Flaws in Recording: DC?

[MRC01]

I've got a recording that has a quiet section that looks like this:

Looks like a DC offset for a couple of seconds. It's not audible, but I figured it could act like a DC offset which could be bad since my preamp & amp are direct coupled.

I figured a high pass filter should fix it, so I applied one with a corner freq of 10 Hz. That makes it look like this:

Yay! No more DC offset.
Now I'm wondering - no professional recording should have artifacts like this, right? Looks like they had some kind of temporary DC offset in their recording process?

 

[DonH56]

Some AVR/AVP/players mute when digital 0 is encountered; I wonder if that is what is going on?

 

[MRC01]

My DAC has a manual switch to control whether it does this. But this never passed through my DAC. This is the original raw data from the actual recording, not processed by me in any way.

 

[pjug]

You are supposed to get a pressure drop in your room and feel your ears pop. Shame if your speakers are not capable of providing this effect.

 

[PaulD]

It could well be an electronic instrument with some DC or some VLF nonsense going on. Many synthesisers will put out something close to DC. Some recording interfaces are DC coupled (often enough so they can be used for synthesis control signals), I have seen this on some MOTU interfaces, and used it to good effect.

It does look like this was missed in the production - it happens. What I find curious in the waveform display above is that the AC signal in the offset part is low, but when it's gone through the high pass filter to remove the DC, the AC signal it is a LOT higher... How did that happen? Hmmm, perhaps if the DAW is using floats or 64bit ints there may be enough resolution to do this. What was the track? How was the file generated? What is the DAW and how is it set-up? Perhaps the DAW used for the production displayed this differently so they did not notice it.

Unfortunately there is not enough info to be able to say much about this.

 

[A800]

DC is bad.
AC/DC is the good one.

 

[KSTR]

Bipolar log view of audio data makes no sense. This is Audacity, isn't it (and and really don't like it for alot of reasons)?
The DC offset is microscopic in reality, below -40dB.
Please use linear view and then it is also easier to see what happened here as indeed any mastering should include a final highpass.

 

[RayDunzl]

The offset is about -0.016V for a 2V DAC

Can't see how long it lasts in your post.

 

[LTig]

What I find curious in the waveform display above is that the AC signal in the offset part is low, but when it's gone through the high pass filter to remove the DC, the AC signal it is a LOT higher... How did that happen?

It's due to the y-axis being exponential (dB scale). With a linear y-axis you'd just see a (insignificant) shift of the signal.

 

[RayDunzl]

1. The offset is not DC, it is AC riding on a DC offset.
2. All decibels are not the same linear (voltage) size*.

Original signal
The signal attenuated by 42dB riding a -42dBfs level
And the attenuated signal without the offset
Displayed as decibel levels

Roughly speaking, and maybe not fully taking into account the logarithmic character of the swings (I'm not a mathematician):

A 0.15 volt wiggle around a 1.0 volt voltage level wiggles by 2.42dB

A 0.15 volt wiggle around a 0.3V volt voltage level wiggles by 7.04dB

a 0.15 volt wiggle around 0 volts would be something like +/- 40dB

But that is why the signal seemed to "jump up" after removing the offset.

Decibels related to the linear scale:

 

[A800]

1. The offset is not DC, it is AC riding on a DC offset.

I am glad it's AC/DC.
Thanks.

 

[Mnyb]

I think small DC offset are to expected and or unfiltered infrasonic disturbances or possibly also ultrasonic junk . So a completely DC -coupled singal path is not that desirable , it was(is ?) a fad in audiophilia ? DC coupling ultra wide bandwith and lack of protection circuits sound familiar for many brands ?

 

[MRC01]

It could well be an electronic instrument with some DC or some VLF nonsense going on. Many synthesisers will put out something close to DC. Some recording interfaces are DC coupled (often enough so they can be used for synthesis control signals), I have seen this on some MOTU interfaces, and used it to good effect. ...

It is this recording: https://play.primephonic.com/album/813543021322
Track 10, 2nd movement of the 1st piano trio, about 5:30 into the track.
So, no synthesizers.
This strange DC shift is not constant but occurs only at a few isolated spots in well over an hour of music.

 

[hege]

I was wondering the same thing, seeing many examples of (intentional?) (AC/)?DC when analyzing tracks..

