Moving Coil Pickups for the Technically Curious

[Steve Conner]

Someone said that the noise output of the differential amp is sqrt(2) times more than a single device. If this is true, then it proves that the choice of common base or common emitter doesn't affect the noise performance.

Why? A long-tailed pair of transistors can be viewed as an emitter follower driving a common-base amplifier. And the sqrt(2) increase means that both devices contributed an equal amount of noise to the total, even though one is an emitter follower and the other a common-base stage.

You can argue that that only holds for differential output between the collectors and not for either collector taken individually. But then you'd have to explain how one collector could show a different noise level to the other, when the circuit is completely symmetrical.

To my mind, the topology choice may affect the current noise: there would be beta+1 times more of it coming out of the emitter than the base. But it doesn't affect the voltage noise, because that is a voltage source in series with Vbe, and it doesn't matter which end of Vbe you're looking at.

If the source impedance were low enough to swallow the extra noise current, then there would be no difference in noise performance.

Joe Gwinn: Got any more information on your dual mixer thingy?

[Mike Sulzer]

If this is true, then it proves that the choice of common base or common emitter doesn't affect the noise performance.

I like that proof. I have used that method of representing a long tail pair before, but never for analyzing the noise.

[Steve Conner]

Well unfortunately in the spirit of this thread, I have to admit that I made a mistake. It proves that the choice of common-collector or common-base doesn't affect the noise performance, but I said: common-emitter or common base.

Then again, if you look at the LTP another way, it's a common-emitter stage too...

[Mike Sulzer]

Well unfortunately in the spirit of this thread, I have to admit that I made a mistake. It proves that the choice of common-collector or common-base doesn't affect the noise performance, but I said: common-emitter or common base.

Then again, if you look at the LTP another way, it's a common-emitter stage too...

That is kind of what I was thinking: the SNR at the output of the emitter follower is almost exactly present in its collector current, and thus in the voltage across the collector load.

[Joe Gwinn]

Oh, I was not questioning the rule; I have used it for almost four decades. When I read this sentence: "The summary rule that bipolar for low impedance and JFET for high impedance assumes common-emitter or common-source respectively.", I assumed that you meant that the rule is only intended for common-emitter or common-source topologies. I am still having trouble reading it any other way.

The key is the word "better" in the original statement of the rule of thumb. You are assuming that this meant for noise only, which is not the case. The issue was general suitability for use with sources in the stated impedance range, and there is a lot more to it than noise. In fact, very few circuits are noise dominated.

In any case, the rule also works for a BJT in common-base and a FET in common-gate.

This part isn't clear, being something to be demonstrated. And, noise is the least of it for the rule.

When you ground the base and connect a low source impedance to the emitter, the noise current source is shunted and has little if any effect. Thus, one is left with the noise voltage source. This is less than the noise voltage of the FET in general, as the rule implies. And so in situations where the rule applies (maybe not so much anymore as FETs get better) the transistor is better for this low impedance application as well as for common-emitter vs. common-source.

The circuit does matter - one configuration shorts out a noise generator.

The sentence is this:
"The question is if use of common-base and common-gate respectively changes this rule of thumb."

Given the first sentence, I must interpret the second one as talking about noise.

As discussed above, not so. The rule of thumb labored under no such limitation.

Yes, the device and the rest of the circuit determine how much noise results. The key thing for a BJT is that there is current flowing into the base of the transistor due to its physics; it has noise fluctuations as any such current must. (The noise behavior of this current is represented by the noise current source in the common model I have mentioned before.) When one increases the source impedance in the common-emitter mode, a bigger noise voltage develops across the base and emitter. That is, when the noise source that is part of the device is shunted by a larger external resistor, a larger noise voltage develops between the base and the emitter.

This fairly describes one such model.

All this is correctly predicted by the commonly used model which has a voltage and current noise sources. I think what you write now agrees with this model, while what you wrote above contradicted it. That is, the model predicts that with low source impedance, common-base and common-emitter have the same SNR.

Well, not quite. Misinterpreting what was said does lead to confusion.

Even if one is able to show that the noise voltage and/or current is the same, is the signal the same? Not usually. There is a reason to attempt to match impedances.

Here are the links to a couple of files professor Marshall Leach at Georgia Tech uses to illustrate some of these concepts:

http://users.ece.gatech.edu/mleach/ece4391/noise_a.pdf
http://users.ece.gatech.edu/mleach/ece4391/noise_b.pdf

The second one is a two page file computing the noise in a BJT and a FET. See, it is not so hard when you make a good model of what is happening and then apply it.

Links initially didn't work for me; now fixed to make them clickable.

