RC filter on MCP4921 DAC Vref

Other Hardware General Electronics
1

Hey guys, I've got a question about an RC filter on the VRef of the same DAC used on Adafruit's Wave Shield.

Here is the datasheet for the DAC in question:

And here is the schematic for the Wave Shield:

You can see they've placed a 16Hz LPF comprised of a 100K resistor and .1uF capacitor on the DACs VRef/REFA pin.

I used this same DAC in a previous design and ended up with a lot of noise from my servos and LED drivers getting into my audio system, so in my next design in addition to including more capacitors to supply power to those I've also been double checking the audio system. I'm not looking for decoupling tips for the LED drivers in this thread, I just want to understand how this filter functions in conjunction with the chip.

Anyway, the reason for my question is I noticed that with this DAC you can enable a "buffered" VRef mode. The data sheet wasn't very clear on what that is, but I'd hoped enabling it would reduce the noise somewhat. Doing so appears to have done nothing however.

The datasheet seems to indicate that buffered mode greatly increases the input impedance on VRef, but it also indicates the voltage reference then needs to be slightly below Vdd. The input to Vref was 5V I assumed. So I'm not sure if enabling buffered mode would be bad for the chip or lead to clipped audio, but again, I tried it on my old circuit and it seemed to function the same as before.

I then did some googling and found this thread on the Arduino forum:
http://forum.arduino.cc/index.php/topic,199374.0.html

Someone there suggested that the 100K resistor was a bad choice and that the 100K resistor in the LPF formed a voltage divider with the input impedance of the VRef pin. In unbuffered mode, that is 165K. In buffered mode, the datasheet doesn't appear to say, other than that it is much higher.

If that forms a voltage divider however I would have assumed my audio volume level would be affected. Passing 4096 to the DAC if Vref is 2.5V after all, should result in a 2.5V output. That doesn't seem to be the case though. Because again, I attempted to enable that buffer, and it did nothing. If the buffer's impedance is far higher than the unbuffered mode, as the datasheet indicates, then I would have expected Vref to be much closer to 5V, and for my audio to therefore be twice as loud. But like I said, in enabling it, I heard no change.

Since I'm redesigning the circuit, I could reduce that resistor, or stick a 1uF cap on there to reduce the LPF to 1.6Hz instead of 16Hz as it is now, and I assume that would reduce the noise further. I could also stick two capacitors and two resistors on there to make a second order filter. But the question remains whether I should even bother. I could leave some pads there and just not populate them I suppose. But if this isn't going to actually reduce the noise in the audio by any perceptible amount, then I'd rather not go overboard and take up board space with additional unnecessary pads.

Any insight into what's going on here would be appreciated. I really need to wrap this design up this week.

2

So long as there is good decoupling on Vref you'll be mainly dealing with grounding
issues - you should be using a star-ground, preferably with the common point under
the DAC chip itself (that's what chip vendors all recommend).

In practice you should have a separate PSU for the analog, floating w.r.t. the
digital, and just connect the grounds together at or near the common point
(ADC or DAC). This keeps all the ground and supply currents separate in the analog
and digital world.

Failing that using RFCs and ferrite beads to isolate the analog circuitry at
high frequencies is usually used - both for supply and ground. I think the way
to do this is just before the linear regulator that supplies the analog rail. Again
a common point under the chip, possibly with ferrite bead is.

Never use a DC-DC converter for analog supply if you want noise-free audio, BTW!

For the MCP492x there's only one supply, this must go to the analog supply,
not digital, note. However its only 12bit so its less critical than with 24bit sigma/delta
chips.

3

Regarding star grounding, I have been reading a bunch of stuff about ground planes the last few days because in my last design I tried having a single point connection to my analog ground plane, and noise still got in, and I've found a number of recent documents stating that a solid ground plane is likely best, and that the layout is more important. Ie, make sure the path for power to your noisy stuff doesn't cut through your analog circuitry.

I lost the links to the papers, but found some discussion of it here:

One fellow towards the end recommends a solid ground plane, but a split power plane. Perhaps I could try that. Seems non-intuitive though to split one but not the other.

4

Never use a DC-DC converter for analog supply if you want noise-free audio, BTW!

Do you mean a switching converter? Because Wikipedia says a linear regulator is a DC-DC converter, and I have to use one or the other.

In this circuit's case, I have a 5V supply from a wall adapter. I was going to put a 100uF cap at the power input and .1uF decoupling caps around the board, and 1uF where specified in datasheets. My power hungry components are right next to the 100uF capacitor.

I could, if necessary use a linear regulator to drop the voltage for the Audio portion of the circuit to 4.5V. The amp is a bit louder than needed, and I would imagine dropping half a volt would be sufficient to smooth whatever noise there ends up being. Unless of course the noise is a result of the ground being noisy. I have room on the PCB for a regulator, it's just cost that is the issue.

5

So I'm still not clear on what's going on with this pin. As I mentioned above, I've been told there is a voltage divider being formed, yet if I am interpreting the data sheet correctly, enabling buffered mode should force the pin impedance to be much higher, which I would assume should make the voltage divider affect the voltage much less, and thus, make the voltage on the output of my DAC higher, and thus increase the volume of my audio... at the very least.

Yet that did not happen when I enabled the buffer. So either I am misinterpreting what the data sheet has to say about buffered mode, or there isn't a voltage divider.

6

Well, I (probably) figured out why nothing happened when I enabled the buffer on the DAC. I'd been expecting a volume change, if in fact the 100K resistor formed a voltage divider with the pin's inductance.

Turns out I must have read that other post two years ago, and taken the advice given in it when I designed my first board, which is the one I tested the buffer on, because when I looked at the schematic to see which resistor I should bypass to test no resistance on the Vref pin to see if that had effect, I saw that I had changed the resistor to 1K and the capacitor with 10uF to keep the 16hz LPF, but avoid most of the voltage drop. The 100K would have dropped 2V, while the 1K should drop only 0.03V.

So that would explain why enabling the buffer, which is supposed to greatly increase the inductance of the pin, and should have reduced the voltage drop a great deal, had no effect on the volume. There's no way I could tell by ear if the volume was a mere 3% louder.