Satellite internet · Session 8 of 8
Defending the result
Can someone who was not there reproduce your test and reach your conclusion?
An explanation only its author can follow is not finished. The test is simple and slightly uncomfortable: hand your model to someone who was not there, let them pick the challenge, and see whether your account still predicts what happens. Working engineers do this deliberately and call it a design review. The useful part is never the praise. It is the one question you cannot answer yet.
By the end
- Demonstrate an end-to-end route, outward and return, naming the physics at each stage.
- Report a result with its uncertainty and the conditions it holds under, rather than as a bare number.
- Distinguish a claim your evidence supports from one it does not, and say which further measurement would close the gap.
The idea
A model earns its keep by predicting something. It does not have to copy reality — every model in this course deliberately omits most of it — but its rules and its limits must be stated clearly enough that someone else can apply them and get your answer. That property is reproducibility, and it is the difference between a result and an anecdote.
Measurements carry uncertainty, and a number without one is incomplete. Your flashlight trials were timed by hand, so each has perhaps ±0.2 s of reaction time; your packet runs are discrete, so a count of 8 is exact but a rate of 0.4 packet/s depends entirely on the window you chose. Repeat a measurement three times and the spread tells you something the single value cannot: whether a difference you observed is real or noise.
Keep calculated, measured and assumed values apart, in your table and in your speech. You calculated v = 7.59 km/s and γ = 8.46°; you measured correct bits and arrival ticks; you assumed 30% panel efficiency, a 25° minimum elevation and a 2 kW load. All three are legitimate, and conclusions that confuse them are not.
The strongest claim is the narrowest one that is fully supported. “Our design never disconnects” is unsupportable from twelve samples. “Under our two-second forwarding rule, the alternate route delivered 7 packets by 20 s after the direct link failed, against 0 without it” names the rule, the condition, the window and the comparison — and is therefore worth something. Narrower claims sound more modest and are harder to dismiss.
A failed test is evidence, not an embarrassment, provided it is reported. The design that missed its target under failure and was then changed and re-tested demonstrates more understanding than one that passed first time, because it shows the measurement was capable of saying no. Hiding a failure destroys the value of every other number in the same table.
Your final account has to integrate six things that were taught separately: information measured in bits, a signal carrying it across a physical channel, an orbit maintained by gravity alone, a geometry that decides who is in view and for how long, an energy budget that survives the dark, and a network whose delay has three distinct sources. Systems fail at the joins between subsystems far more often than inside them.
Gather
Your accumulated model, the mission notebook, a coded message, test cards, a calculator and an audience that will ask hard questions.
Build and investigate
- Prepare a four-minute account: the community and its need, your design, and the physics behind each stage. Trace an outgoing request and its returning reply, naming what changes at each node.
- Have the audience choose a short message. Encode it, route the numbered packets through your model, and reconstruct it at the destination. Report the delivery count and the first-arrival time separately.
- Draw a challenge card and apply it under the same rules you used before: remove a relay, block the direct ground link, add five competing packets, disable a satellite, or move the Sun behind Earth for 35 minutes.
- Present baseline and revised results side by side with units. Repeat at least one test live so the audience can check reproducibility, and state the uncertainty on any hand-timed quantity.
- State one claim your evidence supports and one it does not, in the same breath. Then answer the exit questions independently before comparing with the guidance.
- Record the single question you could not answer, and design — but do not run — the measurement that would answer it. Predict its result first.
Run the numbers
No new physics here. The task is to use all of it at once, correctly, and to say precisely how far each number reaches.
| Relation | Expression | What it means |
|---|---|---|
| Information | N = w × h × d bits; rate R = data ÷ total | Lesson 1 |
| Channel | c = fλ; C = B log₂(1 + S/N) | Lesson 2 |
| Orbit | v = √(GM / r); T = 2πr / v | Lesson 3 |
| Visibility | cos(ε + γ) = (R / r) cos ε | Lesson 4 |
| Energy | P = S·A·η·cos θ; E_batt = P·t ÷ DoD | Lesson 5 |
| Delay | t = t_prop + t_serial + t_queue | Lesson 6 |
Core · Grade 7–8
- Produce one number from each lesson for your own design, each with its unit and its status — calculated, measured or assumed. Six numbers, one table.
