http://electronics.howstuffworks.com/question620.htm
If you google for "1 bit dac" you'll find many more in depth explanations and circuits even
If you google for "1 bit dac" you'll find many more in depth explanations and circuits even
Konnichiwa,
A 1 Bit dac is expressed in the simplest terms a Device that uses a form of pulse width modulation to generate an output signal proportional to a digital input.
Consider it like this:
A squarewave is output at exactly 50% Dutycycle.
Lowpass filter this squarewave and you get a volage equivalent to 1/2 the peakvoltage.
Vary the duty cycle to say 25/75% and your output voltage is 25% the peak voltage.
You can now use a digital mechanism to genertae sucha waveform by varying the pulse width according to a given system.
For example, if we used a clock at 256 Times that of the sample clock we could make for each sample value 256 discrete values simply by varying the width of the output pulse at the sample rate.
Therefore a 1-Bit dac to operate at 16B bit equivalent resolution must operate at 65536 times the sample frequency, in short at GHz frequencies. This is clearly not possible.
Therefore ANY 1-Bit (or delta/Sigma) DAC has a real resolution much lower than 16-Bit and uses a number of mathematical and psychoacoustic tricks to make up for that.
Usually the tricks do not work well causing a very distinct type of sound to appear. This type of sound is either liked or not.
More recent approaches combine Multibit DAC's of low resolution (4 - 8 Bit) with 8 Bit equivalent resolution capable 1-Bit and add noiseshabping to get reasonable performance. These are usually called "Multilevel Delta Sigma" and are a better choice.
Sayonara
percy said:
A 1 Bit dac is expressed in the simplest terms a Device that uses a form of pulse width modulation to generate an output signal proportional to a digital input.
Consider it like this:
A squarewave is output at exactly 50% Dutycycle.
Lowpass filter this squarewave and you get a volage equivalent to 1/2 the peakvoltage.
Vary the duty cycle to say 25/75% and your output voltage is 25% the peak voltage.
You can now use a digital mechanism to genertae sucha waveform by varying the pulse width according to a given system.
For example, if we used a clock at 256 Times that of the sample clock we could make for each sample value 256 discrete values simply by varying the width of the output pulse at the sample rate.
Therefore a 1-Bit dac to operate at 16B bit equivalent resolution must operate at 65536 times the sample frequency, in short at GHz frequencies. This is clearly not possible.
Therefore ANY 1-Bit (or delta/Sigma) DAC has a real resolution much lower than 16-Bit and uses a number of mathematical and psychoacoustic tricks to make up for that.
Usually the tricks do not work well causing a very distinct type of sound to appear. This type of sound is either liked or not.
More recent approaches combine Multibit DAC's of low resolution (4 - 8 Bit) with 8 Bit equivalent resolution capable 1-Bit and add noiseshabping to get reasonable performance. These are usually called "Multilevel Delta Sigma" and are a better choice.
Sayonara
Re: Re: What is a 1-bit DAC ?
I wonder why two of the better bitstream players which I own have low noise like PCM63 players and even lower distortion at -60dB.
By the way 2 PCM63-J and 2 PCM63-K2 chips I recently tested, all were not better than randomly selected PCM56 and have been totally outperformed by very good PCM56 chips.
Question:
R2R-ladder DACs have a number of resistors that depends on the number of bits.
Compared to a 16 bit DAC, statistically it is more likely that a 20 bit DAC got resistors that do not match the desired value, leading to nonlinearity of the DAC chip.
Even more if it is a "colinear" DAC
Other question:
How got Philips 16 bit out of a 14bit 1540 DAC ?
averaging bits / 4 samples ?
Kuei Yang Wang said:
I wonder why two of the better bitstream players which I own have low noise like PCM63 players and even lower distortion at -60dB.
By the way 2 PCM63-J and 2 PCM63-K2 chips I recently tested, all were not better than randomly selected PCM56 and have been totally outperformed by very good PCM56 chips.
Question:
R2R-ladder DACs have a number of resistors that depends on the number of bits.
