A radio link is calculated before you buy it.

It comes from the world of operators who take internet where there is no cable, and it shows: equipment designed to sit on a mast for years, with real calculation and alignment tools. There are two ways of using it. Point to point joins two places: one antenna here, another there. Point to multipoint puts one antenna opening a sector from a tall mast and several small points looking towards it, which is the way to take a network to six huts on an estate without six separate links.

In security it turns up when the civil works are impossible or absurd: a plot on the other side of a road, a quarry, a solar farm, a port, a site that is going to last two years. In those cases the alternative is not another brand: it is not doing it.

The two antennas have to see each other, and everybody knows that. What is less well known is that it is not enough: around the straight line joining them you have to leave a clear space, a sort of spindle that widens in the middle of the span, and if anything comes into that space the link degrades even though the antennas can still see each other. That includes the roof of the building in between, the parapet of a rooftop, the edge of a hill and, very particularly, trees.

Trees deserve a paragraph of their own because they are the number one cause of links that stop working without anyone touching anything. A poplar in February, with no leaves, lets the signal through; the same poplar in July is a wall of water. And it grows too: the link installed with the tree just below the line works for three years and then does not. When there is vegetation in the path, the design counts on the summer and on the growth, or the antenna goes higher.

A link antenna concentrates the energy into a beam, and the more gain it has, the finer you have to aim. Aiming «towards that building over there» by eye can leave the link working, and that is exactly the problem: it works, nobody suspects anything, and it runs with half the margin it should have. It is aligned using the equipment's own reading, moving the antenna slowly, looking for the maximum and comparing it with the value that came out of the calculation. If the calculation said one thing and the measurement gives a good deal less, there is something in the path that was not on the drawing. And there is a classic trap on high-gain antennas: there are secondary positions, to the sides of the good one, where the equipment also locks on and the signal looks acceptable. A link aligned there works on installation day and drops with the first rain.

And one thing about the mounting, not the radio: the bracket. A mast that moves in the wind or a clamp only half tightened undo a well-made alignment in a month. The link is lost and nobody thinks about the bolt.

Interference first, because it is the most frequent in the unlicensed bands, which are the ones most of these links use. Those bands belong to everyone: the industrial estate next door, the neighbour putting up theirs, the installation that gets extended. Your link can degrade on a Tuesday in November because somebody two kilometres away put theirs up on the same channel, and there is nothing you did wrong. That is why you look at what is in the air before choosing a band and a channel, and that is why the link is left ready to change channel without dismantling it.

Rain next. In these bands it subtracts but it is not what decides: it weighs more on long spans and with heavy rain. What does do a lot of damage here is water where it should not be —a badly sealed antenna, an unprotected connector—: the mechanical installation decides more faults than the radio does.

And the margin, which is the central idea of this page. A well-designed link is not designed to work: it is designed to work with plenty to spare, because everything above pulls the same way and adds up over time —the tree grows, the channel fills, the bracket gives, the first real winter arrives—. A link that on installation day is tight and gets accepted because «it works» is a link that is going to fail, and it is going to fail on the worst night of the year.

When a mast shares out to several points, that sector has one capacity and it is shared between all of them. For browsing that works well, because people browse in bursts. Cameras do the opposite: the same rate twenty-four hours a day, all at once and in the same direction.

And the direction is the problem. The split between the downward direction —from the mast to the points— and the upward one —from the points to the mast— is not half and half: it is configured, and by default it usually favours the downward one, because that is how people browse. Video goes in the upward direction, which is the small one. A link that gives a very good figure on a speed test can give a third of it in the direction you really need. It can be adjusted, but you have to know it exists and do the sum with the upward figure and not with the brochure one.

Our rule in these cases: when there are several cameras on the other side, the video is recorded at the point and what travels over the radio is what is needed —alerts, whatever somebody wants to watch, and the retrieval of the pieces that get asked for—. That way a bad radio day is a day without live viewing and not a day without recording.

It fits where the trench is impossible, forbidden or absurd: crossing a road, a railway, a river, someone else's land. In extensive installations with few buildings —quarries, solar farms, farms, ports—. In the temporary: a site, an events ground, a plot while a line is being processed. And as a second route for a fibre that already exists, so that a cut on a construction site does not leave the installation blind.

We would not fit it when there is a viable fibre route and the radio is chosen because this month it comes out cheaper: fibre is paid for once and does not notice the rain, the trees or the neighbour. We would not fit it without line of sight, and in particular when somebody proposes «going between the trees»: that is not a link, it is a link until July. We would not fit it without having measured the spectrum and done the profile of the span, because then we are not designing, we are experimenting with the client's money. And not as the only path for all the video of a large installation without recording at the far end.

Ordinary administration is run by any IT team: look at the state of the link, the quality of the signal and how the last month has gone, update, restart, change channel when interference turns up. We hand over the administrator account in the client's name, the configuration saved and —this is the important part— the calculation report and the signal measurement from the day of commissioning. Without that report, in two years' time somebody will look at the signal, see a number and not know whether it is good or bad; with it, comparing takes a minute.

The boundary is in two places. The calculation of a new link or the modification of an existing one —changing band, moving an antenna, lengthening the span— is done with the terrain profile and the spectrum measurement in front of you; by eye you get a link that works and does not last. And the alignment, which is mast work with an instrument.

And if your team already works with another radio brand, we fit in with it. The brand is the least important part of a radio link: what decides whether it works is the span, the height, the band and the margin, and that does not change with the logo on the box. What we do not do is sign off a link without that calculation done, whatever the brand.