Rockbox Technical Forums

Support and General Use => Recording => Topic started by: mmdats on January 12, 2007, 01:02:12 AM

Title: 20 BIT ??
Post by: mmdats on January 12, 2007, 01:02:12 AM
Would it be possible, or would you have to invent a new file format?
Title: Re: 20 BIT ??
Post by: Llorean on January 12, 2007, 01:10:31 AM
There are already file formats that would support it, but it would require a very significant change to the recording system compared to how it's done right now.
Title: Re: 20 BIT ??
Post by: mmdats on January 12, 2007, 03:41:30 AM
Well that's no good.  

I think I speak for a lot of tapers out there when I say that if you guys could come up with a way to get those four extra bits out of the ADC, this software would become invaluable.  Sales of the iriver would skyrocket-

I would pay good money for a patch that would allow 20bit recording on an h100.  
Title: Re: 20 BIT ??
Post by: mmdats on January 12, 2007, 05:29:11 PM
PETETION FOR 20 BIT!
Title: Re: 20 BIT ??
Post by: preglow on January 12, 2007, 07:35:52 PM
I think you can rest assured that the ADC in the Irivers isn't really high quality enough for the last four bits to contain anything but noise. If I remember correctly, the noise floor of the ADC doesn't even really justify the full sixteen bits we have now completely.
Title: Re: 20 BIT ??
Post by: dwonk on January 13, 2007, 04:47:44 AM
I think mmdats was refering to feeding a 20 bit signal via SPDIF optical.  Alot of concert tapers that use the H120 are using outboard A/Ds like the SONY SBM-1, Grace Lunatec V3, Apogee Mini-Me and Denecke AD20.  A few of these A/Ds are capable of outputs of 20 bit word lengths.  The extra bits would improve detail as well as improve the overall dynamic range (which may be irrelevant considering the material being recorded and self noise of the devices).
Title: Re: 20 BIT ??
Post by: mlind on January 13, 2007, 04:17:30 PM
Amen to that!
Title: Re: 20 BIT ??
Post by: mmdats on January 13, 2007, 09:48:34 PM
I feel like if this is possible then someone should try and do it.  I will do all sorts of testing on it,  I just can't write the code.  

A recording upgrade like this would be very significant among tapers like myself, who run a 24/96 signal to the iRiver to be truncated without the proper UV22HR dithering down to 16 bit.  

Please help us out with the 20bit recording,

Thank you
Title: Re: 20 BIT ??
Post by: saratoga on January 13, 2007, 09:59:03 PM
I don't think dither really matters when your noise floor is at 60dBA.  Since the bottom 30dB is just noise anyway, you're not going to notice any dithering applied.
Title: Re: 20 BIT ??
Post by: Davide-NYC on January 13, 2007, 11:51:16 PM
While full 20 bit sounds great (pun intended) in theory, I'm not sure what actual aural benefit (if any) we would experience. I'm probably wrong but AFAIK it took MikeS some herculean effort to get the big SWCODEC recording update done a couple months back.

Not to discourage features but I'm not certain proper rigorous testing has been done (or at least reported) on the H1x0 units with the current Rockbox code. I know I haven't been doing any testing since the big update.

Basically I'm saying if it's a choice between features and stability (for recording) I always will choose stability.  Not to thread hijack, but how are people finding the current code?  Is it satisfactory?  Any problems?
Title: Re: 20 BIT ??
Post by: mmdats on January 14, 2007, 03:59:01 AM
I use it to tape concerts with the optical input and can say I've had success.

Here is the exact lineage.  The mod done on the UA-5 preamp allows for zero noise at half gain, it's insane really, the self noise is low enough that I believe there would be something to gain if I could capture some of the extra 20khz that the mics have to offer.

Beyerdynamic MC930's > Kind Kable > BM2p+ Mod UA-5> SP glass optical > iRiver [rockBox]

 A 22-20 bit tape is about the ideal size/freq for the taper anyhow.   I'm just saying, if it could be done I would sacrifice the stability and just patch out to avoid a mishap at a show.  For a taper it would be a dream come true, well for this taper anyway.

