I mean...
alvarius, I don't know the size of your recod collection, but it seams to me that it would be more echonomical that you buy a CD or DVD player with upsampling ability.
Anyway, filling the data with "0" to make 24 bits out of 16 and upsampling to a non-integer multiple is not my cup of tea.
Marketing.
You can't get better sound by doing this, maby just different.
alvarius, I don't know the size of your recod collection, but it seams to me that it would be more echonomical that you buy a CD or DVD player with upsampling ability.
Anyway, filling the data with "0" to make 24 bits out of 16 and upsampling to a non-integer multiple is not my cup of tea.
Marketing.
You can't get better sound by doing this, maby just different.
Re: I mean...
Well, unfortunately the practical implementation details are such that it is entirely possible that you're wrong.
Unless you are using one of the throwback non-oversampling DAC designs out there, the truth is that upsampling/oversampling *WILL* happen somewhere in your digital chain. Granted, this is usually an integer-ratio which is easier, but the quality of the filters used for up/oversampling still does vary from one approach to another. Software oversampling using long filters and high-precision floating point math is pretty much as good as it is possible to get, particularly with non-integer ratios. Using a software-upsampled signal of this quality with a DAC that has a poor/marginal upsampling filter can indeed result in *better* sound, not just different.
Of course, if the upsampling filter in the DAC is good enough, then you're right - it won't make any difference. I'm no expert on the quality of the filters in popular DAC chips, but I'm under the impression that they typically use pretty short filters, so I don't think it's too much of a stretch to think that they can be improved upon. Certainly worth the relatively minor trouble to experiment IMHO.
carlosfm said:alvarius, I don't know the size of your recod collection, but it seams to me that it would be more echonomical that you buy a CD or DVD player with upsampling ability.
Anyway, filling the data with "0" to make 24 bits out of 16 and upsampling to a non-integer multiple is not my cup of tea.
Marketing.
You can't get better sound by doing this, maby just different.
Well, unfortunately the practical implementation details are such that it is entirely possible that you're wrong.
Unless you are using one of the throwback non-oversampling DAC designs out there, the truth is that upsampling/oversampling *WILL* happen somewhere in your digital chain. Granted, this is usually an integer-ratio which is easier, but the quality of the filters used for up/oversampling still does vary from one approach to another. Software oversampling using long filters and high-precision floating point math is pretty much as good as it is possible to get, particularly with non-integer ratios. Using a software-upsampled signal of this quality with a DAC that has a poor/marginal upsampling filter can indeed result in *better* sound, not just different.
Of course, if the upsampling filter in the DAC is good enough, then you're right - it won't make any difference. I'm no expert on the quality of the filters in popular DAC chips, but I'm under the impression that they typically use pretty short filters, so I don't think it's too much of a stretch to think that they can be improved upon. Certainly worth the relatively minor trouble to experiment IMHO.
alvaius said:
From what I can determine, Shibatch (shibatch.sourceforge.net) is pretty much as good as it gets in terms of analytically accurate sample rate conversion. Probably worth looking at this before you get too far into things.
If what you're looking to do is 'spectral recreation' - ie re-synthesizing the supra-20kHz components that 'might have been there' before sampling at 44.1, I can't help, but it is certainly an interesting problem.
The issue with buying a CD\DVD player with upsampling capability (which they virtually all inherently do with the exception of the rare non-os ones), is that I have no control over the upsampling.
This gives me a very easy, very low cost (especially in terms of my time), way of playing around with some basic upsampling within the limits of the 192KHz output frequency. Not to mention since I am doing the processing off-line, I have nearly unlimited processing power. At a minimum, I have a 3GHz P4 which is a pretty powerful digital signal processing beast. If I can get really good algorithms to work in real time, I may just consider embedding that in a player myself with a CD\DVD player drive, a small SBC, and my own custom DAC. It may not be as elegant as a custom DSP board, and no doubt in volumn not as cheap, however, in terms of my time, likely the quickest way to get it done.
.... and before anyone starts talking about jitter in a PC environment, I am not really that concerned. The DAC could be completely isolated from the rest of the system, heck, it could even be in a different rack. It would have its own local high quality oscillator and all data coming from the PC would be buffered. It would simply send a signal back to the PC when it was running low on data. There is no reason why it could not even be USB2.0 based (for data transfer only)..... hmmmm... which gives me yet another idea.
You are quite correct in that upsampling does not add any information, however, it certainly does make anti-alias filtering easier as well as not add as much quantization noise as a 16 bit DAC would.
