hey Phil

formatting link

That's slightly better than a BF862 for Ciss * En, and it's a nominally matched pair.

--
John Larkin                  Highland Technology Inc 
www.highlandtechnology.com   jlarkin at highlandtechnology dot com    
 Click to see the full signature
Reply to
John Larkin
Loading thread data ...

matched

I suppose if you don't mind 3x the noise.

>
Reply to
miso

matched

Nice! App notes? GBW?

Reply to
Robert Baer

matched

Hi John,

The figure-of-merit is Ciss * En^2, which is 9e-30 Js (typ) for the LSK489, but slightly better 6.4e-30 Js for the BF862 .

However, Phil tends to explore extremely small-capacitance end of the Z range, so he may find the LSK useful. I work at the diagonally opposite extreme, in the low-Z end. One can couple FETs in parallel to increase the Ciss, but not in series to reduce the Ciss. (Or is there a way? hmm...)

While answering to the post I noticed there's a manufacturer, Moxtek, which I didn't know about. Their Ciss * En^2 seems to be higher than the BF862, I scaled 2.5e-30 Js estimate for the MX-40 ... oh, they specify Cgs rather than Ciss so the numbers are not comparable. Probably hard to get, too.

Regards, Mikko

Reply to
reg

matched

Anyone know somewhere to get single pieces of IF3601 or IF3602?

[...]
--

John Devereux
Reply to
John Devereux

matched

2x, or nearly a tie if you parallel the pair. 862s aren't matched.
--
John Larkin                  Highland Technology Inc 
www.highlandtechnology.com   jlarkin at highlandtechnology dot com    
 Click to see the full signature
Reply to
John Larkin

matched

I was thinking about paralleling a lot of BF862s to really get the voltage noise down. What's your record?

--
John Larkin                  Highland Technology Inc 
www.highlandtechnology.com   jlarkin at highlandtechnology dot com    
 Click to see the full signature
Reply to
John Larkin

matched

Looks like a nice part, thanks. I'll have to get a few to try out. It's probably not 20 cents like a BF862, but the matching would be pretty useful for situations where the transducer capacitance is really small so it's hard to put a snooper on the summing junction.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
 Click to see the full signature
Reply to
Phil Hobbs

At high frequency the noise is dominated by the omega*e_N*(C_iss+C_everything_else) term, so if the transducer capacitance is low, you win.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
 Click to see the full signature
Reply to
Phil Hobbs

matched

Hi, Mikko,

Interesting. What's the application where the FOM is Ciss*e_N**2?

Where I live, it's e_N*C_in, where Cin is C_iss plus whatever else gets hung on the input noode. That's what sets the high frequency noise floor. Since there are a lot more high frequencies than there are low frequencies, that's usually the limit.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
 Click to see the full signature
Reply to
Phil Hobbs

matched

noise

If you don't need to go down to DC, a 1:5 xmfr can replace a lot of JFETs. Eg. parallel 4 for 0.4nV/Hz^1/2 and then at the input get

80pV/Hz^1/2 with reasonable current consumption (but 100x the input capacitance of one, of course, no free lunch).
Reply to
Spehro Pefhany

matched

noise

I've done up to about 12 in a bootstrap, plus another one in the front end.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
 Click to see the full signature
Reply to
Phil Hobbs

Isn't that what remains invariant in impedance transformations? Eg, when paralleling N FETs, Ciss -> N * Ciss and g_m -> N * g_M , I just used the van der Ziel formula En ~ 1/sqrt(g_m) to express this in terms of En rather than g_m. Similarly, putting a 1:K transformer in front, En -> En / K and Ciss -> Ciss * K^2 .

Now it occurs to me that I may have overlooked E_n being in uppercase (I took it as e_n, in volts per sqrt(Hz) ) but does that actually mean the PSD? If this is the case, our FOM's agree. I'm used to write the PSD as S_v or maybe S_en .

Also, John's claim that LSK489 has a lower Ciss * En seemed to indicate that En is in the units of V/sqrt(Hz). The Ciss * e_n is 6 nanovolt-picofarads for the LSK but 8 nV-pF per rtHz for the BF862, so the LSK would indeed be better. But Ciss * S_v -wise the BF862 wins.

Regards, Mikko

Reply to
Mr Stonebeach

e noise

We really didn't do further JFET work after the 2005 ISEC conference; that amplifier contained 4 paralleled BF862's and a 1:4 transformer in front of it; we measured 0.21 nV/rtHz . There must be excess noise in that figure, but we never polished up the amplifier because all our interest went to the cryogenic SiGe.

Regards, Mikko

Reply to
Mr Stonebeach

Hmm, interesting.

In a transformer, the 1-Hz voltage noise goes like 1/N and the capacitance goes like N**2, so the input current noise goes up as N, just as you'd expect.

Paralleling N**2 devices, 1-Hz voltage noise goes like 1/N, capacitance goes as N**2, and noise current goes up as N, which is just the same.

