understanding fet noise curves

[uneumann]

I'm looking at noise curves for the 2SK209 FET. (see below) They show lowest noise with ~10K "signal source resistance". I'm trying to understand what exactly that term means - and then how to achieve the lowest noise in a common-source circuit driven by a low-z source.

My best interpretation is that the device would contribute it's lowest noise if I use a 10K gate resistance (gate stopper). However, that 10k resistor adds noise of it's own. So, does the best circuit noise performance boil down to finding the lowest combination of gate stopper and device noise?

Also, does anyone know of a low noise FET (n-ch) in a through hole package like a TO92?
It seems the 2SK117 was common in the past, but is no longer produced.

[nickb]

I'm looking at noise curves for the 2SK209 FET. (see below) They show lowest noise with ~10K "signal source resistance". I'm trying to understand what exactly that term means - and then how to achieve the lowest noise in a common-source circuit driven by a low-z source.

My best interpretation is that the device would contribute it's lowest noise if I use a 10K gate resistance (gate stopper). However, that 10k resistor adds noise of it's own. So, does the best circuit noise performance boil down to finding the lowest combination of gate stopper and device noise?

Also, does anyone know of a low noise FET (n-ch) in a through hole package like a TO92?
It seems the 2SK117 was common in the past, but is no longer produced.

[ATTACH=CONFIG]44490[/ATTACH]

Try LSK170 New Ultra Low Noise JFETs Target Audio, Instrumentation, Medical and Sensors


jfet300_lsk170familyspicemodels.zip

LSK170A_LSK170B_LSK170C_LSK170D_Low_Noise,_Low_Capacitance,_High_Input_Impedance,_N-Channel_JFET.pdf

[Mike Sulzer]

Try LSK170 New Ultra Low Noise JFETs Target Audio, Instrumentation, Medical and Sensors

Beat me to it, Nick. I would just add that for audio, FETs work best for SNR when the source is transformed to a high impedance. If you must use one at low source impedance, you must go for the lowest nano volts per root Hz, and for ready availability, that is the LSK170. A truly remarkable device.

[Mick Bailey]

I was expecting high noise in an piezo-equipped acoustic guitar FET buffer with 11M Ohm input impedance. No such problem and it's whisper quiet with a 2N5457. Maybe the noise is below the noise floor of my Laney acoustic amp. Noise is relative - what looks noisy on paper depends on the specific application.

[uneumann]

Beat me to it, Nick. I would just add that for audio, FETs work best for SNR when the source is transformed to a high impedance. If you must use one at low source impedance, you must go for the lowest nano volts per root Hz, and for ready availability, that is the LSK170. A truly remarkable device.

Thanks Mike + Nick. It seems supply does follow demand. I'll look around for the LSK170.

On the first issue though. I don't really understand why a high source impedance should lead to lower device noise. I guess I just don't understand the principles and mechanisms involved. Anyone venture a layman's explanation or intuition about why source impedance matters?

Connecting back to guitar amps - would you advise a 10K gate resistor or something lower for a LSK170 amp input stage?
I've been using 470 - 1K ohm gate resistors on J113s and that seems to work OK - but I'm trying for even lower circuit noise.

[darkfenriz]

I'm looking at noise curves for the 2SK209 FET. (see below) They show lowest noise with ~10K "signal source resistance". I'm trying to understand what exactly that term means - and then how to achieve the lowest noise in a common-source circuit driven by a low-z source.

My best interpretation is that the device would contribute it's lowest noise if I use a 10K gate resistance (gate stopper). However, that 10k resistor adds noise of it's own. So, does the best circuit noise performance boil down to finding the lowest combination of gate stopper and device noise?

Also, does anyone know of a low noise FET (n-ch) in a through hole package like a TO92?
It seems the 2SK117 was common in the past, but is no longer produced.

[ATTACH=CONFIG]44490[/ATTACH]

Of the three curves the 10Hz one is clearly the top, so in that area it is the 1/f noise or flicker that dominates the noise figure. It is sometimes useful to think of 1/f noise as a 'drift'.

At low source resistance the JFET flicker will add a lot of noise relative to source resistance noise (thermal and uniform), thus you have higher noise figure.
With source resistance going up - the JFET flicker gets relatively smaller noise source compared to source's inherent resistance white noise - making noise figure closer to 0dB.

