Hans,
The second power supply for the outputs of the FIFO buffer is not 7v. It can be somewhere between about 3.3v and 5.5v depending on what voltages the dac digital inputs need. Also, the reason for two power supplies for the FIFO buffer is because it uses isolation too. There is a clean side and a dirty side.
Mark
Is the transformer output stage something like this as promoted by Bisesik?
Attachments
Marcel explained earlier that this single-bit FIRDAC performance is not particularly dependent on resistance matching or tolerances as such. The opposite phases also have got some sort of interleaving of signals to balance out nicely. It makes sense to me. So, will stick to 0.1% tolerance for the bulk foil type also.
The only reason why I am supporting this change to foil type is possibly material characteristics make it better sounding than premium metal films. Nothing to do with noise comparisons also. So, 32pcs 3k 0.1% 0805 ordered today and will be delivered to you direct from TC
Last edited:
O.k. thank you.
I took these figures from Andrea's site, but I'm happy to change this.
By applying two voltage, just as on the USBoverI2S, is meant that they are from independent sources, just like all other voltages in the diagram.
With the interlink termination I advised to use 100R+2n7, but 100R+1n will be better and at the same time you could remove the 270pF at the Dac side that you added.
This will minimally affect the the level at 20Khz.
Hans
Attachments
Mark,
For differential outputs it's not important that they are equal in amplitude.
Output voltage = A-X*(-A), where A is the output level at one side and X*(-A) is the output level at the other side.
When X=0, you have SE, when X=0.5, output voltage is 1.5A and with X=1 it becomes 2A.
Where it differs is in the suppression of CM, B+X(-B)
When X=0, there is no suppresion at all and you have the full B in your signal, in total therefore A+B
With X=0.5, output signal is 1.5A+0.5B
With X=1, output signal will be 2A.
However linearity is assumed to be perfect in the above calculation, which may not always be the case with a Firdac as I have measured at levels between -10dB and 0dB because digital circuits do not have exactly identical output voltages and output resistances.
Marcel already cleverly addressed this by using two flip flops in parallel for one and the same bit, this makes things better but not fully cures it.
That's why I was curious what 2x8 instead of 4X4 does do to this problem.
Hans
"Sound" is still something - so you as an engineer, what would you say would contribute to better sound if noise isn't it?
- FR
- THD / IMD
- Timing / phase
- sauce... secret
//
Okay. But does "not particularly dependent" mean to -60dB, -100dB, -infinity dB? How far down does human hearing go?
Understood. But now add an error term to each of A and -A due to imperfections in the dac arrays. In time domain view, sometimes a fraction of the error difference will be in the common mode inversion (the error of A and the error of -A will both be moving up or down together, all or a fraction of which is equivalent to a common mode term).
Now suppose that the ear/brain system can recognize A + Ea as intelligible music. You can hear details of all the instruments and so on. Further suppose the same is true for -A and its E-a error component. It sounds pretty close to real music too, just sounds a little different than A and its error term.
But what happens during those moments when part of the error terms are moving in the same direction by the same amount and thus equivalent to common mode noise, and are therefore attenuated? Is what's left equally decipherable to the ear/brain system as real music with credible musical detail? What we find when we try differential summing is the ear/brain system finds some of the credible low level musical information has disappeared, and what we have left is a simpler signal that sounds more like an lossy mp3. IOW, we find it sounds 'better' if we don't differentially sum.
Last edited:
Mark,
Yes fully agree, that’s what I referred to as Firdac’s not being perfectly linear.
Because of that I suppose that more taps may have some beneficial averaging effect.
There is a nice way to “see” this on a scope in X-Y position when playing a mono piece of music (or a mono sine wave of several levels).
In a differential situation you would have to get a perfectly straight line under an angle of ca 45 degrees, depending how equal gain is in both branches.
But every deviation from a perfectly straight line means non linear or some other form of distortion.
Hans
Yes fully agree, that’s what I referred to as Firdac’s not being perfectly linear.
Because of that I suppose that more taps may have some beneficial averaging effect.
There is a nice way to “see” this on a scope in X-Y position when playing a mono piece of music (or a mono sine wave of several levels).
In a differential situation you would have to get a perfectly straight line under an angle of ca 45 degrees, depending how equal gain is in both branches.
But every deviation from a perfectly straight line means non linear or some other form of distortion.
