Time-to-alert: why a few seconds change the approved operation

Chronomètre posé sur une carte aéronautique

The same satellite navigation chain is credited without reservation on an oceanic cruise segment and rejected for an approach with vertical guidance, without a single onboard component having changed. What changed comes down to a single number: the admissible time between the instant the position error becomes unacceptable and the instant the crew is told about it. It is counted in minutes on one end of the scale and in seconds on the other, and this time is nothing other than the factor by which a rate of divergence is multiplied to know how many metres the aircraft will travel outside tolerance without anyone knowing.

Time-to-alert is not a reaction time

The definition is narrow, and almost every misunderstanding in the cockpit starts there: time-to-alert is the maximum interval allowed between the appearance of a condition that makes the position error exceed the alert limit of the operation in progress, and the presentation of the annunciation at a defined output.

Upstream, the clock does not start at the failure. A satellite whose clock correction is drifting is already at fault well before its defect produces an error that crosses the retained threshold. This period is counted nowhere, and rightly so since it is not hazardous; it explains, however, why a post-flight analysis always finds a defect older than the alert.

Downstream, everything depends on which output is chosen, and there are three. The service provider commits on the signal-in-space. The equipment specifications stop at the flag produced by the receiver. Only the installation package speaks to the crew interface. These three numbers differ and do not add up at random: a signal-in-space time-to-alert says nothing about the time an annunciation takes to cross a data concentrator, a message hierarchy and a page that may not be the one selected. As for the crew, it is in none of the three.

Time-to-alert is the maximum allowed delay between when a navigation error becomes dangerous and when the crew is warned. It’s not about how fast the crew reacts, but how fast the system detects and reports the problem. Different parts of the system (satellite, receiver, display) have different time-to-alert values, and the crew’s alert only comes at the end of this chain.

The arithmetic that fixes the number of seconds

The number does not come out of a consensus, it comes out of a division. Two boundaries bracket the trajectory: the one where the monitor commits to speaking, the alert limit, and the one beyond which the segment’s obstacle clearance margin is consumed. Between the two lies a reserve, expressible in metres. During the time-to-alert, the aircraft keeps diverging at a certain rate. The admissible time-to-alert is the quotient of that reserve by that rate.

Both terms move the wrong way as one descends toward the ground. The reserve shrinks because the alert limit follows the protection geometry: a few nautical miles oceanic, about one mile in the terminal area, a few hundred metres for an approach without vertical guidance, a few tens of metres for the vertical term where it exists. The divergence rate, for its part, does not shrink and can even grow.

Hence the usual scale of times: several minutes in cruise, some fifteen seconds in the terminal area, about ten seconds for a conventional approach or a departure, on the order of six seconds as soon as vertical guidance is involved, shorter still for the most demanding modes backed by ground augmentation. The binding values are those of the applicable specification; what matters here is the ratio between the extremes, namely two orders of magnitude for the same equipment.

On an oceanic cruise, the alert limit might be 5 nautical miles, and the divergence rate 0.1 nautical miles per second. The time-to-alert is 5 / 0.1 = 50 seconds. But for an approach with vertical guidance, the alert limit drops to 20 metres, and the divergence rate increases to 5 metres per second. The time-to-alert becomes 20 / 5 = 4 seconds. The same system can’t handle both scenarios with the same time-to-alert.

The divergence rate, the forgotten term of the reasoning

Three contributions build up this rate, and they are not read at the same place.

Two pilots in a cockpit at dusk, closely monitoring flight instruments.
Pilots monitor instruments, where seconds can determine flight safety and operational success.

The first is the fault ramp in the range domain, amplified by geometry. A range error only becomes a position error after multiplication by the slope associated with the faulty satellite, that is, the number of metres of horizontal error produced per metre of range error. A modest clock ramp, on the order of a metre per second, therefore shifts the position by several metres per second if that satellite occupies a poorly redundant direction. Poor geometry does not just degrade accuracy: it accelerates the divergence, thereby shortening the time actually available under an otherwise unchanged time-to-alert.

The second is specific to propagation. An ionospheric gradient crossing the area behaves like a moving front: the gap between the applied correction and the actual error grows as the front passes, at a rate set by the geometry of the phenomenon and not by the electronics.

