L
Admin
[Technical article / FAQ, rfforum Editorial Team]
A 90 dB stopband that appears on a slow VNA sweep can vanish when the same filter is re-measured with a wider IF bandwidth or a faster point rate. The DUT did not change; the instrument noise floor and averaging did. Rejection claims that omit IFBW and averaging are not portable.
1. What the IFBW actually buys
- Narrower IF bandwidth lowers the noise floor roughly as 10 log(IFBW), at the cost of dwell time and sensitivity to source drift.
- A stopband that sits only a few dB above the noise floor is an instrument-limited number, not a filter number. Widening IFBW by 10x can erase 10 dB of "rejection" that was never there.
- Averaging and smoothing hide the same problem: they paint a clean floor over a noisy one without changing the DUT.
2. Freeze list before you ship a rejection spec
- IFBW, point count, sweep time, averaging, and whether the floor was verified with a known thru or a shielded open
- Whether the stopband stays ≥10 dB above the measured noise floor at that IFBW
- A repeat at 10x wider IFBW: if rejection collapses to the floor, the first number was instrument-limited
- Cable leakage and fixture radiation: a "rejection" that moves when you flex the cable is a radiation path, not a filter pole
3. Habit
Never freeze "≥90 dB rejection" from a single slow sweep. Archive IFBW, averaging, and a noise-floor screenshot next to every stopband claim. If a 10x wider IFBW kills the margin to the floor, stop quoting the slow-sweep number as the DUT.
4. Practical five-minute gate
- Note IFBW and averaging on the plot title or traveler
- Compare floor with a shielded load or thru at the same settings
- Re-sweep stopband at 10x IFBW; keep only what survives with ≥10 dB margin to the floor
Discussion: on your last stopband fight, did the fix land in IFBW, leakage, or a real filter redesign — and do you require IFBW on every rejection traveler?
— rfforum 编辑部 / Editorial Team
A 90 dB stopband that appears on a slow VNA sweep can vanish when the same filter is re-measured with a wider IF bandwidth or a faster point rate. The DUT did not change; the instrument noise floor and averaging did. Rejection claims that omit IFBW and averaging are not portable.
1. What the IFBW actually buys
- Narrower IF bandwidth lowers the noise floor roughly as 10 log(IFBW), at the cost of dwell time and sensitivity to source drift.
- A stopband that sits only a few dB above the noise floor is an instrument-limited number, not a filter number. Widening IFBW by 10x can erase 10 dB of "rejection" that was never there.
- Averaging and smoothing hide the same problem: they paint a clean floor over a noisy one without changing the DUT.
2. Freeze list before you ship a rejection spec
- IFBW, point count, sweep time, averaging, and whether the floor was verified with a known thru or a shielded open
- Whether the stopband stays ≥10 dB above the measured noise floor at that IFBW
- A repeat at 10x wider IFBW: if rejection collapses to the floor, the first number was instrument-limited
- Cable leakage and fixture radiation: a "rejection" that moves when you flex the cable is a radiation path, not a filter pole
3. Habit
Never freeze "≥90 dB rejection" from a single slow sweep. Archive IFBW, averaging, and a noise-floor screenshot next to every stopband claim. If a 10x wider IFBW kills the margin to the floor, stop quoting the slow-sweep number as the DUT.
4. Practical five-minute gate
- Note IFBW and averaging on the plot title or traveler
- Compare floor with a shielded load or thru at the same settings
- Re-sweep stopband at 10x IFBW; keep only what survives with ≥10 dB margin to the floor
Discussion: on your last stopband fight, did the fix land in IFBW, leakage, or a real filter redesign — and do you require IFBW on every rejection traveler?
— rfforum 编辑部 / Editorial Team
