If you look at the datasheets in detail, the 2SK3557 and 2SK2394 are essentially the same thing.
The CPH3910 (also ex-Sanyo) would have been a really nice device but for its higher noise.
The NSVJxxxx are IMHO On Semi's attempt to get rid of the Sanyo history.
The CPH5902 is also a good device, but I tend to use J111 as cascode as it is self biasing.
The MMBF stuff are SMD equivalents of the old Siliconix J series and 2N43xx.
Call me biased.
I shall always prefer Japanese FETs (or NXP), as long as I can get them.
Patrick
The CPH3910 (also ex-Sanyo) would have been a really nice device but for its higher noise.
The NSVJxxxx are IMHO On Semi's attempt to get rid of the Sanyo history.
The CPH5902 is also a good device, but I tend to use J111 as cascode as it is self biasing.
The MMBF stuff are SMD equivalents of the old Siliconix J series and 2N43xx.
Call me biased.
I shall always prefer Japanese FETs (or NXP), as long as I can get them.
Patrick
Newark has stock and is showing more to come in 2018. I guess the greedy boys missed some.
Perhaps now might be a good time to design the board-level voltage reference circuit for the next 30 years, optimizing tempco, line rejection, output noise, voltage precision, and dynamic impedance. Using modern components and leading edge design techniques to find one or more vastly preferable final results, in the cost/performance tradeoff space.
Zero tempco LM337 bias string + LT1007 (PSRR 126 dB, en 3.8 nv/rtHz) based 2nd order Sallen Key LPF, with passive post-filter, anyone?
Zero tempco LM337 bias string + LT1007 (PSRR 126 dB, en 3.8 nv/rtHz) based 2nd order Sallen Key LPF, with passive post-filter, anyone?
Both the LM329 and the Jung GLED are 2-terminal devices.
There aren't any alternatives that I know of with same or better performances.
The On Semi TL431 comes close, but for 6.9V would need 2 more resistors and 1 cap in addition.
A super low noise 6.2V Zener would be a good start, preferrably SMD 0805.
My new year wish.
Patrick
There aren't any alternatives that I know of with same or better performances.
The On Semi TL431 comes close, but for 6.9V would need 2 more resistors and 1 cap in addition.
A super low noise 6.2V Zener would be a good start, preferrably SMD 0805.
My new year wish.
Patrick
Sorry, but I live in a sub-5V world
but you can take that as a starting point.
The Nippon Chem. 1000u electrolytics were slightly better
than 1000u/6V tantalums in X-size (somewhat bigger than D)
The last plot are some z-diodes and other references.
The LM329 smells fishy; I got it from Digikey 2 years ago,
still had NS markings and not TI, looks more like a band gap.
The analyzer switches filter BW for each decade, therefore the
50Hz-multiples are smeared above 1000 Hz.
Super-1/f-noise below 50Hz is the amplifier & analyzer.
Attachments
Last edited:
Sorry, but I live in a sub-5V world
Plenty of multi-stage op-amps with high frequency noise peaking similar to your third plot. I never got many comments on disclosing that issue.
The 3rd plot is for voltage references. But it's' a well-known drawback of the LT1028
at 300 KHz.
The blue peak at 400 KHz is probably the purple one @ 900KHz shining through with 1000u instead of 220u.
Last edited:
It seems that the BF862 lives on. No longer from NXP but from
Nexperia, another part of the NXP split.
Waferfab in Hamburg, assembly/test in Asia.
< https://media.digikey.com/pdf/PCNs/NXP/201702021I.pdf >
Gerhard
Last edited:
I think this gives more credibility to my speculation there: http://www.diyaudio.com/forums/soli...e-models-worst-bf862-jfets-9.html#post5282126
The NXP JFETs probably were all manufactured at the Hamburg Fab, which they passed on to Nexperia. I bet that they completely forgot about the JFET line when they drafted the splitoff deal to form Nexperia, because the JFET line was listed amongst the RF products, which officially remained at NXP.
Only months later, last spring, it seems to have dawned on the NXP management that they couldn't make those products any more. Perhaps they thought about contracting manufacturing out to Nexperia, so that they could continue selling the products. After all Nexperia now had the fab, and probably all the documentation. That's what the PCN from Digikey seemed to indicate in March.
But maybe they found that Nexperia wanted too much money for the contract manufacturing, seeing that NXP had cocked up and had no alternative. So the decision apparently was to discontinue the products, which they started doing in June/July, including the BF862. All the JFETs will eventually disappear from NXP's portfolio, they're just selling their remaining stock. All their remaining JFETs are now "not recommended for new designs"; this has changed from when I wrote the above post.
Remember that Broadcom still wants to buy NXP, so they may have more pressing things on their mind than JFETs. Bonuses for example.
Whether Nexperia will pick up on it and start selling the JFETs under their own name is an open question. I wouldn't bet on it. When I spoke to the Nexperia reps on the Nürnberg show last year, they didn't even know what a JFET is.
