Trileru,
Now I think you are getting the right idea about Vref supplies by considering JohnW's advice. Maybe noise isn't the only thing that matters a lot for Vref, or for clocks? I would suspect maybe not. Whatever approach you might decide to try, implementation will always have its effects. Probably best to use a 4-layer board if you can, and try to use good engineering practices.
Now I think you are getting the right idea about Vref supplies by considering JohnW's advice. Maybe noise isn't the only thing that matters a lot for Vref, or for clocks? I would suspect maybe not. Whatever approach you might decide to try, implementation will always have its effects. Probably best to use a 4-layer board if you can, and try to use good engineering practices.
Last edited:
If I recall correctly the circuit of JohnW is from 2004. It would be interesting to ask him what he would use today.
One thing I am sure of ... the ZMR25H won the used anymore with its 65 µV noise. He would use an LT3042 for that
Probably without the NE5534 (are these still produced?). IMHO the circuit is outdated or the parts are. MAX6070 would be a good candidate.
One thing I am sure of ... the ZMR25H won the used anymore with its 65 µV noise. He would use an LT3042 for that
Probably without the NE5534 (are these still produced?). IMHO the circuit is outdated or the parts are. MAX6070 would be a good candidate.
Last edited:
If I recall correctly the circuit of JohnW is from 2004. It would be interesting to ask him what he would use today.
One thing I am sure of ... the ZMR25H won the used anymore with its 65 µV noise. He would use an LT3042 for that
The filter would eliminate whatever the voltage reference might present. We're talking about layout practices dominating the noise at this point, not the reference. There's a point where parts matter less than topology and implementation.
Combustion engine tech has been around 150 years almost. Refrigerator motors are same ol' thing. They work. Maybe manufacture and materials has improved, but there's things that just work. You can use newer parts for the fun of it but that doesn't mean it would translate into any practical gain. Especially since we're talking about sub 100-130dB signals. Maybe measuring apparatus would detect differences but I don't think ears. You might THINK you hear something different, but I don't think you can hear anything in the -170dB range for small differences, even for Vref.
LT3042 has good specs but you need good layout practices to get there. Separate modules without sense lines doesn't make sense. You're better off with lower tier parts mounted near the chip. Just because it's a LT3042 doesn't automatically mean better. It can be worse than a good implemented "lower tier"/older/cheaper part with worse specs.
JohnW tried AD797 instead of NE5534 for this topology and it was worse, oscillated because it's running at unity gain. Noise was worse.
I used the 3V or 3.3V version of MAX6070 for a 3D printer board that I added variable voltage/current to a dc/dc reg to have good control for a range of different laser diodes (blue engraving and lower power uv pcb work). But didn't yet get to write the code and test it.
I need the 5V version for this DAC, might use it.
Gee, it didn't do those bad things for me. Maybe take a look at figures 37, 38 and associated text in the data sheet.
I'm talking specifically for the circuit he presented. He talks about it in this post:
Capcitor type for Vref
He's also mentioning that adding a lower noise reference is not going to help. That noise is taken out by the filter anyway. That's the beauty of the circuit, small part count, simple and cheap.
Capcitor type for Vref
He's also mentioning that adding a lower noise reference is not going to help. That noise is taken out by the filter anyway. That's the beauty of the circuit, small part count, simple and cheap.
As Vref defines the output scale, could one go a little higher, to 5.5V? Absolute maximum Vref voltage is stated as 6.0V while max recommended operating voltage at 5.5V. So a 5.2-5.3V seems reasonable. I'm thinking this would squeeze a bit higher dynamic range?
Also the datasheet states that Vref can be +0.3V higher than AVCC.
edit: actually the Vref value is not that important is it? If using the digital volume control, that means that most of the time the output voltage is always going to be very low usually. I need 40mV rms to my amp for normal listening and that is even a bit on the louder side. So theoretically even the 3.3V reference would do for this. What are the downsides of using a lower reference? Say 1V. You get 1V max swing at max. I presume dynamic range suffers?
Also, why would this Vref stability matter in time? Presumably if you're 100% on the digital pot, the output max swing varies and that may affect the sound? If Vref varies, does that translate into a 20% digital volume setting varying proportionally? Or does it matter for when the signal is hitting the upper Vref rail (digital volume 100%)?
later edit: so AVCC is what actually powers the output drivers, so then if AVCC varies, when 100% output then the output varies because it hits the AVCC rails, so then if Vref varies it would be proportionally translated into the output varying irrespective of the output amplitude. Did I get it right? If so, then to avoid AVCC fluctuations one would be better off staying at max 90% on the digital volume if passively attenuating later in the chain? This way AVCC fluctuations would be somewhat irrelevant.