So it is not noise going down with source resistance, it is the noise figure go lower - remember the noise figure definition is a ratio.
You still have the lowest total noise with low source resistance.

[Mike Sulzer]

On the first issue though. I don't really understand why a high source impedance should lead to lower device noise.

Actually it does not. If you have a signal source with a certain impedance, you can, at least in principle, put a transformer between it and the amplifier and see which transformation ratio gives the best signal to noise ratio. Two things happen when you increase the transformation ratio: 1. the voltage into the amplifier increases, 2. the impedance looking back towards the signal source increases. If the total effective noise at the input of the amplifier due to the amplifier increases when you raise the source impedance, you might find that you should not make the transformation
ratio too high. (The resistive part of the source impedance also contributes noise, so this can get complicated.) In general an amplifying device has two kinds of noise: noise voltage and noise current. As the source impedance increases, the noise current makes a bigger contribution to the total effective noise (E = IR), while the effect of the noise voltage remains the same. At audio, both tubes and FETs are dominated by noise voltage at typical impedances, while for bipolar junction transistors it is usually current.

(OK, it is really a bit more complicated than that, but it has been along time since I have thought about these things!)

[uneumann]

Thanks to all of you ... this is new ground for me. I'm not sure I get it on first read, but you've given me some leads to follow.

I'm going to breadboard some circuits for testing and hopefully I will see some cause/effect relationships that make sense of all of this.

[Gnobuddy]

I don't really understand why a high source impedance should lead to lower device noise.

We need to be careful exactly what we're talking about - noise from the source resistance itself, noise from the amplifying device, total noise, or noise figure.

Noise from the source resistance itself will increase with higher source resistance. The familiar v = sqrt(4kBTR) formula.

Noise from the amplifying device includes both a noise current component, and a noise voltage component. If you imagine the source feeding the device has zero source resistance, then that will actually short out the noise current from the amplifying device. You'll be left with the noise voltage from the source, and the noise voltage from the amplifying device (but no contribution from the noise current of the amplifying device). In such a situation, it would be good to have an amplifying device that has very little noise voltage of its own, but you can tolerate a bit of noise current, as it doesn't affect the total noise at the output much. BJTs usually do better in this sort of role.

On the other hand, if the source has high resistance, then the noise current from your amplifying device, flowing through that high source resistance, generates lots of noise voltage (by Ohm's law). This noisy voltage caused by the flow of the device noise current, is likely to be much bigger than the actual noise voltage generated by the amplifying device itself. So when the source resistance is high, you want an amplifying device that has very little noise current of its own, but a bit of noise voltage may be fine. JFETs usually do well in this zone.

Adding an amplifying device always adds some noise to the signal. Noise figure is, basically, how much more noise you have than before, in decibels. It's a ratio, and it tells you how much worse the amplifying device made things. This is very useful information to the circuit designer - arguably, more important than simply knowing how much noise voltage (or current) you have. If you have a good noise figure, your amplifying device has added a negligible amount of noise to what was already in the signal - which means, you can't really improve on what you already have.

To elaborate, a noise figure of 0 dB would mean you had a magical, perfectly noiseless amplifying device (which, unfortunately, doesn't exist). A noise figure of 3 dB means your device doubled the total noise power at the input (twice the power, +3 dB). A noise figure of 10 dB would be problematic - it means most of the noise at the output comes from your amplifying device, and not from the source itself. (However, it is possible that, if the source was very quiet indeed, a noise figure of 10 dB will still be quiet enough to work just fine.)

If you look at BJT noise curves, you will typically see that they have their lowest noise figures at fairly low source impedances - that's where they do the least amount of damage to the incoming signal, by adding relatively little noise to it.

In the same way, if you look at JFET noise curves, you will typically see that they have their lowest noise figures at fairly high source impedances. That means they do the least amount of damage with a high impedance source. (That doesn't mean a high impedance source has less noise voltage - it means the fraction of the total noise added by the JFET is lower if the source impedance is high. In other words, the source is much noisier than the JFET, which is what you want. The JFET isn't worsening things much.)

