Raspberry Pi clipping on digital outputs?

[tboooe]

Hi guys. There is a very interesting thread on another forum about potential clipping with the Pi3.

 

Here is a portion of his latest findings (direct link to the thread here). I am not sure what to make of his findings. The idea of lowering the volume and losing resolution does not sound like a good solution.

When I connect the Pi to the DAC (with software volume control disabled) over either USB, or Coaxial from my Digi+, an issue occurs whereby the digital signal appears to be peak limited. This results in the THD and IMD test signal being truncated, thus causing the 15% THD results. If the Pi were rendering the digital information correctly, the DAC would give the same output as with any other digital renderer. The only reason why the DAC's output would change is if there were a change in the digital data it receives. I have shown that this issue occurs independently of the HiFi Berry product. I have also replicated the exact same results with: my USB drive unplugged from the Pi, using ethernet, wifi with ethernet unplugged, with wifi and bluetooth disabled, different power supply, various bit depths and sample rates, and with Volumio as well as RuneAudio. The issue must surely reside at the Pi's i2s stage or earlier.

 

I have found that, by enabling software volume control and setting the Pi's volume to 89%, the issue with the altered test signal goes away and the results in RMAA look normal for my DAC. However, by doing this you are essentially removing 2 bits of resolution from the signal. Using the RMAA 16/44 test with the Pi at 87% (the sweet-spot, see results up-thread) gives a dynamic range output of 85 dB which is the equivalent of 14 bits. Disable the volume mixer (so output is 100%) and you get ~16 bits of resolution, but 15% THD.

 

So, my question is: Is the Pi doing something at the i2s (or earlier) rendering stage that results in a clipped signal? I would be inclined at this point to say 'yes' as, to repeat what I wrote earlier, the analogue output of a single file from a single DAC should be exactly the same - regardless of the device used to render the digital information to it - so long as that rendering device is operating in a bit-perfect fashion.

[Pepsican]

Hi guys. There is a very interesting thread on another forum about potential clipping with the Pi3.

 

Here is a portion of his latest findings (direct link to the thread here). I am not sure what to make of his findings. The idea of lowering the volume and losing resolution does not sound like a good solution.

When I connect the Pi to the DAC (with software volume control disabled) over either USB, or Coaxial from my Digi+, an issue occurs whereby the digital signal appears to be peak limited. This results in the THD and IMD test signal being truncated, thus causing the 15% THD results. If the Pi were rendering the digital information correctly, the DAC would give the same output as with any other digital renderer. The only reason why the DAC's output would change is if there were a change in the digital data it receives. I have shown that this issue occurs independently of the HiFi Berry product. I have also replicated the exact same results with: my USB drive unplugged from the Pi, using ethernet, wifi with ethernet unplugged, with wifi and bluetooth disabled, different power supply, various bit depths and sample rates, and with Volumio as well as RuneAudio. The issue must surely reside at the Pi's i2s stage or earlier.

 

I have found that, by enabling software volume control and setting the Pi's volume to 89%, the issue with the altered test signal goes away and the results in RMAA look normal for my DAC. However, by doing this you are essentially removing 2 bits of resolution from the signal. Using the RMAA 16/44 test with the Pi at 87% (the sweet-spot, see results up-thread) gives a dynamic range output of 85 dB which is the equivalent of 14 bits. Disable the volume mixer (so output is 100%) and you get ~16 bits of resolution, but 15% THD.

 

So, my question is: Is the Pi doing something at the i2s (or earlier) rendering stage that results in a clipped signal? I would be inclined at this point to say 'yes' as, to repeat what I wrote earlier, the analogue output of a single file from a single DAC should be exactly the same - regardless of the device used to render the digital information to it - so long as that rendering device is operating in a bit-perfect fashion.

 

 

 

Sounds like this person uncovered a potential hardware failure. Recall perhaps?

[tboooe]

I am actually surprised this issue has not be reported on sooner. Makes me wonder if the tester either has a bad unit or his testing methodology is flawed in some way?

[Axiom05]

Sounds exactly like what happened with my Sonos earlier this year after one of the updates. (Subsequently fixed) The voltage input to the DAC is too high.

[Audio_ELF]

It sounds strange to me ... something doesn't add up if output via USB and output via i2s are both overdriven it can't be down to the i2s implementation as the USB output (surely) doesn't use any of the "audio" hardware?

 

It sounds more like a software bug rather than hardware issue. But then the analogue output doesn't appear to be affected, and that DOES (afaik) use the same software and i2s implementation!

 

PS. I'm not saying there isn't an issue ... just not sure the issue is identified correctly!

[tboooe]

It sounds strange to me ... something doesn't add up if output via USB and output via i2s are both overdriven it can't be down to the i2s implementation as the USB output (surely) doesn't use any of the "audio" hardware?

 

It sounds more like a software bug rather than hardware issue. But then the analogue output doesn't appear to be affected, and that DOES (afaik) use the same software and i2s implementation!

 

PS. I'm not saying there isn't an issue ... just not sure the issue is identified correctly!

 

I agree with you completely. It doenst make sense but I dont know enough about how the Pi is designed. I am definitely going to follow that thread. The interesting thing is I posted my thoughts about using my Pi as a renderer compared to my tweaked pc. In the end I preferred my tweaked pc because I thought the sound was more open and extended. Could this be a result of the clipping in the Pi?