I use the paralleled equivalent of a 2SK146 and a 2SJ73. This is 4 big fets in noise parallel, BUT they tend to cancel their own bias current. I doubt that fets will ever cause much of a problem, but stock bipolars probably will. The bias current compensated AD797 should be just OK. I can't be sure, but 1uA or so is a pretty low current with a few windings as the source.
Fred, as you know, I often go by the initials 'JC' Once, at Harmon Kardon in the late '70's, some engineers and I were discussing a design problem. Someone said, "Even JC could not solve this problem" I chimed in that in fact, I could not. They reminded me that they were not referring to me. ;-)
Christer, a sound card especially 24 bit, the better ones, hardly ever need a preamp in order to measure noise and when you have very low noise you can connect a pre amp but your problem is when you measure a noise source not related to ground. This is your problem and I think this is really hard to solve.
May I ask your interest in this measurement desires? Normally you are only interested in the total noise out and in an amp the place where it's important (really important) is the input stage. If you want to measure the noise in the power supply voltage you don't need any diff amp because the noise is referred to ground.
Now SY and Fred may continue their debate....
May I ask your interest in this measurement desires? Normally you are only interested in the total noise out and in an amp the place where it's important (really important) is the input stage. If you want to measure the noise in the power supply voltage you don't need any diff amp because the noise is referred to ground.
Now SY and Fred may continue their debate....
Per-Anders,
I appreciate that you are trying to help, but it is almost impossible
to have a discussion when you insist on making false assumptions
and jumping to conclusions and base your argumentation on
those. I do not have a battery-powered scope. I do not
have a battery-powered PC. I do not have a 24-bit
soundcard. I do not have have wall outlets with safety
ground (earth?). I do not intend to measure only noise
with this amplifier (and has clearly said so from the start).
Let's try to get som facts straight.
1) The ground of the DUT and the instruments are not necesarily
the same, but may be floating wrt. to each other even have
have a voltage difference of approx. 115V AC caused by PC
mains filter. (Safety note: This is a special situation where the
voltage is current limited. High voltages are generally very
dangerous. Don't try this at home unless you know what you
are doing.)
2) I can easily imagine cases where I want to measure a small
voltage difference between two points, neither being the DUT
ground, and possibly at the same time also use the other
scope/PC channel to measure some other voltage referenced
to DUT ground. (For those with poor imagination, measuring a
current by measuring the voltage over a resistor not connected
to ground at either end would be one such case).
I see the following two possibilities:
A) An instrumentation-type diff. amp where the inputs may
be AC-coupled to overcome limitations in common-mode voltage
range.
B) A single non-inverting op amp where the positive input is
connected through a capacitor to the op amp pos. input and
the negative input is connected via a capacitor to instrumentation
ground. There should probably also be a resistor between the
two inputs on the op-amp side of the capacitors to give a more
well-defined load.
Variant A should work without problems (or maybe not, if we
cannot connect the DUT ground to the amp).
I am not so sure
about variant B, though. Would it work as well? Are there
further possibilities that I am missing?
I appreciate that you are trying to help, but it is almost impossible
to have a discussion when you insist on making false assumptions
and jumping to conclusions and base your argumentation on
those. I do not have a battery-powered scope. I do not
have a battery-powered PC. I do not have a 24-bit
soundcard. I do not have have wall outlets with safety
ground (earth?). I do not intend to measure only noise
with this amplifier (and has clearly said so from the start).
Let's try to get som facts straight.
1) The ground of the DUT and the instruments are not necesarily
the same, but may be floating wrt. to each other even have
have a voltage difference of approx. 115V AC caused by PC
mains filter. (Safety note: This is a special situation where the
voltage is current limited. High voltages are generally very
dangerous. Don't try this at home unless you know what you
are doing.)
2) I can easily imagine cases where I want to measure a small
voltage difference between two points, neither being the DUT
ground, and possibly at the same time also use the other
scope/PC channel to measure some other voltage referenced
to DUT ground. (For those with poor imagination, measuring a
current by measuring the voltage over a resistor not connected
to ground at either end would be one such case).
I see the following two possibilities:
A) An instrumentation-type diff. amp where the inputs may
be AC-coupled to overcome limitations in common-mode voltage
range.
B) A single non-inverting op amp where the positive input is
connected through a capacitor to the op amp pos. input and
the negative input is connected via a capacitor to instrumentation
ground. There should probably also be a resistor between the
two inputs on the op-amp side of the capacitors to give a more
well-defined load.
