Iain McNeill said:
This is something i have been wondering about...
When "you" say that there is a relationship between dynamic range and resolution........
This means that ....
Many records today has a dynamic range of 6dB between the fainthest and the loudest sound, so in theory there is enough with one bit of data and the rest of those 16/24 is redundant....
-So why dont drop those unecessary bits and only have one bit of resolution?
Seriously...
Why is dynamic range and resolution packed together like they had some kind of relation?
Sorry if i sound so sarcastic but ...
-6dB of analog is enough for a kitchenradio but 6dB of digital is useless.
So wouldnt it be better to compare digital steps with distortion and assign an analog distortionequivalent to each bit added instead of todays dynamic range/noise floor?
Hi there.
My two cents:
Why don't you just run everything on the PC.
Buy yourself a professional sound card. Get yourself a Linux installed, run brutefir and FIR filters generated with Acourate or similar.
Running the crossovers and e.g. convolution in realtime with very long filters is not a problem for a realtime Linux system.
Basically you're done at a fragment of cost compared to DEQX or other external solutions. The gained flexibility is IMO much higher. Beside that there is a good chance to get even better results.
Your digital chain is as small as it can get.
I had the pleasure to listen to such a system recently.
For an ambitioned DIYer there is IMO no better solution out there.
Soon my brutefir digital filtered system will be up'n running.
Cheers
My two cents:
Why don't you just run everything on the PC.
Buy yourself a professional sound card. Get yourself a Linux installed, run brutefir and FIR filters generated with Acourate or similar.
Running the crossovers and e.g. convolution in realtime with very long filters is not a problem for a realtime Linux system.
Basically you're done at a fragment of cost compared to DEQX or other external solutions. The gained flexibility is IMO much higher. Beside that there is a good chance to get even better results.
Your digital chain is as small as it can get.
I had the pleasure to listen to such a system recently.
For an ambitioned DIYer there is IMO no better solution out there.
Soon my brutefir digital filtered system will be up'n running.
Cheers
For those of you interested in PC based solutions I invite you to take a good look at Jan's product. I recently have started an evaluation of it and have been very impressed thus far. The software works as advertised and the price that he charges is very reasonable.
http://www.thuneau.com/
http://www.thuneau.com/
In reply to the original question:
An AD-DA chain that is properly designed (including dither noise and good enough antialiasing/reconstruction filters) only has two limitations: A noise floor and the bandwidth limit of the antialias filter (whose cutoff frequency in turn is defined by the sampling rate).
This means that even if the level in the stopband is lowered closer to the (quantisation/dither) noise floor, the noise still has the same amplitude there. Assuming that there is another branch of the filter that "fills in" with high level sound, the net effect are just some added noise from the fact that there is now noise from two branches.
This noise is no worse than the noise from any other digital system, and certainly not worse than that of an old tape recorder or vinyl record.
So in principle: Digital filters are not flawed just because they are digital. There are however, a lot of ways to make such filters sound bad. One would be to use poor antialias filtering or to skip the dither noise, just as in any AD-DA chain. Another would be to use poor filter functions in the crossover.
One of the worst crossover filters imagniable is actually the ideal highpass/lowpass filters with zero phase delay and an infinitely steep slope.
An AD-DA chain that is properly designed (including dither noise and good enough antialiasing/reconstruction filters) only has two limitations: A noise floor and the bandwidth limit of the antialias filter (whose cutoff frequency in turn is defined by the sampling rate).
This means that even if the level in the stopband is lowered closer to the (quantisation/dither) noise floor, the noise still has the same amplitude there. Assuming that there is another branch of the filter that "fills in" with high level sound, the net effect are just some added noise from the fact that there is now noise from two branches.
This noise is no worse than the noise from any other digital system, and certainly not worse than that of an old tape recorder or vinyl record.
So in principle: Digital filters are not flawed just because they are digital. There are however, a lot of ways to make such filters sound bad. One would be to use poor antialias filtering or to skip the dither noise, just as in any AD-DA chain. Another would be to use poor filter functions in the crossover.
One of the worst crossover filters imagniable is actually the ideal highpass/lowpass filters with zero phase delay and an infinitely steep slope.
Thunau said:
I know what you mean, but that wasnt really what i was talking about...
Well i try again...
If you have a D/A with an output of 1volt then if you only have one bit of resolution you can only give the levels of 0volt or 1 volt.
