R
Member
Power amplifiers that pass ACPR at room temperature can easily fail it at temperature extremes. Here's how I track it down in practice.
1. Characterize first, guess never
Sweep the PA from -40 to +85 C on a thermal chuck while running a modulated signal, and log ACPR, gain, and PAE at each step. The curve shape tells you more than any single pass/fail number — look for knees where linearity falls off a cliff.
2. The usual suspects
- Bias shift: quiescent current drifts with temperature, pushing the device into a less linear region. A temperature-compensated bias network is worth its weight.
- Matching drift: capacitors (especially high-K ceramics) change value with temperature, detuning input/output match.
- Memory effects: thermal memory shows up as asymmetric sidebands — if upper and lower ACPR differ, look here.
3. DPD can only do so much
Digital predistortion helps, but it can't fix a PA whose gain has collapsed 3 dB at hot. Get the analog behavior reasonable first, then let DPD clean up the residuals. Also re-train or adapt DPD coefficients across temperature — a room-temp LUT will lie to you at +85 C.
4. Don't forget the test setup
Cables, attenuators, and even the spectrum analyzer's own temperature drift can masquerade as PA drift. Do a thru-cal at temperature before blaming the DUT.
5. Spec it honestly
If the PA only meets ACPR in a 20-degree window, say so in the datasheet. Customers find out anyway — usually at the worst possible time.
How do you handle temp-dependent linearity in your designs?
1. Characterize first, guess never
Sweep the PA from -40 to +85 C on a thermal chuck while running a modulated signal, and log ACPR, gain, and PAE at each step. The curve shape tells you more than any single pass/fail number — look for knees where linearity falls off a cliff.
2. The usual suspects
- Bias shift: quiescent current drifts with temperature, pushing the device into a less linear region. A temperature-compensated bias network is worth its weight.
- Matching drift: capacitors (especially high-K ceramics) change value with temperature, detuning input/output match.
- Memory effects: thermal memory shows up as asymmetric sidebands — if upper and lower ACPR differ, look here.
3. DPD can only do so much
Digital predistortion helps, but it can't fix a PA whose gain has collapsed 3 dB at hot. Get the analog behavior reasonable first, then let DPD clean up the residuals. Also re-train or adapt DPD coefficients across temperature — a room-temp LUT will lie to you at +85 C.
4. Don't forget the test setup
Cables, attenuators, and even the spectrum analyzer's own temperature drift can masquerade as PA drift. Do a thru-cal at temperature before blaming the DUT.
5. Spec it honestly
If the PA only meets ACPR in a 20-degree window, say so in the datasheet. Customers find out anyway — usually at the worst possible time.
How do you handle temp-dependent linearity in your designs?