Variant A should work without problems (or maybe not, if we
cannot connect the DUT ground to the amp).
I am not so sure
about variant B, though. Would it work as well? Are there
further possibilities that I am missing?
Christer, I see what you (did before also) but thing is this common mode thing. Eventually you may notice that what you are trying to do is not an easy task. You have for instance limited common mode rejection and you will also have capacitve coupling. This will not get you clean uV signals. I wonder if the only way is battery powered gear?
I imagine that your are interested in noise in different places but the most important is at the output. This is what counts.
I have a isolation transformer and a stablization transformer at work and I must say when I want to do floating measurements often at 230 VAC potential I have to have the tounge in the right place in the mouth.
I think you want an amp with gain of 1000 at least, with +- 50 V common mode range, 120-140 dB CMRR and 0.1-0.5 uV input noise. I suspect this is a hopeless equation.
I imagine that your are interested in noise in different places but the most important is at the output. This is what counts.
I have a isolation transformer and a stablization transformer at work and I must say when I want to do floating measurements often at 230 VAC potential I have to have the tounge in the right place in the mouth.
I think you want an amp with gain of 1000 at least, with +- 50 V common mode range, 120-140 dB CMRR and 0.1-0.5 uV input noise. I suspect this is a hopeless equation.
I realize now I have been fooling myself that a capacitve
barrier would get around the common-mode problem. It
took half a bottle of Alsace Riesling and some Chévre to
realize I i won't work. (OK, there was some coffee and some
Spice simulations involved too ).
Well, that means it won't be possible to use such an amp
in quite as many ways as I thought. It doesn't matter that
much, I suppose. Since most of the cost will probably go into
a shileded box., connectors, swtiches etc. I thought I might
as well try to make it as versatile as possible, although I
seem to have overestimated the possibilities.
Thanks for you help and suggestions folks.
barrier would get around the common-mode problem. It
took half a bottle of Alsace Riesling and some Chévre to
realize I i won't work. (OK, there was some coffee and some
Spice simulations involved too
). Well, that means it won't be possible to use such an amp
in quite as many ways as I thought. It doesn't matter that
much, I suppose. Since most of the cost will probably go into
a shileded box., connectors, swtiches etc. I thought I might
as well try to make it as versatile as possible, although I
seem to have overestimated the possibilities.
Thanks for you help and suggestions folks.
You bought my ideas finallly? Still, noise is mostly interesting at the output and maybe also the supply voltage. Both places requires only a good preamp (the thing I suggested) in front of a voltmeter or whatever. Bear in mind your ungrounded PC.Capacitors can only block DC. They are useless for common mode problems, which you now have discovered.
When you put input devices in parallel, biasing each at the same current you would otherwise have used for a single input device, the total drain or collector current increases proportionally to the number of devices, the noise current power spectral density increases proportionally to the number of devices and the noise voltage power spectral density decreases inversely proportionally to the number of devices. For example, with 100 input devices, the total collector or drain current increases by a factor of 100 compared to a single device, the noise current increases by 10*10log(100)dB=20dB, and the noise voltage decreases by 20dB.
Since JFET's have a lower product of the input noise voltage and the input noise current at audio frequencies, theoretically, they can win at any source impedance provided that there are no limits to the financially acceptable number of devices and provided that there is no limit on the acceptable supply current.
Since JFET's have a lower product of the input noise voltage and the input noise current at audio frequencies, theoretically, they can win at any source impedance provided that there are no limits to the financially acceptable number of devices and provided that there is no limit on the acceptable supply current.
peranders said:
Eh, what idea were you trying to sell?
Capacitors can only block DC. They are useless for common mode problems, which you now have discovered.
Well, there are cases where capacitive barriers do solve
common-mode problems, but this obviously wasn't one of
those. I can't remember anyone commenting on capacitive
barriers though.
The idea was that your measurement problem isn't easy to solve.Christer said:
I think you have mixed up something. I think that you mean potential shift using capacitors.Christer said:
Think again, you have power supply voltage with let's say 1 V hum (unstablized power) and at the top you have a zener connected to plus and down through a resistor to ground. Then you want to measure across the zener. The results is hum 1 V common mode plus noise. The amp must have at least 120 dB CMRR so you can measure the noise instead of the hum. If you then not have any caps, just DC-connect the amp must additonally have +- common mode range of +-50-100 V.
