“This city is headed for a disaster of biblical proportions, you know, the real Wrath-of-God type stuff. Human sacrifice, dogs and cats living together - mass hysteria.”
Greetings from Baltimore, home of one of the prettiest harbors on the East Coast, one of the nation’s most disappointing baseball teams and, at least for this week, Dr Lance Leventhal’s
Traveling Embedded Difference Engine Exposition and Medicine Show, otherwise known as the
MicroTCA Summit. I don’t take the sudden coziness between DSPs and FPGAs I’ve seen here as an omen signaling the end of the world but, if the buzz at this event is any indicator, Micro-TCA could be the catalyst that legitimizes this mixed-marriage and moves it into the mainstream.
It’s not like signal processing is anything new to FPGAs; most high-end products have been sporting hardware multipliers and other dedicated DSP elements for the last couple of years. But the relatively tight power budgets (65 W suggested) and even-tighter space constraints (180.6 x 73.5 mm) of the MicroTCA advanced mezzanine card (
AMC) means that designers will have to make extremely efficient use of PCB real estate to deliver the processing power required for applications like WiMAX base stations, cellular pico-stations, medical imaging equipment, and mil-spec software-defined radios that seem to comprise so much of the early applications I saw at the show. Instead of having a pair of 20 W - 30 W DSPs aboard an AMC, many designers are discovering that they get much more bang for their board space by using a couple of dozen (or more) specialized offload engines in a 20 W - 30 W FPGA to do most of the repetitive, well-defined tasks and using a single DSP to handle the rest. Another less obvious example of FPGAs co-habiting with DSPs was Actel’s FPGA-based MicroTCA chassis management solution that makes monitoring critical voltages, temperatures and status states easy and space-efficient. With one of these devices handling the management element tasks on your card, you’ll have more space and time to devote to signal processing.
Some more aggressive types (usually the FPGA vendors) have even had the temerity to question the need for any sort of dedicated DSP, claiming that it’s even more efficient to let programmable logic handle all the signal processing tasks. It will be a few years before this territorial debate is settled but, in the meanwhile, FPGAs are proving themselves to be valuable companions to DSPs in other ways, such as providing traffic management, bus protocol bridging, and robust SerDes connectivity across the MicroTCA backplane. Altera shared some pretty impressive test results for carrying 5 Gbit/s PCIExpress 2.0 signals across up to 14 inches of backplane with a 50% eye margin. While supporting these speeds required use of some exotic dielectrics, and back-drilled vias in the PCB, it was evident that you’d have lots of performance margin for 2.5 Gbit/s PCIe with a standard backplane.
If the low cost and versatility of MicroTCA makes it anywhere near as popular as the pundits at the show claim it will be, the AMC card’s high volumes and low component counts will be an ideal vehicle for low-cost hybrid DSP/FPGA development systems. The cards I saw at the show from the likes of
BittWare looked like a great micro-sandbox for the inquiring engineer to play in. I’m pretty sure we’ll see lots more developments like this at next year’s show as this odd, occasionally strained, but very fruitful relationship between hardwired and configurable signal processing elements continues to mature.
Comments? Questions? Interesting applications for FPGA-based DSPs? Write me at LHG at en-genius.net.