This gives me a very easy, very low cost (especially in terms of my time), way of playing around with some basic upsampling within the limits of the 192KHz output frequency. Not to mention since I am doing the processing off-line, I have nearly unlimited processing power. At a minimum, I have a 3GHz P4 which is a pretty powerful digital signal processing beast. If I can get really good algorithms to work in real time, I may just consider embedding that in a player myself with a CD\DVD player drive, a small SBC, and my own custom DAC. It may not be as elegant as a custom DSP board, and no doubt in volumn not as cheap, however, in terms of my time, likely the quickest way to get it done.
.... and before anyone starts talking about jitter in a PC environment, I am not really that concerned. The DAC could be completely isolated from the rest of the system, heck, it could even be in a different rack. It would have its own local high quality oscillator and all data coming from the PC would be buffered. It would simply send a signal back to the PC when it was running low on data. There is no reason why it could not even be USB2.0 based (for data transfer only)..... hmmmm... which gives me yet another idea.
You are quite correct in that upsampling does not add any information, however, it certainly does make anti-alias filtering easier as well as not add as much quantization noise as a 16 bit DAC would.
alvaius
Does the Adobe version of Cooledit support the old plugin interface? I bought a new PC and went to get a new copy of Cooledit and, what a drag, no more basic version. Luckily I found my old installation files and got up and going again. I wrote my own scratch removal plugin using Syntrillium's filter SDK and it worked like a charm in MS C++ 6.0 with no headaches. What you want to do would be only 15-20 lines of C dropped into one of their examples. On my new machine Cooledit opens a 3.5hr mp3 in 3min.
Does the Adobe version of Cooledit support the old plugin interface? I bought a new PC and went to get a new copy of Cooledit and, what a drag, no more basic version. Luckily I found my old installation files and got up and going again. I wrote my own scratch removal plugin using Syntrillium's filter SDK and it worked like a charm in MS C++ 6.0 with no headaches. What you want to do would be only 15-20 lines of C dropped into one of their examples. On my new machine Cooledit opens a 3.5hr mp3 in 3min.
I don't see any problems upscaling the 16bit/44kHz data to 24bit/192kHz. I have a M-Audio Audiophile soundcard and using Steinberg Wavelab.
1. I rip the CD to HD
2. Convert it with Wavelab to 24/192 using high-end Apogee 24bit renderer
3. I plug the soundcard to my preamp and enjoy upscaled 16bit Cd's
There is a definitive difference bertween 16bit original and upscaled 16bit.
1. I rip the CD to HD
2. Convert it with Wavelab to 24/192 using high-end Apogee 24bit renderer
3. I plug the soundcard to my preamp and enjoy upscaled 16bit Cd's
There is a definitive difference bertween 16bit original and upscaled 16bit.
Really ?
Interesting !
Is it upsampling with special processing or just a well implemented sample rate conversion ?
If it is simply a sample rate conversion, if it sounds better, this means it is implemented better than the oversampling in the DAC chips. Question : how ?
Now if it is a special processing, is there any info on the algorithm ????
software...
I am not familiar with the Adobe program but I recommend to anyone trying SoundForge. It's very versatile and has options to configure the bit-depth and sampling rate independently, with different options for dithering, noise shaping, anti-aliasing, and others. The best sound processing software around, IMO. I personally perceived differences when upsampling, (usually more is better) - YMMV.
I am not familiar with the Adobe program but I recommend to anyone trying SoundForge. It's very versatile and has options to configure the bit-depth and sampling rate independently, with different options for dithering, noise shaping, anti-aliasing, and others. The best sound processing software around, IMO. I personally perceived differences when upsampling, (usually more is better) - YMMV.
My JVC receiver that I use for a surround decoder has a circuit in it that does the upconversion to 24/192 from 16/44.1 to reap the sonic benifits. I just got the receiver this week so I don't know if it works yet but I'll let you know if it does. It is the model RX-8030V if you want to look it up.
There a bunch of technical items I want to get into about why upsampling is good in theory and mostly bad in current implemenations but first, there is a piece of software to that does upsampling on WAV files and puts it in DVD-Video 96/24 format so you can burn with Nero. Its a free trial for 14 days:
http://www.eximius.nl/products.php
I'm listening to a few tracks i created from it right now.
Firstly, there seem to be some misconceptions here about upsampling/oversampling and why it is actually beneficial. It has _absolutely nothing_ to do with recreating missing ultrasonic content. Also the idea of taking an entire computer file and upsampling it has incredible benefits compared to the real time implementations out there built into hardware.
