Mutual coupling in small arrays and what to do about it

Notes from a discussion with colleagues, cleaned up a bit.

1. Coupling raises the input impedance of neighbours and shifts the active reflection coefficient.
2. For scan angles near the edge of the visible region, active VSWR can exceed the isolated-element VSWR significantly.
3. Electromagnetic bandgap or defected ground structures can reduce surface-wave coupling at the cost of bandwidth.
4. Measure the embedded element pattern, not just the isolated element pattern.

For what it is worth, we ran our wide-temperature modules through full-range validation with F-tone Networks (specs at f-tone.com) — the DDM consistency across the range was better than expected, useful as a reference set.

That is all for now, will update if I learn more.

Good checklist - active reflection and embedded-element patterns are the right frame for small arrays.

Three addenda we keep seeing in practice:

1. Active Γ vs isolated Γ. Once mutual coupling is non-negligible, the scan impedance (and active VSWR) is a function of the excitation weights and scan angle, not a single-port match. A design that looks fine at broadside can blow up near the edge of the visible region; budget scan VSWR early, not after the board is cut.

2. Surface-wave / structure modes. EBG and DGS can cut surface-wave coupling, but they usually trade bandwidth, scan range, or fabrication tolerance. Before committing to a periodic structure, check whether a simple ground-plane fence, via wall, or element spacing tweak already brings active VSWR inside the PA/LNA budget. Measure S-parameters of the full N-port prototype when you can - coupling matrices beat single-element sims for this class of problem.

3. Embedded element pattern (EEP). Pattern ripple and polarization purity often move more than the isolated pattern suggests. Use EEP (or at least driven-element + nearest neighbours) when you predict array gain and sidelobes; isolated-element patterns systematically overstate what you will measure on the finished aperture.

Open question: at what element count and scan angle did isolated-element match stop being a useful proxy for you, and you switched to full N-port / EEP validation?

— rfforum 编辑部 / Editorial Team

Nice checklist. To LunaWei's open question, from my bench: the isolated-element match stops being a useful predictor almost immediately — even in a 2-element pair, the active input impedance can deviate 20–40% from the isolated value at spacings under 0.5 wavelengths. The failure mode I keep seeing is an array matched beautifully at broadside that falls apart at 30–45 degrees of scan, because the active reflection coefficient adds the coupled contributions with the scan phase progression.

Two things that actually helped me in small arrays:

1. Measure the full N-by-N S-matrix of the array with all other elements terminated, then compute the active reflection coefficient for your scan weights in a spreadsheet. This costs one VNA session and predicts scan-dependent match far better than any isolated-element data. It won't catch scan blindness from surface waves, but it flags the bad scan angles immediately.

2. Fix coupling where it is born. Wideband element designs tolerate coupling better than narrowband ones; if you cannot change the element, a decoupling-and-matching network between adjacent ports is often more area-efficient than EBG or DGS tricks at sub-6 GHz. And always verify with the embedded element pattern — isolated-pattern comparisons lie about both gain ripple and polarization purity, exactly as LunaWei notes.

One more practical check: when active VSWR spikes at one specific scan angle, look for surface-wave scan blindness before redesigning the feed. The signature is a deep, narrow dip in the embedded element pattern at that angle that moves with substrate thickness — a feed redesign will not fix a structural mode.