Why analogue and not digital?

[ausrock]

You know, you've waffled on so fucking much that I don't even know what it is you're trying to say any more

 

[Dr ZEE]

.... I don't even know what it is you're trying to say ...

Which often is THE 'ultimate objective' of waffling .

 

[Theo K]

One thing that no one here has mentioned yet-

(I used to post here as bloomboy by the way, been really busy with college this year so I've just been lurking)

The reason I, and probably many others on this forum, record in analog is the price. The statement (can't remember who made it) that low cost digital sounds better than low cost analog just isn't true in this day and age. I have a TSR-8 that I bought and maintain in perfect condition for about 500 dollars... and it originally sold for what, something like five thousand? closer to eight or nine thousand in today's dollars? I don't have any experience with truly high end digital, so I can't say I know anything about a possible comparison, but that's exactly it- why should I pay thousands of dollars for a much less intuitive workflow and possibly equal fidelity to what I'm using now? I used to have a MobilePre and experience has definitely shown that the exact same signal chain to tape sounds a hell of a lot better.

-Theo

 

[evm1024]

with all due respects

Folks, We have a blind man here. He has touched his part of the elephant and measured it quite precisely. He has listened to the theories of others and that fit his world views quite nicely. He has devised tests that he can "visualize" and they all support his world view. He is SURE he is right and thus other views are wrong....

Perhaps, some day he will discover that there is a lot more to the elephant.

Perhaps some day he will discover that we have 2 ears and not one and that a teeny tiny random artifact in the left ear and a different teeny tiny artifact in the right ear tells the brain something and that those artifacts are never the same moment to moment.

It will be difficult for him to open his eyes. We will be charged to prove to him that he is not blind. And yet all he has to do is step outside his comfortable box and use his nose. And then he will realize that the stuff he is stepping in does not come from the trunk of a tree.

Till then all comments in this thread are not for him but for the others who read this and have their eyes open.

 

[wado1942]

Hello all, I'm the writer of the aformentioned article that's caused so much controversy and can't help but feel a bit flattered. I admit that my initial experimentation was flawed in that I had only used sine waves as my base line for comparison. However, real music and alternate waveforms, though harder to measure are much more significant for listening experiments. My article also only covered the scope of what happens in a purely digital world. Later experiments would show that a whole new world of problems are created when converting from digital to analogue and vice-versa. But I digress.

Now for specifics

correct me if I am wrong here, anytime you digitally eq, filter, adjust volume, mix etc.. , the said file needs to be reprocessed for every adjustment.

No, not unless you do your edits destructively, which no DAW program I am aware of is set up to do

Whether you process destructively or nondestructively, the same processing happens. A simple change of volume by .1dB requires every single sample to be re-estimated. Cutting 100Hz 1dB requires hundreds of delays to be processed and combined with every sample. No matter if you using destructive or non, 24-bit or 64-bit, you WILL introduce distortion, comb filtering and aliasing.

Why can't digital-o-philes simply say: "digital is perfect" and leave the analog-o-philes alone cooking themselves in their own juices sqeezed from a load of fallacies and outright errors about digital "technology"

Why are digitalphiles always asking for proof and thumbing their nose at whoever provides it?

"Do digital oscilloscopes differ in their accuracy from analog oscilloscopes?"

Yes they do. Greately as a matter of fact. I prefer analogue scopes for some things and digital scopes for others. But it takes about 8x the bandwidth for a digital scope to get the same accuracy for high frequency analysis that an analogue scope does. Then again, I hate following that crawling dot on analogue scopes to examine low frequency stuff.

I could bang out a simple ideal digital EQ filter in 15 minutes, which no one would probably use on account of its extremely boring sound.

Not just boring but incredibly harsh and unnatural. Trust me, I've put up with a decade of bad digital EQ. It's only been recently where some decent ones have been put to use. It's not just a matter of emulating artefacts of analogue EQs. In fact, my favorite digital EQ could never easily be created in the analogue world. Likewise, none of my analogue EQs could ever be truthfully recreated in the digital world. They're just 2 different beasts. But this digital EQ I almost like is incredibly high resolution, incredibly accurate, incredibly power hungry and incredibly expensive.


Now, I'm so sick of people using the Nyquist Theorem to prove that digital doesn't miss any information. For starters, Harry Nyquist created his theorem in 1928, well before we had a good working knowledge of digital theory. His statement was simply that in order for a FREQUENCY to stay that frequency, the sample rate has to be at least double that. That's to say nothing about adding distortion below that limit. My recent experiments involving square waves showed significant audible distortion as low as 2KHz. Massive phase rotation, aliasing as high as 6dB below the fundamental frequency and increased output level is normal in this condition particularly as you get NEAR the Nyquist limit. Oversampling DACs further increase the relative peak output level and add even more distortion (because it tries to round off the sharp edges of the waveform) because the convertors were designed to accurately recreate sinewaves only. Your DACs have to work very hard at filling in missing information at even 3Khz. If you're using an 8X DAC, the processor takes the original samples, fills in 7 new samples in between and uses spline curves to essentially "GUESS" what was there based on the previous and next samples. In the case of a nice predictable sine wave, the outcome is relatively accurate to the input signal. In the case of complex music and the like, the errors are greater than the accuracies. Now if nothing above 44.1KHz was necessary, then why do ALL modern DACs use oversampling? Because those hard, brittle, steryl sounding digital recordings of the 80s proved to us that 44.1KHz was NOT enough and we had to figure out how to fill in the blanks.