- For one hand-timed measurement, give the value with an uncertainty and say how you estimated it. For one counted measurement, explain why it has no uncertainty but its derived rate does.
- Write your strongest supportable claim in a single sentence that names the rule, the condition, the window and the comparison. Then write the overclaim you avoided, and say what it would take to earn it.
Deeper · Grade 9–12
- Build a complete one-way latency budget for your design: propagation up and down from your chosen altitude, plus your model's forwarding delay, plus an assumed ground-network component. State which term dominates and what would have to change for a different term to dominate.
- Identify the single assumption your design is most sensitive to. Vary it by ±20% and recompute the affected results. If the conclusion survives, say so; if it flips, you have found the measurement your project most needs.
- Your coverage model samples 12 instants and the geometry is continuous. Using 2γ from Lesson 4 and the angular spacing of your satellites, determine whether a gap exists between samples, and state the sampling interval that would have been needed to catch it.
Measure and record
Report target, baseline and revised result with units or clearly defined counts, the condition each was measured under, and an uncertainty for anything timed by hand. The audience must be able to name the changed variable and repeat your test from your notes alone.
Worked example
A strong claim: “Under our two-second forwarding rule, with the direct B → C link blocked, the alternate route through E delivered 7 of 20 packets by 20 s, against 0 with no alternate route; first arrival moved from never to 8 s. Timing is exact because ticks are counted, not timed.” Every term is pinned: the rule, the failure condition, the observation window, the comparison, and why there is no uncertainty. Compare the weak version — “our design is more reliable” — which names no rule, no condition and no number, and which no one could check or disprove. Now the claim this evidence does not support: anything about live internet speed. The model's ticks stand in for serialisation and processing and correspond to no physical duration, so the correct response to “how fast is it really?” is to produce the propagation arithmetic from Lesson 6 instead, and say plainly that the card model cannot answer it.
Pause and explain
Explain your whole system: how a message becomes a signal, why the satellite stays in orbit, how a handoff preserves a connection, where the energy comes from and what happens when a route fails. Include one calculated value, one measured value and one limitation, and label which is which.
Check your reasoning after your attempt
Look for: an agreed code carried by a controlled change in a physical quantity, with the protocol supplying the meaning; forward motion with gravity supplying centripetal acceleration, so that v = √(GM/r) — not an absence of gravity; transfer to another satellite that satisfies the minimum elevation condition before the first one leaves; solar collection governed by the cosine law with a battery sized by eclipse duration and depth of discharge; and rerouting, or an outage when no route remains. Accept different designs and wording. What is not negotiable: the three value types must be labelled correctly, the claim must match the recorded evidence, and the stated limitation must be specific to this model rather than a general disclaimer.
Save in your mission notebook
The final system diagram with the return path; the six-number table with each value labelled calculated, measured or assumed; baseline and revised results with conditions and uncertainties; your bounded claim and the overclaim you rejected; the question you could not answer and the measurement you designed for it.
Parent / teacher notes
Use the five-criterion rubric in the teacher guide, and weight “limits and consequences” as heavily as “science” — a learner who states a bounded claim has understood something a learner with a tidier model may not have. Ask one new “what if” that was not rehearsed; a prepared speech survives anything else. Accept a narrated diagram, a tactile model or a recorded explanation. Grade 7–8 should produce the six-number table and one bounded claim. Grade 9–12 should do the latency budget and the sensitivity analysis, which is the single most transferable skill in the course.
Suggested timing: 10 minutes predict, 15 read and explain, 30 build and investigate, 25 measure and calculate, 10 independent explanation. Adapt the activities rather than rush unfinished reasoning.
Take it further
Design an experiment that could show your current explanation is wrong — not incomplete, wrong. Predict its outcome under your explanation, predict it under the leading alternative, and identify the observation that would distinguish them. A test that cannot come out either way is not a test.