Hi Rudy
My analysis is NOT based on stations that generate a lot of spots!
My analysis, started from two basic observations:
1. Problems with Busted spots; Non-existent spots on a particular
frequency; small frequency shifting spots; unstable band map with calls
popping-in and popping-out
2. The dynamic of the RBN with a rich and ever increasing number of
spotters, allowing a system that almost "all" validate good calls, and
usually only "one" or "
a few" expose bad calls
This is the very core from where the idea of an error free RBN algorithm
started to evolve.
Therefore there was no approach to the big runners as you suggest, nor
unintended consequences to Small pistols.
A big signal has better changes of being heard and a Small signal has
greater difficulty to break trough.
A station that call CQ for 1 hour or can hold a frequency has better chance
of being spotted. But such is the nature of ham radio. It has nothing to do
with the error free RBN. If you call CQ and three skimmers hear you, no
matter if you are a big gun, a small gun, you get a Good Call tag.
I did present the DR1A example because it is rich full of all variations
and makes a graphical explanation powerful.
I also told clearly the price to such an algorithm. In order to validate a
good call we need to have two so that the third is OK. And I also told that
because of the nature of the rich full RBN network of skimmers, such a
price is almost negligible, because today (with the exception of Pacific
area) when you call CQ, even once, you immediately get spotted by a group
of skimmers.
Also bear in mind that the majority of spots do not come from smaller
stations. In 2012 CW on 40 meters out of the 993.548 spots 90% come from
922 stations out of 8000+ stations that sent logs. These 922 stations are
not the small guns. They are the runners.
In any case the point is, big or small, if you are heard three times, you
get a good call tag.
You seem to be worried by the fact that you are not spotted enough by RBN.
That is not a problem of the error free RBN algorithm because it is not in
place (with the exception of VE7CC-1 that already implemented a version of
it, into his CC Cluster feed). That may be a problem of your QTH or the way
you call CQ, or...
For example in CQWW CW 2012 #1 in the world SOAB was CR3E (myself). While
running I was spotted 1936 times, and #2 in the worl was EF8M with 4534
spots. It is a huge difference. We both have powerful stations and run all
the time, for 48 hours non stop. The difference is in the way both station
call CQ, end QSOs, how often sign the call, variations in speed when giving
a report, etc....there are a lot of variables here.
Your proposed suggestion of making calls good until proven bad is not
possible because of the nature RBN where many spot the good call, and few
spot the bad call. The Bad spot tends to be unique in a time frame and
happen only from time to time. If you think about it you will quickly
realise that with your suggestion all the busts would get through. Also
using SNR to determine would be a mess, subject to signal QSB/variations.
73 José
On Thu, Jul 18, 2013 at 3:39 PM, Rudy Bakalov <r_bakalov@yahoo.com> wrote:
> Hi Jose,
>
> Thank you for your time and energy to improve the usefulness of the RBN,
> an already incredibly useful tool.
>
> I noticed that your analysis is based on stations that generate a lot of
> spots, such as DR1A. I understand why you want to use such stations where
> the many spots give you the opportunity to tweak the algorithm. However, I
> wonder if this approach actually introduced unintended consequences.
>
> Specifically, I wonder if you have considered the impact of your approach
> on small pistol stations being spotted. I have studied the spots for
> N2WQ/VE3 and have found out that despite running full power and using
> decent antennas I never get spotted as often as the big guns, such as DR1A.
> So what happens with even smaller stations that actually constitute the
> majority of the spotted stations? Are we going to end up with a bandmap
> full of "low quality" spots? I suggest a review of your algorithm from the
> perspective of smaller stations and their likelihood of being properly
> spotted.
>
> I also wonder if you can start with the assumption that spots are good
> until proven otherwise, especially if reported at a decent SNR. For
> example, N2WQ/VE3 gets spotted at 15 db SNR. You treated it as good until
> another skimmer reports a similar spot at a better SNR. Or more than one
> skimmer report the similar spot. This concept can be taken a step further
> where you can take the data from multiple skimmers, historical spot data
> from the current contest, and reconstruct the correct spot. So if one node
> reports N2WQ/VE3 and another one reports N2AQ/VE3, and you have heard
> N2WQ/VE3 "recently", you discard N2AQ/VE3 and keep N2WQ/VE3. Similarly, if
> you determine that certain nodes are uncalibrated, you can calibrate them
> on the fly and adjust the spots as reported by RBN.
>
> Rudy N2WQ
> ------------------------------
> *From:* José Nunes CT1BOH <ct1boh@gmail.com>
> *To:* cq-contest@contesting.com
> *Sent:* Wednesday, July 17, 2013 10:26 PM
> *Subject:* [CQ-Contest] Understanding the dynamic error free RBN
> algorithm - The DR1A case
>
> I received several e-mails regarding the understanding of the dynamic error
> free RBN.
> In order to explain this graphically I will use the DR1A case from CQWW CW
> 2012 on 40 meters.
>
> DR1A was the most spotted station on 40 meters.
