Wednesday, April 25, 2012

Some Thoughts on Accuracy

In order to correctly evaluate the accuracy of the RBN versus traditional spotting, I think you need to start with the idea that the RBN is not a traditional spotting network.  Here's what I mean.

Accuracy of a properly set up CW Skimmer in copying callsigns runs right around 99 percent.  There is an as-yet-unresolved problem with copying callers as if they were runners - more on that below.  But let's assume the 99 percent is right for an individual Skimmer, one not assailed by local RFI.

If you consider the RBN, on an open band, with no spot filtering by spotter location, then the picture changes.  Suppose you have 30 Skimmers in zones 4 and 5, all copying an open band.  Each running station will be spotted every 10 minutes by each RBN station, so long as the station remains on its run frequency.  You could, theoretically, have as many as 180 spots of each running station every hour.  In that case, simple probability says there will be roughly 1.8 busted spots per hour of each one.

If you are sitting at a big multi-op, receiving spots from all over the world, or just all over the US, then it is almost inevitable that you will see busted spots a lot more often than from the Cluster network.  Not only are there 10-15 times as many spots, but there will be busts of many running stations, just by the math.  On Sunday, you will see tons of busts, because by that time you will have worked most of the good callsigns, leaving only the busts on your bandmaps.

There are several partial solutions, not all of which will be implementable at a big multi, but let me mention a few:

*  Filter by spotter location, so that you only get spots from stations who are probably hearing the same things as you can.  For example, I filter by spotterstate = MD or PA or VA or NC or WV (have to get my own spots too).  That cuts down the number of Skimmers feeding my bandmap to 7-8, sharply reducing the probability of busts.

*  Use the "Unique > x" filter in AR Cluster version 6.  That filter only passes a spot if at least x+1 Skimmers worldwide have copied the same spot the same way in a relatively short period of time.  This helps a lot to winnow out busts, as you can imagine.  The RBN has a very robust ARC6 node at arcluster.reversebeacon.net, port 7000.

*  Use a logging program that permits displaying unworkable spots (already worked, or not workable in a given contest) on the bandmap.  For example, N1MM Logger displays such spots in gray.  The advantage of doing this is that if LZ9W is running, but you worked him 30 hours ago, and LZ9WL suddenly shows up on the same run frequency, you can see that it's pretty likely to be a bust.  This also helps with callers mistakenly identified as runners; if you see calls appearing one after another on the same frequency as one of the big runners, you're probably safe in skipping past them.


We hope this is helpful.

73, The RBN Team