Compared to a 16 bit DAC, statistically it is more likely that a 20 bit DAC got resistors that do not match the desired value, leading to nonlinearity of the DAC chip.
Even more if it is a "colinear" DAC
Other question:
How got Philips 16 bit out of a 14bit 1540 DAC ?
averaging bits / 4 samples ?
Re: Re: Re: What is a 1-bit DAC ?
Konnichiwa,
First, Delta Sigma eliminates the distortion sources common to multibit DAC's. Therefore distortion according to conventional measurements will not exist. Noisefloor depends upon many factors. Have you ever done a noiseload test?
Conventional measurements show very little relation to "subjectively percieved good sound".
In essence yes. They used primitive noiseshaping and 4 X Oversampling to add in effect 4 more discrete values to the output, in other 2^2 or 2 Bits.
Sayonara
Konnichiwa,
Bernhard said:
First, Delta Sigma eliminates the distortion sources common to multibit DAC's. Therefore distortion according to conventional measurements will not exist. Noisefloor depends upon many factors. Have you ever done a noiseload test?
Conventional measurements show very little relation to "subjectively percieved good sound".
Bernhard said:
In essence yes. They used primitive noiseshaping and 4 X Oversampling to add in effect 4 more discrete values to the output, in other 2^2 or 2 Bits.
Sayonara
Re: Re: Re: Re: What is a 1-bit DAC ?
I use -60dB dithered and noiseshaped signal from a forum member.
What I observe is that DACs with visible distortion tend to sound more warm, brilliant or harsh or all together, while clean DACs sound more detailed, smooth and colder /more neutral.
With exceptions like TDA1540 in CD304 which sounds neutral too, and my very clean CD880 which sounds "yellow".
So for me there is already a relation between measurement and sound.
Kuei Yang Wang said:
I use -60dB dithered and noiseshaped signal from a forum member.
What I observe is that DACs with visible distortion tend to sound more warm, brilliant or harsh or all together, while clean DACs sound more detailed, smooth and colder /more neutral.
With exceptions like TDA1540 in CD304 which sounds neutral too, and my very clean CD880 which sounds "yellow".
So for me there is already a relation between measurement and sound.
Re: Re: What is a 1-bit DAC ?
Ok. So what is the difficulty in implementing a true 16-bit dac ? (the 'how stuff works' link above tried to explain that but it went over the top).
I think I understood the part about Multibit DAC's of low resolution (4-8 bit) in the sense that it must be allowing to use a lower (than Ghz) frequency, right ?
But I didn't understand the part about combining it with '8 bit equivalent resolution capable 1-bit'. Is the '8-bit equivalent resolution 1bit' different from a regular 1bit dac and why 'combine' it ? I'd imagine a 4-8 bit dac by itself would be good enough ?
Kuei Yang Wang said:
Ok. So what is the difficulty in implementing a true 16-bit dac ? (the 'how stuff works' link above tried to explain that but it went over the top).
I think I understood the part about Multibit DAC's of low resolution (4-8 bit) in the sense that it must be allowing to use a lower (than Ghz) frequency, right ?
But I didn't understand the part about combining it with '8 bit equivalent resolution capable 1-bit'. Is the '8-bit equivalent resolution 1bit' different from a regular 1bit dac and why 'combine' it ? I'd imagine a 4-8 bit dac by itself would be good enough ?
Here's an example where objectivists and subjectivists agree:
<<Stereophile also performed a large subjective listening test between some existing high performance recording formats and SACD was even considered to be inferior to PCM 16bit, 176,4 kHz.
Anyway, one should not take too much notice of "how many" said this or that. What is important is how things really are, not how many people believe this or that. Deciding truths by voting is seldom a good way. It can only show what people believe. Some peoples beliefs might of course be right, but to often correct information and relevant listening experiences drowns in common misconceptions, preconceived notions and the media background noise.>>
I won't argue about measurements and what you can hear or not hear. (It's also difficult to make a fair comparison between CD and SACD because of the limited frequency range of the former.) But the poor measurements of the SACD evidently do result in poorer sound.
