PLL tricks

Cheers, James Arthur

On Fri, 19 Sep 2014 08:49:43 -0700 (PDT), snipped-for-privacy@yahoo.com Gave us:

Yeah, but what is that chirpin' twerp laden with?

There is a comedy routine, over there, by Monty Python, IIRC...

Except *that* was an "UNladend swallow".

I like the coconut shell horse trot stuff. :-)

On Fri, 19 Sep 2014 08:49:43 -0700 (PDT), snipped-for-privacy@yahoo.com Gave us:

You are phooly ladend...

Now swallow.

BRL!!

Yes. I agree that close in noise (LF) is limited, and is a basic function of the xtal Q and the 1/f noise of the oscillator transistor.

For cases where the processing requires the oscillator to be squared up, e.g. not where a sine wave oscillator is multiplied by anther sine in a pure continuous demodulate, then the squarer/limiter may be the dominant noise source.

There is a theorem that says a band limited signal going into a limiter which is band limited on its output, has the same output phase noise S/N as its input. i.e. the limier does not increase the noise. The flaw in that theorem is that it is impossible to band limit the intrinsic noise of the input transistors of the limiter.

Simulations (I am lazy) show that an ideal comparator, will, at the minimum, increase its internal noise by 12dB. Additionally, in practice, an xtal oscillator is already massively band limited such that you can't realistically, band limit it again. The result is that if a limiter is used, it tends to dominates the noise floor.

Yes, but it is essentially, impossible to not significantly lower its Q in practice. Base current noise dropped across the impedance at its base is a dominant noise source if the impedance is too large. A bias resistor must be there to bias the transistor. There is therefore a compromise, as low as possible to minimise base current noise and its thermal noise, but high as possible to maximise Q. An inductor can be placed across the resistor to reduce LF noise, but its limited in practice to avoid parasitic oscillations.

Reference Generator Master Oscillator .------------------------------. .------------------------. | VCXO | 720 | VCXO | | .---. .---. .---. .----. | KHz | .---. .---. .---. |

Actually its not mine, many xtal oscillator vendors use the technique. Second its not passive. Multiple stages are done actively, with processing, but I am not at liberty to say what processing :-)

In fact, in contrast to Steve Jobs who was an expert in stealing peoples ideas, and getting other people to do them better, I steal the ideas, and do them better myself.

Indeed.

Detailed simulations on actual designs using that very expensive Cadence R.F Phase noise analysis, which measurements on real hardware have shown to be very accurate, show that the multiplication part of the system does not really dominate the noise. That is, low frequency, close in noise, is insignificant, and flatband noise is not dominated by the tanked multiplier itself. i.e, its largely the theoretical, input noise X N

Principle noise is still due to the input oscillator, and oscillator limiter. So, the tank method can get you pretty close to the theoretical limit, meaning the incentive to investigate better alternatives is not that great. Its the diminishing returns thing.

Kevin Aylward

Leesons equation is only qualitatively, useful. Its F is unknown, so its useless to actually calculate LF PN noise from.

In fact, you can't calculate PN manually in practice, for real systems. The equations are way too intractable. The only realistic way to design for low phase noise is to use a phase noise simulation tools, e.g. Cadence RF or Agilent.

I have paper on my site by A. Demir

The idea of using pen and paper to try and calculate closed form solutions to todays problems is dead. It is a fact that the percentage of systems with closed form solutions compared to systems without, is the limit as x->0.

Its a paradigm shift to understand that using simulation tools to get the answers is the best way and only way. I can an answer and gain real understanding of a problem in minutes compared to pissing about with a pencil and pad for months.

Kevin Aylward

There is still, I think, an academic-leftover prejudice for equations. In a practical, complex, nonlinear system, the math may be enlightening but not practical for quantitative solutions.

Simulation can be enlightening, too.

But then, some physical systems aren't well suited to simulation either, so people still breadboard.

On a sunny day (Fri, 19 Sep 2014 12:16:41 -0700) it happened John Larkin wrote in :

I completely agree. Take my DVB-S circuit for example. I designed it, build it, its basically simple, and it did not work. DVB-S (satellite) receiver did not work on it, no lock, signal strength yes, plenty.

So watsit???? I did not know, so I decided to read the chip registers of the receiver demodulator chipset, to find out what problems the demodulator had. Well, it did not (could not lock)...

So what would be the reason? Step aside, mm could be phase instability, frequencies and processing math, the software, are OK. Mind you I wrote all the soft, nobody is releasing source, I was the first one, but could check output against a close source program. But how instability? I filtered VCO control voltages. Maybe supply? Yes supply was noisy, logic spikes... Added an inductor and big capacitor. Things worked right away, as indicated pico second (2 degrees phase error at 1.something GHz,. Now I have a working circuit, that I can replicate as many times as I like.

Slimulations were never any part of it, and you could not get the models anyways. now I also have a good constellation vectorscope for QAM.

EVEN if the simulation was there, you would wind up with a few "makes you get that warm feeling" screen of 'it should be like this', with no hardware, and you would be exactly where I was before I started. Slimulation is for NASA pictures of alien planets, we all know they are dead and without a ride on a Russian spacecraft they cannot even get it up (uche) for a hundred miles. Pictures of alien planets with cows grazing.

No, that is not going anywhere. Reality is always just a little bit different.