Example: Mr. Fingers - Cerebral Hemispheres (2018)







Result of analog synth? Why would they leave it in mastering? I guess Mr House knows what he's doing. Need to make sure all my sub gear is DC coupled.

 

[Mnyb]

I was wondering the same thing, seeing many examples of (intentional?) (AC/)?DC when analyzing tracks..

Example: Mr. Fingers - Cerebral Hemispheres (2018)

View attachment 81321

View attachment 81320

View attachment 81322

Result of analog synth? Why would they leave it in mastering? I guess Mr House knows what he's doing. Need to make sure all my sub gear is DC coupled.

You must seal all doors and windows , so that you can feel the 0Hz even the weather must be "audible" have a look at the barometer and see that it was sunny day at the recording venue. Sub displacement must ofcourse be a rahther large fraction of the total room size , cover all walls and ceiling with drivers ?

 

[AnalogSteph]

You are supposed to get a pressure drop in your room and feel your ears pop. Shame if your speakers are not capable of providing this effect.

Humor aside:

If I have my math right, with a 0 dBFS level of 103 dB SPL (4 Pa peak) we'd be looking at about 0.03 Pa. That translates to a height difference of a whopping... 2.5 mm, or 1/10 of an inch.

Imagine a pressure difference this small would really make your ears pop. Nodding would be plenty uncomfortable. Forget about getting up, let alone any other more severe vertical movements. Elephants can afford to hear lower frequencies, there just isn't much of any jumping around going on in their life.

BTW, modern audio ADCs (from about the early 2000s onwards) generally include digital highpass filtering for DC offset removal. The earliest reference I found so far was in the datasheet for an AK5392 dated 1997/11. 116 dB(A) dynamic range and 105 dB SINAD in a time when 24 bit was novel and a maximum sample rate of 54 kHz was deemed appropriate.
(To show you how quickly things were moving back then, this part superseded the lowly 113 dB DR, 97 dB SINAD AK5391 dated 1996/11, and by early '99 it was superseded by the 96 kHz capable AK5393 famously used in the Behringer xxx2496 series years later. Mid-2002 then brought the 192 kHz capable AK5394A, which was used in top-grade ADCs for a long, long, long time and only discontinued like last year.)

I have come across devices with recording DC offset once or twice myself, but don't remember what they were. Consumer soundcards from around 2000, onboard audio or a basic USB microphone perhaps.

 

[JIW]

I have recorded the same piece from Tidal HiFi (lossless 16 bit/44.1 kHz) using Audio Hijack at 16 bit/44.1 kHz.

Inspecting the waveform using a Python program does not show the same behaviour. I have checked and Audio Hijack has no high-pass filter. It records a -40 dB negative offset correctly (see below).

Here is the full movement shown in linear scale, i.e. 1 corresponds to 0 dBFS and 0 corresponds to -infinity dBFS. The green line is 0 while the red line is a 40 dB negative offset - -0.01 FS, if you will - as suggested by your figures.

From 5:25 to 5:32 seems to be where you have the -40 dB offset. I don't.

Looking even closer between 5:29.2 and 5:29.5 shows much the same.

Note the levels. The lowest sample value is -96, i.e. -50.66 dBFS.


If you own the files and know your way around Python, I can send you the program.


16 bit/44.1 kHz recording of 5 second -40 dB offset (sample value -328) using Audio Hijack. File made in Python. Playback via QuickTime Player.

 

[Cbdb2]

I've got a recording that has a quiet section that looks like this:
View attachment 70717
Looks like a DC offset for a couple of seconds. It's not audible, but I figured it could act like a DC offset which could be bad since my preamp & amp are direct coupled.

I figured a high pass filter should fix it, so I applied one with a corner freq of 10 Hz. That makes it look like this:
View attachment 70719
Yay! No more DC offset.
Now I'm wondering - no professional recording should have artifacts like this, right? Looks like they had some kind of temporary DC offset in their recording process?

Somethigs wrong with these. How does the quiet section get loud just by removing the DC? I would expext the quiet part to be shifted up to zero but otherwise stay small. Something else was going on, like a bad conection. Try again without the filter.

 

[MRC01]

Somethigs wrong with these. How does the quiet section get loud just by removing the DC?...

It doesn't. That's an illusion caused by the non-linear Y scale of the waveform graph.

 

[Cbdb2]

Dont care what the scales are removing DC from the first file will not increase the AC level the way it lookks in the second image.