Hi Benoit,

<< I'm not talking about the _spectrum_ contained in the file itself, but about the different way the FILTER sounds - and works - at 96kHz. >>

Yes, I understand this. The filters in 96kHz systems are much smoother because they operate over a much larger frequency band that extends to 48kHz. However, if there are no frequencies in the source material in that band then there is nothing for the 96kHz filters to work with. Ergo, how can there be a difference?

Greg

Greg, you can't crank up some frequency that wasn't there to begin with, right so !
But with eq's you do not just crank ONE freq, but a band of freq's, sometimes a wider band, sometimes a smaller band. The upper end of the boosted freq. band might come very close to the filter cut-off point when working in 44.1. The cut-off filter is of the brick wall type, these ones are likely to cause phase anomalies in the audible band.

I think phase coherence has at least something to do with it.

Paul.

Hi Paul,

Yes agreed but we were talking about EQ'ing material originally recorded at 44.1k, so the brick wall filters have already been applied. Importing a 44.1k file into a 96k system and EQ'ing it there should not make a difference because the 44.1k file will not contain any frequencies above 20kHz but will contain the artifacts of the brickwall filtering already applied. It maybe possible to deal with some of these artifacts but there still should not be any frequencies present above 20kHz that will take advantage of the smoother filters in a 96k system.

Greg

Hi Greg,

"It maybe possible to deal with some of these artifacts but there still should not be any frequencies present above 20kHz that will take advantage of the smoother filters in a 96k system."

Again: it's not basicaly about the frequencies the filter has to deal with, but with the different way the filter works. I've been talking about this with an dsp-developper, who gave me a very good explanation of the phenomen - but it's all in french, so give me a little time so that I can translate it in an accurate way! ;-)

Cheers
Benoit

Take your time Benoit. I'd love to read more into this ...

Thanks,

Paul.

Hi Benoit,

<< I've been talking about this with an dsp-developper, who gave me a very good explanation of the phenomen - but it's all in french, so give me a little time so that I can translate it in an accurate way! >>

I'm just going on what I know plus a bit of guess work/common sense. However, I am well aware that the intricacies of digital audio don't always follow the lines of what initially seems like common sense. I'll look forward to your translation.

Thanks,

Greg

Higher sampling frequencies can contribute to a "better sounding" result in many stages of digital sound processing, beginning with conversion itself. A part of every (D/A or A/D) converter is a steep sloped low pass filter, that has a task of cutting off all aliasing and other artifacts produced by the sampling process. As far as I understood, the actual slope of the filter is a compromise between performance and processing delay. Increased sonic performance increases delay; I think John Watkinson described it something like: " you can build a perfect lowpass filter, but won't live long enough to hear it!"
Low order harmonics produced by these filters affect the sound, just remember those first SONY 3324 machines - everyone replaced filters with Apogees and then they sounded better!
What I think could be a reason of better sound of 96K converters is the fact they move these artefacts out of hearing range, or out of signal range, as in case of resampled 44 or 48K material.
Dunno if I'm right, just a guess?
Branko

>>What I think could be a reason of better sound of 96K converters is the fact they move these artefacts out of hearing range, or out of signal range, as in case of resampled 44 or 48K material.
Dunno if I'm right, just a guess?<<

Branko, because of my limited access to gear I'll just have to try to judge from a theoretical point of view:

Your comments *do* make sense, but our discussion has focussed a little on how much sonic benefits there will be when 96 k processed signals are downsampled to 44.1 afterwards, because it has to fit on an ordinary cd.

Now Benoit, any luck so far in translating that paper ? :)

Regards,

Paul.

but our discussion has focussed a little on how much sonic benefits there will be when 96 k processed signals are downsampled to 44.1 afterwards, because it has to fit on an ordinary cd.