And it's also pretty evident, as I see it, that CD, HDCD and DVD-A are the only audiophile formats out there. Unfortunately HDCD will go nowhere. There's simply no money in it. And DVD-A has so far been targeted exclusively for the surround crowd. Which leaves us with CD as the only real audiophile format.
Re: Re: Re: What is a 1-bit DAC ?
Konnichiwa,
There is no real difficulty, merely high [recision, thermally matched resistors are required and other precision analog circuitry, this simply costs and if you make it all on silicone the rate of chips that do not pass muster is high.
On the other hand 1-Bit type DAC's are purely digital devices, very cheap to make.
Yes, it you operate a DAC with 8 Bit Resolution with a 256 X Sample Rate clock then the combination of pulse width modulation and multibit DAC can be combined to make a DAC with a true 16-Bit equivalent resolution.
What it suggests is that by combining the methodes applied to a 1-Bit DAC with a multibit DAC can achieve higher native resolution than either alone.
A DAC with 4-Bit or 8-Bit resolution would be very poor sounding. A 1 Bit DAC with 256 X Sample rate clock is, without applying a lot of trickery to it in effect also merely an 8-Bit DAC.
If you want, windows audio works with 8Bit wav files, just get cooledit and change a Music Filke to 8-Bit and listen.
So, if I now use the 1-Bit part for the lower 8-Bit in audio and the Multibit part for the upper 8-Bit you have a 16-Bit capable device combining the two.
Sayonara
Konnichiwa,
percy said:
There is no real difficulty, merely high [recision, thermally matched resistors are required and other precision analog circuitry, this simply costs and if you make it all on silicone the rate of chips that do not pass muster is high.
On the other hand 1-Bit type DAC's are purely digital devices, very cheap to make.
percy said:
Yes, it you operate a DAC with 8 Bit Resolution with a 256 X Sample Rate clock then the combination of pulse width modulation and multibit DAC can be combined to make a DAC with a true 16-Bit equivalent resolution.
percy said:
What it suggests is that by combining the methodes applied to a 1-Bit DAC with a multibit DAC can achieve higher native resolution than either alone.
percy said:
A DAC with 4-Bit or 8-Bit resolution would be very poor sounding. A 1 Bit DAC with 256 X Sample rate clock is, without applying a lot of trickery to it in effect also merely an 8-Bit DAC.
If you want, windows audio works with 8Bit wav files, just get cooledit and change a Music Filke to 8-Bit and listen.
So, if I now use the 1-Bit part for the lower 8-Bit in audio and the Multibit part for the upper 8-Bit you have a 16-Bit capable device combining the two.
Sayonara
Would it make sense to combine multiple 1-bit DACs ? (or for the matter of fact multiple Multibit DACs ?) i.e. combining two 1-bit 265x DACs ? or combining four 4-bit multibit DACs or eight 2-bit multibit DACs ?
I'd think in order to take full advantage of the best of both the worlds it would be better to mix both the types(1-bit & Multibit)together in the correct proportion.
(did I just answer my own question?)
Secondly, what is a good source to find out detailed architecture of a DAC ? For example the Pioneer DV-563 player that I have uses Burr Brown DSD-1791 DAC for the front L-R channels. I tried looking up the burr-brown(now ti) website but it didn't quite yield much info.
Thank you.
I'd think in order to take full advantage of the best of both the worlds it would be better to mix both the types(1-bit & Multibit)together in the correct proportion.
(did I just answer my own question?)
Secondly, what is a good source to find out detailed architecture of a DAC ? For example the Pioneer DV-563 player that I have uses Burr Brown DSD-1791 DAC for the front L-R channels. I tried looking up the burr-brown(now ti) website but it didn't quite yield much info.
Thank you.
Konnichiwa,
Absolutely. The only you must make sure is that the relative contribution to the output signal is suitably scaled, in other words you need again precision analog. Or you can apply Delta Sigma modulation to the output from a Multibit DAC.
Usually it's datasheet. If not for the specifc device, then usually in the earlier models of the device family.
Funny, page 49 of the datasheet is quite detailed in terms of operation principle.
