Am 01.06.2016 um 01:39 schrieb John Larkin:
Look at the SY58051U also. (Micrel, or whoever owns them today.)
The Johnson noise will drown in the noise of the tail current source. The two transistors are only similar. And you get the noise of the 2 emitter followers that you really do not need.
Yes.
Also interesting: <
regards, Gerhard
If Johnson noise in the mixer is the limiting problem, that method won't help, ISTM. High dV/dphi and low noise is the simpler approach.
Cheers
Phil Hobbs
Hah, send them into laser diodes, mix 'em with beam splitters and send 'em into a fast photodiode.
(someone has been asking about phase locking LD's so I've thinking about that...) George H.
Am 01.06.2016 um 03:51 schrieb Phil Hobbs:
won't help, ISTM. High dV/dphi and low noise is the simpler approach.
But it isn't. We have nearly Volts at the input of the mixer. There is not much available that has less noise than transformers and Schottky diodes. The noise of the Schottkies is even only half-thermal. The real problem is limiting the bandwidth against aliasing. Do not drive the first mixer too hard.
There is a whole industry built on dual mixer systems.
Gerhard
Laser diodes are noisy!
-- John Larkin Highland Technology, Inc lunatic fringe electronics
If you roll back to the start of this thread, you'll notice that John turned down this idea at the very beginning.
-- -TV
Am 01.06.2016 um 09:06 schrieb Tauno Voipio:
oh, sorry, that was gone already locally.
Well, this then is just like High End Audio where they deny that feedback could ever work and then claim that audio is the hardest thing in engineering.
I mean, if it it works easily for two atomic clocks, it should be good enough for what is probably a noisy
155 Mb/s plastic fiber optic link.regards, Gerhard
on the few occasions that I tried foxhunting, I just switched to 3rd, then 5th harmonic of the fundamental as soon as my poor receiver was overloading with the fundamental signal. Every TX makes some signal at odd harmonics. Of course wide RX frontend is a no-no with this approach. I was the winner on the first attempt (then I got bored quickly).
Frank IZ8DWF (sorry for the OT)
Right - maybe John wants thick golden speaker cables.
-- Gruesse, Tauno
I don't think it's quite that simple. The Fourier series for a square wave dies off as 1/f, and has only odd harmonics, so the noise gain vs. the sinusoidal case isn't very large. Since you're LP filtering the output, you only see noise in narrow passbands centred on the harmonics, and the sensitivity goes as 1/N:
noise gain = sum (n=1,2,....) 1/(2n+1)**2
= sum (n=1,2,....) (1/n**2 - 1/(2n)**2)
= 0.75 * pi**2/6 = 0.9 dB.
That's well within the range of performance differences in a single mixer type, and certainly within the range of different types.
In the pure weak-additive-noise case, the phase noise spectrum depends only on the carrier-to-noise ratio:
delta phi = 1/sqrt(2*CNR) .
This is true regardless of whether you've mixed down to a low IF or not. Also, the shape of the phase noise spectrum is identical to that of the additive noise. So in this case the LO-plus-2-mixer trick is bound to make things worse and not better, unless there's a serious difference in the performance of available phase detectors at high vs low frequency.
If the dominant noise source is an irreducible amount of *time* jitter at the input of the phase detector, then you win by mixing down, because a picosecond is 100 ppm at 100 MHz but only 100 ppb at 100 kHz. However I don't think that's the usual case, and it doesn't sound like John's in that position.
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
Continuous-multiplier mixers based on Gilbert-cell topologies that are not hard-swiched tend to be very noisy. It is really not easy to make a multiplier that is as quiet as a switching mixer. It might be easier to imagine it with the LO at, say 1Hz. When the (differential) LO signal is going through zero, each input current is equally divided between two transistors, and even if there were no noise in the input current coming from the gm stage, there would be noise in each current coming out of the current-steering stages. That problem essentially doesn't occur with a switching mixer, except when the (differential) LO signal is going through zero. That is why for low noise (and good linearity) you want to drive the LO port with a nice crisp square wave, not a sine wave. Many people don't believe this but it is fairly straightforward to measure. (These sorts of things are hard to simulate properly without SpectreRF or quite a bit of careful thought.)
