L
Admin
SAW and BAW duplexers / filters get a datasheet Tc and a room-temp S-parameter file. On a handset or IoT board the part that fails at hot/cold is often **not** the resonator Tc you quoted in the link budget — it is the matching network walking, the ground return changing, or self-heating under TX that the chuck never saw.
**1. Separate three drifts**
1. **Resonator frequency** — piezoelectric / package Tc moves passband edges.
2. **Match network** — high-K MLCCs, inductor Q, and pad parasitics move Γin / Γout with temperature and bias.
3. **Fixture / enclosure** — shield can, battery lid, and plastic near the filter change the effective ground and stray C when the product heats.
If you only re-measure S21 of the bare filter on a connectorised eval board, you will sign off (1) and still fail (2)+(3) in the product.
**2. Soak with RF on for TX paths**
Cold/hot chuck with the filter unpowered misses self-heating under uplink. Habit for a TX SAW/BAW path: apply the modulated drive (or at least CW at average TX power) during the dwell, monitor package or nearby board temperature, and take S-parameters / ACLR / desense only after Idc and case temp have settled. A 5-minute timer is not equilibrium on a heavy carrier.
**3. What to log at each temperature**
- Passband edges (or 3 dB / rejection points you care about)
- In-band S21 and the **match** at the system reference plane (not only the eval SMA)
- Adjacent rejection that protects the LNA or the PA
- For duplexers: TX→RX isolation at the aggressor channel, not only mid-band
If S21 mid-band is flat but the match walks and ACLR / desense moves, stop blaming “SAW Tc” and look at the matching capacitors and the ground vias under the part.
**4. Matching parts that lie with temperature**
X7R / high-K caps in the match are a classic: room-temp tune looks perfect, hot corner opens the return loss and the PA sees a different load (or the LNA NF rises). Prefer C0G/NP0 in the RF match where Q and stability matter, and re-check the tune at the temperature corners you ship — not only at 25 °C.
**5. Production habit**
Keep a golden board with a characterised filter + match. After a vendor lot change or a shield-can revision, repeat the hot/cold edge check against that golden. Lot-to-lot Tc is usually small; lot-to-lot package parasitics plus your match tolerance is where programmes get surprised.
**Discussion**
On your last SAW/BAW bring-up, which moved first at temperature — passband edge, input match, or TX→RX isolation — and did fixing the match capacitors buy more margin than arguing about the filter Tc?
— rfforum 编辑部 / Editorial Team
**1. Separate three drifts**
1. **Resonator frequency** — piezoelectric / package Tc moves passband edges.
2. **Match network** — high-K MLCCs, inductor Q, and pad parasitics move Γin / Γout with temperature and bias.
3. **Fixture / enclosure** — shield can, battery lid, and plastic near the filter change the effective ground and stray C when the product heats.
If you only re-measure S21 of the bare filter on a connectorised eval board, you will sign off (1) and still fail (2)+(3) in the product.
**2. Soak with RF on for TX paths**
Cold/hot chuck with the filter unpowered misses self-heating under uplink. Habit for a TX SAW/BAW path: apply the modulated drive (or at least CW at average TX power) during the dwell, monitor package or nearby board temperature, and take S-parameters / ACLR / desense only after Idc and case temp have settled. A 5-minute timer is not equilibrium on a heavy carrier.
**3. What to log at each temperature**
- Passband edges (or 3 dB / rejection points you care about)
- In-band S21 and the **match** at the system reference plane (not only the eval SMA)
- Adjacent rejection that protects the LNA or the PA
- For duplexers: TX→RX isolation at the aggressor channel, not only mid-band
If S21 mid-band is flat but the match walks and ACLR / desense moves, stop blaming “SAW Tc” and look at the matching capacitors and the ground vias under the part.
**4. Matching parts that lie with temperature**
X7R / high-K caps in the match are a classic: room-temp tune looks perfect, hot corner opens the return loss and the PA sees a different load (or the LNA NF rises). Prefer C0G/NP0 in the RF match where Q and stability matter, and re-check the tune at the temperature corners you ship — not only at 25 °C.
**5. Production habit**
Keep a golden board with a characterised filter + match. After a vendor lot change or a shield-can revision, repeat the hot/cold edge check against that golden. Lot-to-lot Tc is usually small; lot-to-lot package parasitics plus your match tolerance is where programmes get surprised.
**Discussion**
On your last SAW/BAW bring-up, which moved first at temperature — passband edge, input match, or TX→RX isolation — and did fixing the match capacitors buy more margin than arguing about the filter Tc?
— rfforum 编辑部 / Editorial Team
