The two couplings no design stage owns

Height is bought in metres and paid for in decibels. Gain is bought in metres and paid for in degrees. Each stage passes on its own.

Every stage passes and the link still underperforms

A microwave hop is designed in stages, and each stage has an owner, a document and a pass mark. The path survey says the obstacle is cleared. The structural analysis says the tower stands. The antenna comes off a datasheet. The link budget closes with margin to spare.

All four are correct. The link can still fall short, because the things that connect the stages are not in any single stage's remit — and the two biggest ones both work in the same direction, which is against you.

Height is bought in metres and paid for in decibels

When a profile comes up short of clearance, the standard fix is to raise both antennas. On a link with an indoor radio, that fix arrives with an invoice.

Every metre of extra tower is a metre of extra feeder — at both ends. At 18 GHz in elliptical waveguide the run costs roughly 0.13 dB per metre, so raising both towers by 28 m adds around 7 dB of loss to the budget.

Seven decibels is not a detail. It is the same order as the whole fade margin the height was bought to protect. It is entirely possible to fix a clearance problem and end up with a worse link than you started with, and for every individual calculation along the way to have been right.

This is the reason split-mount radios displaced indoor units. Put the transceiver at the antenna and only power and data run down the tower; there is no RF feeder, height costs nothing in the budget, and the coupling simply stops existing.

Gain is bought in metres and paid for in degrees

The second coupling is less well known and harder to spot, because it looks like the safe move.

Three facts, each unremarkable on its own:

• Antenna gain rises with the square of diameter. Double the dish, gain up 6 dB.
• Beamwidth falls with the inverse of diameter. Double the dish, beam half as wide.
• Loss from a pointing error rises with the square of that error measured in beamwidths: about 12·(θ/θ_3dB)² dB.

Put them together and doubling the diameter adds 6 dB of gain while multiplying the loss from a given tower deflection by four. In decibels, gain grows logarithmically with diameter and pointing loss grows quadratically. Those curves cross — and past the crossing, every extra centimetre of dish makes the link worse.

At 18 GHz on a tower that deflects a quarter of a degree at design wind, the crossing sits at about 2.8 m. A 6 m dish there carries 6.6 dB more gain on its datasheet and delivers nearly 9 dB less into the link. The tower did not change. Nothing was installed incorrectly. The antenna is doing exactly what its pattern says it does.

The number that decides it lives in another document

An antenna datasheet quotes boresight gain: measured on a range, pointed perfectly, in still air. That figure is right, and it is not what the link receives.

What the link receives depends on how far the tower moves at full design wind — a number that lives in the structural report, in degrees, produced by an engineer who was not asked about radio. The radio planner has the gain and not the deflection; the structural engineer has the deflection and no reason to care about beamwidth.

A stiff monopole might hold a tenth of a degree. A guyed mast with a large dish mounted high can be several times that. On a 0.3° beam — a 3.6 m dish at 18 GHz — one decibel of pointing loss is only 0.09° of movement, which installation tolerance and long-term drift both have to fit inside.

Coordinates, and the precision you are actually carrying

Survey reports, licence applications and the plate on the side of a tower give coordinates in degrees, minutes and seconds. Google Maps gives decimal. Every calculator wants one or the other, and the retyping in between is where a digit goes missing — which on a longitude is a kilometre.

The coordinate fields here take either, in most of the ways either gets written: 51°28'40.5"N, 51 28 40.5 N, N51:28:40.5, 51°28.675'N or 51.477928. The format buttons only change what is displayed; whatever you paste is understood.

What is worth knowing is how much precision each notation actually carries, because converting between them invents digits that were never there:

• A coordinate to whole seconds locates a point to about 15 metres.
• A tenth of a second is about 1.5 m.
• Four decimal places is about 5.6 m; five is 0.6 m; six is 6 cm.

So a position read off a document as 51°28'40"N and written back as 51.477778 has quietly claimed a tenth of a metre for something good to fifteen. The fields show the precision you actually entered for that reason.

One asymmetry worth remembering: a second of latitude is about 31 m everywhere, but a second of longitude shrinks with the cosine of latitude. At 60° it is half that. A longitude written to the same number of places as its latitude is therefore twice as precise on the ground at those latitudes, not equally precise.

And a sign with a hemisphere letter that disagrees — -51°N — is refused rather than resolved. Guessing which one the writer meant is how a site ends up in the wrong half of the world, and the path not closing is a poor way to find out.