Title: Re: 20 BIT ??
Post by: dwonk on January 14, 2007, 07:23:23 AM
David,

I have been using an old build for quite some time (Daliy build 2007-07-17).  I have not used the current MikeS reworked builds at all for recording.  I am curious to know how many people have switched to the newer SWCODEC rebuilds and have success.
Title: Re: 20 BIT ??
Post by: preglow on January 14, 2007, 08:37:45 AM
I think mmdats was refering to feeding a 20 bit signal via SPDIF optical.
Then I wonder why he clearly specified he was talking about the ADC.
Title: Re: 20 BIT ??
Post by: Davide-NYC on January 14, 2007, 10:30:09 AM
David,

I have been using an old build for quite some time (Daliy build 2007-07-17).  I have not used the current MikeS reworked builds at all for recording.  I am curious to know how many people have switched to the newer SWCODEC rebuilds and have success.

I have done many recording with ABSOLUTELY NO PROBLEMS but I do not consider this to be linear, organized, rigorous testing.  Hours and hours of audio without a hitch so far.

I suggest everyone update to the new rec system and report back. The feature set is very cool. I (for one) would like to hear about any glitches though I'd suspect there are none.
Title: Re: 20 BIT ??
Post by: whatboutbob on January 14, 2007, 04:38:10 PM
I think mmdats was refering to feeding a 20 bit signal via SPDIF optical.
Then I wonder why he clearly specified he was talking about the ADC.

I imagine due to slight confusion. I can confirm mmdats is refering to spdif - i run the exact same preamp he does, so using the iriver as a bit-bucket I'd be very interested to see if the extra 4 bits would make a difference....though I understand it is by no means a small change.

Also, saratoga, I might be misinterpreting your post, and it has been a very long time since I tested...but i believe, particularly running spdif-in, the iriver noisefloor is significantly better than 60dB.

By the by, this may or may not be on topic, but below is a post from over at taperssection about level setting with 24 bits - it is lengthy but kinda outlines the virtues of the extra bits:

=================================================================
if you record with a 24 bit word, the noise floor is so low that setting levels that peak well below full scale is fine, still way above the noise floor.

Each bit you add to the word doubles the available values the word can represent, and therefore doubles the dynamic range (signal to noise ratio from full scale down to noise) that you can record.

A doubling of dynamic range equates to 6db. Therefore, each bit in the word contributes 6db of dynamic range. A 16 bit word therefore has a 96db signal to noise ratio, and 24 bit word can express 144db of signal.

In the real world, the audio electronics in the converter provide a higher noise floor than a 24 bit word can represnt, so a good 24 bit converter will give, lets say conservatively, 110db of signal to noise.

This means that if you record your audio with peaks no higher than -14db under Full Scale, you'll still be experiencing a recording with 96db of dynamic range, which is the best any 16 bit CD has every accomplished.


To make the point even more graphically - this all assumed that the source signal has a dynamic range in excess of 96db too. I would bet you a beer that it isn't even close. There's no tube mic that operates that cleanly. Your studio room has noise higher than that. All your hardware compressors and EQs operate with a much higher noise floor.

If you were very careful, and ended up having a source with 70db of dynamic range (congratulations!) you could record it with peaks at -26dbFS (-26 under full scale) and still have preserved every ounce of dynamic range.

So its obvious that hitting full scale isn't necessary at all - why not preserve some headroom just in case?  Let's say you do make it just under full scale. No harm in doing that if you don't go over, right?

Well, what do you do when you want to EQ something +2db? Where does that 2db go? Into clipping of course, unless you lower the input level of the plug in, which is going to lose any hypothetical S/N benefit you had preserved anyway.

Even more importantly, when you record this hot, I've got to ask - what did you do to your preamps, and analog chain to get this level? Most converters are set so that 0dbVU = -18dbFS.

That means that if you're getting -6 below full scale on your converter, that you're +12db over the 0VU point! Many analog electronics can crap out here, but almost all will sound different at least. Some times it may be "better" but usually, its a small accumulation of distortion that builds into a waxy fog that makes people blame "digital recording" for its pristine playback of their slightly distorted, but "pretty on the meter," tracks.

If you record with levels around your 0 point, some thing like -18dbFS or -14dbFS, depending on how your converter is calibrated, you'll have your analog electronics in their sweet spot, headroom for plug ins and summing, an appropriate analog friendly level if you use analog inserts later in the process, and on a modern 24bit converter with 110db S/N, the ability to faithfully record signals with a dynamic range of over 90db.

