Topic: How under-recording at 24 bits can add noise

[Ozpeter]

I've deliberately used a non-specific title to the thread as the tests I've been doing using an R-44 may well be relevant with any comparable device.

It's commonly said that when recording using 24 bits, it doesn't matter if you record with peak levels well below full scale as you can normalise afterwards in your DAW with no noise penalty.

Testing this theory with the R-44 indicates that this may not always be true - maybe it's often not true.  I'd recommend testing your setup carefully before assuming that it holds true for your device.

The R-44 preamp gain has eleven steps of 6dB.  My test was to record (24 bits) at the maximum sensitivity setting (-56) using a 150 ohm resistor as the source, then record at the next setting down (-50) but add 6dB of gain in my DAW, then at -42 but add 12dB of gain in the DAW, etc etc till I reached the lowest sensitivity setting of +4dB and added 60dB of gain in the DAW.

Average RMS dB values of the relative noise floor for each gain value thus obtained are as follows, starting with the R-44 set to highest gain -

-77.81 : -77.87 : -77.46 : -77.33 : -76.40 : -74.23 : -70.64 : -59.63 : -56.37 : -52.21 : -47.39

Applying those figures to a practical example, say you decide to under-record by three steps of the R-44 gain control.  So for instance, if you recorded at the "-38" position let's suppose your recording would have just peaked to full scale.  You decide to record instead at the -20 position to give yourself 18dB of headroom for safety, then normalise in your DAW.  The noise penalty is 77.33 -70.64 = 6.69dB.  If you were a bit less conservative and went for a 12dB headroom figure, the noise penalty would be 77.33 - 74.23 = 3.1 dB.

Note that to keep things in proportion, if we look at the actual noise floor into which we are recording - taking away the gains in the DAW done as part of getting the relative figures - you would be recording (in that last example) into a noise floor of -104.23dB (Av RMS) instead of -95.33dB, then normalising by 12dB - so the noise floor becomes -104.23 + 12 = -92.23dB.

Given that the relative noise level isn't much changed from the -42 sensitivity setting - "noon" - compared with higher gain settings, in the case of the kind of classical recording I do, and using the mics that I do (Sennheiser MKH MS pair), I could probably set the preamp sensitivity and the inner variable gain knobs both to the "noon" position and never bother to look at the meters again - at worst I would be adding about 2dB of noise to any recording.  This would vary with others' mics and recording material.

So what happens when you run the R-44 at a paltry 16 bits?  Well, the comparable line of figures to the 24 bit figures above is as follows:-

-77.81 : -77.70 : -77.12 : -76.22 : -72.27 : -66.33 : -60.33 : -54.35 : -48.35 : -42.35 : -36.35

Apart from exhibiting larger differences in relative noise floor levels than at 24 bits, at 16 bits the R-44 noise floor exhibits a number of odd features including noise spikes and DC offset in some of the lower gain settings.  It would seem to run "happier" at 24 bits.   But having said that, all the concert recordings I have done with mine have been at 16 bits, and ambient noise levels even in the most (supposedly) quiet hall I've recorded in have been at significantly higher levels than the system noise, so in practice none of this may matter anyway...

Finally, it's recently been asserted here that the centre trim knob on the R-44 actually controls an analog gain stage just before the AD converter, and I thought it would be interesting to set that to the minimum and record system noise when in theory at least, only the AD converter is contributing to it.  The result is a file with all bits set to zero, which rather calls into question the statement that the centre knob is other than digital gain - or the AD converter is noise-free.  Hmmmm.

And my point is - if you have an R-44 and you use it with mics plugged in directly, you may be directly interested in the outcome of these tests.  If you have another system, you may feel motivated to check out what actually goes on at the various operating levels you could choose to use, rather than making any assumptions about the effect of different gain settings on overall noise after post-production.

[Todd R]

It's commonly said that when recording using 24 bits, it doesn't matter if you record with peak levels well below full scale as you can normalise afterwards in your DAW with no noise penalty.

But having said that, all the concert recordings I have done with mine have been at 16 bits, and ambient noise levels even in the most (supposedly) quiet hall I've recorded in have been at significantly higher levels than the system noise, so in practice none of this may matter anyway...

Thanks for posting that, lots of good information that I think will take me some time to digest.

As to recording at 24bit with conservative levels and thus having no noise penalty, I think the idea is that there is no noise penalty since in most field music recording situations, the ambient/room noise levels will become the relevant noise issue, not the recorder system noise (as you get at with the second bit I've quoted).  So the no noise penalty idea isn't really an absolute, it incorporates the idea of ambient noise vs recorder system noise.

