Ok, you cannot directly put a cap across a Zener reference and get much noise improvement. You have to add an additional R in series between the reference and the cap to get the cap to lower the self noise significantly. Start with 80 nV/rt Hz or so for 3 series regulatiors adding as to the square root of 3 multiplier and 50 nV/rt Hz, the specified noise of the Zener reference. At Zero Hz the output impedance of the filter will be 3-5K. Want to bet on it?
For everyone else, IF you add an RC filter then you screw up the DC regulation. No big deal, but it should be understood that you are not looking directly at the Zener reference once you add the filter, and its low Z is next to useless.
For everyone else, IF you add an RC filter then you screw up the DC regulation. No big deal, but it should be understood that you are not looking directly at the Zener reference once you add the filter, and its low Z is next to useless.
Last edited:
I'm up to the bet: how much is the filter output impedance to match the JFET Norton equivalent noise at whatever frequency you choose. You are wrong by an order of magnitude, at least. The total output impedance of the LM329 + filter will still be another order of magnitude lower than the JFET Norton.
But then, I have already mentioned that LM329 is not for you, the idea to put 3 x LM329 in series is indeed stupid. I understand that all you want is to reproduce the 30 years old Vendetta performance. You have zero interest in building a modern, better version, gain stage and power supply, which is again understandable. However, teaching and preaching that design in 2009 is like teaching students about the Archeropterix bone structure.
Last edited:
I would like to point out that most electronic concepts go back a long way, often many, many decades. The Norton equivalent concept is one such 'old' concept.
With more modern IC's, some improvements have been noted. It is thus true that the LM329 voltage reference is about 10 times lower noise than a 'low noise' Zener available in the past.
However, I would maintain that using my 'Norton' approach, the typical noise of the Norton equivalent, consisting of a 'low noise' low Gm fet of 4 ma and a 5K resistor would be less and certainly, equal, worst case to 50nV/ rt Hz that the LM329 is rated at. With the 100uf filter cap added across the resistor, the noise performance improves considerably. Those who doubt me, check it out for yourself. Of course, I have the ADVANTAGE of merely changing the value of the resistor to get the required output voltage. This is not possible with the LM329, and more complex circuitry must be added to get variable voltage, and I DOUBT that the approach in the app note to get higher voltage is very quiet, as well. Can't everyone see that a reduction in the self noise of the Norton equivalent at 35 Hz is going to be 100 times lower than 50nV/rt Hz or .5nV/rt Hz? This is due to the filtering action of the cap and the resistor. If I am wrong, show me where my calculations are mistaken.
With more modern IC's, some improvements have been noted. It is thus true that the LM329 voltage reference is about 10 times lower noise than a 'low noise' Zener available in the past.
However, I would maintain that using my 'Norton' approach, the typical noise of the Norton equivalent, consisting of a 'low noise' low Gm fet of 4 ma and a 5K resistor would be less and certainly, equal, worst case to 50nV/ rt Hz that the LM329 is rated at. With the 100uf filter cap added across the resistor, the noise performance improves considerably. Those who doubt me, check it out for yourself. Of course, I have the ADVANTAGE of merely changing the value of the resistor to get the required output voltage. This is not possible with the LM329, and more complex circuitry must be added to get variable voltage, and I DOUBT that the approach in the app note to get higher voltage is very quiet, as well. Can't everyone see that a reduction in the self noise of the Norton equivalent at 35 Hz is going to be 100 times lower than 50nV/rt Hz or .5nV/rt Hz? This is due to the filtering action of the cap and the resistor. If I am wrong, show me where my calculations are mistaken.
What is infuriating is that (a) you reject the concept of (super) regulator to feed a low noise amp and (b) you reject the idea to add some decent PSRR to the Vendetta gain stage (at least a cascode, as mr. Pass does in his XONO design).
Again, I understand your plan to change as little as possible in the Vendetta Research old design, but telling this is an optimal solution (and extending this particular solution as ideal, to other audio applications) is totally incorrect. If I have your permission I can post the entire Vendetta Research schematic, including those ugly paralleled JFETs in the power supply, for everybody to make an educated opinion. BTW, the second stage in your Vendetta Research has feedback, so it's only the input stage that is open loop. Otherwise, the idea to feed the RIAA network from a high impedance source is correct.