Furthermore, people neglect the word-length's contribution to aliasing. An 8-bit recording at -6dB nominal shows significant audible distortion and nobody will argue with that. Now you extrapolate this into the 16-bit world and the same rule applies. Every detail at -54dBfs will have the same distortion in it as the -6dB 8-bit sound did. -54dB is where all the ambience, all the sublety, all the breath of music lies. While a 1/4" tape will typically have a signal to noise ratio of about 60dB without noise reduction at normal operating levels, most of this noise falls below the range reproduced by most systems. 66dB is more common for reproducable ranges. But still, this is noise and not distortion. Detail can be heard as much as 20dB below the noise floor giving you a usable dynamic range of 88dB. Of course this is ignoring the crest factor of the music. My 1/4" deck has a good 15dB of headroom before noticable saturation giving for a total dynamic range of 103dB. Add in Dolby SR and you extend that another 20dB. This is how recordings from the 70s and early 80s, though usually on its 5th tape generation by the time the listener gets it, had so much warmth and ambience. I dare you to get an original CD release (not the remaster) of The Police "Synchronicity" and tell me it doesn't sound fantastic. The CD master was made from a 3rd generation 1/4" tape. You can hear the rooms' effect on the drums, the guitar, the bass (to a lesser extent) and the vocals. Now listen to an original master of "Brothers in Arms", one of the first "DDD" releases and while it's a great recording, it's very brittle and dry sounding even though there's plenty of reverb and bass. Compare that to the 5.1 remix of the same album (done in 2005), LOTS of ambience. You can hear the oscillator leakage on the Hammond B3, the room's effect on the voice even though he's very close to the mic. The only technical difference really was the use of oversampling DACs and 24-bit 96-KHz mixdown medium. It was still mixed on an analogue console, mixed to digital, mastered on analogue equipment to digital. Of course it's a modern master so it's so heavily compressed, the volume actually drops when the drums start. Very dissapointing but whatever.

On a final note, the specs that digital pushers advertise, they REALLY cheat. They use a 1KHz sinewave digitally generated at -.1dB or so to test the S/N ratio and distortion. But the reality is, you NEED to keep your levels around -18dBfs or so to keep both the electronic and digital componants happy and this really changes those specs. Then there's the concept of the fact that their equipment is designed to recreate sine waves faithfully. As you move towards more realistic forms of sound, things change. The same rules no longer apply. Furthermore, the conversion to digital never properly filters out super-Nyquist frequencies. Even at that, the more successful (at filtering) converters have a tendancy to ring for a few ms before and after the occurrance of an HF event. The best converters to date use analogue 5th order filters BEFORE the digital filters. This softens the high end a little but drastically reduces distortion once in the digital domain.



Last words before I go, scientists make mistakes because they can't cover all bases. Apollo 1 was a failure despite extensive testing simply because they didn't think of a certain kind of condition.

People say that tubes sound better and nobody could find any plausible scientific evidence as to why until only about 5 years ago.

Prior to oversampling digital converters, people said digital sounded harsh even though Ole' Harry said it was fine. On paper, digital WAS better, as far as noise and distortion go. With modern advances, we discovered our ears told us something our scopes couldn't. Now that one problem got solved, we still have some more problems that our scopes don't show yet. Digital may one day come of age, but it's not today. There's too much that our scientists haven't even thought to cover yet.

 

[evm1024]

Well said!

Thanks, very well said.

--Ethan

 

[Tim Gillett]

Ethan, have you read the write ups about "The Emperor's new sampling rate " double blind tests? I thought of you when I read them.

Cheers Tim

 

[cjacek]

Is the Revox as good as a true mastering-grade deck?

The poster, who offered up the advice, in the previously mentioned 'violin' thread, gave details as to the tape deck but you had dismissed it as 'comparable to consumer grade digital audio' which is utter non-sense - one is low sampled digital audio and the other is not sampled, a high resolution, linear analogue recording].

In fact, he gave, what I would call one of the no-bullshit, straight forward answers on the subject, not only with regard to using the highest sample rate possible but also the alternative to use a half track, high speed open reel deck.

that was on a thread where the OP (a working musician) wanted a $500 total budget solution for producing CDs for audition. How does the Revox possibly fit in that scenario?

