08-06-2018, 12:56 AM
Hi everyone,
I have run some tests between 3.8.4 (various kernels) and 4.1 but so far haven't been able to collate results to post with graphs yet, sorry. I used 3 x DACS: Topping D30 & D10 (USB), HifiBerry DAC+ Pro (IIS).
I can make a few observations:
- Switching between kernels and moOde versions made no difference to the overall frequency response of pink noise or music tracks. There were no variations that might show EQ differences, etc.
- looking at noise/distortion plots, the only difference I observed was a slight increase in harmonic distortion in 3.8.4 LL kernel in RR mode in the D30 (the DAC with the least amount of harmonic distortion). This was way down in the noise floor, below -110 dB. Other DACs had higher harmonic distortion, so they masked any difference.
There were some major differences but they appear to all be hardware related:
- HifiBerry DAC had higher aliasing artefacts, some of which were in the 0 Hz to 20 kHz range. These were well down in the noise floor and probably inaudible but nevertheless measureably higher than the two USB DACs. There were very noticeable ultrasonic spectra when playing back CD rips at 44.1 kHz but these disappeared in a 24 bit 96 kHz classical recording. I was using the FIR filter setting in the HifiBerry.
- There was a variable amount of ultrasonic noise mostly above 30 kHz on all DACs. This didn't appear to be dependent on kernel or moOde versions but the quality of it was different for each DAC (i.e. each DAC had it's own ultrasonic noise signature). I believe this is some form of interference because it disappeared in a couple of test runs and was not present in a Windows/USB DAC comparison test. Is this possibly a drawback of the RPi -- susceptibility to noise? Anyway this ultrasonic noise is well above human hearing and was buried under the noise floor of a 24 bit 96 kHz recording, so it is down below -100 dB.
In summary: in my measurements so far, I haven't found that changing kernels in 3.8.4 or switching to 4.1 made any demonstrable difference. Differences in hardware and noise susceptibility did make a measurable difference but I wonder how audible they are.
TODO: listening tests; chasing down the source of ultrasonic noise; possibly phase measurements; collation of data.
I have run some tests between 3.8.4 (various kernels) and 4.1 but so far haven't been able to collate results to post with graphs yet, sorry. I used 3 x DACS: Topping D30 & D10 (USB), HifiBerry DAC+ Pro (IIS).
I can make a few observations:
- Switching between kernels and moOde versions made no difference to the overall frequency response of pink noise or music tracks. There were no variations that might show EQ differences, etc.
- looking at noise/distortion plots, the only difference I observed was a slight increase in harmonic distortion in 3.8.4 LL kernel in RR mode in the D30 (the DAC with the least amount of harmonic distortion). This was way down in the noise floor, below -110 dB. Other DACs had higher harmonic distortion, so they masked any difference.
There were some major differences but they appear to all be hardware related:
- HifiBerry DAC had higher aliasing artefacts, some of which were in the 0 Hz to 20 kHz range. These were well down in the noise floor and probably inaudible but nevertheless measureably higher than the two USB DACs. There were very noticeable ultrasonic spectra when playing back CD rips at 44.1 kHz but these disappeared in a 24 bit 96 kHz classical recording. I was using the FIR filter setting in the HifiBerry.
- There was a variable amount of ultrasonic noise mostly above 30 kHz on all DACs. This didn't appear to be dependent on kernel or moOde versions but the quality of it was different for each DAC (i.e. each DAC had it's own ultrasonic noise signature). I believe this is some form of interference because it disappeared in a couple of test runs and was not present in a Windows/USB DAC comparison test. Is this possibly a drawback of the RPi -- susceptibility to noise? Anyway this ultrasonic noise is well above human hearing and was buried under the noise floor of a 24 bit 96 kHz recording, so it is down below -100 dB.
In summary: in my measurements so far, I haven't found that changing kernels in 3.8.4 or switching to 4.1 made any demonstrable difference. Differences in hardware and noise susceptibility did make a measurable difference but I wonder how audible they are.
TODO: listening tests; chasing down the source of ultrasonic noise; possibly phase measurements; collation of data.
