My participation in the ARRL FMT


What is the FMT?

The "FMT" (Frequency Measurement Test) has a long history, with the first one being held in 1931.  Back then, many amateurs could only determine their frequency trusting that their crystal was on the frequency that they thought it was or by comparison with other signals - amateur or commercial - and hope that those signals were where they were supposed to be.  These tests typically consisted of designated amateurs with signals of known frequency stability and accuracy, transmitting signals, with the carrier frequencies being determined by the participants.

These tests were held (more or less) regularly until about 1980.  By this time, technology had progressed to the point that most hams knew, with previously undreamed-of accuracy (within 100 Hz or so) of where they were and reading frequency to within a few hundred hertz was a trivial exercise.  Since then, technology has progressed even farther, decreasing the uncertainty even more.

So, why do it again?

To paraphrase the bard, it's come full circle:  Many amateurs operate new digital modes that are extremely narrow and require frequency accuracy and stability that would have been nearly impossible for anyone (but the best-equipped labs) to attain when the first FMT was held.  Coupled with this is the fact that many new hams (and perhaps some old ones...) may not have a perfectly clear idea as to how the received and transmitted spectrum works out with respect to the modulated signal - especially since most of the new digital modes utilize "audio" (baseband) modulation to generate a non-voice RF signal.  Additionally, with the ability for many amateurs to measure signals with absurd accuracy, there is some skill in being able to determine the precise frequency despite the effects of Doppler spreading due to the changing signal paths or even digging out signals that might be too weak to copy by conventional means.

Another reason to do this has to do with education:  It is painfully apparent to many who operate that some amateurs are unaware of their actual transmitted frequency - often not realizing that the frequency displayed on the radio does not necessarily take into account the aspects of the audio signal modulating their transmitter, nor does it take into account the single-sideband nature of that transmitter!  The result is some confusion when a specific frequency is specified for, say, a PSK31 transmission:  Some amateurs have been observed to simply tune their radios to the specified frequency and not take into account the audio frequency and/or the sideband being used!  If they participate in something like the FMT, they will be required to figure out how everything works!

Additionally, there is (admittedly) a "nerd" factor involved:  It's kind of fun to see if your readings correlate with someone else's - and how well.  With today's technology it's quite practical to have frequency-measurement capability that far exceeds the uncertainty of the propagation medium:  As mentioned above, the ionosphere shifts the frequency by an amount far greater than the measurement uncertainty.

General techniques used for the 2003-2004 FMT:

The method to the madness:
When the sampling rate isn't the sampling rate...

After the 2003 FMT, when I was playing the audio the from the "record" computer into the "analysis" computer, running the ARGO program, I noticed something odd:  The audio frequencies that I'd remembered noting on the record computer didn't match what I was seeing on the analysis computer. 

Scratching my head, I re-checked the recording computer against WWV's precise tones and it looked OK.  I then fed the WWV audio into the analysis computer and found it to be off by several percent. 


Something in the back of my mind then clicked:  I exited ARGO and fired up Spectran and set the sampling rate to 8 kHz:  The WWV frequency was a perfect match (to within a few tens of PPM, anyway.)  Setting Spectran back to a sampling rate of 5512.5 samples/sec I observed the same error that I had seen in ARGO. 

The problem?  My "record" computer runs Windows 98 while my "analysis" computer runs Windows 2000.  I'd read, some weeks earlier, that Windows 2000 and XP always run the sound card at 48 kHz - behind the user's back - doing a conversion to a different sampling rate "on the fly."  The older Windows 98 actually ran the sound card at the sample rate that the program set it to. 

While a sampling rate of 8 kHz is an easy integer divisor of 48 kHz, 5512.5 Hz is not - and 5512.5 Hz was the only sampling rate supported by ARGO.  Apparently, the on-the-fly sample rate conversion from 48 kHz to 5512.5 Hz was "close" but not exact whereas the conversion of 48 kHz to 8 kHz is, as they say, a "no-brainer." 

Whether this is actually the case with all versions of Windows 2000 and/or XP, I'm not certain - but if you try to use your computer for precise frequency measurement - beware!

My method of frequency measurement used for the 2004 FMT was an indirect method and it was was exactly the same as it was in 2003:

For this method to work, several things need to be taken into account: At the time of the FMT, my HF receiver (I used my FT-817:  It had been left on for the day or two previous to allow it to stabilize) had been preset to the announced FMT frequencies for each band and had been connected to the sound card of the computer.  When the first FMT carrier appeared, I quickly adjusted the frequency of the signal generator to be within a few hundred Hz of it and calculated the same offset for test signals on the other bands.

