zprofile

A frequency-domain measurement says how much is wrong with a cable; it cannot say where. For a manufacturer, the difference matters enormously: "return loss fails at 391 MHz" starts an investigation, while "there is a defect 6.0 m from the near end, 0.25 Ω ripple every 26 cm on top" ends one.

The obvious tool — an inverse FFT into the time domain, as a VNA's "time-domain option" does — is quietly biased: on a lossy line the step response keeps rising as √t, so a 15 m automotive pair reads 4 to 9 Ω high by the far end. Real TDR instruments show the same slope. The instrument is not lying; the plain transform is just not the right inverse.

Layer peeling walks the cable cell by cell: read the first reflection, deduce the first impedance step, remove that interface's effect from everything that follows, repeat. On a lossless line the recursion is textbook. On a real line, every later echo has been attenuated and dispersed by everything before it, so the algorithm must amplify each buried reflection by exactly the loss it suffered — and amplification of a measured signal amplifies noise with it. The reconstruction therefore regularises the inverse propagation (Wiener-style, with an adaptive read time) and corrects each cell's read against a lossy remainder — the difference between an algorithm that works on paper and one that survives −85 dB of instrument noise.

Just as important is what surrounds the algorithm: a sensitivity budget. Every processing choice — window, bandwidth, DC estimator, mask — is varied, and the shift of every reported statistic is tabulated. The result does not just say "the mean is 100.2 Ω"; it says how much of that could be the processing.