These are the simplest standard models, most often used with common-emitter amps. Although not explicitly stated, noise_a.pdf analyzes only a common-emitter amplifier. But there are many noise models, each with advantages and disadvantages, and 90% of the battle is knowing which to use for what. Frequency range and circuit configuration both matter a lot. And the wild card is flicker noise (as mentioned in noise_b.pdf), which arises from somewhat different physical causes than the ordinary noise sources.


I poked around in some of my old textbooks:

"Low-Noise Electronic Design", Motchenbacher and Fitchen, Wiley 1973, 358 pages. This is the practical book, and is a classic.

Uses the Hybrid-Pi model of the transistor, with a mixture of voltage and current noise sources. States (on page 82) that the noise figure does not depend on the configuration (CE, CB, or CC), which agrees with the observation that the noise generators of the transistor are what they are. Noise figure relates only to the N part of SNR. Spends many pages on the effect of operating and circuit conditions of the transistor on noise. Discusses the effect of paralleling transistors, and of the differential configuration.

Discusses flicker noise, concludes that the lower the collector current the lower the noise, which conflicts with the high current needed to match low impedances.

I'll have to reread the relevant sections.


"Noise in Solid State Devices and Circuits", Aldert van der Ziel, Wiley 1986, 306 pages. This is the classic theoretical analysis of noise in electronic devices, not just 3-terminal amplifying devices, so the treatments of any one kind of device is truncated, and only the most common circuit configurations are discussed.

Uses a T model of the transistor, with 3 voltage noise sources for low frequencies; but 3 voltage and 2 current sources at high frequency.

Has a whole chapter on flicker noise in various kinds of device, but again the treatment is broad versus deep.

[Mike Sulzer]

The key is the word "better" in the original statement of the rule of thumb. You are assuming that this meant for noise only, which is not the case. The issue was general suitability for use with sources in the stated impedance range, and there is a lot more to it than noise. In fact, very few circuits are noise dominated.

Yes, I did think we were talking about noise in this discussion, and not much else. Yes, I did think that I said that CE and CB would give the same SNR, and that you disagreed. I am glad that one of your textbooks agrees with what I thought I said.

Anyway, I feel as if I am sinking into quicksand and am about to run into Alice and that stupid rabbit, so I think I will retire from this discussion.

[Joe Gwinn]

Yes, I did think we were talking about noise in this discussion, and not much else. Yes, I did think that I said that CE and CB would give the same SNR, and that you disagreed. I am glad that one of your textbooks agrees with what I thought I said.

Almost, but noise is only one component of the signal-to-noise ratio. We started out implicitly but wrongly equating N and SNR.

The assumption (well, hope) is that better impedance matching will increase the S part, without also increasing the N part (which varies with device conditions in complex ways). Flicker noise may prove to be the dominant issue.

I think that while the theory is helpful for generating ideas and directions, the issue will only be settled in the lab.

[Joe Gwinn]

Update on SNR and Flicker Noise

I did a bit of digging into the issues raised above. All sources agree that (excluding flicker noise) the Signal to Noise Ratio (SNR) does not depend on circuit configuration (common emitter, common base, or common collector), as expected; and and some sources also say that adding a series resistor does not help, and in fact usually hurts noise performance. The problem with these two results is that the equations used to derive these results do not apply in the flicker-noise-dominated audio-frequency region of interest to us.

Some sources also say that flicker noise can be modeled as a noise current source in the collector arm of the T model. Which begs the question, because if it is really so easy to model flicker noise, why then is flicker noise excluded in all the standard treatments? Perhaps because nobody could make the models work, or perhaps just because most people who worry about SNR are working far above flicker-noise frequencies. Except for the audio folk. So, it's back to the lab for us.


Anyway, the best treatment I have found so far is not where one would have thought to look:

"Electromagnetic Compatibility Engineering", Henry W. Ott, Wiley 2009. Pages 357 and 366 are particularly on point.

Henry Ott is the Grand Old Man of the EMC/EMI field, and this book is also a good source on the theory of electromagnetic shields.

[Joe Gwinn]

Flicker noise may be multiplicative in nature

For me, it was always a mystery why all the standard texts ran away at the mention of flicker noise, saying it wasn't well understood, although one source (Ott?) did say that in a BJT, the flicker noise generator appeared in parallel with another of the standard noise generators. Very odd: if in parallel, why not combine them? One assumes that the reason is that combining does not work, but no reason was given.

Yesterday, while researching an unrelated issue (low phase noise RF signal sources), I came upon a reason, in a statement that made me sit right up:

"Low frequency flicker noise (actual frequency, not offset frequency) is often considered to be a result of modulation processes and therefore exhibits multiplicative properties." (Page 354 of [1].)

All the other kinds of noise discussed the present thread in reference to amplifier noise are additive, not multiplicative, and the difference matters a great deal. One difference is that while one can overwhelm additive noise, multiplicative noise increases as signal power increases, their ratio remaining constant. This difference will invalidate any analysis that assumes additive noise.