If you have 2 bits you have four steps between 0volt and 1volt (Only 0volt , 0.250volt , 0,500volt and 0,750volt is possible).
and with three bits you get 8 different levels...
(0volt, 0.125 , 0.250 , 0.375 and so on... but never anything in between those values.)
The thing is that with more bits you get a better resolution and not more volume, you can still play 140dB with only one bit of resolution and you can very well play 75dB with 24bits of resolution...
Noone in the analog world would call crossoverdistortion or clippingdistortion for noise so why is it called "noise" in the digital world?
-----------
I once tested a analog chinese amp with a sinus tone and watched the output thru a oscilloscope.
Without load the sinewave was represented as a sinewave with more or less an infinite number of bits.
But when an 8ohm resistor was added to the output, the transistors in the feedbackchain got saturated and it created a stepped response similar to 2bits (Four steps) of resolution.
The amp still had atleast 80-100dB of S/N ratio but it did not in any way have the resolution necessary to trace the inputsignal.
Carl_Huff said:
I am more than interested in serious PC audio.
To start with - a Windows based system I wouldn't call a
serious audio basis.
I am wondering how many taps Thunau is using for his filters?
What's the processing bit-width?
48db filters are not really challenging a DEQX either?
Do I miss something here?
Looks to me like a nice toy on a first glance. That's about it.
Cheers
soundcheck said:
I am more than interested in serious PC audio.
To start with - a Windows based system I wouldn't call a
serious audio basis.
I am wondering how many taps Thunau is using for his filters?
What's the processing bit-width?
48db filters are not really challenging a DEQX either?
Do I miss something here?
Looks to me like a nice toy on a first glance. That's about it.
Cheers
First, majority of the music produced today is produced on Windows boxes. The big studios rely on Mac based ProTools, but outside of Protools (which runs on Windows just as well BTW) it's 90% Windows programs.
Second, I don't use taps- my software uses IIR filters and processing is done with 64bit precision using proprietary algorithms.
Third, Allocator can deliver a transient perfect speaker - for all frequencies reproduced by the speaker- from below box roll off on up.
DEQX can't do anything to correct the group delay of the box/port/woofer combo. You can clearly hear the difference when switching the correction on and off.
Lastly, the steepest slopes are not necessarily best. That's like saying the hottest peppers are the best or the fastest cars are the best. Unless you listened to a speaker with Allocator as the crossover, properly dialed in and with a good sound card reserve your opinion.
Thunau said:
If this is really the case - I am asking myself, why more than 90% sounds as if being produced under Windows.
Thunau said:
I am not a friend of DEQX either.
Thunau said:
You shouldn't forget that you're at DIY-Audio. And it's a free world!
And you as a "Manufacturer" with potentially commercially intentions shouldn't tell non-commercial people to "reserve their opinions".
I never tried your product. That's right. But if everybody in the forum were supposed to reserve their opinions about things they
never tried - this forum would be deadly quiet. And we wouldn't make any progress.
You outlined things I asked for and I appreciate that.
I just bring up a very non-commercial DIY related 64bit long-fir-filter-alternative, which is called brutefir under Linux, which is giving me much more fexibility than an off-the-shelve-product. Of course it needs some DIY spirit to get it going. But that's what this forum is all about, isn't it.
Cheers
With regards to the original posters, and Romy the Cat's opinions, I would like to comment.
The idea that digital can't properly represent small signals is a myth. It's easy to imagine that a 1-bit sine wave looks like a square wave and then just write off the whole idea as a lost cause. This is incorrect, and you can perfectly represent signals with an amplitude very much less than 1 bit in a digital system.
However, this is not done by default, and if you simply try to represent a perfect sine wave with 1 bit it will sound rubbish. The key to the extra resolution is the use of dither whenever you quantise the data. If you add dither before quantisation you will end up, after quantisation, with the original signal (at full resolution) plus some white noise. This is, of course, what you get in an analogue system - the signal and some noise. With the analogue system, the level of the noise depends on many things - circuit layout, component choice, input impedance, etc. With the digital system, the amplitude of the noise depends only on how many bits you use.
And, like an analogue system, you can represent signals smaller than the level of the noise - you won't be able to see them easily on an un-averaged 'scope, but you can hear them "through" the noise and an FFT will easily pick them out.
A very easy demonstration of this is available to anyone with a VST Host (I use Bidule) and the Waves plugins:
- Get some full-resolution audio into your host (I use the wave file player module in Bidule for this)
- Send the audio through a good gain control (such as a couple of cascaded Q1 plugins) and reduce its amplitude by around 40dB.
