RME ADI-2 Pro fs - New firmware update

[Rja4000]

Hi

RME just posted a firmware upgrade for ADI-2 DAC and ADI-2 Pro fs

For the later, they introduce the "M/S-Proc" functionality on the Analog input
This allows averaging (on the left channel) the 2 Analog inputs in the digital domain.

 

[Rja4000]

I had a very impressive SINAD result first time I ran it. (116.75dB !)

Now I can't reproduce it anymore.

I'll come back when I can :-S

From RME manual:
"The other application is to split a single analog channel signal to both analog inputs, then activate M/S-Processing and analyze the digital left channel only. This method performs mono summing, which raises the ADI's incredible SNR by another 3 dB for even better measurement analysis capabilities. "

 

[Rja4000]

OK,

so here we are

This is RME ADI Pro fs in loopback (DAC + ADC)
with firmware FPGA 213, DSP 96, 09/2019

To be noted though, I usually measure with a smaller FFT Window size (this one shows 131072, which is the size I normaly use for 192kHz)

And if I go down to my "normal" size, I get this

I don't really understand how the FFT window size may affect SINAD result here.
Noise level is decreasing by approx 3dB each time I multiply it by 2.
And so does the SINAD.

By the way, when I use an Y cable, instead of both output channels, I get a result a bit lower :

Best results here with Balanced Phone out (Low gain FS = +13dBu) to Analog In (+19dBU range), with M/S Proc on for the Analog input.
(Please note that for last measurement, the red channel is the "S" channel, while in the previous shots, both red and black were from "M")

It's still better than without the averaging with the new Firmware though.

 

[PaulD]

Hi

RME just posted a firmware upgrade for ADI-2 DAC and ADI-2 Pro fs

For the later, they introduce the "M/S-Proc" functionality on the Analog input
This allows averaging (on the left channel) the 2 Analog inputs in the digital domain.

Thanks for the alert! Just a note, I'm pretty sure that M/S-Proc is Mid-Side processing, allowing stereo decode of M/S signals as are common in pro recording situations.

 

[Rja4000]

Thanks for the alert! Just a note, I'm pretty sure that M/S-Proc is Mid-Side processing, allowing stereo decode of M/S signals as are common in pro recording situations.

Yes, it is.
(Well it's actually the reverse in that way : Stereo to M/S)
Mid is a sum (average) of both channels.

 

[GrimSurfer]

Why is the level set at -6 dBFS? Shouldn't it be 0 dB? And if so, what does this do to the SINAD?

 

[Rja4000]

Why is the level set at -6 dBFS? Shouldn't it be 0 dB? And if so, what does this do to the SINAD?

That's the best SINAD figure I can get from my device.

The output is from front phone jacks, in "phone balanced" mode. Each of the 2 stereo front jacks has it's own DAC and, in that mode, the 2 channels of each DAC are giving the hot/cold of a balanced output, left and right.
In that mode, the output has 2 possible gains: "Low" is used here - that's a hair more than +13dBu at full scale (0dBFS).
This is the output level.
In Multi Instrument, it's set in Volts, and, because it's easier to read, I always keep the full scale at 1. Then .5 is 50% of full scale and so on.
Here the output level is 1, so full scale, so a hair over +13dBu.

As this is over +13dBu (+13.06dBu or so), it saturates the input when the input range is set to +13dBu. I can either set the output .1dB lower or change the input range to +19dBu.
As this always gives better results, I do the later. That's why input level is -5.9dBFS.
That's +19-5.9=13.1dBu.
And 51% of input Full Scale.
That's around 3.5Vrms.

 

[Blumlein 88]

Yes doubling FFT size pushes the noise in each frequency bin down 3 db. You also need to look at size vs bandwidth for some purposes.

For instance a 32 k FFT for 24 khz bandwidth would have the same noise per bin as 128 k FFT for 96 khz bandwidth (assuming noise is the same across the band).

 

[Rja4000]

Yes doubling FFT size pushes the noise in each frequency bin down 3 db. You also need to look at size vs bandwidth for some purposes.

For instance a 32 k FFT for 24 khz bandwidth would have the same noise per bin as 128 k FFT for 96 khz bandwidth (assuming noise is the same across the band).

Actually, I typically use the power of 2 just under sampling frequency.
So 32k when freq. Is 48k, 128k for 192k and so on.
The 128k setting was a mistake here, which is why I had difficulties reproducing it at first.

For multitone, one need to increase the window size, otherwise the low frequency resolution is not detailed enough.

Inter frame averaging is also used, but the impact is just visual, to better display the distortion over the noise. Figures remain the same.
Note that Multi instrument shows the Noise Level in the graph, which is helpful (dashed horizontal line)

In its dashboard, Amir said he's using 32k for 44.1 or 48kHz.
He also uses 4Vrms, which is a few tenths of a dB more output level.
To get that level (around +13.7dBu), I'd need to switch the output to Hi gain (+28dBu or more than 19Vrms at full scale) and dial down the output by 14.3dB, and the SINAD figure would then be quite lower.