DC offset

[lmu202]

Could somebody explain the DC offset in general. I exported a project as usual and it looks off. It's a combination of recorded audio and plugins (ones that I use constantly). But when comparing the two parts of the song it's obvious that the first half is badly offset. I can easily correct it in Sound Forge but this is new territory for me. What causes it and why haven't I seen it earlier during 20+ years of using Cakewalk (+Sonar). Surely the hardware has changed during the years but not much recently. And the song has been played with the same gear through the both parts.

[bitflipper]

Are you sure it's actually DC offset? It's possible to have asymmetrical waveforms that look like that, and still be OK. Try placing a HPF on the track and see what that does to the waveform. 

btw, if it does turn out to be DC offset it's not necessary to fix in an external editor, as Cakewalk has a DC removal feature. It's under the Process menu.

[lmu202]

Thanks bitflipper! I tracked the source of this and it's a synth plugin (Dimension Pro, a preset called 'FM pad' if anyone wants to investigate). When this is muted the waveform of the mix looks normal. Weird but I'm happy it's now sorted.  I froze the track and tried using the onboard DC offset removal tool but it didn't work as wished so I exported the track and performed the removal in Sound Forge. All good now ?

Here's the soloed track exported:

 

[bitflipper]

Well, that certainly is unusual. Technically not DC offset, though, which results in an otherwise normal-looking waveform whose center line is shifted up or down. This  is just half a waveform. Unfortunately, I no longer have Dim Pro installed here so can't investigate further. But my assumption would be that as long as it sounds OK it won't cause any problems beyond looking strange. Note that sounds created using FM synthesis often do look weird.

[User 905133]

10 hours ago, lmu2002 said:

I tracked the source of this and it's a synth plugin (Dimension Pro, a preset called 'FM pad' if anyone wants to investigate).

I was curious about this because I have seen waveform displays that are lopsided (e.g., higher % below zero than above) but never took the time to dissect them. So, I looked at patch in realtime using Melda's MOscilloscope.  Sure enough, except for a very tiny portion of the wave, it is almost exclusively below the line.  The normalize display settings change that but very slightly. 

For the most part the underlying sample for the prog is fm 01.flac.  I looked at the sample and I'd say the 1st fifth of the cycle is above 0 and the remainder is below, except for a smidgeon at the very end.

[bvideo]

I'm not sure DC offset can be observed with a waveform drawn at that scale. Actual "area" of top and bottom can't be judged by eye because the thickness of the lines is so large compared to the underlying pattern. That half waveform looks really unusual at that scale, but if it still looks that way after applying DC offset correction, it could look different zoomed way up.

But here's another issue: that FM pad Dimension Pro patch seems to generate a very large very low subsonic component overlaid over the whole waveform. (Maybe less than .5 Hz) In viewing the waveform, parts of it could lie entirely in the positive or negative region. Removing DC offset won't change anything unless you focus on a short portion. Even then, you can still see the effect of the VLF.  Putting a high pass filter (Prochannel EQ) on the Dimension Pro audio track results in a much more normal looking waveform.

Here's the original waveform, about 5 seconds worth, 7 consecutive notes.

After Apply Effects -> DC offset (shifted,but not really looking 0 offset)(edited here)

edit: image replaced

Recorded with Prochannel HP filter enabled, set to 31 HZ:

 

Edited February 19, 2022 by bvideo
wrong image

[bvideo]

Here's an example of viewing a signal with a known 0 offset zoomed out, so it looks horribly skewed negative. (It's not very musical because it's a tone artificially constructed for an example.)

It looks like it has a large negative offset, but it is indeed 0 (as per sound forge).

But when zooming up on this signal, the line drawing is relatively thinner, so you can get a better idea of how the area above 0 might be equal to the area below zero (thus DC offset = 0).

Although this signal is artificially constructed, synth waves are often quite artificially generated too. The point is you can't judge the DC offset by looking at a zoomed out drawing on your monitor.

[User 905133]

OK.  I figured it out. You and I are talking about different things.  Never mind.

Edited February 20, 2022 by User 905133

[lmu202]

15 hours ago, bvideo said:

I'm not sure DC offset can be observed with a waveform drawn at that scale. Actual "area" of top and bottom can't be judged by eye because the thickness of the lines is so large compared to the underlying pattern. That half waveform looks really unusual at that scale, but if it still looks that way after applying DC offset correction, it could look different zoomed way up.

But here's another issue: that FM pad Dimension Pro patch seems to generate a very large very low subsonic component overlaid over the whole waveform. (Maybe less than .5 Hz) In viewing the waveform, parts of it could lie entirely in the positive or negative region. Removing DC offset won't change anything unless you focus on a short portion. Even then, you can still see the effect of the VLF.  Putting a high pass filter (Prochannel EQ) on the Dimension Pro audio track results in a much more normal looking waveform.

Here's the original waveform, about 5 seconds worth, 7 consecutive notes.

After Apply Effects -> DC offset (shifted,but not really looking 0 offset)(edited here)

edit: image replaced

Recorded with Prochannel HP filter enabled, set to 31 HZ:

 

Yes there's a ton of very low frequency content when I looked into it. Much louder than the audible signal. HPF solved the issue.

But if this kind of waveform doesn't produce any problems, when should one be cautious of such phenomena? Surely the 'remove dc offset' function is there for a reason, both in Cakewalk and Sound Forge.

[bitflipper]

Extreme DC bias in a full mix can mess with your speakers' dynamic range, by continually pushing the driver out or pulling it in and thus reducing its throw. Depends on the speaker, though. Many active monitors have their own DC filtering.

As noted previously, just looking odd isn't necessarily a problem. FM waveforms are by nature wildly unnatural and often weird-looking.

Also note that VLF content (subsonic frequencies) can look like DC. At what point do very low frequencies become "DC"? I don't know.  However, I do know that for most genres, anything below ~45 Hz is mostly wasted energy, and anything below 30 Hz is definitely not needed. Even if you're making bass-heavy dance music intended to be reproduced on large full-range systems, there is a point where it's wasted energy that messes with playback systems and biases your mastering limiter to the point where perceived bass and overall loudness can actually suffer.

For this reason, it's a good practice to always have a HPF on your master bus and just get rid of that stuff. If you can't hear it, dump it.

[bitflipper]

Should also note that excessive VLF is quite common, even in sample libraries. Here's the first one I chose at random, from a reputable library developer.  The greatest amount of energy is below the usable range (this is not a low-pitched instrument). The presumption is that you, the mixer, will decide how much of it you want to use.

 

[Glenn Stanton]

22 hours ago, bitflipper said:

At what point do very low frequencies become "DC"?

from my test electronics days - we'd consider anything 5Hz and under to effectively be DC so blocking anything below 10Hz was fairly standard to avoid issues with the DACS. from a musicality standpoint, unless you're doing something with large pipe organs, cutting anything below 20Hz would likely solve this. for VLF "effect" like gaming and theater effects - i'd put those on a separate "effects" track and ultimately route to the appropriate effects channel (if there is one).

[bvideo]

The DC removed by sound forge or cakewalk is only the constant bias that is computed over some length of the signal. Both sound forge and cakewalk offer a "first 5 seconds" option that computes the bias over that segment (instead of the whole clip) and then applies the bias to the whole clip.  Either way, "remove DC offset" certainly fails to remove any VLF energy of any periodic waveform shorter than, say, 4 times the length of the clip.

So except for certain theatrical purposes, it does seem like a good idea to filter out inaudible  periodic signal somewhere in the signal chain.