A measurement of 10 MHz down to 10 uHz

[John Heath]

Hi Guys

To sum up .1 ppb 1700 bucks 10 ppb 300 bucks . and secondly the equipment I am using is not up the the task of 10 uHz at 10 MHz.

I am going to use a heterodyne strategy to reduce the resolution of a 1X10^-13 test down to 1X10^-7 test. Keep in mind I am not sure , only guessing. Any opinions on this , thumbs up thumbs down , would be much appreciated.

!] The rate of change from 0 to 5 K volt frame of reference is 10 seconds so stability down to 10^-13 only needs to last for 10 seconds
 
2] The two 10 MHz reference oscillators are off by 40 Hz then heterodyne together for a beat frequency of 40 Hz. What was 10 Mhz to 10 uHz test is now a 40 Hz to 10 uHz test or 10,000,000.000,010 test to 40.000,010 test.

The burning question is will this work. Theory one thing but application another. Is this a pipe dream or could it work. The heterodyne mixer is a XOR  gate followed by a R/C low pass filter of 20 K resistor and .1 uf condenser to ground. So far test results really suck with with too much noise reducing measurements range to 10 mHz. It has to improve by a factor of 1000 for 10 uHz.

Best guess. Is there any point in continuing with the 40 Hz heterodyne strategy?

[dannyf]

2] The two 10 MHz reference oscillators are off by 40 Hz then heterodyne together for a beat frequency of 40 Hz. What was 10 Mhz to 10 uHz test is now a 40 Hz to 10 uHz test or 10,000,000.000,010 test to 40.000,010 test.

You are measuring phase differentials - see my post on this earlier - not frequency differentials. those ppb/ppm figures are for long-term frequency stability, not phase stability.

Is this a pipe dream or could it work.

Scientific research is not just trial-and-error. You have to have a theory on why you should expect some differences and then you can design an experiment to show such differences.

[uncle_bob]

Hi

The proper term for the heterodyne you are attempting is a single mixer time difference system. A somewhat better approach is a dual mixer time difference system. If you dig into the NIST archives, they have a number of papers on all this.

The weakness of either approach is the limiter that turns the sine wave out of the mixer into a square wave in the counter (or external to it). Noise in this process quickly overwhelms the measurement accuracy.

If you want to go this way, you probably would do better to design, build, and validate a DMTD system.

Bob

[John Heath]

2] The two 10 MHz reference oscillators are off by 40 Hz then heterodyne together for a beat frequency of 40 Hz. What was 10 Mhz to 10 uHz test is now a 40 Hz to 10 uHz test or 10,000,000.000,010 test to 40.000,010 test.

You are measuring phase differentials - see my post on this earlier - not frequency differentials. those ppb/ppm figures are for long-term frequency stability, not phase stability.

Is this a pipe dream or could it work.

Scientific research is not just trial-and-error. You have to have a theory on why you should expect some differences and then you can design an experiment to show such differences.

Hi Danny

In a mood to write so apologies in advance for longer than necessary answers.

As to frequency vs phase measurement. I hear you. This brings to mind Edwin H. Armstrong inventor of FM radio. His first commercial FM broad cast station starts as a phase modulated signal. This was to have better control modulation by changing the amplitude of 0 phase vs 90 phase . This would then go through several resonate frequency multiplying stages with the end result being a 47 MHz frequency modulated broadcast . This would later become television IF and FM radio pushed up to 100 MHz. From this we could say that the difference between phase modulation and frequency modulation  depends only on the period of time the measurement is made. A measurement of 1 cps made in .1 seconds would be a phase difference. A measurement of 1 cps over 10 seconds would be a frequency measurement. A frequency difference measurement to resolve 10 MHz down to 10 uHz would require a measuring time period of 2 days to qualify as a frequency measurement as you rightfully pointed out.

This being said it should be noted that frequency counters have change as technology has advanced. My frequency counter says 400 Hz is 400.000138 Hz with a sample time period of 1 second. How is this possible? Old school frequency counter would say 400 Hz. New frequency counters take advantage of microprocessors to combining both frequency and period measurements and combine these to measurements for best result resolution. The end result is constant resolution regardless of frequency therefore 400 Hz measured over a period of 1 second reads as 400.000138 Hz not 400 Hz. That particular measurement required a 100 second averaging so the real number number is 400.0001 Hz and fluctuating up and down in the 1 mHz range. So close and yet so far.


