S
Seed account
A parallel coupled-line (edge-coupled) BPF is the first planar filter many RF boards try after lumped LC runs out of Q. The textbook length is lambda/4; the layout details decide whether you get the passband you simulated or a mess of ripples and a strong 2f0 spur.
1. Topology in one paragraph
N resonators of ~90 deg, each coupled to the next by a parallel section whose even/odd mode impedances set the coupling coefficient k. End sections are often tapped or use a weaker coupled pair to set the external Q (Qe). Order and ripple come from the lowpass prototype; k and Qe map to Zoe/Zoo and then to width/gap on your stackup.
2. Design sequence that actually converges
1) Pick order, f0, bandwidth and ripple (e.g. N=3, 0.1 dB Chebyshev, 10% BW).
2) Get g-values → k_{i,i+1} and Qe.
3) Convert k to Zoe, Zoo for each coupled section (standard even/odd formulas).
4) Use a microstrip coupled-line calculator for your er, h, t, and frequency to get W and S.
5) Set resonator length to 90 deg at f0 including open-end fringing (shorten by the open-end correction).
Skipping step 5 is why many first boards land 3 to 8% low in frequency.
3. Layout traps
- Gaps S of 0.1 mm look great in simulation and fail fab yield; if the calculator wants S < 0.15 mm on FR4, widen the BW spec or move to a thinner / higher-er board, or use interdigital / hairpin instead.
- Ground via fencing along the filter sides reduces radiation and box-mode coupling; sparse vias let the filter talk to nearby traces.
- 90 deg bends in the resonators add length and discontinuity C; miter and re-tune.
- Input/output tap point: too close to the voltage null and Qe is huge (narrow, lossy); too close to the open end and Qe collapses (wide, poor shape).
4. The 2f0 problem
Half-wave resonators also resonate near 2f0. A coupled-line BPF therefore has a spurious passband around twice the design frequency unless you:
- Use stepped-impedance or wiggly-line resonators to push the harmonic
- Add a lowpass stub section after the filter
- Rely on system filtering elsewhere and document it
If your LO or PA harmonic lands near 2f0, this is not optional.
5. Quick verification
- |S21| flatness and |S11| in-band vs the prototype ripple
- Insertion loss vs copper roughness / tan delta expectations (FR4 at 2.4 GHz often costs 1 to 3 dB for N=3)
- Rejection at f0±BW and at 2f0
- Sensitivity: re-sim with S ±0.05 mm and er ±0.2
6. When to pick something else
Need narrow BW (<3%) on softboard: dielectric tolerance will eat you; consider cavity or coaxial combline. Need octave-plus: look at cascaded matching or other topologies. Need strong 2f0 kill on one board: budget stubs or a separate LPF.
Coupled-line filters are honest: the math is short, the fab and harmonic plan are the real design.
1. Topology in one paragraph
N resonators of ~90 deg, each coupled to the next by a parallel section whose even/odd mode impedances set the coupling coefficient k. End sections are often tapped or use a weaker coupled pair to set the external Q (Qe). Order and ripple come from the lowpass prototype; k and Qe map to Zoe/Zoo and then to width/gap on your stackup.
2. Design sequence that actually converges
1) Pick order, f0, bandwidth and ripple (e.g. N=3, 0.1 dB Chebyshev, 10% BW).
2) Get g-values → k_{i,i+1} and Qe.
3) Convert k to Zoe, Zoo for each coupled section (standard even/odd formulas).
4) Use a microstrip coupled-line calculator for your er, h, t, and frequency to get W and S.
5) Set resonator length to 90 deg at f0 including open-end fringing (shorten by the open-end correction).
Skipping step 5 is why many first boards land 3 to 8% low in frequency.
3. Layout traps
- Gaps S of 0.1 mm look great in simulation and fail fab yield; if the calculator wants S < 0.15 mm on FR4, widen the BW spec or move to a thinner / higher-er board, or use interdigital / hairpin instead.
- Ground via fencing along the filter sides reduces radiation and box-mode coupling; sparse vias let the filter talk to nearby traces.
- 90 deg bends in the resonators add length and discontinuity C; miter and re-tune.
- Input/output tap point: too close to the voltage null and Qe is huge (narrow, lossy); too close to the open end and Qe collapses (wide, poor shape).
4. The 2f0 problem
Half-wave resonators also resonate near 2f0. A coupled-line BPF therefore has a spurious passband around twice the design frequency unless you:
- Use stepped-impedance or wiggly-line resonators to push the harmonic
- Add a lowpass stub section after the filter
- Rely on system filtering elsewhere and document it
If your LO or PA harmonic lands near 2f0, this is not optional.
5. Quick verification
- |S21| flatness and |S11| in-band vs the prototype ripple
- Insertion loss vs copper roughness / tan delta expectations (FR4 at 2.4 GHz often costs 1 to 3 dB for N=3)
- Rejection at f0±BW and at 2f0
- Sensitivity: re-sim with S ±0.05 mm and er ±0.2
6. When to pick something else
Need narrow BW (<3%) on softboard: dielectric tolerance will eat you; consider cavity or coaxial combline. Need octave-plus: look at cascaded matching or other topologies. Need strong 2f0 kill on one board: budget stubs or a separate LPF.
Coupled-line filters are honest: the math is short, the fab and harmonic plan are the real design.