RF·Tools

Coupling-matrix filter

The usual synthesis starts from a prototype: you order an order and get a ladder whose skirts are whatever that order gives. Here it is the other way round — you say where the transmission zeros must be (at the band edges for a steep skirt, or right on an interfering transmitter), and the filter is computed to suit. A plain chain of resonators cannot do that: it needs couplings that bypass neighbours, which is why the diagram looks unfamiliar. What is computed is the coupling matrix, and from it the coupling coefficients and the external Q — the numbers you lay a board out with.

Transmission zeros are frequencies in megahertz, comma separated, outside the passband. The “pick to suit the zeros” topology decides for itself what to make them with: a symmetric pair needs only a quadruplet, a skewed one needs two triplets and six resonators.

Coupling diagram

4 resonators, 2 transmission zeros Tuning sheet
1.0760.9440.7550.9441.076-0.074S1234Lrf.akzami.com

Circles are resonators, the squares at the ends are the source and the load. The arcs over the chain are couplings that bypass neighbours: they are what makes the zeros. A dashed arc means a negative coupling, and that is not decoration — the sign decides which side of the band the zero falls on, and on a board positive and negative couplings are made differently, magnetically or electrically, that is by a different relative placement of the resonators. The numbers under the circles, where present, are the resonator's offset from the band centre.

Response

Computed straight from the coupling matrix, with no line sections and no field solving: these are ideal resonators with the Q you gave.

Couplings

Between What it is M k
S — 1 along the chain 1.0765 0.04485
1 — 2 along the chain 0.9438 0.03933
2 — 3 along the chain 0.7546 0.03144
3 — 4 along the chain 0.9438 0.03933
4 — L along the chain 1.0765 0.04485
1 — 4 cross coupling, makes a zero -0.0744 -0.00310

M is the normalised coupling; k = M·FBW is the coupling coefficient you then set with the gap between resonators. External Q: Qe 20.7 / 20.7 — it sets the tap point or the gap to the feed line.

What you get

Band100 MHz (4%)
Passband loss0.03 dB
Return loss≥ 21.9 dB
Transmission zeros2250, 2560 MHz
Resonators4

The solution is exact: the response matches the ideal one to machine precision.

Resonator tuning

12400 MHz
22400 MHz
32400 MHz
42400 MHz

If the frequencies differ, the resonators are detuned on purpose: an asymmetric response cannot be had without it.

Dimensions on the board

Hairpin arm19.3 mm
Track width1.10 mm
Fold gap1.02 mm
Tap from the fold3.54 mm
Between k Gap, mm
1 — 2 0.03933 0.95
2 — 3 0.03144 1.14
3 — 4 0.03933 0.95
1 — 4 cross -0.00310 out of reach

The resonator is a hairpin; the gaps come from the resonance splitting of a pair and the tap from the slope of the reflection phase — the same model as in the hairpin filter, and verified there. Every gap found is substituted back and checked: whatever does not agree is marked “out of reach” rather than handed over as a number. The placement of the resonators on the board is not drawn here: a cross coupling calls for a horseshoe, and that layout is still done by hand.

computed without loss: no resonator Q was given. In a narrow-band filter that is exactly what eats the decibels, so the plot is kinder than the board will be
the coupling 0.00310 between resonators 1 and 4 is beyond this model: once the hairpins are far apart the resonance splitting sinks out of sight and the gap can no longer be read from it. A coupling that weak is not usually made with a straight gap between the arms anyway — it is obtained by bringing the ends of the resonators closer in a folded layout, which is a different geometry