Practical Analyzer series

I didn't say that there were no possible applications of averaging in phase noise measurement, I just said it was nonstationary, so that trying to average the phase deviation over longer times makes it worse, which it does.

If the phase noise were really Brownian, it would have stationary increments, and you could average those. Phase locking it to a stable source and measuring the phase deviation with respect to time is another method. I'm sure that (as you say) there are others.

The phase deviation among an ensemble of N identical oscillators running in identical conditions will increase without bound as t->infinity. Thus the expectation value of the phase difference is not constant with time, i.e. the phase statistics are not stationary.

You're quite right in saying that that doesn't make it impossible to measure. However, the point I was making in the post you took issue with is that LO noise corrupts spectrum analyzer measurements in a way that is very difficult to compensate for afterwards.

You seem to be saying that all you need to make a good measurement of a

1-Hz wide signal is an IF filter that's narrower than that, even if the LO is some gross noisy mess, and that isn't true. The narrow IF will just give you an exquisitely detailed picture of how badly the LO noise smeared out the input signal.

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 USA 
+1 845 480 2058 

hobbs at electrooptical dot net 
http://electrooptical.net

Sure, but we're not talking about the phase differences themselves, but rather their spectrum. That's what I mean by "stationary" spectral statistics. The spectrum itself is time-invariant, at least until I trip over a cable.

Yep, which is why the averaging techniques involve multiple uncorrelated analyzers. Not much you can do to remove the effects of PN from a conventional SA measurement.

Right, I think we may have conflated two different points. A tracking application doesn't care about the signal, only the DUT's transfer function. The stimulus signal just has to be able to show up -- eventually -- in all of the FFT bins surrounding the passband of the device being measured. In principle it doesn't matter whether the stimulus that finds its way through the DUT and into those bins is "noise" or "signal."

-- john, KE5FX

Sure. That's the same sort of method as the two-MOSFET trick for measuring small noise voltages. Making more measurements than that allows the use of closure, which gives an internal error estimate as well.

That assumes that the digitizer clock is sufficiently stable, though, and that there's no other oscillator in the frequency chain, i.e. that you're digitizing the RF directly. Clock instability is exactly the same sort of problem as the LO stability of a normal spectrum analyzer. Otherwise you might just as well use a noise source instead of a tracking generator.

Using some gross smeary oscillator to test a narrow crystal filter is okay if the RX side is stable, as you describe, but if not, it has exactly the same problem as a normal spectrum analyzer.

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 USA 
+1 845 480 2058 

hobbs at electrooptical dot net 
http://electrooptical.net

That's exactly what I was doing in the 1970's with an HP141T and HP8553L spectrum analyzer and a Wavetek something sweep generator. No tracking generator anywhere in sight. The Wavetek was set to sweep asynchronously across the filter bandpass at some modest rate (about 1 sweep per second) while the spectrum analyzer slowly moved across the same filter bandpass. The minimum IF bandwidth was only 30Hz which was marginal, but still useful. We had a peak-hold feature of sorts as the HP141T mainframe was a storage scope. If I waited long enough, and the HP8553L didn't drift out of lock, I would get a tolerable response filter curve even though the Wavetek sweeper was a very noisy signal source.

Roll forward 40 years, and I'm still using the same boat anchors and methods. Sigh.

--
Jeff Liebermann     jeffl@cruzio.com 
150 Felker St #D    http://www.LearnByDestroying.com 
Santa Cruz CA 95060 http://802.11junk.com 
Skype: JeffLiebermann     AE6KS    831-336-2558

On a sunny day (Sun, 24 Feb 2013 17:12:03 -0500) it happened Phil Hobbs wrote in :

OK, let me try to explain where I want to go with that idea, _not_ _averaging_.

Say you measure a carrier with unknown phase noise both of the own VCO and of that carrier. You do it by taking a n point FFT every x milliseconds. This results in a series of frequency spectra, in my case one every 25 mS, due to phase noise from both the signal and the VCO the spectral component is in a different place after each transform, here a representation:

signal |

-------------------------------- spectrum 1 t0

|

-------------------------------- spectrum 2 t1

|

-------------------------------- spectrum 3 t2

_ | |

-------------------------------- resulting display (for eye)

So we don't know exactly what part of this 'band' is caused by the signal, and what is caused by the VCO instability.

Now I add a rubidium generated stable frequency, 'marker': marker | | |

-------------------------------- spectrum 1 t0

| | |

-------------------------------- spectrum 2 t1

| | |

-------------------------------- spectrum 3 t2

_ _ | | | | | |

-------------------------------- resulting display (for eye)

Now the trick, before displaying each of these spectra I align the marker so those 'overlap' in each display (frame):

| | |

-------------------------------- spectrum 1 t0

| | |

-------------------------------- spectrum 2 t1

| | |

-------------------------------- spectrum 3 t2

A | || |

-------------------------------- resulting display (for eye)

Now what you see, is the signal in A is displayed with the added phase noise of the reference marker, the phase noise of the VCO is canceled. As the Fourier transformed signal is just a one dimensional array here, a simple pointer change will accomplish this picture shift, the 'search' for the marker only needs to happen in a small part of the array, to find a maximum, all very fast, little overhead. You must have seen those programs for video stabilization, picture taken from a gondola in Venice Italy moving up, down left right, pan, tilt, run it through such a program and it becomes as taken with the camera on a big

*boat anchor*. have fun:-)

Hereby donate this in the public domain. Keywords: patent, phase noise, FFT, spectrum analyzer, elimination, prior art, Copyright Jan Panteltje 2013 Mon Feb 25 10:41:54 CET 2013

bs

our

re's a

nd

ESTRIAL RF TUNER':

                         min    typ      max            2600            3900  MHz             2              48         64            1700  MHz   10           20000  Hz              -29           dBc                       -80           dBc/Hz                       -125           dBc/Hz         64             255

RF),

It wasn't too bad, but several locations, one sea-side (but in a closed closet). I'm hoping it isn't salt-rot. More recently it has been living sealed in a cardboard box in the basement.

