Sample rates question

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.
[quote author="bluebird"]


would it be better to use good plugins instead of outboard gear to avoid resampling? I've heard additive EQ with plug-ins is not the best idea.

And If I plan to do alot of "out of the box-back in to the box" recording should I use a higher sampling rate when recording the original material?

[/b][/quote]

That's relative to your converters, outboard gear, and plug-ins. I have yet to hear a plug-in that I like as much as my analog gear, so I do the same as you described... come out of digital, interface with outboard, and then back into digital. You gotta decide if the additional conversion stages are worth the trouble for using outboard. For me, it's not a problem at all. Pretty much a no-brainer, and I don't have the best converters by ANY stretch of the imagination (MOTU1224 DA, Tascam DM24 console AD). I'm at 24/44.1k, and feel fine about the quality of my work. Sure, better converters and a higher SR would be nice, but not immediately necessary. It's all about the big picture more so than the sum of the parts.
 
[quote author="TomWaterman"]If the audio is tracked at 44.1k then there will be no improvement to the audio by sample rate conversion, especially not the stock PT algorithms....[/quote]

Right on.

[quote author="TomWaterman"]I think the real improvement comes when running certain plugins at higher sample rates...[/quote]

Not so. The real improvement comes when running certain plugins at higher bit depths, not at higher sample rates. In fact, Most DAW's have internal processing bit depths that are far greater than your audio bit depth for this very reason. No DAW I know of will let you do this (run a plug at 192KHz on a 44.1KHz session) because it involves two extra unnecessary DSP processes that degrade the signal.

[quote author="TomWaterman"]By upsampling, the plugins can move their anti-aliasing filters right out of the orignal audible bandwidth [/quote]

Wow, wow, hold it! There's no such thing as digital antialias filters! Antialias filters are analog devices by definition. They are there to prevent frequencies over Nyquist from hitting the A/D converters, but once a signal has been digitized at a certain SR, those frequencies are already impossible to reproduce by the system. If the antialias filter missed anything over Nyquist, those frequencies (aliases) are already back in the audio spectrum and there's no getting rid of them!

A clear distinction needs to be made between sample rate and bit depth... They're two very different concepts that are often carelessly - and erroneously - treated the same way.

Peace,
Al.
 
[quote author="bluebird"]I usually take individual tracks out of the box and use my outboard gear (all DIY!) to prosses the signal and record them back in. I record in a one room practice space (no control room) so its hard to use outboard gear on the way in. I use Samplitude in 32 bit floating mode and a Lynx II card. I record at 44.1. I feel this set up is not the best out there but at the high end of whats availible for digital recording.[/quote]

Well, I hope this is a good way to work, as this is my own plan for a personal recording situation, right down to the LynxII card.

I asked about this some time ago, and many people agreed that it is a good way to work.
 
ddt - no I haven't read Bob's book (as you can see) - but I plan to get it....

alk509 - the UAD-1 Pultec EQ upsamples to 192k for its processing and then back down again, you can run it in any sample rate project you want, as such it is quite DSP hungry.....

As far as I knew this was done to remove aliasing artefacts in the process out of the audible band.....

I spent some time last year playing with FIR and IIR filter designs in MATlab and IIRC the spectrum analysis of signals through all of my filter designs illustrated a mirrored harmonic alias above nyquist, this was a digital system?? Maybe I am mistaken, but there is a definite improvement in using plugins at higher sampling rates.

Bit depth matters for sure, but it depends upon which type we are talking about I guess, floating point or fixed point.

I'm afraid I'm out of my depth here but straight from the UA webzine, now don't shoot the messenger:

Upsampling allowed a transformation to discrete-time with minimal warping of the filter characteristics. For upsampling, we chose a linear-phase filter with close to 100dB of rejection in the stopband. Using linear-phase filtering for the upsampling process allowed us to match the phase of the Pultec response throughout the audio band; the large amount of stopband rejection was necessary in order to prevent aliased signal components from degrading the system response. The tight constraints placed on phase response and anti-aliasing properties led to a tradeoff between system latency and high-frequency magnitude response. Thus, in order to maintain acceptable overall system latency, there is a 2db rolloff at 18kHz when running the plugin at 44.1kHz, and a .01db rolloff at 18kHz when running the plug-in at 48kHz. For an analog system, a comparable magnitude rolloff could be problematic, because there would be an associated phase warping which would extend down to the 2kHz range. However, since the anti-aliasing filter is linear-phase, this problem does not occur in the plugin.