You can very easily find a half track deck [not necessarily a revox], a mic and tape for less than $500. Even with minor servicing, it'll still be within budget.

Let's discuss sample rate. [snip]

Sorry to have edited out much of your tech talk, about sample rate and such but I too had a very hard time discerning the content.

I think I know where the problem lies.... It's like trying to describe an object to a certain someone. I try to do it via pictures and you via a mathematical / scientific framework.

Let me set this argument into better focus and restate why analog is objectively better than digital [for sound production, multi-tracking etc..]:

Analog doesn't use sampling [its 'sampling' is effectively infinite] & therefore it is future proof;
Analog is the superior storage medium by far
Analog will outlive [as it already has] every technology past and present
Analog is the easiest and most intuitive technology to master

To put into perspective, a tape recorder from 2 decades ago, with some TLC, maintenance and attention, will continue to not only be serviceable, decades from now but will also be able to outperform and outlive every technology, at least for our lifetime.

Analog is the only stable, predictable and true high resolution recording technology in existence and it has been since its inception.

I note a disturbing tendency here to completely reject any attempt at scientific measurement whatsoever. That is a travesty.

This has been done, with laughable and disingenuous results. Case in point, it was used successfully, over the past decades, to discredit and destroy analog recording and it still continues to this very day. That is a travesty.

I am also increasingly disturbed by frequent claims from tape aficionados to possess not only above-average but ultrasonic hearing.

Again, you're going by specs which can be measured or mathematically figured out, such as frequency response etc....... I'm talking about something else, which is missing in sampled sound, even if it specs out better [on paper] than a linear analogue recording. I'm not particularly interested in the science behind it but there is research to concede that point.

To be perfectly blunt and rather simplistic about it, analogue [in / out] is the way we hear and perceive [that's nature] but when you take a source signal [the 'in'] and you mathematically encode it, sample it, chop it up [to be rude about it ], it is still an incomplete / distorted picture [of the 'in'], no matter if the output has now been converted into an analogue signal. It's still the analogue of a sampled sound. Frequency range, flat frequency, S/N ratio, doesn't mean absolutely anything, cause it's all about the missing 'in-betweens', that sampling, unfortunately causes. That, I believe, is the reason why many find digital sound objectionable. That plus the brick wall limiting of frequencies which should all be present, whether heard or perceived, just as nature intended.

-----

 

[cjacek]

Hello all, I'm the writer of the aforementioned article

Welcome! Great post. I hope you stick around.

-----

 

[clevodrummer]

good points 4 analog good points 4 digital

i use both.

 

[wado1942]

Thanks for the kind words all.

I just thought to add in something else in regard to digital EQ vs analogue. I never said analogue EQ didn't involve phase distortion. PHASE SHIFTS ARE WHAT MAKE IT WORK! But using analogue componants, say an inductor to roll off the high frequencies, there's no delay but rather a phase rotation at certain frequencies. So it's coherant, coherant, coherant until just before you reach the calculated frequency of the filter. At that point, you introduce a 90 degree rotation of voltage vs current. The current change is still instantaneous but the voltage lags just a little. As you continue reading past the key frequency, you approach 180 degree rotation so the higher the frequency, the closer to a 180 degree rotation. As you get closer to that 180 degree rotation, the signal drops as the current/voltage cancel eachother. But again, there's no actual delay. One of the many oversights so many scientists have. They don't differentiate between phase shift, delay and polarity. You have no idea how many manuals I've seen that state that an inverting switch causes a 180 degree phase shift. Well, no, it doesn't. It inverts the signal. If you're looking at just a 1KHz sine wave on a scope, it may LOOK like a phase rotation but they're 2 completely different things.

Digital on the other hand uses descrete delay. So by delaying the signal by 50nS, you will get what looks like a 180 degree phase rotation at one frequency if you're looking at a sine wave on a scope. But above that frequency, it'll appear to go back to 0 degrees again. As soon as a transient passes, or any other combined signal for that matter, it becomes very clear that the time domain has been changed, not a simple phase rotation. There may be what looks like a 180 degree rotation at one frequency, but above that, you start to go back to a 90 and finally a 0, then 90, then 180 (though now you're several cycles apart between the direct and delayed signal. This is not at all what happens in the analogue world because a 1st order filter will never shift more than 1/2 a cycle. Anyways, that continually changing cancellation/reinforcement from the digital delay causes a rippling comb filter effect that spreads across the entire spectrum. So you have to add more delays to try and tame them one at a time. A simple high pass filter can require at least 500 delays to do what a simple 30 cent capacitor can do in the analogue domain. But that's just addressing the frequency/intensity portion of the sound. As I said earlier, a transient passes and you can clearly tell that the signal is no longer time-coherant. You will find echoes following the transient. In the case of linear-phase digital EQ, the process is bidirectional. That means, some delays are applied going forward through the wave form, then identical delays are applied going backward through the wave form. This causes the percieved phase rotation to stay somewhat constant throughout the spectrum except we've caused a new problem. The echoes that would have followed an event now preceedes it as well. So a snare drum hit filtered by a linear-phase EQ will have a little pre-ringing before the snare hit, then you get the usual ringing that happens after the snare hit. This can be very alien to the average listener.