Frankly, I'm still curious about what you get at regarding recording concerts at 16 bits.  If you run 16 bits very conservatively, with levels down 18db's, that means you're only using 13bits of the 16 bits.  13bit resolution allows 13*6 = 78db of dynamic range (assuming I'm working through this all correctly).  I really doubt you would ever find a concert recording situation where the ambient room noise is -78db or better, even in a quiet classical hall.  So it does seem that the argument about running levels conservatively at 24bits should work just as well for running conservatively at 16 bits. 

Still, I keep running at 24 bits, so what do I know.

[it-goes-to-eleven]

Good job on the tests and thanks for sharing!   I've spent quite a lot of time doing similar tests.. And it is one thing to collect your own data and draw your own conclusions and another to try and share it in a sensible way.

I don't think anyone says there won't be noise, but it is gain and recorder dependent.  And you're going to come out ahead at 24 bits.

One gotcha is the funky way in which some all-in-one recorders add gain.  It is often a combination of analog and digital, etc.  Most manufacturers are evasive about the specifics.  So tests like this can be very recorder specific.  Another way to run this test would be to leave the recorder gain constant and use an external pre-amp.   And while the noise floor is important and illustrative of some aspects, there are all those more subjective qualities - does it sound good, is it musical, does it sound thin, etc.

Another gotcha is that quite a few a/d's distort when run hot... OR when run at certain gain/trim settings.  My r09 distorts if the trim is below 8.

I Really like the a/d in the 7xx.  And while the pre-amp is very good, I tend to prefer external pre-amps.  I don't like the a/d in the v3 nearly as much as the 7xx a/d.  I wish the r09 a/d was as good as the 7xx... I do wonder how the newer small recorder a/d's compare at 24 bits with a pre-amp in front.

[cybergaloot]

I can add some real world experience. I recorded Alejandro Escovedo from the balcony of a small club at 48/24. I had my levels set for what was coming off the PA but then he left the stage and played acoustically (directly below my hypercard mics) with no PA. I didn't change my levels and when I boosted the levels for those songs in post, the noise floor came up as well. It is still listenable but the noise is very noticeable. Seems to me that you have to be in the ballpark for the the 24 bit stuff to work. Its not going to compensate for levels being below a certain point.

[Todd R]

Sorry total aside, but what's your input on Alejandro Escavedo?  He's playing a night this summer at the Denver Botanic Gardens with Rickie Lee Jones.  Tickets go onsale for the summer series in a couple of days and we need to decide what shows we want to see.  I like Rickie, but never heard of him.

[it-goes-to-eleven]

I can add some real world experience. I recorded Alejandro Escovedo from the balcony of a small club at 48/24. I had my levels set for what was coming off the PA but then he left the stage and played acoustically (directly below my hypercard mics) with no PA.

Oh yeah - you really have to be ready to tweak the gain when he jumps off stage for the encore!  I've had it where he does one song to my left (turn mics) and then moves to right (turn mics).  So it can be a mad scramble to adjust the gain at the right moment and try and get the mics turned without thumping, etc.  That is a lot easier to do with the v3 (silent gain changes) vs. aerco (noisy changes and small knobs that are harder to coordinate).

Alejandro Escovedo is one of the most talented performers I have seen.  Some of his tours are more folkish, others more rock.  He's on the archive.  He's near the top of my must see list.

[cybergaloot]

Sorry total aside, but what's your input on Alejandro Escavedo?  He's playing a night this summer at the Denver Botanic Gardens with Rickie Lee Jones.  Tickets go onsale for the summer series in a couple of days and we need to decide what shows we want to see.  I like Rickie, but never heard of him.

GO! Alejandro is a great songwriter and performer. An American classic! He plays in a few different contexts, solo or duo acoustic, with a full on rock band and as the Alejandro Escavedo Orchestra that includes a few strings (great stuff!). His style runs the gamut between sensitive acoustic singer/songwriter stuff to his punk rock roots, with some Latin influences thrown in occasionally. Even acoustic you'll get a delicate Latin instrumental one moment and a flat out rocker the next but it all fits. Alejandro was a member of the punk rock band The Nuns, a member of the cow-punk band Rank and File, and a member of the great rock band the True Believers along with one of his brothers from the group Zero. Other brothers are percussionists in the jazz world but also played in an early version of Santana. Alejandro also tells some great stories and can be funny as all hell. He'll probably play his song "Castanets" and the title won't make any sense until you hear the story that goes with it. He'll also probably play a good bit of his new CD "Real Animal" which is somewhat autobiographical. Its a good one!

Back on topic, I should have noted that at the performance I was talking about I was recording with an R-44 using Studio Project C4's with hypercard caps and also using the internal mics. I wanted to see how they sounded and it was just an opportunity to try them out. Both sets of mics had the same problem when the input level dropped and I tried to compensate in post.