Although this is a problem with the PultecEQ emulation, it is clear that the H.F roll-off problem is minimised by running the plug at higher sample rates....I know some guys here don't like UA and I know a lot think plugs sound crap compared to an original EQP but I find the plugin to sound very very good YMMV.

Cheers Tom
 
[quote author="bluebird"]And If I plan to do alot of "out of the box-back in to the box" recording should I use a higher sampling rate when recording the original material?[/quote]

Definitely! As high SR and bit depth as you possibly can. You'll hear the difference!

All this discussion about sticking to one SR throughout is only true if we assume we're digitizing once and that's it. If you know you'll be running the bass through that Gyraf Pultec, and your vocals through your Bloo LA2A at mixdown, by all means sample away at 192Khz/24 bit or whatever.

If you're recording a jazz trio and you want to keep it as clean as possible, get a good sound out of the band, a good level into the computer, record at 44.1/24bit, dither down to 16bit and be done with it. If you're doing Britney Spears (heh, no pun intended) and are using all kinds of outboard gear and hardware and software synths and stuff, sample fast and bring it all down to 44.1 at the end.

Ultimately you have to keep things in perspective: additional A/D-D/A conversions and SR changes are not processes you apply for aesthetic reasons, but out of necessity. They do degrade the quaity of a digital audio signal, so you want to avoid them whenever possible. But this degradation happens to such a small degree, that the disadvantages are almost always offset by the ability to use that vintage Fairchild compressor, or some old-school plate or chamber reverb, or even some crazy, noisy, piece-of-shit fuzz box you want to use for texture!

Peace,
Al.
 
Agreed!!!

I work this way, and the colour of your outboard will more often than not hide the effects of the conversion anyway........

Cheers Tom
 
Sweet.

Some time I will try the higher sampling rates but my system might not be able to handle it. I only have a 2.4G prossesor and 7200 IDE's. doing a session at 192K would most likely allow me about 10 tracks. I suppose not using any plugins I could get away with a bit more.

Thanks for the answers...now could someone tell me why I exist? And would my life be any better if I could somehow squeze my body through an 1176...or better yet a BBE Sonic Maximizer? :green:
 
[quote author="buttachunk"]I think we need to clear up definitions here-- sample rate conversion is not upsampling.[/quote]

OK, here's part of the problem. When I use the word "upsampling", I mean to increase the sample rate of an audio signal by means of a sample rate conversion algorithm. And I think I actually stated that somewhere up there...

[quote author="buttachunk"]Oversampling is averaging. Oversampling reads the wave multiple times and averages those numbers to assure accuracy. Averaging means to take several readings of something and to chose one value that represents the 'mean', or average value.[/quote]

Perfect. But again, oversampling has nothing to do with SR conversion and I can't see how I may have implied that?

[quote author="buttachunk"]Sample rate conversion is approximation-- it has nothing to do with averaging.[/quote]

[quote author="buttachunk"]My original statement was "the number would contain repeats since there is no real way to interpolate the non-exsistent map data" which you said was not true because the SR algorithm averages data...[/quote]

[quote author="buttachunk"]So then, let me restate my original point; there is NO REAL way to accurately interpolate non exsistent map data. As we see above, scaling completely fails to accurately account for any missing information.[/quote]

You're right, SR conversion does not "average" data in the strict mathematical sense of the word ( average of a and b = (a + b)/2 ). "Approximation", as you suggest, is a much more accurate term. Furthermore, you're right in that there is no real way of interpolating what was there before digitizing.

However, upward SR conversion algos DO NOT REPEAT SAMPLES, AND THEY DO INDEED TRY TO INTERPOLATE THE NON-EXISTING INFORMATION. To do this, they look at the level and rate of change of signal amplitude across a range of samples and that way interpolate the interim samples.

Let's look at a similar example to the one above but without using fractions to make it more easily understandable:

A 1 second sample train recorded at 4Hz sample rate looks like this

0 - 4 - 8 - 12

A 4Hz to 8Hz SR conversion algo will not do this:

0-0-4-4-8-8-12-12

Instead it will look at the rate of change of level around two contiguous samples (the wave's second derivative in the given interval) and then assumes that delta-t is sufficiently small that this number must be the same between the samples that it is analysing (or somehow comes up with a better approximation). The algorithm then uses this information to come up with the value of the extra sample it needs to create:

0 - 2 - 4 - 6 - 8 - 10 - 12

... which in this particular example, since the wave is a straight line and its second derivative in this interval is = 0, just happens to be the average between samples, but this is certainly not what usually happens (Although I'm sure there are SR algos that just find the average between samples...)

