AnsweredAssumed Answered

ADAU1761 unexpected attenuation in pass band (9.5 dB vs. input)

Question asked by Lionelwallace Employee on Jun 18, 2014
Latest reply on Jun 24, 2014 by DaveThib

Question from customer:

 

I performed a test on our hardware using the ADAU1761 codec.  Below is a graph of the results.  I inserted a signal into our microphone that passed through a seven-pole, high-pass filter and to the codec input.  Output from the codec was filtered with a three-pole anti-aliasing filter.  There was no shaping done within the codec.

 

SEE ATTACHMENT

 

The blue trace is simply an analog loopback through the codec while the red trace is a digital loopback where the DSP simply routed the ADC output to the DAC input.  The rolloff at 8KHz is predictable from our 16KHz sampling.  What I did not expect was the frequency rolloff starting at about 500Hz.  There is a 9.5dB rolloff from 500Hz to 5KHz!  As mentioned above, we are doing nothing to color the audio, so I expected a flat response until we neared the Nyquist rate.

 

Do you have any explanation for what could be happening?  Is there a way to correct for it and flatten the response?

 

I asked the customer for additional data and received this reply:

 

Answers are embedded below in red.  I have additional information here as well.

 

I have attached the SigmaDSP project file for perusal.  Note that to change to pure analog loopback from the settings shown in this file, I just need to turn on the left input mixer (0dB) and turn off the left DAC from the Playback Output Signal Path.  Thus the only change in the whole path is the source of the output signal (the input signal is identical in the analog realm).

 

As an experiment, I have created another graph with the same analog signal but at different sampling rates.  Note that the rolloff at 5KHz persists until the sampling rate reaches 96KHz!  It gets progressively better, but even at 48KHz the response at 5KHz is down by 4dB.  At the 16KHz sampling that we will be using, 5KHz is down by 8.5dB.

 

I could convince myself that insufficient oversampling is at fault except that there is a 4dB hit even with 10X oversampling (5KHz at 48KHz sampling).  If this rolloff is inherent to the part or the process, we need to perform frequency equalization to flatten out the response.

 

From: Wallace, Lionel [mailto:Lionel.Wallace@analog.com]
Subject: RE: ADAU1761 Frequency Rolloff

 

A few questions:

 

  • Is the 7 pole High Pass filter external to the ADAU1761?  Yes, it is a 7-pole Butterworth, 120Hz HP
  • Is the 3 pole filter external to the ADAU1761  Yes, it is a 3-pole Butterworth, 8KHz LP
    • Can you provide a snip from the schematic if this filter is external?  I do not see how this helps since the straight analog path through the codec shows acceptable filter response
  • Do you have the ADUA1761 simple set up in a flow through mode, or are you performing some processing internally?  There is no processing taking place.
    • If processing, what are you doing?

Outcomes