DR calculation in RTA seems wrong

[nyt]

Same signal, only change was changing the input padding by 10db. DR shouldn't be changing here, should it?

 

[John Mulcahy]

Unweighted THD+N has dropped and likely by a little more weighted, so the DR figure will increase. The test signal levels are too high for a DR measurement, DR should be measured with a -60 dB signal for a result per AES17 - distortion shouldn't be a large contributor, but applying a test signal ensures there is no muting causing invalid results in the absence of a signal. From the help:

An A-weighted dynamic range figure is also shown alongside the THD data. The figure assumes that the system being measured is capable of reproducing a sine wave at -0.1 dBFS (where 0 dBFS is a full scale sine) at distortion of better than -40 dB. If the actual maximum output is known the DR value used should be amended accordingly. For a meaningful result the system should be driven with a 997 Hz sine wave at -60 dBFS, per AES17-2015.

 

[nyt]

Unweighted THD+N has dropped and likely by a little more weighted, so the DR figure will increase.

That's what I would expect, however THD+N and N are actually increased in the measurement listing the higher DR figure, so something seemed off

0.0013% THD+N and 118.8 dBa DR

vs

0.00083% THD+N and 111.9 dBa DR

However, I see the DR is calculated based off of what you say instead of what's actually being measured. This seems kind of strange since it's unlikely the max output of the DUT and the max input of the measurement interface will be exactly the same levels...

 

[John Mulcahy]

The percentage figure is higher but the test signal, which it is a percentage of, has dropped by more so the absolute level of THD+N has dropped. Bear in mind also the DR is an A weighted figure.

 

[nyt]

The percentage figure is higher but the test signal, which it is a percentage of, has dropped by more so the absolute level of THD+N has dropped. Bear in mind also the DR is an A weighted figure.

I edited my earlier post.

The only change was on the input side of things. I understand it's not going to be completely precise by 10db, but it's the same signal, just one measurement is padded 10db more than the other. The noise floor has possibly come up by less than 10db in the second measurement, hence the noise floor difference.

I see the issue with DR is that it's calculated while expecting the signal to be able to reach -0.1dbfs instead of reporting on the actual DR of the measurement.

This seems kind of strange since it's unlikely the max output of the DUT and the max input of the measurement interface will be exactly the same levels...

 

[John Mulcahy]

DR is from THD+N, at the lower signal level the harmonics have dropped by slightly over 10 dB. The idea of using - 60 dBFS as the test signal is that harmonics would usually be in the noise floor, then the noise figure is the determining factor.

A DR figure per AES17 requires a separate measurement of the maximum level (highest while achieving THD+N better than -40 dB).

 

[nyt]

DR is from THD+N, at the lower signal level the harmonics have dropped by slightly over 10 dB. The idea of using - 60 dBFS as the test signal is that harmonics would usually be in the noise floor, then the noise figure is the determining factor.

A DR figure per AES17 requires a separate measurement of the maximum level (highest while achieving THD+N better than -40 dB).

So REW is performing the DR calculation by assuming the max level is -0.1dbfs? This seems like it will rarely ever be accurate. Wouldn't it make more sense to display SNR there? It seems without the ability to enter the max level for the DR calculation that it's mostly useless.

Also, the harmonics level difference is almost exactly 10db, and is visible in the THD calculation, 0.00059 vs 0.00060 percent. The different pads will not be exact, but it's very very close as you can see by the overlay. The noise however is slightly below 10db difference.

 

[nyt]

Maybe still misunderstanding, does REW add 60db to the fundamental signal to calculate dr and just caps it at 0dbfs? Assuming the user has found the max level, then reduced output by 60db? Or does it use the peak sample value?

 

[John Mulcahy]

No, the only use made of the fundamental is subtracting it to find the N+D figure in dBFS.

 

[nyt]

No, the only use made of the fundamental is subtracting it to find the N+D figure in dBFS.

So if it's only using the value of the fundamental compared to the noise floor, then the DR figure seems wrong.

The second image has an increased dynamic range, yet it is reported lower.

The spectra from the second image converted to red and overlaid on the first clearly shows the increased DR, yet DR was listed at about ~112db as opposed to the ~119db in the first image.

 

[John Mulcahy]

We seem to be talking at cross purposes. The DR figure is not stated relative to the fundamental and "the value of the fundamental compared to the noise floor" has no bearing and no use is made of it. The fundamental is subtracted from the signal, what is left is noise and distortion. It doesn't matter what level that fundamental was (beyond its effect on distortion), it is removed. The remainder, noise and distortion, is weighted and gives an absolute dBFS figure that is expressed relative to the maximum dBFS level, or in REW's case relative to -0.1 dBFS.

 

[nyt]

We seem to be talking at cross purposes. The DR figure is not stated relative to the fundamental and "the value of the fundamental compared to the noise floor" has no bearing and no use is made of it. The fundamental is subtracted from the signal, what is left is noise and distortion. It doesn't matter what level that fundamental was (beyond its effect on distortion), it is removed. The remainder, noise and distortion, is weighted and gives an absolute dBFS figure that is expressed relative to the maximum dBFS level, or in REW's case relative to -0.1 dBFS.

Understood, thought that's what you said earlier, but the last message didn't make sense.

Generating a DR figure in this way is only meaningful if the max output of the DUT lines up with -0.1dbfs on REW's input. Chances of this happening are pretty slim. It would likely be better to calculate the SNR of the RTA measurement, or calculate the DR using a historical peak value as the max output instead of -0.1dbfs, or both

Just my $0.02, but the current DR calc does not seem very practical. It's pretty much just a nonsensical measurement based on the noise floor at this point.

 

[nyt]

Any plans to work on this so it's a useful figure?

 

[John Mulcahy]

I think it's useful, or I wouldn't have bothered to implement it in the first place. It provides at least half the answer, since it gives the bottom end of the span required for an A weighted dynamic range figure. If the top end, i.e. the maximum level within the distortion threshold, is at -0.1 dBFS it already does the whole job. Likely to be the case when measuring an ADC, for example. If not, a figure for the maximum level needs to be obtained by the user, there is no way for REW to know or guess what that might be.

 

[nyt]

I think it's useful, or I wouldn't have bothered to implement it in the first place. It provides at least half the answer, since it gives the bottom end of the span required for an A weighted dynamic range figure. If the top end, i.e. the maximum level within the distortion threshold, is at -0.1 dBFS it already does the whole job. Likely to be the case when measuring an ADC, for example. If not, a figure for the maximum level needs to be obtained by the user, there is no way for REW to know or guess what that might be.

No DUT I've ever measured except a specifically calibrated interface has ever had the max output line up with -0.1dbfs, so the number displayed is virtually always wrong.

You could use the peak captured level to calculate (ideal) or let the user enter the max value, otherwise having it display is just misleading.

In fact, the only time this seems like it would ever be useful is for measuring the noise floor through a loopback of a perfectly calibrated device.

When performing acoustical measurements, there's almost no chance it will ever be accurate.

 

[John Mulcahy]

I have replaced the RTA "DR" figure by A-weighted noise plus distortion in the current Y axis units.

 

[nyt]

I have replaced the RTA "DR" figure by A-weighted noise plus distortion in the current Y axis units.

Cool, was paining me seeing irrelevant figures most of the time on generated measurements

If you list DR based on the stored peak value, that would be awesome, or even calculate SNR based on the fundamental level to the noise floor you have now that'd also be nice.