A machine learning model to predict DX propagation in the 20 metre amateur radio band at Vancouver, British Columbia.
Applicable Scenarios and Problems
(Note: this explanation assumes familiarity with amateur radio and CW/Morse code operation. If you're new to this, see "Background info".)
This model attempts to predict how many DX (in this context, non-North American) call signs will be heard by the VE7CC Reverse Beacon Network skimmer in a given hour on the 20 metre band. The model has been built using historical data from the Reverse Beacon Network.
VE7CC was picked because it's close to where I live; 20 metres was picked because it's the band I operate on most often.
Background info (or, "Wait, people still use Morse code?")
Amateur radio is a hobby enjoyed by millions of people worldwide, who use many different forms of radio to contact each other. Amateur radio operators are licensed to operate on different ranges of frequencies, known as bands. Bands are named after the wavelength of the frequency being used; thus, you have the twenty metre band (14.0 - 14.35 MHz), the forty metre band (7.0 - 7.3 MHz), and so on.
Each amateur radio operator has a call sign (similar to call signs
used by commercial radio stations) that identify the operator and what
country (and sometimes state or province) they are from. Mine, for
VA part identifies me as Canadian, and the
7 as being from British Columbia. There may be another call sign in
Canada ending in
UNX -- say,
VE3UNX (Ontario) -- but
unique to me.
One of the many specialties of amateur radio is Morse code operation, also known as CW. Someone who sends Morse code will call themselves a CW operator. While CW is no longer used for professional communications, it's still enjoyed as a fun challenge by operators.
CW has many abbreviations and standardized ways of communicating that are intended to save time. For example: to initiate a conversation with whoever might be listening, I would call CQ:
CQ CQ CQ DE VA7UNX VA7UNX VA7UNX K
Here, I am inviting anyone listening to contact me (
CQ CQ CQ),
telling them my call sign (
DE VA7UNX), with a
K at the end to
indicate I am standing by to hear any responses. The repetitions
ensure that, even if my signal is weak, it can be deciphered by a
Skimmers are computerized radios that will listen to amateur radio
bands for this standard call (
CQ CQ CQ DE [call sign]), and report
the call signs recognized. The Reverse Beacon Network collects the
output of these skimmers, and displays them on its web page; this
allows operators to see how far their signal has been received (and
thus how far they might be heard by other operators), and whose signal
is being picked up locally (and thus who they might talk to). The
Reverse Beacon Network records all this data, and makes it available
Propagation is how far, or how well, your signal travels. How far can it go? That depends! There are a lot of factors that go into it:
- the time of day, and the height of the sun
- the time of year
- the band being used
- how many sunspots there are (which in turn vary in an 11 year cycle)
- the power and equipment used to transmit
- whether there's a contest going on (more on that in a moment)
There are rules of thumb for how each of these affect propagation, but there can also be a lot of variation on the scale of years, weeks, days, hours or even minutes. On the 20 metre band, which this model focuses on, I can reach the east coast of North America from the west coast (about 4000 km/2500 miles) with only 5 Watts of power -- about the same amount of power as a night light uses. On a very good day, I can reach all the way to Europe (7500 km/4650 miles) or New Zealand (11000 km/6800 miles). On a bad day, I might get no further than California (1200 km/750 miles).
DX stations are stations that are from "far away" -- usually defined as "not this country". It's always fun to talk to someone a long way away, so contacts with DX stations are prized by amateur radio operators.
Finally, there are contests: short periods of time (a day or two is typical) where operators try to make as many contacts as they can; there are usually extra points for DX contacts. It's often much easier to make DX contacts during a contest, because everyone is on the air and willing to listen for weaker signals to try and get extra points for distant stations. The ARRL DX Contest is one of the largest such contests.
The model I built tries to answer the question:
- How many DX stations...
- ...will be heard by the VE7CC skimmer (which is located in Vancouver, BC)...
- ...on the 20 metre band...
- ...in a given hour on a given day of the year...
- ...given the number of sunspots...
- ...given whether the ARRL DX contest is on right now or not...
- ...and given how high the sun is in the sky?
(For simplicity, "DX" in the model is defined as stations from outside North America.)
With all that as input, it will return the model's estimate for how many DX stations will be heard in that hour by the Vancouver skimmer.
|sunspots_count||how many sunspots the sun at the given hour|
|hour||hour in UTC (0-23)|
|month||number of month (1-12)|
|annual||day of year / 366|
|weekend||Is it a weekend? 1.0 for yes, 0.0 for no|
|arrl_dx_contest||Is it the weekend of the ARRL DX Contest? 1.0 for yes, 0.0 for no|
|solar_elevation||The sun's azimuth (height in the sky) in degrees|
- All numbers must be floats (1.0, not 1)
sunspot_countcan be obtained from http://sunspotwatch.com
solar_elevationcan be obtained from Heavens Above
- Dates of the ARRL DX Contest can be found at the ARRL contest page; the next dates are February 15-16, 2020.
|prediction_20m||prediction of the number of DX stations that will be heard by the VE7CC skimmer in that hour|