Can someone explain this entire audio resolution thing?

[keyman_sam]

There was a discussion about some sampling rate being at 44Khz or somethin'. i dont understand what PART of the synth is that resolution.

 

Can someone explain all the things that are in a ROMpler that are associated with this audio resolution thing?

 

Thnkx,

[Yoozer]

Generally 2, I think. The stored samples (size) and the DACs (Digital to Analog converters). Of course there's the filters and envelopes you have to deal with, too, but still.

[Artur Meinild]

I can think of at least three: sample resolution, processing resolution and DAC resolution. It appears that some Roland modules have a sample resolution of 16 bit/44 KHZ, a processing resolution of 16 bit/32 KHz and a DAC resolution of 18-20 bit 44-48 KHz (depending on model). Furthermore, the XV-5080 has 24 bit reverbs according to the specs, so that would add a fourth element to the equation (it also has 24 bit converters though).

[keyman_sam]

Exactly how does this work?

 

I mean, how does this affect the sound? If you pass in something at 44Khz, no matter what you do, its 44Khz at its best, no?

 

[J3RK]

If you are just passing raw data through yes, but if you say, take a 16/44 sample, and run that through an internal process (like say an effect or something) that is done at 24/96, the effect of processing the audio at the higher internal rate can sometimes make it sound better, (or at least different.) It also can depend on how good the sample rate conversion is. In some instances, a synth will do all of its internal calculations at a 32 bit depth, so very little changes quality-wise when various things are done to the waveform. Then it is down-sampled to whatever the DACs are capable of. This yields a higher quality output than say keeping the data at 16/44 all the way from start to finish.

[DINpluggedIN]

Originally posted by keyman_sam Exactly how does this work?

A question you didn't ask, but may be important, is "What does this resolution spec mean to me?".

 

We need to take at least two samples at any frequency in order to avoid a type of distortion called "aliasing". Sampling at 44kHz was chosen because that would theoretically allow recording up to 22kHz, but certain limitations mean the practical result is a bit lower. That covers the upper limit for human hearing of around 20kHz (or less, for older farts like me or those who didn't listen to their elders and "turn it down" ).

 

A simple approach to the question of whether 16-bit/44kHZ is acceptable to you is to answer this: "Are you satisfied with how CDs sound?". If you are, then 16/44 is probably fine, because that's the same standard which was decided on for audio CDs.

[keyman_sam]

Thanks for the explanation.

 

Reading this, i got another doubt : Why DO you need 96Khz then? I understand that it has a greater quality, but since we can cover only till 20Khz, shouldnt 96 Khz be unnecessary?

 

sorry i'm asking too much....i want to get the basics clear.

 

Thnkx again,

[BOBA JFET]

There are a few ways 96k can make a difference (sorry if these explanations aren't terribly precise, hopefully they'll at least give you an idea).

 

For one, many sounds have energies up to 100KHz and beyond (a crash cymbal, for instance). While we cannot detect frequencies above 20KHz, some studies have shown that we can detect when these frequencies are absent. I couldn't begin to tell you why, it may have something to do with phase interplay.

 

For another, this also sets the aliasing ceiling much higher, it helps to ensure that rogue energies won't fold back into the audible frequency range at levels that are audible.

 

Also, if you work with plugins a lot, setting the sample rate higher can increase audio fidelity dramatically, but it varies from plugin to plugin (for example, whether the plugin oversamples internally or not seems to make a difference).

[keyman_sam]

Ok, so let me get this straight. Please correct me if i'm wrong :

 

1). The sample resolution is the number of points or values taken per second (when measured as Hz). So, the higher this is, the better interpretation of the original sound. Got this one.

 

2). Though we can't hear above 20 Khz (though sometimes its claimed that we can) we have higher resolution not becos the frequency of the sound itself is higher than 20 Khz, but because the frequency of WHAT WE SAMPLE needs to be higher. That is, HOW FREQUENT we sample the source is important.

 

3). Synths have various components that convert a lot of audio stuff to various resolutions. So, a 44 Khz sample is processed at 48 Khz with effects and the output sample resolution is degraded down to 44 Khz and is supposed to sound better since its processed with 48 Khz rate. (i really dont get this part properly)

 

I understand about the bit that it tells you the maximum number that can be used to represent a signal in the form of 2 powered to the bit depth.

 

 

Ok, a really really nooby question : When there are two different audio resolutions, like, take analog output 1 which comes out as 96 Khz and plug it into analog input 1 which can receive only 44 Khz, what happens?

 

Excuse my poor english, i tried making my questions as clear as possible. Thnkx for the replies again. I'm also doing research on these topics to help me understand better.