The third is the most counter-intuitive, and it is the guidance loop. A position error is not a trajectory error, it is a displayed error. The autopilot conscientiously removes it, in other words it truly moves the aircraft by that same number of metres in the opposite direction, converting a false indication into a real excursion. The more precise the servo-loop, the faster the real deviation grows. At an approach speed on the order of seventy metres per second, an interception correction of a few degrees shifts the trajectory by a few metres per second; vertically, one degree of correction on a three-degree slope changes the rate of descent by about one metre per second. Ten seconds are then worth a few tens of metres of effective displacement, exactly the order of magnitude of a vertical approach alert limit. The calculation closes on itself.

One property of the monitor remains: it sees a large error all the better. The dimensioning fault is therefore neither the abrupt jump, detected almost instantly, nor the infinitesimal drift, which is harmless, but the ramp calibrated just below the test’s sensitivity, the one that maximises the sum of the error already present at detection and the error accumulated during the time-to-alert. The reasoning « such an error would be seen » is therefore worthless: the dangerous class is exactly the one that is not seen.

The scale of requirements by operation: orders of magnitude and what they protect
Operation Alert limit Admissible time-to-alert What consumes the reserve Cost of an unannounced excursion
Oceanic or continental cruise A few nautical miles laterally Several minutes Spacing between organised tracks Convergence still recoverable by air traffic services
Terminal area, arrival and transition About one mile Some fifteen seconds Turns and transients, terrain closer Departure from the area of an already tightened segment
Approach without vertical guidance, departure A few hundred metres About ten seconds Narrowing of the area toward the threshold Crossing the area boundary where the margin is minimal
Approach with satellite-based vertical guidance A few tens of metres vertically On the order of six seconds Height remaining above obstacles Trajectory below the protected surface, with no visual reference
Approach backed by ground augmentation Same order vertically, tightened Shorter still Local ionospheric front Same, plus departure from the local threat model

Time-to-alert is not an electronic constant, it is a budget spent along a chain, and no link shortens for free.

The first is the observability of the divergence. A test built on raw measurements is noisy; it is smoothed, in practice by carrier-smoothing the code measurement with a time constant on the order of a hundred seconds, which lowers the noise and allows a tighter threshold. But a filter that averages over a hundred seconds only restores a ramp with a delay of the same order: what is gained in sensitivity is paid for in latency, and the tension is structural, not an implementation flaw.

Next come threshold crossing, confirmation logic — no serious monitor declares on a single isolated detection, and every persistence cycle is net time-to-alert spent —, the transport of the information through augmented architectures, and finally its restitution to the crew, the only link the operator documents and sometimes chooses.

Breaking down the time-to-alert: what fixes each link and what shortening it costs
Link What fixes its duration What shortens it Price paid elsewhere
Observability of the divergence Smoothing time constant, measurement noise Shorter filter Noisier measurement, higher threshold, larger protection level
Threshold crossing Margin between threshold and noise, accepted false alarm Lower threshold Unjustified interruptions, loss of continuity
Confirmation of the detection Number of cycles required, computation rate Fewer persistence cycles Nuisance alerts during turns or masking
Transport through an augmentation Broadcast rate, loop length Local loop, line-of-sight broadcast Coverage limited to a service volume, per-site infrastructure
Restitution to the crew Display logic, annunciation hierarchy Dedicated, priority, audible annunciation Crowding of the alerting space, desensitisation

A short time-to-alert is paid for in availability, not in safety

This breakdown unravels an everyday flight-deck observation: under the same sky, at the same second, a receiver declares the function usable for lateral guidance alone and unusable as soon as vertical guidance is armed. The constellation has not moved; the pair alert limit and admissible time-to-alert, on the other hand, has changed. A shorter time-to-alert forbids long smoothing, the measurement becomes noisier, the threshold rises to hold the same false-alarm rate, the protection level grows in proportion, and it must now fit under a limit that has tightened at the same time. The two movements add up: availability does not degrade progressively from one operation to the next, it drops in steps.

Another, less intuitive, consequence: tightening the time-to-alert does not buy extra safety compared with a less demanding operation. The targeted safety objective stays of the same order from one end of the scale to the other. What a short time-to-alert buys is the right to operate with a smaller geometric reserve.