- route the audio into L1 Ultramximizer+, set as follows:
-- quantise to: 8 bits
-- dither: none
-- noise-shaping: none
-- everything else as default
- Now, playback the audio and listen (headphones are best for this and you might have to turn up the volume to still be able to hear anything - just be careful not to deafen yourself!). If you've reduced the gain of the audio enough (around 50dB or more), you should hear nothing at all - the audio at all points is less than 1 bit in size and has been rounded to nothing. If you've reduced it by 40dB and the signal is loud, you should hear some hugely distorted sound. The quiet passages will disappear to nothing, and the loud will be accompanied by crackles.
This is what people like Romy the Cat think digital audio is. And if this happens at 8 bits then you can rest assured that the same thing, at a lower level, is happening at 16 bits.
But!
Now we add dither:
- Stop the sound
- In L1+, set Dither to Type 1. You'll hear that a layer of constant white noise has appeared.
- Now play the audio again, with the amplitude still reduced by around 40dB. You'll hear the audio, sounding perfect, but with white noise over the top. To compare dithered and undithered audio, simply switch the dither type in L1+ between Off and Type 1 (ignore Type 2, that's a little different).
You're likely to find that when dithered, the audio sounds fine, albeit with a lot of noise. (You can even reduce the audio beyond the point where, undithered, it would disappear - and still hear it.) The undithered audio, by comparison, sounds horribly distorted. All this is at 8-bits of resolution. More bits is exactly the same, except the noise amplitude is lower - ~6dB for every bit. 24-bit audio, for example, gives perfect linearity (when dithered) and has a noise floor of ~-144dBFS. There's no audible signal that can't be represented with this. And is has absolutely zero distortion.
All of this so far has dealt with fixed-point signals. Floating point processing is very popular, but is inherently unditherable. However, while it suffers from the "bad" kind of noise floor, it has a constant signal-to-noise ratio, regardless of the signal's amplitude, and so quantization noise is generally quieter than for the same number of fixed-point bits.
I personally always use fixed-point arithmetic in the things I write, because unless you want to repeatedly attenuate and then amplify the audio, fixed-point gives better distortion performance. Of course, if you use enough bits, such as 64-bit floating point, the nasties in the noise floor are so quiet that you can just ignore them. I'd be a little wary about 32-bit floating-point though. But that's me.
The idea that digital can't properly represent small signals is a myth. It's easy to imagine that a 1-bit sine wave looks like a square wave and then just write off the whole idea as a lost cause. This is incorrect, and you can perfectly represent signals with an amplitude very much less than 1 bit in a digital system.
However, this is not done by default, and if you simply try to represent a perfect sine wave with 1 bit it will sound rubbish. The key to the extra resolution is the use of dither whenever you quantise the data. If you add dither before quantisation you will end up, after quantisation, with the original signal (at full resolution) plus some white noise. This is, of course, what you get in an analogue system - the signal and some noise. With the analogue system, the level of the noise depends on many things - circuit layout, component choice, input impedance, etc. With the digital system, the amplitude of the noise depends only on how many bits you use.
And, like an analogue system, you can represent signals smaller than the level of the noise - you won't be able to see them easily on an un-averaged 'scope, but you can hear them "through" the noise and an FFT will easily pick them out.
A very easy demonstration of this is available to anyone with a VST Host (I use Bidule) and the Waves plugins:
- Get some full-resolution audio into your host (I use the wave file player module in Bidule for this)
- Send the audio through a good gain control (such as a couple of cascaded Q1 plugins) and reduce its amplitude by around 40dB.
- route the audio into L1 Ultramximizer+, set as follows:
-- quantise to: 8 bits
-- dither: none
-- noise-shaping: none
-- everything else as default
- Now, playback the audio and listen (headphones are best for this and you might have to turn up the volume to still be able to hear anything - just be careful not to deafen yourself!). If you've reduced the gain of the audio enough (around 50dB or more), you should hear nothing at all - the audio at all points is less than 1 bit in size and has been rounded to nothing. If you've reduced it by 40dB and the signal is loud, you should hear some hugely distorted sound. The quiet passages will disappear to nothing, and the loud will be accompanied by crackles.
This is what people like Romy the Cat think digital audio is. And if this happens at 8 bits then you can rest assured that the same thing, at a lower level, is happening at 16 bits.
But!
