24bit vs 16bit, the myth exploded!

[Makiah S]

Quote:

High linearity and low cost. Btw, sigma-delta modulation can be mixed with multi-bit DACs.

whoa explain plox

 

[stv014]

Quote:

whoa explain plox

 
A 1-bit DAC is the easiest and cheapest to implement, as it has only two possible output levels. With only two levels, there is - ignoring other factors - in theory no distortion from the D/A conversion itself. With the current technology, increasing the clock frequency to the MHz range is also cheap, as is the DSP necessary to implement the oversampling and noise shaping. By contrast, a high resolution and accurate resistor ladder (that could match the ability of a delta-sigma DAC to convert 24 bit PCM data at less than 0.001% distortion) is more difficult and expensive. Basically, the complexity is moved from the analog domain to the digital one, where sound quality becomes a function of speed and transistor count, which can be increased at low cost, unlike analog accuracy. The use of oversampling also allows for a digital reconstruction filter, making the analog lowpass filter simpler and cheaper.
 
The reason why multi-bit DACs are still used is that in a 1-bit format, it is impossible to implement a proper dither that makes the quantization error uncorrelated to the input signal (the sum of the dither noise and the signal will get clipped). A low resolution (e.g. 4-bit) multi-bit DAC avoids the dithering problem, in addition to reducing the total amount of quantization noise, and is a good compromise overall.

 

[Makiah S]

Quote:

 
A 1-bit DAC is the easiest and cheapest to implement, as it has only two possible output levels. With only two levels, there is - ignoring other factors - in theory no distortion from the D/A conversion itself. With the current technology, increasing the clock frequency to the MHz range is also cheap, as is the DSP necessary to implement the oversampling and noise shaping. By contrast, a high resolution and accurate resistor ladder (that could match the ability of a delta-sigma DAC to convert 24 bit PCM data at less than 0.001% distortion) is more difficult and expensive. Basically, the complexity is moved from the analog domain to the digital one, where sound quality becomes a function of speed and transistor count, which can be increased at low cost, unlike analog accuracy. The use of oversampling also allows for a digital reconstruction filter, making the analog lowpass filter simpler and cheaper.
 
The reason why multi-bit DACs are still used is that in a 1-bit format, it is impossible to implement a proper dither that makes the quantization error uncorrelated to the input signal (the sum of the dither noise and the signal will get clipped). A low resolution (e.g. 4-bit) multi-bit DAC avoids the dithering problem, in addition to reducing the total amount of quantization noise, and is a good compromise overall.

Well I'm never a fan of cheap quick fixes... so what are some examples of these multi bit DACS?

 

[xnor]

Quote:

Well I'm never a fan of cheap quick fixes... so what are some examples of these multi bit DACS?

These DACs are far from "cheap quick fixes"...
 
An example for a multi-bit sigma delta DAC chip would be the WM8740.

 

[Makiah S]

Quote:

 
A 1-bit DAC is the easiest and cheapest to implement, as it has only two possible output levels. With only two levels, there is - ignoring other factors - in theory no distortion from the D/A conversion itself. With the current technology, increasing the clock frequency to the MHz range is also cheap, as is the DSP necessary to implement the oversampling and noise shaping.
By contrast, a high resolution and accurate resistor ladder (that could match the ability of a delta-sigma DAC to convert 24 bit PCM data at less than 0.001% distortion) is more difficult and expensive.
 
The use of oversampling also allows for a digital reconstruction filter, making the analog lowpass filter simpler and cheaper.

Basically it sounds like the 1-bit DAC's are "simpler and Cheaper" fixes, like for example when your making brownies you can buy a pre made mix and OMG taste GOOD and it was SO simple to make and SO cheap
 
or
From Scratch Brownies
You can CREAM the Butter and Sugar your self then ect... ect... Tastes OMG LIKE SMEX in my mouth, but it took a lot longer and cost me more time due to complexity...
 
