Sub-Microsecond Sampling

The acquisition chain digitizes every analog channel below the microsecond, fast enough to resolve the transients that precede disruptions and plug collapse.

Why sub-microsecond

The control-relevant physics of both machines has structure below the microsecond. Mirnov coils see MHD modes whose rotation and growth are resolved only when the sampler is faster than the mode. In the burner, the ambipolar potential and end-plug density can move on comparably short timescales. Sampling below the microsecond is what lets the precursor features exist at all; a slow sampler simply cannot represent a fast quench precursor.

The Nyquist floor

For a diagnostic with meaningful spectral content up to f_max, faithful reconstruction requires a sample rate above 2*f_max (see Nyquist and aliasing). The fabric sets each channel's rate from its physics bandwidth, not a single global clock rate, then anti-alias filters the front end to that rate. Magnetics and ECE run fastest; slower thermal and strain channels are sampled proportionately.

# per-channel rate from physics bandwidth, then guard
def sample_rate(f_max_hz, guard=2.5):
    # guard > 2 (Nyquist) to leave anti-alias transition band
    return guard * f_max_hz

# a fast Mirnov channel with 500 kHz content
fs = sample_rate(500e3)   # 1.25 MHz  -> ~0.8 us per sample
assert 1/fs < 1e-6        # sub-microsecond

One clock, many channels

All digitizers share a disciplined clock so that cross-channel phase is preserved to the sample. Equilibrium reconstruction and mode analysis are multi-channel inferences; a per-channel timing skew would masquerade as a real plasma asymmetry. Timestamps are distributed by a precision time protocol and audited (see precision timestamping).

Sub-microsecond, not the control deadline

Sampling below a microsecond is distinct from the L1 control deadline. L2 acquires and validates at these rates and streams engineered features toward the sub-10 microsecond control boundary; the hard real-time actuation and the autonomous failsafe live in L1. L2's obligation is that the numbers crossing that boundary are already clean, phase-coherent, and normalized.