From where I stand the only advantage I note in 1-Bit technology is cost. Remove that and Multibit is a better choice by far.
Sayonara
percy said:
Absolutely. The only you must make sure is that the relative contribution to the output signal is suitably scaled, in other words you need again precision analog. Or you can apply Delta Sigma modulation to the output from a Multibit DAC.
percy said:
Usually it's datasheet. If not for the specifc device, then usually in the earlier models of the device family.
percy said:
Funny, page 49 of the datasheet is quite detailed in terms of operation principle.
percy said:
From where I stand the only advantage I note in 1-Bit technology is cost. Remove that and Multibit is a better choice by far.
Sayonara
Re: Re: Re: What is a 1-bit DAC ?
Probably this means you like the pure R2R sound. PCM56 probably is the last thoroughbred R2R DAC ever made. It's a classical 16bit R2R ladder, accurate down to 15 bits. PCM63 is basically several smaller R2R DACs linked in parallel. Those links will change R2R's sound signature.
-finney
Bernhard said:
By the way 2 PCM63-J and 2 PCM63-K2 chips I recently tested, all were not better than randomly selected PCM56 and have been totally outperformed by very good PCM56 chips.
Question:
R2R-ladder DACs have a number of resistors that depends on the number of bits.
Compared to a 16 bit DAC, statistically it is more likely that a 20 bit DAC got resistors that do not match the desired value, leading to nonlinearity of the DAC chip.
Even more if it is a "colinear" DAC
Probably this means you like the pure R2R sound. PCM56 probably is the last thoroughbred R2R DAC ever made. It's a classical 16bit R2R ladder, accurate down to 15 bits. PCM63 is basically several smaller R2R DACs linked in parallel. Those links will change R2R's sound signature.
-finney
Re: Re: What is a 1-bit DAC ?
Decent 1-bit DAC is Delta-sigma generator, and that makes all the difference.
Look at it like this:
Imagine a sinewave where for a full period you have 8 samples. When you go from sample to sample, difference step size is some fraction of full 16-bit resolution. Now if you double the sample rate, you get 16 samples per wave period after interpolation. Now the required step size between samples is smaller - twice at least and much smaller on average. When you increase sample rate enough, your required step size is 1 bit.
This 1-bit step is sent to integrator as a sort of commands "increase output" or "decrease output". Delta signal. When output has to be hold constant, 50% dutycycle is sent.
Now, spectral distribution of audio is such that there is 6db roll-off towards higher frequencies. What this means is that at higher frequencies maximum delta step is smaller already by nature. From this by corollary it occurs that _for each_ doubling of sample rate you gain _2_ effective bits of resolution. So, it goes like that:
1FS - 1 bit effective
2FS - 3 bits
4FS - 5 bits
8FS - 7 bits
16FS - 9 bits
32FS - 11 bits
64FS - 13 bits
128FS - 15 bits
256FS - 17 bits
So there. Benefits of 1-bit DAC are in theory that you need only 1 single reference voltage/current and you get away with all the parallel DAC linearity issues, simple output filters, etc. In practice, well, depends on practice.
Drawbacks are that when audio signal exceeds assumptions of natural roll-off, 1-bit DAC gets overloaded. Effect of this is short-term distortion and DC offset. DC offset is servoed out. Because overloads are very short, affecting only transients, they are not much worse than few samples clipping, even less - 1-bit DAC behaves like slew rate limiter, it enforces high frequency roll-off for strong HF signals, and passes weak HF signals. Sort of compression. This is where psychoacoustic cuts in.
So, for well-behaved audio source, 1-bit DAC can be very good an nice sounding solution. But loudness race and insane compression with very high-frequency heavy content can bring it down. And of course syntetic test signals.
Its not just simple hype that 1-bit DACs can be found in pretty highend devices. At some point they believed it will solve all precision issues. They still do. There are very few pure R2R ladder dac chips these days. Most use hybrid of R2R and delta-sigma. Its almost impossible to go beyond 15-18 or smth bits with R2R ladders.
