Is this normal behaviour for dacs?

[bigshot]

You sure do go on and on…

 

[71 dB]

These noise levels and distortions would be interesting if they were audible, but as far as I can see they are far below audible levels so who cares? Some people apparently. I find it personally boring analysing the performance of well-functioning audibly transparent gear. It is kind of dangerous to see graphs of sound, because we can detect issues we would never hear with our ears.

 

[Ryokan]

It is kind of dangerous to see graphs of sound, because we can detect issues we would never hear with our ears.

They're only meant to be studied by professionals who don't get hung up on what is well below audibility.

 

[71 dB]

They're only meant to be studied by professionals who don't get hung up on what is well below audibility.

Yes, exactly.

 

[Ghoostknight]

i think i can decide myself what i go on and on about

 

[bigshot]

You don’t have to share everything you think up.

 

[VNandor]

the -78.7db are also SNR, maybe this gets calculated into the thd+noise measurement?

I ran some tests with REW with some sines and noise.

I used a 1kHz sine wave with -3dBFS peak (so it's -6dBRMS). I added white noise to it that averages at -46dB and happens to peak at -41dB.

The SNR is at 40dB which is expected because the SNR should come from the RMS values and not from the peak values. The noise alone is reported as -43dBFS. I don't know how REW arrives at this number (this turned out to be an AES17 standard thing). It's too much for average but too little for peak. The high and low pass filters are set to off. According to REW, there is "some" harmonic distortion present around -85-90dB which is below the noise level by at least two orders of magnitude. The harmonic distortion is calculated from the frequency bins at 2kHz, 3kHz and so on... Since there is white noise in the signal, these bins won't be at zero just very near to it. As the FFT length increases, the frequency bin width decreases so it passes less and less parts of the white noise through which is why a longer FFT length can visually "push down" the noise.

As long as there is no harmonic distortion in the signal, the SNR is the same as THD+N. It looks the noise part of the THD+N is relative to the fundamental as well. I knew THD measurements are relative to the fundamental, I didn't know that applies to the noise as well.

I added 2 harmonics below 40dB of the fundamental and removed the noise. The noise drops to the 24bit dither so the SNR is ~140dB but THD+N goes up to -37dB. The reason it's not up to -34dB is because the harmonics aren't perfect copies of each other, their frequencies are different so summing them together increases the average level by only 3dB.

Once I add the noise and harmonics into the signal, the SNR and the THD+N start to differ. This is what indicates that the THD is not negligible compared to the noise.

-3.6db dac gain vs -10dB dac gain:

In the screenhots you posted with different gains, the THD+N and the SNR are the same with -10dB gain and they differ by 0.1dB with -3.6dB gain. For moment, let's just ignore the fact there is no way to tell if this is caused by the DAC, the ADC, or this is just the typical deviation of the measurement setup you use. 0.1dB difference is about 1%. The THD is about 1% of the noise with the -3.6dB gain so it is way, way below the noise floor. The noise is already about 0.006% compared to the signal, and the THD is only 1% of that noise.

If someone took your advice and used -10dB gain instead, they would double the noise level to 0.012% to get rid of that THD that's not even confirmed to come from the DAC and might be so much below the noise that it can't even be measured consistently. Not to mention that music is not at -3dB but rather at -13dB or -23dB so the DAC is not around the peak as much so it will distort less with music. Additionally, the distortion very likely will fade into the noise floor when the DAC is used with music. Music's noise floor is not at dither level as the recording is constrained by the physical word so the noise gets added from the studio, the mic preamps, the ADC, and if more than one mic is used then it gets added to the recording multiple times, then the noise will be further increased during mixing. Even though -10dB gain won't likely cause a problem as long as the noise contained in the music plays over it but it might be heard when no music is playing as you have to make up for it at the amp.

 

[castleofargh]

It's been years since I really looked into the basic settings of REW(if it works, don't touch it), but I do remember something about a default -3dB somewhere(that could be removed). I'm not sure if it was just a visual thing(shown value) or if it was real to limit the risk of clipping, but maybe that still exists?

 

[danadam]

I'm not familiar with REW and I'm not about to get familiar with it.

I ran some tests with REW with some sines and noise.

 

[VNandor]

I noticed I only have to know how to use the RTA module and I can ignore the actually hard stuff

 

[VNandor]

It's been years since I really looked into the basic settings of REW(if it works, don't touch it), but I do remember something about a default -3dB somewhere(that could be removed). I'm not sure if it was just a visual thing(shown value) or if it was real to limit the risk of clipping, but maybe that still exists?

There's an option in preferences->view that sets the full scale sine as 0dBFS RMS that's checked by default. That was it.

 

[Ryokan]

I noticed I only have to know how to use the RTA module and I can ignore the actually hard stuff

Jeez if you think it's hard