Two fade mechanisms, routinely muddled

Ask what takes a microwave link down and you get two answers that are usually run together. They are separate physics, they have separate recommendations, and confusing them produces designs that are protected against the wrong thing.

Multipath, and why Vigants-Barnett is not a rain model

Vigants-Barnett describes multipath. Still, layered air refracts the beam into several paths that arrive slightly out of step and interfere. It is the fade that happens on calm, clear nights over water — not during storms.

Its terrain and climate factors are about how readily the air over a particular path layers: smooth water and coastal humidity are the worst case (C = 4), rolling inland terrain the baseline (C = 1), broken mountainous ground the best (C = 0.25), because rough terrain mixes the air and breaks up the layers. None of that is about rainfall, and using those factors as though they were rain factors gets the answer wrong in both directions.

The shape of the model is worth carrying away on its own:

• Outage goes as the cube of path length. Doubling a hop multiplies multipath by eight.
• It goes linearly with frequency.
• Every 10 dB of fade margin divides it by ten.

The cube is the one that surprises people. Free-space loss only goes as the square of distance, so a long hop is harder than the link budget alone ever suggests — the budget and the fade model do not scale together.

Rain, which is ITU-R P.838 and P.530

Rain attenuation is a different calculation entirely. Specific attenuation comes from the rain rate as γ = k·R^α dB/km, with k and α tabulated against frequency and polarisation in P.838.

Then P.530 applies a path reduction factor, and this is the part hand calculations leave out. A rain cell heavy enough to matter is only a few kilometres across, so on a long hop it never covers the whole path at once. A 20 km link at 18 GHz has an effective rain length of under 10 km — leaving the factor out roughly doubles the predicted attenuation and produces a link nobody would build.

The reference figure the whole method hangs on is the rain rate exceeded for 0.01% of an average year, about 53 minutes. It comes from P.837 rain-zone maps for the region, and there is no sensible default: a temperate maritime 42 mm/h is wrong for most of the world, in both directions.

Polarisation is a free lever

Falling raindrops are not spheres. Air resistance flattens them into oblate shapes, so they present a wider profile to a horizontally polarised wave than a vertical one. Vertical polarisation therefore suffers roughly 10 to 20% less rain attenuation, depending on frequency.

On a 20 km hop at 18 GHz that is the difference between about two hours of rain outage a year and about seventy minutes — for a choice that costs nothing at all, provided it is made before the antennas are mounted and does not conflict with an interference plan.

Which one governs, and why you must not add them

Frequency decides. Below about 10 GHz multipath dominates and rain barely registers. Above about 15 GHz rain dominates and multipath is secondary. Between them both matter.

The two are not added. Multipath needs still, layered air; heavy rain does not occur in still layered air. They are close to mutually exclusive events, and summing them double-counts. Taking the worse of the two is the usual convention and the honest one.

Seeing it in the world

A path drawn on a map tells you what you already knew. A path drawn in Google Earth at the real altitude of the antennas, with the first Fresnel zone as a shape around it, shows you the tree line that becomes a problem in four years.

That only works if the geometry is genuinely three-dimensional and absolute. Heights must be above mean sea level rather than above the terrain, or the beam gets draped over Google Earth's own ground model and the whole point is lost. The export writes sites, masts drawn from ground to antenna, the sampled line of sight, both edges of the Fresnel zone and the clearance limit as its own line — so the question "does the ground come inside 60% of F1" can be answered by looking.

What to actually do about it

Get the deflection figure before choosing the antenna. It is a single number and it changes which antenna is correct.

Prefer split-mount where the band allows it. It removes the height coupling entirely, and above about 6 GHz the feeder loss it avoids is substantial.

Check whether a smaller dish is better. On a soft tower it frequently is, and it is cheaper, lighter, catches less wind — which in turn reduces the deflection that caused the problem.

Treat the datasheet gain as an upper bound. Build the budget from effective gain, or accept that the margin on paper is larger than the margin in the field.

What a lab of this kind cannot tell you

It models one hop, in clear air, with a single controlling obstacle and identical antennas at both ends. Rain fade is a separate mechanism and dominates above about 10 GHz. Diffraction loss from a partially obstructed path is not in the budget. Interference, adaptive modulation, and multipath statistics beyond the simple exponential are all absent.

