Digital Is Digital, do you know someone who uses processors and still sounds "natural& Options

[Ivan Milenkovic]

It's very interesting that we are constantly trying to emulate analog artifacts and rolloffs. I'm guessing we are used to them?

It's even more interesting that there are quite a few extremes among people in the world regarding this matter. While most of people will tell no difference on a blind test, most experts will have their own subjective opinions, based on the sounds they prefer. There is no middle ground with sound, people choose different gear because of the way it alters the original signal.

How can we say what's good, what's not? Frequencies are more complicated than just choosing a range. We can color the sound using analog gear, like a tape machine, or tube preamp, or digital device, like a harmonic generator device, but it will again be - imperfect. Why is that. Is it because it reminds us on analog sounds that we heard while growing up, as members of an age when that type of technology/sound was dominant (and stuck around)?

You are at GuitarMasterClass.net

Don't miss today's free lick. Plus all our lessons are packed with free content!

Don't miss today's free blues, jazz & country licks. Plus all our lessons are packed with free content!

This post has been edited by Ivan Milenkovic: Aug 18 2011, 10:42 AM

[Saoirse O'Shea]

Sorry for taking so long replying - been a bit busy and a bit distracted.

Anyway I think that when people talk about sample rates wrt digital audio that they presume a higher rate = higher resolution etc. It doesn't all a high sample rate does is extend the frequency range. So you go from 0Hz-44.1kHz through to 0-96kHz to 0-192kHz. It's probably worth stating that the frequency range always starts at 0Hz - you don't get extended low end at a higher SR.

A higher SR in digital can be useful as we've already said as it moves the Nyquist frequency - and this is important because an ADC contains a filter to cope with alias distortion folding down in to the audible range and an anti-image filter on a DAC to cope the production of every image of the baseband frequency.

Since the late 80s ADCs have been designed to operate at 64 or 128 times the base band frequency but over a reduced bit range (4-5 bits) at the modulator. At 44.1kHz baseband a 128 oversampling ADC converter already runs internally at 5.6 MHz. This is done as any noise in the converter can then be spread and shaped over a very large frequency spectrum and the noise is shifted above the audible range. Put it another way a well designed ADC at 44.1 already has sufficient frequency range for mixing and mastering.

Any oversampling by the way has to be downsampled to the final destination rate. For anyone who takes their digital audio to CD the final destination rate is 44.1kHz, Nyquist of 22.05kHz and that provide a full spectrum that adequately covers the human audible range of 0-20kHz. Supersonic noise, including hyper harmonics, is filtered out at this point (there's a stop band filter @20kHz in an ADC). The only way you can retain an extended higher sampling frequency is to maintain the same destination rate - if you record and mix at 96 then everything you do up to and including your final destination must remain at 96.

(It's possibly worth noting here that the issue of hyper harmonics is more problematic than just the destination rate. In order to achieve a recording with accurate hypersonics requires that it is recorded with that intention: you need microphones and preamps capable of capturing the hyper frequency without rolling off and/ or adversely colouring the audio. It's perhaps worth remembering that the majority of mics start to attenuate by 3dB per octave from 20kHz up: preamps may be worse. All your processing must run at the intended extended rate and not introduce any downsampling or stop band. All of the monitoring chain must also run at the extended range. IN the case of many DACs they are designed to run only at 44.1 or 48 and cannot use 96 or 192.)

So to downsample you use a decimator and that contains the anti-aliasing filter- in oversampling the DAC has an anti-image filter. As Ivan says earlier it is the quality of these filters that are key here. What many people (and this has been AB and ABX tested and demonstrated lots of times) hear and think is a better sound at 192 compared to say 48 is not specific to the increase in sample rate but due the quality of the filtering.

Now this may sound like a 192 kHz SRC has a better filter than one at 44.1kHz. Actually it's usually the reverse. The purpose of an ADC and a DAC is to convert the signal accurately and that means with as little colouration as possible. In prosumer models the SRC and the filtering takes place on a single chip. The filtering then has to cope with a range of sample rates and is compromised. Furthermore at low frequencies if you use a high SR there is very little discernable change in the signal between samples. You therefore need a very long filter (i.e. a gentle slope) in order to capture the same accuracy as if you used a slower sampling rate. With high SRs it becomes more and more difficult to design and produce such a filter - all prosumer manufacturers buy and use mass produced chips and these have quite sharp filters in order to keep production costs down. Ironically oversampling, particularly to 192, actually needs a gentle filter despite the extended Nyquist because it is more prone to innaccuracy, distortion and ripple in the pass band but due to costs and compromises the slope is often sharp. SRC to 192 actually results in a degradation of the audio: What people hear as an improvement is this colouration (distortion, ripple etc). Whilst some might like it is actually counterproductive in recording, mixing and particularly mastering because such colouration is not an accurate portrail of the original signal and is likely to be system specific and so not reproducible elsewhere.

