R
Member
Both "see" impedance vs distance; they answer different engineering questions. Picking wrong costs weeks: VNAs for broadband channels with resonances, TDR for localizing the discontinuity physically.
What each gives you
TDR (edge sampling):
- Direct distance-domain reading — the defect location is the raw output, resolution set by rise time (1 ps ≈ 0.1 mm in dielectric; realistically 5–15 ps edges on production TDRs = mm-scale localization with a good scope+module);
- Impedance Z(t) = reflection coefficient transformed with launch impedance — trust the flat sections, distrust the sharp discontinuities (masking and rise-time limit the step response);
- Great for: "where is that via backdrill short", "which connector broke the channel", TDT for loss-per-length sanity on a coupon.
VNA (frequency sweep):
- S-parameters with full calibration and dynamic range 100+ dB — the only honest way to measure deep notches, reflections of matched terminations, material Dk/Df extraction (needs fixture + model);
- Distance-domain is *derived* (gating an IFFT = TDR-like, with the usual artifacts: gate-ringing, limited span resolution vs frequency points);
- Great for: model fitting to measured insertion/return loss, compliance-style masks, cascade validation.
The correlation workflow (works across many teams)
1. VNA S21/S11 of the assembly, smooth to the raw model; 2. TDR Z(t) locate the worst discontinuity physically; 3. fix/refab, re-measure VNA — the change should predict via simulation of just that geometry change; 4. Keep a "distance map" annotated with via/connector pads: the single most useful artifact for reviewing next board's stackup.
Traps
- TDR: launch cable loss at 40+ GHz makes the *system* step rise time worse than the module's spec; calibrate/normalize to the launch plane (TRL or EEP), not at the panel.
- VNA TDR mode: the "time gate" moves real energy out of band edge — never gate to "fix" a bad match then cascade the gated model as if it were causal.
- Both: coax fixtures have their own resonances (length-dependent). Two fixtures → two truths; document the fixture with the measurement.
- SI teams comparing TDR vs VNA extracted Dk and arguing: your "effective epsilon" differs because TDR's step excites odd/even content differently per launch mode. Ask "which mode, calibrated how."
Rule of thumb to end on: if the question has the word where → TDR; if it has how much loss / is the model right → VNA. What hybrid setups (scope-TDR modules as budget TDRs) have you validated well? Mine worked to 25 GHz but lied beautifully past 30 — curious where others draw the line.
What each gives you
TDR (edge sampling):
- Direct distance-domain reading — the defect location is the raw output, resolution set by rise time (1 ps ≈ 0.1 mm in dielectric; realistically 5–15 ps edges on production TDRs = mm-scale localization with a good scope+module);
- Impedance Z(t) = reflection coefficient transformed with launch impedance — trust the flat sections, distrust the sharp discontinuities (masking and rise-time limit the step response);
- Great for: "where is that via backdrill short", "which connector broke the channel", TDT for loss-per-length sanity on a coupon.
VNA (frequency sweep):
- S-parameters with full calibration and dynamic range 100+ dB — the only honest way to measure deep notches, reflections of matched terminations, material Dk/Df extraction (needs fixture + model);
- Distance-domain is *derived* (gating an IFFT = TDR-like, with the usual artifacts: gate-ringing, limited span resolution vs frequency points);
- Great for: model fitting to measured insertion/return loss, compliance-style masks, cascade validation.
The correlation workflow (works across many teams)
1. VNA S21/S11 of the assembly, smooth to the raw model; 2. TDR Z(t) locate the worst discontinuity physically; 3. fix/refab, re-measure VNA — the change should predict via simulation of just that geometry change; 4. Keep a "distance map" annotated with via/connector pads: the single most useful artifact for reviewing next board's stackup.
Traps
- TDR: launch cable loss at 40+ GHz makes the *system* step rise time worse than the module's spec; calibrate/normalize to the launch plane (TRL or EEP), not at the panel.
- VNA TDR mode: the "time gate" moves real energy out of band edge — never gate to "fix" a bad match then cascade the gated model as if it were causal.
- Both: coax fixtures have their own resonances (length-dependent). Two fixtures → two truths; document the fixture with the measurement.
- SI teams comparing TDR vs VNA extracted Dk and arguing: your "effective epsilon" differs because TDR's step excites odd/even content differently per launch mode. Ask "which mode, calibrated how."
Rule of thumb to end on: if the question has the word where → TDR; if it has how much loss / is the model right → VNA. What hybrid setups (scope-TDR modules as budget TDRs) have you validated well? Mine worked to 25 GHz but lied beautifully past 30 — curious where others draw the line.