I'd like my DAC to be like those from scratch brownies... AWESOME, because "good" is not GOOD enough
 
Still though I wish I understood more about electronics so I can avoid putting my foot in my mouth! <3

 

[xnor]

Quote:

Basically it sounds like the 1-bit DAC's are "simpler and Cheaper" fixes, like for example when your making brownies you can buy a pre made mix and OMG taste GOOD and it was SO simple to make and SO cheap

So DSD (SACD) is also a "simple and cheap" format? Seriously, 1-bit SDM is ingenious.

 

[jcx]

 
table from: http://web.archive.org/web/20070118012711/http://www.iet.ntnu.no/~ivarlo/files/School/PhD/Report_audiodac.pdf
 
a 2005 graduate level report on audio DAC internals - if you really want to know more about multi-bit delta-sigma

 

[Makiah S]

Quote:

So DSD (SACD) is also a "simple and cheap" format? Seriously, 1-bit SDM is ingenious.

See, there my foot is now in my mouth <3
 
*sigh* I DONT WANT TO BE IGNORANT! *reads article* 

 

[BernieW]

Hi, just joined this forum because this topic has interested me for a while. I work in pro audio, mostly live. I've been reading a bit about this debate, including xiph's and Lavry's digital audio articles. I recently bought a computer audio system, and I try to get 24/96 or other high data rate tracks whenever possible.
 
I accept that 16/44.1 CD format is adequate, in theory, for almost everybody. I'm also willing to believe that some people perceive (not HEAR, but PERCEIVE) frequencies above 20kHz. I DON'T believe "hi-res" media's purpose is to reproduce frequencies above 20kHz. The filter in most DAC's would account for that anyway (so flame away, but I don't care. I have a thick skin, I've been online for years).
 
So here's a question I haven't seen addressed yet. (The topic was actually about bit depth, not sample rate, but this thread has covered a lot of ground). Sampling theory says 44.1kFs/s is enough to fully encode an analogue waveform with content up to 20kHz... in theory. But that's where I see the problem. Surely that represents an ideal, and all practical systems are going to fall short of that ideal. The output wave will differ from the original thanks to that short-fall, meaning distortion. So wouldn't a stream of more samples per second mean reproduction will deviate less from the ideal? (because it's being corrected/referenced more often).
 
The problem I've always had with CD audio is that there's enough information to perfectly recreate the original waveform, but only just. Won't economics mean there's so much compromise in most systems that deviation from that ideal is pretty high? After all, those 20 buck portable players DO sound awful. I'm suggesting imperfect (real world) hi-res playback should have an advantage over equivalent standard rate playback. Very little equipment, only the most expensive, is near-ideal. So do economics and the real world equal a case for high-accuracy formats?

 

[xnor]

Quote:

So here's a question I haven't seen addressed yet. (The topic was actually about bit depth, not sample rate, but this thread has covered a lot of ground). Sampling theory says 44.1kFs/s is enough to fully encode an analogue waveform with content up to 20kHz... in theory.

Nope, it can contain frequencies up to just below 22.05 kHz. Giving filters some room to work we can say that 20 to 21 kHz is a practical upper limit.
 
Quote:

But that's where I see the problem. Surely that represents an ideal, and all practical systems are going to fall short of that ideal. The output wave will differ from the original thanks to that short-fall, meaning distortion.

Cue: oversampling DACs.
 
Quote:

So wouldn't a stream of more samples per second mean reproduction will deviate less from the ideal? (because it's being corrected/referenced more often).

More samples would just be redundant. And too many (e.g. 192 k) samples per second will cause real-world performance to suffer.
 
 
Quote:

The problem I've always had with CD audio is that there's enough information to perfectly recreate the original waveform, but only just. Won't economics mean there's so much compromise in most systems that deviation from that ideal is pretty high? After all, those 20 buck portable players DO sound awful. I'm suggesting imperfect (real world) hi-res playback should have an advantage over equivalent standard rate playback. Very little equipment, only the most expensive, is near-ideal. So do economics and the real world equal a case for high-accuracy formats?

We've got a lot closer to the ideal over time, still do.
A $ 20 portable player sounding awful has little to do with the CD format.

 

[stv014]

Quote:

The problem I've always had with CD audio is that there's enough information to perfectly recreate the original waveform, but only just. Won't economics mean there's so much compromise in most systems that deviation from that ideal is pretty high?

