3
\$\begingroup\$

I need to get the higest possible precision from an analog to digital convertor (and by high I mean 24bits or more), 10 times per second. To this end I've decided to investigate using an ADC that is designed for audio. Unfortunately, after vast amounts of reading and research the only way that I can interface it with my microcontroller, which I can't connect directly to it because of timing concerns. The only way that I can interface it is with an FPGA, which is alot of work for such a seemly simple task. I refuse to believe that nobody has wanted 24bit ADC precision on a simple non DSP microcontroller such as the Adurino or Fez.

Does anyone know of an IC that can interface the I2S bus to either SPI, I2C, CAN, Serial, etc for the purpose of obtaining the 24bit value of the analog signal. Note: The data is not being using for high speed Digital Signal Processing.

\$\endgroup\$
  • 4
    \$\begingroup\$ Many audio ADCs filter out DC. Read the datasheet carefully. \$\endgroup\$ – markrages May 2 '11 at 6:44
  • 4
    \$\begingroup\$ Remember, getting anything close to 20+ bits, let alone 31 requires extreme attention to noise issues. To manage something like 22+ bits, the whole system pretty much will have to be isolated, and powered by a custom-designed low-noise DC-DC power supply. Grounding will be extremely critical, and a 4 layer PCB is almost absolutely necessary. Remember, 1 bit from a 24 bit converter w/4.096v reference is 244 nanovolts. Even with excellent PSRR in the ADC, your rails have to be dead-quiet. \$\endgroup\$ – Connor Wolf May 2 '11 at 7:38
  • \$\begingroup\$ @markrages The device that I am investigating has the option to leave the DC component in :) I checked that at least. @FakeName won't a regular regulated supply work? I understand about the nanovolt precision, and I know that I may not need 24 bits, but I know that 16 is not enough. \$\endgroup\$ – user3045 May 2 '11 at 16:38
2
\$\begingroup\$

If you only need 10 Hz sampling then an audio ADC seems like overkill. I would bet most Audio ADCs that are 24 bit would be at least 48 or 96 KHz.

Instead I would recommend going with something like this that provides you with SPI like you want and has a selectable sample rate between 6.25 SPs to 3840 SPs.

I am also not sure why a microcontroller would be unable to handle reading that data.

\$\endgroup\$
  • \$\begingroup\$ The complications that would arise are spoken about in the post by ajs410 below. \$\endgroup\$ – user3045 May 2 '11 at 17:04
4
\$\begingroup\$

This page on the TI website lists a dozen 24 bit ADCs interfacing with SPI. Speeds start at 15 sps.

A thought on resolution:
24 bits or more! You do realize that 20 bit offers 1 ppm resolution, and that 1 LSB in 24 bit is 0.06 ppm of FS? Just looking at the circuit will give you several LSB of error, so to speak. For one thing, you'll need temperature control to within tenths of a degree to take advantage of the full measurement range.
If you really need more than 24 bits, TI has the ADS1281 which offers 31 bits.

\$\endgroup\$
  • \$\begingroup\$ Hey temp control, didn't think about that. Thanks for the advice! \$\endgroup\$ – user3045 May 2 '11 at 16:40
4
\$\begingroup\$

You could use a CPLD instead of an FPGA in order to get a more cost effective digital hardware solution. However, there are a number of reasons you shouldn't use an audio ADC for certain applications.

Audio ADCs tend to be sigma-delta architecture. You can't just sample them whenever you like by asserting some input. You are required to pump the data from them at a constant rate, and there's only a limited number of specific frequencies that they can use for operation. In addition, if you need to phase-lock onto the signal, that will be almost impossible without careful manipulation of the ADC's oscillator, and such manipulation is almost certainly going to cause artifacts that could be detected at 24-bits, and that's assuming that you aren't overwhelmed with all kinds of other noise already.

Another problem with audio ADCs is that they make a lot of assumptions based on the typical audio frequency range of the human ear. These assumptions may interfere with your sampling, particularly if it is outside of typical audio frequency range.

\$\endgroup\$
  • \$\begingroup\$ :( Didn't know that either, and it makes sense. Wish I could mark all as correct. \$\endgroup\$ – user3045 May 2 '11 at 16:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy