A useful way to diagnose a bias tee that stops working below its stated lower frequency is to model it as three separate impedances rather than treating the cutoff as one component value. The RF blocking inductor must present high shunt impedance relative to the system reference impedance, but its self-capacitance, ferrite loss, and package resonance can make the effective impedance collapse well before the nominal SRF. The DC blocking capacitor must present low series impedance, but its real limit is usually set by the required insertion-loss/return-loss budget, not simply by Xc = 50 ohms. Finally, the bias source return and board layout define the actual common-mode and differential-mode paths; a large capacitor placed far from the active device can still leave a long, inductive bias loop.

A reproducible bench sequence is to first measure the blocking capacitor alone on a calibrated VNA through a controlled fixture, using port extension or a proper de-embedding method, and record where return loss crosses your design threshold. Then measure the RF choke or inductor as a shunt element in a grounded fixture, extracting both impedance magnitude and phase rather than relying on datasheet SRF. Combine those measured one-ports in a linear circuit model with the intended 50-ohm system and compare the predicted S21/S11 with the assembled bias tee. If the assembled result is substantially worse, inspect layout inductance, ground-via placement, ferrite saturation, and coupling between the inductor and RF trace before changing component values.

For troubleshooting, keep the bias supply disconnected or heavily filtered during the first VNA check so the supply's low-frequency output impedance does not dominate the result. Then reconnect the supply and repeat the sweep; any new low-frequency notch or ripple points to the bias network or cable resonance rather than the bias tee capacitor alone. The main limitation is that ferrite and high-K capacitor behavior are strongly dependent on DC bias, current, temperature, and fixture geometry, so small-signal bench data should be validated under the actual operating bias point. AI-generated technical discussion; verify against authoritative documentation.