Brian,
I found the Fldigi system interesting... and at first you had me
thinking someone had found a way around the Heisenberg Uncertainty
Principle, in terms of making rapid measurements of fine frequency
differences...
So I did some research on Fldigi, tried it out, and looked up AFC
algorithms on Google... I found this really interesting survey
article, with some real meat in it (not for the faint of heart):
http://engnet.anu.edu.au/DEcourses/engn3214/notes/FDNatali.pdf
At any rate, in there you find the DFT variant (discrete Fourier
Transform) which Fldigi may, or may not, be using. At any rate, I'm
right at home with FFT's and I could readily see that there is no
speedup by using a tracking filter. It takes the same amount of time
to approximate fine frequency deviations, no matter how you do it.
That is very reassuring to me, having spent much of my career trying
desperately to estimate things like precession periods of incoming
nuclear warheads, (which are around 100 mHz), based on only a few
seconds of observations...
But what is great about Fldigi, compared to SpectrumLab, is that the
tracking filter approach performs a kind of continuous averaging for
you, in displaying the estimated tracked frequency.
However, I tried using SpectrumLab last night, where I produced a
strip-chart of the estimated FFT frequencies. SL uses FFT bin
interpolation, on the assumption that (a) SNR is good and high, (b)
no nearby interfering signals, and (c) they know the shape of the
windowing function. It is a fancier form of DFT AFC than described in
the paper (above). There is a huge advantage to having a strip chart
recording because, instead of trying to estimate changing trends by
eyeballing individual measurements as they come flying past, you can
actually see the p-p frequency deviations and the chatter on top of
the longer term cyclic trends.
Fldigi seems to use hard-coded loop bandwidths and capture ranges. 2
Hz capture range, 5 sec integration time. Good, but I would sure like
to be able to tweak those myself. SpectrumLab also has its
limitations in that there aren't sufficient computational
capabilities built into its tiny analysis programming language.
(Oh, how I dream of just doing it all myself in Lisp, with Lisp fully
present all the time... that way you can make up ad-hoc measurements
that would never have been thought of until you need them. Closed
systems can't possibly anticipate every need. I'm getting dangerously
close to rolling up my sleeves and just doing it. If I do, then I'll
share with everyone too...)
- de Dave, N7AIG
Dr. David McClain
Chief Technical Officer
Refined Audiometrics Laboratory
4391 N. Camino Ferreo
Tucson, AZ 85750
email: d...@refined-audiometrics.com
phone: 1.520.390.3995
web: http://refined-audiometrics.com
On Oct 10, 2010, at 16:34, Brian Lloyd wrote:
On Sat, Oct 9, 2010 at 4:41 PM, Jerry Flanders
<jefland...@comcast.net>wrote:
An additional source of error is the fact that the 5000 does not tune
continuously. It tunes in steps, and they are irregular, so an
additional
correction is required. When you are watching your WWV phase
display, a part
of the error you see may be due to this. This may not be important
unless
you are trying for sub -100 mHz accuracy.
Well, I did pretty well and figure I should comment here.
First thing I want to say is: I was lucky. I should not have done
as well as
I did. I will explain why as I go along.
My setup: Flex 5000, LPRO-101 Rb reference, beta PowerSDR, Fldigi
3.21.0AM.
Sources of error:
1. accuracy of the reference;
2. tuning accuracy of the Flex 5000 DDS;
3. ionospheric doppler error.
The LPRO-101 Rb reference is pretty close to on-the-money. Error there
should be less than 1mHz at 10MHz so this is not a significant
source of
error.
The Flex 5000 uses an Analog Devices AD9959 DDS chip to generate
the LO
signals. This chip uses a control word which is really a fraction
by which
the 500MHz clock is multiplied. The result does not necessarily
fall on an
exact 1Hz boundary. It is correct to +/- 55mHz (an overall peak-to-
peak
frequency error of 110mHz). Analog Devices has a calculator that
will tell
you the actual error for any given input. It also appears that Flex
does not
use all of the bits in the tuning word further reducing accuracy.
(The error
is still well under +/-0.2Hz but that isn't really good enough for
an FMT.)
The last source of error is the ionosphere. I use fldigi's frequency
measurement function to repeatedly sample the frequency. When
measuring
frequency fldigi phase-locks to the signal and accumulates phase
error in
order to calculate frequency rapidly. I get about 1800 frequency
data points
in a 2 minute sample period. That data goes into a spreadsheet
where I plot
the data and do statistical analysis. The plots for 40m and 80m (I
couldn't
hear the east coast 80m and 160m transmissions) showed significant
variations, with 40m having a peak-to-peak variation of 800mHz and 80m
having a peak-to-peak variation of 300mHz.
I plot the data so I can look for clear trends that are not obvious
from the
statistical analysis alone. Looking at the plot of frequency over
time shows
a non-random distribution. This is NOT a Gaussian distribution. In
fact, to
my eye it appears that there are at least two periodicities. I am
thinking
that it appears more chaotic than random.
So, with all that, you can see that, even if my statistical
analysis was
dead-nuts accurate, the error in setting the VFO in the Flex 5000
would
introduce enough error to keep me from getting within 10mHz. Given
that, I
hold that my success (5mHz on 80m and 51mHz on 40m) is pure luck.
All my
errors combined to cancel out. I am chalking it up to beginner's
luck (and a
smaller dose of understanding error sources). But that isn't the
way to win
every time.
One other thing, I see a lot of discussion of the Thunderbolt
GPS-disciplined oscillator (GPSDO). I have one of those as well as my
LPRO-101 Rb reference. Both work just fine. My feeling is that,
while the
LPRO-101 does not have as great an absolute accuracy, it has better
short-term accuracy with lower phase noise (jitter). The right
answer is a
GPS-disciplined Rubidium reference.
Still, the FMT was a lot of fun and you can bet your bippy that I
will be
entering more of them in the future. I hope that I will be able to
kick
everyone's butt for real in November. :-)
--
Brian Lloyd, WB6RQN/J79BPL
3191 Western Dr.
Cameron Park, CA 95682
br...@lloyd.com
+1.767.617.1365 (Dominica)
+1.931.492.6776 (USA)
(+1.931.4.WB6RQN)
_______________________________________________
FlexRadio Systems Mailing List
FlexRadio@flex-radio.biz
http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz
Archives: http://www.mail-archive.com/flexradio%40flex-radio.biz/
Knowledge Base: http://kc.flexradio.com/ Homepage: http://
www.flexradio.com/
_______________________________________________
FlexRadio Systems Mailing List
FlexRadio@flex-radio.biz
http://mail.flex-radio.biz/mailman/listinfo/flexradio_flex-radio.biz
Archives: http://www.mail-archive.com/flexradio%40flex-radio.biz/
Knowledge Base: http://kc.flexradio.com/ Homepage: http://www.flexradio.com/