Case Study: Designing the Navigation for a Mars Orbiter

"You do not aim a spacecraft at Mars. You aim it at a point in empty space, months ahead of the planet, and then you spend the whole cruise finding out how wrong you were — while there is still time to fix it."

Executive Summary

In the first case study we determined the orbit of an object already in front of us. Here we do the harder, forward-looking job: we design the navigation for a mission that has not flown — a science orbiter cruising to Mars — deciding in advance what to measure, how to estimate, and how accurate we must be to survive Mars Orbit Insertion (MOI). We will set an accuracy requirement from the physics of the capture burn, assemble a Deep Space Network observation plan, architect a batch-then-filter estimation pipeline, and — the analytical heart — quantify how the plan buys accuracy: how stacking Doppler passes drives velocity error down as $1/\sqrt{N}$, and how a single delta-DOR measurement collapses the plane-of-sky error from tens of kilometres to a fraction of one. It is orbit determination run as engineering: not "where is it?" but "what must we build so that we will know where it is, well enough, when it matters?"

Skills applied

• Turning a mission constraint (a survivable MOI) into an orbit-determination accuracy requirement (§13.1).
• Selecting and combining observation types for a deep-space regime (Ch. 26, §13.2).
• Architecting a batch-least-squares-then-Kalman estimation pipeline (§13.5, §13.6).
• Sizing error reduction quantitatively: $N$-pass averaging and inverse-variance sensor fusion (§13.5, §13.6).
• Using residuals to make a pre-burn go/no-go decision (§13.5; Ch. 12).

Background: why deep-space navigation is a different animal

A Mars-bound spacecraft is far past radar and GNSS. It is tracked by the Deep Space Network, whose measurements are the radiometric trio of Chapter 26:

The defining hardship is geometry. Ranging and Doppler are superb along the Earth–spacecraft line, but by themselves they barely constrain the two directions across that line — and it is exactly a cross-line (plane-of-sky) error that turns into a missed aim point at Mars. Our design must buy down that plane-of-sky error deliberately. (All precisions here are Tier 2/3 — representative of DSN capability, not a specific mission's numbers.)

Phase 1: Set the accuracy requirement from the burn

Navigation accuracy is not an aspiration; it is derived from what the mission cannot survive. Mars Orbit Insertion fires the main engine once, at closest approach, to bend the hyperbolic arrival into a bound orbit. The aim point is described in the B-plane — the target plane perpendicular to the incoming trajectory — and the periapsis altitude at Mars is exquisitely sensitive to it. Come in a few kilometres too low and the spacecraft grazes too deep into the atmosphere or captures into the wrong orbit; a few kilometres too high and the geometry for the science orbit is wrong. A defensible requirement:

Deliver the spacecraft to the MOI B-plane with a cross-track (plane-of-sky) error no worse than a few kilometres, and a timing error small enough that the burn starts within a second of the plan.

Hold that number — "a few kilometres" — because Phase 4 will show that the choice of observations decides whether we meet it or miss by an order of magnitude.

Phase 2: The observation plan

From the menu, the plan writes itself once we know what each type constrains:

• Continuous Doppler through the cruise. It is cheap (it rides on the normal communications link) and it pins the velocity superbly — and velocity error, integrated over months of cruise, is position error at arrival, so nailing velocity early is the highest-leverage thing we can do.
• Periodic ranging to anchor the absolute line-of-sight distance.
• Delta-DOR near key events — after each trajectory-correction maneuver and on approach — to attack the plane-of-sky error that Doppler and ranging leave loose.
• Optical navigation in the final weeks, imaging Mars and its moons Phobos and Deimos against the star background, which measures position relative to the target itself and removes any lingering error in Mars's own ephemeris.

Phase 3: The estimation architecture — batch, then filter

We use both estimators of this chapter, each where it is strongest:

1. Cruise: batch least squares (§13.5). Weeks of Doppler, ranging, and delta-DOR are reduced together in a batch differential-correction fit, using a propagator that includes the perturbations of Chapter 12 — solar gravity, planetary perturbations, and, crucially for a light spacecraft, solar radiation pressure. Batch is right here because cruise is slow and the premium is on precision: reprocess the whole arc, study the residuals, iterate. Each batch solution also plans the next trajectory-correction maneuver.
2. Approach: an extended Kalman filter (§13.6). In the final days, optical-nav images arrive frequently and the situation changes fast; we hand the batch solution — the state estimate and its covariance — to an EKF that folds in each new image in real time. This same filter is the navigation block of the GN&C loop (Chapter 27) that will execute MOI.

Batch for the patient, precise cruise; filter for the fast, live approach. The handoff — batch covariance becoming the filter's initial uncertainty — is the seam where this chapter's two estimators meet.

Phase 4: Size the accuracy — two calculations that make or break the mission

(a) Stacking Doppler passes drives velocity error down. A single Doppler pass measures the line-of-sight velocity with some noise, say a per-pass component uncertainty of $\sigma_1 = 0.5\ \text{mm/s}$. Because each pass is an independent measurement of a slowly-changing quantity, the least-squares combination of $N$ passes reduces the error like the mean of $N$ samples: $$\sigma_N = \frac{\sigma_1}{\sqrt{N}}.$$ To reach $\sigma_N = 0.1\ \text{mm/s}$ we need $$N = \left(\frac{\sigma_1}{\sigma_N}\right)^2 = \left(\frac{0.5}{0.1}\right)^2 = 25\ \text{passes}.$$ Twenty-five tracking passes — a few weeks of routine Doppler — sharpen the velocity fivefold. This is the quiet workhorse of deep-space navigation: nothing dramatic, just the $1/\sqrt{N}$ law of least squares, paying off over a long cruise.

