ESS recommends +3.3v +-5% for AVCC for each of the two AVCC supplies. They also specify low noise in order to minimize DAC distortion and noise.
As can probably be seen there are long discussions in the forum about all the different ways to design low noise references, and the various trade offs to consider. I don't think we should get sidetracked into that this early in the going. We are trying to focus on the bigger picture and not get deeper into the weeds than we have to.
Regarding Tom Jung's GLED431, here are Walt's later comments he appended to original article:
"Addendum of March 8, 2016:
The circuit originally shown with a LT6200 op amp was/is intended as a conceptual example of low noise reference voltage scaling, not a complete working regulator!
The LT6200, like many low noise op amp devices, also has a very high bandwidth. From that it follows that it can readily oscillate. There are past caveats here and elsewhere about using very high BW parts within regulators (AD797, etc). Always proceed with caution, and make good use of a scope to monitor the output."
If someone really wants to get into it and is equipped to do the design and testing for a +3.3v version with known good PSRR, stability, etc., then let me know and I will see about getting Walt's updated paper to you.
As can probably be seen there are long discussions in the forum about all the different ways to design low noise references, and the various trade offs to consider. I don't think we should get sidetracked into that this early in the going. We are trying to focus on the bigger picture and not get deeper into the weeds than we have to.
Regarding Tom Jung's GLED431, here are Walt's later comments he appended to original article:
"Addendum of March 8, 2016:
The circuit originally shown with a LT6200 op amp was/is intended as a conceptual example of low noise reference voltage scaling, not a complete working regulator!
The LT6200, like many low noise op amp devices, also has a very high bandwidth. From that it follows that it can readily oscillate. There are past caveats here and elsewhere about using very high BW parts within regulators (AD797, etc). Always proceed with caution, and make good use of a scope to monitor the output."
If someone really wants to get into it and is equipped to do the design and testing for a +3.3v version with known good PSRR, stability, etc., then let me know and I will see about getting Walt's updated paper to you.
Regarding use of spare AD797, may I ask if they are new and therefore SMD? Old and possibly DIP? Assuming SMD, are you comfortable prototyping in SMD?
Probably best to make some updates to the original ESS recommendations, but they are a great start in multiple ways.
OSB.rtf circuit is laid out for dual opamps and DIP sockets, so it has been tried with opamps like LME49720 and OPA1612.
Probably best to make some updates to the original ESS recommendations, but they are a great start in multiple ways.
OSB.rtf circuit is laid out for dual opamps and DIP sockets, so it has been tried with opamps like LME49720 and OPA1612.
Hi Marc,
chips are DIP, I'll follow the ESS recommendations with my 797's see how it measures up.
ta
Okay. You planning to do stereo 3-opamp output stages and stereo AVCC buffers with the AD797s?
Last edited:
Sorry, TL431 is a noisy pig. I don't use it.
If you really need a low noise reference built out of whatever parts you have on hand, then Terry was basically right that LEDs or VBE-multipliers are the way to go. GLED-431 is basically a buffered VBE-multiplier. What Terry didn't say was that there is more to making one than his simple description might suggest. Such references have limited PSRR and line ripple and other noises can creep in from the power supply that powers the reference. Sure, if you run the reference on a battery then no problem, but in the real world we don't usually want to use batteries. So, in addition to a simple reference you need a clean power supply for it. That adds cost and complexity unless you have parts laying around already to build one, or something already built you don't might using for a dac project. Easiest thing to do is to use a decent low noise IC reference. All things considered, it might end up being cheaper than making a good discrete reference for a one-off project.
If you really need a low noise reference built out of whatever parts you have on hand, then Terry was basically right that LEDs or VBE-multipliers are the way to go. GLED-431 is basically a buffered VBE-multiplier. What Terry didn't say was that there is more to making one than his simple description might suggest. Such references have limited PSRR and line ripple and other noises can creep in from the power supply that powers the reference. Sure, if you run the reference on a battery then no problem, but in the real world we don't usually want to use batteries. So, in addition to a simple reference you need a clean power supply for it. That adds cost and complexity unless you have parts laying around already to build one, or something already built you don't might using for a dac project. Easiest thing to do is to use a decent low noise IC reference. All things considered, it might end up being cheaper than making a good discrete reference for a one-off project.
Last edited:
No, not even close. The circuit doesn't multiply a d*mn thing, but adds and outputs the LED Vf with a Vbe
https://www.diyaudio.com/forums/att...-reference-cell-_walts-blog-2014_092418_r-pdf
No, not even close. The little circuit divides the dynamic impedance of the LED plus the current source resistor approximately by the transistor Beta, being therefore very low, well under 1 ohm, and the frequency response is way up into the MHz, much higher than any regulated power supply that may feed this circuit. Not sure how you define the PSRR for this circuit, but almost nothing would creep into such a low dynamic impedance.