The entire FMT was recorded to one .WAV file, notes were taken as to exactly when the recording was started, exactly which band was being recorded at what instant, the time indices of the various band changes, and the frequencies of the generated local reference signals.  At conclusion of the FMT I immediately tuned the receiver to WWV on 5 MHz and applied the same techniques to be used for possible calibration and as a "sanity check" to verify the method and the math.

Following the FMT recording session, I played back the recording of the FMT from one computer (the "record" computer) into another (the "analysis" computer.)  The analysis computer's sound card had also been verified for accuracy and any offsets noted and using the analysis computer and the "Spectran" program, I was able to  accurately measure the audio frequency of the locally-generated signal and the audio audio frequency of the W1AW test carrier.  The Spectran program allows the use of some extremely narrow detection bandwidths so not only was it possible to determine the audio frequencies very accurately, but the narrow bandwidth allowed significant enhancement of the W1AW signal's signal-noise ratio, making it possible to measure it even if it was too weak to hear with one's ear well enough for "conventional" measurement techniques.  The ability to "loop" the playback also permits even longer integration of a reading than the original duration of the W1AW test carrier would allow.  Because several different "test carrier" periods were available for each band, these could be analyzed separately and from this, a statistical average frequency could be calculated.
Spectral diagrams from the 2004 FMT (from the Spectran program) of the 75 meter signal (top) and the 40 meter signal (bottom.)
Click on either image for its full-sized version.

Knowing the RF frequency of the locally-generated signal, the audio frequency of that signal as produced by the receiver and the precise audio frequency of the W1AW signal allows the one remaining unknown to be easily calculated:  The RF frequency of the W1AW signal.  Knowing this allowed the final calculation of the W1AW "audio" frequency.

Because this is an indirect system of frequency measurement and because it is based on relative measurements, the frequency accuracy of the receiver itself is unimportant:  The onus is on accuracy of the locally-generated signal.  What is important is that the receiver be stable enough during the test carrier that its drift is a small fraction of the observed frequency error.  Having a locally-generated signal can help mitigate this to some extent, as any receiver drift that does appear can be observed, measured, and taken into account and, in some cases, counteracted using audio processing software.  In my case, I observed only one or two 10ths of one Hz drift over the duration of the entire FMT - this stability being the result of having had the radio on for a day or so in advance and the ambient room temperature being stable.

Of course, the "twist" to the 2004 FMT was that there was this "hypothetical" SSB carrier on a previously-stated frequency, and the goal was to determine what audio tone modulated onto that carrier would yield such an RF frequency.  Being that I know how to add and subtract, this extra step was rather trivial in comparison to the others.

Conditions in 2003:

The 75 meter signal:

As it turns out, the copy of the 75 meter signal was quite rough due to some QRM on a nearby frequency, making the voice announcements were difficult to copy.  Despite the QRM, using a detection bandwidth of 0.031 Hz in Spectran yielded a signal/noise ratio of approximately 45 db for this signal with minimal Doppler spread.

The 40 meter signal:

On 40 meters, it was even worse as the W1AW signal had to contend with QRM from a foreign broadcast station:  On this band I could copy none of the voice - but I could just hear (using the "naked ear") what appeared to be the test carrier.  The spectral diagram (again, using a detection bandwidth of 0.031 Hz) showed a signal/noise ratio of approximately 10 db.  This is actually somewhat misleading as the energy is spread over nearly 1 Hz and is "diluted" amongst several of the FFT bins and is actually better in a slightly wider bandwidth.  Trying to determine the actual frequency from a display like this is somewhat problematic:  It takes a bit of staring and mental averaging to decide where the "center" of the energy mass might be.

As a reminder, note that the audio frequency shown on the spectral display is not exactly that of the audio frequency:  The displayed frequency was referenced to the locally-generated carrier and the actual RF carrier frequency was thus determined.  Also note that the waterfall display is incomplete as I did not wait long enough for it to build.  Finally, the test segment of the .WAV file containing the test was set to loop (repeat)  in order to allow easily-repeated observation of the segment.