This is why when I work in the digital domain, I think very carefully about my poison. If I'm making very steep corrections (narrow Q), I'll lean towards linear EQ because otherwise, narrow Qs tend to introduce a lot of phase distortion. However, with wide Qs, phase distortion is not a huge issue and thus I'll use starndard Infinite Impulse Response EQs to avoid pre-echoes.

Cool thread BTW, I love discussing this stuff and hearing other people's ideas.

 

[evm1024]

Interesting read

Ethan, have you read the write ups about "The Emperor's new sampling rate " double blind tests? I thought of you when I read them.

Cheers Tim

Thanks Tim, An interesting read.

What struck me quite strongly was the statement that "women ...getting 37.5% right".

If the 2 sampling rates were indistinguishable then we would expect a 50% correct, 50% incorrect answer.

Back to the women.... If they were able to tell which was which 37.5 of the time then they were detecting a difference 12.5% of the time! The sign does not matter. It is still a detection.This is worse for those who say there should be no detection....



--Ethan

 

[wado1942]

That's very true. Of course probability is involved here which is why you want to use a large cross section of people. The blind test I conducted in my communications class 10 years ago between analogue and digital showed me some interesting results. I asked for people to differentiate. The only guesses as to which were digital out of 40 people were wrong. In other words, they liked the way the analogue example sounded and thus thought it was digital based on the media hype.

Of course Alesis conducted a study comparing 20-bit and 16-bit when the 20-bit XTs were being prototyped. They tested hundreds of people blind and about half said the 20-bit sounded better while the other half couldn't tell the difference. Granted ADATs with their many whirring motors always had jitter issues.

 

[evm1024]

16,20,24 ...

An ex wife of mine could barely tell the difference between stereo and mono.... I have great faith in a future digital which is uncompromisingly musical. Not sure what sample rate and bit depth that might end up at. This faith is not constrained by ROI calculations or computational requirements. Having started computing at the tail end of the tube computer era I have seen 24 moore's law periods go by with the resulting 16 million times increase in computational power (at least). There was a time where a single CD contained more data than was stored on all the worlds computers and a single second of the computers here at work have more cycles than the entire world had in a year.

Needing a terabyte per minute does not phase me as I sit here with 160 TB in my computer room and a 16 GB flash in my hand. That is the least of my worries.

What I do think about is the uncorrelated spatial information that is generated by the random generation of quantization errors of that 6 to 8 bit amplitude signal you spoke about. We really do not have a good way of measuring this and in general the whole notion of uncorrelated sounds and their interaction in the brain is not considered or considered important in design engineering.

It is not a question of hearing a tone or not or hearing the differences in 16,20 or 24 bit depths. There will always be those who do not hear and those who are acutely aware. It is a question of long term listening patterns.

I would rather see a test where we looked at the listening patterns of a large group of people for a year or two who some are given 16 bit systems and others 24 bit systems . Same music, same room, speakers etc.....

The complexity of human audio perception is not subject to a few basic tests. As I recall we are quite capable of putting in tones that do not exist in the program material based on the harmonic content. This correlated sound is inserted by our brains much as we don't see our blind spot where the optic nerve attaches to our retina....Optical designers do not even think about this....and yet it is there (or rather not there).

And yet there are people who hear the difference between a digital recording and an analog recording. They cannot tell you what to measure but they hear a difference. And we do not have a good understanding of the difference. Some look at first order effects and say that everything else does not matter....

Regards, Ethan

 

[mshilarious]

Hello all, I'm the writer of the aformentioned article that's caused so much controversy and can't help but feel a bit flattered.

Feel more flattered, as yours is the ONLY post thus far to actually address technical issues.

I admit that my initial experimentation was flawed in that I had only used sine waves as my base line for comparison. However, real music and alternate waveforms, though harder to measure are much more significant for listening experiments.

That is fair enough. The information in that post seemed around seven or eight years old to me, so I am encouraged you have better information.

I agree that more complex waveforms are required to poke around what really happens in a digital converter. And so I and most other designers I would expect are considering these criteria.

My article also only covered the scope of what happens in a purely digital world. Later experiments would show that a whole new world of problems are created when converting from digital to analogue and vice-versa. But I digress.

OK, we can and I have distinguished from a pure digital world, and the implementation of those principles.

As we move on to your specifics below, let's keep in mind to test both the pure digital concept and its real-world implementation.

Whether you process destructively or nondestructively, the same processing happens. A simple change of volume by .1dB requires every single sample to be re-estimated.