If the two SR's are not multiples of each other, the process is still similar, except that every single sample will be interpolated by the computer.

[quote author="buttachunk"]most songs on the radio are recorded to Protools, mixed on an SSL or neve, and the mix is converted back to digital. After that, Bob Ludwig and all of the other mastering engineers convert back to analog to eq and compress the material then re-convert it back to digital.[/quote]

True, but this is only done because putting your mix through a Neve or an SSL or a Manley compressor or some other cool piece of gear largely offsets the losses caused by to extra D-A-A/D conversions. Read my previous post on the matter.

Peace,
Al.
 
alk,

If I had to take sides :wink: , I think I would have to agree more with Tom Waterman, as I vaguely remember something similar in Bob Katz's book (which I haven't got at hand, unfortunately) to what Tom was saying in the post where he quotes the UA info (about aliasing).

I however agree with you comlpetely on what you're saying about using the different sample rates! :guinness: :guinness:

d.
 
Hi folks. This is a great forum - loads of dedicated people doing things that I like doing too, and have been for the last 40 years. I normally lurk, but just occasionally I don't...

[quote author="alk509"]
Wow, wow, hold it! There's no such thing as digital antialias filters! Antialias filters are analog devices by definition. They are there to prevent frequencies over Nyquist from hitting the A/D converters, but once a signal has been digitized at a certain SR, those frequencies are already impossible to reproduce by the system. If the antialias filter missed anything over Nyquist, those frequencies (aliases) are already back in the audio spectrum and there's no getting rid of them!
[/quote]
I'd certainly agree about the situation regarding anti-alias filters at the original pre-digitising point - but it's perfectly possible to implement one digitally in a resampling situation where you convert, say, a 30kHz sine wave at 192k down to 44.1k. If you don't carry out an anti-alias digitally at this point, you end up with the difference frequencies reflected back very clearly into the audio band. Any competent sample rate converter (like the one in Audition, for instance) will forcibly include a digital anti-aliasing filter within the downsampling algorithm. If you generate my example as a 16-bit integer signal in Audition's tone generator, and sample rate convert it to 44.1k, you are left with no more than 1-bit dither noise. If you carry out the same operation with a 32-bit FP sample, you still end up with dither noise - but it's at the 24-bit noise floor level. And that's done with a digital anti-alias filter!
 
[quote author="SmG"]but it's perfectly possible to implement one digitally in a resampling situation where you convert, say, a 30kHz sine wave at 192k down to 44.1k[/quote]

:shock:
I stand corrected. You're totally, absolutely right. The original question was about converting up from 44.1 to 192KHz, and this is what I was refering to.

Regarding the UA webzine article:

The key phrase in there is:

[quote author="UA webzine"]Upsampling allowed a transformation to discrete-time with minimal warping of the filter characteristics[/quote]

The aliased components the UA people talk about in their webzine would be a result of the upsampling process itself had they not used those cool linear-phase filters. They claim upsampling lets them more accurately model the original Pultec (which may well be true, since the plug has gotten rave reviews), but the whole process is very specific to the UA pultec plug-in. Read the whole thing here.

I'm curious now... Does anyone know if there's other plugins that up/down-sample like that?

Peace,
Al.
 
Hi Al!

I agree - the digital anti-aliasing filter is used to remedy the effects of upsampling......and would not be needed if it wasn't for the upsampling process. I believe there are other plugs that upsample - I think Waves LinearPhaseEQ may be another example.....maybe some of the voxengo stuff (I'm unsure about that one).

By running the whole session at a higher rate, there may be a noticeable improvement in the sound of certain plugins (I know many other have found this)......of course this would probably be pointless if you have to run all of your tracks through an SRC prior to mixing.

I think the best advice is to stick at whatever rate the audio was recorded at, unless someone knows of an absolutely awesome SRC in software form?

My lynx-2 will do SRC on digital sources but I haven't tried it - anyone who has? My rigs in storage right now...