1) For any digital to analog conversion you need an analog filter at the end to filter out the "images" of the frequency content above the nyquist frequency. Ideally you would have a "brick wall" filter which is perfect with a gain of 1 and linear phase below nyquist/2 (22.05 KHz for CD) and infiite attentuation everywhere else if the DAC produced infinitely short impulses of the samples. In the real world case where the DAC keeps the sample constant until the next one, you need a brick wall filter where the gain actually curves up from 1 as frequency goes up. This is impossible to build with analog components since for it to work at any instant it needs knowledge of the entire future conent of the signal, it is non-causal. Any analog you try to build to get a steep enough drop at Nyquist/2 will have gain an phase distortion in the pass band. In the case of CD this means in the audible portion of the spectrum. Non-upsampling DACs don't use an analog filter but expect your equipment and ears will do a good enough job of filtering. The disadvantages is that it may make some equipment unstable or introduce IM distortion, and striclty speaking a gain 1 passband is not actually correct for holding a sample constant till the next sample.
2) The goal of upsampling/oversampling is to recaculate the signal in the digital domain to a higher sample frequency, filter out the "images" in the digital domain treating them like impulses instead of sample and hold, and then having the same signal at a higher sample frequency so you can use a simple analog filter with very stable pass band, a gentle rolloff well below the new Nyquist/2 and an ability to ignore sample and hold filters because their gain increase is negligible in the actual audio band. This best way to do this is to increase sample frequency AND bit depth so that your calculations end up with a resolution of 1/256 of a regular sample (going from 16 to 24 bit) which is pretty darn good. The problem with doing this stuff in real time is that to make the ideal filter you need all past and future values of a signal for any given sample calculation. It a DAC in real-time you only get a small window of the data to work with at any given time. As well, you are relying on the clocks and any jitter on the input in ASRC gets encoded into the new calcuated samples, and you get addition jitter from the output clock.
3) This is where the computer file based upsampling gets very interesting. In theory you can implement the ideal digital filter to get rid of the frequency "images" above the original Nyquist/2 because you actually do have all the information about the future and past for each sample calculation if you want to use it. Also, jitter will not get encoded into the samples since all the clock in this calculation are just numbers, completely virtual. Ideally the new frequency would be an exact muiltple of the original so the original samples remain. Unfortunately you don't get to buy a player that can play a disc at 176.4Khz sample rate. You can use a DVD player to play 96Khz/24 bit or 192Khz/24 bit. This is why those rates are targeted in a compluter software based upsample. Other interesting things can be done this way to like room/speaker correction and equalisation in the digital domain. You could build all of this into a single algorithm where everything is done 64bit floating point and requantised down to 24 bit at the end. I think you could do some very cool stuff this way.
And all of this is simply to allow a much nicer analog filter to be used at the end of the D/A conversion. This is by no means a waste of time.
http://www.eximius.nl/products.php
I'm listening to a few tracks i created from it right now.
Firstly, there seem to be some misconceptions here about upsampling/oversampling and why it is actually beneficial. It has _absolutely nothing_ to do with recreating missing ultrasonic content. Also the idea of taking an entire computer file and upsampling it has incredible benefits compared to the real time implementations out there built into hardware.
1) For any digital to analog conversion you need an analog filter at the end to filter out the "images" of the frequency content above the nyquist frequency. Ideally you would have a "brick wall" filter which is perfect with a gain of 1 and linear phase below nyquist/2 (22.05 KHz for CD) and infiite attentuation everywhere else if the DAC produced infinitely short impulses of the samples. In the real world case where the DAC keeps the sample constant until the next one, you need a brick wall filter where the gain actually curves up from 1 as frequency goes up. This is impossible to build with analog components since for it to work at any instant it needs knowledge of the entire future conent of the signal, it is non-causal. Any analog you try to build to get a steep enough drop at Nyquist/2 will have gain an phase distortion in the pass band. In the case of CD this means in the audible portion of the spectrum. Non-upsampling DACs don't use an analog filter but expect your equipment and ears will do a good enough job of filtering. The disadvantages is that it may make some equipment unstable or introduce IM distortion, and striclty speaking a gain 1 passband is not actually correct for holding a sample constant till the next sample.
2) The goal of upsampling/oversampling is to recaculate the signal in the digital domain to a higher sample frequency, filter out the "images" in the digital domain treating them like impulses instead of sample and hold, and then having the same signal at a higher sample frequency so you can use a simple analog filter with very stable pass band, a gentle rolloff well below the new Nyquist/2 and an ability to ignore sample and hold filters because their gain increase is negligible in the actual audio band. This best way to do this is to increase sample frequency AND bit depth so that your calculations end up with a resolution of 1/256 of a regular sample (going from 16 to 24 bit) which is pretty darn good. The problem with doing this stuff in real time is that to make the ideal filter you need all past and future values of a signal for any given sample calculation. It a DAC in real-time you only get a small window of the data to work with at any given time. As well, you are relying on the clocks and any jitter on the input in ASRC gets encoded into the new calcuated samples, and you get addition jitter from the output clock.
