PCM telemetry: frames, sync and link budget

From sensors to a single bit stream

An instrumented aircraft carries hundreds — sometimes thousands — of measurement points: strain gauges, accelerometers, thermocouples, pressures, control surface positions, plus everything already available on the avionics buses. All of it has to reach the ground through one radio carrier.

The answer, unchanged since the 1960s and still the right one, is time-division multiplexing. Instead of giving each parameter its own carrier, every parameter gets a slot in a repeating time sequence. Each sample is digitised into a fixed number of bits — that is the pulse-code modulation part — and the slots are transmitted one after the other as a single continuous serial stream.

Nothing in that stream identifies a parameter. There is no header, no tag, no name: a value is recognised purely by its position in the sequence. That is what makes PCM so efficient, and also why the whole scheme rests on the receiver knowing exactly where the sequence starts.

Words, minor frames, major frames

The elementary slot is the word, typically 8 to 16 bits. A 12-bit converter usually ends up in a 16-bit word, the spare bits carrying status or simply being wasted — alignment is worth more than a few percent of bandwidth.

A fixed sequence of words forms the minor frame. It carries a synchronisation pattern that marks the frame boundary and, almost always, a counter identifying the frame inside the cycle — the SFID. IRIG 106 bounds the minor frame: at most 8 192 bits and at most 1 024 words, which is generous but not infinite.

A set of N minor frames forms the major frame, N being the range of the frame counter. The complete format repeats once per major frame. That two-level structure is what lets a single stream carry parameters with very different sample rates.

Sub-commutation and super-commutation

A slow parameter does not deserve a slot in every minor frame. Give one slot to a different parameter each minor frame and you divide its sample rate by N: that is sub-commutation, and it is how a hundred temperatures fit into the space of one fast channel.

The opposite works too. Give a parameter several slots in the same minor frame and you multiply its rate: that is super-commutation, used for vibration or anything with real dynamics. Spacing those slots evenly across the frame matters — unevenly spaced samples produce jitter that no amount of filtering on the ground will remove.

The map from slot position to parameter is the format. It must be identical onboard and on the ground, which is exactly what the TMATS setup record carries in an IRIG 106 recording. Get the format wrong and the data does not look wrong — it looks plausible, which is far worse.

Choosing the sample rates

Two lines of arithmetic drive the whole design. The frame rate is the bit rate divided by the minor frame length in bits. The sample rate of a parameter is the frame rate multiplied by the number of slots it holds in a minor frame — or divided by the depth of its sub-commutation.

Take a 5 Mbit/s stream with minor frames of 512 words of 16 bits, so 8 192 bits per frame. The frame rate is 610 frames per second. One slot gives a 610 Hz parameter; eight evenly spaced slots give 4 883 Hz; a parameter sub-commutated over 32 minor frames gives 19 Hz. Everything else is bookkeeping.

The classic mistake is to set sample rates from the Nyquist limit. Twice the highest frequency of interest is the theoretical floor, not an engineering target: five to ten times is what produces time histories an engineer can read, and it is what leaves room for filtering and resampling later.

Frame synchronisation: how the ground finds the start

The receiver gets a continuous stream of bits with no natural boundary. It first has to recover the bit clock, then find the frame. Both steps have practical consequences on the way the stream is built.

Clock recovery needs transitions. A long run of identical bits starves the recovery loop and the receiver drifts, which is why raw NRZ-L is rarely transmitted as is: IRIG 106 specifies a 15-bit randomiser that guarantees transitions without adding overhead, and bi-phase codes guarantee one transition per bit at the cost of twice the bandwidth.

Frame lock then comes from the synchronisation pattern. The decommutator searches the stream for it, checks that it reappears exactly one minor frame later, and only then declares lock. Good patterns are chosen for their autocorrelation properties, so that a shifted or noisy version does not look like a match; the 24-bit 0xFAF320 is the one you will meet most often, and longer patterns are used on noisier links.

The search, check and lock strategy is tunable, and both extremes hurt. Too strict, and a single bit error drops lock and costs you a frame; too tolerant, and the decommutator locks onto a pattern that happens to appear inside the data. Sync losses are worth logging as an engineering parameter of the flight — they are the honest measure of link quality.

From bits to RF: codes and modulation

The bit stream is then modulated onto a carrier. IRIG 106 defines a small set of options, and in practice the choice is between robustness and spectral efficiency.

Frequency bands follow the range and the country, not the equipment. What a test transmitter offers is a tuning range: L band around 1 400 to 1 600 MHz, S band 2 200 to 2 400 MHz with an extended option, and C band around 5 090 to 5 250 MHz with a lower option. Which slice you are actually allowed to radiate is a frequency allocation question, settled with the range authority long before the flight.

The one relationship to keep in mind: occupied bandwidth is proportional to bit rate. Doubling the rate doubles the bandwidth, which doubles the noise power in the receiver and costs 3 dB of margin. There is no way around it, only a choice of where to spend it.

The link budget, line by line

A telemetry link budget is a single column of decibels. Everything the transmitter provides, minus everything the path takes away, compared with what the demodulator needs. Free-space loss is the dominant term and is worth memorising: 32.44 + 20 log(distance in km) + 20 log(frequency in MHz).

Thirteen decibels of margin at 200 km sounds generous, and it is exactly what you want, because the table is optimistic on the line that matters least in theory and most in practice: the airborne antenna gain. A blade antenna has deep nulls, the airframe masks it, and a bank away from the ground station can cost 10 to 20 dB in an instant. Multi-symbol detection buys 2 to 3 dB back on PCM/FM; a second antenna with best-source selection buys much more.

What actually breaks the link

• Antenna nulls and airframe masking — the leading cause, and the reason two antennas plus best-source selection or diversity combining is standard practice.
• Multipath at low elevation angles, over water or along a runway, where the reflected ray arrives almost as strong as the direct one.
• Tracking: an autotracking antenna that loses the aircraft, or acquires the wrong side of its own pattern.
• Spectrum: another user on the range, a spurious emission, or an allocation narrower than the occupied bandwidth you actually radiate.
• The mundane: a damaged coaxial run, water in a connector, a preamplifier that is not at the feed, a pointing calibration that was never redone.

Recording is the safety net

A telemetry link drops out. A recorder does not. The normal architecture keeps both: a reduced-rate stream over RF so the flight-test engineers can watch the test live and call the next point, and the full-rate data written onboard for the analysis that actually produces the answers.

That only works if the two views share a time reference. The recorder, the acquisition units and the ground station must all be locked to the same source — IRIG-B, GPS or PTP — otherwise merging a telemetry segment with the recorded file produces an offset that nobody can explain afterwards.