So that's why combining didn't work. Not that they knew the reason back in the day.

I will chase this multiplicative-origin thread down. I assume that it's a relatively recent discovery.


Ref [1]: Oscillator Insights Based on Circuit Q", Roger L. Clark, IEEE 1991, 45th Symposium on Frequency Control, pages 352-359.

[Mike Sulzer]

... the flicker noise generator appeared in parallel with another of the standard noise generators. Very odd: if in parallel, why not combine them? One assumes that the reason is that combining does not work, but no reason was given.
....

"Low frequency flicker noise (actual frequency, not offset frequency) is often considered to be a result of modulation processes and therefore exhibits multiplicative properties." (Page 354 of [1].)

I have always assumed that it was easier to leave them separate because they have different frequency responses.

It can only be partially a modulation process if the signal can be involved as opposed to just other noise.

To see this, consider a device in which the flicker noise is caused by modulation of the another noise source, and this modulation applies to a signal as well, when it is applied. With no applied signal, assume that the flicker noise component is large enough to be clearly heard at low frequencies. Then as a signal is added, one should hear the modulation on the signal as well. Since on does not, one or more of the assumptions is wrong.

[Joe Gwinn]

I queried the author of the article referenced above. He allowed that the wording of the article is a bit misleading on this point. The discussed multiplicative effect is how the existing baseband flicker noise becomes phase noise (that is, becomes phase and amplitude modulation on an RF signal). How that baseband flicker noise was created in the first place is not addressed.

I suppose one could regard flicker noise as arising from a multiplicative effect that modulates the DC bias currents, but I don't know that this view reveals anything new. Nor is it likely that I'm the first to think of this approach; flicker noise has been a research item since the invention of the vacuum tube amplifier: Triode - Wikipedia, the free encyclopedia.

[Rick Turner]

Saw a "string return" moving coil pickup on a violin in 1973. Lane Poor and Ron Armstrong then fought over who came up with it for guitar/bass, but in fact, neither had invented it. Lane tried to sell it to Martin Guitars. They didn't go for it.

Like the Lightwave, there are a number of "inventions", ideas, whatever that are wonderful "science projects"...and terrible products. If you want to do science or art projects, just know that that's what they are and have fun with them. I've learned to differentiate the art projects from the money projects, thank heavens... I still do the art projects but...

Plug in a '58 Strat into a Marshall stack and burn the house down. To hell with the hum. So what if it buzzes when you take your hands off the strings...

But if we could have it all...

[David Schwab]

Plug in a '58 Strat into a Marshall stack and burn the house down. To hell with the hum. So what if it buzzes when you take your hands off the strings...

But if we could have it all...

Speaking of which... I've mentioned this before here, but I came across a post you did at MIMF back in 2000.

You wrote:

You can enlarge the area picked up by the dummy coil and reduce the number of turns..... A series dummy coil made the same size as the perimeter of a control cavity and wound with fairly large wire might cancel a fair bit of hum with minimum effect on tone.....Just a thought.

Apparently that idea worked well enough for someone to get it patented... five years later. Makes me wonder where he got the idea? Seems to work very well too. I'll assume you know the system I'm speaking of. Anyway, seems like you had the idea first.

[Rick Turner]

Hi, David...

Yeah, I know about that patent; I'll have to get a copy and read it. I also know the guy who assigned or licensed the patent to John Suhr; he used to work in the engineering department at Seymour Duncan and apparently developed the hum canceling coil thing on his "off time". The folks at SD were extremely generous in not claiming the intellectual property which they had every right to do.

And yes, my having posted that is legally "prior art", and would probably negate the core of that patent if it ever came up as an issue.

One of the interesting things that Illich did was to put in a CR network to get cancellation of 60 Hz harmonics to more or less match what's going on with the actual pickups.

I might come back to this idea in the future; there's this guitar I want to design...

[David Schwab]

Hi, David...

Yeah, I know about that patent; I'll have to get a copy and read it. I also know the guy who assigned or licensed the patent to John Suhr; he used to work in the engineering department at Seymour Duncan and apparently developed the hum canceling coil thing on his "off time". The folks at SD were extremely generous in not claiming the intellectual property which they had every right to do.

And yes, my having posted that is legally "prior art", and would probably negate the core of that patent if it ever came up as an issue.

One of the interesting things that Illich did was to put in a CR network to get cancellation of 60 Hz harmonics to more or less match what's going on with the actual pickups.

I might come back to this idea in the future; there's this guitar I want to design...

It certainly is prior art... I read your post after his patent was out and was shocked how similar it was. I've been meaning to play around with the circuit, but there's way too much on my list of things to try out as it is.

Here's the patent PDF.

By the way, I've certainly borrowed a lot of your ideas over the years, but I never pretend they were mine! I think anyone who makes "modern" basses has to give you a tip of the hat.