In conclusion, your assertion that 1-bit 256FS dac is never better than 8-bit ladder DAC is misleading. Its the implementation details and intended usage that makes the difference.
Of course there is much hype around 1-bit dacs and truely cheap crp has flooded down the market, making it impossible to make any judgement about 1-bit dacs by specs. Two devices both with proudly advertized 1-bit DACs can be very different.
No, this is not so trivial. Pure 1-bit DAC is not dutycycle generator, and GHz frequencies are not required.Kuei Yang Wang said:
Decent 1-bit DAC is Delta-sigma generator, and that makes all the difference.
Look at it like this:
Imagine a sinewave where for a full period you have 8 samples. When you go from sample to sample, difference step size is some fraction of full 16-bit resolution. Now if you double the sample rate, you get 16 samples per wave period after interpolation. Now the required step size between samples is smaller - twice at least and much smaller on average. When you increase sample rate enough, your required step size is 1 bit.
This 1-bit step is sent to integrator as a sort of commands "increase output" or "decrease output". Delta signal. When output has to be hold constant, 50% dutycycle is sent.
Now, spectral distribution of audio is such that there is 6db roll-off towards higher frequencies. What this means is that at higher frequencies maximum delta step is smaller already by nature. From this by corollary it occurs that _for each_ doubling of sample rate you gain _2_ effective bits of resolution. So, it goes like that:
1FS - 1 bit effective
2FS - 3 bits
4FS - 5 bits
8FS - 7 bits
16FS - 9 bits
32FS - 11 bits
64FS - 13 bits
128FS - 15 bits
256FS - 17 bits
So there. Benefits of 1-bit DAC are in theory that you need only 1 single reference voltage/current and you get away with all the parallel DAC linearity issues, simple output filters, etc. In practice, well, depends on practice.
Drawbacks are that when audio signal exceeds assumptions of natural roll-off, 1-bit DAC gets overloaded. Effect of this is short-term distortion and DC offset. DC offset is servoed out. Because overloads are very short, affecting only transients, they are not much worse than few samples clipping, even less - 1-bit DAC behaves like slew rate limiter, it enforces high frequency roll-off for strong HF signals, and passes weak HF signals. Sort of compression. This is where psychoacoustic cuts in.
So, for well-behaved audio source, 1-bit DAC can be very good an nice sounding solution. But loudness race and insane compression with very high-frequency heavy content can bring it down. And of course syntetic test signals.
Its not just simple hype that 1-bit DACs can be found in pretty highend devices. At some point they believed it will solve all precision issues. They still do. There are very few pure R2R ladder dac chips these days. Most use hybrid of R2R and delta-sigma. Its almost impossible to go beyond 15-18 or smth bits with R2R ladders.
In conclusion, your assertion that 1-bit 256FS dac is never better than 8-bit ladder DAC is misleading. Its the implementation details and intended usage that makes the difference.
Of course there is much hype around 1-bit dacs and truely cheap crp has flooded down the market, making it impossible to make any judgement about 1-bit dacs by specs. Two devices both with proudly advertized 1-bit DACs can be very different.
Re: Re: Re: What is a 1-bit DAC ?
Konnichiwa,
You realise that what you explain here cannot work using a conventional integrator or any such analogue circuit and given that you only have the ability to switch the output of the DS DAC between Vdd and Vss you are limited to a variation of PWM. However you cut that, without noiseshaping or other methodes.
The process you describes CANNOT work as described, a simple 50% dutycycle modulation MUST result in an output that will return from wherever it was when the signal was switched to 50% dutycycle to a point miway between Vss and Vdd according to the timeconstant of the integrator.
Therefore, switching to a 50% Dutycycle will not result in a constant output but one that will successively approximate "digital silence" more, the longer it is retained.
Therefore the claim that the 1-Bit (or so-called delta sigama) DAC's work by actually only encoding the "delta" and not the absolute value must be considered case unproven. I have come repeatedly about claims to the effect, but failed to observe the actual operation of the DAC to be anything like the claims, as far as is possible to ascertain from the documention(s) available.