And a zilion(tm) neurons with more than half a century experience cannot be beaten by a couple of incomplete by design formulas. Never.

I can get microwave layout software, but for Gods sake, try brain first. The consensus is not (very often actually) what really is. Take the ground plains fetish for example. :-)

But the Leeson equation predicts that the oscillator exhibits a huge noise gain close in, which the comparator doesn't.

Cheers

Phil Hobbs

Yes. Its job security. What would they do if they did not produce equations

It took me a while to eliminate my own prejudice. I once spent an inordinate amount of time learning all sorts of ways to solve differentia equations.

Enlightening sometimes, often one cant see the forest because the trees are in the way. The equations are usually unmanageable. Middlebrook appreciated and stressed this, illustrated by his reduced element theorem. He didn't go far enough.

Very enlightening. Regarding phase noise, bench work simply can not compete with the tests you can do in the virtual world. e.g. the sims spit out the value of each contributing noise noise value, in order.

Yes. Simulation only works when the individual models accurately reflect reality, and correct theory expressed in a numerical solvable equations exist. The reverse/forward recovery diode thread shows that there are still areas lacking.

Kevin Aylward

Whatever gets results is good. I wonder how intern errors sum. I don't think it's a half power law, although I suppose it depends on the intern. Maybe you can hire the re-incarnation of Claude Shannon.

I'm biased toward lots calculation and simulation before building and fiddling - for complicated systems anyhow. I think it's from working early in my career without much theory. That was frustrating as hell.

Anyway, I make your 155.52 MHz oscillator with 0.2 ps rms jitter as containing 15.0 bits of information. When it degrades to 12.7 bits you're on the edge of your 1 ps error budget. That and how fast it degrades should set the performance requirements on the feedback loop.

ChesterW

Seems to me that more modulation of a sine wave carries more information, not less.

Does a perfect sine wave convey an infinite amount of information?

I think your approach would suggest that total broadband noise would convey infinite information.

He is talking about the loss of info from the presence of noise. Remember that guy Shannon?

That's actually right, if you're the source of the noise. ;)

Efficient modulation schemes have spectra that look a lot like noise.

Cheers

Phil Hobbs

I was thinking about weather records, as in "the highest temperature in Mule Kick Arizona on this date in recorded history".

Start observing gaussian noise on a storage scope. The longer you wait, the more likely you'll see a record peak. The more recently you started saving data, the more likely you'll see spikes.

Weather stations used to be widely scattered and infrequently read, and there was a start date for "historical records." So naturally, as time goes on, and more weather stations report more often, new records will be set, here and there.

Am 19.09.2014 um 16:11 schrieb snipped-for-privacy@yahoo.com: to 52Km/hr only increases draw from 10.4 to 14.9W.

Arthur, as a king you are are supposed to know these things!

:-) Gerhard

You can attempt to steer the subject to any direction you wish, but you'll fail miserably.

The question was reading of the schematic, not understanding how it works. I understand how it works and there for I saw no mistakes nor did I see a problem with the drawing.

Trying to move the goal post to blur your lack of understanding isn't doing much for your image.

Suck it up..

Jamie

The answer depends on un-supplied details.

In the sense of sending data using a carrier, then more modulation can mean more information content. It depends on the determinism of the modulation.

In the sense of sending data using a carrier, then a perfect sine wave with infinite extent in time conveys zero information.

In the sense of a time reference, a perfect sine wave with zero error conveys infinite information.

So how about that, the same signal conveys both zero and infinite information, depending on the question. Reminds me of a story about imaginary cows that I may post later.

In your oscillator problem I modeled the system as a clock with each reference tick happening at 1/155.52 MHz. The 0.2 ps rms jitter is the uncertainty in the time measurement. For these numbers, that's about 1 part in 32k, which requires about 15.0 bits to specify. Since the jitter is random from cycle to cycle, this represents the maximum precision of the system which I take as the information content of the (perfectly aligned) system related to timing.

When the error from sources besides jitter accumulates to your spec of 1 ps, then the system is precise to about 1 part in 6400, or about 12.7 bits. So, when random occurrences cause your system to drift from perfectly aligned to just out of spec, the time reference has lost about

So, if you know the drift rate of the timing of the system, then you can calculate the data loss rate in bits per second. Since bits per second, bandwidth, and SNR are related in the channel capacity formula, this may be a tool to evaluate feedback schemes. If the feedback can't supply enough bps, then it can't control the loop. I think this might set a necessary but not sufficient condition, which at least may keep you from wasting time building something that has no chance of working, which of course is the purpose of analysis in engineering.

ChesterW

Dear Rick,

Your posts make we want to be a better man, and specifically one who has learned to use the filters in his news reader.

ChesterW

The other place simulation fails is when the sim would have to run for days or years to produce an accurate result. Like a 2-minute event simulated at 10 fs time steps. Or something that #$%^& LT Spice refuses to converge on.

Don't get me wrong, some things, like filter design, are crazy to do by simulation or experiment. Ditto things that involve real physics fundamentals.

The diode recovery thing is not suited to math analysis, because we don't know stuff like doping profiles of diodes that we can buy, and Spice doesn't handle diffusion anyhow. We may as well use a behavioral model based on part measurement, limited to modeling the behavior that happens to matter right now. The textbooks tell us what sorts of general behaviors we can expect in diodes.