Poking around the ham pages, worst-case is a gassy CRT.

I'm chasing a nasty SMPS RFI problem, so I whipped out my homemade backup analyzer. Strictly bush-league performance--it uses a CATV tuner front-end--but surprisingly helpful.

--
Cheers, 
James Arthur

I was calculating the spot noise (specified by Elonics at 10 kHz) and extrapolating down to 1 kHz by taking the three-halves power. That's why I got (very roughly) 60 dB, which I think is correct.

With a 500 MHz f_RF, the 8568B hits the additive noise floor at about 1 kHz offset, because it has about a 30 dB noise figure, typical of most spectrum analyzers. From a specsmanship POV the Elonics looks better than it really is. The integrated noise over a wide band is not nearly as important to me as the close-in phase noise, which is why I ignored the wideband spec altogether.

The 8568B blows all modern entry-level spectrum analyzers right into the bushes--I was considering mortgaging my firstborn and getting a Tek MDO4104-6 mixed-signal scope and 6 GHz SA, until I looked at the noise specs. (I wish that the modern ones were better, because it's a pain keeping boat anchors running, but they aren't.)

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

Ten years from now try and keep one of the newer instruments running. It'll be a much different story and you will wish you still had your 8568B.

a

RF TUNER':

max

3900 MHz 48 1700 MHz 20000 Hz

dBc

dBc/Hz

dBc/Hz

255

There is a glut of analog CATV SLMs on Ebay, and some go down to 5 MHz for about $25 if you keep an eye on the listings.

I bought a Texscan 'Spectrum 700' that covers 5MHz to 600 MHZ in six bands and will measure signal levels from -30 dBmv to =60 dBmv.

RF TUNER':

max

3900 MHz 48 1700 MHz 20000 Hz

dBc

dBc/Hz

dBc/Hz

255

It can come with a time lag. I was (unpleasantly) surprised how much of our stuff developed nasty corrosion and also defects after just one international move across an ocean. Everything was in a sea container but that didn't protect it enough.

If it is, contact the guys at Signalhound. They started their business not with their own spectrum analyzers but with LCD replacement screens for HP analyzers and such.

McGyver type tools are often the best. Or as they say, a kludge lasts the longest.

--
Regards, Joerg 

http://www.analogconsultants.com/

Then it'll repeatedly say "An irrecoverable error has occurred and the instrument will be shut down" ... and that's it. Oh, and of course it'll also be "out of support".

--
Regards, Joerg 

http://www.analogconsultants.com/

Hobbs

of your

there's a

gh-end

TERRESTRIAL RF TUNER':

                           min    typ      max              2600            3900  MHz               2              48           64            1700  MHz

te8)   10           20000  Hz

               -29           dBc                         -80           dBc /Hz                         -125           dB c/Hz           64             255

z f_RF),

s

ix

Hey, cool. That tip led to all sorts of eBay treasure. Surplus is wonderful.

--
Cheers, 
James Arthur

obbs

your

here's a

-end

RRESTRIAL RF TUNER':

                         min    typ      max              2600            3900  MHz             2              48           64            1700  MHz 8)   10           20000  Hz              -29           dBc                       -80           dBc/Hz                       -125           dBc/Hz           64             255

f_RF),

s

I wondered how those work. I'm not sure how a generic LCD could replace & interface with the HP-141T's variable-persistance storage function. (A digital replacement would be nice though.)

I found the problem. A shielded inductor was inducing 60mV in the (broken) ground "plane". The input was bypassed to that, imposing a

60mV 500KHz _square_wave on the d.c. input. Nasty. Actually finding it wasn't so bad as fixing it with no "ground" in sight--everything was bouncing.

Copper foil'd a proper ground without big currents flowing in it, added ferrite + bypasses. Fixed.

--
Cheers, 
James Arthur

a

RF TUNER':

max

3900 MHz 48 1700 MHz 20000 Hz

dBc

dBc/Hz

dBc/Hz

255

Well, that one may be a tad too old but it can't hurt to ask. It's just one email.

A floating ground? It still beats me why people design like that. OTOH this also brings me work and income 8-)

--
Regards, Joerg 

http://www.analogconsultants.com/

Hobbs

your

there's a

high-end

TERRESTRIAL RF TUNER':

typ max

3900 MHz 48 1700 MHz 20000 Hz

dBc

-80 dBc/Hz

-125 dBc/Hz

255

f_RF),

You're welcome. With a few mods, you can connect a scope . A very few modles had a video output to use it as a demod in the headend to look at video quality. Another interesting surplus item is the 'Agile Demodulator' which could feed one of those SDR for an improved SA by connecting it at the IF loopthrough.