The April 2024 issue of IEEE Spectrum is here!

Close bar

New Class of Digital Signal Processor Wipes Out Wasted Power

Hearing aids, power converters, medical implants, and telecommunications could benefit from continuous-time digital signal processing

4 min read

Samuel K. Moore is IEEE Spectrum’s semiconductor editor.

5 November 2008—Digital signal processors—those practically ubiquitous circuits that make cellphone conversations understandable and MP3 players possible—come in a great many varieties, but until recently, there was one variety no one had even thought to make. Called continuous-time digital signal processing (CT DSP), it has the ability, unique among such circuits, to consume dynamic power in proportion to the intensity of the signal it processes. When there’s no signal, such as during the silent spots of a cellphone conversation, the processor is practically inactive. But when the signal appears, it kicks into gear. Its inventor, Columbia University electrical engineering professor Yannis Tsividis, says the device’s miserly management of power could make it attractive for small systems—such as biomedical implants and remote sensors—that deal with ”bursty” signals in need of real-time processing. Industrial firms are also interested in using the technology for telecommunications and power conversion.

”It’s a different way of looking at a problem that people have looked at for a long time,” says Rajit Manohar, an expert on clockless circuits and professor of electrical engineering at Cornell University. ”There are a lot of benefits to the approach.”

A signal can be either analog or digital, and the system that processes it can do so either continuously or in discrete time, that is, sample the signal at regular intervals. But digital signal processing has always been practically synonymous with discrete time. ”This has been bothering me for many years,” says Tsividis, an IEEE Fellow.

In conventional discrete-time DSPs, the clock signal that triggers the sampling must have a frequency of at least twice the highest frequency of interest in a signal. But the clock has to go on nonstop at that frequency whether or not there’s a signal and whether or not that signal has any high-frequency component to it at a given time, which is a waste of power.

A CT DSP, by contrast, has no clock. Instead, when the signal changes by a set amount, the CT DSP produces a digital pulse. If it changes by that amount again, a second pulse occurs, and so on. A separate digital signal marks whether the input is increasing or decreasing. Those digital pulses are put through a programmable circuit that digitally filters the signal in continuous time and then converts it back to analog.

When there is no input signal, there is also no digital output, and little energy is expended. And when the input signal changes quickly—rising or falling steeply—the system produces digital pulses at a higher frequency. So the circuit automatically adjusts its sample rate to match the needs of the input signal. ”The signal samples itself,” says Tsividis.

Aside from the potential power advantage of a CT DSP, it just might make things sound better. Conventional DSPs suffer from two types of spurious signals: one, called aliasing, happens because the input signal mixes with the clock frequency. With no clock, CT DSPs are free of aliasing. The other, quantization error, is produced by the inexactness of turning the analog input into a digitized signal. In a conventional DSP, quantization error is spread over all frequencies, but for a CT DSP, it occurs only at multiples of the frequencies in the signal, so it’s easier to filter out and there’s less of it to begin with. Also, such ”harmonic error” is less objectionable to the ear.

The one downside, says Tsividis, is that CT DSPs do not allow you to store the processed data in digital memory, so they are useful mainly in applications where that’s not needed. (Of course, you could always sample the CT DSP’s output using a conventional analog-to-digital converter and store the data digitally, but that would introduce a clocked signal and draw more power.)

Tsividis first developed the theory for CT DSPs in the summer of 2003. Rather than have a graduate student work on it, Tsividis took it on himself ”in case it was just a crazy theoretical idea.” In two weeks, he’d built a low-fidelity (4-bit) version on a breadboard. He plugged in the output of a CD player, and it worked, sounding much better than the awfulness produced by a conventional 4-bit DSP. After getting those initial results, he worked with fellow Columbia electrical engineering professor and IEEE Fellow Ken Shepard and Shepard’s student Yee William Li to produce an integrated prototype. Tsividis then further developed the technology with his own graduate student Bob Schell, who designed a high-performance 8-bit chip. Schell and Tsividis describe the chip in this month’s IEEE Journal of Solid-State Circuits.

Tsividis says the CT DSP has caught the attention of industry more quickly than he expected. Following the chip’s debut at last February’s IEEE International Solid-State Circuits Conference, he heard from companies in the digital music synthesis field, who are interested in the processor because it has no aliasing. His group at Columbia has also begun a collaboration on a telecommunications application with an industrial firm he would not name. Separately, Dieter Brückmann at the University of Wuppertal, in Germany, is working on a CT DSP version of a wave digital filter, a class of filters found in Bluetooth radios and GSM phones, among other devices.

CT DSPs are creating a lot of buzz in the power electronics industry as well, says University of Toronto electrical engineering professor Aleksandar Prodic. As soon as he saw a presentation of the concept by Tsividis in 2006, he knew CT DSP could make for better, cheaper dc-dc converters—crucial circuits that maintain a steady voltage supply to chips in cellphones and computers even in the face of noisy inputs and sudden changes in the amount of current that must be delivered. He and his student Zhenyu Zhao built a CT DSP�based converter that responds to sudden changes in power demand with ”virtually the fastest recovery time possible.” The CT DSP could lead to both size and cost reductions for dc-dc converters, too, says Prodic. Its response is so fast that it should allow for the use of smaller, cheaper capacitors, which today are the largest components in converters.

This article is for IEEE members only. Join IEEE to access our full archive.

Join the world’s largest professional organization devoted to engineering and applied sciences and get access to all of Spectrum’s articles, podcasts, and special reports. Learn more →

If you're already an IEEE member, please sign in to continue reading.

Membership includes:

  • Get unlimited access to IEEE Spectrum content
  • Follow your favorite topics to create a personalized feed of IEEE Spectrum content
  • Save Spectrum articles to read later
  • Network with other technology professionals
  • Establish a professional profile
  • Create a group to share and collaborate on projects
  • Discover IEEE events and activities
  • Join and participate in discussions