John Curl's Blowtorch preamplifier part II

[john curl]

Too many cooks spoil the soup. Here, simple, is replaced by complex. It is not 'rocket science' to understand Gm Rl is the key.

 

[iko]

I don't know Scott, sometimes all the little facts add up and we can draw a conclusion.

I've now made sense (at least on paper) of what's going on. To summarize, syn08 showed the derivation for the noise when the Norton equivalent zener made of a jfet and series resistor works with the jfet at Idss, meaning G and S pins are tied together. His conclusion was that the gain of the circuit

Au = gfs R = Vo / Vt

and therefore it depends on the on the jfet pinch-off voltage.

The unanswered question for me was, what happens if we insert a resistor between the gate and source of the jfet. This is the recommendation in any source I found which analyzes a CCS made from a jfet because it improves regulation. The sources I cite here being the Vishay AN103 app note and the book I mentioned previously. However these sources don't have anything to say about the noise if one builds a voltage reference with this current source.

Note: in practice the resistor gets bypassed with a capacitor and this noise becomes very low anyway, but it's instructive to be able to compare these different solutions.

Here we go, the Vgs = Vt case, as in syn08's analysis:
R = resistor below the jfet which establishes the desired voltage
Vo = output voltage (across R)
gfs = forward transconductance of the jfet
Au = voltage gain of the circuit
en = jfet noise

circuit noise = sqrt( (Au en)^2 + (resistor noise)^2)
but let's just worry about Au for now.

syn08 derived

Au = Vo / Vt (0)

therefore the circuit noise depends mainly on en / Vt. So, if no Rs is used, and the jfet is used at Idss, then the jfet chosen should have low noise and large pinch-off voltage.

The case where we insert a resistor Rs in between G and S pins of the jfet.
gfs' = forward transconductance
Rs = resistor between G and S pins
R' = resistor below the jfet that establishes desired output voltage
Vo = output voltage
en = jfet noise
A'u = circuit gain
Id = current through the circuit, due to using Rs

We want to derive a similar formula for A'u as syn08 derived for Au:

When Rs is used,

gfs' = 2 Idss/Vt (1 - Vgs/Vt) (1)

also

Vgs = Vt (1 - sqrt(Id/Idss)) (2)

substitute (2) in (1) and get

gfs' = 2 sqrt(Idss * Id)/Vt (3)

A'u = gfs' R' (4)

We also can express

Idss = Vo/R (5)
Id = Vo/R' (6)

substitute (3), (5), and (6) in (4) and get

A'u = 2 Vo/Vt sqrt(R' / R) (7)

But R' > R, and therefore A'u > Au. Larger gain circuit when using Rs, therefore the noise multiplier is larger.

So, if low intrinsic noise is desired, the overall conclusion here would be to choose a jfet with lowest noise, largest Vt, and largest Idss (to be able to use low value R).

However, the CCS has poor regulation, hence the reference will too. Do we get output voltage that's wiggly but low noise?

 

[scott wurcer]

Too many cooks spoil the soup. Here, simple, is replaced by complex. It is not 'rocket science' to understand Gm Rl is the key.

John, a little birdie told me that you should get on the blower to your friends at Linear Systems. I'm not sure they are up on the high end microphone market.

 

[Salas]

However, the CCS has poor regulation, hence the reference will too. Do we get output voltage that's wiggly but low noise?

We simply pick up hum and buzz easier from the layout. That is why datasheets like the TL431 one notice low impedance as a benefit to advertise. TL431 is best for self noise at 1mA BTW. Tc comp Zeners with some RC filtering towards the business end of the supplied cct are low imp too. Though not availale in small quantity. For manageable led strings up to say 20V, filtered in the same way as above, poor man's good Vref is near if a good for noise generic led and a good Vt JFET with decent noise spec are in stash. Much likely so.

 

[iko]

In practice it may still make more sense to bypass R with an appropriate capacitor and built the CCS for best regulation. Then you get both good regulation and low noise.

 

[Salas]

Yes, we do that already in our regs but appropriate caps are relatively high value electrolytics and we progressively lose the benefit up very high in frequency, and RF can creep in. So better start best IMHO.