Also the datasheet states that Vref can be +0.3V higher than AVCC.
edit: actually the Vref value is not that important is it? If using the digital volume control, that means that most of the time the output voltage is always going to be very low usually. I need 40mV rms to my amp for normal listening and that is even a bit on the louder side. So theoretically even the 3.3V reference would do for this. What are the downsides of using a lower reference? Say 1V. You get 1V max swing at max. I presume dynamic range suffers?
Also, why would this Vref stability matter in time? Presumably if you're 100% on the digital pot, the output max swing varies and that may affect the sound? If Vref varies, does that translate into a 20% digital volume setting varying proportionally? Or does it matter for when the signal is hitting the upper Vref rail (digital volume 100%)?
later edit: so AVCC is what actually powers the output drivers, so then if AVCC varies, when 100% output then the output varies because it hits the AVCC rails, so then if Vref varies it would be proportionally translated into the output varying irrespective of the output amplitude. Did I get it right? If so, then to avoid AVCC fluctuations one would be better off staying at max 90% on the digital volume if passively attenuating later in the chain? This way AVCC fluctuations would be somewhat irrelevant.
Last edited:
All true but no designer will choose a 65 µV noise part if there are way better around in 2020. Nothing to do with practical gain but just the time one lives in.
Regarding the LT304x: I thought you haven't used that part until now (virtually, in a PCB layout program). Please first try and test in real life situations. No sane designer will use the oldest part or the part he has in a drawer. Even if the newer part is slightly worse, which is unlikely, the newer part is available in abundance. In the industry many PCB revisions are just because availability of a certain part is in question.
The schematic you are pushing (by JohnW) is outdated. I therefor would adapt it to 2020 parts if one likes to do things the most difficult way instead of choosing what is among the best on the market. It makes no sense to keep looking back and even to think of designing a PCB with those parts if they are hard to find, worse than their newer cousins or antique. That is why, just like LT304x, I suggested MAX6070. Again one that may fit your project. We will see if the part is in the BOM
Last edited:
The analogy with combustion engines/refrigerator motors is flawed. Please check a recent A+++ refrigerator to see where the condensor is, how the valve works and how the compressor starts and stops. The basis may be the same as 20 years ago, the way things are done in real life are way different.
If we would choose things that "just work" there would be no progress at all.
If we would choose things that "just work" there would be no progress at all.
The main advantages of an LT3042 are the ease with which you can put a filter between the reference and the output buffer and the performance of the output buffer. The reference itself is not particularly good, at least not when you look at the white noise (I haven't checked the 1/f noise). An 85A2 glow discharge tube and an LT1236 buried Zener reference are far better in that respect.
https://www.diyaudio.com/forums/power-supplies/359652-fine-ic-voltage-regulators-12.html#post6334678
https://www.diyaudio.com/forums/power-supplies/359652-fine-ic-voltage-regulators-12.html#post6334678
Design change: A 85A2 glow discharge tube it will be In a separate casing with also separate HV PSU . We drop the 85V with naked tantalum resistors, buffering with ECC101 with gold pins. DC filament PSU of course.
We could also go the hardest way, just a MAX6070 with some caps at the output without any added stuff
In a separate casing with also separate HV PSU . We drop the 85V with naked tantalum resistors, buffering with ECC101 with gold pins. DC filament PSU of course.We could also go the hardest way, just a MAX6070 with some caps at the output without any added stuff
Last edited:
The schematic you are pushing (by JohnW) is outdated. I therefor would adapt it to 2020 parts if one likes to do things the most difficult way instead of choosing what is among the best on the market. It makes no sense to keep looking back and even to think of designing a PCB with those parts if they are hard to find, worse than their newer cousins or antique. That is why, just like LT304x, I suggested MAX6070. Again one that may fit your project. We will see if the part is in the BOM
Using the better specs part in a vacuum doesn't yield the best results. AD797 instead of NE5534 results in worse noise in the same circuit.
You keeping saying "it's 2020 man use new parts" doesn't instill that much confidence. Riddle me this: how come AK4468, 8 channel DAC, released last year, has worse specs than ES9018 which was released 11!!! years ago.
