The issue is the number of unique runs you need to get representative statistics, even if you know the actual noise characteristics of the parts.
The other approach would be to measure all the partial derivatives of everything with respect to those noise sources, and then do the noise analysis analytically afterwards. That requires linearizing the system, though, which doesn't work well in strongly nonlinear systems.
Flicker noise especially is often far from Gaussian.
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics
160 North State Road #203
Briarcliff Manor NY 10510
hobbs at electrooptical dot net
http://electrooptical.net
Is flicker noise included in the LTspice modeling? I do not take that ito account, but think it's possible to add.
One run is all I've been using to get within that few percentage window.
Like I said, run .tranoise and out comes a time waveform with fuzz on it, and out omes a spectral display with a calibrated noise floor. And, can bounce back and forth between time/frequency. apply filtering etc.
It's in the built-in noise models for transistors, at least, but of course that only works in .noise mode. Flicker noise is ugly in the time domain, so it would be hard to include in a transient simulation.
Do you have a script or something to add the noise sources to your .asc files, or do you maybe have a library of noise-enriched subcircuits?
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics
160 North State Road #203
Briarcliff Manor NY 10510
hobbs at electrooptical dot net
http://electrooptical.net
What are the amplitude/time characteristics of flicker noise?
I can tell you DON'T try to use the built-in 'white' function inside LTspice. everytime you use it it is the SAME! and it's 'rounded' off so check the spectral content, you'll find you have to have sampling of the white function 10 times your interest, else the frequency has a downward, rather linear, tilt of over 10dB
Depends on its origin, but it often has a telegraph/popcorn character, like a badly cleaned circuit board. It's also statistically nonstationary in general.
Hmm, that's odd. I'd have expected it to be a Box-Mueller generator or something like that, with a different random seed each time. Do you have LTspice set to re-use the same random seed for each run? (That's a control panel option iirc.)
So I'm still curious about whether you built a magic subcircuit library of noisy components, or what?
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics
160 North State Road #203
Briarcliff Manor NY 10510
hobbs at electrooptical dot net
http://electrooptical.net
Doesn't matter since more importantly, WITHIN a single run it's identical, so two noise sources become correlated.
Mike said he purposely made it the same, so you could compare results as apples to apples.
But the irksome part for me, is that even *if* you use it; unless you set the maximum step size more than 10 times smaller than your interest, you get misleading results with that rounded and tilt to the spectral content.
Wow! I thought I said exactly how to do it.
I can send a sample resistor circuit to you [I think] it completely depends on size AFTER zip [remember I'm on dialup here]
the models tend to be large. The MC1496 model is over 35MB AFTER zip. but as I said, correlates to better than 2% overall between 'stepping' voltages .noise and duplicating using .tranoise Yes, that's sometimes EXACT and sometimes up to maximum of 2% just depends on that run. but, hey, this is noise and noise doesn't repeat.
So the MC1496 is, like, 100 MB of text? Wow, that must have been a lot of typing. Or is that all PWL models of a given run of noise?
Cheers
Phil Hobbs
--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics
160 North State Road #203
Briarcliff Manor NY 10510
hobbs at electrooptical dot net
http://electrooptical.net
this MUST be set: .options plotwinsize=0 that gives you uncompressed 'results' to get white() to work well you MUST set the maximum step of interest to
1/10th what you're really looking for so that the filtering makes the spectrum a bit flatter, still be slightly less than 1 dB tilt, but acceptable, because after all this is statistics.
Yes, big. No, very little typing. Autogenerated. But be better *if* could get inside LTspice. Plus, once created the files under certain conditions are reuseable. I only have to create 'sets' of indepedent ones once. It's like calling up subcircuits and each call bumps the number by one. as in file001, file002, file003, file004, ....
Right now I'm trying to find out why I get 20 times the noise density that I expected. 2 times I can understand with a 6 dB mixer loss, but the other factor of 10 ?? Do you have any tutorials on mixer modulation/demodulation noise, or know any URLs ?
Hmmm. I did notice that the noise from a resistor looks like a smooth wide 'blur' along the time waveform
The output from an OpAmp that had the 1/f noise also has a blur line, but due to the 1/f looks more like there are also those 'meandering' DC levels, It does look just like what you see on a scope trace at the output of an OpAmp. Or, on a data sheet for the low frequency time waveforms.
Maybe simply characterizing with 1/f noise that goes all the way to the lowest frequency of interest is enough to represent flicker noise. Then, if 1/f is identical to flicker noise, I have it represented also.
y
e
100 MHz is not "microwave." When sine wave drive is done at higher frequen cies, it is at least partially done for reasons of dispersion. They would do square wave drive if they could (felt confident of generating and mainta ining the wave shape.) At low frequencies, a good square wave drive is usu ally achievable and desirable.
.
Can you recommend a DB Mixer that has great carrier suppression to operate between ...1MHz to 100MHz? Be best if pure linear to remove EVERY unwanted harmonic, but not likely to find that one.
If not as a mixer component, what about transistor arrays that can be adapted? And, any experience in getting great balance there?
That's wrong! The IC solution (six-transistor ensemble) does depend somewhat on matching transistors, but so do MiniCircuits (matched diodes). The linearity is dependent on the signal AMPLITUDE; if you keep the transistors out of the switching regime, it gets better the lower the input voltage swings. In analyzing the Gilbert-cell type mixer, inputs A and B determine a smooth, continuous function output C, so it can be expanded in series
By balancing and differencing one eliminates the constant term, and the first-degree A, B terms; A^2 and B^2 have the wrong symmetry, so they vanish, too. The leading term that DOES NOT VANISH is A*B, and that term will always dominate at sufficiently small A, B values.
Well I am sorry, but I am not sure if you need "great suppression." I've b een working on the receiver side for more than a decade, so I don't do the IQ thing anymore (assuming that is why you need it)---it is usually bandpas s sampling for RX'er's these days, and OFDM tolerates carrier leakage prett y well anyway.
The last Mini Circuits mixer I used (a couple years ago) was the ADE-1MH+, but I think you can probably get something better centered (for 1-100) if y ou look at the catalog. It isn't clear on their specs that they segregate isolation from imbalance regarding the LO to other port isolation. I am pr etty sure these two things are conflated in the isolation spec. Vague. :-(
I question your desire to avoid the conversion at the LO harmonics ("pure l inear") in any case. That stuff is easy to filter out.
I suppose so. The HFA3101 is an array, just already hooked up in gilbert c ell form. I mean, Harris/Intersil makes a bunch of UHF transistor arrays, and the HFA3101 may well be the same process as the rest. I haven't used i t, but have been aware of it since around the turn of the century. Just pa ssing it on...
The last experience I had with active IQ compensation/balance was more than a decade ago and at 374 MHz with proprietary ICs for WLAN. I was able to get carrier suppression down to about 45 dBc, but the limiting factor there was running out of bits, not isolation or what the analog chip could have otherwise done. Good enough for OFDM. These nulls are obviously very sens itive and thus stability is an issue.
I once did analog compensation for a 1496 IC, but it was only at 455 kHz ou tput. lol. But I could get > 80 dBc.
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.