[kgreener]

Back on topic, I should have noted that at the performance I was talking about I was recording with an R-44 using Studio Project C4's with hypercard caps and also using the internal mics. I wanted to see how they sounded and it was just an opportunity to try them out. Both sets of mics had the same problem when the input level dropped and I tried to compensate in post.

Pardon my ignorance, but you can do this?  Run a pair of external mics and the internal mics simultaneously?  Well color me silly if you can.  What channels would you plug the external mics into, 1 & 2, or 3 & 4?

Thanks.

[cybergaloot]

Back on topic, I should have noted that at the performance I was talking about I was recording with an R-44 using Studio Project C4's with hypercard caps and also using the internal mics. I wanted to see how they sounded and it was just an opportunity to try them out. Both sets of mics had the same problem when the input level dropped and I tried to compensate in post.

Pardon my ignorance, but you can do this?  Run a pair of external mics and the internal mics simultaneously?  Well color me silly if you can.  What channels would you plug the external mics into, 1 & 2, or 3 & 4?

Thanks.

The internals take channels 1 & 2, you plug the externals into 3 & 4. You have set the input to use both in the menus and set the recorder to stereo X 2. Actually it sounded fairly decent mixing the two recordings.

[kgreener]

very cool!  had no idea this could be done, even after reading the manual.

thanks!

+t 

[Gutbucket]

I didn't change my levels and when I boosted the levels for those songs in post, the noise floor came up as well.

In all likeliness, the noise floor you boosted on your recording was the contribution of the room and you'd notice the same noise level increase if you had instead increased the preamp gain by the same amount.

[it-goes-to-eleven]

No doubt.  Anyone who has tried mic testing at high gain knows how hard it is to isolate room noise.. Even in the basement, with acoustic foam, etc.  It seems like you can hear/see every dust particle in the air hitting the mic.

[cybergaloot]

I didn't change my levels and when I boosted the levels for those songs in post, the noise floor came up as well.

In all likeliness, the noise floor you boosted on your recording was the contribution of the room and you'd notice the same noise level increase if you had instead increased the preamp gain by the same amount.

Could be. It was a general white noise type background sound. If I ever record him from the balcony in that club again I think I'll aim two mics at the stage and two straight down then use whatever combination sounds best for each song.

[DSatz]

I've been baffled by the idea some people here have of deliberately under-recording by 10 dB or even more. Some people here talk as if it's a virtue for the highest peak level on a recording to be at 12 - 15 dB below full scale, rather than, say, at -2.

It's one thing to allow a few dB of headroom because a show might get louder than you expect; that's just common sense. And if you have no opportunity at all to check levels in advance, or if you can't watch levels during the show, I can see being even more cautious. This would be compounded even further by the use of unfamiliar equipment.

But if you keep so much dynamic range in reserve that (say) 6+ dB of it never gets used, you're simply wasting it; you should carefully raise your recording levels somewhat.

I can understand feeling proud that you predicted the levels so accurately that your peaks came in 2 dB below full scale; I developed that skill back when live recording was all analog and the equipment was all (in effect) 12-bit. But I can't understand feeling proud not to have any peaks above (say) -15 the way some people here talk. In the latter situation I would probably think, "At least I was recording 24 bits; I can bring the levels up when I transfer to 16 for the client's CD, and there won't be a huge noise penalty." Even if the client doesn't notice the noise, I know that I blew it, and the techonology covered up for me to a large extent. I would get no pleasure or self-esteem from deliberately creating that situation.

--best regards

[it-goes-to-eleven]

I've been baffled by the idea some people here have of deliberately under-recording by 10 dB or even more. Some people here talk as if it's a virtue for the highest peak level on a recording to be at 12 - 15 dB below full scale, rather than, say, at -2.

It's one thing to allow a few dB of headroom because a show might get louder than you expect; that's just common sense. And if you have no opportunity at all to check levels in advance, or if you can't watch levels during the show, I can see being even more cautious. This would be compounded even further by the use of unfamiliar equipment. But if you keep so much dynamic range in reserve that (say) 6 dB of it never gets used, you're simply wasting it; you should carefully raise your recording levels somewhat.

Disagree.  There is a factor you aren't taking into account - some gear sounds worse when driven too hot.  It distorts.

In some cases the concern is the pre-amp, in others the a/d.  The amount of headroom to reserve, for me, is gear and material specific.  I have long wondered whether the average rms levels may ultimately have more impact on sound quality than where the peak falls.

There has been a lot written on why...  Some pre-amps were designed in the pre-digital era.  As you know, 0 VU is still -20 dbFS.  Of course that doesn't explain cases where modern devices sound better when driven not quite so hard..  But sometimes we just need to follow our ears.