True, a decent regulator has feedback, but then again, only a poor designed regulator has overshoot, and getting 1milliohm output impedance up to 10MHz from a small parallel regulator is not a problem using modern components. Similarly, the noise can be made arbitrary low.
P.S. Dmitri's calculations are correct. J201 has though 6nV/rtHz at best, so you may double those numbers. And K246 is not at all cheap. Here you go.
Last edited:
Agreed. That's another good reason to provide a decent PSRR to the low noise gain stages, then the pressure to build ultra low noise power supplies would be significantly reduced. In my 0.32nV/rtHz preamp, I live happily with around 100nV/rtHz power supply noise.
Force and sense wiring and proper screening. And SMD parts, of course. And that 0805 inductive reactance @10MHz would be of course 30 MILLI ohm. And 20nH @10MHz would be 1.2 ohm.
Last edited:
Thank you Dimitri, for adding a reasoned input here. I got the same calculation with MY J203's made with Siliconix parts. Yes, 3nV/rt Hz is what the parts were originally characterized at. NP geometry, p 4-37 'Siliconix FET Data Book' 1977 Gm = .5ma/.2V or 2.5K (umhos) That gives me 3 (nV) times 2.5K times 5K divided by 1 million or 37.5 nV/rt Hz max, without cap.
I am stating typical noise here, from my own parts that I have many hundreds of. It is true that you could get a bad batch or a cheap and dirty equivalent clone that would be noisier. Syn08 is pushing toward that goal.
Now what about the LM 329 to an experienced engineering eye?
Well 50uV/rt Hz is pretty good. However, the worst case is MUCH WORSE!
They don't state it that way, BUT they do say 7uV is typical total noise. Guess what, WORST CASE IS 100uV! That is the noise between 10 Hz and 10K Hz.
Now, how does this correlate to 50 nV/rt Hz shown on the graph? p. 6
Well, a 10KHz bandwidth with some spill-over is approximately 150 times 50nV to equal about 7 uV. Now what does 100uV (worst case) imply? 660nV/rt Hz, about the same noise as a typical low noise low voltage Zener, perhaps 4.7V.
Now what is worst case and do manufacturers actually give you parts approaching it? YES! Has anyone had this experience with National Semi? YES, me. I have a bunch of LM411 fet input IC op amps, that don't get anywhere near typical spec. Therefore, can the LM 329 be trusted to be as quiet as they claim? If so, why a worst case spec that is more than 10 times worse than typical. Random chance? I doubt it. This is what you learn from EXPERIENCE, the equivalent of getting your hand burned on the heater when you are young. You are more careful, once you have had the experience.
Now this does not preclude this part from being valuable, but it is not the answer to all low noise applications that it is touted to be.
I am stating typical noise here, from my own parts that I have many hundreds of. It is true that you could get a bad batch or a cheap and dirty equivalent clone that would be noisier. Syn08 is pushing toward that goal.
Now what about the LM 329 to an experienced engineering eye?
Well 50uV/rt Hz is pretty good. However, the worst case is MUCH WORSE!
They don't state it that way, BUT they do say 7uV is typical total noise. Guess what, WORST CASE IS 100uV! That is the noise between 10 Hz and 10K Hz.
Now, how does this correlate to 50 nV/rt Hz shown on the graph? p. 6
Well, a 10KHz bandwidth with some spill-over is approximately 150 times 50nV to equal about 7 uV. Now what does 100uV (worst case) imply? 660nV/rt Hz, about the same noise as a typical low noise low voltage Zener, perhaps 4.7V.
Now what is worst case and do manufacturers actually give you parts approaching it? YES! Has anyone had this experience with National Semi? YES, me. I have a bunch of LM411 fet input IC op amps, that don't get anywhere near typical spec. Therefore, can the LM 329 be trusted to be as quiet as they claim? If so, why a worst case spec that is more than 10 times worse than typical. Random chance? I doubt it. This is what you learn from EXPERIENCE, the equivalent of getting your hand burned on the heater when you are young. You are more careful, once you have had the experience.