So as you say, the number C_iss*e_n**2 is invariant for a single device type, regardless of what you do with them, and you can go either way to adjusting the tradeoff, providing you can really make a transformer that good in the required frequency range.

OTOH:

If you wire up a FET with a noiseless cascode on its drain, so it has voltage gain but runs at exactly C_iss, its input current noise will be

i_N = 2*pi*f*C_iss *e_N.

So if you use a BF862 and a single section of an LSK489, the noise currents are proportional to C_iss * e_N, and the LSK489 will be

-20*(log(1.5 nV * 4 pF)-log(0.7 nV * 10 pF)) = 1.3 dB quieter.

So in situations where the lowest current noise from a barefoot device is the issue, the LSK489 wins if the additional capacitance on its input is negligible.

When the transducer capacitance is large, you win by using enough FETs or a large enough transformer ratio that the multiplied C_iss is roughly the same as the transducer capacitance. The optimum is pretty broad, though, so practicality usually dictates not going quite that far.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
 Click to see the full signature
Reply to
Phil Hobbs

matched

It's probably best for fairly low frequency stuff, where phemts get really nasty noisy.

A small phemt has a phenomenal Gm/Cg ratio, and RF noise figures in the 0.25 dB area, but their noise corner is really high.

Getting nostalgic, I remember how excited I was as a kid when the frame grid tubes were introduced, with relatively huge Gm/C values. I've always been a real geek. 6DJ8?

--
John Larkin         Highland Technology, Inc 

jlarkin at highlandtechnology dot com 
 Click to see the full signature
Reply to
John Larkin

matched

Yup. 0.35 nV/Hz for an ATF38143, but with a 10-MHz 1/f corner frequency, and a drain impedance of, like, 200 ohms. You really can't use them for anything except common-source amps.

The SKY65050 is a bit noisier, and has a 30-MHz 1/f corner, but has a way higher drain impedance, so you can use it for followers and diff pairs and stuff like that.

Last time I designed-in a tube was in about 1988. It was an 811A, and I used it to put 1 kV pulses on a grid in an air ionization experiment. There's still nothing to touch a tube for high voltage at low capacitance.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
 Click to see the full signature
Reply to
Phil Hobbs

...

I have developed a habit to think in terms of the (matched) noise temperature Tn = e_n * i_n / 2 k_B and noise matching resistance Ropt = e_n / i_n, rather than directly in terms of e_n and i_n. (If e_n and i_n are correlated then the match Zopt is complex, but that is too ..err.. complex for idle everyday thinking.)

That 'mysterious invariant' is just the the matched noise temperature of the JFET. What makes it harder to recognize as Tn is that its expressed in terms of e_n, rather than g_m which appears in van der Ziel formulas.

That is a bit confusing IMO, it makes it look like i_N is somehow just the reactive current driven by the e_N through the C_iss (implying eg. that the two would be perfectly correlated). My understanding is that although e_N and i_N, input-referred, both arise from the Johnson noise of the channel, e_N is driven by one linear combination of the fluctuating elementary currents, and i_N is driven by a different linear combination. Hence they are not (completely) correlated.

If you can neglect e_N and only consider i_N, then your generator must be above the noise match Ropt - or, because JFET input is capacitive, at every spot frequency your generator reactance must be above the u_N / i_N of the JFET.

If you're above the noise match, you could in principle improve your SNR by putting a transformer in front of the FET, and use it to trade u_N for the lower i_N, until the u_N and i_N contributions are equal. A more convenient impedance transform would be to cross- connect several FETs - but you can only connect them in parallel to lower the Ropt, whereas in the i_N-dominated case they should be connected in series - which is not possible (or is it? some extension of the diff pair?). This is what I meant in my first post by you still gaining from the use of the LSK's.

You of course know all the above already, just thinking aloud...

Regards, Mikko

Reply to
Mr Stonebeach

They're pretty well completely correlated by the time the j omega mechanism takes over--well, since that mechanism puts a 90 degree phase shift on at all frequencies, I suppose that that makes them completely uncorrelated, but hardly unrelated since one is just the derivative of the other. The actual current coming out of the gate is really small in almost all cases of interest, so the physical noise current is

My main application is bootstraps, where you need the voltage gain to be

1-epsilon, with 0 < epsilon < 0.01ish. Transformers aren't much good there, because there's no way I know of to hold the gain accurately enough and maintain low enough capacitance to ground--normally it needs to be a fraction of a picofarad at most.

The noise match of a photodiode is usually at some fairly inaccessible impedance level, except in the SWIR and beyond. So most of the time I'm trying to match something that looks pretty much like a current source with some gross capacitance in parallel.

Right now I'm building a fibre-coupled SWIR spectrometer for blood constituents, which is fun. I really am going to have to get a Sherline one of these days--cutting sheet brass with a Dremel wheel is too fiddly.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs 
Principal Consultant 
 Click to see the full signature
Reply to
Phil Hobbs

Google dremel nc mill

--
John Larkin         Highland Technology, Inc 

jlarkin at highlandtechnology dot com 
 Click to see the full signature
Reply to
John Larkin

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.