The NXP JFETs probably were all manufactured at the Hamburg Fab, which they passed on to Nexperia. I bet that they completely forgot about the JFET line when they drafted the splitoff deal to form Nexperia, because the JFET line was listed amongst the RF products, which officially remained at NXP.
Only months later, last spring, it seems to have dawned on the NXP management that they couldn't make those products any more. Perhaps they thought about contracting manufacturing out to Nexperia, so that they could continue selling the products. After all Nexperia now had the fab, and probably all the documentation. That's what the PCN from Digikey seemed to indicate in March.
But maybe they found that Nexperia wanted too much money for the contract manufacturing, seeing that NXP had cocked up and had no alternative. So the decision apparently was to discontinue the products, which they started doing in June/July, including the BF862. All the JFETs will eventually disappear from NXP's portfolio, they're just selling their remaining stock. All their remaining JFETs are now "not recommended for new designs"; this has changed from when I wrote the above post.
Remember that Broadcom still wants to buy NXP, so they may have more pressing things on their mind than JFETs. Bonuses for example.
Whether Nexperia will pick up on it and start selling the JFETs under their own name is an open question. I wouldn't bet on it. When I spoke to the Nexperia reps on the Nürnberg show last year, they didn't even know what a JFET is.
... and I was thinking that I was paralleling lots of stuff 
< Low-noise Broadband Receive Amplifier for Real-Time Magnetic Particle Imaging (PDF Download Available) >
200 BF862, there is also a plot for 400!
They claim 37 pV / rtHz
< Low-noise Broadband Receive Amplifier for Real-Time Magnetic Particle Imaging (PDF Download Available) >
200 BF862, there is also a plot for 400!
They claim 37 pV / rtHz
Last edited:
... and I was thinking that I was paralleling lots of stuff
< Low-noise Broadband Receive Amplifier for Real-Time Magnetic Particle Imaging (PDF Download Available) >
200 BF862, there is also a plot for 400!
They claim 37 pV / rtHz
At some point don't you think DIYaudio should appear in the bibliography. Last I looked Professor Fano still had an office at MIT, now that's longevity.
Well we showed them. I have to look again, did they actually build it? The inductive gate stoppers are going to make a mess of that complex input impedance.
EDIT - I see they end up not worrying about the match. They say they use 12 FET modules, I have not done that many without stoppers (even one oscillates in a Quantech).
Last edited:
I think there is no place for gate stoppers. Assume one BF862 delivers
1nV/rt Hz, then 200 of them deliver 1/sqrt(200), or 70.7 pVrt Hz.
They clearly need some low-loss up transformation. It seems that they
use the inductive signal source and the input capacitance of the FETs
as a series resonant circuit.
I don't have the paper here, and I'm network-challenged with
Laptop & cell phone in the radio shadow.
I also think that gate stoppers cure only the symptoms and not the
cause.
The usual circuit that everybody shows: CS Jfet, Cascode transistor,
pullup, inverting op amp, gain setting divider ---> back into the source
is not as harmless as it looks.
First thing, the input FET is not CS. At least not only. For the feedback,
it is common gate, so the gate needs some gnd reference. At low
frequencies, the signal source will do. At 2 MHz, a few meters of
cable will present funny transformations, even when we are not
interested in 2 MHz. So, some capacitance from gate to gnd is
badly needed for the feedback to work.
A 0.2 or even 1 Ohm resistor from source to gnd looks harmless.
Everybody thinks "That's as common source as it gets, cannot be real feedback!"
But by action of the loop through the op amp and the gain divider
it is nearly 100% feedback. The source follows the gate completely.
At the same time, the drain looks into the cascode transistor and sees
effectively a short circuit.
So, for the input FET, the world looks as if it was operating as a source
follower into a current source load.
It gets worse. The feedback voltage lags the input voltage because of the
cascode / op amp delay. So the source current comes later than the input
voltage and that looks like a capacitor in parallel to the current souce.
To sum it up, the JFET feels like it is in a capacitively loaded source
follower, working into a current source and having a short on the drain.
That is the recipe behind most oscillators. The standard text book
explanation must be applicable: If we measure into the gate, we see
a negative resistor in series to a capacitor.
SamG in Linear Audio talks here about a negative capacitance that he
tries to compensate with a positive capacitance; that cannot be correct
imho. Negative capacitance is inductance.
But it helps to get the RF reference for the feedback right.
BTW he found an unexplained slight deviation of the gain from the
computed value. That comes from the non-zero output impedance
of the op amp / inverting stage. This can easily be proved by inserting
a voltage-controlled voltage source to buffer the feedback divider in
Spice.
The negative resistance in series to the input must generate the
appropriate thermal noise, so we are punished twice, not only for
the gate stopper.
I wonder if the average simulator gets that right when it constructs
the G matrix. It could say -50 Ohms + 70 for the stopper, that makes 20
The only real solution is to massage the loop gain that there is no
excessive delay @ positive loop gain. I have played with the various
loop gain probes that are on offer for LTspice, without much success.