Now this does not preclude this part from being valuable, but it is not the answer to all low noise applications that it is touted to be.
I wish to state that I am trying to help aspiring audio designers here, not the competition, necessarily, but amateurs mostly, to address a difficult design problem with noise. It was ONLY used with the Vendetta Research input stage, a complementary folded cascode input stage that was sensitive to power supply noise, because sound quality was my first concern, not ease in manufacture. Other designs of mine have never been given the Norton equivalent reference. Usually, cap multipliers work very well, are just as quiet, easier to make, and cheaper. The cap multiplier on each gain block, coupled with a standard 3 terminal regulator pair, works pretty well for me. and uses the same parts as Syn08 would have to use for a single regulator. Quieter, too! Is this so 'out of fashion'?
So far no schematic I found on the internet or this forum has ever simulated an output impedance smaller than 1mohm at 10MHz. And this is simulation, i.e. ideal pcb and components. Sorry, I have a hard time accepting a real implementation that measures such output impedance. Meaningful measurements at such frequencies are not easy to obtain. I have my doubts. I'd be happy to know though, numbers for up to 500kHz.
As far as Vref noise is concerned, I will happily use the j201 before the lm329 because it fits better in what I do. This has nothing to do with what John Curl is doing, in fact I arrived at the solution before he divulged that he uses a similar setup with the j203 jfet. I am not concerned with sub microvolt noise because most people cannot even measure it, let alone achieve it in a circuit. I live in Toronto, like syn08, and the ground is infested with noise. BTW, for what is worth, Bob Pease has a nice article about ground noise filtering
ground noise rejection article
Everyone should use whatever makes them feel warm and cozy at night. But be not mistaken, what I see done and reported around here is not science. Call it what you will, but to me, an experiment to be called science will have to be a lot more rigorous than what I've seen so far reported here. That's why I reserve the right to doubt some of the numbers that get thrown around. One can choose to be like the French, who say "It works in practice, but does the theory hold?" or like others who say it the other way around.
As far as Vref noise is concerned, I will happily use the j201 before the lm329 because it fits better in what I do. This has nothing to do with what John Curl is doing, in fact I arrived at the solution before he divulged that he uses a similar setup with the j203 jfet. I am not concerned with sub microvolt noise because most people cannot even measure it, let alone achieve it in a circuit. I live in Toronto, like syn08, and the ground is infested with noise. BTW, for what is worth, Bob Pease has a nice article about ground noise filtering
ground noise rejection article
Everyone should use whatever makes them feel warm and cozy at night. But be not mistaken, what I see done and reported around here is not science. Call it what you will, but to me, an experiment to be called science will have to be a lot more rigorous than what I've seen so far reported here. That's why I reserve the right to doubt some of the numbers that get thrown around. One can choose to be like the French, who say "It works in practice, but does the theory hold?" or like others who say it the other way around.
Last edited:
Ikoflexer, it might be a good idea to learn the analog engineering simplified computations that Dimitri and I both used (without consulting each other first) which shows the easiest way to predict noise performance in a Norton equivalent or many other circuits.
As we all know, learning engineering or physics takes a lot of math, Calculus, Differential equations, Vector algebra, Linear algebra (matrix algebra) etc. Yet, typical design engineering does NOT involve this math as much as you might think. Usually a scientific calculator, slide rule, or even, pencil and paper, is enough to design many sub-circuits, or individual stages. People tend not to learn these simple approaches to circuit design, or remain suspicious of them. Tech's are often intimidated by 'engineers' because the techs lack the math background, even though it is usually unnecessary for making good quality audio circuits. 'Engineers' sometimes take advantage of this, by 'lording over' the uneducated amateurs. Please note this technicians.
Now Ikoflexer has stated that he uses a J201, while I use a J203. What is the difference? Just Idss The J201 has a lower Idss than the J202 or the J203. It is the American system of sorting devices. The Japanese would have just put a color: Y,GR,BL,V after a single part number. No big deal. If Ikoflexor made a Norton equivalent, could he use the J201 to do so? Yes. The only real change would be the load resistor that might go from 3-5K, to 10-20K. However the MAX Gm of the J201 is ONLY 1000 umho, so the combined gain and therefore potential noise will be about the same as a J202, or a J203. Is this clear enough, everyone?
