Upcoming ARRL Frequency measurement test

[cncjerry]

I've kicked around a lot of ways to participate in the upcoming test.  I have a GPSDO, 4 or more spectrum analyzers, frequency counters, etc.  Not looking for trade secrets or anything, but I was wondering how best of breed participants get down to 1/4hz? 

I was thinking of building a crystal filter on the frequency and feeding it right into a counter using my GPSDO as a base.  I could also feed the signal right into a spectrum analyzer using the GPSDO as a reference.  Based on some testing I have done this gets me to 1hz.  The question is the next level getting below 1hz accuracy.  I think one of my counters has .1 hz resolution at that frequency.  So if I used my crystal filter, amplified it and fed it into my counter using the GPSDO as the reference, that should get me below 1hz, assuming the filter doesn't start ringing then I would just be measuring my filter resonant frequency.  Pipe dream?  I know people use SDR panadapters which I could also do but I think that would be guess work unless you could see a reference in the same band spread as the signal in question, no?  One of the signals is going to be around 7.040Mhz.  I was thinking that if I used my reference to feed a synthesized signal generator like my HP 8648C, I could then do an FFT on my signal next to theirs with a large number of bins.  That should get me below 1hz I think.  I've seen rotating phase shift detectors that if you count the rotations per second you can get the frequency delta to a greater level of accuracy.

Other ideas?  I wanted to participate last year but I was traveling.

Thanks.

[Kleinstein]

A usual way to get very high frequency resolution is to do I/Q mixing (much like many SDR systems do it), down to a frequency low enough to digitize the signal. Digitize after moderate analog filtering and than do a least square (or approximation of that) fit of an complex harmonic signal. Slow signals can be digitized directly.

An analog high Q filter can be contra-productive (especially directly at the input), as drift of that filter would cause low frequency phase noise. So the input filtering would be more like lower Q and higher order.

I have not used this one high frequencies - I used it in the audio range to get down to µHz resolution, even with not so great input signal.

[cncjerry]

That's what I was thinking though.  take my GPSDO and feed a DDS to make a near perfect 7Mhz signal, or whatever the best signal would be depending on the AD9854 DDS.  That chip outputs I/Q signals into a downconverter that will then feed SDR Sharp running on my PC.  So I'll be using one reference to receive the signal and another AD9854 to act as a signal generator.  I'll then use the FFT function in SDR sharp set to like 512K bins to measure the frequency vs the reference oscillator.  That should get me to sub hz I would think.  Using two AD9854's locked to the reference should be pretty accurate.  I have to do some reading on the AD9854 to see what the purest signal would be around 7.040Mhz as they have spurs all over the place.  I have 4 or 5 of the AD9854 evaluation boards and some early boards from Richard Hoskins when he built one of the first I/Q VFOs.

[NA5WH]

Last year I picked up an IC-7300, which is a direct sampling SDR. I've had some interest in FMT for awhule, sadly I have nothing to produce a quality reference signal, but have thought that injecting a known reference signal and using spectrum lab or something like it to do the comparison.  I know it can also be done with comparing against WWV and beat-toning... but haven't learned all the details how to do that side.

Many years ago did it with a friend, as our method was maybe somewhat crude... but was somewhat similar using a reference oscillator and an old tektronics oscilloscope in x-y mode looking for a 1:1 Lissajous pattern....   I can't remember all the details, but it took too long for the .. 3 minute transmission I think it was... especially with atmospherics that do evil things to signals.

[Kleinstein]

Many mixers don't need a sine wave, but can as well be driven from a square wave.

You get a clean 7 MHz square wave from an PLL using an VCXO. It won't be 7.04 MHz, but 7.000 MHz and might need a custom made VCXO (e.g. 14 MHz ?) - still usually more stable than an DDS driven by an PLL derived clock. With todays HW 40 kHz IF sounds very reasonable.

[cncjerry]

So you are suggesting to use a fractional PLL with the 10Mhz reference instead of a DDS?

[German_EE]

I've done a couple of frequency measurement tests and they are an interesting exercise, here's what I did:

1) Warm up. Close all the doors and windows in the shack and let everything including the computer warm up for two hours. This reduces the drift problem to more manageable levels.

