noise question

Yes, just inverted?

(also, "negative" resistance?)

Alas, it doesn’t have imaginary noise.

There’s no general formula, because a negative resistance can’t exist in thermodynamic equilibrium. You have to figure it out from the actual circuit.

When it’s just an LTspice spherical-cow resistor, I don’t know what happens in a .noise simulation.

Cheers

Phil Hobbs

A negative resistance is an active circuit, and you have to add up the noise contribution from all the real components in the actual circuit.

Imagining that you could treat a negative resistance as if it were an isolated component is a trifle unrealistic.

I just tried it: In LTspice the sign doesn't matter, only the absolute value. Also, if you put a positive resistor in series with negative one, the noise voltages add RMS-wise, like you'd expect of independent sources.

In real life, a negative resistor may have more or less noise than an actual resistor, depending on the low-noise design skills of the designer.

I think you knew that...

Jeroen Belleman

Cool. Thanks.

Sure, I was considering an ideal neg resistor, without added noise from active parts.

As a college project, I built a 2-terminal negative resistor and plugged the negative value into a bunch of equations (voltage dividers, RCs, LRCs, things like that) and demonstrated that they worked that way in real life. That was fun.

What I was thinking lately was about making an LC oscillator with very low phase noise, namely low jitter in my world. The finite Q of the parallel LC is equivalent to a shunt resistor so I'd expect it to have the Johnson noise of that equivalent resistance. Then the active stuff must look like a negative resistor, which is noisy too.

LT Spice noise analysis is very limited. I have sometimes added some random-noise BV blocks in series with resistors and such, so I can do genuine nonlinear sims with noise. It's actually easier to breadboard.

Sure. I = -E/R.

Connect that to a battery, and it charges the battery. Across a capacitor, you get an exponentially increasing voltage.

That works in Spice.

Yes, that's what I'd expect too.

BTDT. What's with the nonlinear bit? LTspice noise analysis is basically an AC analysis, no?

Jeroen Belleman

LT Spice noise analysis is weird. You need one signal source, even if you don't use it. And you can only probe one node. It must be entirely linear.

I want to simulate jitter in a LC oscillator, and of course an oscillator always has some nonlinear amplitude limiting mechanism.

You can probe *components*, rather than nodes, to see how much each contributes to the output noise.

I think I'd open the loop somewhere and treat the thing as an amplifier. I dabbled with the conversion between voltage noise and oscillator jitter, but the details have faded somewhat.

Jeroen Belleman

LTspice has time domain noise sources. They’re actually deterministic, of course, but you can use a big time offset to fix that.

The scaling iirc 1 sample per second, so you wind up using

noise(1e8*time+2e9)

and stuff like that.

Cheers

Phil Hobbs

Am 15.07.24 um 20:04 schrieb john larkin:

What irritates me is, that I need to edit .noise and .ac commands if I want to see noise and some gains; I want want both at the same time. The way it's now is clumsy.

At least, that seems to work. <

If you want nonlinear noise, you need Keysight's Advanced Design System or such. Be prepared to a 5 or 6 digit price tag, depending on options. The keyword is harmonic balance simulator.

cheers, Gerhard

I've done a couple of LC oscillators using an asymmetric current mirror as the amplitude-controlling mechanism. That stays linear as long as the amplitude is stable, and it only gets very slightly non-linear while it is getting the amplitude back to where it ought to be.

As a jitter source it should be innocuous.

I've also simulated a Wein Bridge with an Analog Devices AD734 as the gain-adjusting device, which is essentially the same idea (but with a linearised current mirror) and Spice suggests that it introduces less distortion than the usual gain-adjusting mechanisms.

Use two of them and you can control frequency (over a limited range) as well as amplitude. I've never found an application for that.

Nah. The LTspice noise() and white() functions give you time-domain noise, with uniform or Gaussian amplitude statistics, and are good enough for many things.

If you want separate voltage and current noise contributions, you can use a voltage controlled current source.

To keep the complexity down, you need to do a little analysis to figure out what the dominant noise sources are going to be, so that you can just model those.

Cheers

Phil Hobbs

random()

and white() functions give you-domain noise,

Isn't that negative voltage, as written?

I should probably have more coffee before trying to figure this out

That assumes that oscillators are mostly linear, but they aren't. Switching on a really linear oscillator would imply an explosion soon after power on.So there must be a large scale non-linearity to limit the growth.

There are even proposals that say that there is an optimum point in the cycle to inject all the feedback for best phase noise.

(and books on it)

Our ex-regular Kevin Aylward, WardenOf The King's Ale <

interesting: <

Gerhard

No, it doesn’t. It’s the normal .tran simulation, not .noise or .ac.

Try it sometime. It works fine, as long as you scale and offset the argument reasonably thoughtfully, to get the right bandwidth and enough independent runs for good statistics.

Using white() or random() as part of a bv or bi expression will let you model the changes in the noise with bias.

Sure, oscillator theory is fun. Try looking into the math of injection locking, if you like that sort of thing—it’s full of bifurcations and strange attractors and limit cycles and stuff.

Most of the time I try to be relentlessly practical, but often fail. ;)

Cheers

Phil Hobbs

They can be remarkably close to it, once the amplitude has settled down.

If you use a AD734 four quadrant multiplier as your gain-setting device you can do even better, which is something I've only done in simulation,but should work in real life.

No. You just have to be able to control the gain around the loop to hold the amplitude of oscillation at a stable level. Obviously that gain has to be changed from time to time, but it doesn't have to changed all that much or all that fast.

Seems implausible.

Having just finished my first cup of Peets, I'm way ahead of you.

The minus sign just means that the current flows the opposite way from what an ordinary resistor would do. A negative resistor doesn't load down a signal, it helps it.

In my case, I have basically a negative resistor across a parellel LC. So instead of a ringy thing dying out exponentially, it increases exponentially. That's an oscillator.

Of course, an exponentially increasing oscillation can't keep increasing forever, so something has to limit the swing or else Planet Earth will be incinerated. If you Spice the case of paralleled L C -R, it swings to teravolts and runs out of floating-point range before we are all killed.

(Of course you have to goose it to get it started.)

I guess the L C -R can still have a low Q, in which case it would be stable. Sure, it doesn't have enough gain to oscillate. I've done that.