R
Member
Bias-tee self-resonance sets an RoF link's low-end corner — and it hides in the S21 you blame on the MZM
Most RoF trouble reports start with "the MZM is rolling off below 50 MHz". Nine times out of ten the modulator is fine and the bias network is the culprit, because the bias tee's inductor stops behaving like an inductor well inside the band people think they designed for.
The mechanism, in the order it actually bites:
1) A 1 mH chip inductor with ~2 pF of stray capacitance self-resonates around 3.5 MHz. Below SRF it is high impedance and the RF sees the bias supply. Above SRF it looks capacitive, impedance falls with frequency, and the RF path now loads the bias supply's output impedance — which is usually an op-amp output with tens of ohms and a big output capacitor. The result is a low-end corner plus a ripple whose periodicity follows the bias loop's own poles, not the modulator's.
2) The DC return path is the second trap. Plenty of MZM packages terminate the RF electrode into 50 ohm to ground, but some bias tees are AC-coupled on the bias side. If the return is broken, the termination is no longer 50 ohm at DC, the operating point walks with the average signal, and you get a slow envelope-dependent drift that looks exactly like thermal bias drift. Check it with a DMM before you check it with a VNA.
3) Bias supply noise is third-order for NF but first-order for SFDR at low offset. A switching bias supply with 200 mVpp ripple at 300 kHz will show up as close-in spurs once it modulates the bias point, and no amount of RF shielding fixes a spur that enters through the DC pin.
What I do now on a new build, in this order:
- Ask the bias tee vendor for SRF, not just inductance, and require SRF > 3x the highest frequency of interest. If the datasheet only gives L and DCR, assume the worst.
- Sweep S21 from 10 kHz, not from 10 MHz. The low-end corner is invisible if you start the sweep where the problem lives.
- With the laser off (CW, no modulation), sweep S21 again. Any difference between laser-on and laser-off tells you the electrical path is interacting with the optical one.
- Look at S22 with the bias supply connected and then with a 50 ohm load replacing it. The delta is your bias network's contribution.
- For narrowband links (say 100 MHz - 1 GHz), a 10 kΩ resistor in series with the bias feed plus 100 nF to ground often beats a broadband bias tee, because a resistor cannot self-resonate.
A number worth remembering: moving from a 1 mH part with 2 pF stray to a 100 uH part with 0.5 pF stray moved a measured low-end -1 dB point from 45 MHz to 8 MHz on the same board, with no change to the modulator or the TIA.
So: what are people using for the 10 MHz - 18 GHz range where a single bias tee cannot cover it? Active bias tees with a current source and a controlled output impedance look attractive on paper, but I have not seen one that survives a 20 Vpp drive without adding spurs. If anyone has field data on active bias tees in RoF links, I would like to see the S21 at 5 MHz and the SFDR at 1 MHz offset.
Most RoF trouble reports start with "the MZM is rolling off below 50 MHz". Nine times out of ten the modulator is fine and the bias network is the culprit, because the bias tee's inductor stops behaving like an inductor well inside the band people think they designed for.
The mechanism, in the order it actually bites:
1) A 1 mH chip inductor with ~2 pF of stray capacitance self-resonates around 3.5 MHz. Below SRF it is high impedance and the RF sees the bias supply. Above SRF it looks capacitive, impedance falls with frequency, and the RF path now loads the bias supply's output impedance — which is usually an op-amp output with tens of ohms and a big output capacitor. The result is a low-end corner plus a ripple whose periodicity follows the bias loop's own poles, not the modulator's.
2) The DC return path is the second trap. Plenty of MZM packages terminate the RF electrode into 50 ohm to ground, but some bias tees are AC-coupled on the bias side. If the return is broken, the termination is no longer 50 ohm at DC, the operating point walks with the average signal, and you get a slow envelope-dependent drift that looks exactly like thermal bias drift. Check it with a DMM before you check it with a VNA.
3) Bias supply noise is third-order for NF but first-order for SFDR at low offset. A switching bias supply with 200 mVpp ripple at 300 kHz will show up as close-in spurs once it modulates the bias point, and no amount of RF shielding fixes a spur that enters through the DC pin.
What I do now on a new build, in this order:
- Ask the bias tee vendor for SRF, not just inductance, and require SRF > 3x the highest frequency of interest. If the datasheet only gives L and DCR, assume the worst.
- Sweep S21 from 10 kHz, not from 10 MHz. The low-end corner is invisible if you start the sweep where the problem lives.
- With the laser off (CW, no modulation), sweep S21 again. Any difference between laser-on and laser-off tells you the electrical path is interacting with the optical one.
- Look at S22 with the bias supply connected and then with a 50 ohm load replacing it. The delta is your bias network's contribution.
- For narrowband links (say 100 MHz - 1 GHz), a 10 kΩ resistor in series with the bias feed plus 100 nF to ground often beats a broadband bias tee, because a resistor cannot self-resonate.
A number worth remembering: moving from a 1 mH part with 2 pF stray to a 100 uH part with 0.5 pF stray moved a measured low-end -1 dB point from 45 MHz to 8 MHz on the same board, with no change to the modulator or the TIA.
So: what are people using for the 10 MHz - 18 GHz range where a single bias tee cannot cover it? Active bias tees with a current source and a controlled output impedance look attractive on paper, but I have not seen one that survives a 20 Vpp drive without adding spurs. If anyone has field data on active bias tees in RoF links, I would like to see the S21 at 5 MHz and the SFDR at 1 MHz offset.
