Normalize gain scaling overshoots the set level

[Johnny Z]

I noticed that using normalize gain scaling on an LP or 78 RPM file usually overshoots the set level by 1-2 dB in DC10.51 and 10.62. For example, setting at -2.0 dB results in a nearly-clipped file (~ -0.5 dB peaks) and setting at -1.0 dB will usually clip.

I also have DC 7.15 and 8.50 installed on the same Windows 8.1 machine and using either of those earlier versions gives much more accurate results (setting -2.0 dB gives -1.5 dB typically), so I've been gain-scaling on DC7 for some time now.

I tested the function with 400 Hz sine waves from Make Waves and all 3 versions give about the same results with only a small overshoot (setting -2.0 dB gives -1.5 dB, -1.0 dB gives -0.75 dB). But gain-scaling actual music (classical, jazz, pop) with DC10.62 results in significant (and annoying) overshoots. Suppose I could just "under-spec" the setting and hope to get the desired level (usually -2.0 dB) but why do the earlier versions work so much better for gain-scaling?

[Craig Maier]

I am not seeing this using DCArt10.62. I do know that we intentionally set the clip led to light below 0dB - - - the reason for that is to prevent clipping - - - it is sort of a clip warning. The actual value of the clip led trip-point is somewhere around -1 dB as I recall. btw - what bit depth and sample rate are you using?

Craig

[Craig Maier]

Clip Test.docx Here is a make-waves 400 Hz signal that started out at -10 dB. I used normalize gain scaling to 0.0 dB. Note that it is not clipping - - - it just touches 0 dB, but the LED does light since it is set to illuminate slightly below 0.0 dB:

Clip Test.docx

[Johnny Z]

I use mostly 16-96 but 16-44 files also overshoot. The resolution doesn't seem to be the problem, but rather the type of material. Music that's already dynamically compressed like rock is less likely to overshoot, but wide dynamic range music like classical (or even spoken word like a comedy album) overshoots quite often.

For example, the attached screenshots show the results of -1 dB and -2 dB gain scaling on a live Brahms Symphony 1 LP. You can see that the music peak in the 4th movement overshoots the set levels in both cases: -1 dB makes the clip LED light and gives a slightly flattened waveform that touches 0 dB (see zoomed pic) and -2 dB gives a -0.7 peak. There's no audible distortion in the first case.

With classical, these results are typical, or slightly better or worse. The screenshots show DC10 tabs for a few other LP's tested. The 18th Century Harpsichord album gave the same results as the Brahms. The Dylan album gave more accurate results (-1 dB gave -0.75) and the 2000 Year Old Man record overshot a little less than the classical (-0.3 and -1.1 dB respectively).

So the overshooting is program-dependent, but since I'm recording classical most often, it's happened numerous times. As I mentioned, gain scaling with DC7 and DC8 are fine. I first noticed the problem in DC10.51 but when I recently upgraded to 10.62, it remained.

I usually gain scale to -2 dB to allow some headroom, right before setting up the MultiFilter. So if it overshoots the gain scaling, it may clip in the MultiFilter. I no longer chop into tracks, but leave each record as a preset in my "Giant Jukebox."

[Craig Maier]

Several things: 1. We have not changed the algorithm between DC7 and DC10.6x. 2. We do not see any overshooting in your screen shots; when a signal 'touches' 0 dB, that is not overshoot. 3. We can not reproduce your results here, so we are unsure what is happening on your end.

Here is a 400 Hz triangle wave (created by the Make Waves Generator in the DC10.6x). It was recorded at - 10 dB and than Gain Normalized to 0.0 dB. Note that it precisely touches 0 dB:



Call us at 973-316-9111 so we can work together and find out what is different between your system and one of our systems.

Thanks,

Craig

[Craig Maier]

I just researched and then tested the main VU meter Clip LED. It is designed to trip on (red) whenever the signal exceeds -1.0 dB. I verfified that to be the case on DCArt10.62 using a triangle wave set to various values between -1 dB and 0 dB.

Craig

[Craig Maier]

This is not specifically related, but did you know that each channel can be Normalize Gain Scaled separately? Just chose L or R rather than L/R. That is useful when the channels end up substantially out of balance with respect to one another.

[Johnny Z]

Perhaps the term "overshoot" is unclear. I use it to refer to any gain scaling that overshoots the value I enter, not necessarily values that equal or exceed 0 dB or cause the LED to light. That's why I said "...overshoots the set level by 1-2 dB" in my first post. If I specify -2 dB (as I usually do), I expect that the file peak should be closer to -2 dB (as it is in DC7 and DC8) than 0 dB (as it is in DC10, depending on the type of music as I said). Yes, the peak in the right channel at 40:01 touches 0 dB, but that's not what I want when I specify -1 dB. I could've given you dozens of other examples of "overshoot" I've documented, these are just a few.

You say the algorithm is the same in DC7/8/10, but I'm getting different results with actual music files. The previous versions give me what I want every time, but DC10 often does not.

I'd rather not do separate left/right channel gain scaling, since that may change the balance enough to cause soundstage shifting. As you probably know, in classical, the usual orchestra seating is unbalanced left/right in both frequency and often volume, with higher pitched instruments on the left and lower pitched on the right, and depending on the piece, it's not unusual that one channel or the other has considerably louder peaks than the other due to scoring. Bruckner's 4th Symphony with all those strong trombones on the right comes to mind. So I'd rather not alter the balance unless I've got a good reason.

I would like to discuss this further, as suggested, along with a couple other points. Give me a good day and time to call, and I'll contact you, Craig. Today, Friday 10/9 probably won't work, but next week will.

[Craig Maier]

Hi,

It is the same algorithm, so no difference in performance can happen. Please call us at 973-316-9111 and ask for Craig. I will sit down in front of a computer here while you do the same thing and we shall walk through it and identify the anomoly. It is hard to do that on the forum - - - we need to do a step by step analysis of the situation.

Craig

[Craig Maier]

Johnny Z and I chatted on the phone about this today. It appears that he is running DCArt10.6x on Win7 and Win8. This produces numerous issues as we discovered on the phone. DCArt10.6x was compiled on and for the Win10 operating system. Extensive Alpha and Beta testing was performed on that operating system alone. So, we recommend using DCArt10.6x on Win 10 only since it has never been evaluated on the earlier operating systems.

Craig