And by all means, 0dbVu is no glass ceiling like 0dbFS is. Keeping levels around 0dbVU doesn't mean that peaks won't exceed that by 6 (or more) db. If they do, your ability to record 96db of S/N (if you even have it in the source, and you don't), just like the best CD you ever heard will be preserved, if peaks don't exceed -12dbFS! More if they do.


) A 24 bit PCM word can express a theoretical limit of 144 db of S/N.

2) The analog electronics in the converter limit the performance to a functional 100 db of S/N. (slightly more in some cases, but I'll use a conservative figure and make the point even without those extra 6 db)

3) As long as the noise floor in any recording system is lower than the noise floor in the signal you're recording, you will record the full dynamic range perfectly.

4) No source you've ever recorded had a signal to noise ratio higher than 80 db, and most would be much much lower. Lynn suggests that he RARELY sees the source's noise floor lower than 70 db down, and even then, rarely. Assuming that his peaks are not at full scale, his typical source S/N ratio must be in the 50-60 db range?

This means that if you record your (best ever) 80db S/N source into a converter so that the highest peak just reaches -19 dbFS (below full scale) on the meter, that the noise floor in your signal will be louder than the noise floor in the converter. You needn't record it any hotter than that.

In the real world, you could get away with peaks around -28 dbFS, and be PERFECT. Any higher than that is totally unnecessary.

Conclusion: There is absolutely NO benefit to tracking hot.

But does it hurt to do it? Read on...

1) Your microphone preamp is set to perform best (gritty distorted choices aside) peaking around 0dbVU. This is where you'd have it set if you were recording to analog tape, hitting 0 on the VU meter. Plug that same source into most converters, and you get peaks around -20dbFS to -14dbFS, depending on how the converter is setup.

The scientists who developed this system understood the situation, even if the guys who wrote the digidesign manual don't! They EXPECT you to record with peaks around 0VU (-18dbFS on the digital scale). They KNOW about the signal to noise deal I explained earlier. That's why they chose to put the nominal level so "low" on the meter.

When you record hotter, with peaks at -6dbFS, lets say. You're driving your mic preamp 12 db hotter than you did yesterday in the analog world! That's going to add a subtle layer of distortion to your project. And they say analog sounds so much better than digital - maybe its because most people use their analog gear incorrectly when recording to digital. Maybe the "problem with Pro Tools summing" is really the effect of tracking too hot?

I've heard people say "My Neves can handle outputs +24db according to the spec, so what's the big deal?" My Neve 1073s are great sounding workhorses. They are rated for a LOT of gain. Still, they definitely sound very different even at +12. Very different. Maybe a good choice in some cases, but not the norm.

2) If you have a peak at -2dbFS, and you try to boost a mid range frequency +3db on an equalizer, you're going to clip.

Another unintended detriment to tracking hot is that you no longer have any headroom in your plug ins! It is true that in Pro Tools, you can recover lost headroom in the mix bus by lowering the master fader. This isn't true in an analog console, where the distortion has happened in a summing amp "upstream" on the master fader. In that case, the master fader only lowers the volume of the distorted signal, which remains distorted.

In Pro Tools, the master fader is actually a co-efficient with each individual fader before summing. This means that if you're clipping the mix bus, you can pull the master fader down, and fix it. Great. But what about the plug ins across each fader? They aren't affected by the master fader (thank god, or your compression levels would change etc) but neither are they protcted by the master fader. If you're clipping the mix bus, and have your master fader at values lower than unity, then odds are that you're clipping some plug ins too.

3) Most analog gear doesn't like inputs that are 12db and more over 0, even if the spec says they can take it. If you track hot, you're causing a nightmare for analog gear that you may choose to insert during the mix. Keep your levels around 0dbVU, and you can leave the digital domain freely without adding more sonic grunge.

Conclusion: Tracking hotter than 0dbVU can easily cause distortion in any number of places in the chain.

So, to reiterate:

1) There is absolutely NO benefit to tracking hot.

2) Tracking hotter than 0dbVU can easily cause distortion in any number of places in the chain.

If you want to hear the result of tracking too hot, and what it does to Pro Tools, listen to any Lenny Kravitz record. believe me, he uses all the best vintage gear, with gobs of headroom etc. There is no shortage of Neve, Helios, Fairchild, Neumann, Telefunken or whatever on his sessions. The sound of those records is entirely due to the tracking and mixing levels.

