Hello
I'm certainly not an expert with LTSpice but many years an electronics designer ..
I'm doing a hopefully low noise jfet pre-amp for the 34mm condenser insert I just bought but not sure if I have LTSpice doing the right thing, it's reporting an unusually low-noise level on the output of the pre-amp. I temporarily have the input 1G resistor set to noiseless to get a noise floor level on the pre-amp itself, but the noise floor seems way too low for the humble 2N3819 to me, the gain of the pre-amp is only *9 to *10 voltage gain, but that ought to be enough for input to a 24-bit 100dB+ S/N ADC.
The aim isn't for best THD etc, the desire is to get the lowest noise floor with the available parts we have laying around here. Spending £30+ on digikey etc just for a single low-noise op-amp design is not desireable (at present).
Any ideas what I might be doing wrong with LTSPice ? .. it's a new install with default everything.
I'm certainly not an expert with LTSpice but many years an electronics designer ..
I'm doing a hopefully low noise jfet pre-amp for the 34mm condenser insert I just bought but not sure if I have LTSpice doing the right thing, it's reporting an unusually low-noise level on the output of the pre-amp. I temporarily have the input 1G resistor set to noiseless to get a noise floor level on the pre-amp itself, but the noise floor seems way too low for the humble 2N3819 to me, the gain of the pre-amp is only *9 to *10 voltage gain, but that ought to be enough for input to a 24-bit 100dB+ S/N ADC.
The aim isn't for best THD etc, the desire is to get the lowest noise floor with the available parts we have laying around here. Spending £30+ on digikey etc just for a single low-noise op-amp design is not desireable (at present).
Any ideas what I might be doing wrong with LTSPice ? .. it's a new install with default everything.
Why do you think there is anything wrong?
Could you put a cursor on the white noise floor? I would expect something between 10 and 15 nV/√Hz.
Could you put a cursor on the white noise floor? I would expect something between 10 and 15 nV/√Hz.
Yes sure, anything.
You don't say what frequency so I just left clicked (took me a while to look up how to add a cursor, that's how new I am to LTSPice).
I was expecting a higher noise floor because I'm using an old jfet, though I haven't actually made the PCB yet, might do that today to do a real world test. If I had any LS844's etc laying around I'd be most likely using them, but I hasn't.
edit, might help to attach the image !
You don't say what frequency so I just left clicked (took me a while to look up how to add a cursor, that's how new I am to LTSPice).
I was expecting a higher noise floor because I'm using an old jfet, though I haven't actually made the PCB yet, might do that today to do a real world test. If I had any LS844's etc laying around I'd be most likely using them, but I hasn't.
edit, might help to attach the image !
Last edited:
Swapping to LSK170B's (+ biasing adjustments to suit) doubles the noise floor level but also the voltage gain goes from about *10 to about *45 ..
The noise floor with the 2N3819 is of the same order as what I get out of a calculation assuming that its mean-square channel noise is 2/3 of the thermal noise belonging to its transconductance, which is 12.21 nV/√Hz across the 1.5 kohm resistor if the transconductance is 5 mS.
As soon as you turn on the noise of that 1 Gohm and that weird 10 kohm series resistor, they will dominate, the 1 Gohm at low and the 10 kohm at high frequencies.
As soon as you turn on the noise of that 1 Gohm and that weird 10 kohm series resistor, they will dominate, the 1 Gohm at low and the 10 kohm at high frequencies.
OK, that's good to know. Thank you @MarcelvdG !
Yes I know the 1G input resistor will pump the noise floor up quite a bit, just needed to ensure the amp itself was low enough noise to begin with.
Yes I know the 1G input resistor will pump the noise floor up quite a bit, just needed to ensure the amp itself was low enough noise to begin with.
Last edited:
LTSPICE includes a set of equations which attempt to approximate the electrical behavior of JFET semiconductor devices; they are "mathematical models" of real world components.
The equations inside LTSPICE which model a JFET, include lots of user-supplied coefficients. Why? Because there are lots of different JFET part-numbers out in the world, with lots of different performance characteristics. Some JFETs are optimized for low leakage, some JFETs are optimized for VHF operation, some JFETs are optimized to operate at 100uA while other JFETs are optimized to operate at 50mA, etc. To model each of these accurately, LTSPICE requires a different set of user-supplied modeling coefficients for each one, in order to achieve the best possible "fit" between real world behavior, and LTSPICE modeled behavior.
Now an uncomfortable question: how do these sets of model coefficients originate? Who creates them? Who tests them to verify best possible fit? Which ones do you trust, and why? For example, do you trust sets of model coefficients just because they were included in your LTSPICE distribution? Why or why not?
Do you trust sets of model coefficients just because somebody you don't know, uploaded them to diyAudio two years ago?
And a hard question: how much do you trust a set of modeling coefficients that you didn't create yourself? Will you tell yourself "these are probably accurate to within 5% or 10%"? Do you suppose they are accurate to within 20-30%?
You are betting the success of your design on the accuracy of the LTSPICE modeling. What are the odds your bet will pay off?
The equations inside LTSPICE which model a JFET, include lots of user-supplied coefficients. Why? Because there are lots of different JFET part-numbers out in the world, with lots of different performance characteristics. Some JFETs are optimized for low leakage, some JFETs are optimized for VHF operation, some JFETs are optimized to operate at 100uA while other JFETs are optimized to operate at 50mA, etc. To model each of these accurately, LTSPICE requires a different set of user-supplied modeling coefficients for each one, in order to achieve the best possible "fit" between real world behavior, and LTSPICE modeled behavior.
Now an uncomfortable question: how do these sets of model coefficients originate? Who creates them? Who tests them to verify best possible fit? Which ones do you trust, and why? For example, do you trust sets of model coefficients just because they were included in your LTSPICE distribution? Why or why not?
Do you trust sets of model coefficients just because somebody you don't know, uploaded them to diyAudio two years ago?
And a hard question: how much do you trust a set of modeling coefficients that you didn't create yourself? Will you tell yourself "these are probably accurate to within 5% or 10%"? Do you suppose they are accurate to within 20-30%?
You are betting the success of your design on the accuracy of the LTSPICE modeling. What are the odds your bet will pay off?
Well I guess I'm questioning all that @Mark Johnson, not knowing much about (or trusting) LTSpice was the reason for my first post.
All very interesting answers ! thank you
All very interesting answers ! thank you
R4 and R11 + caps are to help reduce the high levels of RF (10's of MHz and up) not far from the mic getting back into the pre-amp from the cables, might well need some extra LC filtering there but real world test will tell us that.
I suggest incorporating the gain into the analysis; namely, divide output noise by the gain (function of frequency) so that the you have noise density referred to the input (RTI). This aids assessment of design changes and comparison of active devices. Transistor data sheets will often show noise density vs. frequency, so RTI analysis lets you assess performance vs. manufacture's specs.
Any SPICE can calculate equivalent input noise. What they usually can't do is split it up in current and voltage noise, they just give the total for the given source impedance.
You can do this by renaming the onoise signal to inoise... I wish it just plotted both automatically.
R4 (noisy series input resistor replaced), added LP roll-off cap C8.
Could parallel up the jfets to help reduce noise but it's diminishing returns. 2.2nV/sqrt(Hz) as it is now.
Could turn the output PNP buffer into a gain stage to somewhat increase but not sure as it's needed until real world test.
Could parallel up the jfets to help reduce noise but it's diminishing returns. 2.2nV/sqrt(Hz) as it is now.
Could turn the output PNP buffer into a gain stage to somewhat increase but not sure as it's needed until real world test.
Last edited:
No not yet tried a BF862, was trying to limit myself to what's currently lying around here without restoring to spending £30 on an order.
Is there a good LTSpice model anywhere around for the BF862 ? LTSpice doesn't have one.
Is there a good LTSpice model anywhere around for the BF862 ? LTSpice doesn't have one.