This is a neat piece of making up numbers. You fail to illustrate that:
A) An actual DAC can ACTUALLY WORK according to the principle you claim it should work like.
B) "that _for each_ doubling of sample rate you gain _2_ effective bits of resolution"
You claim A) and B) without any material references or derived math.
There what?
Yes, my point. Much cheaper to make than precision silicone.
In practice, well, depends on practice.
Hmmm. If this is indeed so (which I assert it is not and cannot be, in effect) then the dynamic range of a 1-Bit DAC is large only at DC and progressively narrows towards high frequencies with a rolloff of the actual MOL (maximum output level) with frequency. This is illustrably not the case.
We do not observe a falling amplitude characteristic (MOL or FR) with DS DAC's in audio but a rising noisefloor with frequency, which is what I'd predict for noiseshaped (dithered) PWM.
Yet surprisingly, to my ears I note the failure of 1-Bit DAC's especially with puurely acoustical music, recorded without compression and the failure is gross, differences become less and less as music approaches modern high loudness, high compression material.
Sadly I must disagree. Moreover, i have noted that in accordance with the claims for their performance most (if not all) DS DAC's indeed provide the same sound in low price devices as they do in high prise ones, very bad sound at that too.
I never claimed that the observed resolution cannot be larger, it can be, by using resolution enhancing trickery of all sorts. However the same trickery can also be used to extend the performance of low wordlength multibit DAC past the theoretical boundaries. This is indeed implementation related.
What I did note was that without any further trickery, which invariably amounts to throwing the baby (the music or the holy signal) out with the bathwater the NATIVE resolution is limited to 2^8 = 256 or 8-Bit equivalent.
We can use agressive noiseshaping (and most 1-Bit systems do so VERY AGRESSIVELY, including SACD) to re-distribute the 8-Bit Noisefloor and then attempt to filter out the redistributed noise, leaving whatever our mathematical noiseshaping has left of the original signal. I for one find such an approach to run counter to any attempt to transmit a signal that has reliable semblence to what was fed into the transmission chain. A simple look at the output of an SACD Player using a wideband 'scope confirms my point visually.
Sayonara
Konnichiwa,
wimms said:
You realise that what you explain here cannot work using a conventional integrator or any such analogue circuit and given that you only have the ability to switch the output of the DS DAC between Vdd and Vss you are limited to a variation of PWM. However you cut that, without noiseshaping or other methodes.
The process you describes CANNOT work as described, a simple 50% dutycycle modulation MUST result in an output that will return from wherever it was when the signal was switched to 50% dutycycle to a point miway between Vss and Vdd according to the timeconstant of the integrator.
Therefore, switching to a 50% Dutycycle will not result in a constant output but one that will successively approximate "digital silence" more, the longer it is retained.
Therefore the claim that the 1-Bit (or so-called delta sigama) DAC's work by actually only encoding the "delta" and not the absolute value must be considered case unproven. I have come repeatedly about claims to the effect, but failed to observe the actual operation of the DAC to be anything like the claims, as far as is possible to ascertain from the documention(s) available.
wimms said:
This is a neat piece of making up numbers. You fail to illustrate that:
A) An actual DAC can ACTUALLY WORK according to the principle you claim it should work like.
B) "that _for each_ doubling of sample rate you gain _2_ effective bits of resolution"
You claim A) and B) without any material references or derived math.
wimms said:
There what?
wimms said:
Yes, my point. Much cheaper to make than precision silicone.
In practice, well, depends on practice.
wimms said:
Hmmm. If this is indeed so (which I assert it is not and cannot be, in effect) then the dynamic range of a 1-Bit DAC is large only at DC and progressively narrows towards high frequencies with a rolloff of the actual MOL (maximum output level) with frequency. This is illustrably not the case.
