--
| James E.Thompson | mens |
| Analog Innovations | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| San Tan Valley, AZ 85142 Skype: skypeanalog | |
| Voice:(480)460-2350 Fax: Available upon request | Brass Rat |
| E-mail Icon at http://www.analog-innovations.com | 1962 |
I love to cook with wine. Sometimes I even put it in the food.
It's difficult to analyze in general terms no matter what tools you use. Yes, you can simulate a system and get an answer for a particular set of numbers, but you can't really get a general solution.
The approximation, on the other hand, is always wrong, but for small phase noise (which is usually the case) it's not wrong by much, it's wrong by a predictable amount, and you can use a whole lot of techniques from linear systems theory that simply do not apply to the nonlinear case where it's 'pure' phase noise.
If you want phase noise in radians, then take Ac (constant) and An(t) (noise process), and use
Ac * (cos(w * t) + An(t) * sin(w * t))
Then for An(t) always small enough, An(t) is the phase noise in radians.
--
Tim Wescott
Wescott Design Services
http://www.wescottdesign.com
You already have a Spice model... just add AF= to it. ...Jim Thompson
-- | James E.Thompson | mens | | Analog Innovations | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | San Tan Valley, AZ 85142 Skype: skypeanalog | | | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | | E-mail Icon at
formatting link
| 1962 | I love to cook with wine. Sometimes I even put it in the food.
The i didn't cover the phasenoise term, so you had some form of AM.
This is *way* too high a max deviation angle, and puts you deep into the nonlinear modulation region.
A really bad phase noise level is -60 dBc, so the largest likely rms deviation is 0.001 radians (0.057 degrees).
AM and PM sidebands do look alike; the difference in in the phase.
Read up on the Armstrong method of generating FM signals.
This is because phase modulation is a nonlinear phenomena, and as the angular deviation increases, more and more sidebands start to arise, stealing energy. AM does not behave this way.
You really need to read up on modulation theory, or none of this will ever make any sense. There are many lectures on the web, written by various EE professors for their students.
For oscillators and phase noise, you are trying to rediscover the Leeson effect. Again, some reading is in order. I would start with Rubiola's book on Phase Noise. He also has many articles available gratis.
.
He also explains a bit of modulation theory, but on modulation he is a bit terse for my taste, so I'd read about modulation elsewhere first.
Try harder. It's hard to beat, despite all the false alarms.
I assume that BW is bin width not bandwidth. Yes, that will eliminate truncation errors from spraying error energy everywhere.
As for windows, my preferences run more towards Taylor and Dolph-Chebyshev.
This has nothing to do with flicker noise. Depends on the physics.
If one is for instance measuring the impedance of something that's largely capacitative, or something that is magnetically coupled, 1/f will level the SNR versus frequency.
Mainly because any amplitude limiting hits the AM, but not the PM. PM survives a hard limiter undamaged.
Narrowband is a relative term here. The noise can be flat within its band.
Actually, phase noise from an oscillator is usually pink noise. There is a large literature. Leeson (mentioned in another reply) was the first to explain why - see Rubiola.
I do this too, but step one usually involves reading books and articles.
Yes, the input can speed things up. But you have some homework to do.
Jim, thanks that matches exactly all the information in MicroSim's Manual for PSpice, but all in one spot. However, I do not know how to go from the manual describing "what the model is and what each term is" to taking the data sheet for the MC1496 and deriving the individual NPN's that would go into that model. just a bit beyond me.
To make progress, I simply made up a sample mixer circuit using the 'application' shown in the sheet, used simple 2N3904's from the library and added all the resistor noise sources [internal and external] to see what comes out. the 500, 1k, and 3.9k resistors. No power supply noise.
Drove the mixer with fc=1MHz and fm=10kHz, interestingly the spikes are indeed spread after doing the analysis.
Surprisingly, or maybe expectedly, the noise is low, but at least above numerical analysis values by quite a bit.
I'm still floundering around on what I'm looking at, but this is fun.
Thanks! makes sense since the origin is very small 'rattle' and nothing major.
difficult to read. very dyslexic, didn't read until 8.
Especially don't trust a lot after seeing things in print which are pure misleading garbage and especially after an incident on a weekend's rush and crush project and getting bit really hard by trusting an article in McGraw Hill publication [which was wrong] I keep going back to basics, build up my own tool set.