Starting with low loop gain and wide bandwidth seems to help, even
to the point that the gain stops depending only on the feedback divider.
That softens the virtual CCS at the FET's source, and low impedance
there is good for stability.
But then, Cgs is no longer mostly removed.
You cannot have your Kate and Edith, too.
I have no idea if the following is stable. Completely untested.
The Carl Battjes style compound transistor drops gain at -6dB/octave
instead of -12 dB like a Darlington. That is more feedback friendly,
although beta is only twice that of one transistor. It rose the upper
-3dB corner from 1 to 4 MHz.
The integrator has been tuned for maximum flatness at the lower
passband edge.
C3 is ugly, the cascode needs to compensate for gain loss below
10 Hz. That goes away for a _REALLY_ large capacitor.
source_ac is a test point only, like ac coupling on the scope.
The voltage divider on the bias line speeds up powerUp.
The 30 FETS have enough bias current to drop a volt across
the 44Meg. In simulation and real live. :-(
And I found out with the network analyzer that ferrite beads
are just inductors at low frequencies and do not help. At frequencies
where they do work, they also produce thermal noise.
1nV/rt Hz, then 200 of them deliver 1/sqrt(200), or 70.7 pVrt Hz.
They clearly need some low-loss up transformation. It seems that they
use the inductive signal source and the input capacitance of the FETs
as a series resonant circuit.
I don't have the paper here, and I'm network-challenged with
Laptop & cell phone in the radio shadow.
I also think that gate stoppers cure only the symptoms and not the
cause.
The usual circuit that everybody shows: CS Jfet, Cascode transistor,
pullup, inverting op amp, gain setting divider ---> back into the source
is not as harmless as it looks.
First thing, the input FET is not CS. At least not only. For the feedback,
it is common gate, so the gate needs some gnd reference. At low
frequencies, the signal source will do. At 2 MHz, a few meters of
cable will present funny transformations, even when we are not
interested in 2 MHz. So, some capacitance from gate to gnd is
badly needed for the feedback to work.
A 0.2 or even 1 Ohm resistor from source to gnd looks harmless.
Everybody thinks "That's as common source as it gets, cannot be real feedback!"
But by action of the loop through the op amp and the gain divider
it is nearly 100% feedback. The source follows the gate completely.
At the same time, the drain looks into the cascode transistor and sees
effectively a short circuit.
So, for the input FET, the world looks as if it was operating as a source
follower into a current source load.
It gets worse. The feedback voltage lags the input voltage because of the
cascode / op amp delay. So the source current comes later than the input
voltage and that looks like a capacitor in parallel to the current souce.
To sum it up, the JFET feels like it is in a capacitively loaded source
follower, working into a current source and having a short on the drain.
That is the recipe behind most oscillators. The standard text book
explanation must be applicable: If we measure into the gate, we see
a negative resistor in series to a capacitor.
SamG in Linear Audio talks here about a negative capacitance that he
tries to compensate with a positive capacitance; that cannot be correct
imho. Negative capacitance is inductance.
But it helps to get the RF reference for the feedback right.
BTW he found an unexplained slight deviation of the gain from the
computed value. That comes from the non-zero output impedance
of the op amp / inverting stage. This can easily be proved by inserting
a voltage-controlled voltage source to buffer the feedback divider in
Spice.
The negative resistance in series to the input must generate the
appropriate thermal noise, so we are punished twice, not only for
the gate stopper.
I wonder if the average simulator gets that right when it constructs
the G matrix. It could say -50 Ohms + 70 for the stopper, that makes 20
The only real solution is to massage the loop gain that there is no
excessive delay @ positive loop gain. I have played with the various
loop gain probes that are on offer for LTspice, without much success.
Starting with low loop gain and wide bandwidth seems to help, even
to the point that the gain stops depending only on the feedback divider.
That softens the virtual CCS at the FET's source, and low impedance
there is good for stability.
But then, Cgs is no longer mostly removed.
You cannot have your Kate and Edith, too.
I have no idea if the following is stable. Completely untested.
The Carl Battjes style compound transistor drops gain at -6dB/octave
instead of -12 dB like a Darlington. That is more feedback friendly,
although beta is only twice that of one transistor. It rose the upper
-3dB corner from 1 to 4 MHz.
The integrator has been tuned for maximum flatness at the lower
passband edge.
C3 is ugly, the cascode needs to compensate for gain loss below
10 Hz. That goes away for a _REALLY_ large capacitor.
source_ac is a test point only, like ac coupling on the scope.
The voltage divider on the bias line speeds up powerUp.
The 30 FETS have enough bias current to drop a volt across
the 44Meg. In simulation and real live. :-(
And I found out with the network analyzer that ferrite beads
are just inductors at low frequencies and do not help. At frequencies
where they do work, they also produce thermal noise.
Attachments
Last edited:
The inductive gate stoppers would seem to benefit from some parallel and series resistance.
(Was simulating Samuel's measurement preamp...)
Thankfully, I don;t have to do this for a living...