[danatkorg]

I'll add a few points:

 

Sample rate and bit depth are two different issues, which affect two different aspects of the audio.

 

Sample rate affects frequency response, aliasing, and other frequency-related issues. Bit depth affects the noise floor, and the degree to which quiet sounds can be accurately represented (aka dynamic range).

 

Sample rate, aliasing, etc.:

 

Often, aliasing is thought of as an issue when sampling analog audio, but it can (and does!) happen during internal processing. VA and sample playback oscillators have aliasing issues; so does any process which modfies amplitude quickly, such as compressors and limiters.

 

Filters and EQs also have frequency-related issues, caused by the relative closeness of the maximum representable frequency (the Nyquist frequency, at half the sample rate), in comparison to the theoretically infinite frequencies of the analog domain.

 

For more on both aliasing and the Nyquist-related issues of filters, see under the "Myth 7" heading of this article:

 

http://emusician.com/mag/emusic_debunking_digitalaudio_myths/index.html

 

- Dan

[keyman_sam]

Thnkx for the response Dan.

 

I'll check out that link. Thnkx again.

[danatkorg]

 

Originally posted by keyman_sam 3). Synths have various components that convert a lot of audio stuff to various resolutions. So, a 44 Khz sample is processed at 48 Khz with effects and the output sample resolution is degraded down to 44 Khz and is supposed to sound better since its processed with 48 Khz rate. (i really dont get this part properly)

 

 

Sample rate conversion is sometimes used, but it's computationally expensive, so it's avoided when possible. (Sample interpolation, which is used by all modern systems to change a sample's pitch by transposition, pitch-bend, etc., is a form of sample rate conversion.)

 

The most common case would be:

 

(Samples at different rates: e.g. 24kHz, 32 kHz, 44.1kHz, 48kHz) --> (Interpolation to system sample rate: e.g. 48kHz) --> (Filters, effects, etc. at the system sample rate) --> (D/A converters at the system sample rate)

 

 

Ok, a really really nooby question : When there are two different audio resolutions, like, take analog output 1 which comes out as 96 Khz and plug it into analog input 1 which can receive only 44 Khz, what happens?

 

 

Note that "resolution" refers to bit depth, and not to the sample rate.

 

It's not possible to connect different sample rates directly, as you describe above. It is possible, however, to convert from one rate to another in realtime; there are a number of devices on the market which do just that, including some multichannel computer audio interfaces. For instance, what you describe above might be done as:

 

(96kHz signal) -- AES/EBU cable --> (Input to device) --> (Sample rate conversion from 96kHz to 44.1kHz) --> (Converted signal sent onward into the 44.1kHz system)

 

Hope this helps,

 

Dan

[SushiFugu]

I'd like to take this opportunity to ask a question that I've had nagging for awhile, I will try to avoid a thread hijack here

 

 

So, people scream all the time "Bit depth affects the noise floor!", "Use 24bit recording instead of 16, it kills the noise!", and so on and so forth. First of all, how does bit rate affect noise ratios? And how do I take advantage of this? Because I've done a fair number of direct comparison recordings using a few different setups and I can never hear nor see a change in line/channel noise at all between 16 and 24 bit resolutions. Pure sine waves aswell as compositions show exact duplicate noise levels using both. What's the deal?

 

Thanks!

[pagan]

I think that final 24-bit recording will make effect when creating audio-CD. But as much as I understand, my final MP3-file will be max. 192 kbps/44 kHz and doesn't matter, is it 16-bit or 24-bit. Please correct me if I'm wrong!

[Roald]

if you have a higher bit depth, the available amount of numbers can be higher

 

In a sampler, you sample voltages as a computer number. If we represent a voltage as a 1 bit number, its value is either

 

0 > -10volt

1 > +10 volt

 

if we take a 2 bit number we have

 

00 > -10volt

01 > -3.3 volt

10 > 3.3 volt

11 > 10 volt

 

 

so...the higher the number of bits, the more different voltages out sampler or soundcard can handle. This has the following advantage: When you record it is not important how 'loud' it is on your hard disk, what is important is that you capture the dynamics well.