In 32 bit float calculation, the precision required to do this with complete accuracy well below the point of inaudibly has been frequently demonstrated. In fact, anyone with a properly functioning DAW can try it with a series of gain changes offset by attenuation on the master bus. It should be noted that certain DAWs apparent don't accurately label the gain changes on different controls, and that could throw you off a bit.

Anyway, the DAW I use (Wavelab v5) performs this test to a complete null, even when the inter-process peaks are driven well above 0dBFS.

The point about nondestructive editing is still important. Both digital and analog recording formats allow nondestructive mixing and equalizing, and it would be foolish to do otherwise. Digital also offers nondestructive edits; analog can't manage that without copying the source material, which cannot be done without adding some amount of distortion and noise, small though it may be.

In the real analog world, you can't stack up an arbitrary number of offsetting gain changes without suffering a severe penalty in terms of noise or headroom. Hence, no one works that way, whether analog or digital. I mean you could in digital, but what's the point?

Cutting 100Hz 1dB requires hundreds of delays to be processed and combined with every sample. No matter if you using destructive or non, 24-bit or 64-bit, you WILL introduce distortion, comb filtering and aliasing.

At what level? If there is comb filtering, that can be easily measured. Post some sample .wavs if you don't mind, and describe the EQ you used to generate that result so I can repeat it.

If you take a full-spectrum signal (pink noise, white noise, program material all works about the same), copy the signal, delay it a few msec or usec, attenuate or boost it, flip phase or something. Anything like that. You will see very large and very audible comb filtering effects. But if you try to use a reasonably designed digital EQ (which is just about any plug I didn't design), you only see the intended change in the signal.

You seem to be implying all minimum-phase digital EQs will generate the former and not the latter desired result, but that is something I find very difficult to measure. In fact I can't measure any change in other than the intended EQ at all. This could be because they are very small, say less than 0.1dB in random places in program material. We would then need a test of audibility to demonstrate that it mattered.

Let's also reiterate that all analog EQs generate phase distortion. Because that's what they are designed to do! It is not tenable to suggest that the desired digital distortion is bad, but the desired analog distortion is good. That is merely a preference for different types of distortion.

Yes they do. Greately as a matter of fact. I prefer analogue scopes for some things and digital scopes for others. But it takes about 8x the bandwidth for a digital scope to get the same accuracy for high frequency analysis that an analogue scope does. Then again, I hate following that crawling dot on analogue scopes to examine low frequency stuff.

That's a good answer! I like how the good Dr. criticized me for not questioning my tools, when I had indeed questioned them but he had not answered.

Still, I don't know if anybody makes an analog o-scope anymore. I don't need one (A or D) for what I do anyway . . .

Not just boring but incredibly harsh and unnatural. Trust me, I've put up with a decade of bad digital EQ. It's only been recently where some decent ones have been put to use. It's not just a matter of emulating artefacts of analogue EQs. In fact, my favorite digital EQ could never easily be created in the analogue world. Likewise, none of my analogue EQs could ever be truthfully recreated in the digital world. They're just 2 different beasts. But this digital EQ I almost like is incredibly high resolution, incredibly accurate, incredibly power hungry and incredibly expensive.

Well, I don't mind the UAD stuff, and some of that has been out, what, 5 or 6 years? It was good enough for Mark Linnett to use on Smile, which is easily one of the best recordings I've ever heard. Better than the Beach Boys vinyl I have, and better than the CD remasters of their analog catalog. He deserved the Grammy for that. He used a lot of analog stuff too. Actually I think he used everything! Good article here:

http://www.soundonsound.com/sos/Oct04/articles/smile.htm

I tend to use the Pultec somewhere on every mix. I like the Helios too. There is no analog EQ at a similar price that I've used that is anything close. Plus once you consider the per-instance price . . .

My recent experiments involving square waves showed significant audible distortion as low as 2KHz. Massive phase rotation, aliasing as high as 6dB below the fundamental frequency and increased output level is normal in this condition particularly as you get NEAR the Nyquist limit.

Is this a problem with the digital theory or a particular converter's implementation? May I have specifics so I can reproduce these tests.

In the case of complex music and the like, the errors are greater than the accuracies.

Fourier's theorem states that any complex waveform can be shown to be comprised of sine waves. It is true that a single sine wave isn't going to suss out all of the characteristics of a circuit. But program material cannot easily be used, becase the results can't be easily analyzed. This is because in the sample length that a Fourier transform requires for the desired degree of resolution, the program material's content will overwhelm any errors present.

So, we have to compromise and use a more complex series of sine waves that allow us to see the distortions. I am not telling you anything you don't know, this is background for other readers.

As I said earlier, designers do use such complex test signals. I would enjoy seeing your test methodology, since it could be useful not only for evaluating digital theory but different brands of converters.

I tend to use three types of signals; first, a series of four high-frequency sine waves with different intervals; second, a series of wide-ranging sine waves (for example, 200Hz, 1kHz, 5kHz, 20kHz) that are phase aligned, and finally, short bursts of various frequencies.