Now we add dither:
- Stop the sound
- In L1+, set Dither to Type 1. You'll hear that a layer of constant white noise has appeared.
- Now play the audio again, with the amplitude still reduced by around 40dB. You'll hear the audio, sounding perfect, but with white noise over the top. To compare dithered and undithered audio, simply switch the dither type in L1+ between Off and Type 1 (ignore Type 2, that's a little different).
You're likely to find that when dithered, the audio sounds fine, albeit with a lot of noise. (You can even reduce the audio beyond the point where, undithered, it would disappear - and still hear it.) The undithered audio, by comparison, sounds horribly distorted. All this is at 8-bits of resolution. More bits is exactly the same, except the noise amplitude is lower - ~6dB for every bit. 24-bit audio, for example, gives perfect linearity (when dithered) and has a noise floor of ~-144dBFS. There's no audible signal that can't be represented with this. And is has absolutely zero distortion.
All of this so far has dealt with fixed-point signals. Floating point processing is very popular, but is inherently unditherable. However, while it suffers from the "bad" kind of noise floor, it has a constant signal-to-noise ratio, regardless of the signal's amplitude, and so quantization noise is generally quieter than for the same number of fixed-point bits.
I personally always use fixed-point arithmetic in the things I write, because unless you want to repeatedly attenuate and then amplify the audio, fixed-point gives better distortion performance. Of course, if you use enough bits, such as 64-bit floating point, the nasties in the noise floor are so quiet that you can just ignore them. I'd be a little wary about 32-bit floating-point though. But that's me.
Wingfeather said:
This was exactly my point, but you explained it better.
If anyone wants to listen to the effects of dither noise, please try these 4-bit snippets of Stravinskji's Firebird:
http://www.tolvan.com/quant/firebird4none.wav
http://www.tolvan.com/quant/firebird4tri.wav
Listen to how the weak sounds are audible below the noise in the dithered version, but how they drown in distortion in the undithered version.
I'd like to comment on this approach, because I think it's a little bit flawed - while you will be retaining all of the 16 bits of original data, you're still going to be introducing quantization errors at, presumably, the 24-bit level.
While it's not clear-cut as to whether anybody is really going to care about a single 24-bit quantization, it would seem to be a much better (and simpler) approach to perform the entire volume control digitally (but properly dithered!). This would not only remove the need for a fiddly two-stage volume control, but also give you full-resolution control of the volume for audio that's up to 24-bits in depth - with no loss of detail.
As far as I can tell, this approach has no drawbacks. Comments are welcome, though.
I think I'm learning..a little!
Maybe back a bit to Romy's comments, as in all engineering there are compromises, so where would digital x-overs beat passive, where would passive beat digital??
I'm a real computer idiot, where would I find an easily understood article on plugins etc along the lines spoken of in this thread?/ I tried reading Shinobiwans big writeup in this area, but for a complete newbie like me it seemed to have a bit of 'pre-supposed' knowledge and went way beyond what I could use (right now).
There seem to be differing viewpoints on something like the deqx, well at least for me it provided a reasonably understandable step by step approach (even if I have no idea of the mechanics behind it). I can understand that some of the other aftermarket type stuff found on the net could, if you knew what you're doing, provide greater control of the results perhaps, but again it's then back to computing knowledge.
Another possible difference is that unlike most people on a forum like this, a lot of audio people wouldn't want neccessarily a computer based sound system, and would feel more comfortable with a 'traditional' box based setup for audio.
Maybe back a bit to Romy's comments, as in all engineering there are compromises, so where would digital x-overs beat passive, where would passive beat digital??
I'm a real computer idiot, where would I find an easily understood article on plugins etc along the lines spoken of in this thread?/ I tried reading Shinobiwans big writeup in this area, but for a complete newbie like me it seemed to have a bit of 'pre-supposed' knowledge and went way beyond what I could use (right now).
There seem to be differing viewpoints on something like the deqx, well at least for me it provided a reasonably understandable step by step approach (even if I have no idea of the mechanics behind it). I can understand that some of the other aftermarket type stuff found on the net could, if you knew what you're doing, provide greater control of the results perhaps, but again it's then back to computing knowledge.
Another possible difference is that unlike most people on a forum like this, a lot of audio people wouldn't want neccessarily a computer based sound system, and would feel more comfortable with a 'traditional' box based setup for audio.
I do like my modded out DEQX.
And I can't sell folks a computer based solution..just yet. Soon though.
