L
Admin
[Technical article / FAQ, rfforum Editorial Team]
An FR2 link that looks clean when you sample mid-dwell and fails right after a TCI or beam switch is often a transient story, not a steady-state EVM story. Settling of the RF chain, AGC, and beamformer weights owns the first symbols; averaging them away invents a false pass.
1. What the switch owns
- Beam / TCI changes move phase shifters, switches, and sometimes PA bias. The RF and baseband loops need time before constellation metrics are meaningful.
- Mid-dwell EVM skips the worst symbols. A release gate that only looks mid-dwell will ship a UE or gNB that drops the first slots after every switch.
- Scheduler assumptions about guard symbols must match measured settle time, not datasheet switch time alone.
2. Freeze list before you blame the modem DSP
- Switch type (TCI / beam / panel), commanded vs measured RF settle to within 1 dB / target EVM
- EVM vs symbol index for at least one dwell after the switch, not a single mid-dwell number
- AGC and DPD state: frozen, fast-adapt, or hold across the boundary
- Cable and chamber: confirm the transient survives a conducted test so the air interface is not inventing it
3. Habit
Never freeze "beam switch OK" from mid-dwell EVM alone. Archive first-N-symbol EVM and settle time next to every TCI release. If adding guard symbols or holding AGC kills the fail, the equalizer was never the patient.
4. Practical gate
- Plot EVM for the first 1–2 slots after switch
- Compare hold vs adapt AGC across the boundary
- Write measured settle time on the same sheet as the scheduler guard
Discussion: on your last FR2 switch fail, how many symbols were trash, and do you require first-symbol EVM on every beam/TCI acceptance?
— rfforum 编辑部 / Editorial Team
An FR2 link that looks clean when you sample mid-dwell and fails right after a TCI or beam switch is often a transient story, not a steady-state EVM story. Settling of the RF chain, AGC, and beamformer weights owns the first symbols; averaging them away invents a false pass.
1. What the switch owns
- Beam / TCI changes move phase shifters, switches, and sometimes PA bias. The RF and baseband loops need time before constellation metrics are meaningful.
- Mid-dwell EVM skips the worst symbols. A release gate that only looks mid-dwell will ship a UE or gNB that drops the first slots after every switch.
- Scheduler assumptions about guard symbols must match measured settle time, not datasheet switch time alone.
2. Freeze list before you blame the modem DSP
- Switch type (TCI / beam / panel), commanded vs measured RF settle to within 1 dB / target EVM
- EVM vs symbol index for at least one dwell after the switch, not a single mid-dwell number
- AGC and DPD state: frozen, fast-adapt, or hold across the boundary
- Cable and chamber: confirm the transient survives a conducted test so the air interface is not inventing it
3. Habit
Never freeze "beam switch OK" from mid-dwell EVM alone. Archive first-N-symbol EVM and settle time next to every TCI release. If adding guard symbols or holding AGC kills the fail, the equalizer was never the patient.
4. Practical gate
- Plot EVM for the first 1–2 slots after switch
- Compare hold vs adapt AGC across the boundary
- Write measured settle time on the same sheet as the scheduler guard
Discussion: on your last FR2 switch fail, how many symbols were trash, and do you require first-symbol EVM on every beam/TCI acceptance?
— rfforum 编辑部 / Editorial Team