Until Benoit posts something from his friend's DSP paper, my theory is that the downsampling process will do the most damage to the file in that digital filtering creates a "ripple" effect around the cutoff frequency, due to overshoot, which is minimized by smoothing or windowing functions. But this is still a factor and is a characteristic of filters in general. There is no such thing as an absolute, squared off cutoff filter. Only varying degrees of smoothing/windowing, or higher-order filters are able to minimize the effect, which varies from one type of function and filter to another (Haning, Hamming; FIR, IIR, etc). It seems possible to me that the effects of EQ at 96k within the 44.1k bandwidth might sound different, or perhaps even a little more silky than a 44.1k bandwidth EQ at the same frequencies, for the reason alone that it uses more than twice the number of samples per time interval. However, the downsampling cutoff filter will affect the higher frequencies, and may, in some cases, outweigh the benefits of processing at 96k, at least in the higher frequency range up to 20k. In general downsampling does more damage than dithering, which is also a noticeable artifact-producing process. EQs introduce much more noise in the LSB range of the word than dithering does due to the effects I listed above. Dithering adds noise below the bit range's lowest end of dynamic (SQNR) range, whereas filters often create side-effects around the selected frequency, at levels above the lowest end of the dynamic (SQNR) range - esp. in the case of cutoff filters, where there is a sharp change in the transfer function. Consider that the effect of downsampling from 48k to 44.1k is noticeable enough to require very high qualtiy converters for the process - especially for mixes (i.e. no SRC algorithms sound good enough to my ears for this, and I doubt any would suffice for 96k to 44.1k conversions). I actually stay at 44.1k/24 bit through the recording process for this reason. I gain more from extra bit depth than higher sample rate. It is audible enough to be a significant consideration. Just my theoretical DSP .02.

Hello Dedric.

Sample rate conversion *is* an issue.
As I'm currently recording directly to 44.1, I don't have an awfull lot of experience in this area. I vaguely remember I once had to convert from 48 to 44.1 with Cool Edit Pro, and this sounded bad, if memory serves me well.
Now, if and when going to higher sample rates I would definately opt for 88.2 when cd releases come into play. Transferring from 88.2 to 44.1 is reported to be easier, and sure , this makes sense to me.

But then, the mix wouldn't be ready for future DVD release, Blast !

Regards,

Paul.

Paul,

SRC is the quandry with higher rates for everyone - i.e. does one plan for DVD later (when it is more prevalent), or record for CD now? Yes, Cool Edit's SRC isn't very good - in fact no software SRCs really compare to high-end sample rate conversion. If I were working towards 96k, I would probably record at 44.1, and mix for CD at 44.1, but leave my mix options open for 96k when my projects required a 24/96 mix for DVD. This seems to me to be the best middle ground for both media. The only other option is start with 96k now and invest in a really nice outboard converter. You might even do as well for the money to use a high quality audio sound card such as an RME to another RME on another computer for the SRC - D(96) to A to A to D(44.1k). Not the digital purist method, but it will sound better than any SRC algorithm - going through an analog path has a nice way of smoothing out the effects of the conversion process, as long as the analog I/O is very good (RME's are some of the best from what I hear - comparable to Apogees perhaps). And the second machine could serve for final edits, track assembly and CD burning. The third, but probably the best method would be to find a mastering house to do the conversion during the mastering process and simply bring in 24/96 mixes.

Dedric,

I was thinking about it but I'm glad you said it: Of course transferring with back-to-back converters is an option ! We might even throw in a little 2 track Reel to Reel now we're back in the analog domain ...

>>The only other option is start with 96k now and invest in a really nice outboard converter. You might even do as well for the money to use a high quality audio sound card such as an RME to another RME on another computer for the SRC - D(96) to A to A to D(44.1k).<<

This could be done inside one computer with two soundcards, perhaps even one card with a separate clock source for the 44.1 input ?
I think my humble RME / SEK'd Prodiff 96 Pro can handle this.

>>Not the digital purist method, but it will sound better than any SRC algorithm - going through an analog path has a nice way of smoothing out the effects of the conversion process, as long as the analog I/O is very good<<

Only so much for digital purists' processes !
These *theoretically* fine methods have brought us a lot of sonic trouble in the past.
I only got excited about the idea of all digital processing with the advent of 24 bit and 32 bit FP processing ...

>>(RME's are some of the best from what I hear - comparable to Apogees perhaps).<<

This what I've heard too, isn't that apogee ad8000 a little overpriced ?

>>The third, but probably the best method would be to find a mastering house to do the conversion during the mastering process and simply bring in 24/96 mixes. <<

I think no arguing necessary here, but for a home recordist with limited budget and only an occasional underground release ... :(

Paul.

This could be done inside one computer with two soundcards, perhaps even one card with a separate clock source for the 44.1 input ?
I think my humble RME / SEK'd Prodiff 96 Pro can handle this.

Humble RME? :) Sounds like a good choice for the job.

Same computer, two cards, sure. If your card is the same as the RME Digi 96 pro (sounds like it is) then you could use the one card, with separate input and output rates - good idea Paul - more economical than a two computer solution. :)

Originally the discussion was concerning a 44.1k track imported into a 96k system, EQ'ed at 96k then downsampled back to 44.1k. My contention, baring in mind the limits of my knowledge, is that EQ'ing a 44.1k track at 96k should not be an improvement over staying at 44.1k throughout.

<< "RME's are some of the best from what I hear - comparable to Apogees perhaps)."
This what I've heard too, isn't that apogee ad8000 a little overpriced ? >>

I haven't heard the RME's but I would be extremely surprised if they were comparable with the AD8000, let alone the truely high end converters. From the opinions of others I would guess that the RME's are the best of the more budget end of the market and at least comparable with many of the ADCs in the middle range of the market. With the AD8000 we aren't just talking about the good quality converter chips but also the high quality analog components as well as some of the excellent additional features like Soft Limit and UV22. If we are talking about the very top of the converter market I heard a two channel Lucid converter about 6 months ago which totally blew me away, easily the best converter I've heard. However, the price also blew me away, I think it was about $12k!

Greg

Originally the discussion was concerning a 44.1k track imported into a 96k system, EQ'ed at 96k then downsampled back to 44.1k. My contention, baring in mind the limits of my knowledge, is that EQ'ing a 44.1k track at 96k should not be an improvement over staying at 44.1k throughout.

I think that the only difference in my conclusion was that it *might* be different since you would be covering more samples per time interval, *BUT* since it would be downsampled back to 44.1, you effectively only have a possible slight variation in the EQ at any given point, or in the general behavior of the EQ algorithm, which may or may not actually sound any better than staying at 44.1. The biggest effect would be in the downsampling filtering EQ - as that is where the most destruction occurs. So I would have to agree for the most part with your conclusion Greg. The advantage of upsampling to EQ at a higher rate would be outweighed by the negative effects of downsampling back to 44.1.

I haven't heard the RME's but I would be extremely surprised if they were comparable with the AD8000, let alone the truely high end converters. From the opinions of others I would guess that the RME's are the best of the more budget end of the market and at least comparable with many of the ADCs in the middle range of the market.

I haven't heard them either, and should have qualified my comment (rather misleading wasn't it?) that the users I've heard comments from probably compared them to lower end/mid-priced Apogees - so, no, I too doubt they compete with AD8000's. Greg - have you heard other Lucid converters? I have seen several models advertised and on their website, but none in the $12k range. I have wondered how their converters compared.

going on a limb here...but are you sure you don't mean $1,200 not 12K for 2 channels of Lucid converters. Unless they've got a pair that they're not showing on their site...
And yes, i've heard great things about BOTH the new RME converters as well as the Lucid.

<< are you sure you don't mean $1,200 not 12K for 2 channels of Lucid converters. >>

To be honest the sound quality and the price were the only two things that stuck in my mind. It was all a bit incidental once I'd heard the price as it is outside my budget. I was sure it was made by Lucid although it could have been a Prism or even another make.

Greg

I have two RME 96k converters. I think they sound very good. Smooth but not too dark, with good ambience retention. For the money, they sound amazing., but I don't think you'll mistake the sound for a multi-thousand dollar audiophile box. Still, they do sound good, the only real problem for me is a relative lack of input headroom. +19dbu when most of the stuff I plug into is well over +20, which means I gotta be real careful about levels.

I've always thought that the goal is to get what you're hearing in the speakers to the end format. In my case that almost always includes an analog compressor and eq, which obviously need to be included. The only way to do that is to re-record the analog monitor signal, or console output.

So let's say you're working at 24/96 and like the sound a lot. Why not simply record what you're hearing to a CD recorder using a top-quality 16/44.1 a/d converter?

There is such a thing - the old Manley a/d's sound really good to me, better than some of the 24-bit stuff, and WAY better than the old Apogee 500 and 1000 stuff. Use that to get it on a CD-R, then import the audio files into a new, 44.1k DAW session to sequence/edit, and burn copies.

That's kind of where I'm heading with this stuff - run the multitrack session at 24-32/96, fold the analog signal back to the computer's a/d for archival purposes, and to a top-quality 16-bit a/d converter on a CD recorder for the outside world.

