ECSS Communications — Time Distribution (space-systems/ecss/e50-time-distribution)
Use when a time reference has to be distributed to the on-board or proximity users that need it, per ECSS-E-ST-50C clause 5.7.2.8 — whether what arrives at each user is still inside the accuracy that user was promised.
Domain quick reference
- The obligation is met at the consumer, not at the source. A master clock holding nanoseconds proves nothing about a user three hops down a distribution tree; the number that matters is what is left after the chain and the holdover have spent their share.
- Hop contributions split into two kinds and do not combine the same way. Uncompensated transport delay, path asymmetry and timestamp quantization are biases that repeat every message, so they add; jitter is independent per message, so hops combine root-sum-square. Adding jitter linearly overstates a long chain badly enough to buy hardware nobody needed.
- Asymmetry and quantization each enter at half. A two-way exchange cancels the common part of the path and leaves half the difference; a timestamp resolved to one step is wrong by at most half a step.
- Compensation changes which number a hop contributes, not whether it contributes. A hop declared compensated still carries the residual the compensation left behind, and a residual nobody measured is an unknown, not a zero.
- Holdover is usually the dominant term and it is the one under design control. Between two distribution messages the consumer runs free at its own drift rate, so the distribution period and the oscillator together decide most of the budget — which makes the period the first knob to reach for, and the inverse calculation worth reporting.
- A chain that spends the whole allowance on its own cannot be rescued by distributing more often, and saying so is more useful than a period that rounds to something unbuildable.
Workflow
- Describe the chain hop by hop: transport delay, whether it is compensated and with what residual, path asymmetry, timestamp step and jitter. Reject an unnamed hop or an unknown field rather than silently defaulting it.
- Reduce each hop to a systematic and a random contribution, taking the residual for a compensated hop and the full delay for one that is not.
- Sum the systematic contributions; combine the random ones root-sum-square; keep the per-hop breakdown so the dominant hop is visible rather than inferred.
- Add the holdover the consumer accumulates over one distribution period at its drift rate. A consumer with no declared drift holds over without error, which is a statement about the data, not a guarantee.
- Compare the total against that consumer's required accuracy with a relative tolerance, so a budget landing exactly on the allowance passes on every build host.
- Invert the same model for the remedy: the longest distribution period that still fits. Report zero where the chain alone exhausts the allowance, and report that the period is unconstrained where there is no drift to accumulate.
- Grade every consumer, name the worst, and check the stated period remedy back through the model before offering it. Confirm in the same pass that the graded set holds every node the network has to serve with time, and that all of them trace back to the one reference: a node nobody distributes to, or a second source of time alongside the first, leaves the network without a single reference however comfortably each budget closes.
Obligations
| Item | Step |
|---|---|
| ECSS-E-ST-50C Rev.2 5.7.2.8a | 7 |
Pitfalls
- Budgeting at the master clock. Reference stability is an input to the budget, not the answer to it, and a distribution tree can lose orders of magnitude between the source and the user.
- Adding jitter terms linearly across hops. Independent random terms combine in quadrature; summing them makes a compliant chain look non-compliant and sends the design after a problem it does not have.
- Taking a compensated hop as contributing nothing. Compensation leaves a residual, and an unmeasured residual is a hole in the budget rather than a zero in it.
- Charging the full asymmetry or the full timestamp step. Both enter at half, and doubling them is the quiet way a correct design fails its own review.
- Deciding compliance with a bare inequality on the computed total. Two arithmetically identical budgets can straddle the allowance on different machines, and the verdict then depends on the build host.
- Answering an exceeded budget with a shorter period without checking whether the chain already spends the allowance. Distributing more often cannot buy back a bias that is present in every message.
Behavior contract (gate 3)
Hop and consumer validation, the systematic and root-sum-square split, the compensated-hop residual, the holdover term, the exact-allowance boundary, the longest-period inverse and its zero and unconstrained cases, and the worst-consumer selection are exercised by the gate 3 contract test: scripts/test_e50_time_distribution.py against scripts/e50_time_distribution_logic.py (stdlib unittest, offline). Run: python3 scripts/test_e50_time_distribution.py
Compliance
- ECSS standards are freely downloadable (ESA); cite the source and paraphrase per standards-map.yaml.
- compliance: STANDARDS-REF, gated: false.