Cheers Tom
 
[quote author="JPrisus"] The topic of PT's horrible math affecting audio is NOT open for debate IMHO. It's damn near been proven that it hurts, and this is an area where Digi proceeds to skimp constantly.[/quote]

This is of very great interest to me, and I would like to know where I can find "the proven facts" about this, and also when you talk about the digi math..where in the signal chain do you say it hurts..??? everywhere ??? or is it specific area´s of processing you find it to be bad...???

And when you say that the digi math "hurts" what do you mean by this ???

Thanks

Kind regards

Peter
 
Hey Peter!

There was a huge post in GMs forum over at PSW with Nika Aldrich covering the facts of PT math - however I can't remember if it touched on sample rates......it seemed bit rate focussed, on the internal mixing algorithms and math.

A quick search should find it...it was about 20 pages long I think....

Sorry I can't help more

Tom
 
Hold on I think that was the Dan Lavry sample rates discussion, IIRC the PT math discussion was in the old PSW forum.....I'm not sure if its online - they must have an archive.

I'll check.

It was a big discussion with fixed point (PT) and floating point (Nuendo).

Tom
 
[quote author="TomWaterman"]Hold on I think that was the Dan Lavry sample rates discussion, IIRC the PT math discussion was in the old PSW forum.....I'm not sure if its online - they must have an archive.

I'll check.

It was a big discussion with fixed point (PT) and floating point (Nuendo).

Tom[/quote]

Tom,

I seam to remember that one, and if I recall this correctly the problem was regarding the old PT mixer math. This "old" problem has been solved, (or should I say should have been taking care of if thats the case)..today there seams to be nothing more wrong with PT than with other platform of DAWs..I mean nothing is "perfect" yet imho.

So I would like to know if JPrisus could tell me as to what "new" math problems he refers to when he so utterly speaks of bad PT math, and which problems the bad math are the results in, and where in the signal chain it goes so wrong *S*

This is meant as interest and absolutely NOT a PT vs XXX or anything like that.

btw I do agree greatly with Dan Lavry, and Poul Frindle on the 96Khz isue ;-)

And I do also agree that the difference among floting point and fixed point math is today not worth much...Meaning if one takes the exsact same 24, or xxx audio racks and mix them in either of the 2 math systems
one has to be damm good to hear the difference all else beeing exsact the same.

The place where alot of works still has to be done is imho among the plugins...Among the plugins I have used the only ones that does not "fuck up audio" would be the MDW EQ, and Oxford Eq/comp, that beeing said I havent tried all out there...*S*


Kind regards

Peter
 
[quote author="Consul"]Something I found while surfing around:

http://www.jamminpower.com/PDF/48-bit%20Audio.htm

Making the case for 48-bit integer over 32-bit floating point.[/quote]

This is such an emotive topic with many customers that I talk to :)

But let me put it like this... i'd rather work in 32bit float, and then jump to 64bit float for bass end filters and their coefficients than be stuck with a slow processor working overtime to do 48bit fixed all day long.

Also, one of the reasons that the Sharc's are used in the Audio industry is that they incorporate a 40bit floating point mode... essentially using 8more bits for accuracy and the rest for dynamic range.

Forcing PT to do some good maths (i.e. pushing it into 48bit as well as running decent algo's) will give good results... but as Peter Simonsen mentioned... there's only a few third parties (such as Sony Oxford & Focusrite) that really know enough to push those DSP's -- and not blush at the higher DSP use.
 
Hi Peter!

I also agree with Dan Lavry and Paul Frindle about 96kHz.....

You are right about the bad PT math......its an old old issue as far as I'm aware. The issue was with the mix buss summing. However for those that think PT runs at 24-bit, I recall its actually 48-bit fixed point system (or is that just LE?).

IIRC fixed point, when used at large enough bit depth can actually perform more accurate calculations than 32-bit float, although its more costly in terms of DSP overhead.

Fixed point is also more suitable to small range calculations, however if run at large bit depths such as 48, 64 there are enough bits spare to facilitate many more processes before rounding errors creep in. The spare bits are padded with 0's I believe.

Floating point, such as Nuendo's 32-bit system has the advantage that its more efficient and provides massive headroom, so that large number processes (like big chunks of 24-bit audio) can be processed while minimising the rounding error.

However, the problem with floating point is that once the rounding errors occur, there is no way to prevent them escalating and increasing with the number of processes.

I've got to get me a powercore to try the Sony plugs.....thats for sure! The elements are a bargain right now.

Cheers Tom
 

Latest posts

Back
Top