In a professional quality ADC or DAC the filtering at Nyquist is on a very gentle slope and in a pro DAC the upsampling is usually done on one chip that sits in the circuit before a specially designed chips with the filtering. The second chip is then set at 88.2 or 96 kHz and does not have to cope with a range of SRs as the first chip sends it a set SR. There is thus no compromise in the filter design due to any need for it to cover a range of SRs.

This though isn't just about filter design - albeit that that is very important. It's also about data size, accuracy and precision. In digital audio accuracy requires that the circuit can track and reproduce the incoming signal properly. In the circuit components however take a finite time to react to the signal and this results in innaccuracy, drift, etc. This becomes more of an issue as frequency and sample rate increases. To date the best compromise between SR and accuracy sits @60kHz and in the real world 44.1kHz is good enough. Go beyond that and a/ the human ear can't discern any real change (ignoring filtering issues) and b/ you are losing accuracy just to extend the frequency spectrum along with an increase in distortion, more data to store and process, a need for a more powerful DPS chip to do the maths and so on.

It is worth remembering whilst we talk about accuracy and distortion that everytime you change the SR you also increase the amount of processing and thus distortion to the original signal. In many prosumer and vst effects the equipment may downsampling at output. Thus a processing chain here may actually be a series of Upsampling and Downsampling as the signal moves from through the chain.

From timing inaccuracy you can then move on further in to a discussion about the increased need for accurate clocking to cope with the increased potential for jitter. Prosumer kit is designed and manufactured to a price point and may well not be good enough to deal with the increase needed to minimise jitter as SR increases. Jitter occurs at both the interface and sampling: in order to reduce interface jitter the clock signal and the audio data need to be separated. Interfaces that are common on prosumer gear like firewire, USB etc don't do this you need professional AES/EBU connections and interfaces to do this.

Furthermore the internal clock on prosumer kit tend to suffer from timing inconsistencies (the clocks aren't very accurate) - this becomes more apparent and more of an issue as SR increases and intermodulation becomes an audible issue long with an increasing noise floor. At 44.1kHz you need a jitter below 25psec to achieve a -120dB noise floor for 20 bit: at 88.2 (ie 2xFS) you need a jitter below 12psec., at 192 you need a jitter better than 5psec peak to peak. If you run a 192 SR with the clock jitter at 25psec peak to peak you will reduce your noise floor by more than 12dB. Put another way for a 20 bit system you'd have gone from -120dB at 44.1 kHz to worse than -108dB at 192. To even get close to the 120dB noise floor of a 44.1kHz system a 192SR must achieve better than 5psec jitter and that needs a considerable amount of electrical shielding and isolation to protect the clock circuit. The cost of such an accurate clock and all the additional shielding places it out of the realm of prosumer kit.

Anyway I could go on at length but I'll summarise instead:

IMHO there is currently no need to use 192kHz as a SR and it does not result in better audio particularly if your final destination is CD. 96kHz is more than good enough and in the vast majority of cases 44.1kHz and 48kHz should be preferred.

You are at GuitarMasterClass.net

Don't miss today's free lick. Plus all our lessons are packed with free content!

Don't miss today's free blues, jazz & country licks. Plus all our lessons are packed with free content!

[Ivan Milenkovic]

Once again, Tony comes to the rescue Thanks for this detailed post, you're a library of knowledge Tony

You are at GuitarMasterClass.net

Don't miss today's free lick. Plus all our lessons are packed with free content!

Don't miss today's free blues, jazz & country licks. Plus all our lessons are packed with free content!

[fkalich]

FKALICH: (Hardware limitations) This just seems impossible now. In five years it won't be. Moore's Law of MicroProcessor advancement marches on. Hardware gets fasters, bus width gets bigger, hard drives get larger, etc. It's just a matter of time. We will wonder how we ever got by with these low sample rates. I remember when 44.1 was "Enough" for most folks and 48k was high end. Now it's 96. Give it a few years.

Here is a abstract to an article talking about the effect I'm on about here. It's called the "Hypersonic Effect". Essentially, encoding and reproducing sounds well beyond the audible limit of most hardware are pleasing to the ear much like a real instrument. It's still a new area of research in the Audio Engineering Society, even though the initial research goes back more than 10 years. This is from the

JOURNAL OF NEUROPHYSIOLOGY
http://jn.physiology.org/content/83/6/3548.abstract

I read a book recently "The Disappearing Spoon", a history of elements of the periodic table. I have forgotten exactly the mechanism, but it described how the silicon switches on a circuit will be replaced with switches operating on a molecular level.