"But how do I get my product hot?"

There is a point to having a final mix that peaks at -0.1dbFS. if you are going to have a 16 bit version, if you want to be commercially competitive, if you like to see all the lights light up - sure, I do it every time. The point is i bump it up LAST in plug ins across the master fader. That way, the mix is all properly gain staged, with lots of headroom right up until the last thing juncture. Then if I raise the result to just below clipping after having the benefit of proper levels all the way through, everything is beautiful.

If you are a non believer, try it. The amount of air, detail and image is astonishing. In fact, eventually you may find that Pro Tools is actually TOO CLEAN and transparent! Then you'll start introducing purposeful distortion in your mix - distortion that YOU control at the mix is a very different animal than the unwitting accumulation of crud that comes from tracking too hot all along.

So all this means is that the noise is SO low in a good modern 24 bit converter that you can keep gobs of headroom for proper interfacing with analog gear, and still get the full 96+db of dynamic range, just 12 to 18 db lower on the meter. Your analog gear will thank you too.

So all of this results in a pristine, beatiful, airy, detailed 24 bit mix with peaks around -12dbFS? Cool! .[/i]
Title: Re: 20 BIT ??
Post by: saratoga on January 14, 2007, 07:47:28 PM

Also, saratoga, I might be misinterpreting your post, and it has been a very long time since I tested...but i believe, particularly running spdif-in, the iriver noisefloor is significantly better than 60dB.

60dBA == accoustic noise, not electrical noise.  At a concert, you will be limited by the noise of the event, not the electronics (and certainly not the software).

By the by, this may or may not be on topic, but below is a post from over at taperssection about level setting with 24 bits - it is lengthy but kinda outlines the virtues of the extra bits

Actually he concludes that its useless, and I agree with him.

That said, hes a little off on one of the arguments in favor of higher bits:

Quote
Well, what do you do when you want to EQ something +2db? Where does that 2db go? Into clipping of course, unless you lower the input level of the plug in, which is going to lose any hypothetical S/N benefit you had preserved anyway.

Any good EQ normalizes downward, not upward.  So if you want to EQ to +2dB, the entire track gets lowered by 2dB, and then the EQ is applied. This way you don't clip.

Title: Re: 20 BIT ??
Post by: mlind on January 15, 2007, 11:51:41 AM
60dBA == accoustic noise, not electrical noise.  At a concert, you will be limited by the noise of the event, not the electronics (and certainly not the software).
So you want that acoustic noise to be recorded and stored by 4-5 bits?

And not all of us use Rockbox to record music without any dynamics!
60 dB? Not the things that I usually record.

One of the most notable advantages with higher resolution in recordings is the way reverbtails sound when they fade out into the noise from the acoustic environment or analog recording equipment, as opposed to the digital noise of having too few bits for your recording.
Quote
Any good EQ normalizes downward, not upward.  So if you want to EQ to +2dB, the entire track gets lowered by 2dB, and then the EQ is applied. This way you don't clip.
That's not "any eq".
I want to control any aspect of the eq myself.
Title: Re: 20 BIT ??
Post by: saratoga on January 15, 2007, 12:19:56 PM
60dBA == accoustic noise, not electrical noise.  At a concert, you will be limited by the noise of the event, not the electronics (and certainly not the software).
So you want that acoustic noise to be recorded and stored by 4-5 bits?

And not all of us use Rockbox to record music without any dynamics!
60 dB? Not the things that I usually record.

You appear to have not read anything I wrote.  Please reread my original post about dither, rather then assuming I'm talking about dynamics or whatever it is you're posting about.

One of the most notable advantages with higher resolution in recordings is the way reverbtails sound when they fade out into the noise from the acoustic environment or analog recording equipment, as opposed to the digital noise of having too few bits for your recording.

Yes but as I and the previous poster already explained, this is not an issue here.  Please read the thread.

Quote
Any good EQ normalizes downward, not upward.  So if you want to EQ to +2dB, the entire track gets lowered by 2dB, and then the EQ is applied. This way you don't clip.
That's not "any eq".
I want to control any aspect of the eq myself.