Hans
Last edited:
Marcel,
Not exactly. What I was trying to explain is that differential summing sounds like a lossy encoded MP3. Why?
Please recall that there is only one pressure wave impinging on each eardrum. From that, the brain has to deconstruct it into an assemblage of instruments (including vocal instruments). We don't know exactly how the ear/brain does that trick, we just know that it does.
Empirically we also know that if we take a SE output, that it sounds like there is more low level musical detail along with some increased distortion/noise.
OTOH if we differentially sum, then we find get less low level musical details along with less distortion/noise.
Okay, let's for the moment assume that low level musical details, distortion, and or noise are separable problems. If we focus on the low level musical details problem, we find that differential summing can lose information that is actually on a recording. That's why some people refer to getting all the musical information from a recording as 'retrieval.' Differential summing is IMHO an easy way destroy high quality retrieval and reduce a high quality to recording to more like MP3 quality. IOW, its not a good thing for quality. Maybe its a good thing for low cost.
The goal for high quality reproduction should be to retain/reproduce all low level musical information and deal with distortion and noise by better design in the first place so there is less distortion and noise to bother the listener.
What metal foil resistors may bring to equation is they are likely to help reduce signal-correlated noise. From the numbers someone might not think it can be a problem, but after have tried some different so-called low-noise metal film resistors I could find no other physical explanation for the difference in sound other than resistor Excess Noise. It is known that metal foil resistors are tend to be lower in excess noise than metal film due to construction details. It is also known that manufactures of resistors such as Susumu make resistors intended for audio use. The audio resistors are lower noise than standard Susumu resistors, but the tradeoff is that some of the other specs aren't as good.
Not exactly. What I was trying to explain is that differential summing sounds like a lossy encoded MP3. Why?
Please recall that there is only one pressure wave impinging on each eardrum. From that, the brain has to deconstruct it into an assemblage of instruments (including vocal instruments). We don't know exactly how the ear/brain does that trick, we just know that it does.
Empirically we also know that if we take a SE output, that it sounds like there is more low level musical detail along with some increased distortion/noise.
OTOH if we differentially sum, then we find get less low level musical details along with less distortion/noise.
Okay, let's for the moment assume that low level musical details, distortion, and or noise are separable problems. If we focus on the low level musical details problem, we find that differential summing can lose information that is actually on a recording. That's why some people refer to getting all the musical information from a recording as 'retrieval.' Differential summing is IMHO an easy way destroy high quality retrieval and reduce a high quality to recording to more like MP3 quality. IOW, its not a good thing for quality. Maybe its a good thing for low cost.
The goal for high quality reproduction should be to retain/reproduce all low level musical information and deal with distortion and noise by better design in the first place so there is less distortion and noise to bother the listener.
What metal foil resistors may bring to equation is they are likely to help reduce signal-correlated noise. From the numbers someone might not think it can be a problem, but after have tried some different so-called low-noise metal film resistors I could find no other physical explanation for the difference in sound other than resistor Excess Noise. It is known that metal foil resistors are tend to be lower in excess noise than metal film due to construction details. It is also known that manufactures of resistors such as Susumu make resistors intended for audio use. The audio resistors are lower noise than standard Susumu resistors, but the tradeoff is that some of the other specs aren't as good.
Last edited:
In most textbooks resistor noise is divided into thermal noise and current noise. Excess noise is just another terminology used for current noise. And as I already showed in post #1625 current (or excess) noise in this case is very small in relation to thermal noise. As you haven't done proper AB testing the obvious explanation for the perceived difference in sound is expectation and subjective bias.
BTW for more information about resistors noise this over 60 year old paper is good read.
https://conradhoffman.com/papers_lib/TI_Noise_Prec_Resistors.pdf
You are guessing, nothing more. Small numbers do not mean inaudible. It is already known that signal-correlated noise is much more audible than Johnson noise. Two very different things. Besides, how do you know that end cap noise is white? Guessing again. It may sound nothing like what you are assuming. How do you do know it isn't popcorn noise or frying noise? You don't.
Not guessing. The resistor noise in this case is almost all from thermal noise, you can't escape that even with your audiophile mumbo jumbo.
What you fail to understand is that metal foils do not have lower excess noise than low noise thin films. Metal foils have better tolerance and stability.
What you fail to understand is that metal foils do not have lower excess noise than low noise thin films. Metal foils have better tolerance and stability.