Regarding the study, well they state that there is activity in the Occipital lobe, but that is the visual lobe...the temporal lobe deals with sound. They also state activity noted (and correlated with the above) in two lower brain regions (which would not be conscious regions). I don't know about the psychological testing that they mention, I would have to see more on that to see if it was a convincing test study.

edit: Not wanting to jump to conclusions with my initial and typical cynicism, I looked further into this. This study is very disputed in the scientific community regarding the statistical analysis and methodology, and does not seem to be supported by follow up studies. Even if the study were solid, finding a neurological reaction in no way would imply that this reaches the level of consciousness. As the study is old, and is in this much dispute, I feel very skeptical about both the results and the conclusions.

You are at GuitarMasterClass.net

Don't miss today's free lick. Plus all our lessons are packed with free content!

Don't miss today's free blues, jazz & country licks. Plus all our lessons are packed with free content!

This post has been edited by fkalich: Aug 25 2011, 07:15 PM

[Todd Simpson]

Sorry for taking so long replying - been a bit busy and a bit distracted.

IMHO there is currently no need to use 192kHz as a SR and it does not result in better audio particularly if your final destination is CD. 96kHz is more than good enough and in the vast majority of cases 44.1kHz and 48kHz should be preferred.

Now that is a fine post! Pretty much covers everything and I have to agree again with everything mentioned. If your deliverable is a 44.1 CD or an MP3, or any digital file, you are wasting your time to go beyond 96. Until everything changes including the delivery / playback system, it's pointless.

I read a book recently "The Disappearing Spoon", a history of elements of the periodic table. I have forgotten exactly the Even if the study were solid, finding a neurological reaction in no way would imply that this reaches the level of consciousness. As the study is old, and is in this much dispute, I feel very skeptical about both the results and the conclusions.

It's just one study, only used as a brief example. That's it, nothing more. I used to have this discussion (about the future of audio/sampling/digital) with a guy named Jim Zumpano who is a great engineer here in town. It's more of a coffee talk than anything else. Somehow it seems to have been taken to be a bit more.

You are at GuitarMasterClass.net

Don't miss today's free lick. Plus all our lessons are packed with free content!

Don't miss today's free blues, jazz & country licks. Plus all our lessons are packed with free content!

[Todd Simpson]

Sorry for taking so long replying - been a bit busy and a bit distracted.

Anyway I think that when people talk about sample rates wrt digital audio that they presume a higher rate = higher resolution etc. It doesn't all a high sample rate does is extend the frequency range. So you go from 0Hz-44.1kHz through to 0-96kHz to 0-192kHz. It's probably worth stating that the frequency range always starts at 0Hz - you don't get extended low end at a higher SR.

A higher SR in digital can be useful as we've already said as it moves the Nyquist frequency - and this is important because an ADC contains a filter to cope with alias distortion folding down in to the audible range and an anti-image filter on a DAC to cope the production of every image of the baseband frequency.

Since the late 80s ADCs have been designed to operate at 64 or 128 times the base band frequency but over a reduced bit range (4-5 bits) at the modulator. At 44.1kHz baseband a 128 oversampling ADC converter already runs internally at 5.6 MHz. This is done as any noise in the converter can then be spread and shaped over a very large frequency spectrum and the noise is shifted above the audible range. Put it another way a well designed ADC at 44.1 already has sufficient frequency range for mixing and mastering.

Any oversampling by the way has to be downsampled to the final destination rate. For anyone who takes their digital audio to CD the final destination rate is 44.1kHz, Nyquist of 22.05kHz and that provide a full spectrum that adequately covers the human audible range of 0-20kHz. Supersonic noise, including hyper harmonics, is filtered out at this point (there's a stop band filter @20kHz in an ADC). The only way you can retain an extended higher sampling frequency is to maintain the same destination rate - if you record and mix at 96 then everything you do up to and including your final destination must remain at 96.

(It's possibly worth noting here that the issue of hyper harmonics is more problematic than just the destination rate. In order to achieve a recording with accurate hypersonics requires that it is recorded with that intention: you need microphones and preamps capable of capturing the hyper frequency without rolling off and/ or adversely colouring the audio. It's perhaps worth remembering that the majority of mics start to attenuate by 3dB per octave from 20kHz up: preamps may be worse. All your processing must run at the intended extended rate and not introduce any downsampling or stop band. All of the monitoring chain must also run at the extended range. IN the case of many DACs they are designed to run only at 44.1 or 48 and cannot use 96 or 192.)

