In the 4th edition of Valve amplifiers Morgan introduces the ‘statistical regulator’. It is made out of a CCS feeding a string of 5,6V zeners. Nothing new feeding a zener with a CCS, but MJ has a series of arguments and measurements to show that the stack of 5,6V zeners does much better than normal high voltage zeners or other common shunt devices.
On page 176 of the 3rd edition of VA, MJ writes that the 6,2V zener is the best, as it combines the true zener action with the avalanche effect: less temperature dependent and less noise. Also datasheets show that the impedance of a 6,2V zener is (much) lower than that of a 5,6V zener (on one datasheet I saw 2 vs 7 ohm). I am sure MJ has his reasons to prefer the 5,6V to the 6,2V, but I would like to understand why. Can someone shed a light on this?
many thanks! Erik
On page 176 of the 3rd edition of VA, MJ writes that the 6,2V zener is the best, as it combines the true zener action with the avalanche effect: less temperature dependent and less noise. Also datasheets show that the impedance of a 6,2V zener is (much) lower than that of a 5,6V zener (on one datasheet I saw 2 vs 7 ohm). I am sure MJ has his reasons to prefer the 5,6V to the 6,2V, but I would like to understand why. Can someone shed a light on this?
many thanks! Erik
Probably a question of best match.I suspect thea temp.dependecy of the CCS is best 🙂 balanced by the zeners of 5V6.
6V2 is oke standing alone but with an extra base-emitter junction voltage p.x. better take a 5V6.
Mona
6V2 is oke standing alone but with an extra base-emitter junction voltage p.x. better take a 5V6.
Mona
From the horse's mouth:
"It seems to depend on manufacturer as to what you get between 5.6V and 6.2V, although most seem to agree that 6.2V gives the lowest slope resistance. However, the near-zero temperature coefficient and low slope resistance at 6.2V is due to a balancing act between avalanche mode and true Zener mode. In general, Zener mode is quieter, and dominates below 5.6V, so it is possible that 5.1V would be even quieter. However, I had the very practical limitation that my tag board I had could accommodate 35 diodes, and the roll of 5.6V diodes that I had would give exactly the 195V I needed. My main point was that a lot of quiet diodes in series is even quieter than you would think, and ideal for circuits (such as cascodes) that have miserable power supply rejection. It is quite possible that 5.1V or 6.2V might be quieter. 12V is much noisier."
"It seems to depend on manufacturer as to what you get between 5.6V and 6.2V, although most seem to agree that 6.2V gives the lowest slope resistance. However, the near-zero temperature coefficient and low slope resistance at 6.2V is due to a balancing act between avalanche mode and true Zener mode. In general, Zener mode is quieter, and dominates below 5.6V, so it is possible that 5.1V would be even quieter. However, I had the very practical limitation that my tag board I had could accommodate 35 diodes, and the roll of 5.6V diodes that I had would give exactly the 195V I needed. My main point was that a lot of quiet diodes in series is even quieter than you would think, and ideal for circuits (such as cascodes) that have miserable power supply rejection. It is quite possible that 5.1V or 6.2V might be quieter. 12V is much noisier."
Zener voltage adds linearly, but presumably zener noise adds quadratically? This assumes no correlation, but I suppose there is a possible correlation mechanism via the common current.
The pedant in me can't help but poke a hole in Jones' assertion that "noise is due to the granularity of current, we should expect the noise of a reference to be proportional to the inverse square root of operating current".
In fact the the opposite is true- shot noise is proportional to the square root of current, not its inverse (otherwise no current at all would give infinite noise...). The reason his noise measurements get worse at low currents is because the noise current density is constant but the diode slope resistance increases. If you take that into account, his measurements conform exactly to the noise predicted by Schottky's theorem: iN2=2qIdcB
Hi SY,
thanks for asking Morgan and if you want forward my thanks to him as well. So 5,6V is a safe bet: I'll start from there and maybe try other values later.
His arguments in favor of the Statistical regulator are better ripple attenuation and short circuit proof (when comparing to a Maida). I therefore barely dare to ask 🙂 but did MJ comment anything about a perceived/subjective difference in sound between different regulators?
Erik
thanks for asking Morgan and if you want forward my thanks to him as well. So 5,6V is a safe bet: I'll start from there and maybe try other values later.
