Digital signal processing systems have long enabled interaction of the digital electronics world with the real, analog world. It all started with simple devices that performed sums of products in a single instruction cycle. In 1979, Intel Corp. introduced the 25bit I2920, an integer processor, and NEC came out with uPD7720, a 16bit processor with a 512-word program memory and a 128+512 data memory. By 1980, Texas Instruments had offered the TMS32010 and Fujitsu, the MB8764—both of them, 16bit programmable ICs.

The DSP in those days found wide application in speech synthesis and recognition, echo cancellation, modems, disk drives and instrumentation. Now, digital signal processing is found in a much wider variety of applications—consumer electronics, computers and peripherals, cellular handsets, wired and wireless communications infrastructure equipment, industrial control and measurement, as well as car electronics. Not surprisingly, application requirements have changed.

A device used for digital signal processing now needs to be judged for power consumption, in-application flexibility, cost and development time. ASICs are typically power-optimized and cost-effective in high sustained volumes, but offer the least flexible solution. Reconfigurable FPGAs offer high flexibility, but it may take longer to develop algorithms for an FPGA than for a DSP.

Several types of devices and technologies have recently emerged to address different requirements, including those posed by next-generation wideband communication standards like 3G and WiMAX. picoChip's picoArray combines hundreds of array elements, each with a processor and its local data and program memory, connected by high-speed interconnect. The array elements can be completely re-programmed to work together as a group.

Another company, Intrinsity Inc., makes use of dynamic logic, a design methodology made nearly obsolete in the 1980s. Dynamic logic uses a clock signal to evaluate combinational logic. Processors developed using dynamic logic are substantively faster and more efficient while those developed using static logic can be quickly and cheaply designed and manufactured. Intrinsity claims to have lowered the cost of developing dynamic-logic-based products to a point where performance justifies the additional expense.

A company called Stretch Inc., on the other hand, integrates its proprietary Instruction Set Extension Fabric (ISEF) configurable hardware inside a Tensilica processor core. By embedding the logic inside the processor architecture, the logic can be programmed using a conventional, software environment.

Meanwhile, traditional processors have also made performance advancements—Intel's Xeon processors were used in the recent field trials of the software-defined GSM base station by Vanu Inc., a company selling an FCC-approved software-defined radio.

While each approach has its benefits, it is clear that both traditional DSP companies and FPGA vendors will soon be gearing up their product road maps to new competition. And as companies battle to find an application niche for their processor architecture, the Embedded Systems Conference-Taiwan (ESC-Taiwan) to be held July 27-28 in Taipei will host the forum titled, "The Future of Processors for Signal Processing." MIPS Technologies will propose adding DSP functionality to the RISC architecture, while Xilinx Inc. will talk about using an FPGA parallel architecture to boost DSP performance. Broadcom Corp., on the other hand, believes that multicore processors are the answer to DSP needs, especially in communications.

I invite you to participate and help both vendors and users define the processor architecture for the next-generation of applications.

Email the Editor

Back to top