Published Nov 20, 2026 · Last updated Nov 20, 2026 · 5 min · IndieRF

Two Combined 10 W PAs: 21.2 W Out, 21.5 W of Heat

42.64 W of DC, 9.91 W of heat per PA.

Two HMC1114 PAs combined through TTM 3 dB hybrids put 21.2 W into the antenna at 3.5 GHz from 42.64 W of DC, so the board sheds 21.46 W: 9.91 W per PA, 1.26 W in the combiner. One PA plus 3 dB in a spreadsheet says 22.44 W.

The lineup and where each number comes from

An exciter drives an ADL5321, a quadrature hybrid splits its output, each arm is an HMC1114, and a second hybrid recombines them. The quadrature split and the quadrature combine cancel, so the two arms arrive in phase and add as voltages.

HMC1114 Table 2 (3.2 to 3.8 GHz, 28 V, IDQ 150 mA) prints PSAT 40.5 dBm at PIN 16 dBm and 53% PAE. It prints the 150 mA quiescent current, not the current under drive, so the link takes the current the printed PAE implies: (11.22 − 0.04)/0.53 = 21.1 W, or 0.753 A at 28 V. If the real PAE at this point is 50%, each PA draws 0.799 A and sheds 11.18 W instead of 9.91; at 56%, 8.78 W. The same table prints a 25 dB power gain beside PSAT, half a dB more than 40.5 − 16. The link uses the two printed powers; the gain row would put 23.77 W at the antenna port.

The XMC0204F1-03G guarantees 0.25 dB maximum loss over the split from 2700 to 3700 MHz and handles 40 W CW at 85 °C. The ADL5321 at 3.5 GHz and 5 V is 12.0 dB of gain, 25.7 dBm P1dB and 90 mA, class A, so its current holds under drive. It has to deliver 16 + 3.01 + 0.25 = 19.26 dBm, 6.4 dB under its P1dB, which takes 7.302 dBm from the exciter on Cascade's compression curve.

Per block, with the solved RF powers in watts:

heat=PDC+Pin−Pout,PAE=Pout−PinPDC\mathrm{heat} = P_{DC} + P_{in} - P_{out}, \qquad \mathrm{PAE} = \frac{P_{out} - P_{in}}{P_{DC}}

Each PA reads 21.1 W of DC, 9.91 W of heat and 53.0% PAE. The combining hybrid takes 22.44 W and passes 21.19, so it keeps 1.26 W. The splitter keeps 4.7 mW and the driver 0.37 W. The project line counts each of the seven blocks once: 0.45 W of driver plus two PAs is 42.64 W, and 42.64 W in less 21.19 W delivered is 21.46 W of heat. System PAE, antenna-port power less exciter power over all the DC, is 49.7%.

What a spreadsheet does with two arms

A cascade sheet is one row per stage, so a second arm has nowhere to go. It either models one PA and adds 3 dB, or lists both PAs as rows in series and doubles the gain. The first is the common one, and it reads 22.44 W at the antenna: no splitter, no combiner loss, no combiner heat. Its DC column counts the PA once, 21.55 W, and its system PAE comes out at 104%, which is the giveaway. Fix that by copying the whole chain per arm and the driver is counted twice: 43.09 W.

Cascade carries the split as two paths that meet at the combiner, adds the arms coherently at the merge, and sums DC and heat over the union of the paths. Each block counts once however many paths it sits on.

Rule of thumb: combining loss comes straight off system efficiency. System PAE is about the PA's PAE times 10^(−L/10), so 0.25 dB costs 5.6% of the output and takes 53% to 50.0%. It breaks when the driver's DC is a real share of the total (here it is 1.1%), and when the arms stop matching.

The worst mismatch is an arm that turns off. The live PA's 11.22 W splits in the combining hybrid: 5.3 W reaches the antenna and 5.3 W goes to the isolated port, so the termination there needs a rating above 5 W, though it only sees power when an arm fails.

One, two and four PAs, and a lossier combiner

Every row is Cascade's budget on the linked lineup, checked against the sums above worked by hand. Each PA sits at its printed point, so heat per PA stays 9.91 W.

LineupAntenna port (W)Antenna port (dBm)Combining loss (W)DC (W)Heat (W)System PAE (%)
One HMC1114, no hybrids11.2240.500.0021.5510.3352.1
Two, XMC0204F1-03G pair (0.25 dB)21.1943.261.2642.6421.4649.7
Two, combiner 0.6 dB19.5442.912.9042.6423.1045.8
Two, combiner 1.0 dB17.8342.514.6242.6424.8241.8
Four, two hybrid levels40.0046.024.8884.8344.8447.1

A 1 dB combiner, typical of a lossy printed Wilkinson, keeps 4.62 W in a part with a far smaller heat path than the PA's paddle. Four PAs reach 40.00 W, but the last hybrid takes 42.37 W in, over its 40 W rating, so a four-way build needs a bigger final combiner.

Cascade holds each block's DC at V·I at every drive level. Back the exciter off and each PA still draws 21.1 W in the budget, while a real class AB HMC1114 draws less, so heat at back-off reads high. The rows above all sit at the printed saturated point, where the convention holds.

References

  • Analog Devices, HMC1114 10 W, GaN Power Amplifier, 2.7 GHz to 3.8 GHz, data sheet Rev. A, page 3, Table 2 (3.2 GHz to 3.8 GHz, TA = 25 °C, VDD = 28 V, IDQ = 150 mA).
  • Analog Devices, ADL5321 2.3 GHz to 4.0 GHz ¼ Watt RF Driver Amplifier, data sheet Rev. D, page 3, Table 1, frequency 3.5 GHz, 5 V.
  • TTM Technologies, XMC0204F1-03G Hybrid Coupler 3 dB, 90°, data sheet Rev G, Detailed Electrical Specifications, 2700 to 3700 MHz.

Questions? Contact

Discussion

No comments yet. Start the thread with a measurement, a correction, or a worked example.