2) Calibration. Tune your receiver either 1KHz high or 1 KHz low of a standard frequency transmission. In my case I'm using DCF77 on 77.5 KHz and I have the receiver tuned to 77.4 KHz, this generates a beat note of approx 1 KHz. Feed the audio signal into a PC based spectrum analyzer and measure the frequency down to at least 1 mHz, in my case I measured 999.9813 Hz. If your software allows you to calibrate this to exactly 1 KHz then do so but your alternative is to multiply all readings by 1.0000187 to get the same result

3) Measurement. Tune in the test transmission using the same receiver and the same VFO and sideband settings (important) and vary the VFO until you get a beat note of approx 1 KHz. The frequency of transmission is Displayed VFO Fq + (Displayed Audio SA Fq x 1.0000187)

This system works if your receiver has a DDS based VFO, if it's a mixing rig with a different crystal oscillator for each band then it won't work. Using a good oven based frequency counter to measure the 1 KHz signal is an alternative but in this case you are forced to work out the correction factor manually.

[Kleinstein]

No need to use a fractional PLL to go from 10 MHz to 7 or 14 or 70 MHz. This can be an integer PLL. The close in range is set by the PLL loop anyway - so one could get away with a more standard VCO.
An DDS would use something like an PLL to get a higher internal clock anyway. Not sure if you really need to build your own specialized receiver.

The signal quality will be limited due to radio transmission anyway. So ne need to go for absolute best performance. Just make sure to get enough measurement time and good reception. The worst part would be mixed reception from different paths. A don't know if there are suitable antennas to avoid this.

The easy solution would be an of the self receiver in SSB mode, recording to something like a sound-card and parallel measurement of the LO frequency (or frequencies) - used with a good counter. This could be indirect via the tuners crystal reference.

[w2aew]

I've done a couple of frequency measurement tests and they are an interesting exercise, here's what I did:

1) Warm up. Close all the doors and windows in the shack and let everything including the computer warm up for two hours. This reduces the drift problem to more manageable levels.

2) Calibration. Tune your receiver either 1KHz high or 1 KHz low of a standard frequency transmission. In my case I'm using DCF77 on 77.5 KHz and I have the receiver tuned to 77.4 KHz, this generates a beat note of approx 1 KHz. Feed the audio signal into a PC based spectrum analyzer and measure the frequency down to at least 1 mHz, in my case I measured 999.9813 Hz. If your software allows you to calibrate this to exactly 1 KHz then do so but your alternative is to multiply all readings by 1.0000187 to get the same result

3) Measurement. Tune in the test transmission using the same receiver and the same VFO and sideband settings (important) and vary the VFO until you get a beat note of approx 1 KHz. The frequency of transmission is Displayed VFO Fq + (Displayed Audio SA Fq x 1.0000187)

This system works if your receiver has a DDS based VFO, if it's a mixing rig with a different crystal oscillator for each band then it won't work. Using a good oven based frequency counter to measure the 1 KHz signal is an alternative but in this case you are forced to work out the correction factor manually.

Another option without relying on the accuracy of the audio frequency measurement would be use a standard frequency broadcast (as suggested above, but the higher the better), but listen/measure the audio frequency of the heterodyne produced in the upper and lower sidebands when tuned away from the carrier.  For example, tune 1kHz above the carrier and listen for the 1kHz heterdyne on USB, then tune 1kHz below the carrier and listed to the tone on LSB.  If you do this on a radio with two VFOs, you can set one for the USB test and one for the LSB test, and quickly switch between them.  If your radio is "dead on", you'll hear no difference between the two tones.  This is actually easier to "hear" if you choose a smaller offset, like 100 or 150Hz.  If you have the ability to adjust the reference in your radio, you can tweak it until the tones match in frequency.  By ear, you can usually get the difference to a few Hz or less.  If you are listening to something like WWV at 10MHz, this means you've adjusted the rig to a few tenths of a ppm.  If you can't adjust the reference, at least you know how much you're off (by noting the difference in tone frequencies.

[cncjerry]

Audio methods won't get anywhere near close enough.  To be competitive you need to be in the .05PPM range. 

Using a reference signal measured to .01hz or better and recording that signal next to the unknown frequency would enable me to get close enough to be within the Doppler shift error.  Handling the Doppler shift is another feat.