(b) Delta-DOR collapses the plane-of-sky error. Now the cross-line direction, where the mission is won or lost. Suppose Doppler-and-ranging alone leave the plane-of-sky angular error at roughly $\sigma_{\text{DR}} = 200\ \text{nrad}$ (nanoradians), while a delta-DOR measurement determines that same angle to $\sigma_{\text{DD}} = 2.5\ \text{nrad}$. At an Earth–spacecraft distance of $D \approx 2\times10^{8}\ \text{km}$ near arrival, convert angles to cross-track position errors (position $=$ angle $\times$ distance): $$\text{Doppler/ranging alone:}\quad 200\times10^{-9}\times 2\times10^{8}\ \text{km} = 40\ \text{km},$$ $$\text{with delta-DOR:}\quad 2.5\times10^{-9}\times 2\times10^{8}\ \text{km} = 0.5\ \text{km}.$$ Fusing the two by inverse variance (§13.6) gives a combined angle dominated by the sharper delta-DOR: $$\frac{1}{\sigma^2} = \frac{1}{200^2}+\frac{1}{2.5^2} = 0.000025 + 0.16 = 0.160\ \text{nrad}^{-2} \;\Rightarrow\; \sigma \approx 2.5\ \text{nrad},$$ i.e. $\approx 0.5\ \text{km}$ of cross-track error. This is the decisive result. Doppler and ranging alone deliver a $40\ \text{km}$ plane-of-sky error — an order of magnitude worse than our "few kilometres" requirement, a miss that could doom the capture. Add delta-DOR and the same requirement is met with margin, at $0.5\ \text{km}$. The mission's survival was decided not by the spacecraft but by putting the right observation in the plan.

Phase 5: Residual monitoring and the go/no-go

In the final approach, the residuals (§13.5) become the mission's conscience. If the optical-nav and delta-DOR residuals settle into structureless noise around zero, the solution is trustworthy and MOI is go. If a slow trend appears in the residuals, the model is missing a force — most likely mis-modeled solar radiation pressure or a small unbalanced thruster leak quietly pushing the spacecraft off the fitted arc — and the navigation team must re-fit, possibly insert a late trajectory-correction maneuver, or, in the worst case, wave off. A structured residual before an irreversible burn is not a nuisance; it is the system telling you, while you can still act, that your knowledge is wrong. Reading it correctly is the last and most important act of orbit determination.

Discussion Questions

1. Phase 1 derived the accuracy requirement from the MOI burn. Why is a cross-track (plane-of-sky) error more dangerous at arrival than an equal along-track (line-of-sight) error?
2. In Phase 4(a), reducing velocity error from $0.5$ to $0.1\ \text{mm/s}$ took $25$ passes. To halve the error again to $0.05\ \text{mm/s}$, how many total passes would you need, and what does the $1/\sqrt{N}$ law say about the cost of the last increment of accuracy?
3. Delta-DOR shrank the cross-track error from $40\ \text{km}$ to $0.5\ \text{km}$. Why can Doppler and ranging, however precise, never do this on their own? (Hint: what direction do they measure?)
4. Why hand off from a batch estimator to a Kalman filter for the approach rather than simply running batch the whole way?

Your Turn: Extensions

• Option A (analysis). Recompute the Phase 4(b) cross-track errors for a closer arrival distance of $D = 1\times10^{8}\ \text{km}$. Does delta-DOR still meet a "few km" requirement? Does Doppler-only?
• Option B (computation). Write passes_needed(sigma_single, sigma_target) returning $(\sigma_{\text{single}}/\sigma_{\text{target}})^2$, and fuse(sig1, sig2) returning the inverse-variance combined standard deviation. Reproduce the $25$ passes and the $\approx 2.5\ \text{nrad}$ fusion. (Do not run it — hand-trace and add # Expected output:.)
• Option C (your mission). For your mission (Track A/B/C/D), write the navigation architecture: the observation types, the batch/filter split, the single accuracy number that would make or break a critical event, and the observation that buys it. Record it in your MDR's tracking-and-navigation note.

Key Takeaways

1. Navigation accuracy is derived, not assumed. The MOI burn sets a "few kilometres" plane-of-sky requirement, and every observation choice is judged against it.
2. Match the observation to the direction you must constrain. Doppler and ranging pin the line of sight; only delta-DOR (or optical nav) pins the plane of sky — and it is the plane of sky that decides the aim point.
3. Two estimators, each in its place. Batch least squares for the precise, patient cruise; a Kalman/EKF for the fast, live approach — with the batch covariance seeding the filter.
4. Accuracy is bought by two laws. Averaging $N$ independent measurements buys $1/\sqrt{N}$; fusing a sharp sensor with a dull one buys the sharp sensor's precision. Both are just least squares and inverse-variance fusion at work.
5. Residuals make the final call. Structureless residuals mean go; a trend before an irreversible burn means the model — and your knowledge — is wrong while there is still time to fix it.