The trap. The time-to-alert stops at the annunciation, while the demonstration of obstacle margin assumes the manoeuvre already engaged. Between the two opens an interval that nobody has budgeted and that the crew fills. Going to check on the other screen, seeking visual confirmation, allowing a few seconds to see whether it stabilises: on approach, each of these gestures consumes the equivalent of the entire regulatory time-to-alert, drawn not from a provision but from the obstacle margin itself. The trap also closes on the installation side, where a signal-in-space time-to-alert of a few seconds guarantees nothing if the annunciation appears on a page that is not selected.

Augmentation systems: the time-to-alert becomes a shared budget

As soon as an augmentation system enters the picture, the time-to-alert is split between a ground segment that monitors, a broadcast means that transports, and a receiver that decodes and then annunciates. Loop length therefore becomes a design parameter, and it explains the hierarchy of performance.

A satellite-based augmentation system observes the constellation from a wide network of stations, centralises the processing, uplinks the message to a geostationary satellite and rebroadcasts it at a fixed rate over the whole covered area. Long loop and imposed rate set a floor that no receiver will go below; the ground segment therefore only has a fraction of the total time-to-alert.

A ground-based augmentation system works the other way round. Its reference receivers see the same satellites through practically the same ionosphere as the approaching aircraft, and the broadcast is local, line-of-sight, at a high rate. Short loop, hence shorter time-to-alert and a tighter vertical limit. The counterpart is an assumption of locality: the protection only holds within the service volume, and the local threat model, an ionospheric front foremost among them, must be covered by ground monitors whose effectiveness assumes the aircraft crosses an ionosphere comparable to the one the station observes.

What the time-to-alert does not bound

The time-to-alert is a conditional commitment: it holds for the failure modes of the model, and only from the moment a monitored quantity crosses a monitored threshold. Four families escape it, and none of them announces itself with a clock that starts.

  • Coherent spoofing. The measurements remain mutually consistent, no residual is abnormal, nothing is crossed. The clock never starts, which is the nominal behaviour of a monitor fed false but consistent data.
  • The path-definition error. A mis-coded waypoint, a wrong segment altitude, a database with a valid seal but erroneous content: the position is correct, it is the reference that is wrong. Nothing to detect a divergence from, hence no applicable time-to-alert.
  • The flying error. The monitor bounds the error of the navigation system, not the way the aircraft follows the displayed trajectory. A flying deviation triggers no clock, however long it lasts.
  • The inhibited or non-delivered annunciation. An alert produced on time but suppressed, or displayed where nobody is looking, leaves the time-to-alert formally respected and the design assumption destroyed.

For the first three, the recourse is external to the monitor: cross-checking by a means of a different nature, plausibility of the route and the remaining distance, comparison of altitude sources, checking segments against the chart. For the fourth, the recourse is documentary and is prepared on the ground.

The right reflex. Treat the annunciation as an order for immediate execution, not as information to verify: the verification time has already been spent by the monitor before the message appears, and that is precisely what the time-to-alert was for. At the briefing, name three things instead of one: the exact form the alert takes on your installation, the point in the procedure where the equipment switches to the tightened regime, and the action taken without discussion. An annunciation whose form is known in advance is recognised in a fraction of the time it takes to read an unexpected message, and that is the only link in the budget on which a crew can still gain.

What must be found in one’s own documentation

The reasoning above is generic; its application depends on installation and approval choices that are, themselves, written down. Five points deserve to be found in black and white.

  1. The output on which the credited time-to-alert is demonstrated: signal-in-space, receiver output or crew interface. It is this that says whether the display chain is included in the number or added on after it.
  2. The form and hierarchy of the annunciation, which determine the recognition time, the largest unbudgeted variable in the chain.
  3. The inhibition possibilities and the displayed state after inhibition: an inhibited alert removes a load-bearing piece of the obstacle margin without changing anything on the screen.
  4. The point in the procedure where the equipment moves to the tightened regime, because that is where the alert limit and the time-to-alert change, and hence where availability can drop.
  5. The behaviour on loss of vertical guidance alone, which decides whether a legitimate lateral operation remains and to what minima.

The grey area, and what the topic is worth in a checkride

A weakness remains, and it is one of construction. The admissible time-to-alert is computed at the most constraining point of the segment, then applied as a constant. Above that point it is comfortable; right at the bottom, the height remaining above obstacles becomes so small that the geometric reserve tends toward nothing, and what then holds safety together is no longer the number written in the specification but the crew’s reaction convention. Yet no operational document publishes the share of the time-to-alert reserved for that reaction: reading the specification, one will never know whether the second one takes is drawn from a provision or from the obstacle margin. This is exactly the kind of practice gap from one operator to another revealed by professional exchanges.

For anyone preparing a type rating or a proficiency check in performance-based navigation, the topic is never decided on the number of seconds, but on three answers: what the time-to-alert bounds, what it leaves out, and why it divides by a considerable factor between cruise and the final segment. Knowing that an approach with vertical guidance requires a few seconds is only useful if one can explain that a quotient of reserve by divergence rate leaves no choice, and that the guidance loop works against the crew by faithfully converting a false indication into a real excursion. This is the level of understanding targeted by the FL330 preparations devoted to satellite navigation.

This article describes mechanisms and does not set any operational value. In the event of a discrepancy, the applicable regulatory texts, the specifications of the installed equipment, the charts and notices in force, and the operator’s approved documentation prevail over what is written here, including where they hold time-to-alert values or alert limits different from those discussed.

Questions fréquentes

What should be done about an integrity alert that occurs in the final moments of an approach with vertical guidance?

The annunciation means the guidance is no longer guaranteed, not that the trajectory is already lost: the verification time has precisely already been spent by the monitor before the display. Since no provision exists for a second in-cockpit verification, the consistent action is the immediate abandonment of vertical guidance; continuing is only open to discussion if the required visual references have been acquired, and it then falls solely under that operation’s visual criteria. The case is settled at the briefing rather than in flight, because the alert limit and the admissible time-to-alert tighten precisely in that segment. What is actually authorised is fixed by the published minima and the operator’s approved documentation.

Does the time-to-alert include the crew's reaction time?

No, it stops at the presentation of the annunciation at a defined output, with no provision whatsoever for recognition, decision, action and the aircraft’s response. Yet the demonstration of obstacle clearance margin assumes the manoeuvre has actually been engaged. Between the annunciation and the action there therefore opens an unbudgeted interval which, on approach, is drawn directly from the obstacle margin. This is why an integrity annunciation is executed rather than verified.

Is a time-to-alert demonstrated at signal-in-space level enough to qualify an installation?

No, because three different outputs coexist depending on the document: the signal-in-space for the service provider, the receiver output for the equipment specifications, the crew interface for the installation package. The time taken to cross a data concentrator, a message hierarchy and a page that may not be selected adds on to the published number. Only the installation documentation states at which output the credited time-to-alert was demonstrated.

Why can a receiver be available for lateral guidance and unavailable for vertical guidance at the same instant?

Because the tightening of the time-to-alert and the narrowing of the alert limit add up. A shorter time-to-alert forbids long smoothing of the measurements, which increases the noise, forces the detection threshold up to hold the same false-alarm rate, and thereby grows the protection level. This larger level must then fit under a smaller alert limit. Availability does not degrade progressively from one operation to the next: it drops in steps.

What type of fault is most constraining for the monitor's design?

Neither the abrupt jump, detected almost instantly because it produces a clear residual, nor the infinitesimal drift, which remains harmless. The dimensioning case is the ramp calibrated just below the test’s sensitivity, the one that maximises the sum of the error already present at the moment of detection and the error accumulated during the time-to-alert. This invalidates the reasoning that a dangerous error would be seen: the class that dimensions the system is precisely the one that is not seen.

Why can a ground-based augmentation system promise a shorter time-to-alert than a satellite-based one?

Because its monitoring loop is short: the reference receivers see the same satellites through practically the same ionosphere as the aircraft, and the broadcast is local, line-of-sight, at a high rate. A satellite-based augmentation system, by contrast, must centralise the processing of a wide network, uplink to a geostationary satellite and then rebroadcast at a fixed rate, which sets a floor independent of the receiver. The counterpart of the short time-to-alert is an assumption of locality, valid only within the service volume.

Commentaires

Leave a Reply

Découvrez l’écosystème FL330

FL330, c’est toute l’aviation professionnelle — explorez nos autres espaces.