[jbay808]

> In fact, a remarkably common response to a diagnosis of resistor noise is to seek a source of "good" resistors, with "good" being defined as without thermal noise. This is impossible.

It's impossible to make a totally noiseless resistor, but it's also important to understand that all resistors are not created equal.

Most resistors have noise levels that are orders of magnitude above the Johnson limit. Potentiometers are especially bad.

If you want "good" resistors for noise-critical applications, I recommend metal thin-film resistors. They hardly cost any extra anyway.

Also, in cases where resistors are used to set DC signals such as offsets and biases, you can add capacitors to filter the heck out of those lines to decrease their noise contribution.

[xenocyon]

Electrical engineer here. Thermal noise is the same for all resistors with a given R, regardless of their method of manufacture. You cannot have thermal noise either less or more than this, so it's not a Johnson "limit" but rather a definite value. You are correct that there are other sources of noise such as 1/f noise. But more importantly, the manner in which noise manifests in the end result has to do with the circuit as a whole.

For noise critical applications you should do a noise analysis of the circuit as a whole rather than make ad hoc selections of components.

[jbay808]

I think you "well actually"'d a bunch of things I specifically didn't say.

I use "Johnson limit" synonymously with "thermal noise limit" because they are the same, and it's the limit of how low noise can be after removing all other sources of noise.

Most people, if they even learn about resistor noise, will only learn about thermal noise. If they're lucky enough to identify a resistor as the noise troublemaker in a circuit, they might not have any idea that they can potentially cut the noise 1000x by changing the resistor from thick film to thin film, with no other design changes, at the cost of one cent. It's not common knowledge, as illustrated by the fact that an article like this, dedicated to the topic of resistor noise, doesn't even mention it. And instead laughs at someone even considering to look for a "good" resistor as though it were superstition.

In fact, thick film resistors are far more common, so if you're in the situation where you're reading this article because you have a noisy resistor in your circuit and don't already know about Johnson noise, you almost definitely don't know about current/flicker noise, and since you got here because of a noisy resistor in the first place, a "good" thin film resistor is overwhelmingly likely to be the cure.

I'm not advocating ad hoc selection of components to reduce noise any more than selecting ad hoc components to reduce cost. A noise analysis can help you find the problem but won't help you solve it if you think your only option is to change the resistor's value, rather than its type.

[ldarby]

Elsewhere in this thread there's a link to https://dcc.ligo.org/LIGO-T0900200/public, which backs up what you're saying:

"Metal film or thin film resistors have little excess noise in these tests and thick film resistors show large excess noise"

[exmadscientist]

Yep. Any time I actually care about the value of a resistor (which is to say, for precision analog work, not power supplies or logic pull-ups or LEDs or ...), I'll pick a cheap thin-film part, even without doing analysis. This does several things for me:

1. I get a "pretty good" (i.e., >80% of best achievable performance for <<20% of the effort) resistor in my important location.

2. I get a separate BOM line item for that resistor, so it stands alone (or with similarly important parts) and is easy to single out to, say, ensure correct ordering.

3. The part will be physically distinctive, so I can also distinguish it on the circuit board. If I have space, I'll even go up one size, to make it impossible to swap out with the less critical parts. (It would work fine, if expensive, to put thin-films everywhere. The reverse is not true.) It also sends the "pay attention to this" signal to anyone down the line who only has the board to work from.

I favor Susumu thin-film resistors, as they're generally cost-competitive, very high quality, available in very tight tolerances if needed, and visually easily identifiable as Susumu parts (though the series is not identifiable). This means they have to be thin-film, because Susumu doesn't make anything else. RR and RG series are the go-to parts here. If cost is a concern, Yageo RT is also excellent, but not as noticeable on the board. For analog work, I'll go no smaller than 0603 (1608M) to ensure values are printed on the parts. This pays for itself after only one gain resistor mixup on the bench....

[robocat]

Your comment looks like it disagrees, but actually you are both agreeing that thermal noise is not the only noise source and that other design factors matter.

For other readers: some resistors have much lower flicker noise[1] where “wire-wound resistors have the least amount of flicker noise. Since flicker noise is related to the level of DC, if the current is kept low, thermal noise will be the predominant effect in the resistor, and the type of resistor used may not affect noise levels, depending on the frequency window.”

https://en.wikipedia.org/wiki/Flicker_noise

[colin_mccabe]

Thanks for the comment... good to have some EEs drop in.

[analog31]

I recommend measuring the noise. Systems guru Phil Hobbs said that you should know where every dB of noise comes from in your design. Of course a dB could be a little or a lot in your application, but the point is that you should perform a noise budget and then test your assumptions.

It's not necessarily easy, but recommended if possible. I was doing it with DIY equipment, so I don't claim traceable results.

In one case, I literally measured the noise of some resistors, and within the parameters of what I cared about, I found no measurable difference between metal film and carbon film. I was passing no DC current through the resistor. Some sources of "excess noise" are proportionate to DC current and can be corrected by appropriate filtering.

[wellthereitis]

Great tips, thank you!

[prpl]

… and overspec them everywhere reasonable