For those of you with a memory, I was RIGHT at the tip of the point that introduced the HTPC and 'media box' to the world. I laid out the specs to the guy who did the first one. I suggested the entire process and who to contact, who to develop the boards, components, I outlined the challenges, etc. I laid out everything ..and the guy followed it -- to the letter. And the first true commecial grade successful Scaler/HTPC/media box was born. So yeah, I was instrumental in speeding up the death and/or change of direction at Faroudja, etc, sheerly by accident.
So I'm not afraid of PC based crossovers..I simply haven't tried one yet.
Thuneau, we may be talking, sometime soon. I very much like what the DEQX does, but the bit rate is not high enough for any large changes in data due to 'algorithm twisties', ie, steep slopes and large levels of correction can cause minute changes in where a given bit may be after the algorithms and bit depth are done with it.....when related to the other modified bits. The DEQX guys are likely to come out with a 192/24 bit unit or plug-in soon, one suspects. Perfection in location of relative bit info before vs after is critical, to maintain fidelity. Which is part of why lower slope crossovers, even in the digital domain, are critical to good sound with a digital unit. I definitely hear good stuff going on.
Did I mention I do one hell of a clock? No-one knows how important a low jitter clock is until they hear a really good one. My personal design (which I hope to patent soon) is decent enough that everyone single person who's heard it thinks I'm actually playing a LP on my analog rig, it's so good. And this is playing a 44.1/16 CD. I tend to apply it to everything digital, especially recording gear!
Point is, low jitter in digital crossovers, compared to that of CD reproduction..is ..at LEAST.....4 times as critical!!
Passive crossovers beat digital..when they are OUT of the crossover area. Ie, if the digital crossover is 24db at 2khz, and the same for the analog crossover version....then..from about 1khz to about 3.5khz ..the digital unit RULES. But below that and above that frequency area...micro detail and phasing agreeance between the drivers absolutely RULES in the analog passive crossover. Critical micro detail is preserved in the analog one. It is screwed in the digital one. Totally pooched.
But in the analog crossover..in the crossover region, that approx 1khz to about 3-3.5khz..the analog crossover pahsing agreeance is totally pooched. A total soggy mess. One can only do so much to save it. Lord knows I do an excellent analog crossover. But we can only do so much.
Again, in that crossover region...(1khz to 3-3.5khz) phasing agreeance in the digital one is nearly perfect, making for a VERY nice 'snap' to a very critical frequency area of human hearing function. Critical. Except all that micro-detail that we pay the extremely long green ($$,$$$) for...is totally pooched. Everywhere, as well. Not just in the crossover region that it just 'saved'. Like someone saving you at an accident scene....they carry you safely away...and at the same time, mixes your blood with his..and gives you TB in the process.
Some is good..some is bad. Such is life. Too bad. Neither is perfect. Both have their strengths. But Personally, I cannot truly enjoy the heart and soul of the music with the digital unit. Granted, it does sound great, but a totally analog crossover, even with its glaring faults, still blows digital out of the water.
Unless you listen via total analog systems, ie LP's...done right... (I have a three chassis octal based Tube preamp, I'm using right now), you are likely unfamiliar with the loss of fidelity I'm speaking of.
All I can say, is generally, the further down the road the person in the musical quest is..the more they tend to understand that 'analog rules'. For all the right reasons. A brutally long story.
And I can't sell folks a computer based solution..just yet. Soon though.
For those of you with a memory, I was RIGHT at the tip of the point that introduced the HTPC and 'media box' to the world. I laid out the specs to the guy who did the first one. I suggested the entire process and who to contact, who to develop the boards, components, I outlined the challenges, etc. I laid out everything ..and the guy followed it -- to the letter. And the first true commecial grade successful Scaler/HTPC/media box was born. So yeah, I was instrumental in speeding up the death and/or change of direction at Faroudja, etc, sheerly by accident.
So I'm not afraid of PC based crossovers..I simply haven't tried one yet.
Thuneau, we may be talking, sometime soon. I very much like what the DEQX does, but the bit rate is not high enough for any large changes in data due to 'algorithm twisties', ie, steep slopes and large levels of correction can cause minute changes in where a given bit may be after the algorithms and bit depth are done with it.....when related to the other modified bits. The DEQX guys are likely to come out with a 192/24 bit unit or plug-in soon, one suspects. Perfection in location of relative bit info before vs after is critical, to maintain fidelity. Which is part of why lower slope crossovers, even in the digital domain, are critical to good sound with a digital unit. I definitely hear good stuff going on.
