• State
  • Locked Locked
  • Replies 2 replies
  • Subscribers 29 subscribers
  • Views 633 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

TLV320AIC3104 Channel-to-channel bias/DC offset variation

James Long
Prodigy 150 points
Other Parts Discussed in Thread: TLV320AIC3104

We have a custom board which uses a couple of TLV320AIC3104 Audio CODECs for recording microphones.

Each CODEC has two microphones AC-coupled and connected differentially and to input 1 Left and right 

mic1+ = pin 12

mic1- = pin 13

mic2+ = pin 10

mic2- = pin 11

We have built approximately 50 units so far, they sound great and things are mostly as expected, but on a recent unit, we see a large DC offset in the output of the ADC on one of the CODECs on one board. We are not using the in-built high-pass filter as we filter the signal later, but on all of the units we've built so far, when we disconnect the microphone and record the input we see a very low value - just a few LSBs of noise centred around 0. This is what we expect.

On the recent unit, there is a significant offset of approx 1% full scale, such that the output codes are centred at 300 with a few LSBs noise, rather than 0. This offset is also present at the input of the ADC, where we can measure approx 10mV at DC between the +ve and -ve inputs of the CODEC with nothing connected. We tried removing the AC-coupling caps also, so that there was nothing at all connected, but the result is the same.

We expect that there would be some DC-offset to the output signal as it is expected that the bias point on the +ve and -ve inputs would be slightly different, but 1% / 10mV / 300 LSB seems like quite a lot, esp as we have seen so little elsewhere.

Can you give us an upper bound for the offset we should expect to see?

Does 10mV sound like a reasonable value?

Thanks

James.

  • 0
    TI__Guru 57800 points

    Hello James,

    Is strange that the noise is still there with nothing connected. I suppose that the register setup is the same for all parts, can you confirm this?.

    Built-in High pass filter is intended to be used to remove the remaining DC offset of analog inputs, Could you try verifying if using the filter removes/minimizes the offset?. 

    Thanks and Best Regards,

      -Diego Meléndez López
       Audio Applications Engineer

  • 0 in reply to Diego Melendez
    Prodigy 150 points

    Hi Diego

    I wouldn't call the offset 'noise' as there is no energy above DC, so it's really just an offset and we fully expect a small amount.

    The High-pass filter does indeed remove the offset as expected and our samples are then centred at 0.
    All the register settings we program are programmed identically on each device. Are there any register settings beyond the default ones which affect the input bias voltages? If you could suggest any, then we could check/program these also.

    However, we would still like to know a bound for the difference between the bias points of channels on the same part as a large difference will limit the headroom available for the signal. We've seen the bias point result in an offset of +1% full scale, could it be as high as 10%? That would badly limit the headroom for signals and probably cause loud signals to clip.
    If you can't provide a bound, could you at least provide a distribution measured from factory test?
    So far, our test results show approx 100 CODECs where the DC-offset measures very close to 0LSBs and one which measures at 300 LSBs. That's quite a strange distribution!

    Thanks

    James.