-Gnobuddy

[nickb]

Try LSK170 New Ultra Low Noise JFETs Target Audio, Instrumentation, Medical and Sensors


[ATTACH]44491[/ATTACH]

[ATTACH]44492[/ATTACH]

As an addendum to this post, I have my doubts about the accuracy of the SPICE model. I suspect the drain to gate capacitance is wrong as I'm seeing a surprisingly early HF rolloff. For a common source amplifier with a gain of 24.7dB it was showing -3dB down at about 1.2KHz with a 100K source resistance. That an equivalent input capacitance of 1.3nF. The data sheet specifies Cdg and Cgs as 5pF and 20pf respectively. That comes to about 110pf with the miller effect. Seems that something is arwy here.

My plan was to give a plot of total noise vs freq for various source resistances. I'll not do that now as it won't be helpful.

[Mike Sulzer]

More generally, noise figure fails as a convenient measure. This is mostly a result of the the need to refer to a particular reference noise temperature. Point a low loss antenna around the sky, and there is a huge variation in the noise temperature as a function of where you point and frequency; that is, the sky temperature varies, and the ground temperature can matter. It is simpler to work with a characteristic of the device alone, its noise temperature, still a function of frequency and source impedance.

Even for audio, it can be important to work with noise figure as a function of frequency. An example relevant here is the guitar pickup. It might be 8K resistive at low frequencies, but at the resonance of the pickup coil and the coil plus cable capacitance, it could be 100K, leading to an unexpectedly large contribution from noise current. This of course favors tubes and FETs over BJTs even more than might be expected.

[Mike Sulzer]

As an addendum to this post, I have my doubts about the accuracy of the SPICE model. I suspect the drain to gate capacitance is wrong as I'm seeing a surprisingly early HF rolloff. For a common source amplifier with a gain of 24.7dB it was showing -3dB down at about 1.2KHz with a 100K source resistance. That an equivalent input capacitance of 1.3nF. The data sheet specifies Cdg and Cgs as 5pF and 20pf respectively. That comes to about 110pf with the miller effect. Seems that something is arwy here.

My plan was to give a plot of total noise vs freq for various source resistances. I'll not do that now as it won't be helpful.

In [180]: 10.**2.47
Out[180]: 295.1209226666387


In [181]: 295*5
Out[181]: 1475


In [182]: 1./1.e5/(1475.e-12 + 20.e-12)/2./pi
Out[182]: 1064.5815591431128

Getting nano volt noise performance, good BW and good dynamic range is not so easy, but there are circuits around.

[Gnobuddy]

...the guitar pickup...might be 8K resistive at low frequencies, but at the resonance of the pickup coil and the coil plus cable capacitance, it could be 100K...

And that's if the volume pot in the guitar is turned all the way up. It can be worse, a lot of humbucker-equipped guitars have 500k volume pots, which provide 125k of source resistance (!) if set to half-resistance with an imaginary zero-impedance pickup, and even more than that with a real guitar pickup.

Interestingly, though, I have never noticed any of my guitar amps hissing more loudly when I turn down the guitar volume (though many times I've noticed the dreaded "tone suck", treble loss caused by the same increase in source resistance).

Perhaps input noise is dominated by something else, perhaps I don't hear it because I tend to play using relatively low gain, and at modest volume.

All the back-of-the-envelope calculations I've done invariably point to JFETs as being the quietest input devices at typical guitar source impedances. Not all JFETs, though, as we also want low flicker noise down to 80 Hz.

-Gnobuddy

[nickb]

In [180]: 10.**2.47
Out[180]: 295.1209226666387


In [181]: 295*5
Out[181]: 1475


In [182]: 1./1.e5/(1475.e-12 + 20.e-12)/2./pi
Out[182]: 1064.5815591431128

Getting nano volt noise performance, good BW and good dynamic range is not so easy, but there are circuits around.

The voltage gain Av was 24.7dB
=> Av= 10**(1.235) = 17.1 not 295
So Ceff= (Gds*(Av+1) + Cgs) = (5 * 18.1 +20) =110pF
and F-3db = 14468.6

[Mick Bailey]

Interestingly, though, I have never noticed any of my guitar amps hissing more loudly when I turn down the guitar volume (though many times I've noticed the dreaded "tone suck", treble loss caused by the same increase in source resistance).
-Gnobuddy

I do have one guitar that gets really noisy as the volume control approaches halfway - more hum and hiss that isn't there on full. I thought the hum was the ground-referencing, but I never could rationalize the hiss that crept in.