By Lunibyte team · November 1, 2026 · 6 min read

What Can Quantum Algorithms Do for Satellite Networks?

Satellite networks create planning problems before a single user packet is sent. Which available route should carry a demand? When should a planner change routes? What happens if two candidate routes share the same failed link? These questions are about decisions under a model, and some can be expressed as combinatorial optimization problems.

Quantum algorithms are one research direction for such computations. The meaningful starting point is a precise problem and a credible comparison, rather than a promise of faster internet. Our LuniQ SatMesh preview uses a deliberately small synthetic instance to make every decision and constraint inspectable.

Begin with routes, snapshots and choices

The toy network contains ground endpoints A and B, plus four satellite-labelled relay nodes S1 through S4. There are three candidate complete routes. P0 follows A–S1–S2–B, P1 follows A–S1–S3–B, and P2 follows A–S4–B.

The planner chooses one route at snapshot t0 and one at snapshot t1. Those labels denote abstract planning snapshots, not orbital timestamps or contact durations. We assign one binary variable to each snapshot-and-route pair, giving six variables in total.

A 1 means that route is selected for that snapshot. Six variables produce 64 possible bitstrings, but most are not valid schedules. Exactly one route must be chosen per snapshot, and every hop in that route must be available there. Six qubits encode these decisions in the simulator; the count is unrelated to the number of physical satellites.

Availability changes the valid answers

In the normal scenario, P0 and P2 are available at both snapshots. P1 is available at t0, but the S1–S3 hop is unavailable at t1. Of the nine schedules satisfying the one-route-per-snapshot rule, six also satisfy availability.

The degraded scenario additionally removes S1–S2 at t1. P0 becomes unavailable there, leaving P2 as the only valid t1 route. A separate shared-link scenario removes A–S1 at t1, invalidating both P0 and P1 together.

That shared link is an important detail. Different route names do not guarantee independent backups. A planning model needs to check actual hops and common dependencies. Our two outage scenarios happen to produce the same feasible schedules and costs, even though the failed links differ.

Give the objective an honest meaning

The toy objective adds three components: assumed one-way snapshot latencies, a risk preference weighted by 10, and a penalty of 4 for changing route IDs between snapshots. The combined value is called planning cost.

For normal P0 followed by P0, the assumed latencies are 12 and 14. The two risk inputs are 0.02 and 0.04. There is no route change. The calculation is 12 + 14 + 10 × (0.02 + 0.04), giving 26.6.

After the degraded outage, the optimum changes to P0 followed by P2. The calculation becomes 12 + 21 + 10 × (0.02 + 0.01) + 4, giving 37.3. The 33-millisecond latency sum and the total planning cost are different fields. The risk inputs are illustrative preferences, not estimated outage probabilities; neither total is a measured ping.

Turn the rules into a mathematical cost

A QUBO is a quadratic unconstrained binary optimization formulation. It expresses a cost using binary variables, individual terms and pairwise products. Constraints can be represented by added penalty terms, although the original rules still need explicit verification afterward.

In SatMesh, choosing other than one route for a snapshot adds 100 times the squared deviation from one. Selecting an unavailable route adds another penalty. The goal is to make invalid choices unattractive in this fixed instance.

The number 100 is not a universal constant for satellite planning. It is justified against this toy objective and its feasible alternatives. Changing the allowed costs, topology or constraints would require revisiting that reasoning. A large-looking penalty alone is not a proof that the model works.

Where QAOA enters the experiment

The QUBO can be converted into an Ising-form cost operator, which supplies the circuit's problem-dependent phase operations. The circuit alternates those operations with mixer rotations. A classical optimizer searches a bounded set of circuit-angle choices, then the resulting distribution is sampled.

This is a hybrid computation. The original QAOA paper supplies the algorithmic foundation, not a guarantee that our satellite-labelled example outperforms classical planning. Our lab implements its circuit using a noiseless classical simulator.

A penalty encourages valid outcomes but does not ensure every measured string satisfies the constraints. The lab therefore checks each observed schedule against the original graph and selection rules. If no feasible quantum sample is observed, it shows no quantum recommendation; it does not insert the classical answer into that panel.

Why the classical baseline is essential

For six variables, an ordinary exhaustive solver can enumerate all 64 strings. It checks feasibility directly and calculates the exact best valid planning cost. That gives the small experiment a reference answer instead of relying on the quantum result's own energy.

Matching 26.6 in the normal case or 37.3 after an outage means the best observed feasible sample matched that reference for this instance. It does not establish faster computation, a better physical network or a general performance guarantee.

Look at the whole distribution as well. A run can find an optimal sample while most shots are invalid. More circuit layers do not guarantee improvement under our bounded optimizer: in recorded degraded/shared runs, p2 had lower feasible probability than p1. That is a result to explain, not a detail to discard.

What a physical extension would need

Satellite-labelled nodes are not an orbital model. A geometry-based extension would need positions at a common epoch, a consistent coordinate frame, Earth-obstruction checks, ground elevation rules and contact intervals. Operational planning could additionally need terminal limits, capacity, weather and traffic information.

For a physical link, propagation delay depends on distance and the medium. NOAA explains relevant radio-path effects in its satellite communications overview. A planning algorithm cannot make these effects disappear by optimizing a different cost function.

Two abstract snapshots also cannot establish uninterrupted service between them. A route might be available at both endpoints of a time interval while becoming unavailable in the middle. That is a modelling question to solve before interpreting a schedule as an operational plan.

Try the experiment as a comparison

Open LuniQ SatMesh, begin with the normal scenario and inspect both snapshots. Run p1 with the default seed and shots. Compare the observed best with the classical panel, then inspect invalid outcomes and the objective breakdown.

Change to an outage and notice that the previous results clear. Run again and explain why the selected t1 route changes. Finally, try the disconnected scenario: the correct outcome is no feasible schedule, with quantum execution skipped.

This is the useful role of our preview: make a small planning formulation, a real simulated circuit and its limitations visible together. Read the hands-on QAOA guide next if you want a repeatable procedure for examining the results.