Now if nothing above 44.1KHz was necessary, then why do ALL modern DACs use oversampling? Because those hard, brittle, steryl sounding digital recordings of the 80s proved to us that 44.1KHz was NOT enough and we had to figure out how to fill in the blanks.

OK, you are now arguing against 20 year old digital implementations. Oversampling was well discussed in the Lavry whitepapers. Also, 44.1kHz is a data rate, not a conversion rate. No modern A/D converter uses a 44.1kHz sample rate; it is constructed from a stream of much high sample rate data, for a similar reason that DACs using oversampling. Again, it's all in Lavry.

However, 44.1kHz is an adequate if not ideal data rate. Lavry argues for a minimum 60kHz rate, but no more than 96kHz. I have tested his theories and arrived at his result.

Furthermore, people neglect the word-length's contribution to aliasing. An 8-bit recording at -6dB nominal shows significant audible distortion and nobody will argue with that. Now you extrapolate this into the 16-bit world and the same rule applies. Every detail at -54dBfs will have the same distortion in it as the -6dB 8-bit sound did.

You can say that -54dBFS is required for ambience, and that's true. But the distortion is something like -50dB below the signal, which puts it in the -104dBFS range, and thus extremely difficult to hear. If you turned up the -54dBFS peak 16-bit signal to -6dBFS, you'd have the same result as your 8-bit example. This is why nobody in their right mind would record 16 bit at -54dBFS peak. In fact, in the days of 16-bit converters (mid 1990s and earlier), people did make a point to hit the converters fairly hot to preserve dynamic range.

24 bit converters have no such trouble. I mean if you are an incompetent engineer, you can always find a way to break something. I guess somebody could walk in and record a track on a 24 bit converter with peak level at -102dBFS, but then I could walk up to a 24 track deck and spill a reel of tape on the floor too

OK, but before we leave this topic, two more points: first, you are guilty of an error you warned against earlier: overreliance on a single sine wave to describe a system. It's true that you want good performance with a simple signal. But it is also true that distortion at -104dBFS with normal listening levels (say 0dBFS = 94dBSPL) with a signal at -54dBFS (44dBSPL) is really hard to hear. It is a bit easier to hear than those figures imply, since it's not a single distortion product but multiple products, it would be a higher level than -10dBSPL, something more like 5dBSPL if I did the math right).

So what's the problem? Repeat the experiment with a more complex wave, and the distortion products are greatly diminished. In fact, Ethan Winer used a similar argument against the need for dither, that any real-world signal was sufficiently complex to in essence be self-dithering.

While we're on the topic of dither (whether as a process or Ethan's real-world view), it prevents this type of distortion from occurring when bit depth is reduced, so that quantization distortion does not occur at levels audible above noise.

I did quite a bit of work both with test signals designed to simulate simple and quiet real-world signals. Generally this involved a simple overtone series, such as a flute, recorder would have, and a low level of white noise designed to simulate acoustic and electrical noise. With such signals, I could demonstrate quantization distortion at a very low level, potentially but not easily audible (this is essentially a demonstration of a digital 'flaw', for anyone not following this).

But to date I haven't been able to reproduce this with a real-world signal. Thus I have to admit Ethan is probably right; that is, although quantization distortion can be easily demonstrated with sine waves, it's incredibly difficult if not impossible to generate audible quantization distortion on an actual recording.

Therefore I have confidence, even though I very recently thought otherwise, that your scenario above is not relevant to any modern digital audio recording. If you are still comparing analog to a recording done poorly on a 16 bit converter, I have no (further) response to that

I dare you to get an original CD release (not the remaster) of The Police "Synchronicity" and tell me it doesn't sound fantastic. The CD master was made from a 3rd generation 1/4" tape. You can hear the rooms' effect on the drums, the guitar, the bass (to a lesser extent) and the vocals.

I believe I have the original release, let me check . . . well, it's the 1995 release, stated as remastered on a 20 bit converter. I also have the LP Singles, I did a comparison and ignoring what seem to be severe deficiencies in my turntable (working on that), I believe the production is essentially the same.

All that to say I have never contested any specification of any analog recorder. On the contrary, I suspect engineering departments used to have more sway over marketing than they do these days. But I accept all manufacturers' specs unless I know them to be unreliable. To date, that has only happened with one (rather new) manufacturer, and they don't make a recorder.

As I mentioned earlier, I have a better variety of Beach Boys recordings in various formats. I have to say I really think Smile sounds better than any of them, even though it has the handicap of missing two Wilsons. Most of those were recorded or remastered by the same engineer, Mark Linnett.

