The sum of gains minus the sum of losses, compared against the receiver's sensitivity - and the difference is the fade margin that keeps the link up in bad weather.

Add the transmitter's output power and the gain of both antennas. Subtract the free space path loss for the distance and frequency, plus losses in cables and connectors at each end.

The result is the received signal level. Compare it against the receiver's sensitivity for the data rate you want - sensitivity is worse at higher rates, which is why a link can be reliable at a modest speed and unreliable at its maximum.

The margin between them is the fade margin. Rain, fog, humidity, foliage and small alignment drift all eat into it, and a design with a small margin will be intermittent in weather.

Aim for a substantial margin rather than the minimum that works on the day of installation. Vendors publish planning tools that do the arithmetic; the judgement is in how much margin the application deserves.

The radio path is an ellipsoid around the straight line, not a line - and an obstruction inside it degrades the link even when the antennas can see each other.

Radio energy travels in a three-dimensional zone that is widest at the midpoint of the link. Objects intruding into that zone reflect and diffract the signal, and the reflections can arrive out of phase and cancel part of the direct signal.

The practical consequence is that visual line of sight is not sufficient. A link that just clears a treeline may have half its Fresnel zone obstructed and perform far below prediction, with no obvious cause.

The zone is larger on longer links and at lower frequencies, so a long link needs substantially more height than intuition suggests.

Trees are the classic problem because they grow, and because they are much worse when in leaf and wet - which is why some links degrade seasonally.

Calculate the clearance at the design stage and add margin for growth. Raising a mast later is expensive.

Unlicensed for most links because it is quick and cheap; licensed where the link is critical or the band is congested.

Unlicensed bands need no application and no fee, and equipment is inexpensive and widely available. The trade is that anyone else may use the same spectrum. In a city, unlicensed bands can be crowded enough to make a link unreliable in ways nobody controls.

Licensed bands give exclusive use of a frequency on a path, coordinated by the regulator so no one else may interfere. That buys predictability and is the right answer where the link carries something that must not fail. The costs are the licence fee, the application process, and more expensive equipment.

Regulations differ by country, including power limits and which bands are available, so confirm locally rather than assuming.

A reasonable approach is to survey the spectrum at both ends first. If the unlicensed band is clean at that location, it may serve for years; if it is busy, licensed is worth the cost.

Coarsely by sight or compass bearing, then finely using the equipment's own signal readout - and both ends must be done, iteratively.

Start with the calculated bearing and elevation, and get both ends roughly pointed. Most links will establish at this stage, well below their potential.

Then refine using the received signal level the equipment reports. Adjust one end slowly through the peak while watching the reading, lock it, then do the other end, then return to the first - because improving one end changes the optimum at the other.

Beware side lobes: antennas have secondary peaks, and it is entirely possible to align onto one and get a working but much weaker link. Sweeping through the full range rather than stopping at the first peak avoids it.

High-gain antennas have narrow beams, so alignment is more critical and more sensitive to mast movement afterwards.

Record the final signal levels. They are the baseline that shows whether a later problem is alignment, weather or equipment.

From tens of megabits to several gigabits depending on equipment, band and distance - and the honest figure is the one the link achieves in poor weather, not the best.

Modern bridges in unlicensed bands routinely deliver hundreds of megabits over a few kilometres with good clearance. Millimetre-wave equipment can exceed a gigabit over shorter distances. Licensed microwave links can be engineered for whatever the application needs.

Throughput falls with distance, with reduced fade margin and with interference, because the link drops to a more robust modulation. A link that runs at its maximum on a clear day may run at a fraction of that in heavy rain, and the design should be based on the worst acceptable case.

Higher frequencies carry more capacity but are attenuated far more by rain, which is why millimetre-wave links are short and engineered with substantial margin.

Specify the required throughput at the required availability - a figure such as ninety-nine point nine per cent of the year - and let that drive the equipment and band choice.

Very - a high-gain antenna has a narrow beam, and mast movement of a degree or two measurably degrades the link.

The higher the gain, the narrower the beamwidth. A large dish may have a beam only a few degrees wide, so movement that would be irrelevant for an access point takes the link off its peak.

Masts flex in wind, buildings move thermally, and lightweight brackets twist. On long high-gain links this shows as throughput that varies with the weather in a way that looks like rain fade but is actually mechanical.

So mount on something rigid: a substantial pole properly braced, or structural steel. Avoid tall thin poles and roof edges that vibrate.

Wind loading on a large antenna is significant and must be part of the structural assessment, not an afterthought.

Where very stable mounting is impossible, a lower-gain antenna with a wider beam is more forgiving - trading link margin for tolerance, which is often the right trade on a mast that moves.

Vegetation growth, alignment drift, connector and weatherproofing degradation, and new interference - roughly in that order.

Trees are the most common. A link installed with adequate clearance loses it as growth intrudes into the Fresnel zone, and the degradation is gradual and seasonal, which delays diagnosis.

Alignment drifts from mast movement, fixings loosening or a bracket corroding. It shows as a steady decline in received signal at both ends.

Weatherproofing degrades: tape systems perish in ultraviolet, water reaches a connector, and the loss increases. A corroded connector can cost several decibels.

New interference appears when somebody else deploys equipment in the same unlicensed band nearby, and it can be sudden.

All four are distinguishable if the installation baseline was recorded. Trend the received signal level and the modulation rate; a slow decline points to alignment or connectors, a seasonal pattern to vegetation, and a step change to interference.