We do not observe a falling amplitude characteristic (MOL or FR) with DS DAC's in audio but a rising noisefloor with frequency, which is what I'd predict for noiseshaped (dithered) PWM.
wimms said:
Yet surprisingly, to my ears I note the failure of 1-Bit DAC's especially with puurely acoustical music, recorded without compression and the failure is gross, differences become less and less as music approaches modern high loudness, high compression material.
wimms said:
Sadly I must disagree. Moreover, i have noted that in accordance with the claims for their performance most (if not all) DS DAC's indeed provide the same sound in low price devices as they do in high prise ones, very bad sound at that too.
wimms said:
I never claimed that the observed resolution cannot be larger, it can be, by using resolution enhancing trickery of all sorts. However the same trickery can also be used to extend the performance of low wordlength multibit DAC past the theoretical boundaries. This is indeed implementation related.
What I did note was that without any further trickery, which invariably amounts to throwing the baby (the music or the holy signal) out with the bathwater the NATIVE resolution is limited to 2^8 = 256 or 8-Bit equivalent.
We can use agressive noiseshaping (and most 1-Bit systems do so VERY AGRESSIVELY, including SACD) to re-distribute the 8-Bit Noisefloor and then attempt to filter out the redistributed noise, leaving whatever our mathematical noiseshaping has left of the original signal. I for one find such an approach to run counter to any attempt to transmit a signal that has reliable semblence to what was fed into the transmission chain. A simple look at the output of an SACD Player using a wideband 'scope confirms my point visually.
Sayonara
yeah I missed the little 'download' link hiding in the cluttered page. I downloaded a few datasheets of different types of DACs (multilevel D-S, advanced segment, monolithic) to look at each one's architecture.
Interesting thing this Advanced Segment is. Looks like its proprietary TI/BB. They claim to have overcome the drawbacks of Multibit DACs with this architecture.
The way I understand it is that in Advanced Segment the ICOB converter/decoder is combined with a Multilevel Delta-Sigma.
However, I failed to find out if the ICOB converter/decoder is same as a multibit decoder or something different ?
Regarding bit-gain with oversampling:
It is actually half a bit approx that one gains per doubling of sampling rate (without doing anything else like noise-shaping). I.e. we gain 3 dB per doubling. The simple non-mathematical explanation is that the quantisation-noise POWER of our converter stays always the same, regardless of sampling rate.
But with the doubling of the sampling rate it is spread over double the frequency range and therefore the part that is contained in the audio range is halved. And voila: Half the power = -3dB.
With noise-shaping one further influences the spectrum of the noise by the use of some sort of frequency dependant negative feedback scheme (to use a simple explanation).
The overall signal to noise ratio is still only 6 dB, but the error (i.e. quantisation noise) is pushed above the audio range, where it can be filtered out by the use of a lowpass filter.
The sigma-delta signal is not the same as PWM - it is better described as PDM (i.e. Pulse Density Modulation) since it is noise-like in its properties. With PWM it shares the property that it can be "demodulated" by averaging, i.e. filtering (LOW-pass filtering in our case).
An example for how such a delta-sigma signal looks like can be found here:
http://www.diyaudio.com/forums/showthread.php?postid=570516#post570516
Regards
Charles
It is actually half a bit approx that one gains per doubling of sampling rate (without doing anything else like noise-shaping). I.e. we gain 3 dB per doubling. The simple non-mathematical explanation is that the quantisation-noise POWER of our converter stays always the same, regardless of sampling rate.
But with the doubling of the sampling rate it is spread over double the frequency range and therefore the part that is contained in the audio range is halved. And voila: Half the power = -3dB.
With noise-shaping one further influences the spectrum of the noise by the use of some sort of frequency dependant negative feedback scheme (to use a simple explanation).
The overall signal to noise ratio is still only 6 dB, but the error (i.e. quantisation noise) is pushed above the audio range, where it can be filtered out by the use of a lowpass filter.
The sigma-delta signal is not the same as PWM - it is better described as PDM (i.e. Pulse Density Modulation) since it is noise-like in its properties. With PWM it shares the property that it can be "demodulated" by averaging, i.e. filtering (LOW-pass filtering in our case).
An example for how such a delta-sigma signal looks like can be found here:
http://www.diyaudio.com/forums/showthread.php?postid=570516#post570516
Regards
Charles
Oversampling for the curious the furious and the damned
Well, first up, a "Hello" to all. My first post here.