Did I mention I do one hell of a clock? No-one knows how important a low jitter clock is until they hear a really good one. My personal design (which I hope to patent soon) is decent enough that everyone single person who's heard it thinks I'm actually playing a LP on my analog rig, it's so good. And this is playing a 44.1/16 CD. I tend to apply it to everything digital, especially recording gear!
Point is, low jitter in digital crossovers, compared to that of CD reproduction..is ..at LEAST.....4 times as critical!!
Passive crossovers beat digital..when they are OUT of the crossover area. Ie, if the digital crossover is 24db at 2khz, and the same for the analog crossover version....then..from about 1khz to about 3.5khz ..the digital unit RULES. But below that and above that frequency area...micro detail and phasing agreeance between the drivers absolutely RULES in the analog passive crossover. Critical micro detail is preserved in the analog one. It is screwed in the digital one. Totally pooched.
But in the analog crossover..in the crossover region, that approx 1khz to about 3-3.5khz..the analog crossover pahsing agreeance is totally pooched. A total soggy mess. One can only do so much to save it. Lord knows I do an excellent analog crossover. But we can only do so much.
Again, in that crossover region...(1khz to 3-3.5khz) phasing agreeance in the digital one is nearly perfect, making for a VERY nice 'snap' to a very critical frequency area of human hearing function. Critical. Except all that micro-detail that we pay the extremely long green ($$,$$$) for...is totally pooched. Everywhere, as well. Not just in the crossover region that it just 'saved'. Like someone saving you at an accident scene....they carry you safely away...and at the same time, mixes your blood with his..and gives you TB in the process.
Some is good..some is bad. Such is life. Too bad. Neither is perfect. Both have their strengths. But Personally, I cannot truly enjoy the heart and soul of the music with the digital unit. Granted, it does sound great, but a totally analog crossover, even with its glaring faults, still blows digital out of the water.
Unless you listen via total analog systems, ie LP's...done right... (I have a three chassis octal based Tube preamp, I'm using right now), you are likely unfamiliar with the loss of fidelity I'm speaking of.
All I can say, is generally, the further down the road the person in the musical quest is..the more they tend to understand that 'analog rules'. For all the right reasons. A brutally long story.
It is a little bit more subtle then that. There can be a lot of different opinions on how audible digital filtering and digital level control is, but the fact remains that digitally lowering the level of a signal decreases its dynamic range.
The noise in the DAC remains at the same level when you digitally lower the signal (meaning the signal is represented by less bits). That decreases the dynamic range, and your 16 bit resolution is now down to 14 or 12. Again, this may or may not be audible, depending on a lot of other things.
If you lower the level analogically you lower not only the signal amplitude but also the noise so to a first order your dynamic range is not decreased.
I have both a Behringer DCX2496 and a DEQX and on the Behringer I can hear loss of resolution near the max level change of 15dB.
Jan Didden
The noise in the DAC remains at the same level when you digitally lower the signal (meaning the signal is represented by less bits). That decreases the dynamic range, and your 16 bit resolution is now down to 14 or 12. Again, this may or may not be audible, depending on a lot of other things.
If you lower the level analogically you lower not only the signal amplitude but also the noise so to a first order your dynamic range is not decreased.
I have both a Behringer DCX2496 and a DEQX and on the Behringer I can hear loss of resolution near the max level change of 15dB.
Jan Didden
KBK, all I can say is... wow.
Well firstly, and probably most importantly, it's difficult to compare digital and passive with each other. If you want to compare digital and analogue techniques you're really going to have to compare digital with active analogue - they're much closer to being equivalent.
Analogue active and analogue passive have been compared before - they both have their followers. While active is "technically" superior in essentially every way, a lot of people prefer the sound of passive. If you don't like active analogue, it would seem you're very unlikely to like digital. However, if you do like it things get interesting.
What's important to remember is that digital and active analogue really are doing the same thing. Ignoring for a moment the extra stuff you can do with digital filters, the maths of the filter design is all the same, and all the same things are possible with digital that are possible with analogue. It's also important to remember that general filtering performance is identical - amplitude and phase responses of basic filter types are the same whether you implement them with op-amps or with some DSP. I don't know what KBK was talking about saying phase is "messed up" with one kind of crossover or other - those statements are demonstrably wrong.