 

For example, you record me playing on my wildly dynamic Rhodes. So, to prevent clipping you make the gain on the preamp so low, that when hit really hard, the voltage doesn't exceed the maximum input voltage of 10 volts. So when I play silently, I only 'use' a bandwith of -2 to +2 volts. When the bitrate is low, there is a small number of voltages between -2 and 2, say 150 possible values. So, when you apply gain after the recording to get the sound up to the right level, you get a grainy sound, because you amplified the 150 different values. (some hiphop people like this and put all stuff through 12bit samplers like akai950s and stuff...kind of ridiculous imho)

 

So...when you record you generally want to have the input level as high as possible to get the highest audio quality

 

 

When you work with a high bit depth, you will still have a large number of voltages between -2 and 2 volts, so when you amplify afterwards, you still keep a smooth sound. In other words, you can record with a 'safer' lower gain on the input! That's great news, since you want AND sound quality AND not spoil the recording with clipping.

 

I can't really see why lower bitrate recording would bring extra noise. Maybe, the effect is that you can record with lower input gain, and the input amplifier will give less noise.

 

hope I explained it a bit....

[DINpluggedIN]

Originally posted by SushiFugu ...I've done a fair number of direct comparison recordings using a few different setups and I can never hear nor see a change in line/channel noise at all between 16 and 24 bit resolutions. Pure sine waves aswell as compositions show exact duplicate noise levels using both. What's the deal?

It's a matter of the theoretical versus the real world. Using 16 bits should result in a signal-to-noise ratio of at least 90dB (IIRC, the theoretical maximum dynamic range is 96dB -- 6dB times 16 bits). That's pretty darn good, and many people would be hard-pressed to detect the noise. In fact, the S/N of analog amplification equipment is often worse than that. So, even if you get another 20/30/40dB of S/N by using greater bit depth, from a practical standpoint you may not be able to detect the difference because your amp (or other equipment in the chain) produces noise that overshadows it.

 

Roald, I was going to explain how bit depth affects audible noise, and how "dithering" is used to mask problems, but I found a Web page that has some decent explanations and will save me the time needed . Check out http://www.pcrecording.com/dither.htm

 

Sam, I think you'll find that page helpful as well.

 

--Barry

[Roald]

so Barry, if I understand it correctly, a DA builds up the signal like a 'staircase' function? I can imagine that in that way you add square wave high harmonics, and the higher the resolution the lower the level of noise. Are there also DA converters that build up the A-signal with piecewise linear functions? The little kinks in such a signal would bring much less noise, I think.

 

 

I'm just a humble mech. engineer, so I was just wondering. Maybe this is electronically difficult

 

 

About the bandwidth issue....I don't really care for the extra quality, but I do like to record on the safe side! In many cases I record myself on Rhodes, or friends on e-guitars and like to have as much bandwidth as possible, just to be able to record with low input levels.

[DINpluggedIN]

Originally posted by Roald so Barry, if I understand it correctly, a DA builds up the signal like a 'staircase' function? I can imagine that in that way you add square wave high harmonics, and the higher the resolution the lower the level of noise.

Yes, bit depth determines how many "steps" will recreate the signal amplitude-wise. If you have enough steps (sufficient bit depth), filtering/smoothing techniques make the transitions mostly inaudible. The noise results primarily from "quantization" errors, which is explained pretty well in the link I gave above. The point is that this is a shortcoming in the digital domain, but results in what sounds like noise when converted to analog.

 

Are there also DA converters that build up the A-signal with piecewise linear functions? The little kinks in such a signal would bring much less noise, I think. I'm just a humble mech. engineer, so I was just wondering. Maybe this is electronically difficult

There are converters that use a different technique than what we've been discussing here, and they come a lot closer to a linear approach. If you're interested, do a Google search on "1-bit delta-sigma converter".

 

And I don't for a second buy the "humble mech. engineer" bit. You seem to be grasping this quite well! Or, if we're playing that game, I'm just a humble elec. engineer.

 

About the bandwidth issue....I don't really care for the extra quality, but I do like to record on the safe side! In many cases I record myself on Rhodes, or friends on e-guitars and like to have as much bandwidth as possible, just to be able to record with low input levels.

Well, increased bit depth will certainly increase available dynamic range, so if you want to use it to avoid clipping without sacrificing S/N much, that will do it.

 

--Barry

[Khazul]

Another way to look at sampling rates (number of samples per second) and sampling bit depth (accuracy of each sample) is to find yourself a basic anologue synth with sine waves, square waves and saw tooth etc.

 

Create a pure tone without any filtering - and listen to a square wave on middle C.

 

Now change wave form to a sine wave - its nice a soft and smooth compared to the square wave, and perceptably alot quieter. That is beecause fo the square wave harmonics adding entry energy at octaves above middle C (search for harmonic series or similar on the net - will tell you precisely what harmonics you are hearing).

 

Now imaging that the sound you wanted was a sine wave, but what you actually got was the square wave - you have just simulated 1 bit sampling at 512Hz.