> DR1A was also busted many times and those busted spots were sent to the RBN
>
> Overall there were 8223 spots coming from the real DR1A on the bands:
>
> 8114 DR1A spots
>
> 109 bust spots of DR1A with the following occurrences - DR1S (24), NR1A
> (15), DR1T (14), IR1A (11), DR1I (8), DM1A (8), DR4A(7), MR1A (4), KR1A
> (4), LR1A (3), DR2A (2), DK1AX ( 2), DR1U (2), DD1A (1), DR1AE (1), XR1A
> (1), DR1N (1) and HR1A (1)
>
>
> As you can see RBN skimmers are very "creative" when it comes to busting
> good calls. But bear in mind that Skimmer is more accurate than humans.
> Humans are even more creative when it comes to busting calls, as can be
> seen in CQWW rpt log checking reports.
>
> CT1BOH dynamic error free RBN algorithm catches this dynamic flawlessly.
>
> CT1BOH algorithm output for all these 8223 spots is the following:
>
> Good Call 7643 92.9%
> Good call, New Frequency 468 5.7%
> DR1A Busts 109 1.3%
> ?Spots 2 0.0%
> Busted 1 0.0%
>
>
> Let's put only the good spots in a graph. Click the links below to see.
> http://www.qsl.net/c/ct1boh//dl/good.png
>
> The X axis shows the 48 hours of the contest from minute 1 (00:00 Saturday)
> until minute 2880 (23:59 sunday).
> The Y axis shows frequency.
>
> The scattered dots in the first three hours is DR1A trying to find a good
> run frequency. If I zoom only for the first three hours this becomes more
> clear
> http://www.qsl.net/c/ct1boh//dl/first3.png
>
> Let's go back to the original graph with the good spots during the 48
> hours, and add the 109 DR1A Busts that Skimmers generated
> http://www.qsl.net/c/ct1boh//dl/goodandbusts.png
> CT1BOH algorithm catches all the busted spots and can filter them out of
> the network
>
> Let's now add to the graph the "Good call, New frequency?" spots. By now
> you know that these spots are the first and the second spots when a new run
> frequency is established (these are good spots - I know they were 49 spots
> out of the 468) and the true bad frequency spots from uncalibrated skimmers
> (I know they are 419 spots)
> http://www.qsl.net/c/ct1boh//dl/goodbustgnf.png
>
> Let's just go a bit deeper in this graphical analysis.
> Where are all these bad frequency spots coming from?
> 93 different skimmer from around the world spotted DR1A or busted DR1A
> call.
> But not all of them were uncalibrated.
> http://www.qsl.net/c/ct1boh//dl/skimmers.png
>
> If I remove from the graph the six uncalibratted skimmers (MM1PTT/P,
> F5MUX, HA6PX, DJ4DI, G4HYG, W2RDX)
> The graph goes from this http://www.qsl.net/c/ct1boh//dl/goodbustgnf.png
> into this http://www.qsl.net/c/ct1boh//dl/remove6skimmers.png
>
> Amazing isn't it. Just removing six uncalibratted skimmer makes a huge
> difference.
> One final graph. If we now look at all the spots from those six skimmers
> http://www.qsl.net/c/ct1boh//dl/datafrom6skimmers.png
> we clearly see, that they still produce a lot of good spots. They just have
> to fix that IQ image N4ZR explained in his post.
> Until then the dynamic error free RBN algorithm will easily filter them out
> on the spot.
>
> By now, I hope I have explained how effective a dynamic free RBN algorithm
> works.
> This algorithm is so effective because the network is so big that the good
> spots will always outnumber the bad spots.
>
> Also I would like to explain the price of running this algorithm
> The price is the first two good spots in a run until the third comes to
> validate it is a good call.
>
> In the case of DR1A the price is 49 spots that are filtered out so that we
> can drop 520 bad spots
> Because the network is so big, this price is negligible, i.e. even if we
> drop the first two spots of a run, chances are that the third spot will be
> in the same minute.
>
> To answer UA9CDC:
> Parasitic AM Modulation will be much less spotted that the good signal. The
> good signal will outnumber them, and since they are +/- 1.5 KHz away they
> will be dropped as "Good call, New Freq?" spot
>
> Second and third harmonic will be spotted as good, just like they are
> today. These are very, very, insignificant cases in the overall number of
> spots. The algorithm could be enhanced to try to predict these cases, but I
> don't advice doing it. I'd rather have a false positive than a false
> negative.
>
> No problem with MM stations. The algorithm runs per band. My code was just
> for one band for demonstration purposes.
>
> I understand VE7CC-1 just implemented his version of the dynamic erro free
> RBN. Also I understand N1MM and AR cluster is also implemented.
> Soon, busted spots ans wrong frequency spots will be a thing of the past
> from this extraordinary tool - Reverse Beacon Network
>
> 73 José Nunes
> CT1BOH
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
>
> --
> José Nunes
> CONTEST CT1BOH - http://www.qsl.net/ct1boh
> _______________________________________________
> CQ-Contest mailing list
> CQ-Contest@contesting.com
> http://lists.contesting.com/mailman/listinfo/cq-contest
>
>
--
José Nunes
CONTEST CT1BOH - http://www.qsl.net/ct1boh
_______________________________________________
CQ-Contest mailing list
CQ-Contest@contesting.com
http://lists.contesting.com/mailman/listinfo/cq-contest
|
