NOISE FIGURE OF A BIPOLAR TRANSISTOR

In theory it's possible to synthesize a 50 ohm resistor using the Miller ef fect, and end up with a resistor that has less than 4KTR noise. Assuming th e amplifier you use for the Miller effect is ultra-low-noise. This could bu y you 3db in the limit, assuming you used this resistor as a termination. D on't know how practical this really is.

Back to the OP's topic, generally real products use JFETS. You can get them with less than 2 nv/root-hz, I think. Some of those parts have probably go ne obsolete.

Bob

A pad? As in, resistors?

/Imagines Phil dragging his fingernails across a chalkboard ;-)

Tim

C'mon, I have nice fingernails.

Cheers

Phil Hobbs

effect, and end up with a resistor that has less than 4KTR noise. Assuming the amplifier you use for the Miller effect is ultra-low-noise. This could buy you

with less than 2 nv/root-hz, I think. Some of those parts have probably gone obsolete.

Not only is it possible in theory, it's commonly done in front ends. Not Miller, which is capacitive, but a similar idea--you use a quiet inverting amplifier to jiggle the opposite end of a resistor to make it look smaller.

Cheers

Phil Hobbs

BTW:

The effective noise Z of a mag PU is about 4000 ohms, when RIAA equalised. In real RIAA pre amps, JFETs have disadvantages (ie non linearity, low gain & large basic parameter variations) that outweigh any tiny noise advantage.

... Phil

effect, and end up with a resistor that has less than 4KTR noise. Assuming the amplifier you use for the Miller effect is ultra-low-noise. This could buy you 3db in the limit, assuming you used this resistor as a termination. Don't know how practical this really is.

em with less than 2 nv/root-hz, I think. Some of those parts have probably gone obsolete.

Linear Technology's LT1011 [I think it is] has around 1 nV/rtHz input noise.

Again, from memory Supertex makes some FETs with less than 1nV/rtHz.

effect, and end up with a resistor that has less than 4KTR noise. Assuming the amplifier you use for the Miller effect is ultra-low-noise. This could buy you

with less than 2 nv/root-hz, I think. Some of those parts have probably gone obsolete.

The BF862 is around 0.7 to 0.8 nV, with the usual JFET noise corner at 1 kHz or so. The LT1028 and its ilk are around 0.9 nV, but the original LT1028A has a nasty noise peak around 300 kHz that they don't tell you about. The datasheet noise plot conveniently ends well below that, even though it's a 100 MHz op amp. (Marketing again.)

The ADA4898 is a nice well-behaved part with around 0.9 nV noise.

Cheers

Phil Hobbs

I said "power match" but should have said "impedance match". I believe that Hayward stated that the reason is to make the various filters in the system happy.

But your general tone of superiority makes it obvious that Wes Hayward, and his work (neither books, nor, presumably, decades of Tektronix spectrum analyzers) exist.

Silly me. And here I've been laboring under the delusion that I've used those spectrum analyzers, and read some of the books, to boot.

On Feb 13, 2:23 am, Phil Hobbs wrote:> Not only is it possible in theory, it's commonly done in front ends.

Exactly. The trick traces back to the vacuum tube era: W.S.Percival, "An Electrically Cold Resistance", the Wireless Engineer, May 1939, p. 237. It is necessary as a standard practice in room-temperature front ends of SQUID readouts, where the generator resistance indeed has the 50-ohm Johnson noise *but* it is located in LHe.

Nowadays eg. the VCA2611 and AD8331 use the technique. I'm in impression that most low-noise rf/microwave gain blocks utilize that technique to move the noise match and power match to roughly the same impedance.

The technique is effectively the same as the thought experiment of damping an indicator needle, discussed in many thermodynamics textbooks. I think I got first exposed to the idea in the Kittel's book, without realizing how widely it is applicable.

Regards, Mikko

Actually, in Transiton Edge bolometers the Johnson noise is modified because of the correlated temperature fluctuation in the heat bath - because the electrical power present in each upward (downward) voltage swing is drawn from (dumped into) the thermal bath.

Vinante et. al. have used a SQUID and feedback to cool a metal bar weighting a ton into microkelvin range of temperatures (although the above wording cheats a bit).

Regards, Mikko

Indeed. That undocumented peak spoiled our noise-cancelling readout back in 1994. There's another peak at 2.5 MHz. We had to move to the AD797.

Regards, Mikko

It's quite practical. I've built amplifiers with 50 Ohm matched inputs and 300pV/rtHz input-referred noise. I use them in the beam trajectory measurement system of a particle accelerator.

The signal source isn't resistive. It's a capacitive position pick-up. I'd have preferred to use Hi-Z amplifiers directly on the pick-ups, but the radiation would kill them.

A BF862 does 0.8nV/rtHz. There are lower noise JFETS, but none with Yfs/Cg as good as this one. Well, perhaps some RF devices can beat that, but those lose out on 1/f noise.

Jeroen Belleman

T/2

h"

Hi Tim,

I did not sense that Phil objected Haywards conclusion about moving the noise match and impedance/power match to the same Z. I think that's completely feasible and done on a regular basis.

I think Phil's point was that the noise match is more important than the power match. If you cannot squeeze as much power gain out of your stage as you would with the optimal power match, you can always add stages. Gain is cheap. But if you spoil your SNR with non-optimal noise match, there is no way you could get rid of the added noise in the subsequent stages. It'll sit in your signal forever.

The above does not contradict with this fact: if the earlier stages want to see a 50-ohm amplifier input (e.g. for proper freq response of filters, termination of directional couplers etc.), you better arrange such an input or risk screwing the whole circuit. In this case (as you indicated) the 50-ohm input is not needed for the maximum power transfer, but for proper damping of the previous stages.

Regards, Mikko

I didn't see your followup post that Mikko is replying to, but he's right, I didn't intend to contradict you. I'm not a big 50-ohm RF guy myself--I generally buy stuff for that, and certainly until I do the math myself, I'll take the word of somebody who has.

When it comes to front ends, my shtick is salvaging good performance out of inconvenient corners of the design space, and impedance matching is usually a very peripheral issue.

Cheers

Phil Hobbs

The familiar TIA is sort of the logical conclusion of that process, i.e. cranking the gain up as high as you can to make the input resistance close to zero, but that isn't always what you want.

Front ends are fun.

Cheers

Phil Hobbs

Two electrically-connected resistors have an equivalent thermal conductivity, through their Johnson noise, but it's many orders below the thermal conductivity of any electrical conductor. The effect is probably unmeasurable.

er effect, and end up with a resistor that has less than 4KTR noise. Assumi ng the amplifier you use for the Miller effect is ultra-low-noise. This cou ld buy you 3db in the limit, assuming you used this resistor as a terminati on. Don't know how practical this really is.

them with less than 2 nv/root-hz, I think. Some of those parts have probab ly gone obsolete.

THANKS!!! first 1028, not 1011 Brain dead here. second, I got bit by that spike! Not having any instrumentation of value had to 'noodle' out what was wrong and did not understand until now.

Really sad, because Linear parts are usually 'normal'

effect, and end up with a resistor that has less than 4KTR noise. Assuming the amplifier you use for the Miller effect is ultra-low-noise. This could buy you

with less than 2 nv/root-hz, I think. Some of those parts have probably gone obsolete.

The price you pay in bipolars like the LT1028 is current noise. In the non-inverting config, you wind up with a 20 ohm feedback divider. I sometimes have the 1028 drive a follower opamp which in turn drives the feedback divider, so the feedback current doesn't make thermal tails in the LT1028.