His arguments in favor of the Statistical regulator are better ripple attenuation and short circuit proof (when comparing to a Maida). I therefore barely dare to ask 🙂 but did MJ comment anything about a perceived/subjective difference in sound between different regulators?
Erik
Last edited:
I think that will be highly dependent on the circuit it's attached to. The main raison d'etre is low noise, and that it indeed does achieve. The use of a Kelvin connection cap is icing on the cake.
Hi Erik,
Just saw this and not sure if you're still interested but a few years ago VoltSecond posted a number of times on the AudioAsylum on zeners. He stated there that the 6.2V value is noteworthy as it is high enough to regulate well yet low impedance and low noise. He also pointed out that lower voltage zeners don't regulate well. The figure I remember is anything below 5.1V.
BTW, I found this thread because I was looking to see if you'd taken your idea for the HV shunt reg (originally with GU50 and bipolars) any further. How'd it go?
Hi Hearinspace
Thanks for your reply! That sounds good: the results MJ obtained with the 5,6V are very good indeed, but I am going to try the 6,2V as well, one day.
Regarding the GU50 shunt regulator: that one is still on the 'drawing board' 🙂 We had a move and so in the last months, but I think I'll start testing and building soon again.
Cheers, Erik
Thanks for your reply! That sounds good: the results MJ obtained with the 5,6V are very good indeed, but I am going to try the 6,2V as well, one day.
Regarding the GU50 shunt regulator: that one is still on the 'drawing board' 🙂 We had a move and so in the last months, but I think I'll start testing and building soon again.
Cheers, Erik
Thanks Erik, please post if you do. I'm looking at building a version of Rod's shunt cascode RIAA circuit and thinking about shunt regulators there. Also for same in power amp driver stage. Particularly interested in the BJT pair - something I have little experience with yet.
Sy,
The noise formula, 1/sqrtI what is the output DB? percentage? I want to use a Zener regulator for an SV83 for G2 voltage. Current will be limited to 8.3ma. So at maximum, 1/sqrt .0083 = 11. Eleven what? And is that high? The tube will only be running at about 13ma so actual current will be more like 4ma for both G2.
(And pardon me if this is an inane question.)
The noise formula, 1/sqrtI what is the output DB? percentage? I want to use a Zener regulator for an SV83 for G2 voltage. Current will be limited to 8.3ma. So at maximum, 1/sqrt .0083 = 11. Eleven what? And is that high? The tube will only be running at about 13ma so actual current will be more like 4ma for both G2.
(And pardon me if this is an inane question.)
It's a proportionality, not an actual formula for finding absolute noise. It's also only an epirical approximation that works OK over a range of currents, but does not derive from any fundamental mechanism and does not work at all currents. For technically correct results you need to find the shot noise current in the string of zeners (assuming true zeners, not avalanche diodes):
in = sqrt (N2qIdcB)
Where:
N = number of zeners
q = electron charge
Idc = zener current
B = bandwidth
You can then multiply this by the Thevenin resistance of the Zener string in parallel with whatever external resistances you have, to find the noise voltage across the string. Of course, if you have a bypass cap then this will shunt the noise voltage quite effectively, possibly to the point where a single avalanche diode is actually no worse than a statistical regulator.
Last edited:
It's shown right in the manufacturer's data sheets that zener diodes below about 6 volts have much lower noise. This is nothing new. I don't know about the quality of regulation though for those lower voltage zeners.
I thought that noise sources added as the square root of the sum of the squares?
I thought that noise sources added as the square root of the sum of the squares?
FWIW, I've done some PCBs for people interested in following this up - there's a thread on this at Audio-Talk.... MJ Statistical Regulator PCBs - audio-talk
Code:
However, the near-zero temperature coefficient and low slope resistance at 6.2V is due to a balancing act between avalanche mode and true Zener mode.According to available datasheets lowest temp coefficient is for Zeners at around 5.1V, this is also what I learned long time ago, see here for an example http://www.farnell.com/datasheets/1739884.pdf
however it seem to vary somewhat with manufacturers, some show lowest temp coeff for even lower voltages.
High voltage "Zener diodes" are not Zener diodes at all as they use avalanche effect, they are much noiser and have high temp coeff as MJ says, in one application for a bias control board I use 10 series connected 5.1V zeners instead of a 51V with much better results, (I am more concerned about temperature drift than about noise in that application).
Is there a theoretical or practical reason to omit the noise filtering capacitor(s)? Are they excessively large or ridiculously expensive?