However, the details are what makes the debate interesting. Let's take some examples. An 8th-order analogue filter (Linkwitz-Riley or whatever) will probably be implemented as a cascade of 2nd-order Sallen/Key filters. Each of these 2nd-order sections will add a little bit of distortion (not much, of course - and it depends on what components you use - but a little) and some noise. Four such stages in cascade can start to add reasonably significant amounts of noise. And this is just for the basic crossover section of the system. If you want a Linkwitz Transform in there, baffle-step correction, system EQ, driver EQ, any of that, the nonlinearity and noise is going to add up. Indeed, there's a thread open at the moment about improving Linkwitz's own active crossover board by amalgamating his multiple-stage processing into as few stages as possible - and this should provide some improvement.
If you do these sorts of operations digitally (but properly, i.e., with dither), you get no distortion added anywhere and the amount of noise added at each stage depends only on how many bits you use to represent the audio. You're also guaranteed that the noise floor will be predictably pure white - this isn't something that all analogue components can give you.
That is quite right. But people seem to be under the mistaken impression that you need all of this dynamic range. The signal is quieter! After the level change, the quietest parts of the signal are now inaudibly quiet - it is really immaterial whether they are lost in the white noise floor or not.
But ignore that view of it for a second. Let's say we have our 24-bit digital system. The noise floor of this system is at around -144dBFS. If we calibrate all our analogue gains so that this noise floor is at 0dBSPL, (which, I'm sure we'll all agree, is inaudible in a room), then the dithered digital system can perfectly represent any signal that's up to 144dBSPL in amplitude. If your signal happens to only be around 96dBSPL in amplitude, then 16 bits can represent it perfectly. If you make this quieter, down to 84dB, then 14 bits can represent it perfectly. This system is easily capable of (again, perfectly) representing signals that are far below the threshold of audibility, and signals will be lost to our hearing long before they are lost in the noise floor of the digital system.
Exactly right! That's part of my whole argument! The noise ("noise" being defined as the difference between the infinite-resolution original signal and the finite-resolution digital representation of it) output doesn't ever change in level. Only the signal does.
Finally, it's worth adding that just because you can hear a difference in quality from a DCX2496 when you use its gain control, does not mean you can condemn digital audio as broken. I don't think any of us here know how Behringer have implemented the arithmetic in this unit.
Well firstly, and probably most importantly, it's difficult to compare digital and passive with each other. If you want to compare digital and analogue techniques you're really going to have to compare digital with active analogue - they're much closer to being equivalent.
Analogue active and analogue passive have been compared before - they both have their followers. While active is "technically" superior in essentially every way, a lot of people prefer the sound of passive. If you don't like active analogue, it would seem you're very unlikely to like digital. However, if you do like it things get interesting.
What's important to remember is that digital and active analogue really are doing the same thing. Ignoring for a moment the extra stuff you can do with digital filters, the maths of the filter design is all the same, and all the same things are possible with digital that are possible with analogue. It's also important to remember that general filtering performance is identical - amplitude and phase responses of basic filter types are the same whether you implement them with op-amps or with some DSP. I don't know what KBK was talking about saying phase is "messed up" with one kind of crossover or other - those statements are demonstrably wrong.
However, the details are what makes the debate interesting. Let's take some examples. An 8th-order analogue filter (Linkwitz-Riley or whatever) will probably be implemented as a cascade of 2nd-order Sallen/Key filters. Each of these 2nd-order sections will add a little bit of distortion (not much, of course - and it depends on what components you use - but a little) and some noise. Four such stages in cascade can start to add reasonably significant amounts of noise. And this is just for the basic crossover section of the system. If you want a Linkwitz Transform in there, baffle-step correction, system EQ, driver EQ, any of that, the nonlinearity and noise is going to add up. Indeed, there's a thread open at the moment about improving Linkwitz's own active crossover board by amalgamating his multiple-stage processing into as few stages as possible - and this should provide some improvement.
If you do these sorts of operations digitally (but properly, i.e., with dither), you get no distortion added anywhere and the amount of noise added at each stage depends only on how many bits you use to represent the audio. You're also guaranteed that the noise floor will be predictably pure white - this isn't something that all analogue components can give you.
That is quite right. But people seem to be under the mistaken impression that you need all of this dynamic range. The signal is quieter! After the level change, the quietest parts of the signal are now inaudibly quiet - it is really immaterial whether they are lost in the white noise floor or not.