The 20 meter "signal:"

Just for completeness, I also listened and recorded on 20 meters, but heard nothing using my ear.  Just for 'yuks I decided re-analyze my .WAV file now that I know the precise frequency on which the carrier was transmitted to see if there are any traces of it at all.  As you can see from the spectral diagram, the results are somewhat inconclusive:  While a vague peak within a few 10ths of Hz of the expected frequency was detected, it does not distinguish itself from the background noise to enough of a degree to have been spotted amongst everything else.
Comparisons between conditions during the 2003 and 2004 FMTs:
Spectral diagrams from two recorded test transmission on 20 meters during the 2004 FMT.  The detected peak closest to the expected frequency is marked with a red "V"
Click on the image for a larger version.

Conditions in 2004 were much better than in 2003:  In '03 conditions were somewhat disturbed due to solar activity, causing a significant amount of Doppler spread.  Also, owing to the moving terminator (the line between Day and Night) it appeared that there was a bit of Doppler shift due to the shifting of the ionospheric layers.  Both of these factors raised the uncertainty of the frequency measurement resulting in a frequency error of 0.21 Hz and 0.13 Hz for 80 and 40 meters, respectively.  Another possible source of error in 2003 may have been exciter drift:  I noted a pronounced drift in the 80 meter signal that, at the time, I wrote off to changes in propagation.  Later, I noted that another 2003 FMT participant had reported observing exactly the same amount of drift, but he was on the other side of the continent. Was it likely that the propagation changes affected us equally, even though we were over 2000 miles apart?  If so, would it have affected an East/West path in the same way as a mostly North/South path?  I don't think so!

The "calmer" conditions in 2004 allowed a much more precise frequency measurement:  A reported measurement error of 0.02 Hz and 0.06 Hz on 75 and 40 meters, respectively.

It is interesting that the ARRL chose to change the way the FMT was run in 2004, having the participants determine the audio frequency rather than the actual RF carrier frequency.  While the relative merits of these rules as compared to previous years' rules may be debated, both methods require attention to detail.

The 2005 FMT:

The format for the 2005 Frequency Measurement Test (held on November 16) was the same as that of the 2004 FMT.  Because of good results in years past, I used the same general techniques that I had used in 2003 and 2004 - that is the generation of a weak "reference" signal on a precise, known frequency within the receiver's passband and then recording the entire FMT on a computer for later analysis.  The only notable difference was that the signal on 20 meters was replaced with one on 160, owing to the reported poor coverage of that band during that time of day from that location.

I was able to hear, with the "Naked Ear," the test transmission on all bands, but the 75 meter test signal seemed, audibly, far stronger than the other two.  Conditions on 75 meters seemed to be slightly poorer than in 2004 as I was not able to copy more than a few occasional syllables of the voice announcement.  Being that I was using a single receiver (the FT-817 again) I simply changed bands (and frequency on the GPS-referenced signal generator) after each test carrier, making notations as I had done so.

As expected, there was some detectable "Doppler spreading" of the 40 meter signal (perhaps 0.5-0.75 Hz) which caused some "bin dilution" on the FFT-based spectrum display (using the Spectran program) but this spread was minimal on 75 meters and more-or-less nonexistent on 160.  Using a detection bandwidth of 0.061 Hz, I observed about 25 dB C/N on 40 and 160 meters (with, perhaps, 10-15 dB of dilution on 40 due to the Doppler spread) and about 40 dB C/N on 75 meters.

These C/N readings would seem to be about right:  The minimum detectable C/N for a single tone for the "Naked" ear correlates roughly to 0-3 dB C/N in a 50 Hz bandwidth, when the audio spectrum is populated with just noise - and this approximately matches the "just audible" tone heard on 40 and (especially) 160 meters.

In the days following the FMT, I observed some interesting "chatter" on a mailing list concerning the hypothetical possibility of removing one source of frequency measurement error by being able to measure the frequency of the suppressed carrier of the signal.  In theory, this could work, but there are several potential problems with this method:

Despite all of this, it would still be fun to see if one could detect the suppressed carrier, anyway - but it would certainly take a bit of work!

The 2006 FMT:

The format of the 2006 FMT reverted to tradition, being pretty much the same as that of the 2003 FMT - that is, simply determining the frequency of a transmitted carrier rather than calculating the frequency of the audio tone generated by an SSB transmitter given a hypothetical carrier frequency.

At the moment of commencement (at 0245Z) the signals from W1AW were comparable to those last year, but the 80 meter signal rapidly gained ground over a few minutes, getting to be quite strong.  The signals on 160 meters were fairly strong, but so is the normal QRN on that band, so the overall Signal to Noise ratio was not particularly good - but the CW was generally copyable.  The signal on 40 meters was similar to that experienced last year:  Difficult copy on CW and fairly weak signals.  Owing to the narrowband detection techniques, this did not prove to be any real problem.

