Very good performance is hard enough to achieve, but when you need to squeeze out that last bit of performance (pun intended), you have to look at, and beat down, second- and even third-order error and noise sources. I was reminded of this eternal engineering lesson when I met with Alison Steer, Product Marketing Manager for mixed-signal products at Linear Technology Corp.
She was discussing the LTC2208 family of A/D converters, which emphasizes the large spur free dynamic range (SFDR) needed in digital receivers. These converter ICs address both internal shortcomings, as you would expect, but also external factors. Internally, the design implements a dithering circuit (transparent to the user) which spreads the input signal across a wider portion of the transfer function. This dithering thus reduces the effect of even small, localized integral nonlinearity (INL) errors on low-level signals.
But even if the converter produces a “perfect” result, digitally induced noise and transients, often coupled from the output lines, may corrupt the sensitive analog signals on the board despite careful layout. To reduce this effect, the converter also lets users randomize the digital output lines by XORing them with the output least significant bit (LSB). Of course, the processor or FPGA which receives the randomized data must perform the reverse operation and decode them, but that's a straightforward operation. LTC claims this scheme can reduce residual tone which results from digital feedback by 10 to 15 dB.
My point here is not to single out and praise one vendor's efforts. If you look closely at any truly high-performance system design, you see that such performance is achieved by using the right–but not necessarily the absolute best–components, and with attention to details at all levels. Each source of error or shortcoming has to be investigated, understood, assessed, and overcome, which is achieved with one or a combination of approaches, such as using better components, better layout, clever software, tighter ir more stable calibration, or even inelegant, brute force techniques such as a larger thermal mass, depending on the application priorities. That's the real engineering challenge.
-x x x-