Now listen to an original master of "Brothers in Arms", one of the first "DDD" releases and while it's a great recording, it's very brittle and dry sounding even though there's plenty of reverb and bass. Compare that to the 5.1 remix of the same album (done in 2005), LOTS of ambience. You can hear the oscillator leakage on the Hammond B3, the room's effect on the voice even though he's very close to the mic. The only technical difference really was the use of oversampling DACs and 24-bit 96-KHz mixdown medium. It was still mixed on an analogue console, mixed to digital, mastered on analogue equipment to digital. Of course it's a modern master so it's so heavily compressed, the volume actually drops when the drums start. Very dissapointing but whatever.

I don't quite follow this vis-a-vis the Police . . . you are saying that more recent converters are much better than older converters? I would agree. It's hard to know exactly what to attribute the better later remaster of Dire Straits to though--it is a better analog front-end, a better engineer, a better A/D, or the sample rate? Or some of all of the above? I don't know how to isolate those factors once the recording is done. I'd have to be there to tell the guy to burn me a 44.1kHz version for more direct comparison.

I will say again, I have never argued that 44.1kHz is perfect, just that it's pretty good. Audiophiles might want a little higher, 96kHz wasn't quite necessary but it doesn't hurt anything.

On a final note, the specs that digital pushers advertise, they REALLY cheat. They use a 1KHz sinewave digitally generated at -.1dB or so to test the S/N ratio and distortion. But the reality is, you NEED to keep your levels around -18dBfs or so to keep both the electronic and digital componants happy and this really changes those specs.

Well, I don't know who those digital pushers are. If you stroll over to the Mixng & Mastering or the Recording boards, all the old-timers ALWAYS make a point to tell people not to track hot. And all gear is spec'ed at its maximum range. Like earlier when you talked about hitting tape really hot to increase its dynamic range. That isn't the full story either; first, it's an artistic decision to saturate the tape. Second, you need an analog front end that can push those levels without having its own problems with distortion. On the bottom-feeding end of the market, that isn't very common.

The good news is, I guess, from my experience the uninitiated can't understand specs, and the more knowledgeable know what they actually mean.

Prior to oversampling digital converters, people said digital sounded harsh even though Ole' Harry said it was fine.

OK, Harry had one very important criterion: a bandwidth-limited range. It took designers a while to figure out the best way to execute that. It took analog recordings, what, 80 years to stop sounding like ass? I think digital is probably 95% of the way there. That last 5% is simply making the high-end performance available today affordable for the masses.

Which presupposes that manufacturers will ever put money into analog front-ends, which recent experience demonstrates is unlikely

I thank you for your well-reasoned response.

 

[mshilarious]

This is why when I work in the digital domain, I think very carefully about my poison. If I'm making very steep corrections (narrow Q), I'll lean towards linear EQ because otherwise, narrow Qs tend to introduce a lot of phase distortion. However, with wide Qs, phase distortion is not a huge issue and thus I'll use starndard Infinite Impulse Response EQs to avoid pre-echoes.

Cool thread BTW, I love discussing this stuff and hearing other people's ideas.

Again I thank you!

I don't like making steep changes at all. Sometimes one has to . . . I would prefer to move or change a mic or an instrument or a musician But when it's real bad, I try to break out the linear-phase multiband. That sounds horrible, but used right, it's an EQ that's only on some of the time

My overall message is simple: there really aren't any excuses anymore. If my recording sucks, it's not because I didn't use analog, or if I used analog, because I didn't use Dobly It's because I suck!

 

[mshilarious]

In fact, he gave, what I would call one of the no-bullshit, straight forward answers on the subject, not only with regard to using the highest sample rate possible but also the alternative to use a half track, high speed open reel deck.

But it could be a bad idea to use 192kHz. Remember the price range. 192kHz could actually sound measurably worse than 96kHz. It would depend on the box itself, but the theory is there.

You can very easily find a half track deck [not necessarily a revox], a mic and tape for less than $500. Even with minor servicing, it'll still be within budget.

I believe that, but the OP needed a CD as a final product, so you need to include that in your budget. If you're going to go with Soundblaster-quality conversion, why bother with the interim tape step? You'd have warm-sounding violin covered with digital ass.

See? Digital can sound really bad! I can admit that freely! But SBs sound really bad mostly because of terrible analog front-ends.

Sorry to have edited out much of your tech talk, about sample rate and such but I too had a very hard time discerning the content.

Everybody has to edit responses, no apology required.

Analog doesn't use sampling [its 'sampling' is effectively infinite] & therefore it is future proof;
Analog is the superior storage medium by far
Analog will outlive [as it already has] every technology past and present
Analog is the easiest and most intuitive technology to master

Some of this I address two posts ago, but you bring up a new argument too: the Steve Albini argument about archiving. That's a good argument. There are ways to archive digital data, but good engineering practive in that respect is often lacking.

To put into perspective, a tape recorder from 2 decades ago, with some TLC, maintenance and attention, will continue to not only be serviceable, decades from now but will also be able to outperform and outlive every technology, at least for our lifetime.

