DC Offset - variance

[amanise]

I'm sure we all get a bit of DC offset to clean up from time to time. Whenever it pops into any recorded track I've just added to a project I clean it up at source (i.e. in the recorded wav file) so that it doesn't get layered up as the project grows. This is just habit now whenever I bring in a new overdub.

The variance in its occurrence has got me thinking whether or not some kinds of gear are more prone to it than others, and in some cases, would it indicate that a bit of kit was on its way out?

At what point to the pro audio folks start to pay attention to it?

[Hugh Robjohns]

I've not come across a DC offset in any recorded material I've been involved with in decades. Often people mistake normal waveform asymmetry as a DC offset...

The waveform stats function in iZotope's RX will measure and display DC offsets, and I'm sure there are other tools that can do the same.

Genuine DC offsets only occur as a fault — typically a converter fault but occasionally as an analogue hardware failure or plugin coding error.

The last example I remember was of a commercial Erik Satie piano works CD recorded by a forum member over 10 years ago.

I'm astonished you are commonly finding DC offsets. That's really not normal. Can you track down the offending source in your recording chain?

[Wonks]

It simply shouldn’t exist in modern kit. It could exist quite severely in early A/D converters, hence the addition of ‘remove DC offset’ options. But these days any DC offset is minuscule. Anything significant you see these days is down to faulty equipment, not design.

Not all waveforms are symmetrical around the x-axis, so don’t assume that because a waveform doesn’t look symmetrical, it has a DC offset.

[amanise]

I've not come across a DC offset in any recorded material I've been involved with in decades. Often people mistake normal waveform asymmetry as a DC offset...

The waveform stats function in iZotope's RX will measure and display DC offsets, and I'm sure there are other tools that can do the same.

Genuine DC offsets only occur as a fault — typically a converter fault but occasionally as an analogue hardware failure or plugin coding error.

The last example I remember was of a commercial Erik Satie piano works CD recorded by a forum member over 10 years ago.

Can you track down the source in your material?

Yes - its RX8 which is letting me know about it. I can certainly never detect anything by ear, but getting shot of it gives me so much more headroom its always worth doing. That's why I've built it into my routine for adding new tracks. Sometimes I don't get any at all - other times its obvious which bit of kit is the culprit. For example, I used to use a very cheap external audio interface (I'm talking £15 cheap) for grabbing snippets of audio out of headphone sockets to my laptop. I always got a few hundredths of a percent of DC offset on both channels which would clean up easily. I eventually forked out for a proper USB interface - which always gets zero DC offset. I suspect that my Marshal JTM45 cranked through a Bugera attenuator and fed into my Mackie desk is another culprit - but that's probably the Bugera - another budget bit of kit. I'm happy with that also. I also use an outboard Behringer 32 channel EQ and limiter unit from time to time - which I think adds a bit when pushed. My answer to that as with any Behringer kit is to not push it.

I understand what it is (I think) - but is it telling me anything I should be worried about is what I'm wondering.

[Wonks]

The JTM45 will be the amp output. Slightly different outputs from the two output valves. The Bugera box is just a resistive load, so won’t do anything. It’s all part of the sound.

A plugin can calculate a DC offset, but if the numbers are very small, it’s probably an artefact of the calculation on an asymmetrical waveform, rather than a real DC offset introduced during the recording process.

[amanise]

The JTM45 will be the amp output. Slightly different outputs from the two output valves. The Bugera box is just a resistive load, so won’t do anything. It’s all oarg of the sound.

A plugin can calculate a DC offset, but if the numbers are very small, it’s probably an artefact of the calculation on an asymmetrical waveform, rather than a real DC offset introduced during the recording process.

That's interesting - thanks. It's always in hundredths of percents - which sounds very small, but the headroom you get on a waveform after asking RX8 to correct it is always in whole dBs - i.e significant if you want to use that later for adding punch and body without running down your integrated LUFS budget.
Cheers!

[Hugh Robjohns]

Sounds like the sources are ultra cheap gear, poorly designed, or slightly misaligned power amps. I'm not surprised they present a DC offset.

What does surprise me is that your recording gear doesn't reject it. Most inputs are capacitively coupled and so prevent DC offsets getting in.

As I said, it's not something I come across in my recording life... but it is obviously very easy to correct with RX and most DAWs have a DC-removal option somewhere if you detect it.

I routinely use a stereo vectorscope (goniometer) when recording and mixing, and that reveals a DC offset very clearly as the squiggly line starts and ends at some point along the X or Y axis rather than at the origin.

You can also see it on a DAW waveform display (if enlarged sufficiently) as the waveform builds and decays to a point above or below the marked centreline.

[amanise]

Yes - you can also see it on the vertical axis in Audacity even - if it's large enough or you adjust the display to magnify it.

My recording gear doesn't reject anything - not even my singing

[James Perrett]

What you are seeing when you try to remove DC offset is possibly phase rotation. The process alters the phase relationship between different frequencies in the signal to produce a more symmetrical waveform. This is often used in broadcast processors to increase the apparent level.

I've not seen a genuine DC offset since I stopped using the Turtle Beach Multisound in the late 1990s.

[sonics]

Good metering and phase adjustment tools can be very helpful with this type of analysis, but like the others here, I've not had to correct a DC offset issue for a couple of decades I think. Phase problems due to bad recording or over-enthusiastic use of plugins is another thing!

[amanise]

Interestingly... I got a new clue from the cosmos (well, my computer) on this last night. I was using the DN1-E soft synth in ACID (supplied with Pro V10 by Magix if memory serves) and had selected the ubiquitous 'Simple Saw Tooth' voice for a passage of about 16 bars of medium velocity midi placed for future use as backing to the inevitable face melting guitar solo. It nice here in the past.

I rendered the clip out to reload it as a wav 'one-shot' - because for some reason my DAW always fails to remember that voice after exit and when started again next time always loads 'Spanish Guitar' instead. Its only ever that sawtooth voice it does it on - and one of the reasons to love the DAW. So, its less pain to convert those clips to wav clips if they don't look like they're going to need tweaking.

As is my habit - I then took a quick look at the clip, before reloading it, in RX8 to give it a lick and a polish - and what did I find? About 0.01% DC offset each side. Now that must be being generated by my sound card - surely? Which is a Soundblaster Audigy 5 from memory - so getting on a bit now.

[Wonks]

Unless you are recording with a mic or using a line input, it has absolutely nothing to do with your sound card. Everything else is all down to the computer and the code and sample data etc.

[amanise]

Unless you are recording with a mic or using a line input, it has absolutely nothing to do with your sound card. Everything else is all down to the computer and the code and sample data etc.

Software generated DC Offset? That's a confusing idea (I'm easily confused). I don't think it's a sample driven soft synth - so no sample data. It would certainly be a realism selling point for the software

So maybe part of the rendering process from sound card to disc write? That sounds bad news to me - there's no new computer in my runes for a very long time indeed. Good job it cleans up in RX and I know what to do about it.

[Wonks]

As has been said before, it’s probably not true DC offset, but phase issues, and I’d ignore it.

The sound card has two functions. A) to capture sound and to convert it to digital format, and B) to convert the digital files it’s sent to an analogue signal to be sent to your monitors.

Unless you are actively recording then the sound card is only providing the output sound, and plays no part at all in how the digital wave files are manipulated. (You could argue that a soundcard with DSP like a UAD Apollo can take part in the sound computing process, but if not processing the sound on the way in, the DSP acts just like the stand-alone UAD DSP cards and is only part of the sound card for convenience).

The soundcard doesn’t write any files to the disk, it’s only the computer that does that.

The code is the code, and should run the same on any computer (bar coding errors for different operating systems). So Acid and Ozone on your computer should behave in exactly the same way as on a brand new top-of-the-line PC.

[amanise]

As has been said before, it’s probably not true DC offset, but phase issues, and I’d ignore it.
....

Yes, it's not killing me so you're right. Just a cheeky little conundrum to speculate on. The other thing I wondered in the shower was (given what you said about samples etc) - could it have been 'leaking' though from the drum samples in that section of audio - even though they were muted by the DAW? I have a sneaky feeling those old 808 samples of mine my be a bit dodge down at that level. They're also at the base of most things I do. But muting should be muting - right?

[Wonks]

Muting is muting.

You can of course try muting everything, export the mix and then inspect the resulting file, which should either be a flat line along the X-axis, or just show dither noise if it applies it automatically on mixdown.

[amanise]

Muting is muting.

You can of course try muting everything, export the mix and then inspect the resulting file, which should either be a flat line along the X-axis, or just show dither noise if it applies it automatically on mixdown.

Now that's a GREAT idea! Next job on the list! Cheers!

[James Perrett]

Maybe RX8 is wrong...

Do you know that there is a version 8.5 which RX8 owners can upgrade to?

[amanise]

Maybe RX8 is wrong...

Do you know that there is a version 8.5 which RX8 owners can upgrade to?

Well - I have done Wonks' suggestion of a render with everything muted. I also ramped up the master fader to a ridiculous level to make sure there was good 'pressure' behind the empty render. I got the predicted flat line with zero output in RX8 and zero DC Offset. -inf on both channels and everything as you'd expect it to be.

I'm beginning to wonder if there was a bug in my version of RX8 - so it's handy you say that. If it's am RX8 bug that would explain a lot. Next job will be a trawl of RX8 bugs.

Thanks both!

[Wonks]

It’s not particularly easy to detect DC offsets unless you are inputting a very symmetrical waveform, such as a sine or sawtooth, when an offset from zero is very noticeable and you can simply use the difference in the +ve and -ve amplitudes to calculate the offset amount.

But many natural (and unnatural) waveforms aren’t symmetrical and have peaky initial transients, so it’s easy for a generic algorithm to conclude there is an offset when in reality there isn’t.

[amanise]

I do have a habit of using stereo delays with one side having double the delay time as the other......

[amanise]

Maybe RX8 is wrong...

Do you know that there is a version 8.5 which RX8 owners can upgrade to?

Just to round this one off for completeness. This is iZotope's view on the possibility of a bug causing erroneous offset reports;

"Jul 25, 2023, 16:02 GMT+2
There are no known issues with the iZotope DC offset detection.

It has not been changed in many years. It's been the same for a long time.

If RX says there is DC offset, then it's telling the truth. If you hover over the DC offset box in waveform stats, it'll tell you how many dB it is.

Usually, any DC offset -65dB or quieter is fine. You don't always have to have it set to 0. If it's reading louder than -65dB then you may start to hear issues in the low end of the file."

So - now I have a threshold to work to. That should do it!

[Hugh Robjohns]

Just checked a couple of my recordings in RX9 and the worst case I could find was 0.001% in one channel which translates to -99dB. Not going to worry about that! Most score 0.000% and translate to around -150dB.

If I run the Filter DC offset option in the dehum model the numbers fall to 0.00% and translate as -170dB or lower!

Checking the commercial Satie recording I mentioned previously — the only DC offset I've been able to find (without creating one deliberately) — scored +0.634% (and slightly less on the right channel) which translates to -44dB.

As discussed before, there really shouldn't be any significant DC offset. If you find you're getting offsets regularly, work through the signal path step by step to find out what is creating it... and then get it fixed or throw it away!