[time-nuts] AM vs PM noise of signal sources

Attila Kinali attila at kinali.ch
Mon Jan 8 09:02:24 EST 2018


Disclaimer: I am by far not an expert in oscillators. Please correct me if
I am wrong.

I am putting my replies to a few mails together into one, as not to clutter
the mailinglist too much.

On Sat, 6 Jan 2018 19:53:20 +1300
donald collie <donaldbcollie at gmail.com> wrote:

> So to be lowest noise, an oscillator should have the highest Q resonator
> possible in its feedback loop, operate in class "A" [for maximum
> linearity], and utilise active amplifier device(s) that contribute the
> least noise [both amplitude, or 1/f], and phase. 

Actually, it shouldn't. At least not if you want low phase noise.
The sensitivity of the output phase noise to the internal noise sources
changes during the period of the oscillator. AFAIK this has been first
noted by Hajimiri and Lee in [1], you can also find it mentioned in [2].
The small problem with that is, that it will lead to an increase of AM
noise, which in turn is turned into 1/f^2 and 1/f^3 PM noise through
the oscillator.

Also, you might want to back down a bit on the loaded Q, if you can
significantly improve the noise performance of the sustaining amplifier,
by better matching. (Gregory Weaver mentioned this during a discussion)

> Resistors in the oscillator carrying DC make 1/f noise - the best in this
> respect are the metal type, I think - so use metal resistors or WW.

The 1/f noise of resistors is lower than that of the semiconductors
involved, unless you are using carbon resistors. Hence most people
just simply ignore it. The commonly accepted theory is, that the 1/f
noise in resistors comes from the electron traps at material faults.
In carbon resistors, these are formed by the edges of the carbon particles.
In other these are formed by the way how the resistive material is deposited.
Hence thin-film and metal-foil have the lowest 1/f noise.

On Sat, 6 Jan 2018 20:12:16 -0600
Dana Whitlow <k8yumdoober at gmail.com> wrote:

> I've long wondered if a very slow AGC might avoid the nonlinear mechanisms
> issue except, of course, for things happening within the AGC loop's
> bandwidth.
> Is anybody reading this aware of what the truth really is?

The truth is complicated. There are so many effects that one has to
analyse and keep track that it's hard to say which one dominates in
a design, without doing extensive calculations or simulations.
At least that's my impressen, when I read papers on noise in electronics.

On Mon, 8 Jan 2018 01:02:11 +1300
donald collie <donaldbcollie at gmail.com> wrote:

> Does any limiter, soft or hard, [and perhaps any nonlinearity  of power
> term 3 or greater in the amplifier of an oscillator] cause the "baseband
> 1/f noise to translate up to the resonator frequency [a form of
> crossmodulation]?.

The upconversion happens regardless of the limiting circuit. It stems from
the sustaining amplifier being non-linear. Even running a transistor completely
in a class A configuration will lead to upconversion, because not all the noise
sources are at places where the transfer function through the transistor to the
output is linear.

> I wonder this because
> phase noise vs freq plots look a bit like the 1/f plots of a resistor, or
> active device, or power supply. 

I do not understand this question. Noise looks "the same" for all devices.
Their only difference is the relative levels of 1/f^a noise. As such, it is
hard (impossible?) to say which device causes the noise at the output of
an oscillator by just looking at its output.

> Ceramic caps, and resonators [I`m thinking
> of quartz crystals] don`t pass much DC, and as I understand it, 1/f noise
> is associated with dc passing through resistors, or semiconductors.

1/f noise is generally associatiated with semiconductors and (carbon)
resistors, yes. But the crystal itself has its own 1/f noise and depending
on your circuit that might be actually the limiting factor and not the
electronic circuit.

Also keep in mind that a lot of electronic components are electro-mechanical
in nature. Ie they convert mechanical noise (aka vibrations) into electrical
noise. A prime culprit of this behaviour are capacitors and inductors, but
also semiconductors are known for this.

> So the
> best way to go might be to have a very linear amplifier, which exhibits
> very low noise [perhaps 150dB below the operating level], with an AGC loop,
> that sets the operating levela little below the level at which the amp
> starts to clip - this could be done with a thermistor to avoid the AGC loop
> altering the [optimised] operating conditions of the amp. Alternatively you
> might be able to use a tetrode device like a dual gate MOSFET, and apply
> the AGC to the second gate. Thus you could keep the extremely linear amp
> extremely linear. [150dB below 1Volt RMS is 0.032uV RMS].

If you google for Rohde/Poddar and noise/oscillator, you will find quite
a few papers and articles on how to build low-noise oscillators that are
only limited by the thermal noise in the 50Ω source resistance, Ie oscillators
that have a white noise floor at almost -174dBm (note: dBm not dBc).

It is "known" how to build such oscillators, but it doesn't mean it's easy :-)

			Attila Kinali

[1] "A General Theory of Phase Noise in Electrical Oscillators",
Hajimiri and Lee, 1998

[2] "How Low Can They Go?", by Poddar, Rohde, Apte, 2013
It is upon moral qualities that a society is ultimately founded. All 
the prosperity and technological sophistication in the world is of no 
use without that foundation.
                 -- Miss Matheson, The Diamond Age, Neil Stephenson

More information about the time-nuts mailing list