 

[john curl]

Salas, you make a good point. However, I always put a .005-.01 uf film cap at the input of the follower, in order that the jfet does not try to be an amplifier, or an oscillator, at RF. However, that couples with the synthetic L from the LM regulators, and the wire inductance (about 8ft worth) to create a pretty good 12dB/octave RF filter. Yet, you might be right, in the fact that 1000pf or so across the 100uF cap, would be a good idea.
The very idea of using a Norton equivalent WITHOUT a 47-1000uF cap across the resistor is inappropriate. I hope that this is now obvious.
Scott, why don't you look on the Linear Systems website, and learn what they also make in Jfets?

 

[iko]

I agree, start with the best reference for the job. Definitely even more issues; it's good to remember that a reference does not make a regulator by itself. The rest of the circuit has still lots of influence to the noise, and also that low noise isn't the only feature that characterizes a good supply.

I think I'm ready now to move on from voltage references; guys, sorry about holding you back with the confusions of a beginner!

John, syn08, Pavel, I'm ready for another lesson

 

[Salas]

Salas, you make a good point. However, I always put a .005-.01 uf film cap at the input of the follower, in order that the jfet does not try to be an amplifier, or an oscillator, at RF. However, that couples with the synthetic L from the regulators, and the wire inductance (about 8ft worth) to create a pretty good 12dB/octave RF filter. Yet, you might be right, in the fact that 1000pf or so across the 100uF cap, would be a good idea.
The very idea of using a Norton equivalent WITHOUT a 47-1000uF cap across the resistor is inappropriate. I hope that this is now obvious.
Scott, why don't you look on the Linear Systems website, and learn what they also make in Jfets?

I used to recommend 470-1000uF high quality electrolytics there in the shunt reg threads. And then we saw diy people that were changing all types of caps and values for that bypass because they were pretty sure it was changing the subjective signature in their own words. If they are right then the clean decoupling low and high limits, depending on value and the ESR and the cap's self distortions should have a clear impact, especially in a polluted layout and box. All are to a degree, especially average diy. Best cap is no cap if we want to be absolute. Which is not my realm. Wanna be relative and adequate, and I have learned much about Vrefs from you guys arguing and measuring. Thanks a bunch.

 

[syn08]

So, if low intrinsic noise is desired, the overall conclusion here would be to choose a jfet with lowest noise, largest Vt, and largest Idss (to be able to use low value R).

And here comes what Scott said. Idss and Vt are not really independent. They are related by Idss/Vt^2=constant where the constant is JFET design dependent, named Beta. Each device model has it's own Beta, Idss and Vt values are floating with the manufacturing process variations.

All these in the first approximation, of course.

 

[scott wurcer]

Scott, why don't you look on the Linear Systems website, and learn what they also make in Jfets?

The question is does the noise of a resistor go up or down with its value? The answer is both. The voltage noise goes up as the square root of R and the current noise goes down as the square root of R. This simple fact escapes too many people.

This result is fundamental to thermodynamics.

 

[Salas]

After seeing all you guys contributing it dawned on me that someone important is missing...Erno Borbely is another seminal analogue angel missing from such threads. Where is EB? Does he read us at all? I sit and wait for EB to join DIYA at a point. He would have a lot to teach. But there is fight here many times too...

 

[Joshua_G]

Erno Borbely doesn't participate in forums.

 

[Salas]

I am a disappointed fan then.

 

[anatech]

Hi Salas,
There are a few engineers who do not enjoy open forums. Erno Borbely has his reasons and I can't say I can fault him for it.

The better people behave, the more interest from other professionals we may have in the future. Open debate may not be the way to win their hearts and minds.

-Chris

 

[Salas]

I wonder who invented democracy.

 

[Wavebourn]

I wonder who invented democracy.

Ancient Egyptian priests who taught Greek Philosophers used to say that Gods taught people democracy. The question still remains unanswered, who were that Gods actually?

 

[dimitri]

I'm not sure they are up on the high end microphone market.

Scott, they don't have the right part(s)

 

[Salas]

The question still remains unanswered, who were that Gods actually?

They were DIYers, they made themselves.

Seriously now, when stoic philosophers debated the question if virtue is something you are born with and entitled to or it can be taught and aquired, they concluded the latter, and then the aristocratic system started to be questioned. The rest is...history.

 

[stinius]

who were that Gods actually?

Odin and Tor