Coming back to the reference, MAX6070 has nice specs. One problem I need to solve is the fact that the 5V version has a 5.5V max input, and 5.2V min input. AK4468 datasheet states lower performance for a 3.3V Vref, not by much but it's there. I'd go with the 4.096V MAX6070 as I want to power it from AVCC. Else I need another reg for it. Then people would recommend LT3042 and I'd get in the stupid position of using a LT3042 to power a reference for Vref. Or I can bump AVCC to 5.3-5.4V and I can directly power the voltage reference.
You could use your favourite 7805/LM317 for that A germanium diode out of the drawer from the GND pin to GND (7805) and you are done. Just joking as you like to make it complicated. I would not play too much with raising Vref. I would follow datasheet advice. You would need MAX6071B for 5V.
Using the worse specced part in a vacuum doesn't yield the best results either. You don't need confidence in me (worst thing one can do) but keeping to old stuff one already has gives the same feeling. I see you are slowly getting used to the fact that better stuff exists.
AD797 is famous, it is excellent but a wild horse and much fakes are made. AKM versus ESS: maybe the latter just makes better stuff? Why you would choose an AKM with worse specs when you know ESS has better parts is beyond me?!
A germanium diode out of the drawer from the GND pin to GND (7805) and you are done. Just joking as you like to make it complicated. I would not play too much with raising Vref. I would follow datasheet advice. You would need MAX6071B for 5V.Using the worse specced part in a vacuum doesn't yield the best results either. You don't need confidence in me (worst thing one can do) but keeping to old stuff one already has gives the same feeling. I see you are slowly getting used to the fact that better stuff exists.
AD797 is famous, it is excellent but a wild horse and much fakes are made. AKM versus ESS: maybe the latter just makes better stuff? Why you would choose an AKM with worse specs when you know ESS has better parts is beyond me?!
Last edited:
I assume you are referring to AKM4458. As this thread is not specifically related to that chip it would clarify the discussion if you would mention the chip you are referring to. AKM4458 has a different power supply architecture compared to e.g. AKM4490 & AKM4493 so generalizations should be avoided.
LM317+denoiser has over/undershoot for two minutes or so until it stabilizes. Not by much, but I didn't see any info for 5V. for 15V it goes to 14/16V and settles at 15. I don't know if it's a percentage or it does +-1V at any output voltage. But even as a percent, sizing to 5.3V means an upswing to 5.65V which would be over the MAX6070 limit. AK4468 VACC is ok with anything to 6V especially for two minutes. LM317+denoiser is perfectly fine for +-12 to 15V for opamps.
The new parts present new challenges. They are more touchy to implement. Besides the higher part price I have to deal with layout headaches and limitations. If the application does not absolutely mandate a LT3042 I'm not going to risk creating more problems than solving.
The new parts present new challenges. They are more touchy to implement. Besides the higher part price I have to deal with layout headaches and limitations. If the application does not absolutely mandate a LT3042 I'm not going to risk creating more problems than solving.
You see that it is exactly the LM317 and denoizer causing this don't you? A regulator like that will blow up almost any digital circuit! It is the old parts with updated circuitry that cause trouble here, not the new parts.
The dogma to desparately wanting to implement LM317 and denoizer
You could do it with just 1 small IC and a small 5V local reg like MIC5205 or the like.
Last edited:
MAX6071 seems to have a faster settle time than 6070. Also the A variants have better tempco than the B.
It is irrelevant if the behavior is not affecting the circuit if it's taken into account. It isn't intrinsically good or bad. It just is. If that behavior is affecting the circuit in a bad way then you do not use it. For example, you don't make a 5.3V supply for a 5.5V max chip, with a lm317+denoiser. That would be a bad example of using it. Using it for opamps supply is perfectly fine as +-1V means nothing, especially for only two minutes. It's just the idea that bugs you, and that is a human problem, not a technical problem. It's just something technical that happens. It is not a problem. It is a problem for the audiophile because it doesn't make him/her feel good. That is another issue and it has to do with psychology, not with tech.
It is irrelevant if the behavior is not affecting the circuit if it's taken into account. It isn't intrinsically good or bad. It just is. If that behavior is affecting the circuit in a bad way then you do not use it. For example, you don't make a 5.3V supply for a 5.5V max chip, with a lm317+denoiser. That would be a bad example of using it. Using it for opamps supply is perfectly fine as +-1V means nothing, especially for only two minutes. It's just the idea that bugs you, and that is a human problem, not a technical problem. It's just something technical that happens. It is not a problem. It is a problem for the audiophile because it doesn't make him/her feel good. That is another issue and it has to do with psychology, not with tech.