Probe
The monitor observes one channel and returns detuning or an equivalent signal.
AGE explicitly represents shared dynamics, local perturbations and serial observation. The gain comes from structure, not from applying a more elaborate filter to each channel in isolation.
For an N-channel bank, detuning is decomposed into a shared contribution and one device-specific contribution per channel. Parameters are not assumed universal: they are identified or bounded for the platform under study.
The model is a testable hypothesis. If residuals show the common/local structure is insufficient, it must be extended or rejected.
state xₖ = [ cₖ, ℓ₁ₖ, …, ℓₙₖ ]ᵀ dynamics cₖ₊₁ = a_c cₖ + w_c ℓᵢₖ₊₁ = aᵢ ℓᵢₖ + wᵢ detuning Δλᵢₖ = cₖ + ℓᵢₖ measurement yₖ = H(channelₖ) xₖ + vₖ # every probe updates the common state
Under round-robin probing, each channel may go many periods without its own measurement. Yet any measurement can update the shared component of every channel.
The monitor observes one channel and returns detuning or an equivalent signal.
The innovation is allocated between common and local components under the model.
Every channel posterior changes through the shared common state.
The state is propagated to the next probe with explicit uncertainty.
Improvement in the same window used to fit parameters is not evidence of predictive performance. The recommended route therefore separates identification and evaluation chronologically.
The evaluation uses exported data. A later integration adds monitor and control adapters without assuming every platform exposes the same interfaces.
Detuning, wavelength, power, temperature, channel state and timestamps.
Effective parameters, covariance structure, noise and residual tests.
Joint state, uncertainty and between-observation prediction.
Corrections constrained by power, latency, range and system safety.
The actuation layer is platform-specific and requires experimental validation. The current open package demonstrates offline estimation in simulation.
Download the technical note and reproducibility package. Both distinguish the nominal scenario from future measured validation.