The recent article on envelope-tracking RF power amplifiers by Gerard Wimpenny of Nujira appeared in my Inbox on the same day as an article on the “Ideal A/D Converter." And that triggered a train of thought that leads to a pair of New Year's resolutions I’d like to suggest to engineers. Specifically, go learn about the “other stuff.” And find one of “those other guys” and teach him something about what you do.
Lots of technology companies have a large team of digital designers who are trying to squeeze a few million more transistors on a chip and make them go faster while consuming less power. And a smaller team of analog designers, trying to squeeze a little more accuracy out of a few hundred transistors and mediocre passive components. And off in a dark corner of the lab sit the RF/microwave engineers - only a few of them, since the radio is the smallest part of the system, at least in terms of transistor count. These lonely souls are tweaking a handful of devices, including oddly bent bits of metal pretending to be electronic components.
When computing hardware and A/D converters became widely available at reasonable prices, about 1975 or so, it became practical to replace some low-frequency analog functions with digital equivalents. When DSP chips with features like single-cycle multiplication became available in the 80s, more complex functions and higher frequency signals could be processed in software. Hardwired DSP functions like filters, interpolators, and decimators could produce sigma-delta A/D and D/A converters with higher resolution than conventional designs and eliminate complex analog filters. This changed the design of A/D and D/A converters radically. I was actively involved in this transition. Customers wanted us to specify distortion and SNR, not linearity.
In the 90s, the RF design world collided with analog circuit design as the mainstream silicon semiconductor processes reached speeds comparable to then-exotic materials like gallium arsenide. This led to a radical change in the approach to RF circuit design. Early RFICs used as few transistors as possible, because transistors were expensive. A typical microwave LNA/mixer designed in GaAs might have had 10 transistors, and fairly poor performance over temperature. When the analog guys starting doing RF chips, they would routinely throw 20 transistors into each stage for a temperature-compensated bias circuit. And the RF chip performance got a lot better.
I was in the middle of this transition too, and spent a lot of time helping translating the microvolts-and-milliamps and time-domain language of the analog domain into dBm and frequency-domain terms for the RF guys.
Now we’re seeing the next transition…digital processing compensating for analog/RF shortcomings. Amplifier linearization is a hot topic, since the RF signals of today’s systems are not single-carrier constant-envelope waveforms tolerant of amplifier nonlinearity. Rather, they are complex waveforms with high peak-to-average ratios, carrying either multiple carriers or dense quadrature constellations, and any distortion will degrade signal quality or cause adjacent-channel interference. The simple solution - a high-linearity amplifier - generally is a bad choice since it would be very inefficient and waste power most of the time.
Techniques like digital predistortion and envelope-tracking draw from all three domains…digital, analog and RF, and allow fundamentally non-linear (but power-efficient) amplifiers to behave like linear amplifiers.
So go introduce yourself to one of those “other guys”, and cook up the next problem that can be solved by a tasteful merger of the three domains. One of those guys might have a simple solution to something that’s your biggest problem!
hm, agree with you that the distant relative will eventually be acknowledge and use in more areas that already are. However, these distant relatives STILL NEED TO INTERACT WITH their Analog counterparts to do any common-good. ;-)
Blog Doing Math in FPGAs Tom Burke 14 comments For a recent project, I explored doing "real" (that is, non-integer) math on a Spartan 3 FPGA. FPGAs, by their nature, do integer math. That is, there's no floating-point ...