I read and reread bcarso's comments on the theoretical snr of the RN Marsh discrete op-amp and after thinking about it worked the same analysis, and conclude he got it correct - this is at least the way I analyze my own designs when not being too lazy...I understand the argument, but the actual feedback components are not improved by feedback so adding the feedback margin to the calculated open loop noise performance doesn't look right to me either. Scaling the resistor values in a downward direction which looks possible and reasonable here would help.
IMO the SNR is probably more than good enough, and I don't get me wrong I think this is a pretty cool design.
Well since Wave BLEW my door prize, it will have to be simply the pleasure of knowing how things work that motivates the understanding of how the discrepancy evolved.
I could tell you what I think happened, but I'd like to see the other answers. Simulations are interesting, but just reasoning about how the circuit works is even better. And again, I agree that it is a perfectly nice design and quite acceptably low noise.
Last edited:
Agree on both counts. Can we continue on that path?
Good applications examples to show the other possible limitations of a discrete op amp as well. PSRR, noise, and offset in higher gain applications are interesting challenges with such simple circuits.
Dave
Hi Scott,
The 2SK2145 looks very interesting. Noise performance still needs to be tested. Q1/Q2 and Q8/Q3 could be a matched NXP pair. The PMP4201 and PMP5201 look reasonable. For Q15/Q16 the BC847BVN pair might work. All have more gain and lower Ft than the 4401/4403. Still worth a look as they are duals and should work reasonably well for thermal balance. Reasonably matched pairs for the mirrors as well in a single package.
For SMT diodes or other small transistors see the NXP leaflet - http://www.nxp.com/documents/leaflet/75016734.pdf
I have not simmed it so I am shooting from the hip on the NXP duals.
How critical is matching for the output buffer?
Dave
Since no one has jumped in with suggestions I have a couple for the gain half–
The 2SK2145 looks very interesting. Noise performance still needs to be tested. Q1/Q2 and Q8/Q3 could be a matched NXP pair. The PMP4201 and PMP5201 look reasonable. For Q15/Q16 the BC847BVN pair might work. All have more gain and lower Ft than the 4401/4403. Still worth a look as they are duals and should work reasonably well for thermal balance. Reasonably matched pairs for the mirrors as well in a single package.
For SMT diodes or other small transistors see the NXP leaflet - http://www.nxp.com/documents/leaflet/75016734.pdf
I have not simmed it so I am shooting from the hip on the NXP duals.
How critical is matching for the output buffer?
Dave
Hey no fair Wave!
It's okey Brad, here is more: Why "Chaldean" Numerology is NOT Chaldean - The Mystery Explained
Error message -
I went today and dug up some equipment and got at the output (closed loop) more like -100db ref 1v. One channel a little less, one a little more. BW limited to 100hz to 10KHz is the best i could do towards an A weighted figure. No load and input shorted. -RNM
[should do better with Rf/Rg lowered and a couple other simple things. I'll work on it.]
I went today and dug up some equipment and got at the output (closed loop) more like -100db ref 1v. One channel a little less, one a little more. BW limited to 100hz to 10KHz is the best i could do towards an A weighted figure. No load and input shorted. -RNM
[should do better with Rf/Rg lowered and a couple other simple things. I'll work on it.]
Last edited:
My next preamp design (if I ever get time for audio again
PMA,
I would be interested in hearing your (or anyone else's) take on why the CFA topology would be superior for audio? i'm not opposed just curious as to where the trades offs rate in "audio" performance.
Many thanks
-Antonio
If we could get someone to do some pcb of the Marsh design and offer them as a group buy I'd be in for 4..
They are done already... go over to the headphone section and see the one with Marsh in the subject line. I didnt make the pcb but they are available.
The 2145 is two 2SK117 chips, probably from adjacent locations on a wafer. The 117 is roughly 1/2 of a 2SK170, that is, the same channel length with about half the width, thus same figure-of-merit of gm/Cgs, except for parasitics. The gate leakage for a sample Wayne sent was below 1 pA at 9V G-S. I'll get a noise measurement of it soon, but according to Dimitri they manage about 1nV/sqrtHz. Not much not to like except the committed channels connection, which limits versatility just a bit, and of course the difficulty of breadboarding with a five pin SM package, and finally the dissipation limit, although I think at least the gate leads have fairly low thermal resistance to the chips.