And deflection is an input, never an output. No radio tool can produce that number; it comes from the tower analysis, and using a guess for it defeats the purpose of looking.

Frequently asked questions

Why does clearing a hill cost fade margin?

Because raising both antennas adds feeder at both ends. At 18 GHz in elliptical waveguide that is about 0.13 dB per metre, so 28 m of extra height on both towers costs roughly 7 dB — the same order as the fade margin the height was bought to protect. Split-mount radios remove the coupling because there is no RF feeder to lengthen.

How can a bigger antenna make a link worse?

Gain rises with the square of diameter, beamwidth falls with its inverse, and pointing loss rises with the square of the error measured in beamwidths. Doubling the dish adds 6 dB of gain and quadruples the loss from a given tower deflection. In decibels gain grows logarithmically and pointing loss quadratically, so the curves cross. At 18 GHz on a tower holding 0.25°, a 6 m dish has 6.6 dB more gain than the 2.8 m optimum and delivers nearly 9 dB less.

Where does the tower deflection figure come from?

The structural analysis, in degrees at full design wind. No radio tool can produce it. A stiff monopole might hold 0.1°; a guyed mast with a large dish high up can be several times that. Guessing it defeats the purpose of checking.

Is the datasheet gain wrong?

No — it is boresight gain, measured on a range, pointed perfectly, and it is correct. It just is not what the link receives, because the link's antennas move. Build the budget from effective gain or accept that the paper margin exceeds the field margin.

What is the 12·(θ/θ₃dB)² approximation?

The standard main-lobe model for pointing loss. By construction it gives 3 dB at half the 3 dB beamwidth, which is what that beamwidth means. It holds well inside the main lobe; past about one full beamwidth off axis the real pattern is sidelobes and the number stops describing anything.

Would a smaller dish really be better?

On a soft tower, often yes. It is also cheaper, lighter and presents less wind area — which reduces the deflection that caused the problem, so the benefit compounds. It is worth checking rather than assuming.

Can I enter coordinates in degrees, minutes and seconds?

Yes. The coordinate fields accept DMS, degrees with decimal minutes, or plain decimal degrees, in most of the ways each gets written — 51°28'40.5"N, 51 28 40.5 N, N51:28:40.5, 51°28.675'N or 51.477928. The format buttons change only what is displayed; whatever you paste is understood.

How precise is a coordinate given to whole seconds?

About 15 metres. A tenth of a second is about 1.5 m. In decimal, four places is about 5.6 m, five is 0.6 m and six is 6 cm. Converting DMS to decimal invents digits — a point read as 51°28'40"N and written back as 51.477778 claims a tenth of a metre for something good to fifteen — so the fields report the precision actually entered.

Why is a second of longitude worth less than a second of latitude?

Because meridians converge. A second of latitude is about 31 m everywhere; a second of longitude is that times the cosine of the latitude, so at 60° it is about 15 m. A longitude written to the same number of places as its latitude is twice as precise on the ground at those latitudes.

Is Vigants-Barnett a rain model?

No. It is a multipath model. It describes still, layered air refracting the beam into interfering paths — the fade that happens on calm clear nights over water, not during storms. Its terrain and climate factors are about how readily the air layers, not about rainfall. Rain has its own recommendations, ITU-R P.838 and P.530, and the two are calculated separately.

Why is rain attenuation less than the specific attenuation times the path length?

Because of the path reduction factor. A rain cell heavy enough to matter is only a few kilometres across, so on a long hop it never covers the whole path at once — a 20 km link at 18 GHz has an effective rain length under 10 km. Leaving the factor out roughly doubles the predicted attenuation and produces a link nobody would build.

Should I use vertical or horizontal polarisation?

Vertical, for rain, if nothing else decides it. Falling raindrops flatten into oblate shapes and present a wider profile horizontally, so vertical polarisation suffers roughly 10 to 20% less rain attenuation. On a 20 km hop at 18 GHz that is about two hours of outage a year against about seventy minutes, for a choice that costs nothing.

Do rain and multipath outages add together?

No, and adding them double-counts. Multipath needs still, layered air and heavy rain does not occur in still layered air, so the two are close to mutually exclusive events. The usual convention is to take the worse of the two, which is what this does.

Is anything sent anywhere?

No. Path lengths, frequencies, site data and equipment figures are calculated in your browser and are not uploaded.