In the FMT announcement, it was noted that WA6ZTY, in the San Francisco Bay area, would also provide a test signal starting at 0330Z, but on approximately 7029 kHz.  I listened for this signal - but could detect absolutely nothing, other than other casual CW QSOs going on nearby - and I even checked at 7039 and places in between.  This would seem odd, as I would expect that the 40 meter path between Utah and California would exist at this time of day, even if the band did get a bit "long" - and it seems clear that in years past, those located about the same distance from W1AW as I am from Northern California did not have too much difficulty in hearing the 40 meter signals.

I have since heard from several participants in the Southern California, Ontario, Ohio, Los Angeles area, and various places along the Eastern seaboard who had exactly the same (lack of) luck - yet a smaller number of people (one in Oklahoma and another in Vancouver reported having heard the transmissions just fine.  I also got an email from WA6ZTY himself who mentioned that he was running 500 watts into a full-length dipole favoring radiation in the East/West directions:  He, too, was disappointed in the propagation.

Comparison of results:

The following table compares my results with those of others that have made theirs known.

Note:  For the 2003 FMT, the object was to determine the absolute carrier frequency, while the object of the 2004 and 2005 FMT's was to determine what audio frequency would need to be modulated using an SSB transmitter to produce the signal detected, assuming a hypothetical RF carrier on the frequency given in ARRL's announcement.  For 2006, the format reverted to that used in 2003 where the object was simply to determine the precise frequency of the transmitted carrier.

It is worth noting that the 2004 announcement, specified a frequency measurement accuracy of +-0.1 ppm (100 ppb) for the transmitted frequencies, and that the 2005 announcement specified a 0.25 ppm (250 ppb) accuracy.

ARRL-published frequency (Hz):
("hypothetical" RF carrier freq. noted for '04 and '05)

Error in Hz (ppb)
compared to ARRL
[compared to median]
N8UR's error in Hz (ppb)
(compared to ARRL):
2003 FMT Results:

80 meters:  3585383.7
3585383.489 Hz
-0.21 (-58.6 ppb)
3585383.685 Hz
-0.015 Hz (-4.18 ppb)
40 meters:  7050409.9
7050409.775 Hz
-0.13 (-18.4 ppb)
7050409.949 Hz
0.049 Hz (6.93 ppb)
20 meters:  14050075.7
(nothing detected)

(nothing detected)

15 meters:  21053399.1
(nothing detected)

(nothing detected)
2004 FMT Results:

75 meters:  1105.02 (RF = 3990.0 kHz)
1105.0 Hz
-0.02 (-5.0 ppb)
1105.013 Hz
-0.007 Hz (1.75 ppb)
40 meters:  1108.26 (RF = 7290.0 kHz)
1108.2 Hz
-0.06 (-8.2 ppb)
1108.279 Hz
0.019 Hz (2.60 ppb)
20 meters:  1117.22 (RF = 14290.0 kHz)
(nothing detected)

(nothing detected)
2005 FMT Results:

160 meters:  1050.92 (RF = 1855.0 kHz)
1050.547 Hz
-0.373 (-201 ppb)
[0.047 (+23 ppb)]
1050.532 Hz
-0.388 (-209 ppb)
75 meters:  1047.19 (RF = 3990.0 kHz)
1047.165 Hz
-0.025 (-6.3 ppb)
[0.065 (+16.3 ppb)]
1047.057 Hz
-0.133 (-33.3 ppb)
40 meters:  1056.81 (RF = 7290.0 kHz)
1056.733 Hz
-0.077 (-10.56 ppb)
[-0.067 (-9.2 ppb)]
1056.512 Hz
-0.298 (-40.9 ppb)
2006 FMT Results:  Official ARRL data is pending

160 meters:  Approx. 1853 kHz
1,854,314.74 Hz
- see notes below)

75 meters:  Approx. 3586 kHz
3,587,117.907 Hz

40 meters:  Approx. 7039 kHz
7,038,805.905 Hz

40 meters:  Approx. 7029 kHz
(nothing detected)

Notes on 2006 measurements:

NOTE on 2005 data:  "Median" data (where available) was obtained from N8UR's "Raw Data" analysis of results obtained from the ARRL that were within +-1 Hz.   N8UR's data may be seen at his website along with some analysis of the Raw Data from entrants.

Analysis of 2005 data:

Any questions? Send some Email.



(C) KA7OEI - Updated 20070119

Since 12/2010: