A chief limiting factor of analog-to-digital converter (ADC) resolution is noise, while the digital section of a design can be a chief contributor of noise. This antagonism reminds me of the westerns where the cattlemen and the sheepherders are fighting it out.
I like the scene where some cowboys grab a lone sheepherder, wrap him in barbwire, and drag what’s left of him back to his wailing mother. A young boy asks the hired hero why cattlemen and shepherds just can’t live in peace. The hero gives some stoically sensitive answer that justifies the serious carnage he is about to reign down on the cattle baron’s men.
There comes a similar point when increasing ADC resolution doesn’t easily coexist with the digital circuitry. This point is generally at 12 bits. (I suppose a guy with 20 cows has no problem with his neighbor’s five sheep. It’s when you have a cattle baron with 10,000 cows and a whole community of immigrant sheepherders that you get a movie.) Can analog and digital live together? Yes, if some basic rules are followed.
ON THE BENCH
A customer sent me the schematic in Figure 1. It includes two major functions: reading an analog sensor with a 1.024-V range 16-bit ADC, and controlling the intensity of an LED with a pulse-width modulator (PWM). The customer had called to complain about our ADC’s poor performance. He was laying out his boards in the same way he had when he was designing with his old (10-bit) ADC.
After asking him a few questions, I determined that the problem was the digital signal return path and finite trace resistance interfering with the analog circuitry. The digitized signal was nowhere near 16 bits. With 32 times the sensitivity, layout won’t be as simple. I made some recommendations, and the new ADC performed with the resolution expected of it.
First, the resolution of the ADC is 1.024 V/216 or an LSB of 15.6 µV. The board is two layers. Four layers with a ground plane would have been nice, but a two-layer board can be used to save cost. The analog and digital grounds are tied together.
The designer said he did this because his company couldn’t afford separate power supplies, so there was no need to separate the grounds. This is incorrect. Any noise at a power input is reduced by the analog power-supply rejection ratio (PSRR). However, any noise that gets into the ground directly affects the measurement.
In this case, the 4-mA LED pulse wave returns through the digital ground pin to the system ground. There is a finite resistance to the trace. The trace is 8 mils wide by 0.25 in. long and is made from 0.5-oz (or 0.7 mil thick) copper. Traditionally, circuit boards have been made with 1-oz copper. But as power requirements got smaller, 0.5 oz has become more common. A 0.008- by 0.25-in. trace of 0.5-oz copper is 31.25 squares for a resistance of 31.25 mΩ.
The 4-mA LED current returns through the digital ground pin causing a 125-µV or 8-LSB voltage drop across the trace. The analog ground for the ADC, tied directly to the digital ground pin, is now 125 µV different from the sensor ground.
If this trace is increased to 32 mils wide and the board is fabricated with 1-oz copper, the round trace resistance is lowered to 3.9 mΩ for a trace voltage of 31 µV or 2 LSB—better, but still not good enough. It is going to be necessary to separate the grounds.
HEADING FOR MECCA
There should be an analog ground path and digital ground path, and they should be connected at a single point. Back in the days before multi-layer boards, we referred to that point where all ground returned to as “Mecca.” This is the point where shields were also connected.
It would be preferable to have separate analog and digital power supplies. With a typical PSRR of 90 dB, 100 mV of noise on the digital supply is 3 µV of noise at the ADC input. If dual supplies aren’t an option, then treat the supply as if it were two. This means giving both the analog power pin and digital power pin their own bypass.
For the digital supply, a capacitor bypass is adequate. For the analog supply, you might consider an R-C or L-C bypass. Also, instead of sinking the LED current, it should be sourced to digital ground. This keeps the current path out of the chip ground pin. Lastly, the way to read the voltage at the sensor relative to its ground is to measure it differentially. When selecting an ADC, it is good to find one that can be used with either a single input or differential inputs.
Figure 2 illustrates these changes, which are minor in terms of cost. These changes require an additional resistor and capacitor to bypass the analog supply. The digital and grounds are separated and brought together in a star formation. The board is fabricated with 1-oz copper-clad two-layer circuit board material with the ground trace made as wide as possible.
Ideally, making flooded ground planes helps reduce the resistance back to Mecca. In addition, you might as well keep just as much copper on the board. You paid for the whole layer and you got no refund for the amount that dissolved away during fabrication.
I received an e-mail from reader Ed Wetherhold about capacitors. In a previous column, I gave an example about rules of thumb where I talked about NPO (en pee oh) capacitors (see “Don’t Let Fear Of Risk Dictate Your ADC Selection”). Ed wrote to correct me that it is in fact NP0 (en pee zero). We called them COGs (see oh gee) 30 years ago, and I sometimes get them confused.
I did a quick survey among the engineers here at Cypress, and most of them got it wrong too—the application engineers, that is. The chip designers didn’t even know what I was talking about.
People often say “oh” instead of zero when they’re saying a number. I guess they say it wrong because it rolls off the tongue more easily. I blame the phone company for deciding to use zero to dial the operator. So, good catch, Ed!