Until NXP is persuaded to make a BF862 dual (and the latter was developed for AM radios, so the prospect may be unlikely) they are very attractive.
Brad
Thanks for checking that out Richard!
I did some work with a 2SK170/2SJ74 pair in sim with a somewhat different topology, including a complementary DMOS output and more cascoding, necessarily a good deal more parts and higher rails, to explore the possibilities. Of course without SJ74 it is impractical
I may show it anyway. I pushed the feedback divider down to 66.5/20 ohms, and the adjustable values in the sources to a net ~5 ohms. As mentioned, the low pinchoff FETs are close to their zero tempco operating points near Idss to begin with, which helps the offset stability concerns.Now you are getting practical! These other aspects are often viewed as an afterthought, some of which that can be brushed aside by invoking a better power supply etc. Considering them at or near the outset is wise.
Another consideration is to look throughout, especially near the input stage, at how much signal swing changes power dissipation, i.e., thermal distortions, and also how the design responds to more-than-zero source impedance. And --- how does it handle overloads?
Dave, I found your proposal very interesting.
But now, what circuit are you talking about?
Generally, are we talking about several circuits, every one heading for another direction?
Where's the fun in that?? I was referring to that old joke..""of all the things I've lost in my life, I miss my mind the most.""
For me, it would be green olives stuffed with blue cheese...
I've a friend with JC disease..expectation is less than a year left, his demeanor changes several months ago leading to the diagnosis.
Loss of mind is very bad indeed.
The thermal model requires DC stability, DC identical, and a way to fast couple devices which should be tracking one another. You're two examples were incongruent as you they had different dissipation ratio's. As you described, the two example would certainly respond differently, but you ignored the power change.
The subject of thermal modelling is a very good one to have when dealing with discretes, as the IC guys cheat by putting tracking devices nanometers apart, something we have to deal with differently for an assembly.
I believe you meant panthd.
jn
I think a few of you guys need to take the tone down a notch or two...
I agree, please resist promoting one solution over another. Let's stick with real performance issues if possible.
I know how you feel. My one lone bottle of 1971 Petrus, which has at least enjoyed continuous ideal storage, is now valued so highly by some that it is truly difficult to decide whether to drink or merely sell.
Well, to the amplifier, I'm reminded of The Hitchhiker's Guide to the Galaxy, when Deep Thought announces that it's ready to reveal the answer to life, the universe, and everything, and says "You're not going to like it" (recollection of the BBC serialized version of Doug Adams' multimorphing creation).
What I should really do is show the performance as a function of parts, as it quickly exceeds the eight of Richard's headphone buffer just getting started, as it gets steadily "better", at least in eliminating various effects like thermal distortions, and reducing noise. I also should do some sims of settling time after the renewed interest in that topic in another thread. I suspect the compensation needs to be tweaked to instill better single-pole-approximating behavior.
Soon. I actually have some paying work right now, so that's going to take priority for the moment.
and the results are?
What was the s/n and harmonic et al data from the sim? Finding good compliments is a problem and guess it will continue to be.
IF matched jFETs are used without source resistance, the H2, H3 is very, very low and higher harmonics are virtually non-existant. [notice John didnt use source resistors - makes matching more critical but THD and Noise is lowest - purist approach]
But as soon as source resistance is added, you might as well go to bipolars as the distortion isnt lower and H5,7.9th can be made so low as hard to measure with the best equipment available. But bipolars need extra circuitry and size, cost grows and it is harder to do without a dc servo appendage. The end result isnt any better but it is easier to make. -Dick
Thanks for checking that out Richard!
I did some work with a 2SK170/2SJ74 pair in sim with a somewhat different topology, including a complementary DMOS output and more cascoding, necessarily a good deal more parts and higher rails, to explore the possibilities. Of course without SJ74 it is impractical
What was the s/n and harmonic et al data from the sim? Finding good compliments is a problem and guess it will continue to be.
IF matched jFETs are used without source resistance, the H2, H3 is very, very low and higher harmonics are virtually non-existant. [notice John didnt use source resistors - makes matching more critical but THD and Noise is lowest - purist approach]
But as soon as source resistance is added, you might as well go to bipolars as the distortion isnt lower and H5,7.9th can be made so low as hard to measure with the best equipment available. But bipolars need extra circuitry and size, cost grows and it is harder to do without a dc servo appendage. The end result isnt any better but it is easier to make. -Dick
Last edited:
