Need ultra low noise preamp

[carbontracks]

For a project I'm doing I need to amplify a 14MHz signal (very narrowband, though), and the amp needs to have input-referred noise on the order of a few tens of nanovolts rms.  I've tried using the AD8331 chip, but I was unable to achieve its specified noise characteristics, and even if I did I'm not sure they would be good enough.  I can't seem to find a lower noise integrated amp for my application, so I'm open to the possibility of building a discrete preamplifier.  The problem is, my frequency isn't really high enough to be considered "rf," making it difficult to search for low noise parts (it's not a parameter in most search engines).

So does anyone know of a good integrated or discrete part for my frequency?  I'm willing to spend a decent amount of money, if necessary.

Thanks

[solutions]

No a whole lot of details to go on (gain, circuit topology, LAYOUT (the biggest variable), impedances, passives used, output voltage, dynamic range, isolation requirements ).  i assume you are using the amp in a broadband configuration (gain block) vs a tuned amplifier for 14MHz - if the amp is tuned, in theory you should be able to lower the noise in the passband since you're not integrating it over the bandwidth of the amp....you're not RF, but you can steal the principles used in LNAs, which may be another place to look for ideas.  That said, you might also look at a lower noise technology like JFET or SiGe.  Can't use GaAs below 50MHz due to charge trapping.

Then there's always liquid nitrogen :-)

[nawab]

Hi!

Most of the IF chips would work for you. Try MC1350. If still lower noise figure is required go for a JFET preamplifier. The best discreet would be to pre amplify using a discreet norton noiseless negative feedback preamplifier. There are lot many other solutions also available. Much would depend upon application.

Regards

[carbontracks]

No a whole lot of details to go on (gain, circuit topology, LAYOUT (the biggest variable), impedances, passives used, output voltage, dynamic range, isolation requirements ).  i assume you are using the amp in a broadband configuration (gain block) vs a tuned amplifier for 14MHz - if the amp is tuned, in theory you should be able to lower the noise in the passband since you're not integrating it over the bandwidth of the amp....you're not RF, but you can steal the principles used in LNAs, which may be another place to look for ideas.  That said, you might also look at a lower noise technology like JFET or SiGe.  Can't use GaAs below 50MHz due to charge trapping.

Then there's always liquid nitrogen :-)

I'd like at least 15db of gain.  Circuit topology is flexible, and so is layout (though I don't want anything huge).  I'll likely be working with 50ohms, but I have no problem using passive matching networks (since my application is very narrowband, as I mentioned before).  My input signal is going to range from 100nV to 2uV rms, and will have a bandwidth of maybe 10KHz.

Cutting the noise by filtering away everything but my desired bandwidth is definitely going to be necessary, but even if I'm able to make a good filter with a Q of 1000, a spectral noise density of just 1nV per root hertz would leave me with a significant amount of noise.   That's why I need the amplifier itself to have excellent noise.

Is charge trapping the reason GaAsfets are terrible at low frequencies?  I had looked at them but none had good noise characteristics for the HF band, as you implied.

Hi!

Most of the IF chips would work for you. Try MC1350. If still lower noise figure is required go for a JFET preamplifier. The best discreet would be to pre amplify using a discreet norton noiseless negative feedback preamplifier. There are lot many other solutions also available. Much would depend upon application.

Regards

The MC1350 doesn't specify noise spectral density, but it's noise figure (>6dB) implies it's way too noisy for me.  The norton noiseless amp is pretty cool, but it seems to only address noise induced by feedback resistors by replacing them with transistors.  I'm not planning on using feedback in my preamp stage, so that's not my concern right now.

[solutions]

Can't you get a higher gain antenna/transducer to simplify your life a bit?

Perhaps something in this schematic (I absolutely LOVE the font they use) of a 14MHz receiver might help:

http://www.qsl.net/yo5ofh/projects/14mhz_ssb_trx/14mssb10.gif

[carbontracks]

Can't you get a higher gain antenna/transducer to simplify your life a bit?

Perhaps.  I don't know much about receiver coil design, but in any case I'm quite sure I'll need the low noise.

Perhaps something in this schematic (I absolutely LOVE the font they use) of a 14MHz receiver might help:

http://www.qsl.net/yo5ofh/projects/14mhz_ssb_trx/14mssb10.gif

the preamp is the part in the bottom right, correct?  I don't see what's so special about a resistor-biased transistor amp.

[2N5109]

A simple JFET common source amplifier should give you the lowest noise possible.  A J309 or 2N4416 has less than 5nV rms noise in a 1Hz bandwidth at 14MHz.  Noise current is insignificant (<femtoamps).  If you need high input impedance, build a cascode circuit.  Junction FETs have lower noise than MOS FETs; not sure why.  If you have a problem with using a discrete FET in a circuit let me know and I'll upload a circuit.  But remember at this frequency your system sensitivity is also limited by external (atmospheric and man made) noise so device noise may not make a lot of difference.  

[carbontracks]

A simple JFET common source amplifier should give you the lowest noise possible.  A J309 or 2N4416 has less than 5nV rms noise in a 1Hz bandwidth at 14MHz.  Noise current is insignificant (<femtoamps).  If you need high input impedance, build a cascode circuit.  Junction FETs have lower noise than MOS FETs; not sure why.  If you have a problem with using a discrete FET in a circuit let me know and I'll upload a circuit.

One of the alternative's I'm considering is using a JFET in a common source or even common drain configuration and exploiting the high input impedance by using passive, narrow band impedance transformations to get gain before the input of the amplifier, thus increasing my SNR.  The problem with most JFETs is that they normally don't spec both their input noise current and voltage spectral densities, meaning I don't know the optimal impedance for a noise match.  And finding it empirically would be a huge pain in the ass....

I've also heard that paralleling many devices, or using larger geometry devices (power transistors) can decrease their effective noise.  I don't really care about the resulting drop in input impedance, so such an option might be feasible.

But remember at this frequency your system sensitivity is also limited by external (atmospheric and man made) noise so device noise may not make a lot of difference. 

I definitely suspected this at first (our lab has crappy CFLs above), so I tried testing it in a very good shielded room, but it didn't help.  I'm certain the noise in intrinsic in my circuit.

[2N5109]

There is a copy of a Vishay-Siliconix application note which gives you models and formulas for JFET noise sources at:  

http://www.qsl.net/aa1ll/jfetNoise.pdf

My last post was incorrect about the noise current; in the shot noise region above ~15kHz the noise current is equal to the Johnson noise from a resistor equal to the input impedance of the device at the frequency of interest.  This would apply at 14MHz.  Fairchild Semiconductor has pretty good info on their JFETs since they bought all the National Semiconductor parts and IP.  They give enough information which, taken with the app note above should enable you to model the design, and determine optimum generator impedance (except if it includes an antenna).    

http://www.fairchildsemi.com/ds/MM%2FMMBFJ309.pdf

What did you test in a shielded room--was it a very short antenna?  I suggest you work out the antenna/preamp interface first, then see what the difference is between having the antenna exposed to the outside world vs. being in a screen room.  Paralleling a bunch of FETs will only lower the signal level out of a high impedance antenna or E-field probe by putting more capacitance across it.  I have never heard of doing this when you are sensing the RF voltage developed by a wire.  

Atmospheric noise at 14MHz is 30-40dB above kToB.  Check it out on Wikipedia.com:  

http://en.wikipedia.org/wiki/Atmospheric_noise

[carbontracks]

I uploaded a copy of a Vishay-Siliconix application note which gives you models and formulas for JFET noise sources.  It is at: 

http://www.qsl.net/aa1ll/jfetNoise.pdf

My last post was incorrect about the noise current; in the shot noise region above ~15kHz the noise current is equal to the Johnson noise from a resistor equal to the input impedance of the device at the frequency of interest.  This would apply at 14MHz.  Fairchild Semiconductor has pretty good info on their JFETs since they bought all the National Semiconductor parts and IP.  They give enough information which, taken with the app note above should enable you to model the design, and determine optimum generator impedance (except if it includes an antenna).

This may be pretty useful, thanks.  However, looking back at my receiver coil (it's not really an antenna, since I'm in a near-field regime) I'm wondering if I'll be able to transform my very low coil impedance (around 0.3ohm) up to the necessary noise match for a JFET (maybe 10K).  Doing so would require some very high Q, very accurate components.  Yesterday I came across some MAT02 low noise BJTs, so I've been playing with those to see if I can get some improvements.  At some point I'll assess my ability to use JFETs effectively with my coil, in which case this documentation will really be handy.   

What did you test in a shielded room--was it a very short antenna?  I suggest you work out the antenna/preamp interface first, then see what the difference is between having the antenna exposed to the outside world vs. being in a screen room.  Paralleling a bunch of FETs will only lower the signal level out of a high impedance antenna or E-field probe by putting more capacitance across it.  I have never heard of doing this when you are sensing the RF voltage developed by a wire. 

In our shielded room I just took the AD8331 I was working with, shorted its input to ground, and measured its output noise.  No antenna involved.  I simply wanted to measure the input referred noise of the amp itself, not the external noise picked up by my antenna.  I'm fairly sure my signal, weak as it is, is significantly above the background noise.  If not, then welp, my adviser gave me an impossible project to research!

And I've read that it's not uncommon in practice to parallel many jfets (or other devices) in order to decrease noise voltage while increasing noise current.  It makes it easier to match to low impedance sources.  Some sources even suggested using power JFETs with large geometries in order to get a relatively low noise impedance.

[SBW]

This may be pretty useful, thanks.  However, looking back at my receiver coil (it's not really an antenna, since I'm in a near-field regime) I'm wondering if I'll be able to transform my very low coil impedance (around 0.3ohm) up to the necessary noise match for a JFET (maybe 10K).  Doing so would require some very high Q, very accurate components. 

I think you are on the right track here: choosing the right transistor will affect the SNR by factors of two.  Getting the matching for a low impedance source right can affect the SNR by factors of ten.

L-C circuits have Qs that are limited by the L (at room temperature--my friends who make superconducting resonators complain that they can't make a good C).  At HF, coils can have Qs of hundreds, up to about a thousand.  (If the junk you put in the coil is lossy, then obviously the Q will go down.)  I would consider using a varactor for part of the C to resonate the coil, so that you can tune the resonance onto the correct frequency without having to reach in to the coil.  Varactors are more lossy than capacitors, but if you provide (say) 90-95% of the C with capacitors, and 5-10% with a varactor, you will still be limited by the L.

Ideally, you want the loading on the resonator (losses) from the source (the stuff inside the coil and the coil) to be equal to the loading on the resonator (source impedance) of the amplifier.  But this is probably back to factors-of-two land.

[carbontracks]

Okay after fooling with some low noise BJTs I'm pretty sure JFETs are going to be the way to go.  I've got some J310s on order, and I'm planning to implement a common source cascode like AA1LL suggested.  I'm assuming the device I use for the cascode doesn't have to have as low noise, so I'll probably go with a BJT (due to higher transconductance).

I still think the limiter is going to be the passive components I use in my impedance transformer.  What I have to work with right now is a kit of high Q air core SMD inductors, a large array of SMD RF caps, and some very good tunable capacitors.  I'm thinking that the inductors specifically are going to be the bottleneck, since at my relatively low frequency I need high values (like 4.7-10uH), which result in poor Q (like 10-30).  Does anyone know of a series of better RF inductors for this application?  If possible I'd like to get a small designers kit.

I would consider using a varactor for part of the C to resonate the coil, so that you can tune the resonance onto the correct frequency without having to reach in to the coil.  Varactors are more lossy than capacitors, but if you provide (say) 90-95% of the C with capacitors, and 5-10% with a varactor, you will still be limited by the L.

I don't think varactors will be necessary, since it's not a problem to just twist a knob on a trim cap here and there.

Ideally, you want the loading on the resonator (losses) from the source (the stuff inside the coil and the coil) to be equal to the loading on the resonator (source impedance) of the amplifier.  But this is probably back to factors-of-two land.

I'm certainly not going to get an impedance match to my JFET input.  That would really require superconducting components.  I'm hoping for, at best, an impedance of maybe 10K or so.

[2N5109]

Yes inductors always limit the unloaded Q of resonators in this frequency range.  Forget the SMT parts; you would be better off winding your own coil on a plastic pill bottle.  You probably will need an inductance of several tens of microhenries.  You might consider building a helical resonator to get unloaded Q's approaching 1000 at 14MHz.  You can tune them somewhat with a small capacitor on the open circuited end.  (Maybe put the FET gate there too.)  At this frequency you can build one into a large aluminum can or some other cylindrical structure about 0.1 meter in diameter and length.  More volume, higher Q.  Line it with copper tape and solder the seams to get really low losses.  You could build an autotransformer this way and the impedance at the open circuit end would be EXCEEDINGLY high.  It  would be tapped at the grounded end and your low impedance sensor would attach there or magnetically couple to it there. 

Try googling this: 

itt handbook helical resonator

It should put you on the right track! 



73, DE 2N5109

[SBW]

I second 2N5109's comment: at this frequency, you will want to wind your own coil, rather than use a SMT one.  It is kinda fun, and liberating, once you accept to the idea of making your own components instead of buying them.

The ARRL Handbook for Radio Amateurs has formulas for inductance and Q for a variety of coils, including helical resonators.    Their formula is Qu=50 (diameter/in) sqrt (frequency/MHz), which gives 736 for 2n5109's 10 cm example--which is about what he said.  (The same formula is also in the ITT Reference Data for Radio Engineers, with more information about the assumptions and such.  I also agree with 2n5109's recommendation of this book.)

And I also agree that you could get an impressive impedance transformation out of a properly built (helical) resonator.  However, if your source impedance is really a fraction of an ohm, you're not going to be able to have any cable (or connectors) between it and the preamp without serious SNR degradation.  Which is why you might want to resonate the coil your signal comes from, to increase the impedance to something that you could put through a cable if you need to.  To get all the way up to the ideal impedance for a FET, you might need both the impedance transformation at the pickup coil and a (helical resonator) transformer on the preamp input.

[carbontracks]

I second 2N5109's comment: at this frequency, you will want to wind your own coil, rather than use a SMT one.  It is kinda fun, and liberating, once you accept to the idea of making your own components instead of buying them.

The ARRL Handbook for Radio Amateurs has formulas for inductance and Q for a variety of coils, including helical resonators.    Their formula is Qu=50 (diameter/in) sqrt (frequency/MHz), which gives 736 for 2n5109's 10 cm example--which is about what he said.  (The same formula is also in the ITT Reference Data for Radio Engineers, with more information about the assumptions and such.  I also agree with 2n5109's recommendation of this book.)

A helical resonator would be overkill, I think.  Plus, I still have to deal with the parasitic input capacitance of my JFET, which is a lossy capacitor.  That alone will prohibit any incredible Q factors.  A hand-wound inductor is definitely feasible, and will probably get the job done.

However, if your source impedance is really a fraction of an ohm, you're not going to be able to have any cable (or connectors) between it and the preamp without serious SNR degradation.  Which is why you might want to resonate the coil your signal comes from, to increase the impedance to something that you could put through a cable if you need to.  To get all the way up to the ideal impedance for a FET, you might need both the impedance transformation at the pickup coil and a (helical resonator) transformer on the preamp input.

Yeah, I'm already using a local transformation network to step the coil up to 50 ohms, which then terminates to coax.  I've already built a simple proof of concept circuit to use it with.  It has a simple series LC step up network into a two-stage common source cascoded Jfet amp.  I'm getting around 50db of gain out of it (about half from the amp, half from the LC), which is plenty.  Problem is it's oscillating somewhat at a very high frequency, making noise measurement difficult.  Probably just a layout issue.

[2N5109]

I've had good agreement between measured and simulated noise from LF to VHF using LTSPICE even though it's (gasp) free.  The models for JFETS have been developed over the years and seem pretty good.  This way you could also run a Transient Simulation of the same circuit including layout parasitics and see if it oscillates and possibly how to stop it if it does. 

Download at: 
http://ltspice.linear.com/software/LTspiceIV.exe

73, 2N5109

[carbontracks]

I've been using LTspice for years, but never its noise analysis function.  I did try to get it working, but couldn't make heads or tails of its output.  Seems to be fairly poorly documented.  As for parasitic oscillation, there's no way it will be able to help me with that, since I can't accurately quantize my parasitics.