It's not 155.52 and it's kilometers of singlemode fiber, but the problem is essentially the same. I want to characterize the link phase noise spectrum down to -180 dBc if possible, and I need a really good phase detector to do that. The DMTD idea will just add more mixer phase noise to the problem.
I'm thinking two fast XOR gates, each driving a diode current-steering network and amp/filter, then dual ADCs and some FFT/correlation math. Averaging washes out the mixer noise.
DMTD does a lot of averaging, too.
-- John Larkin Highland Technology, Inc lunatic fringe electronics
That aligns with my thinking: two ECL square waves with really fast edges over-driving a Gilbert multiplier, or an XOR over-driving a differential pair or some current-steering diodes. The transistors have base-current shot noise, but that's pretty far down.
Transistor Ft has to be really high, 10 GHz or so, to keep the base currents down. That's why diode current steering is interesting.
I want to graph phase noise vs frequency, out to 10s of KHz, so a low IF, or any IF, doesn't work. A single mixer looks best.
-- John Larkin Highland Technology, Inc lunatic fringe electronics
Sounds right. Adding the phase noise of multiple mixers can't help.
Multiple mixers help if they are used in parallel. I could literally parallel four mixers (XOR or Gilberts have fairly high input impedances, so string them along differential transmission lines) and sum their outputs, and gain 6 dB s/n. But I may as well digitize all four outputs. Then I can sum digitally, or do correlation tricks to get way better than 6 dB. There are commercial phase-noise measurement systems that do correlation of two mixer outputs, so it would only take a really smart math guy to extend that to four mixers. I have access to really smart math guys.
-- John Larkin Highland Technology, Inc lunatic fringe electronics
Am 01.06.2016 um 16:37 schrieb John Larkin:
Can't you lend a E5052B or an R&S FSUP if it's only for characterization? They also run out of steam at -180dBc, but there has already gone a lot of verfication work into them.
In this league you must have everything twice including the reference oscillators and include everything in the three cornered hat / cross FFT.
There are surprises like correlated thermal noise that comes in antiphase out of a power divider and that happily averages away, but the true result is much worse.
I'm also interested in building sth. simple for this purpose, probably in Wenzel's down mixer style, but in stereo with
2 references and using an Agilent 89441A to do the cross-FFT stuff. For fun & self education, but on the job there is a E5052B for reality checks.A cheaper alternative might be a Microsemi/Symmetricom Timepod.
regards, Gerhard
'tain't that hard. With N mixers, you get (N**2-N )/2 independent cross-correlations, so asymptotically your noise power goes down by
1/N**2. That's pretty favourable compared with straight averaging, which goes as 1/N.That's how you measure really low noise levels as well--multiple FET buffers, which don't contribute current noise (which wouldn't be uncorrelated like the voltage 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
Am 01.06.2016 um 17:39 schrieb John Larkin:
Hey, hardware multiplication is MY fetisch :-) I did it with 20 ADA4898 opamps to get to 220 pV/sqrt Hz, built but not yet verified with 8 Interfet IF3601 (should be somewhat better and avoid those huge capacitors at the input) and already made the layout for 1:8 dividers / combiners and ring mixers with MAcom baluns and Avago low 1/f schottkys.
Yes, that reduces the averaging time until you hit the limit. Thermal noise is hard to beat.
Gerhard
Given two mixers with the same input signals, I guess you'd digitize the output of both and do the complex FFT of each. Then at each spectral point, do the dot product if the vectors and sum the result into a third, scalar array. Do that lots of times, and the correlated noise accululates in the scalar array but the random (mixer/amp/ADC) noise doesn't.
For multipler mixers, just pick all possible pairs and sum the dot products into that same single scalar array. The result is eventually the phase noise spectrum. Something like that.
-- John Larkin Highland Technology, Inc lunatic fringe electronics
Yeah, youd' have to lock the laser to something.. too fussy. This is probably a stupid idea, but you could send both signals into the same laser diode.. and mix the result in the photodiode. (I think that works..?) Most likely horrible IMD.
George H.
It isn't hard to show analytically, either. An easy-to-remember rule of thumb is that if you feed a BJT diff pair a current with full shot noise, the collector currents each have exactly full shot noise regardless of delta-V_BE. (This assumes infinite beta and zero R_ee'.) You can figure out other cases from that one.
Plus there's the noise of the base resistance. Both of those go away if you use a quiet bias current and switch hard.
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
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.