Cross-Coupled Topology
How Non-Adjacent Couplings Shape the Response
A conventional bandpass filter is a chain of resonators in which energy hops only from one resonator to its immediate neighbor. That single through-path produces an all-pole response (Butterworth or Chebyshev) whose stopband rolls off monotonically. A cross-coupled filter adds at least one coupling between resonators that are not neighbors, for example between the first and fourth resonator of a quadruplet. Energy now travels two ways at once: the long way around the main chain and the short way across the bridge. At the frequency where the two arrivals are equal in amplitude and opposite in phase they cancel, driving the transmission coefficient S21 to zero. Each such cancellation is a finite transmission zero, and it can be parked exactly where the design needs the most rejection.
The standard analysis tool is the N×N coupling matrix M, whose diagonal entries hold the resonator frequency offsets and whose off-diagonal entries hold the normalized couplings. Main-line couplings sit on the first sub-diagonal; a cross coupling appears as an entry away from that diagonal, such as M14 or M16. Synthesis routines first build a transversal or arrow-form matrix from the prescribed reflection and transmission polynomials, then apply a sequence of similarity rotations to drive it into a physically realizable form such as a folded canonical, cascaded-triplet (CT), or cascaded-quadruplet (CQ) network. Because a similarity transform preserves the eigenvalues, the response is identical while the topology becomes buildable in real hardware.
The sign of a cross coupling is as important as its size. A coupling that is opposite in sign to the main couplings (typically capacitive against inductive main irises) places a symmetric pair of zeros on the imaginary axis, steepening both skirts. A same-sign coupling pushes that pair onto the real axis as a complex-conjugate pair, leaving the amplitude Chebyshev-like but equalizing the in-band group delay. A single triplet gives one asymmetric zero, useful when only one band edge needs extra rejection, for instance to protect a transmit band sitting just above a receive filter.
Realizing the Cross Coupling in Hardware
In cavity and dielectric-resonator filters the bridge coupling is created with a probe, a coupling screw, or a slot in the wall between non-adjacent cavities; rotating a probe or moving it from a magnetic-field region to an electric-field region flips the sign. In planar and combline structures the cross coupling can be a short capacitive finger or an extra inductive line laid between non-adjacent resonators. The folded arrangement is popular because it physically places the input and output resonators side by side, so the bridging element between them is mechanically convenient and easy to tune.
Governing Relations
[A] = [M] − jΩ[U] + [q] , with Ω the normalized frequency
S21(Ω) = −2j / (q1qN)1/2 × [A]−1N1 , S11(Ω) = 1 + 2j / q1 × [A]−111
Maximum finite transmission zeros:
Ntz ≤ N − 2 (canonical), one zero per triplet, two per quadruplet
Zero location vs. sign of cross coupling Mcc:
Mcc < 0 ⇒ zeros on jΩ axis (selectivity); Mcc > 0 ⇒ real-axis pair (group-delay equalization)
Where [M] = coupling matrix, [U] = identity with zeros at source/load, [q] = scaled external-Q terms, Ω = (ω/ω0 − ω0/ω) / FBW, N = filter order, FBW = fractional bandwidth.
Topology Comparison
| Topology | Finite zeros | Zero symmetry | Tuning difficulty | Typical use |
|---|---|---|---|---|
| All-pole (inline) | 0 | None | Low | Wideband, simple preselectors |
| Folded canonical | Up to N − 2 | Symmetric | High | Satellite output multiplexers |
| Cascaded triplet (CT) | 1 per triplet | Asymmetric | Medium | One-sided rejection, duplexers |
| Cascaded quadruplet (CQ) | 2 per quad | Symmetric pair | Medium to high | Quasi-elliptic base-station filters |
| Extracted-pole | 1 per section | Independent | High | Precise zero placement, diplexers |
Frequently Asked Questions
How many transmission zeros can a cross-coupled filter produce?
A fully canonical N-pole filter can realize up to N−2 finite zeros, since two couplings must stay in the direct through-path. A folded 6-pole with a single 1-to-6 bridge yields a symmetric pair (2 zeros); a cascaded triplet gives one asymmetric zero each and a cascaded quadruplet gives a symmetric pair each. An 8-pole canonical filter can reach 6 finite zeros. The coupling sign decides whether the zeros sit on the imaginary axis (selectivity) or the real axis (group-delay equalization).
What is the difference between a positive and negative cross coupling?
The sign sets where the zeros land in the complex frequency plane. A negative (capacitive relative to inductive main couplings) cross coupling puts a symmetric pair on the jΩ axis just outside the band, steepening the skirts. A same-magnitude positive coupling moves that pair onto the real axis as a complex-conjugate pair, keeping a Chebyshev-like amplitude but flattening in-band group delay. Physically the sign comes from whether the bridge is dominantly electric (probe, capacitive iris) or magnetic (loop, inductive aperture).
Why use a cross-coupled topology instead of adding more resonators?
Each finite zero adds close-in rejection that would otherwise need extra resonators. A 4-pole quasi-elliptic filter with one cross-coupled quadruplet matches the near-band rejection of a 6-pole Chebyshev, saving two resonators and the associated insertion loss, size, and mass. That trade matters for satellite multiplexers and base-station front ends. The price is design and tuning complexity: the bridge must hit the right magnitude and sign and it interacts with the main couplings, so synthesis is done through a coupling matrix.