Anatomy of a telemetry ground station

What has to happen in under a second

A telemetry ground station has one job: take a radio signal off an aircraft and put a number in front of an engineer fast enough for that engineer to act on it. Everything in it is a means to that end, and the deadline is set by the test itself — a flutter point, a stall approach, an envelope expansion step gives you seconds, not minutes.

The propagation delay is a millisecond. Everything else in that budget is equipment: demodulation, bit synchronisation, frame lock, decommutation, and the refresh of a display. A well-built chain lands in the tens of milliseconds; a badly built one adds buffering at every stage and hands the test conductor a picture of the recent past.

The antenna, and why tracking decides everything

Antenna gain and beamwidth are the same choice seen from two sides. A 2.4 m dish at S band has a beam about four degrees wide — which is what gives it the gain to close the link at two hundred kilometres, and what means it must be pointed within a couple of degrees or you have nothing at all.

So the pedestal and its autotracking are not an accessory to the receiver, they are the first stage of the measurement. Tracking works by comparing the signal across slightly offset beams — conical scan or monopulse — and steering to null the difference. It needs a signal to track, which is why acquisition is a separate problem: a wide-beam acquisition aid, a slew to a predicted position from the flight plan, or an operator with a hand controller and good reflexes.

Two smaller practical points that decide more campaigns than they should: the low-noise amplifier belongs at the feed, not at the end of thirty metres of coaxial cable, and the pointing calibration deserves to be redone rather than assumed.

The receiver, and the value of a second antenna

The receiver down-converts, filters and demodulates. Its intermediate-frequency bandwidth has to match the occupied bandwidth of the signal: too wide and you are adding noise for nothing, too narrow and you are clipping the spectrum of your own data. It is a setting, it is on a menu, and it is wrong more often than anyone admits.

Demodulation depends on the modulation. A classical PCM/FM link uses a discriminator; a multi-symbol trellis detector on the same signal buys two to three decibels for free, which at the edge of the range is the difference between data and a dropout. SOQPSK needs its own demodulator and a receiver that knows about it.

The most cost-effective improvement to any ground station, though, is a second antenna. An airborne antenna has deep nulls and the airframe masks it, so the signal fades in a way that is uncorrelated between two receive polarisations or two ground sites. Selecting the better of two sources removes most dropouts; combining them before detection adds up to three decibels on top. Either is worth more than a bigger dish.

Bit and frame synchronisation

The demodulator hands over a stream of bits with no boundaries. The bit synchroniser recovers the clock — which is why the transmitted stream is randomised or bi-phase coded, so there are always transitions to lock to. The frame synchroniser then finds the frame pattern, verifies that it recurs exactly one minor frame later, and declares lock.

Both stages have tolerance settings, and both extremes cost you. A frame lock that drops on a single bit error throws away frames that were perfectly usable; one that is too forgiving locks onto a pattern that happens to appear inside the data and delivers convincing nonsense. The right setting depends on the link, which means it is a campaign decision rather than a factory default.

And both stages produce something worth recording: the number of sync losses, and their duration. That is the honest measure of link quality, it costs nothing to log, and it is the only thing that will settle the argument about whether the data was good during the third point.

Decommutation: from a stream to parameters

Decommutation applies the format: this word position is that parameter, sub-commutated words are reassembled across minor frames, super-commutated ones are interleaved back into a single time series, and the raw counts are converted into engineering units with the calibration.

Which means the ground station holds a copy of a contract written onboard, and the whole chain is only as correct as that copy. A format that has drifted from the aircraft configuration does not produce an error message; it produces plausible values in the wrong places. Loading the setup record that came with the recording, rather than the one in somebody folder, is the cheapest protection there is.

Beyond the format, this is also where derived parameters are computed in real time — a load factor from accelerations, a margin from a measured and a limit value, a flag when two channels disagree. Those are often what the test conductor actually watches.

The display nobody has time to read

A thousand parameters arrive; a human can act on perhaps six. The design of a real-time display is therefore an exercise in subtraction, and the most common failure of a ground station is a wall of numbers that is technically complete and operationally useless.

• One screen per test point, designed before the flight, containing what would cause an abort and nothing else.
• Limits and envelope masks drawn on the plot rather than checked mentally — the eye catches a curve crossing a line far faster than it reads a number.
• Strip charts for anything with a trend, digital readouts only for what genuinely needs a value.
• An explicit indication of link and data health, so nobody argues about whether a flat trace is a flat parameter or a lost signal.
• And the same time reference as the aircraft, displayed, because every call the conductor makes is timestamped by somebody.

Recording on the ground, and replaying

The ground station records too, and it should record the raw bit stream as received, not only the parameters it decommutated. The reason is simple: a wrong format, a wrong calibration or a missing derived parameter can all be fixed afterwards and re-applied to a raw stream. None of them can be fixed on a file that only contains the result.

That ground copy is also the insurance policy. If the aircraft is lost, or a cartridge is damaged, or the onboard recorder simply did not start, the telemetered subset is what remains — degraded, partial, and infinitely better than nothing.

Replay closes the loop. Feeding a recorded stream bit-for-bit back through the same synchronisers and decommutator is how you validate the ground chain before a campaign, how you reproduce an anomaly, and how you re-run an analysis after correcting a format. A station that cannot replay its own recording is a station you cannot debug.

What breaks a ground station

• Pointing and tracking: a keyhole pass, an acquisition that never completed, a calibration nobody redid.
• An intermediate-frequency bandwidth that does not match the signal, quietly costing decibels all campaign.
• A format that has drifted from the aircraft — the failure that produces data instead of an error.
• No signal-strength and no sync-loss logging, which turns every dropout into an opinion.
• A ground time base that is not locked to the same reference as the aircraft, which makes correlation impossible after the fact.
• And recording only the decommutated result, which throws away every chance of a second attempt.

None of that list is exotic, and none of it is about the quality of the equipment. A ground station is a chain of well-understood stages, and it fails at the settings and the housekeeping — which is also why a day of end-to-end validation before a campaign pays for itself the first time it is needed.