Yes agreed but we were talking about EQ'ing material originally recorded at 44.1k, so the brick wall filters have already been applied. Importing a 44.1k file into a 96k system and EQ'ing it there should not make a difference because the 44.1k file will not contain any frequencies above 20kHz but will contain the artifacts of the brickwall filtering already applied. It maybe possible to deal with some of these artifacts but there still should not be any frequencies present above 20kHz that will take advantage of the smoother filters in a 96k system.

I'm definitely not an expert in digital audio, but my understanding is that the artifacts of steep filtering haven't existed in a long while due to oversampling converters, and digital filtering.

Old a/d converters employed a very steep analog filter, but oversampling allows for a much more shallow filter curve. Also, the filtering is new done in the digital domain. Between the two, the brickwall filtering problem is basically solved - if the digital thing in question still sounds like shit, as many do, that isn't why.

But there still isn't any content above 22k or 24k when using 44.1 or 48k a/d sampling rates. However, if you add any processing that stirs up harmonics, like some of the tube, tape, and analog eq emulationplug-ins there might be. You boost 15k and some 30k and 45k harmonics might be generated. If the destination bitstream is limited to 24k those harmonics are history, but if the bitstream can hold information up to 48k, they get passed along to the next process, like summing, or the d/a.

My guess is that's why a 96k digital eq generally sounds better, even on a 48k audio file. I've noticed it too, but that's the only explanation I can think of for it.

>>I'm definitely not an expert in digital audio, but my understanding is that the artifacts of steep filtering haven't existed in a long while due to oversampling converters, and digital filtering.<<

But this has little to do with the downsampling process when going from 96 k to 44.1 k when you want it to fit on a cd.
The filter in the cd player does something else , to merely keep it's own sample rate related noise out of reach, cos' these frequencies will foldback into the audible spectrum, due to aliasing.

>>Old a/d converters employed a very steep analog filter, but oversampling allows for a much more shallow filter curve. Also, the filtering is new done in the digital domain. Between the two, the brickwall filtering problem is basically solved - if the digital thing in question still sounds like shit, as many do, that isn't why.<<

But this is still of concern only to the playback side of things, in consumer's homes.

All you said about it is probably right, but our discussion has steered towards the drawbacks of sample rate conversion ;) .

Paul.

Hello all ...

is this thread still alive ?

Benoit, any luck so far in translating that French paper ? :)

Paul.

Hi Paul,

I'm back after a wild working time.
I hope that the translation will come this week! ;-)
This guy has an interesting theory btw: he is currently developping an eq going up to 40kHz. He thinks that there is a mathematical interest in doing that, and that there are good chances that it could be a benefit sonically too - the problem not beeing able to hear what happens above 20kHz, but hear indirectly that something
changed below 20KHz after a change at 30kHz for ex. Modifying gain at 25,30 or 35KHz could improve dither, quantization, generate
a better error of calculation, etc.

Cheers
Benoit

O.K. Benoit,

I'll be patient ... :)

Paul.

Hi Paul,

Sorry, I didn't have a minute... :-(
In the meanwhile, maybe the beginning of an answer:
http://www.sonyplugins.com/

Cheers
Benoit

Hello Benoit,
thanks for the link.

This Sony website nicely explains how plug in designers are returning to classic analog style EQ characteristics, instead of the previously applied "surgical precision" early digital styles.

They (Sony) even stipulated the Hi freq problems in digital systems because of the limited bandwidth of a certain sampling frequency. They stated that they have not gone the route of up- and downsampling, I'm curious to learn how they managed to tackle the problem without that. Unfortunately, they website doesn't tell how. Have you got any idea how they did it ?

Paul.

Just bumpin' this topic up, and wondering what happenned with that translation?
Branko

Hi Branko.

I think that translation is still on it's way .... :) I think Benoit is way to busy, good for him !
Anyway, I think he'll get a mail that there is again some activity in this thread.

Paul.

Another month has passed.... still no translation.... I'm waiting.... as many of us are waiting for that translation.
Branko

Hey Branko,

good old Benoit better return as a rich engineer, otherwise we''ll accuse him of hanging around in bars too much :)

Paul.

I hate to be the one that's complaining all the time, but I'm waiting for that translation.... desperately! I keep on mixing in my DAW (PT), without any external mixer, without knowing if I'm wrong or right!

Hi there Branko !

I would have thought this topic would have long gone dead ... :)
Patience my friend . As long as you're able to deliver the goodies in the meantime ...

Regards,

Paul.