 
DAC chips that cost only a few dollars are close enough to the ideal now, it is not that hard for an oversampling DAC to reconstruct a signal sampled at 44.1 kHz without audible artifacts.

 

[AstralStorm]

/me = slowpoke. The thread has 60 pages. I'm almost certainly sure the points have been mentioned multiple times.
 
 
Quote:

 
DAC chips that cost only a few dollars are close enough to the ideal now, it is not that hard for an oversampling DAC to reconstruct a signal sampled at 44.1 kHz without audible artifacts.

 
Heck, even 0.5$ DAC can be pretty good. The real issues are usually in the analog part further down, that is, amp. I can make a pretty decent DAC+amp for $20 in parts, definitely good enough for 16-bits. The trick is to put these parts in correct topology and be careful about grounds. Granted, it won't be exactly Objective O2 class amp, closer to, say, FiiO E5. (sans the bass roll and THD from undersize cap as well as with better channel separation and bit less noise in general - those are sacrificed for battery life and size.)

 

[stv014]

Quote:

Heck, even 0.5$ DAC can be pretty good. The real issues are usually in the analog part further down, that is, amp.

 
That is not really relevant to the question of whether CD format is good enough with a typical reconstruction filter, though, since the digital filter is in the DAC chip itself.

 

[xnor]

Quote:

 
A) Dither does not eliminate quantization errors, only decorrelates the error from the signal, making it less detectable/more pleasant, but not any less measurable. Decorrelated noise put into an integrator (averaging system) will be greatly reduced at the output, how much depends on the integration length. Correlated noise causes frequency response spikes similar to comb filtering.

That's why he wrote "eliminate quantization distortion" and: "Essentially during the conversion process a very small amount of white noise is added to the signal, this has the effect of completely randomising the quantisation errors."
 
Quote:

B) [...] Nyquist also deals with continuous signals, but temporal resolution of DACs/amps is also great. Any issue there is nothing higher digital bit depth can fix - unless the DAC happens to use a different reconstruction filter for bit depths - that is more common with different sample rates, but usually if one of the filters is broken, they all are.
(Hello, Hifiman HM-801; cheap chinese noisy amps. Also high output impedance effects.)

I don't understand what you're saying here. Could you rephrase this please or expand a bit?
 
Quote:

C) Higher digital SNR allows you to reduce volume in digital domain (subtract) quite a lot without truncation loss or noise. However, remember that 50% loudness is just about 6 dB cut. (@ 1kHz) This difference means that you will likely increase loudness in analog domain, which might or might not introduce more noise than that.

Sure, but you can play a 16-bit track using a 24-bit or 32-bit DAC. Doesn't change that 16-bit seems to be enough to store audio.
 
 
Quote:

This mostly comes into play when you're doing any DSP, such as equalization, where that ~20 dB extra dynamic range might be useful, otherwise the truncation might be just barely audible. (since there are 70 dB SNR remaining or so) Even if the recording is louder, you've added noise by boosting the reference volume, the noise might or might not be perceptually masked. That's why good equalizers work with at least 24-bit precision and dither back if required. Certain lousy equalizers work with the same bit depth as input and/or output though.

When doing equalization for playback you usually correct the FR of the headphones/speakers. If there's a 10 dB dip in the FR you boost by 10 dB. The signal can again be played through a higher bit DAC. The end result should be the same, as if played on a flat headphone/speaker without eq.
 
 
Quote:

D) You can sometimes hear the difference between 16-bit and 24-bit version of the same track on different medium, because the 24-bit one doesn't use any dynamic range compression and/or has different mastering. It's like with any other remaster.

Yeah, that sucks. They should release properly masters tracks on CD too... in some genres we're down to 3 dB dynamic range.

 

[dbbloke]

I don't have time to read this thread. It's bollocks.
24 bit is an improvement over 16bit. There is more air around things, there is more description of the reverb of the room, things sound more lifelike and slightly more subtle. I know this 100% for sure, it's a fact. No matter of flawed science can prove it otherwise.
 
Whether or not it makes THAT much difference or that much MORE of a music experience though ... probably not much. If anything like a good DAC, it can ruin with extra vision like 1080p sometimes does.