3) This is where the computer file based upsampling gets very interesting. In theory you can implement the ideal digital filter to get rid of the frequency "images" above the original Nyquist/2 because you actually do have all the information about the future and past for each sample calculation if you want to use it. Also, jitter will not get encoded into the samples since all the clock in this calculation are just numbers, completely virtual. Ideally the new frequency would be an exact muiltple of the original so the original samples remain. Unfortunately you don't get to buy a player that can play a disc at 176.4Khz sample rate. You can use a DVD player to play 96Khz/24 bit or 192Khz/24 bit. This is why those rates are targeted in a compluter software based upsample. Other interesting things can be done this way to like room/speaker correction and equalisation in the digital domain. You could build all of this into a single algorithm where everything is done 64bit floating point and requantised down to 24 bit at the end. I think you could do some very cool stuff this way.
And all of this is simply to allow a much nicer analog filter to be used at the end of the D/A conversion. This is by no means a waste of time.
I didn´t read the thread here but the headline leads to this:
http://www.hydrogenaudio.org/index.php?showtopic=19467&
Wombat
http://www.hydrogenaudio.org/index.php?showtopic=19467&
Wombat
Thanks Werner.. I looked up the spec and you are right - I didn't know they had included 88.2 and 176.4 in the spec - I haven't seen a disc done as such.
This is great for experimenting with software - unfortunately with my DVD-A player (Panasonic RP-91) you can only use the internal DACs for DVD-A and the player does not sound that good. I've been playing with DVD-V audio since it can go to an external DAC (in my case a battery powered CS4398 eval board). I've tried the RP-91 88.2 upsampled output to the DAC and that doesn't sound great at all.
Are there any DVD-A players out there that allow you to get the digital signal out?
This is great for experimenting with software - unfortunately with my DVD-A player (Panasonic RP-91) you can only use the internal DACs for DVD-A and the player does not sound that good. I've been playing with DVD-V audio since it can go to an external DAC (in my case a battery powered CS4398 eval board). I've tried the RP-91 88.2 upsampled output to the DAC and that doesn't sound great at all.
Are there any DVD-A players out there that allow you to get the digital signal out?
After some listening tests I come to this conclusions:
1. Upscaled Cd's sound much better
2. Playing from harddisk sounds better than playing the CD
3. 24bit/44kHz sounds better than 16bit/96kHz
4. 24bit sounds much more rythmic than 16bit
5. 96kHz gives much more detail and space than 44kHz
In the end playing a CD directly sounds like crap, upscaled (24bit/192kHz) CD's played from harddisk have a musicality/openness/clarity and rhythmical correctness nobody can deny.
1. Upscaled Cd's sound much better
2. Playing from harddisk sounds better than playing the CD
3. 24bit/44kHz sounds better than 16bit/96kHz
4. 24bit sounds much more rythmic than 16bit
5. 96kHz gives much more detail and space than 44kHz
In the end playing a CD directly sounds like crap, upscaled (24bit/192kHz) CD's played from harddisk have a musicality/openness/clarity and rhythmical correctness nobody can deny.
macaque said:There a bunch of technical items I want to get into about why upsampling is good in theory and mostly bad in current implemenations but first, there is a piece of software to that does upsampling on WAV files and puts it in DVD-Video 96/24 format so you can burn with Nero. Its a free trial for 14 days:
http://www.eximius.nl/products.php
I'm listening to a few tracks i created from it right now.
-----------------------------------------------------------------------------
Sounds just what I need! Does this burn onto CDRW using CDR disca, or do I require a DVD burner?
And will discs play on a DVD-V hardware player?
One thing that comes to mind when talking about upsampling, etc, is the delivery medium given to the mastering engineer. If they have received a finished mix on a 16bit,44khz DAT tape, then run it through some analogue processing, then it makes sense to me that artefacts (from the mix on the DAT) would have been irreversibly rendered into the mastered recording.
unless the mastering engineer was smart and used upsampling technology when decoding digital recordings too?
I've been wondering about this for ages...
...Steve
unless the mastering engineer was smart and used upsampling technology when decoding digital recordings too?
I've been wondering about this for ages...
...Steve
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.