 

 

OK - digital sampling a playback effectively adds a square wave to you signal at basically half the sampling rate. (It not quite as simple as that, the the end result effect is about the same). The average ampitude of this square wave is between 1 and 2 times max voltage/(2^) . So, 8 bit samples, 256 possble values, say 1 v peek to peek, gives 4-8mv - ie about -24dB if we say that 1v is 0dB

 

So if you were sampling at say 16KHz with the above bit depth, that tone is going to be sounding at 8Khz and quiet but very audioable. At the frequency you will high the harmonics of a square wave which adds alot of sound energy compared to a sine wave (remember hearing the difference between a sine and a square?)

 

OK - now increase the bit depth - say double it to 16bit - our square wave ampitude is now appox 15uv (microvolts) - ie at around -48dB - ie you can still hear it in *very* quiet bits.

 

OK - now crank up the sample rate to say 32KHz - ie , the square wave frequency goes up to 16KHz - you *may* still hear it, but you ears are much less sensitive at this frequency - so crank it up again - well beyong hearing to say 44Khz - CD sampling rate.

 

The sound is still there at -48dB, but the freuqncy is so high you cant hear it.

 

Now is you cranmk the sampling rate up again - the the level of the square wave goes down loads more... etc. By now you probably get the picture.

 

 

Next is why both going beyond 44Khz?

 

Well lets take a sound at say 12Khz. If you have tried tuing a guitar by comparing one string with another that is fretted to the same frequency, you hear a beat effect.

 

The beat effect actually cycles at a frequency that is the difference between the two strings - that why you are dead on and can acheive incredably accurate tuning by ear - just fiddle until the phaseing is gone and you well with a tiny fraction of 1Hz to the other string.

 

Back to our sampling at 44KHz and sampling our 12Khz sample. When sampling higher frequencies, there is an error introduced - think of a potato that cas been cut into nice straight chip with nice straight ends as well - now reassemble the original potato - it has odd bits sticking out all over the place - the difference between whgere the surface of the original potato was and where is is now is the error. The end result in audio smapling can give rise to beat effects - in this case half sampling rate - frequence of singla your sampling - ie 10Khz - which is going to be very audiable as a distortion if you are unluckly or try sampling a continous perfect 12Khz tone. In practise you dont notice that particularly exept under certain circumstances:

 

Soft sibilance from vocals can be turn into rather harsh siblance.

Extreme highs get harsher and louder with a stronger tonal quality to them then they should have.

Cranking up the resonance and cutoff frequency on an analogue synth isnt a cool effect anymore - it just plain hurts you ears.

 

You cant exactly tell what the difference is - its just difference.

 

And thats just sampling - if you now process the samples digitally, then you can effectively be resampling and effectively adding more error and therefore more tonal damage - making everything worse.

 

So - how to elliminate this? - simple - crank up the bit depth some more (that reduced the level of the errors) but much better - crank up the sampling rate even more so even beat effects from sampling are taken way out above audiable range - hence why 96KHz can into being as that satifies all those needs.

 

However - because you are sampling music - ie a very variable tone then lots of beat effects come out of the sampling process and recombine to create more beat effects and eventually they are back in the audio range again (but *very* low level) - so the response was very simple - crank up the sampling rate again - so we have 192KHz. I think the theory is that 192KHz is considered to be high enough at 24bit such that the composite effect of beat effects results in less than 1 bit errors (ie less than 1 part in 16million - 66nV at 1vP-P).

 

I cant imagin any human ever being able to tell the difference betwen 96Khz and 192KHz sampling - but you *can* tell the difference between 96Khz and 192Khz processing and particularly synthesis - high rates just sound smoother, easier on the ears at best.

 

 

As best as I understand it - thats what has driven up sample rates - Im not an audio engineer, or a physist, just an ex-electronics engineer who has messed around with sound alot like most us here - so this is intended to give you a rough idea - as with everything, the reality is slighly different and probably way beyond my limited knowlegde.

 

(Feel free to jump in and correct the numbers - I just very roughly approximated them.)

[keyman_sam]

Thanks to you guyz, i'm learning a lot of new stuff.

 

 

This thread needs to bumped!

[Jeebus]

GODDY!

 

[Allerian]

Cool thread. Now I have to record 8 bit music at 96khz. Maybe I can use the other 88khz for distortion.

[DINpluggedIN]

Originally posted by Allerian Now I have to record 8 bit music at 96khz. Maybe I can use the other 88khz for distortion.

Please, a little respect; Harry Nyquist is probably rolling over in his grave.