then analog-to digital-conversion. In this way the instantaneous dynamic range (DR) requirements of the ADC is reduced and economically practical systems become possible. Originally, systems focused on 10-bit converters (60dB instantaneous DR), but more recently, 12- and 14-bits are routinely designed into these devices. The trend is always higher.

Reducing the instantaneous dynamic range of an ADC makes the design much easier (and cheaper) since dynamic range is closely intertwined with speed and power. For portable systems, power is a huge issue. A high transducer count system might consume as much power as a high-performance desktop PC. I guess that's why portable in a hospital means it comes along with a cart on wheels.

I make it a point to regularly track the progress in data converter technology. Awhile ago, I ran across an excel database put together by Dr. Boris Murman of Stanford University in Northern California (Figure 2) that easily summarizes the historical progress over the past few years. It is interesting to see in his chart how the fundamental tradeoffs of speed, power, and dynamic range in ADCs form an implicit performance barrier shown as the dotted line.

Figure 2

ADC performance 
Source: B. Murmann, 'ADC Performance Survey 1997-2013' 
http://www.stanford.edu/~murmann/adcsurvey.html

ADC performance
Source: B. Murmann, “ADC Performance Survey 1997-2013”
http://www.stanford.edu/~murmann/adcsurvey.html

The x-axis in his chart is speed of a converter (i.e., bandwidth, BW) whereas the y-axis represents a computed figure of merit (FOM) that shows how the three hallmarks of an ADC — speed, power, and DR — are related. I transformed Dr. Murman's FOM equation into the one below to more easily show the relationships.

I think most analog designers know that precision drops as bandwidth rises so the flat section of the dotted line makes sense. And then at some point the speed requirements are so high that power is the only thing that helps — and even then, only so much. This, in part, explains the falloff at the curve's knee. Something tells me that Dr. Murman's dotted line is probably a contour of constant price also. Maybe someone wants to check this out?

Several manufacturers have multi-channel analog front-end ICs for ultrasound systems. The most recent introductions are eight channels. That means several of these are required in a single system. I suspect the future will see more and more of these types of high-dollar products containing en masse precision amplifiers with en masse precision data converters. Power is probably a key area to focus on as the Internet of Things evolves.

Related posts:

3 comments on “Data Converters in Massively Parallel Analog Systems

  1. eafpres
    April 15, 2013

    Hi Scott–the numbers certainly are daunting.  I recall seeing some articles over the last year about miniaturization of the probe–the part that touches the body and transmits the ultrasound waves and receives them back.  I think some of this development is to allow ultrasound imaging to be used in conjunction with laproscopic procedures and other minimally invasive procedures.  I can't lay my hands on an article but I wondered after reading your article if improvements in the probe end are one factor which works to reduce the power and dynamic range requirements on the electronic side?

  2. DEREK.KOONCE
    April 15, 2013

    The chart and slope is interesting. Yet you have two general sets of data 2012 and “1997 to 2011”. If you break down the year grouping into something smaller and have several groups, you might also see that the level and slope, in my guess, would be rising and pushing out to the right as technology develops.

    Shrinking of the transducer head and moving electronics into the head will also show performance improvement.

  3. Brad Albing
    April 30, 2013

    Not sure if this is an area about which Scott has knowledge, but we can poke him and see.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.