Start with the receiver

A ground receiver must track an incoming beam through changing conditions and deliver useful light to its detector. The relevant loss depends on the instrument: pointing error, wavefront distortion, calibration drift, and sensing delay call for different measurements and corrections.

For a fiber-coupled receiver, the spatial overlap between the incoming field and the receiving mode matters. For other architectures, the endpoint must follow the detector and optical design. A smaller wavefront error alone does not establish better coupling or more delivered data.

How much information does a scalar error miss?

Our public diffraction component computes overlap between a propagated optical field and a Gaussian receiving mode. Equal phase variance need not produce equal coupling: the spatial structure of the aberration matters.

Declared optical caseRetained numerical result
Clear pupil, no aberrationCoupling efficiency 0.814512, versus the ideal Gaussian limit 0.814529 at the declared mode radius.
Defocus, 1 rad RMS phase0.36829 of the zero-aberration coupling.
Astigmatism, 1 rad RMS phase0.50416 of the zero-aberration coupling.

An exponential depending only on phase variance gives 0.36788 for both aberrations. The difference shows the information lost by that scalar summary in these monochromatic examples.

The study retains 13 cases on 64², 128², and 256² pupil grids; the examples above use the finest grid. These are optical-component calculations, not controller gains or measured fiber performance. The receiving mode is an ideal Gaussian, not a measured fiber eigenmode. Inspect the retained comparison ↗

Control evidence is a separate step

In Prediction Boundary, five controller families were compared across 600 held-out trials of a delayed adaptive-optics simulation. A two-tap predictor reduced residual wavefront error relative to the reference integrator in five of twelve conditions. Seven comparisons remained uncertain; the learned mixture did not beat both reference controllers in any condition.

That result motivates testing prediction against strong classical baselines. It does not establish a receiver benefit. The fiber-overlap study does not re-score those controller trials, and neither study measures delivered communication data.

A proposed receiver evaluation

Our pilot protocol begins with one instrument, one measurable problem, and synchronized sensor, actuator, calibration, and receiver records. The existing controller provides the baseline.

  1. Offline. Check timing and calibration, reproduce the baseline, and evaluate predictions on held-out recordings. A model-based counterfactual remains dependent on the model.
  2. Shadow mode. Measure predictions, latency, and failure behavior alongside the operating system, without sending commands to the hardware.
  3. Controlled bench comparison. With the instrument owner’s approval, compare bounded interventions and measure the receiver outcome. Only this stage can establish the effect of changing the controller on that bench.

The primary endpoint depends on the architecture and available measurements. For a fiber receiver it may be coupled power or fade duration; delivered bits require modem counters. A thresholded coupling proxy is not measured throughput. Missing records, loss of lock, and failed runs belong in the evaluation.

This is a prospective protocol. No ground-station pilot, hardware control gain, or on-sky performance result is established by the studies above.

Code, methods, and results