This circuit would beat hands down any integrated reference noise, even the LTC6655, at the price of voltage precision and temperature coefficient, which would be comparatively poor. And it costs $1 to build.
Okay, I stand corrected on memory of basic GLED Cell topology. Thank you. However, I will stand with my previous statements agreeing with Jung in reference to his 2016 GLED431 addendum:
"Addendum of March 8, 2016:
The circuit originally shown with a LT6200 op amp was/is intended as a conceptual example of low noise reference voltage scaling, not a complete working regulator!
The LT6200, like many low noise op amp devices, also has a very high bandwidth. From that it follows that it can readily oscillate. There are past caveats here and elsewhere about using very high BW parts within regulators (AD797, etc). Always proceed with caution, and make good use of a scope to monitor the output."
I would add that the opamp is powered from an external supply which should be clean.
For the sake of completeness, Jung also added:
"Addendum of December 26, 2016:
As noted, the GLED431 circuit is not one for high Vout temp stability. If that is critical, use a TL431 or other IC reference. But, it is nevertheless also true that the best drift will be realized if the temp-sensitive parts D1/Q1 are buffered from room air currents. A cheap and simple way to do this is to place them back-to-back, and assemble the circuit on a two-pin header strip. Once soldered up and checked for function and correct Vout, then slide a length of plastic tubing over the completed circuit, as follows. In doing this final step, first use squeeze some silicone compound within the tubing, before applying it over and around the GLED431circuit. This step surrounds and isolates D1/Q1 within the silicon, as well as holds the tubing in place, once the silicon has set up. The sample photo to the left shows the general idea. The components of the GLED431 circuit can just be seen through the translucent tubing. I have no doubt that this will greatly minimize the temperature bouncing of Vout."
Last edited:
You may stand wherever you want, but the claim that Walt's circuit doesn't have enough immunity to outside noise and other harmonic perturbation is incorrect. Please go ahead and implement and measure the circuit and compare with whatever integrated reference you want.
Otherwise, the poor precision and thermal stability were both already mentioned, so please stop introducing diversions (including about the opamp potential stability issues, nothing to do with your noise claim).
Otherwise, the poor precision and thermal stability were both already mentioned, so please stop introducing diversions (including about the opamp potential stability issues, nothing to do with your noise claim).
Hm, I wonder do we have a complete PS which is as good as GLED431 voltage reference and qualifies for the voltage reference usage as well?
In the power supply forum section, we’ve been discussing Elvee’s denoiser circuit, which turns ordinary LM or LT regulator in the one with over 120 dB PSRR, output noise voltage of 0,8 µV and output impedance in the micro ohms range.
D-Noizator: a magic active noise canceller to retrofit & upgrade any 317-based V.Reg.
Output voltage drifts about 10 - 20 mV, for the 12V nominal output, with transition from cold state to long term load on my evaluation design which is probably not good enough?
Could You enlighten me, how could an LM317 with.. Maybe.. 10kHz bandwith, to be transformed into a real full 1MHz regulator, like an LT3042? (with 2nV/sqrtHz noise floor..)
As far as I understood, the additional noise subtraction and other functionality works until the base regulator is keeping it's full effect. After that..
As far as I understood, the additional noise subtraction and other functionality works until the base regulator is keeping it's full effect. After that..
I’m sure you can read past first post in the referenced thread, what is likely all you have looked at, and find answer yourself (and real measurements comparison where pimped up LM317 beats LT3042 in the noise department). 
Look around post #519 (before and after)
D-Noizator: a magic active noise canceller to retrofit & upgrade any 317-based V.Reg.
Look around post #519 (before and after)
D-Noizator: a magic active noise canceller to retrofit & upgrade any 317-based V.Reg.
TL431 is not actually much worse than LTC6655. Considering it's simplicity
and price, it can be a very useful part. The noise can easily be filtered with
RC filter.
SYN08 is correct WRT to the GLED-431.
The LED's OP impedance is buffered so you basically get the OP impedance
of the BJT. This will be below 10 ohms for moderate currents and should
have no stability issues.
You just need a very high impedance current source pulling the reference up.
2 resistorrs and 2 JFets can yield a 5 mA current source with >10 Megohm
OP impedance. That will give you well over 100dB PSRR on any shunt ref of
10 ohms or less impedance.
I recently found these measurements by Gerhard.
https://www.diyaudio.com/forums/pow...ultra-low-noise-vref-gled431.html#post4589769
So in actual fact, if the parts prove to be consistent, a simple BZX84C3V3,
3.3V zener pulled up with a very high Z current source is all you need.
Follow it with some RC LPF and you are done. That would constitute a super low noise reference.
TCD