Resample

Resample converts a signal from its native sample rate $f_s$ to a new rate target_rate ($f_t$, in Hz) using sinc-based, anti-aliased interpolation. It takes one signal input and produces one result signal carrying the same physical unit but a new time base defined by $f_t$. The output length scales by the rate ratio: an $N$-sample input becomes approximately $N\cdot f_t/f_s$ samples.

The core is bandlimited interpolation: each output sample is reconstructed from a windowed $\mathrm{sinc}$ kernel over neighboring input samples. On downsampling ($f_t<f_s$) the kernel cutoff is lowered to the new Nyquist frequency $f_t/2$, suppressing frequencies that would otherwise alias. On upsampling ($f_t>f_s$) the kernel interpolates between existing samples without adding spectral content above the original Nyquist.

Unit is preserved unchanged — resampling is purely a time-axis operation and does not touch the sample magnitude dimension. The uncertainty ($\sigma$) channel is propagated through the same linear interpolation weights; because each output sample is a weighted sum of input samples, correlated smoothing of the sigma field is expected rather than independent-error scaling.

This node is stateless: each evaluation resamples its input block independently with no carried filter memory or phase accumulator between calls. target_rate is constrained to $\ge 1.0$ Hz.