Corporate Combining
How a Corporate Combining Tree Scales Solid-State Power
Solid-state devices such as GaN and GaAs transistors are inherently power-limited; a single die at 10 GHz might deliver only a few watts. To reach the tens or hundreds of watts that radar transmitters and satellite uplinks demand, designers run many identical amplifiers in parallel and sum them coherently. The corporate combiner, named for its resemblance to a corporate organizational chart, is the most common way to do this. It is a balanced binary tree: at the input a divider splits the drive signal into 2, 4, 8 or more equal paths, each path feeds its own amplifier, and a mirror-image combiner tree on the output recombines the amplified signals into one port.
Symmetry is the defining property. Every branch traverses the same number of junctions and the same physical line length, so to first order all paths share identical insertion loss and identical electrical delay. That equality is what allows the signals to add in phase at the output. Any departure from it, whether a manufacturing tolerance, a thermal gradient across the amplifier bank, or a damaged line, shows up directly as amplitude or phase imbalance and erodes combining efficiency. Practical layouts therefore keep the tree planar and geometrically regular so that nominally identical branches really are identical.
The second defining property is isolation. When the two-way junctions are Wilkinson sections, each combiner places a resistor across its two input arms. That resistor absorbs any out-of-phase or reflected power without sending it back into the amplifiers, giving better than 20 dB of port-to-port isolation. As a result a single failed transistor cannot pull down its neighbors; the array keeps running at slightly reduced power, the hallmark of graceful degradation that makes corporate trees the default for high-reliability transmitters.
Stage Count, Loss, and Combining Efficiency
Because each level of the tree doubles the number of paths, an N-way corporate combiner requires log2(N) stages, which constrains N to a power of two. Each stage contributes roughly one combiner's insertion loss, so total tree loss grows logarithmically with device count rather than linearly. This logarithmic scaling, together with the per-branch isolation, is precisely the trade space a designer weighs against radial or spatial combiners.
S = log2(N) (N must be a power of 2)
Total combiner insertion loss:
Ltree ≈ Lstage × log2(N) dB
Combining efficiency:
ηcomb = 10(−Ltree / 10) × cos2(Δφ / 2)
Output after a single device failure (of N):
Pout / Pfull = ((N − 1) / N)2
Where N = number of amplifier branches, Lstage = per-stage loss (≈0.15 dB for a Wilkinson at 10 GHz), Δφ = peak phase difference between the two branches at a combining junction. Example: N = 16, Lstage = 0.15 dB → Ltree ≈ 0.6 dB; one failed device → (15/16)2 ≈ −0.56 dB.
Corporate vs. Radial vs. Spatial Combining
| Architecture | Topology | Practical Device Count | Loss Scaling | Branch Isolation | Best Application |
|---|---|---|---|---|---|
| Corporate (binary tree) | Cascaded 2-way junctions | 2 to 32 (power of 2) | ∝ log2(N) | > 20 dB (Wilkinson) | High-reliability SSPAs, T/R modules |
| Radial / N-way | Single multi-port junction | 8 to 50+ | Nearly flat with N | Low to moderate | Compact medium-power combiners |
| Spatial (waveguide) | Free-space / oversized guide | 16 to 100+ | Lowest per device | Coupling dependent | mmWave very-high-power amplifiers |
| Chain / serial | Traveling-wave couplers | 4 to 16 | ∝ N (worst) | Directional | Wideband, distributed gain |
Frequently Asked Questions
How many combining stages does a corporate combiner need for N amplifiers?
A binary tree needs log2(N) stages, so N must be a power of two. An 8-way uses 3 stages (7 two-way junctions); a 16-way uses 4 stages (15 junctions). Tree loss is per-stage loss times log2(N): about 0.45 dB for an 8-way and 0.75 dB for a 32-way Wilkinson tree at 10 GHz with 0.15 dB per stage. The symmetric layout keeps every input the same electrical distance from the output, which is what guarantees equal amplitude and phase.
What is the combining efficiency, and how does an amplifier failure affect output?
Efficiency is summed output power divided by the arithmetic sum of branch powers, set by tree loss plus amplitude and phase imbalance. With 0.5 dB loss and tight matching it is near 89%; at 1.0 dB about 79%. A 20° phase spread costs roughly 0.13 dB, 40° about 0.5 dB. When one device fails, output drops by ((N−1)/N)2, so a single failure in a 16-way combiner costs only about 0.56 dB. That graceful degradation is the main reason corporate trees are used in high-reliability transmitters.
When should I use corporate combining instead of spatial or radial combining?
Choose a corporate tree for modest device counts (2 to 32), moderate bandwidth, and when you need branch isolation so a failed device does not load its neighbors; Wilkinson trees give 20 dB or more. The drawback is that loss grows with log2(N), so beyond about 16 to 32 devices it erodes the gain from adding amplifiers. Radial and spatial combiners merge all devices in one junction for higher efficiency in large arrays, but give up the per-branch isolation and the simple, repeatable layout of a corporate tree.