I don't know what it is, but it can be determined pretty easily. However, bear in mind that the test signals are not really tones - they are band-limited random noise. I mention this because in order to get an accurate reading (for example, in terms of RMS volts), you would need a meter that is true RMS. Many 'budget' volt meters are desigend with measuring a simple, single sinusoid in mind (i.e. 50 or 60 Hz as those are the mains frequencies around the world), and most will not accurately measure something like band-limited random noise.

Having said that, and assuming that you had such a meter, then you could measure this for yourself, but you would need a reference file against which you could compare the observed values, and the reference file would also have to be known in terms of its scale (i.e. how many dBFS).

So, let's suppose that you had a 60 Hz wav file that was 0 dBFS (all bits hi). You could play this through the 990 (with amplifiers disconnected) and for a given setting on the volume control, observe its value in RMS volts.

Since I don't know that actual scaling of the 990's outputs, I can't say (or guess) what you would see with the volume control set to 0 dB (although I think you can only get to -1 dB, but I am not sure about that. Anyway, you could play the 60 Hz wav file and note its voltage. Then, you could turn on the test tones and measure their RMS voltage. Since you know the reference file is full scale, then the RMS value observed for the band-limited random noise could be used to compute how many dB down the test signal (random noise) is.

For example, suppose the 60 Hz wav file (for a given setting on the volume control - and you DO NOT change the volume control setting for your measurements) produced 500 mV RMS. For that SAME setting on the volume control you measure the test signals (band-limited random noise) and observe that 330 mV RMS are produced. Given these two values you could make the calculation for dB relative to Volts. That is:

dB = 20 * log (V1/V2)

substituting, we get

dB = 20 * log (330/500) --> dB = 20 * log (0.66) --> dB = 20 (- 0.180) --> dB = -3.61

That is, based on the observed voltages (using the control signal (0 dBFS) as your reference) you would know that the test signal (band limited random noise) was 3.6 dB down from full scale, or -3.61 dBFS. You could say this because the file against which you are comparing the band-limited noise is known to be a full-scale file.

Assuming that the D/A has high linearity then this relationship should hold for any 'reasonable' volume control setting. In other words, whether you did this experiment with the volume control set to -30, -20, -5, or 0 (assuming the volume control can reach 0 dB) the RELATIVE difference observed in the two signals' RMS values should be the same.

So, let's assume that the D/A is very, very linear, and you conduct this experiment at several volume control settings..., write them down, and compute the difference in their magnitudes, in dB. Now...the values that I am showing below were pulled out of the air for the sake of illustration, but pay attention to the difference, in dB - it is conserved as a function of the volume control setting on the 990:

-10 dB SETTING
Vref file: 500 mV, Vtest tone:330 mV, delta dB = 3.61

-13 dB SETTING
Vref file: 353.5 mV, Vtest tone:233 mV, delta dB = 3.61

-16 dB SETTING
Vref file: 250 mV, Vtest tone:165 mV, delta dB = 3.61

-20 dB SETTING
Vref file: 50 mV, Vtest tone:33 mV, delta dB = 3.61

-30 dB SETTING
Vref file: 15.8 mV, Vtest tone:10.44 mV, delta dB = 3.61

-40 dB SETTING
Vref file: 5 mV, Vtest tone:3.3 mV, delta dB = 3.61

If you were to do this over the full range of the volume control, at some point (most likely, as the volume control's setting went further and further negative), the values that you would observe between the two files would no longer 'track', and you would start to see the delta change from (in this example) 3.61 dB to something otehr than that, and this is because of non-linearity in the D/A as well as other elements in the circuit topology that could affect linearity.

Likewise, if you did this experiment at (assuming that you can get there) 0 dB on the 990's volume control, you should see this same relationship. Again, I am NOT saying that the values that I have shown in my example are what you would observe as actual voltages - they are only to illustrate my point and methodology.

So far so good, however, keep in mind that getting an accurate RMS estimate of these values depends upon that which you are using to measure them. A more appropriate way would be to compute the signal power using Fourier analysis, and using the same settings for each analysis. That is, PC-based spectrum analyzers that have a highly linear A/D front end for the measurement and the requisite signal processing software to do these calculations would be helpful.

Using such an animal, you could either calibrate it with a known reference voltage (to get dB relative to a known reference, and in so doing, also be able to know the actual voltage of the signals measured), or just rely upon the relative difference in the values.

For example, if you used 1 VRMS as a calibration signal for the analyzer, then you would know that in the ANALYZER that 0dB would equate to 1.0 VRMS; were you to choose a calibration file that was 100 mV RMS then 100 mV RMS would be your reference, and as such, when the analyzer measured 100 mV RMS it would show up on the analyzer as 0 dB.

Thus, if you were to calibrate the analyzer using your 0 dBFS file for the HIGHEST volume control setting (i.e. 0 dB on the 990), then your first measurement would show that the reference file was indeed 0 dB, and when you play the test tones, the number of dB down (from full scale) would be calculated for you, and this would be the number of dB down from full scale the test tones are.

Likewise, as you work your way down the volume control (going more negative) then there would just be an offset to the values you see in the analyzer, expressed in dB. For example, if you calibrated the analyzer with the 990's volume control at 0 dB, then when you measured the full-scale 60 Hz wav file at a volume control setting of - 10 dB, this would be its magnitude in the software analysis (i.e. - 10 dB), but again, the delta you observe (when you play the test tones) should be the same DELTA as when you took the measurement with the 990's volume control at 0 dB - there would only be an offset.

It doesn't really matter what value you use as a reference (unless you want to know the actual value in volts...and then it matters), but as long as you do not alter any gain settings, the RELATIVE differences that you measure, in dB will be correct.

Does this help?