Do you see the problem?
Phred what is your opinoin in this matter? Do you have a ready solution for Christer?
peranders said:
I see. Well, since I didn't understand that was what you were
trying to tell me, I guess I didin't buy it then.
No, nothing mixed up, but the problems are somewhat
different in those cases. (Hint, one case where it is used can
be found in the Elfa catalogue).
I already had realized that it doesn't work for the case we
were discussing.
As I said before, it is not the case that I have a measurement
problem that must be solved. Rather, since I mistakenly thought
the capacitive barrier would work, I thought I could get this
extra functionality. Now, when I realize I was wrong, it is
probably not worth the trouble to achieve this functionality, since
I guess there is no simple solution. BTW, Fred has also suggested
(by email) that I simply go for a single op amp without capacitive
barrier.
Yes, and by this simple "tool" you can investigate rather much in combination of a True RMS DVM, infact you don't have to true rms if you only want to compare different solutions. With a battery powered preamp and a DVM you don't have a common mode problem either.Christer said:
If you don't have true RMS Linear Tech has a cool TrueRMS chip (uses delta-sigma).
LTC1966
BTW: Has anyone used LTC1966? Opinions?
Marcel, I noted your earlier post. Actually, JFET's can be paralleled with relatively less increase in overall operating current, BECAUSE the transconductance of the JFET's drops as the square root of idle current, rather than directly like a transistor does. This keeps the noise from increasing as fast as you 'starve' the JFET of current. An additional advantage comes from lower internal self heating of the individual JFET's. This improves noise performance, as well.
John, I agree completely. I just tried to show with my earlier post that groups of JFETs can in principle outperform bipolar transistors in all cases at audio frequencies if there are no supply current budget limitations. If there is only a limited supply current available, biasing each JFET close to threshold may indeed be advantageous, as this results in the highest transconductance to drain current ratio.
Do you know whether JFETs at very low drain currents become exponential, similar to the weak inversion region in MOSFETs? I vaguely remember having read somewhere that JFETs and thermionic valves do become exponential and that their transconductance tends to Id/(kT/q) when the drain current per device tends to zero, but I don't remember where I read it.
Do you know whether JFETs at very low drain currents become exponential, similar to the weak inversion region in MOSFETs? I vaguely remember having read somewhere that JFETs and thermionic valves do become exponential and that their transconductance tends to Id/(kT/q) when the drain current per device tends to zero, but I don't remember where I read it.
Marcel, you are correct when you speak of the transconductance of fets at VERY LOW CURRENT going exponential (or linearly changing Gm with current) but not in a practical case mentioned here by me. It is time for some of you to actually look at an app note. For example, look at the 2SK146(147) Toshiba fet spec sheet if you can find it. You will find that at 10ma, a single device can have a noise of 0.7nV/rt-Hz and a Gm of about 55,000 uMho's.
The same device at 1ma, the noise only rises to .9nV/rt-Hz, and the Gm drops to about 15,000. The theoretical noise is: .67/Gm, which is a noise eq resistance of 45 ohms at 1 ma and 12 ohms at 10 ma, so both conditions have added noise or the specified measurement are conservative.
Now, paralleling 10 devices, would give you a transconductance of 150,000 uMhos at 10 ma and a noise resistance of .67/.15, about 5 ohms, or 0.3nV/rt-Hz. Pretty darn good! Now a bipolar transistor would have an equivalent Gm of 400,000 uMhos, which is even better! The actual noise resistance (ideal) would then be .5/.4 or about 1.3 ohms. But the added base resistance and the current noise will eat up any intrinsic voltage noise advantage
The same device at 1ma, the noise only rises to .9nV/rt-Hz, and the Gm drops to about 15,000. The theoretical noise is: .67/Gm, which is a noise eq resistance of 45 ohms at 1 ma and 12 ohms at 10 ma, so both conditions have added noise or the specified measurement are conservative.
Now, paralleling 10 devices, would give you a transconductance of 150,000 uMhos at 10 ma and a noise resistance of .67/.15, about 5 ohms, or 0.3nV/rt-Hz. Pretty darn good! Now a bipolar transistor would have an equivalent Gm of 400,000 uMhos, which is even better! The actual noise resistance (ideal) would then be .5/.4 or about 1.3 ohms. But the added base resistance and the current noise will eat up any intrinsic voltage noise advantage
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.