A hairpin filter with a displaced passband needs a different investigation from one with the correct center but the wrong bandwidth or ripple. Start by separating resonator-frequency errors, coupling between resonators and loading at the input and output. They interact, so the aim is to test controlled hypotheses before changing several dimensions together.
This worksheet is for debugging a civilian RF bandpass filter. Its numerical responses are invented teaching examples, not simulations or measurements. Apply the method to your calibrated S-parameters and fabrication dimensions, using the microwave-filter reference for the wider design context.
Establish a comparable baseline before calling it detuning
Save the original target mask, synthesized response, electromagnetic model and measured two-port data. Record port impedance, frequency grid, calibration, reference planes, temperature and connector assembly. Compare the same physical network: a model ending at the filter feed cannot be compared directly with a measurement that includes unmodeled launches.
Inspect the fabricated line widths, neighboring gaps, dielectric spacing, copper thickness and surface coating. Confirm the exact laminate and its modeling data. A board described as RO4003C still needs a defined construction. Rogers’ discussion of PCB parameters affecting impedance identifies substrate thickness, conductor dimensions and Dk as relevant inputs. Do not use Dk as a free fitting parameter to conceal an incorrect cross-section.
If the discrepancy changes after reconnecting a cable, check measurement repeatability first. The connector-launch transfer checklist helps identify a fixture or stackup transition that could obscure the filter response.
Compare frequency shift and shape separately
Choose a consistent definition for lower and upper passband markers. For this hypothetical example, each edge is 3 dB below that response’s maximum transmission. Define the arithmetic midpoint as (fL + fH)/2, bandwidth as fH − fL and fractional bandwidth as bandwidth divided by that midpoint. This bookkeeping midpoint is not a substitute for your synthesis convention.
| Response | fL / fH | Midpoint | Bandwidth | Fractional bandwidth |
|---|---|---|---|---|
| Reference example | 4.800 / 5.200 GHz | 5.000 GHz | 0.400 GHz | 8.0% |
| A: scaled downward | 4.704 / 5.096 GHz | 4.900 GHz | 0.392 GHz | 8.0% |
| B: narrower, same midpoint | 4.880 / 5.120 GHz | 5.000 GHz | 0.240 GHz | 4.8% |
Response A has a 2% lower midpoint while preserving the example’s fractional bandwidth. If the full ripple and return-loss patterns also align after frequency normalization, a common electrical-length error is a useful first hypothesis. The two edge numbers alone cannot establish that conclusion.
Response B directs attention toward coupling and loading, but does not prove either is the only error. Unequal resonator detuning or loss can also change apparent bandwidth. A third possible response might retain the reference edges while developing uneven ripple or poor matching at one edge. That case requires the full S11, S22 and S21 traces, not just a center-frequency correction.
Keep absolute insertion loss alongside these relative edge measurements. A lossy filter can appear acceptable after normalizing its amplitude while still failing the product’s loss budget.
Test resonator frequency with a limited length sweep
Use a weakly excited single-resonator model to establish a reference, then examine the resonators in the assembled layout. Keep the substrate, conductor model, feed arrangement and mesh policy fixed. Sweep a small, manufacturable open-end length change and record resonance movement. Check whether changing the perturbation size gives a consistent local sensitivity.
The ADS Passive Circuit DesignGuide, HPFilter section, distinguishes length tuning that centers the response from width and spacing changes used to shape it. Treat that as a useful division of work, not perfect independence between variables.
For a nearly uniform frequency offset, test a common length change before making every resonator different. If ripple remains distorted after recentering, inspect individual sensitivities. An open-end change near another conductor can affect both frequency and coupling. The wavelength and phase tool provides a scale estimate, while the bent, coupled structure still needs its electromagnetic model.
Measure coupling in a controlled two-resonator model
Build a pair with the intended orientation and surroundings. First check that the isolated resonances coincide; detuning itself can split the response. Use weak excitation and a suitable extraction method, then sweep the physical gap or overlap while holding resonator lengths fixed.
Hong and Lancaster’s hairpin-resonator coupling study relates coupling to split modal frequencies. For an appropriate synchronously tuned pair, the common magnitude estimate is |k| = (fHigh² − fLow²)/(fHigh² + fLow²). Hypothetical modes at 4.95 and 5.05 GHz give about 0.0200. These are pair-mode frequencies, not the passband edges in the table. The magnitude alone does not identify a cross-coupling sign.
The same study shows that mixed electric and magnetic coupling can make the gap relationship nonmonotonic for some orientations. Do not assume that closing any gap always produces the correction you want. Build the curve for the actual geometry, and compare it with the coupling target from the original synthesis.
Keep input and output loading as separate variables
After the internal resonators and couplings are understood, review the feed tap, coupled feed section or other input/output structure. External Q describes loading through the external connection; it is different from the resonator’s unloaded Q. Evaluate the two ends separately when the measured reflection responses differ.
Daniel Swanson’s folded-hairpin port-tuning tutorial separates resonance and tap-point setup, then converts tuning sensitivities into physical changes. It also checks the final layout after removing tuning ports. Its alumina example and dimensional corrections should not be copied into a Rogers PCB design.
A feed change can improve matching while perturbing the end resonator. Recheck the complete filter after each accepted change. If an optimizer uses temporary tuning elements or ports, the deliverable is the verified physical layout with those aids removed.
Choose the next revision from a small experiment matrix
| Observed pattern | First controlled comparison | Record before accepting the change |
|---|---|---|
| Similar shape shifted in frequency | Common resonator-length perturbation | Midpoint, normalized ripple, loss and all return-loss extrema |
| Midpoint close, bandwidth wrong | Relevant adjacent-pair gap or overlap sweep | Pair-mode frequencies, coupling trend and full-filter bandwidth |
| Uneven ripple after recentering | Individual resonance and coupling sensitivities | Which perturbation improves several landmarks consistently |
| Input/output matching differs | Separate feed-loading sweeps after fixture checks | S11, S22, end-resonator movement and insertion loss |
Preserve one baseline and label every trial with its changed parameter. Confirm that mesh refinement and frequency sampling do not change the apparent winner. Use the controlled-impedance guide to document the fabrication dimensions behind the model.
Before releasing artwork, rerun the complete structure with realistic loss, copper thickness, launches and enclosure conditions. Check the original passband and stopband masks, including any transmission zeros. Review dimensional tolerance cases and verify a new prototype. Recentring one trace is not sufficient evidence that the filter meets its specification.
Request a review with the error pattern attached
Send filter artwork, stackup, target mask, calibrated S-parameters, connector model, measured line/gap dimensions and original synthesis assumptions. Include the sweep matrix so the review starts from the hypotheses already tested. Request a hairpin-filter PCB review, or organize the prototype requirements with the RFQ checklist builder.