But there is no quicker way home than to read some of the classic texts. Or take a course, if one is available.
One saving grace is that the part of modulation theory needed here is a small fraction of the typical book on modulation theory.
There are well-respected books on modulation theory. They too have errors (all publications do), but are usually far better than random web sites.
Authors also differ in their approach, so I find it best to read a few books in parallel, and compare. Nor do I understand everything at first. Usually, it's read a little, do a little, read some more, do some more, ...
WOW! Thanks for that URL! I see why you like that window!
I'm going to try it out and see what it does. I need something that does NOT distort the 'grass' outside.
The following did not need windowing: For what it's worth, just tried LTspice analysis of a MC1496 mixer made up from data sheet's simplified schematic implemented in the sample application, using fc 1MHz and fc 10kHz. Compared three forms: PSpice/LTspice .tran, then added the noise sources from the resistors, then added the noise sources from the transistors [in this case the 2N3904 NPN's in LTspice library]
VERY interesting results: well, at least to me.
the basic .tran shows NOTHING of what's going on with noise
adding only resistor noise shows noise floor and 'skirts' on the sides of all the tones that get generated. ...but don't know how to interpret that, sigh.
adding additional noise from the 9 NPN's made very little addition to the mixer's noise, being too small [or I made a mistake, will double check that today] so complicating the schematic with NPN noise was not as important as complicating with R noise.
Thanks again for that window, can't wait to try it out on my noisy RC oscillator to see what pops out of the spectrum.
NEW QUESTION: Any idea what amount of noise to expect in a standard 2 NPN flip flop oscillator? You know, 2 NPN's, two base R's, two collector R's and two cross coupled caps. I've got the model set up to run at approx 700Hz from
Define "distort". All window functions distort in the sense that they are optimized for one property at the expense of other properties.
Sounds like the Leeson effect. Rubiola has a good discussion of this.
Depends mostly on how quiet the transistors are. Flicker noise (also known as 1/f noise) is often the dominant kind of noise. This is a matter of how good (~clean) the semiconductor manufacturing process is, and may vary from manufacturer to manufacturer, even if the part number is the same.
By the way, it's best to power test circuits from a battery, to eliminate power supply noise issues, simplifying the diagnosis process. Also put battery and test circuit into a diecast metal box with RF connectors mounted in the walls, to prevent local broadcast stations from obtruding. Not to mention 60 Hz and harmonics.
Had a problem trying to add 1/f. No problem for me to easily add 1/f, BUT I'm having difficulty getting any representative numbers of what to set the models' flicker noise parameters to, default is always zero. If I use an OpAmp's spec sheet and manually add 1/f for Op Amp's works great, ...as long as simulation is only concerned/stays within the GB Product window, else once outside; all simulated input noise sources' contributions tend to drop to zero. And zero is unrealistic because the noise out of an OpAmp can skyrocket sometimes when outside the gain. You can see the output noise jump to 3X during skew rate limit, if you get a chance to see the results of my fully modeling the IC schematic for Jim Thompson's old design, the MC1530, again implemented with 2N3904's, but sadly NO 1/f term! I didn't get any values from Jim, whom I thought from his experience with so many foundries would give me some 'guesstimates'.
Good idea with the battery box. I design PCB mount SMPS systems to be equivalent to battery supplies for really, really low noise Data Acquisition Systems. Interestingly, [or maybe not] learned that most of the App Notes from the suppliers are not very good implementations. Rather have to play games with the filtering to 'control' the impedance looking into AND out of the filters. People don't realize that filtering is not some cookbook, but rather a design process, complete with performance specs to be met. And sometimes very difficult to meet, example of performance spec: less than 0.5 ohm from DC to 1GHz looking into or out of the noise filters. But do that and you have a pretty quiet supply AND pretty quiet power bus, else you can get some wildly high impedance, 10 to
100 ohms, at some frequency and that frequency is ALWAYS somewhere in the operation of your circuit! [Murphy's Law] But, you knew that.