As to theory and why I would expect an oscillator to have a voltage coefficient frame of reference. When developing a theory there comes a time when the theory starts to write itself. At this point you find yourself in the back seat while the theory drives the car. I have come to personify this theory as Mr VEPS and I can say that his driving skills leave room for improvement , somewhat like a bull in a china shop. However after the pieces are glued back together he is within the limits of empirical evidence.  In short it is out of my hands as I am in the back seat at the moment. He is insisting that relative size of a system not the relative inertia will cause time to dilate and that the relative size is changed by charge parity. The only way to remove Mr VEPS from the drivers seat is to test for a small change in time caused by a change in voltage framer of reference , 5 K volts. If it has a negative result I can kick him out of the drivers seat and take back control. If the results are positive I will be stuck in the back seat again. The back seat is not as bad as it sounds. Just sit back and enjoy the ride with no responsibilities.

In summary new frequency counters appear to be within range for this measurement , just in range. The real problem is jitter , hydro hum , IC noise , 5 volt regulators and probably a few more that I am not aware of. And the theory that makes this prediction , bad driver Mr VEPS , is also in place. All that is left is a solid measurement for a thumbs up or thumbs down.

 

[John Heath]

Hi

The proper term for the heterodyne you are attempting is a single mixer time difference system. A somewhat better approach is a dual mixer time difference system. If you dig into the NIST archives, they have a number of papers on all this.

The weakness of either approach is the limiter that turns the sine wave out of the mixer into a square wave in the counter (or external to it). Noise in this process quickly overwhelms the measurement accuracy.

If you want to go this way, you probably would do better to design, build, and validate a DMTD system.

Bob

Hi Bob

I am in a leaning curve for precise frequency measur4ement as you can tell so I appreciate you taking the time to help myself through this. I can confirm you position that a sine wave is better than a square wave. My main frequency counter comes with a biult in oscillospope to eye ball the input. If I over drive a sine wave into a square wave the readings become unstable in the mHz range. If I back off the gain to the origenal sine wave the rerasding are more stable. This seems counter intuitive but test results comfirm your position that sine wave is better than overr driven square wave. This is good to know but I feel a little uncomfortable not knowing why. Not knowing why will always bite you in the end if it is not nipped in the bud. e a sine wave .  If you could expand on why a over driven sine to square wave is less stable. From my shoes a clean and decisive high slew rate up and down should be more stable and yet it is less stable as you said?

As to DMTD . This one googled up nicely. I found a nice link that included a diagram. Somehow diagrams get the point across in a way that words can fail. A common language understood in all languages.

Link to DMTD

http://www.wriley.com/A%20Small%20DMTD%20System.pdf

The money shot , the diagram.

[John Heath]

Hi Bob again

Still getting use to picture attachment that caused a premature posting.

As you can see from the diagram it is a isolation transformer with a full bridge rectifier followed by band pass filters for side band analysis. As my coupling is fiber the isolation is already there. A 1n GHz full bridge diode mixer , TUF R3SM , is over kill so I should get by with a 74hc4066 quad FET switch driven by the reference oscillator as a full bridge mixer. Maybe another 6 to 12 DB reduction in noise?? I can hear you banging your fist on the table as I speak these words. I know it is a reference oscillator square wave that spells trouble from your post. Then again is there a difference between this DMTD class D chopping a bridge rectifier mixer and my proposed 74hc4066 bridge mixer? It will only take a day to prototype it and give it a run.

Not sure what a dual mixer time difference system is without digging into the NIST archives. Will speak to it later.

A plan B which deals with software interface4 with GPS satellites. If a corona bust comes from out sun the earth it will be sprayed with electrons and protons. Our earth's magnetic field will cause electrons to go right and protons  to go left. This means GPS satellites on the right will be in a positive voltage framer of reference and left in negative. As a GPS satellites receiver will average out information from all satellites it can receive it will make a best guess from data it has to my position. If I had access to individual GPS satellite gdata as to my distance from that satellite then a very precise assessment can be made of any change to time dilation of that satellite by comparing east to west GPS rendering of my position. Not sure if my meaning survived the translations into words. To state this in another way. If all east satellites say I am 2 feet east off my real position and all west satellites say I am 2 feet west of my true position when a corona bust from the sun hits the earth then there is a second confirmation that voltage frame of reference will change time dilation. It is for this reason that I have a keen interest in software interface with GPS satellites to gain access to this information. Is there a way to receive raw untouched information from GPS satellites and display it on a computer screen??

[uncle_bob]

If all east satellites say I am 2 feet east off my real position and all west satellites say I am 2 feet west of my true position when a corona bust from the sun hits the earth then there is a second confirmation that voltage frame of reference will change time dilation. It is for this reason that I have a keen interest in software interface with GPS satellites to gain access to this information. Is there a way to receive raw untouched information from GPS satellites and display it on a computer screen??

Hi

Sure you can. As long as your desk is above the ionosphere you can get things done pretty easily. The same solar burst that gives you the sudden electron flood bumps the ionosphere and you get a delta delay of (maybe) 50 ns. At 1 ns per foot, that is about a 50' jump. Indeed you can take that down to 10 or 20 ns, but doing so involves processes that would take out the GPS jump.

Bob

[uncle_bob]

If you could expand on why a over driven sine to square wave is less stable. From my shoes a clean and decisive high slew rate up and down should be more stable and yet it is less stable as you said?

Hi

If you are an IEEE member, there are a few papers behind their pay wall that dig into all the details of turning slow sine waves into square waves. The square wave is indeed more stable into the counter than the sine wave. The details are all in how you turn the sine wave into a square. With a "high gain" down converter, the sine waves will always be slow / low frequency. Limiter design is not simple.

Drive into a good mixer (diode ring) does not really matter sine vs square. If both are done correctly they can work equally well. Again the key word is correctly.

To get to the levels you are after, JPL can barely do it with all of the right parts. Substituting other "stuff" is likely to degrade things by 10 to 10,000X. There really isn't a shortcut when it comes to the parts you need to use.

Looked at like a radio:

You are trying to receive a 2 KHz wide sideband signal at 0.05uV and get 20 db s/n.  You can indeed do it. Doing it with just any radio and any preamp hooked up any which way.... not going to happen.

These are not toss it together measurements. They require either high quality commercial gear or stuff built with a through understanding of all the issues.

Bob

[John Heath]

If all east satellites say I am 2 feet east off my real position and all west satellites say I am 2 feet west of my true position when a corona bust from the sun hits the earth then there is a second confirmation that voltage frame of reference will change time dilation. It is for this reason that I have a keen interest in software interface with GPS satellites to gain access to this information. Is there a way to receive raw untouched information from GPS satellites and display it on a computer screen??

Hi

Sure you can. As long as your desk is above the ionosphere you can get things done pretty easily. The same solar burst that gives you the sudden electron flood bumps the ionosphere and you get a delta delay of (maybe) 50 ns. At 1 ns per foot, that is about a 50' jump. Indeed you can take that down to 10 or 20 ns, but doing so involves processes that would take out the GPS jump.

Bob

Yes I see your point. From what I google changes in the ionosphere alter the path the RF signal. Putting my lab bench above the ionosphere would exceed my budget , ha.  Thanks for the input.

I will post a picture of my super frequency counter. It is free to anyone with a android phone. Would like to know what you think of it.

     

[DimitriP]

I'm partial to the Function Generator ...

[jpb]

If you could expand on why a over driven sine to square wave is less stable. From my shoes a clean and decisive high slew rate up and down should be more stable and yet it is less stable as you said?

Hi

If you are an IEEE member, there are a few papers behind their pay wall that dig into all the details of turning slow sine waves into square waves. The square wave is indeed more stable into the counter than the sine wave. The details are all in how you turn the sine wave into a square. With a "high gain" down converter, the sine waves will always be slow / low frequency. Limiter design is not simple.


Bob

I am trying to improve my Alan Deviation measurements with such an approach and the paper that I've found most useful (because it is simple) is

ZERO-CROSSING DETECTOR
WITH SUB-MICROSECOND
JITTER AND CROSSTALK

http://www.dtic.mil/dtic/tr/fulltext/u2/a515384.pdf

I've not built anything yet so I can't comment on difficulties, but my direct measurements of the output of a mixer and the DDS output of an Agilent 33522A with the same 10MHz as a reference input are in good agreement with a crude calculation based on slope and typical counter noise of 300-500uV. The sin output is around 130mV peak at 1Hz so the slope is around 0.8V/s so noise of say 500uV gives an uncertainty of 6E-4 seconds and, surprisingly to me, my measurements on my cheap counter are slightly better than this but of the right order of magnitude (noise floor around 1E-11 i.e. 1E-7 gain in going from 10MHz to 1Hz and then 1E-4 measurement noise in measuring the 1 second period).

My plan is to follow a similar scheme to the paper and try and improve the slope by a large factor without increasing the noise significantly. I guess that I will discover all the other noise sources in the system on the way but it will be interesting to find out.

[PointyOintment]

The Faraday cage at 5 K volts is offsetting charge parity therefore the clock should change in frequency.

I would think this would require all components of the oscillator, or at least the crystal element, to be at 5 kV as well as the cage. Not being directly connected to the cage will cause their voltages to not track the cage's voltage well, I think.

I will post a picture of my super frequency counter. It is free to anyone with a android phone. Would like to know what you think of it.

Infuriatingly skeuomorphic.

[John Heath]

I'm partial to the Function Generator ...

Ya his function generator is a step above as well. He seems to raise the bar with everything he touches. His name is Dr Owen Thomas from the UK. The days when PC virtual test equipment is considered to be a cheap substitute could be coming to an end. Much like digital cameras took over film with the purest and their chemicals kicking and screaming all the way.

[John Heath]

The Faraday cage at 5 K volts is offsetting charge parity therefore the clock should change in frequency.

I would think this would require all components of the oscillator, or at least the crystal element, to be at 5 kV as well as the cage. Not being directly connected to the cage will cause their voltages to not track the cage's voltage well, I think.

I will post a picture of my super frequency counter. It is free to anyone with a android phone. Would like to know what you think of it.

The oscillator is hermetically sealed in a metal case that is grounded to the 5 K volt source. There could be some voltage gradients do to the coarse 1/2 inch spacing of the Faraday cage for 7805 regulator?? as well as the fiber cable leaving the cage.

Infuriatingly skeuomorphic.

 It tickles the ear to hear the English language well executed. Nicely done. As Bob Dylan said " The Times They Are a-Changin". Virtual test equipment is our future. The notion that we would solder components together instead of using a simulator could be a source of curiosity 50 years from now. Some what shivering in the cold when all that needs to be done is light a fire. 

[John Heath]

If you could expand on why a over driven sine to square wave is less stable. From my shoes a clean and decisive high slew rate up and down should be more stable and yet it is less stable as you said?

Hi

If you are an IEEE member, there are a few papers behind their pay wall that dig into all the details of turning slow sine waves into square waves. The square wave is indeed more stable into the counter than the sine wave. The details are all in how you turn the sine wave into a square. With a "high gain" down converter, the sine waves will always be slow / low frequency. Limiter design is not simple.


Bob

I am trying to improve my Alan Deviation measurements with such an approach and the paper that I've found most useful (because it is simple) is

ZERO-CROSSING DETECTOR
WITH SUB-MICROSECOND
JITTER AND CROSSTALK

http://www.dtic.mil/dtic/tr/fulltext/u2/a515384.pdf

I've not built anything yet so I can't comment on difficulties, but my direct measurements of the output of a mixer and the DDS output of an Agilent 33522A with the same 10MHz as a reference input are in good agreement with a crude calculation based on slope and typical counter noise of 300-500uV. The sin output is around 130mV peak at 1Hz so the slope is around 0.8V/s so noise of say 500uV gives an uncertainty of 6E-4 seconds and, surprisingly to me, my measurements on my cheap counter are slightly better than this but of the right order of magnitude (noise floor around 1E-11 i.e. 1E-7 gain in going from 10MHz to 1Hz and then 1E-4 measurement noise in measuring the 1 second period).

My plan is to follow a similar scheme to the paper and try and improve the slope by a large factor without increasing the noise significantly. I guess that I will discover all the other noise sources in the system on the way but it will be interesting to find out.

I saw a clever design in a electronic organ. Take the frequency of interest and use it as a clock driver for a bucket brigade delay line , comb filter. This indeed will get us into trouble if the signal to noise is too high breaking down to chaos. On the bright side if the signal to noise is small enough the signal to noise will be much improved. A balancing act on the bubble between chaos and improved signal to noise. Have a diagram of it. 

[uncle_bob]

I am trying to improve my Alan Deviation measurements with such an approach and the paper that I've found most useful (because it is simple) is

ZERO-CROSSING DETECTOR
WITH SUB-MICROSECOND
JITTER AND CROSSTALK

http://www.dtic.mil/dtic/tr/fulltext/u2/a515384.pdf

I've not built anything yet so I can't comment on difficulties, but my direct measurements of the output of a mixer and the DDS output of an Agilent 33522A with the same 10MHz as a reference input are in good agreement with a crude calculation based on slope and typical counter noise of 300-500uV. The sin output is around 130mV peak at 1Hz so the slope is around 0.8V/s so noise of say 500uV gives an uncertainty of 6E-4 seconds and, surprisingly to me, my measurements on my cheap counter are slightly better than this but of the right order of magnitude (noise floor around 1E-11 i.e. 1E-7 gain in going from 10MHz to 1Hz and then 1E-4 measurement noise in measuring the 1 second period).

My plan is to follow a similar scheme to the paper and try and improve the slope by a large factor without increasing the noise significantly. I guess that I will discover all the other noise sources in the system on the way but it will be interesting to find out.

Hi

With any reasonable offset frequency you will get to about 1x10^-11 with that circuit.

Bob

[jpb]

I am trying to improve my Alan Deviation measurements with such an approach and the paper that I've found most useful (because it is simple) is

ZERO-CROSSING DETECTOR
WITH SUB-MICROSECOND
JITTER AND CROSSTALK

http://www.dtic.mil/dtic/tr/fulltext/u2/a515384.pdf

I've not built anything yet so I can't comment on difficulties, but my direct measurements of the output of a mixer and the DDS output of an Agilent 33522A with the same 10MHz as a reference input are in good agreement with a crude calculation based on slope and typical counter noise of 300-500uV. The sin output is around 130mV peak at 1Hz so the slope is around 0.8V/s so noise of say 500uV gives an uncertainty of 6E-4 seconds and, surprisingly to me, my measurements on my cheap counter are slightly better than this but of the right order of magnitude (noise floor around 1E-11 i.e. 1E-7 gain in going from 10MHz to 1Hz and then 1E-4 measurement noise in measuring the 1 second period).

My plan is to follow a similar scheme to the paper and try and improve the slope by a large factor without increasing the noise significantly. I guess that I will discover all the other noise sources in the system on the way but it will be interesting to find out.

Hi

With any reasonable offset frequency you will get to about 1x10^-11 with that circuit.

Bob

I am hoping to get below 10^-12 as I'm getting around 1E-11 without the circuit (i.e. directly feeding the mixer output via a LPF to the counter) - but perhaps we're talking at cross-purposes. I'm taking a figure averaged over 48hrs of measurements not a one-shot measurement value. (That is I take measurements every second and then average (f1-f0)^2 as per the ADEV expression).

The authors of the original paper were getting below 10^14 (sub microsecond on the 1 second intervals plus 10^8 on mixing down from 100MHz).

But as I said, I've not yet built anything so it is very premature to set myself targets other than just a general improvement on what I currently have.

[uncle_bob]

Hi

If you are using OCXO's for this, they will not hold anything close to 1x10^-11 for 48 hours. That's in an absolutely perfect temperature stable environment. The only practical way to do it is with something that gives you resolution in a second and cycle the high voltage at something like a 10 second per cycle rate. If you slow things down cycle and reading wise, you need to get much better standards (Cesium Atomic clocks or Hydrogen Masers). For a 48 hour setup, the 5071's at ~$75K each would be a lower cost approach. 

Bob

[jpb]

Hi

If you are using OCXO's for this, they will not hold anything close to 1x10^-11 for 48 hours. That's in an absolutely perfect temperature stable environment. The only practical way to do it is with something that gives you resolution in a second and cycle the high voltage at something like a 10 second per cycle rate. If you slow things down cycle and reading wise, you need to get much better standards (Cesium Atomic clocks or Hydrogen Masers). For a 48 hour setup, the 5071's at ~$75K each would be a lower cost approach. 

Bob

We have perhaps been talking at cross purposes - I realise that I've perhaps gone off topic from the original poster to the general problem of zero crossing and Alan Deviation measurements. I have no budget or plans to try and acquire an oscillator stable to the degree you mention, only to try and get the intrinsic noise of my setup down when measuring a reference against itself. I've been measuring rubidium references against each other.

It would be nice to have a Hydrogen Maser though

[John Heath]

Hi jpb

There is an alternative method to jitter , noise and frequency measurements by separating the measurements into three parts to avoid 75K , ouch , for hydrogen masers. In my case frequency is the only measurement of interest. Old school TVs found a way to separate noise from frequency for horizontal sync by using high Q resonance with only a 2 to 5 percent composite video sync injection. By doing it this way a 25/75 noise to video ratio could still sync up horizontal scanning rate. I use the same method by a 5 percent coupling of both oscillater inputs to 10 MHz crystals tuned to a slightly lower frequency. 90 % of jitter and noise information is lost. However signal to noise is much improved for a frequency measurement. In your case jitter and noise measurement could be a higher priority. For this a phase detector between high Q resonant crystal and oscillator source would improve jitter and noise measurements but frequency information would be lost. In summary by sacrificing jitter and noise information frequency measurement is improved or by sacrificing frequency information noise and jitter information is improved.

Related to this in an odd way is energy conservation laws. In physics there is a growing seed of information theory that has gained the status of energy conservation laws. Both energy and information follow the same conservation laws. There are examples where you can slip nature a few bucks under the table to look the other way to violate energy conservation laws. The currency of this less than ethical exchange is energy today with an IOU note for energy tomorrow.  Neutrino e/m/tau flavor oscillation is an example of it. As information laws are tracking energy conservation laws the same exchange can be made with nature to give her jitter and noise information in exchange for greater fidelity in frequency information or the compliment of that.
 

To date information theory from physics has not ventured into electrical engineering. If you were to apply information conservation laws to improve noise and jitter measurements it would be a first. I smell a paper here that could be a feather in your cap.

[jpb]

Related to this in an odd way is energy conservation laws. In physics there is a growing seed of information theory that has gained the status of energy conservation laws. Both energy and information follow the same conservation laws. There are examples where you can slip nature a few bucks under the table to look the other way to violate energy conservation laws. The currency of this less than ethical exchange is energy today with an IOU note for energy tomorrow.  Neutrino e/m/tau flavor oscillation is an example of it. As information laws are tracking energy conservation laws the same exchange can be made with nature to give her jitter and noise information in exchange for greater fidelity in frequency information or the compliment of that.

I guess quantum tunneling is another example of this (you could look at it as "borrowing" energy to get up and over the barrier rather than "tunneling" through it).

And of course Heisenburg's classic principle. Though much of this comes down to Fourier Transforms and the fact that if you localize something in one domain (e.g. time domain) then it is uniformly spread out in the other domain (such as frequency).

At present I'm too ignorant to fully understand exactly what I'm measuring in terms of whether it is changes in frequency (frequency modulation) or better looked at as jitter. I must admit to not fully understanding your post.

I remember a few years ago reading that there was some discussion in the physics community regarding black holes and conservation of information as information is apparently lost if light can't escape. I think the conclusion was it all was retained in the event horizon - I know nothing about black holes but this is the gist of what was reported in the popular press.

Fascinating stuff but way off topic!

To date information theory from physics has not ventured into electrical engineering. If you were to apply information conservation laws to improve noise and jitter measurements it would be a first. I smell a paper here that could be a feather in your cap.

I thought information theory started with electronics rather than physics (Shannon)? Though I have a background in academia and research it is no longer my day job so I do this stuff just for fun - the OP sounds more serious though.

[John Heath]

 Measuring 10 MHz down to 10 uHz or your efforts to improve noise and jitter measurements is all about information theory. This is not off topic rather on topic. Claude Elwood Shannon. Glad you put him on the table as it leads to great googling that I am still enjoying.  I can see from Wikipedia that he is using entropy as his foundation. I would add that just 5 years ago Dr Susskind  raised the bar yet again to say information equals energy. From this he calculates the precise diameter of a worm hole between to black holes. A little over my head but he makes his  reasoning is clear , energy = information = surface area of an event horizon. This way one does not use information when it falls in as it is still there stuck on the outside surface frozen in time. So historically it starts with thermodynamics , entropy , statistical thermal dynamics , entropy = organized vs chaos , energy = degree of organized and the last kick at the can information = energy.  There are different ways to connect the dots as there were so many bright minds that contributed to information theory as it stands today.

In your case of over driving a full bridge mixer with a high slew rate you have lost symmetry information. By this I mean was it a pulse or a square wave? That information is lost and the only way to gain it back is to divide by 2 returning to the symmetry of a square wave.  However in doing this frequency information is 1/2.  However with over driven high slew rate jitter is easier to measure just by syncing it on a scope and eyeball the right side of the scope. In this case information of symmetry was lost but jitter information was gained. Information theory applied as an energy conservation law is starting to show itself in this example. Most of us already know this intuitively but when it is stated as a law then a greater confidence can come from it to try new ideas.

In my case I look for ways to lose jitter and noise information with a blind faith that nature will provide more frequency information in exchange for jitter and noise. A practical application of this in my case is to take a 36 Hz beat frequency between test and reference 10 MHz oscillators and frequency multiply 36 Hz by 1024 to 36 KHz with a 74hc4046 PLL and a 74hc4040 ripple counter. Results so far are miserable but faith in information theory says it should work. It is just a matter of fiddling with the pedestrian details of getting a 74hc4046 to do what the spec sheet says it will do.

As to your distinction between jitter and frequency. For my 2 cents I do not see a difference between jitter and random frequency modulation. It is more the source of the noise than the technical classification of the noise. I hope my ambiguous response puts a smile on your face knowing I have no idea either. 
 

[uncle_bob]

Measuring 10 MHz down to 10 uHz or your efforts to improve noise and jitter measurements is all about information theory. This is not off topic rather on topic. Claude Elwood Shannon. Glad you put him on the table as it leads to great googling that I am still enjoying.  I can see from Wikipedia that he is using entropy as his foundation. I would add that just 5 years ago Dr Susskind  raised the bar yet again to say information equals energy. From this he calculates the precise diameter of a worm hole between to black holes. A little over my head but he makes his  reasoning is clear , energy = information = surface area of an event horizon. This way one does not use information when it falls in as it is still there stuck on the outside surface frozen in time. So historically it starts with thermodynamics , entropy , statistical thermal dynamics , entropy = organized vs chaos , energy = degree of organized and the last kick at the can information = energy.  There are different ways to connect the dots as there were so many bright minds that contributed to information theory as it stands today.

In your case of over driving a full bridge mixer with a high slew rate you have lost symmetry information. By this I mean was it a pulse or a square wave? That information is lost and the only way to gain it back is to divide by 2 returning to the symmetry of a square wave.  However in doing this frequency information is 1/2.  However with over driven high slew rate jitter is easier to measure just by syncing it on a scope and eyeball the right side of the scope. In this case information of symmetry was lost but jitter information was gained. Information theory applied as an energy conservation law is starting to show itself in this example. Most of us already know this intuitively but when it is stated as a law then a greater confidence can come from it to try new ideas.

