| A new sophistication is emerging among users of data converters. While not resorting to specifying a custom-made component or mapping intellectual property onto a custom system-on-chip, designers are nonetheless seeking a better fit between the requirements of a particular application and the performance of a particular part. Resolution, speed and power consumption are no longer the only criteria. |
Converter manufacturers thus have redoubled their efforts to tailor their offerings to the finely specified parameters of each end application. The result is an array of devices reflecting each company's philosophy on what matters most, from sheer breadth of offerings to parts targeting specific customers' needs for resolution or speed.
Typically, data converter resolution gets retuned-on manufacturers' data sheets and customer spec sheets-to reflect not just bit resolution but also effective number of bits (ENOB), signal-to-noise ratio (SNR) and spurious-free dynamic range (SFDR). While a high-speed analog-to-digital converter is intended to track a fast-changing signal by digitally sampling its amplitude over time, oversampling techniques such as delta-sigma flatten the amplitude changes, along with the resolution required to capture these changes within specific intervals. It is also possible to use a front-end programmable gain amplifier rather than the data converter itself to make finer gradations in signal amplitude. Thus, for ac signal capture, a 2-bit switched-capacitor device now captures the same audio signal once caught by a full 16-bit A/D converter.
As Kevin Kattmann, Analog Devices Inc.'s product line director for high-speed converters, told EE Times, "If a particular customer wants 9-1/2 ENOB, they'll pay for just that-and not a penny more."
Indeed, ADI (Norwood, Mass.) is in the forefront of companies trying to fit data converter building blocks to the requirements of particular applications without actually building a custom part. Such applications as video signal capture, medical imaging and software radios fine-tune converter requirements beyond the immediate issues of resolution and speed.
Cellular basestations, for example, depend less on raw bit resolution than on SNR (which enables more cellular channels) and SFDR (which allows detection of weak signals in the presence of strong interfering signals). And while set-top boxes require megahertz sampling rates for video signal capture, they also require 10- and 12-bit resolution-even though the human eye cannot discern better than an 8-bit pixel depth for each RGB color. Set-top makers are asking for higher resolution to accommodate overlays and changing broadband communications standards, Kattmann said.
Perceived resolution and speed will often be engineering considerations apart from the data converter's capabilities, such as clocking systems and pc-board signal routing, noted Maher Matta, business manager for high-speed data converters at Maxim Integrated Products (Sunnyvale, Calif.). This is especially true for applications like ultrasound, which uses a phased-array antenna, capturing as many as 256 or 512 separate signals.
In addition to high sampling rates, the converters must provide high channel isolation (up to 110 dB), Matta said.
Ultrasound equipment makers also value SNR performance, for improved image quality, said ADI's Kattmann. But attached to this requirement is an overrange recovery on the input to allow sharpened image quality for surfaces both near and far from the sensor arrays.
The traditional war-still waged by ADI's competitors, including Texas Instruments Inc. (Dallas), Maxim, Linear Technology Corp. (LTC; Milpitas, Calif.) and National Semiconductor Corp. (Santa Clara, Calif.)-juggles trade-offs among amplitude resolution, conversion speed (sampling rate) and power consumption. In the absence of detailed customer requirements, such as high channel counts, data converter manufacturers assume users will gravitate toward both the fastest and the highest-resolution converters they can find, modulated by power consumption and cost.
Though it has been technologically prohibitive to get ultrahigh resolution and ultrahigh speed from the same A/D converter, the envelope defined by those axes keeps expanding outward. And there is always a "lunatic fringe" that will pursue the ultimate in performance no matter what the cost, observed Dave Robertson, product-line director of ADI's high-speed converter group.
On the speed axis, the fastest commercial A/D converters will sample at 1, 1.5 or 2 GHz, but bit resolution is only effectively 8 bits. On the other axis-bit resolution with dc accuracy-successive approximation register (SAR)-based converters extend out to 16 or 18 bits. The resolution of a 16-bit A/D, for example, in one part is 65,535 (1/2N], where N is the ENOB). The sample rate of these parts is on the order of 1 or 2 MHz-several orders of magnitude slower than the fastest samplers.
If delta-sigma devices-DSP-enhanced converters-are included in the mix, bit resolution extends up to 24 bits (1 part in 16,777,216). Sampling rates at this resolution could accommodate audio signals with no dc reference. With dc accuracy, delta-sigma devices like the CS5550 from Cirrus Logic (Austin, Texas) will capture no more than a few ksamples per second-again, orders of magnitude slower.
Texas Instruments is promoting parts that appear to push the envelope in resolution and speed. Its ADS1271 is a dc-accurate delta-sigma converter, with a previously unheard-of sample rate of 105 ksamples/second. It enables users to capture minute variations in signals with 50-kHz bandwidths (such as motor-bearing wobble). The part serves as a "bridge" for those who need to combine dc and ac measurements, said Jim Todsen, TI product line manager for delta-sigma converters.
Earlier, TI introduced one of the industry's highest-resolution SAR converters-the 18-bit ADS8381-which takes 580 ksamples/s. "Both delta-sigma and SAR architectures have a broader spread today," Todsen said. "Both are doing things they haven't done before." Divergent philosophies
TI follows ADI in the data converter market, which generated between $2.1 billion and $2.4 billion in total revenues, according to market trackers like Gartner Dataquest (San Jose, Calif.), iSuppli Corp. (El Segundo, Calif.) and Databeans (Reno, Nev.). Consensus gives ADI roughly 40 percent of the market and Texas Instruments 15 percent; smaller shares go to Maxim, Linear Technology and National Semiconductor.