Any URL for the 'values' in the Dolph-Chebyshev window? When I followed the description in the wikipedia article got garbage [I know, my fault, but still garbage] However, because of your URL and your discussing other windows, I suddenly noticed LTspice provides an incredible range of options when you plot their FFT! Never saw that before, and definitely had no idea what most of the terms meant to 'adjust' those windows, except for that excellent wiki URL. Mike Engelhardt had even included Dolph-Chebyshev Window! Even allows you to plot the window itself! So I can get some rough values to compare to at least.
Hmm. If the specsheet for a part doesn't give a phase noise curve, and you care about 1/f noise, assume that the part is useless for your application, and move on.
Absolutely. I need such a box for far less sensitive measurements.
Here is one example I calculated for another purpose. Dolph-Chebyshev weights. Generated by the MatLab R2012b "winwrite(sigwin.chebwin(32,90))" function from the Signal Processing library, for 32 samples and 90 dB sidelobes.
Google really does not work well with dialup access, and yet Google is the key to finding the information you need soon enough to matter. You might consider upgrading to broadband. The difference is the ratio between kilobits and megabits -- 1000:1.
The other thing to do is to find out what the classic textbooks are, and buy them used on Amazon. Most of the classics were printed in great number, so used copies in reasonable condition are cheap.
I bet this was also explained in the LTSpice documentation.
good point, but a lot of parts are rarely specifically designed for how they really end up being used.
how sensitive? I'm routinely into the subnanovolts and femtoteslas.
is ok, BECAUSE of your mention and successful search on google ended up finding an undocumented window in octave!, called chebwin()
Because of your wikipedia URL on Dolph-Chebyshev it was easy to learn how to use the chebwin function like this N=100000; w=chebwin(N,100); w=w'/(sum(w)/N);
where 100 because it's in terms of dB, not alphs then w becomes a 'normalized window that can multiply times the data and removes everything down to the 'known' noise floor ...at the cost of broadening the tones, also by a known amount.
THANKS again! never would have found this function, never would have known such window was possible.
Now I'm looking at the output of the two transistor oscillator to see if the phase noise shows up. looks like the 1kHz oscillator has 'skirts' of
+/- 40Hz down at -40dB, which seems like a lot of rattle, but still interesting tools.
Can't get high speed here. rural area. 17,261 feet from station so DSL sucks too.
couldn't find it! but then again can't find a lot in those stupid [at least to me] help menu thingies.
I get 'circular help lists, and sometimes what is sid works, doesn't. so discourages me from exploring that stuff. I just go back to my basic MultiSim PSpice manual! which was EXCELLENT! but new stuff isn't included,
Depends on the volume and cost sensitivity, and if there is a backup for when the original manufacturer changes his process to optimize some datasheet parameter, at the expense of parameters not specified.
Microvolts in that case, but I also worry about phase noise in radars, and that does get to nanovolts.
There are a lot of phase-noise related scripts on the Mathworks site, and many of these scripts will work under Octave.
Welcome. But get thee to the library.
Actually, what are you trying to build?
So, you only need to drive a few miles, to a coffee shop or public library, or a friend's house, to be able to do research at megabit speeds. You will be amazed at the difference.
Or, move the house by a few miles.
Using dialup fro real work is simply crippling, and while people will suggest places to look and search terms to use, they won't do your work for you, or at least not fast enough to be competitive.
Right. I didn't mean online help, which is generally useless. I mean books, be they pdf or paper. I prefer paper, for ease of reading, and of annotation.
Set up a spice schematic to simulate the conditions of its noise figure v frequency graph. I would suggest the RS=200, 1ma, @ 0.1Hz one to get the max LF variation. Then trial and error AF and KF to get a match.
I did a quick go and got KF= 2e-17, Af=1. It seems to be approximately ok for the 50u and 1ma graphs. Do check though.
I am quite tempted to use my daily availability of the $100k/year Cadence Virtuoso system to run one of your example tests to see if any of the method/results actually match up with what is known to work. A PSS Noise run for this type of schematic will probably take about 2.34 seconds. Unfortunately, probably hours to draw the schematic....
I have not really been following this thread in any detail, but pm/am noise is quite complicated. I am under the vague impression that you are using ac noise data to feed into a trans noise source. If so, an issue with this is that the noise is a continuous function of the large signal currents in the transistors. AC noise only gets one spot point. Cadence PSS does this all in the wash, by calculating what the noise is at each point in the cycle.
Post an exact schematic, and I will see what I can do in my lunch time :-)
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.