Funny, after a few years away from the Internet audio forums, I have poked my head in here and, well in 15 years nothing has changed. (I have not been away that long however.) Years ago I used to post this little essay when confusion about the various DAC (and ADC) implementations got over the top. It was written by Max Hauser, who works professionally in the field. Amusingly, after 12 years this posting is still archived, and saves a huge posting here.
http://members.chello.nl/~m.heijligers/DAChtml/Digital Theory/hauser.txt
This link contains some useful treatments as well. It is again amusing to see how old this stuff is.
http://members.chello.nl/~m.heijligers/DAChtml/Digital Theory/Digital theory.html
Well, first up, a "Hello" to all. My first post here.
Funny, after a few years away from the Internet audio forums, I have poked my head in here and, well in 15 years nothing has changed. (I have not been away that long however.) Years ago I used to post this little essay when confusion about the various DAC (and ADC) implementations got over the top. It was written by Max Hauser, who works professionally in the field. Amusingly, after 12 years this posting is still archived, and saves a huge posting here.
http://members.chello.nl/~m.heijligers/DAChtml/Digital Theory/hauser.txt
This link contains some useful treatments as well. It is again amusing to see how old this stuff is.
http://members.chello.nl/~m.heijligers/DAChtml/Digital Theory/Digital theory.html
Re: Oversampling for the curious the furious and the damned
Konnichiwa,
The points Master Hauser makes are quite valid and accurate. However, the mathematical presented by digital domain oversampling is far from simple. Even in this day and age of nearly unlimited DSP power most Digital filters operate using lookup tables (similar to Math Coprocessors) at a very limited resolution.
Yes, 1-Bit, MUSH, Delta-Sigma etc. systems plus using noiseshaping allied to multibit systems place the burden of performance in the realm of math instead of the analogue domain. This does not mean that the math used is any good though.
The black box nature of the digital filter means people simply assume the job was done well.... May I remind people of the "Pentium inside doesn't divide". As usual, when you assume you AssUMe....
And the bottom line also stands, without extra processing 256 X Oversampling "1-Bit" DAC = 9Bit resolution (follow the logic why 4 X OS = 2Bit Extra resolution), or:
1 X OS = +0Bit
2 X OS = +1Bit.
4 X OS = +2Bit
8 X OS = +3Bit
16 X OS = +4Bit
32 X OS = +5Bit
64 X OS = +6Bit
128 X OS = +7Bit
256 X OS = +8Bit
All that of course assumes perfect floating point math in the DF....
Sayonara
PS, okay, I admit, i was by 1Bit resolution claiming 256 X OS 1-Bit DAC = 8Bit, it's 9Bit, I should have calculated it through)
Konnichiwa,
Francis_Vaughan said:
The points Master Hauser makes are quite valid and accurate. However, the mathematical presented by digital domain oversampling is far from simple. Even in this day and age of nearly unlimited DSP power most Digital filters operate using lookup tables (similar to Math Coprocessors) at a very limited resolution.
Yes, 1-Bit, MUSH, Delta-Sigma etc. systems plus using noiseshaping allied to multibit systems place the burden of performance in the realm of math instead of the analogue domain. This does not mean that the math used is any good though.
The black box nature of the digital filter means people simply assume the job was done well.... May I remind people of the "Pentium inside doesn't divide". As usual, when you assume you AssUMe....
And the bottom line also stands, without extra processing 256 X Oversampling "1-Bit" DAC = 9Bit resolution (follow the logic why 4 X OS = 2Bit Extra resolution), or:
1 X OS = +0Bit
2 X OS = +1Bit.
4 X OS = +2Bit
8 X OS = +3Bit
16 X OS = +4Bit
32 X OS = +5Bit
64 X OS = +6Bit
128 X OS = +7Bit
256 X OS = +8Bit
All that of course assumes perfect floating point math in the DF....
Sayonara
PS, okay, I admit, i was by 1Bit resolution claiming 256 X OS 1-Bit DAC = 8Bit, it's 9Bit, I should have calculated it through)