I don't even know what this is supposed to mean.  The ability of EQ to avoid introducing clipping is essential to actually using EQ.  This has nothing to do with "control".
Title: Re: 20 BIT ??
Post by: jhMikeS on January 15, 2007, 11:08:00 PM
It probably wouldn't be too hard to introduce other depths into recording. I kind of had it in mind while implementing the updates but don't see any real advantage with the S/N ratios in the chips hardly making 16 bits. I doubt it would bring about stability issues. The real tricky part of it all was being sure I covered all the possible cases in delaying file creation.

The pcm formats will need 24 bits/sample since a sample must be an even byte size. AIFF needs them left-justified and WAV I don't know atm.
Title: Re: 20 BIT ??
Post by: mmdats on January 15, 2007, 11:23:33 PM
Please let me try to record at whatever bitrate you can give me.  If it says 24 and is actually truncated to 20 bit I would be happy anyway.

Luke
Title: Re: 20 BIT ??
Post by: jhMikeS on January 15, 2007, 11:44:00 PM
Surely at some point I'll confirm its possibility for starters but preglow's got me on this waveform synthesizer trip now and I'm trying to get my math head back and have pages of equations scribbled all over the place. Haven't been around at all in IRC as a result since it's got me totally hooked. ;D I must focus but am doing the forum rounds now to catch up on issues in software I had a hand in. If all goes well with that a better SID codec might be in the works too. (I know, off topic).
Title: Re: 20 BIT ??
Post by: mmdats on January 17, 2007, 07:02:32 PM
Ive been talking to fellow tapers about this and this is what they think:


http://taperssection.com/index.php/topic,78249.0.html
Title: Re: 20 BIT ??
Post by: xlarge on January 18, 2007, 05:59:08 AM
Ive been talking to fellow tapers about this and this is what they think:


http://taperssection.com/index.php/topic,78249.0.html


I think shaggy's got a valid point;

"Please, might I remind anyone thinking of bugging the RB developers...DON'T BOTHER THEM WITH THIS REQUEST.  Just let them take their time and do it.  THese guys are the only ones that know the hardware and can write code for RB..."
Title: Re: 20 BIT ??
Post by: mlind on August 11, 2007, 03:54:09 AM
Despite the last previous post:

(moved from another topic)
They were just a couple experimental bits preglow and I made up to prove things out - flanger and ring mod.
jhMikeS: Sorry for possibly being a pain in the...  but have you had any further thoughts about 20 bit recording for iRiver?
I remember you some months ago saying something about it being not too far fetched...

Surely at some point I'll confirm its possibility for starters but preglow's got me on this waveform synthesizer trip now...(snip)
Title: Re: 20 BIT ??
Post by: jhMikeS on August 13, 2007, 03:25:53 PM
mlind:
Not too much really. It's probably not even worth the trouble unless S/PDIF can actually send more than 16 bits/sample. Using it with the ADC would be pretty much pointless.
Title: Re: 20 BIT ??
Post by: mlind on August 14, 2007, 04:16:51 AM
mlind:
Not too much really. It's probably not even worth the trouble unless S/PDIF can actually send more than 16 bits/sample. Using it with the ADC would be pretty much pointless.

But using it with s/pdif IS the WHOLE POINT.

http://en.wikipedia.org/wiki/Spdif says max resolution of S/PDIF is "20 bits (24 bits optional)".
Further reading here: http://www.epanorama.net/documents/audio/spdif.html

Example of a Micpre+AD that would match a 20 bit S/PDIF recording device:
"Denecke AD20"
http://www.denecke.com/Products/Audio%20Accessories/audioaccessories.htm#AD20

I can't find specs for the S/PDIF input of the H1x0 on the wiki, at least before coffe...
But someone can probably tell us that it can handle at least up to 20 bits.

And making Rockbox able to handle more bits might possibly open for more interesting ports in the future. Am I wrong?
Title: Re: 20 BIT ??
Post by: Llorean on August 14, 2007, 10:27:19 AM
It's more "More interesting future ports might inspire Rockbox developers to make things handle more bits." If something's going to take a lot of time and effort that could go into improving something more useful to everyone else (as opposed to only useful to a very few people who have 20-bit inputs for S/PDIF) then often enough people will work on the thing where the work will be most useful (either to them, or to others).

You have to remember, people work on what interests THEM, not YOU. And usually this means things more globally useful. Meanwhile, you're also more than welcome to put in the work, as nobody would object to a properly formed patch I assume.