ENOB IS ENOUGH

The front page of the QF4A512 datasheet touts a 16-bit ADC, but the fine print says it’s actually only 12 bits. What gives? Is this another example of market speak run amok?

This is an opportune moment to bring the issue of ADC specsmanship into the spotlight. The fact is, using resolution as the single figure of merit for an ADC isn’t only prone to marketing manipulation, but has little to do with what actually happens when the electrons start flowing.

A timely, if unfortunate, analogy is the way the accounting records at a cook-the-books enterprise are kept to the penny, even as they hide megabucks of mischief. A wrong answer can be presented with great precision, but it’s still wrong.

The truth about ADCs can only be found in the myriad of specifications that make up the fine print (INL, DNL, DRIFT, SNR, SINAD, SFDR, and on and on). However, different manufacturers use different specifications, and they even interpret the same specifications differently, making comparing chips an apples-to-oranges exercise. To make a long story short, no N-bit ADC ever delivers the accuracy N bits of resolution implies.

The problem starts with the ADC itself, which isn’t perfect. The conversion process itself inevitably introduces noise and distortion not found in the input signal. And that’s in addition to errors introduced before the signal ever hits the ADC, such as electrical interference or analog front-end glitches (e.g., gain and offset error).

When contemplating the capabilities of a particular ADC relative to your application needs, one specification you might find useful is ENOB, which stands for Effective Number Of Bits. The specification derates the marketing number of bits, or MNOB (i.e., resolution), by recognizing the effects of noise and distortion introduced by the front-end and converter itself. In effect, the MNOB describes how many bits of resolution a particular ADC would have if it (and the analog front-end) were perfect (i.e., no noise or distortion), while ENOB takes those problems into account.

So is the QF4A512 12 or 16 bits? The answer is arguably both and, in light of ENOB, everything in between. As stated earlier, under the hood is a 12-bit converter. The chip earns 16 bits bragging rights by oversampling, exploiting the fact that the converter is a super-fast pipelined design running at up to 75 MHz. Each factor of four oversampling is equivalent to adding a bit of resolution, and the chip supports up to 256× oversampling. That explains where the extra 4 bits of MNOB resolution (from 12 bits in the hardware to 16 bits on the datasheet) come from. 

Meanwhile, calculating ENOB is easy because it’s a simple function of the signal-to-noise-and-distortion ratio (SINAD).

Plugging in the specification for SINAD from the QF4A512 datasheet, you find that the ENOB is nominally about 13 bits, but it depends on the sample rate, which is limited by the over-sampling ratio (see Table 1). Check out the datasheets for real 16-bit ADCs, even much slower ones, and you’ll find that an ENOB of 13 to 14 bits is par for the course, arguably justifying the Quickfilter chip’s 16-bit aspirations.

Don’t feel bad. Remind yourself that 13.2 bits of ENOB translates to about 0.01% precision, which is just fine for plenty of applications.