Calibration stages

Calibration is split by the domain where a correction is physically meaningful. Every configured array is shape-checked; a mismatch raises an error rather than silently disabling calibration.

ADC domain

adcFrequencyResponse and adcChannelGain act before range FFT. Frequency response corrects sample-domain complex gain, while channel gain corrects emission/RX response. Arrays must broadcast exactly according to the documented canonical ADC axes.

This stage is appropriate for measured front-end gain/phase response and static per-channel imbalance represented in the capture’s native channel layout.

Range domain

After range FFT and MIMO decoding, rangeCoupling is subtracted from (loop, virtual, range) data. It models direct TX/RX leakage and static coupling as a complex range profile:

\[ X_\mathrm{clean}[\ell,m,k] =X[\ell,m,k]-C[m,k]. \]

Subtracting a slow-time mean is not equivalent: mean removal also removes all stationary scene reflectors.

Array domain

arrayPhase multiplies virtual channels after Doppler compensation and before DoA:

\[ \tilde{X}[r,d,m]=X[r,d,m]G_m. \]

The gain convention is correction gain, not measured error. Calibrate before coherently merging duplicate phase centres.

Capture profiles

Dataset adapters parse native JSON, MAT, or text calibration files and construct Calibration. Parsed physical units must be converted to SI once at the reader boundary. A profile should record source filenames and calibration assumptions in frame metadata for reproducibility.

Overlap antenna pairs identify repeated phase centres used for TDM velocity disambiguation. The implementation evaluates candidates within the raw chirp-rate Nyquist interval and selects the calibrated phase-consistency minimum. Invalid pair indices or array-calibration shapes are errors; zero-energy evidence falls back to ambiguity order zero.