So to downsample you use a decimator and that contains the anti-aliasing filter- in oversampling the DAC has an anti-image filter. As Ivan says earlier it is the quality of these filters that are key here. What many people (and this has been AB and ABX tested and demonstrated lots of times) hear and think is a better sound at 192 compared to say 48 is not specific to the increase in sample rate but due the quality of the filtering.

Now this may sound like a 192 kHz SRC has a better filter than one at 44.1kHz. Actually it's usually the reverse. The purpose of an ADC and a DAC is to convert the signal accurately and that means with as little colouration as possible. In prosumer models the SRC and the filtering takes place on a single chip. The filtering then has to cope with a range of sample rates and is compromised. Furthermore at low frequencies if you use a high SR there is very little discernable change in the signal between samples. You therefore need a very long filter (i.e. a gentle slope) in order to capture the same accuracy as if you used a slower sampling rate. With high SRs it becomes more and more difficult to design and produce such a filter - all prosumer manufacturers buy and use mass produced chips and these have quite sharp filters in order to keep production costs down. Ironically oversampling, particularly to 192, actually needs a gentle filter despite the extended Nyquist because it is more prone to innaccuracy, distortion and ripple in the pass band but due to costs and compromises the slope is often sharp. SRC to 192 actually results in a degradation of the audio: What people hear as an improvement is this colouration (distortion, ripple etc). Whilst some might like it is actually counterproductive in recording, mixing and particularly mastering because such colouration is not an accurate portrail of the original signal and is likely to be system specific and so not reproducible elsewhere.

In a professional quality ADC or DAC the filtering at Nyquist is on a very gentle slope and in a pro DAC the upsampling is usually done on one chip that sits in the circuit before a specially designed chips with the filtering. The second chip is then set at 88.2 or 96 kHz and does not have to cope with a range of SRs as the first chip sends it a set SR. There is thus no compromise in the filter design due to any need for it to cover a range of SRs.

This though isn't just about filter design - albeit that that is very important. It's also about data size, accuracy and precision. In digital audio accuracy requires that the circuit can track and reproduce the incoming signal properly. In the circuit components however take a finite time to react to the signal and this results in innaccuracy, drift, etc. This becomes more of an issue as frequency and sample rate increases. To date the best compromise between SR and accuracy sits @60kHz and in the real world 44.1kHz is good enough. Go beyond that and a/ the human ear can't discern any real change (ignoring filtering issues) and b/ you are losing accuracy just to extend the frequency spectrum along with an increase in distortion, more data to store and process, a need for a more powerful DPS chip to do the maths and so on.

It is worth remembering whilst we talk about accuracy and distortion that everytime you change the SR you also increase the amount of processing and thus distortion to the original signal. In many prosumer and vst effects the equipment may downsampling at output. Thus a processing chain here may actually be a series of Upsampling and Downsampling as the signal moves from through the chain.

From timing inaccuracy you can then move on further in to a discussion about the increased need for accurate clocking to cope with the increased potential for jitter. Prosumer kit is designed and manufactured to a price point and may well not be good enough to deal with the increase needed to minimise jitter as SR increases. Jitter occurs at both the interface and sampling: in order to reduce interface jitter the clock signal and the audio data need to be separated. Interfaces that are common on prosumer gear like firewire, USB etc don't do this you need professional AES/EBU connections and interfaces to do this.

Furthermore the internal clock on prosumer kit tend to suffer from timing inconsistencies (the clocks aren't very accurate) - this becomes more apparent and more of an issue as SR increases and intermodulation becomes an audible issue long with an increasing noise floor. At 44.1kHz you need a jitter below 25psec to achieve a -120dB noise floor for 20 bit: at 88.2 (ie 2xFS) you need a jitter below 12psec., at 192 you need a jitter better than 5psec peak to peak. If you run a 192 SR with the clock jitter at 25psec peak to peak you will reduce your noise floor by more than 12dB. Put another way for a 20 bit system you'd have gone from -120dB at 44.1 kHz to worse than -108dB at 192. To even get close to the 120dB noise floor of a 44.1kHz system a 192SR must achieve better than 5psec jitter and that needs a considerable amount of electrical shielding and isolation to protect the clock circuit. The cost of such an accurate clock and all the additional shielding places it out of the realm of prosumer kit.

Anyway I could go on at length but I'll summarise instead:

IMHO there is currently no need to use 192kHz as a SR and it does not result in better audio particularly if your final destination is CD. 96kHz is more than good enough and in the vast majority of cases 44.1kHz and 48kHz should be preferred.

I"m reading through this again and I really think it would make a great addition to the Wiki. I"m gonna see if Fran Agrees. Really great post!

You are at GuitarMasterClass.net

Don't miss today's free lick. Plus all our lessons are packed with free content!

Don't miss today's free blues, jazz & country licks. Plus all our lessons are packed with free content!