While all of those manufacturers offer building blocks designed to serve multiple markets, they're pursuing slightly different philosophies in their approaches to customization. ADI can satisfy seemingly specialized customer requirements with the sheer breadth of its product portfolio. Specialized semiconductor processes and packaging, under the umbrellas of iCMOS and iBipolar, for example, satisfy industrial users' needs for high-voltage protection and extended temperature operation.
Linear Technology, in contrast, avoids targeting one customer or narrow market areas but will introduce products to respond to broad market areas with at least five known customers (preferably 10), according to Bob Reay, vice president and general manager of the company's mixed-signal products group.
ADI's competitors similarly will extend their product portfolios, but with careful emphasis on the three most relevant specs: resolution, sampling speed and power consumption. TI is pushing the edges for resolution; Maxim and National are pushing speed thresholds to 1 GHz and beyond. Examples of 8-bit GHz samplers include National's ADC-081000, Maxim's MAX104 and Atmel's AT84AD001B. Atmel (San Jose), in fact, is promoting radar processors-including the TS83102G0B-that sample 10 bits (8 bits ENOB) at 2 GHz.
Linear Technology believes it has a special position along the lower-power axis. Typically, there is a direct relationship between conversion speed and power consumption: The faster you go, the more power you'll consume. "We'll optimize parts for the lowest possible power consumption, then try to find the highest performance we can get at that power level," said LTC product marketing manager Todd Nelson. "We achieve, typically, one-third to one-fourth the power of our competition."
Examples include the LTC1407A, a 3-megasamples/s 14-bit A/D converter that consumes 14 mW typical, and the LTC2255, a 125-Msamples/s 14-bitter that consumes 395 mW.
It is not just handheld portables that require low-power consumption, Nelson explained. The need extends to space-constrained servers and telecom system card cages. As room-sized equipment shrinks to the size of a "dorm refrigerator," Nelson said, board spacing becomes tighter. The converter ICs must take up a smaller footprint, exhibit a lower profile and-most important-dissipate minimal heat.
National Semiconductor's position, articulated by product line director Antonio Visconti, is to satisfy customers' performance requirements (i.e., sampling rate and resolution) first, then concentrate on power consumption. The 1-Gsamples/s ADC08D1000, for example, uses parallel comparators in a "folded" architecture (effectively a two-stage flash converter) to get its high sampling rate. With current technology (cost-reduced 0.35-micron CMOS), there is no way to clock so many parallel comparators so fast and not pay a power penalty. Still, the 1.6 watts attributed to the 8-bit National part is significantly lower than its competitors' draw in that speed range. (The MAX104, for example, is specified at 4.88 W. But that device is a radar processor, observed Maxim's Maher Matta, adding, "Military customers are more concerned with performance than with power consumption.")
While the dual-converter architecture is intended as a general-purpose building block for communications or ATE (automatic test equipment), National's Visconti believes that it will support I-and-Q extraction in digital radios, such as cellular basestations.
But cellular basestation receivers-where practically all high-speed data converter makers now compete-constitute an area where the applications sophistication of the user has begun to displace the mad rush for ever-higher resolution and speed. Here, cellular basestation manufacturers are sampling RF spectrum, said Linear Technology design leader Richard Reay.
The need for sampling speed seems obvious: the higher the sampling rate, the wider the swath of RF spectrum that can be captured. Initially, the market was carved up by manufacturers of 12-bit 65-Msamples/s converters, which would capture (with Nyquist anti-aliasing criteria) a 30-MHz slice of the desired spectra.
While competitors in this part of the data converter market began increasing their sampling rates, basestation manufacturers took advantage of the availability of similar part types from multiple sources. The use of relatively standardized building blocks from a range of sources, including distributors, ensured faster time-to-market than would be available with ASICs; it also ensured competitive pricing, said design leader Reay.
When suppliers began promoting 14-bit parts for the same application, the need for higher resolution seemed obvious: The wider SFDR of the 14-bit parts would allow basestations to capture weaker cell signals in the presence of very strong ones. Put another way, the wider dynamic range (i.e., higher bit resolution) would prevent weak signals-those in the least-significant-bit area-from being drowned out by strong ones.
Consequently, a number of manufacturers are now producing 14-bit data converters with 125-Msamples/s rates. Examples include Linear Technology's LTC2255, ADI's AD9445 and TI's ADS5500.
It would seem that this application would have an almost insatiable thirst for resolution and speed, but that is not the case, said Reay. "Basestation manufacturers are under severe price pressure," he said. "They are now driving costs rather than performance. They are looking at cheaper ways to do what they already do."
In many cases, basestation designs have backed off 14-bit converters and are reverting to 12-bit parts supported by front-end gain-ranging amplifiers-strictly analog components.
This is not to say there is no market for, say, a 16-bit, 125-Msamples/s converter, Richard Reay said. But the users might be manufacturers of communications test equipment, rather than basestations, as previously thought.
See related image
See related image