But ignore that view of it for a second. Let's say we have our 24-bit digital system. The noise floor of this system is at around -144dBFS. If we calibrate all our analogue gains so that this noise floor is at 0dBSPL, (which, I'm sure we'll all agree, is inaudible in a room), then the dithered digital system can perfectly represent any signal that's up to 144dBSPL in amplitude. If your signal happens to only be around 96dBSPL in amplitude, then 16 bits can represent it perfectly. If you make this quieter, down to 84dB, then 14 bits can represent it perfectly. This system is easily capable of (again, perfectly) representing signals that are far below the threshold of audibility, and signals will be lost to our hearing long before they are lost in the noise floor of the digital system.
Exactly right! That's part of my whole argument! The noise ("noise" being defined as the difference between the infinite-resolution original signal and the finite-resolution digital representation of it) output doesn't ever change in level. Only the signal does.
Finally, it's worth adding that just because you can hear a difference in quality from a DCX2496 when you use its gain control, does not mean you can condemn digital audio as broken. I don't think any of us here know how Behringer have implemented the arithmetic in this unit.
Wingfeather said:
dither is not a magic weapon for preserving resolution. All it does is it randomizes quantization error, while increasing the noise floor. That's where noise shaping comes in- good dither introduces noise where it is least audible. But, when you digitally attenuate a signal you lose resolution bits with or without dither. As it was pointed out in this thread before, dither is most useful with low amplitude signals- this is due to the way we perceive loudness. We hear in a logarithmic fashion, sampling quantization is linear. Therefore we can hear the same absolute error much better when signal is low. Once the signal resides in the mid to upper range, the rounding errors become less and less perceivable. They simply become ever so smaller fraction of a dB with louder signals. That's a good reason to run your D/A with solid signals.
Had 16 bit AD and DA converter been designed to work the way we hear- with sampling steps every 96/65536 or every 0.00146484375 dB, most all the resolution problems would be moot and CD music would sound the same with loudly mastered or not so loudly mastered sources.
Right now, the first steps of AD and DA converters are very large, the last steps are tiny (to our ears). There are as many steps from -6dBFS to 0dBFS as there are from -6dBFS to -96dBFS.
Actually, it kind've is. I guess we need to be clear on our terminology here. I am not saying you do not lose information by attenuating digitally - you do. You also lose information by attenuating with an active analogue system, because some of the signal is lost into the noise floor. There's no way to amplify it again without amplifying the noise floor along with it. What I am saying is that a properly-dithered digital system has infinite resolution, in that the signal you get is made up of the pristine original signal, plus some amount of white noise. Any repetitive signal can always be recovered, to any degree of accuracy, if you average out the noise for long enough. In a 16-bit dithered digital system you can put a sine wave of -300dBFS in if you want to. A long enough FFT (it will have to be really, really long for this one) will see this sine wave perfectly, because the dithered system has infinite resolution. Just like an analogue system.
It is most useful with low-amplitude signals, but not for that reason. The reason is that with large signals the "natural" quantisation noise is essentially white - there is very little correlation between the signal and the error. You get good linearity without actually having to apply dither. With small signals, this is no longer true, and the quantisation noise will be harmonically related to the signal itself. This is the source of distortion in an undithered digital system, and gets worse as the signal gets smaller. The use of dither forces the error noise to be always white regardless of the level and character of the signal you put in.
Well I sort of see what you're getting at here, but I don't see how it applies. You're saying that as the signal gets louder, you're less able to hear the noise floor. Well, of course! The noise floor is at a constant level and the signal got louder.
What I thought we'd established earlier was that we're using 24 bits - so many that this noise floor is well below the threshold of hearing. And since we use dither it's better to think of it less as a "rounding error" (which I think implies it has some kind of correlation with the audio), than as random white noise.
Digital makes for pretty numbers on a piece of paper but the reality says otherwise, when it comes to being complementary to the human hearing function.
I note that in 2006,as a strictly high end audio show..I went around as a manufacturer, and questioned quite a few of the manufacturers of maximum quality loudspeakers on their thoughts on digital crossovers. 'Not yet ready for prime time' was the answer, almost to a man.
This, specifically concerning 'cutting edge' sonics. All were waiting for the next generation of digital hardware and software, to begin considering commercial implementation.
I note that in 2006,as a strictly high end audio show..I went around as a manufacturer, and questioned quite a few of the manufacturers of maximum quality loudspeakers on their thoughts on digital crossovers. 'Not yet ready for prime time' was the answer, almost to a man.
This, specifically concerning 'cutting edge' sonics. All were waiting for the next generation of digital hardware and software, to begin considering commercial implementation.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.