In my case I look for ways to lose jitter and noise information with a blind faith that nature will provide more frequency information in exchange for jitter and noise. A practical application of this in my case is to take a 36 Hz beat frequency between test and reference 10 MHz oscillators and frequency multiply 36 Hz by 1024 to 36 KHz with a 74hc4046 PLL and a 74hc4040 ripple counter. Results so far are miserable but faith in information theory says it should work. It is just a matter of fiddling with the pedestrian details of getting a 74hc4046 to do what the spec sheet says it will do.

As to your distinction between jitter and frequency. For my 2 cents I do not see a difference between jitter and random frequency modulation. It is more the source of the noise than the technical classification of the noise. I hope my ambiguous response puts a smile on your face knowing I have no idea either. 

Hi

Precise frequency measurement is a heavily studied area. If you dig into the NIST archives, they have at least 50 years of papers there on various approaches. They also have far more papers on the statistics associated with these measurements that you would ever care to read. Their archives are nice simply because they are free to access. They represent < 10% of the studies in this area.

When you talk about an oscillator (or frequency source) in a precise way, you use things that have exact definitions. ADEV is one of those terms. It's whole reason for existing is that back in the 60's they needed a number that was precise as opposed to the other odd ways of describing things. The gear to measure ADEV is the same thing you need to use to measure your two oscillators. The techniques you are proposing have been in use since at least the 1940's in various forms. HP produced and sold pre-built setups that did these things starting in the mid 70's.

You can indeed re-invent this all from scratch. That's fine and it can be a lot of fun. You can build up this and that to see what happens. If the target is the original experiment, that sounds like a pretty long many years of detour. If the objective is to simply keep busy, it's as good as anything else to do. I'd suggest starting by breadboarding a single mixer setup and seeing what you get. Then try to optimize it for better performance.

What isn't quite in the cards is to have others read a few hundred papers for you and summarize each of them to you step by step. That's going to get really old really fast. Also, much gets lost when that sort of thing is done. You never really get the understanding that is critical to getting the whole system to work. It's like a tech who has a repair manual to follow, but has no idea at all how a system actually works. He can fix a broken system. He will have a very hard time building a system from scratch.

Bob

[John Heath]

First of all to the many who have responded thank you. The advise given leads to a redesign and fine tuning to resolve 10 MHz down to 10 UHz. Resolution is now 10 mHz ,, sort of . Still off by a factor of 1000 but it is the direction that counts. An odd spectrogram observation is made that I find myself taxed to explain. Perhaps someone else has seen this and would know what it is.

The test set up:

Two 10 MHz oscillators with a hetrodyne beat frequency of 3390 Hz . The focus of the spectrogram is the 3390 Hz beat frequency. The beat frequency is drifting at about 1 Hz per second every 2 seconds or so. The problem is a close side band that hints there is a way to cheat the system. Cheat in the sense of having more information than one deserves. The frequency change in the side bands 3390 Hz is 100 fold greater in delta frequency than the main frequency. Them question is why and what is causing it to be there. It is my hope that someone has seen this in side bands and can explain why it is there. 

[uncle_bob]

First of all to the many who have responded thank you. The advise given leads to a redesign and fine tuning to resolve 10 MHz down to 10 UHz. Resolution is now 10 mHz ,, sort of . Still off by a factor of 1000 but it is the direction that counts. An odd spectrogram observation is made that I find myself taxed to explain. Perhaps someone else has seen this and would know what it is.

The test set up:

Two 10 MHz oscillators with a hetrodyne beat frequency of 3390 Hz . The focus of the spectrogram is the 3390 Hz beat frequency. The beat frequency is drifting at about 1 Hz per second every 2 seconds or so. The problem is a close side band that hints there is a way to cheat the system. Cheat in the sense of having more information than one deserves. The frequency change in the side bands 3390 Hz is 100 fold greater in delta frequency than the main frequency. Them question is why and what is causing it to be there. It is my hope that someone has seen this in side bands and can explain why it is there.

Hi

One of your OCXO's has a component in it that is oscillating ... It could be a regulator,or the oven controller, or a buffer amp or something entirely different.

Bob