As we all know, learning engineering or physics takes a lot of math, Calculus, Differential equations, Vector algebra, Linear algebra (matrix algebra) etc. Yet, typical design engineering does NOT involve this math as much as you might think. Usually a scientific calculator, slide rule, or even, pencil and paper, is enough to design many sub-circuits, or individual stages. People tend not to learn these simple approaches to circuit design, or remain suspicious of them. Tech's are often intimidated by 'engineers' because the techs lack the math background, even though it is usually unnecessary for making good quality audio circuits. 'Engineers' sometimes take advantage of this, by 'lording over' the uneducated amateurs. Please note this technicians.
Now Ikoflexer has stated that he uses a J201, while I use a J203. What is the difference? Just Idss The J201 has a lower Idss than the J202 or the J203. It is the American system of sorting devices. The Japanese would have just put a color: Y,GR,BL,V after a single part number. No big deal. If Ikoflexor made a Norton equivalent, could he use the J201 to do so? Yes. The only real change would be the load resistor that might go from 3-5K, to 10-20K. However the MAX Gm of the J201 is ONLY 1000 umho, so the combined gain and therefore potential noise will be about the same as a J202, or a J203. Is this clear enough, everyone?
Last edited:
John, 
although I'm not an engineer I do have a math background. Certainly enough to follow Dmitri's noise calculations. Still, even though it's a good theoretical exercise, it's still a theoretical exercise. I am a bit skeptical by nature, not of theory, but you know yourself how the real world is somewhat messy.
I tell you what I'd like to see. Someone draw a random sample of twenty one lm329 devices, and twenty one j201/j203 devices. Then hook them up to a calibrated scope with a sensitivity of 10uV/div, and show me 42 images of the result. The details of building the reference are left out.
although I'm not an engineer I do have a math background. Certainly enough to follow Dmitri's noise calculations. Still, even though it's a good theoretical exercise, it's still a theoretical exercise. I am a bit skeptical by nature, not of theory, but you know yourself how the real world is somewhat messy.I tell you what I'd like to see. Someone draw a random sample of twenty one lm329 devices, and twenty one j201/j203 devices. Then hook them up to a calibrated scope with a sensitivity of 10uV/div, and show me 42 images of the result. The details of building the reference are left out.
Ikoflexer, the numbers come out almost exactly the same as the calculation. You just have to AVERAGE the noise.
I have a QuanTech noise analyzer. It is an old, and tired piece of test equipment. Two other people known to some here have QuanTech's, just like me, they are: Demian, and Kirkwood Rough, a consultant to Linear Systems, who gave a talk and demo at the Burning Amp Festival, last week. We rely on our test equipment to tell us if we are on the right track. It rarely fails.
I have a QuanTech noise analyzer. It is an old, and tired piece of test equipment. Two other people known to some here have QuanTech's, just like me, they are: Demian, and Kirkwood Rough, a consultant to Linear Systems, who gave a talk and demo at the Burning Amp Festival, last week. We rely on our test equipment to tell us if we are on the right track. It rarely fails.
Last edited:
Much of engineering design is making simple calculations based on what one has learned from experience, to make a new product. It is not always new and different. It is not really very mysterious either. Computer models are great if they can be done accurately, but often the parts that we use in audio are not very well characterized, and we just get a high decimal point approximation of what we could have put together and measured faster than if we had to model the circuit on the computer. The Norton equivalent reference is one such example. However, many of us can even do better. We can MEASURE the audio noise of the devices we want to use, quickly and easy with a QuanTech or equivalent noise tester, and know how the devices that we are going to use actually measure. We can equally easy do a curve tracer measurement (I use a TeK 577) to measure the Gm at Id. That is IF we are really suspicious, like you appear to be, Ikoflexer. Usually we just look up the device on a data sheet and presume they are not going to lie to us about Gm. Noise maybe, but Gm? I think not.
And finger crossed, the voltmeter itself has much less noise than the device itself.
The 3581A will do a great job, if the IF bandwidth is set to minimum, at 3Hz. Painful and time consuming, but certainly workable.
- Status
- Not open for further replies.