I cannot predict the future. Digital technology does advance quickly and sometimes leaves behind old formats. I can still read all of my digital files and play all analog recordings I have, but again good engineering practice will be required to manage a digital library.

This is another aspect of the Albini argument, and a reasonable one.

This has been done, with laughable and disingenuous results. Case in point, it was used successfully, over the past decades, to discredit and destroy analog recording and it still continues to this very day. That is a travesty.

Once again, I have NO DESIRE to discredit analog recording.

 

[mshilarious]

Regards, Ethan

Ethan, I really wish a consumer 24 bit format had caught on. It might have alleviated the volume wars, for one. That's because before you drop to 16 bit, you have to increase your working level to preserve dynamic range. That is, if you are smart and work around -18dBFS in 24 bit, you will increase to -0.1dBFS (or a little less) peak before going to 16 bit.

The gain change itself isn't the problem; I think it is the temptation. First you say, ah what the heck, I don't want to bother with exactly figuring out peak level, I'll just set a limiter to trim a hair off it. No real harm there, maybe even gain some punch.

But then you say, hey I've got this limiter on, let's juice it a bit.

Next thing you know it's 10 years later and there's a new Grammy category for best white noise recording

But for whatever reason . . . the high-res formats are hated

I am probably being naive . . . in the world of mp3 players, the volume wars were inevitable.

 

[Tim Gillett]

Thanks Tim, An interesting read.

What struck me quite strongly was the statement that "women ...getting 37.5% right".

If the 2 sampling rates were indistinguishable then we would expect a 50% correct, 50% incorrect answer.

Back to the women.... If they were able to tell which was which 37.5 of the time then they were detecting a difference 12.5% of the time! The sign does not matter. It is still a detection.This is worse for those who say there should be no detection....



--Ethan

Point taken. Women apparently (on average) have better hearing in some respects than men and maybe that's related to the reported tendency to observe a some sort of a difference for some of the time.

But it doesnt help those who say the difference is obvious and profound, and that CD audio is crap. I dont need to repeat here that that line has been peddled VERY strongly on this forum, and for a long time. Anyone who dared to offer an alternative opinion was treated very poorly. There's no need to cite examples. They would fill a book.

I was impressed that these guys actually took the time to conduct the tests. To me properly conducted real world double blind listening tests are worth a whole lot more than words, words and then more words, and it seems surprising that such an obvious approach should have taken so long to be finally conducted. Sure, it would have costed time and money.

I hope those who disagree with some aspect of the tests get to it and set up their own. Why be afraid of the truth, which ever way it comes out?

I feel that for some people who for so long have strongly asserted a view, the real problem is not to do with audio but ego and saving face before their friends. It gets to a point where backing down gracefully is too much of an embarrassment, and so one sticks to "the party line" because it's easier than admitting one has been wrong, and for such a long time.

Cheers Tim

 

[Dr ZEE]

I hope those who disagree with some aspect of the tests get to it and set up their own. Why be afraid of the truth, which ever way it comes out?

I see no reason for being afraid of the ruth.
If you set up a listening test for let's say 100 of individuals and ask then to "determine" something based on what they hear. And, let's say, the results are: 99 of those individuals have determine "it" incorrectly and 1 of them have determined it correctly.
So, based on the results of your test, the truth is this: 99 of the individuals that you've tested have determined "it" incorrectly and 1 of them have determined it correctly. That is where the truth starts and that is exactly where that truth ends. Plain and simple.
Your test DOES NOT and I repeat DOES NOT provide you with any more of a truth, nor with any other truth, nor with any extension nor expansion of the one and only true fact that 99 of the individuals that you've tested have determine "it" incorrectly and 1 of them have determined it correctly.

So? What can be so scary about the truth? I see nothing scary about the truth as long as you actually stick to it.

I feel that for some people who for so long have strongly asserted a view, the real problem is not to do with audio but ego and saving face before their friends. It gets to a point where backing down gracefully is too much of an embarrassment, and so one sticks to "the party line" because it's easier than admitting one has been wrong, and for such a long time.

You can feel things, but since the OBJECTIVE TRUTH is what you "really" after, than you must recognize, that feelings do not provide you with the truth. That is not to say that there's something wrong with expressing you feelings.
***********

Feel more flattered, as yours is the ONLY post thus far to actually address technical issues.

Hmmmmm. Looks like mshilarious ignored "medical advice". That's cool. Be it so. So why bother to even going through 100s lines of waffling? - Good question.
One thing I'd have to address though:

The good news is, I guess, from my experience the uninitiated can't understand specs, and the more knowledgeable know what they actually mean.

There's even BETTER news, which may be actually a NEWS for mshilarious, and it is that there is one more "category of viewers of specifications" - those who ALSO KNOW what those specifications DO NOT MEAN, those who actually have some knowledge about what knowledge actually is, where knowledge comes from and where knowledge is applicable.

/later