I hooked my GPSDO to my spectrum analyzer reference input tonight and got within .4hz measurement of WWV.  I tried measuring some local AM stations and they were all over the place.  Tomorrow I am going to try some of the other disciplined stations to see how close I can get to their frequency.

[LaserSteve]

If your in the US or Canada, look at the Pilot carrier of a DTV station. They are supposed to be -/+ 2 Hz.  Should show as a nice spike above the data on your spec an.

S.

[German_EE]

"Audio methods won't get anywhere near close enough."

The last frequency measurement test I triedmy entry was 2.2 mH out, not bad as I was monitoring a signal from the USA here in Germany. Most of that error was probably due to Doppler effects.

[retrolefty]

I'm an extra class ham so I should know I guess, but what is the reference used by the ARRL transmission during this contest? I recall reading a QST article about the contest decades ago but haven't ever participated.

[cncjerry]

"Audio methods won't get anywhere near close enough."

The last frequency measurement test I tried my entry was 2.2 mH out, not bad as I was monitoring a signal from the USA here in Germany. Most of that error was probably due to Doppler effects.

I'm referring to using your ear to zero beat.  Using audio as in I/Q input into an SDR software where you then run a high resolution FFT is something else.  You have to make sure your sampling rate on the PC audio card is accurate but you can minimize that error by using an accurately generated and measured reference frequency beside the desired carrier. Then you just have to run the FFT to calculate the offset from the reference. I'm wondering if you can calculate an approximation of the doppler shift.  Just starting looking into that one.

[Kleinstein]

One way to estimate Doppler-shift and similar atmospheric trouble might be looking at a second similar radio station too. AFAIK there are a few official reference frequency stations a round that could be used. If they are the same direction, similar path one might use them to estimate atmospheric effects. This might be interesting for those in Europe that are more influences by it.

With a good model for transmission one might be able to estimate Doppler shift from amplitude variations. The frequency spectrum (e.g. how much phase noise) might also give a clue. Also experience based on the time might be a first approximation.

[German_EE]

This is roughly what I did. My equipment calibration is based on DCF77 which is only about 33Km from where I am. I then switched to WWV  which (I think) is in Colorado. Knowing that the transmission was on exactly 10.00000000 MHz I could then make an attempt at measuring the doppler effect from the USA to Germany.

[cncjerry]

Doesn't the rotational speed of the earth impact Doppler?  Isn't the west coast spinning toward Connecticut at 1000mph after the signal is transmitted therefore raising the frequency like a train moving towards you?  I was thinking I could calculate the Doppler shift based on some simple math.  Over simplified as usual.

[wn1fju]

Well, here was my experience with the ARRL FMT.  Years ago, I used a piece of equipment we built at work, which consisted of a Trimble
Thunderbolt timing receiver (GPSDO), feeding its 10 MHz output to a x6 multiplier driving a 60 MHz 14-bit Analog Devices A/D.  The data
was then fed through an Analog Devices digital down-converter (with 32-bit frequency control) and stored to disk.  A high resolution FFT
was then done on the decimated output and the frequency determined.  Should be near perfect, right?  I fail to understand where a frequency
error would occur other than 1) Doppler, or 2) a slight (millihertz) offset due to the resolution provided by my 32-bit control word. 

Yet the ARRL said I was 0.8 Hz off.  I looked at all of the entries who got an "award" by being <1 Hz and histogrammed the results.  There
was a huge spike at 0.8 Hz error!!!

In other words, a whole bunch of us were "wrong" by the same 0.8 Hz and the ARRL was "right."

That was my last ARRL FMT.

[cncjerry]

I think I remember the year where .8 hz or thereabouts was talked about.  Doesn't one particular GPSDO report a little low in frequency?  I wonder if that had something to do with it.  I was also wondering if there is a phase shift that could be introduced in an antenna tuner meaning that the actual transmitted frequency would be a little off from what is dialed-in.  Haven't thought about that much to determine if it is possible.  As far as the contest I doubt I'll submit an entry as it is more for my satisfaction.

The guys generating the frequency are just bozos like us using the same equipment.  Actually I would think our GPSDOs are closer to spec than many of the rubidium oscillators that haven't been calibrated in 6 months.  What the heck, if I can measure WWV on 10Mhz to .01hz I should be able to measure a random 7Mhz signal just as well, no?