We tend to think of digital signal processing as being computationally demanding, since it typically involves applying math-intensive algorithms to real-time streaming data. And, compared with many other types of embedded processing tasks, digital signal processing does indeed make heavy demands on processors. But that doesn't mean that DSP always requires expensive hardware.
Lately, I've been impressed by the growing array of microcontrollers with digital signal processing features. Often called "digital signal controllers," or "DSCs", these chips combine the attributes of microcontrollers (high integration, on-chip flash memory, low cost) with those of digital signal processors (instruction sets and memory architectures tuned for efficient signal processing algorithm implementation).
DSCs have been around for many years, but lately they've been sporting some impressive specs, including clock rates on the order of 100 MHz and prices around $2. While these chips aren't going to set any DSP performance records, a 100 MHz clock rate coupled with a DSP-oriented instruction set often yields ample performance for signal processing tasks in realms such as sensor signal conditioning, motor control, speech and audio. As a result, it's becoming possible to implement valuable signal processing features in very inexpensive chips. And because DSCs, like their MCU cousins, feature high integration, a low chip price may very well enable a DSC system to be built with a low total component cost.
Beyond performance, integration and price, however, there remain obstacles to widespread use of DSCs. The first is simply awareness. In many MCU applications, product developers aren't experienced in digital signal processing, and haven't yet recognized the ways that their products could be improved through digital signal processing functionality. There's an opportunity here for DSC suppliers to help build awareness—and, in the process, grow their markets. (How about some cool demos, folks?)
Another key obstacle is ease of application development. In DSP-centric applications such as wireless communications, product development teams include engineers who specialize in the efficient implementation of digital signal processing algorithms in software. In other applications, engineering teams are much less likely to have this expertise, so the onus is on the chip supplier to make the DSC's signal processing features readily usable by non-specialists.
DSC floating-point capabilities in are an important part of achieving ease of application development. Most algorithms start out in floating-point form, typically running on a PC. As I’ve written about before, even for signal processing specialists, converting an algorithm that uses floating-point math to a form that uses only fixed-point math can be a real chore. For engineers who are not DSP specialists, it can be a show-stopper. So, DSCs that deliver floating-point DSP capabilities not only speed product development, but also significantly expand the number of product development teams that can incorporate digital signal processing into their products. Although floating-point DSCs are not new, arguably it's only the past year that they've begun to become mainstream.
Floating-point capabilities are important—but in my view, not sufficient. Bearing in mind that most microcontroller users are not digital signal processing specialists, there's a real need for off-the-shelf software to enable effective use of DSCs. This includes low-level building-block function libraries, high-level task libraries (including features like speech compression, for example), and application frameworks. There are already some promising offerings in this direction, but there's much more to be done, and doing it will require a significant and sustained investment by DSC suppliers.
Not so many years ago, digital signal processing was a niche technology, and any product that used it was likely developed by DSP specialists. These days, digital signal processing is well on its way to becoming ubiquitous, and low-cost DSCs are a major enabler of this trend. I believe that the DSC vendors that will be most successful will those that recognize the importance of ease of use, and make significant, persistent investments in improving ease-of-use, both within their chips and through application software development infrastructure.
Jeff Bier is president of BDTI and founder of the Embedded Vision Alliance. Post a comment here or send him your feedback at http://www.BDTI.com/Contact.
Add new comment