The quietest number in a Stirling power system: its voltage

What the work claims

This is a design-trade study, an analytical comparison, not a test campaign. Its claim: for a pair of free-piston Stirling convertors totaling 140 W, the alternator voltage level can be chosen analytically with real space-qualified components, and under the study's assumptions the voltage band between 45 and 63 volts rms gives good SWaP (size, weight, and power) performance. Supporting claims: total power loss varies modestly over a broad good range of 25 to 110 volts rms; the H-bridge loss falls as voltage rises until the MOSFET's rated voltage is reached, while the buck converter loss rises with voltage; and the smallest, lightest DC-link capacitor occurs near 60 volts rms. The authors state plainly that different Stirling technologies have historically used different alternator voltages and that no quantitative analysis of that choice existed before this study.¹

How it works

The analysis starts from a fixed point in physics: a Stirling alternator winding with constant copper area has constant conduction loss no matter how the voltage-to-current ratio is split. The study derives this directly: with winding area held fixed, alternator resistance scales with turns squared and loss depends only on flux swing, winding geometry, and resistivity, so P_loss is invariant with voltage. That single assumption converts a multi-dimensional redesign problem into a one-parameter sweep, and everything interesting then happens in the electronics around the alternator.¹

The controller architecture is the High Voltage Convertor Controller lineage: the alternator's own inductance serves as the boost inductor, an H-bridge running near 30 kHz handles power-factor-corrected rectification and piston control, and a synchronous buck converter, also near 30 kHz, produces the spacecraft DC bus. The H-bridge DC rail sits 30 V above the alternator peak, so the two heritage designs bracket the study: the low-voltage controller at 17 V peak alternator voltage (45 V DC rail) and the high-voltage controller at 85 V peak (105 V DC rail). Harnessing resistance is held at 0.1 ohms per convertor, and housekeeping plus miscellaneous loads consume about 9 W, constant across the sweep.¹

The parts are real and qualified: Infineon IRHNS9A7064 MOSFETs screened to MIL-PRF-19500, and MIL-PRF-49470 ceramic capacitors, derated 0.75 and 0.6 respectively per NASA's EEE-INST-002 parts instruction. Their physics sets the trade. The MOSFET's on-resistance rises with rated voltage while its output capacitance falls, so the H-bridge conduction and switching penalty worsens as voltage climbs, but the current being switched shrinks, which is why H-bridge loss falls with voltage until the device rating is hit. The buck converter runs the other way: its loss rises with voltage, from switching loss at the high rail plus conduction loss through the now higher-resistance FETs. The filter inductor is assumed to hold a constant 1.5 W loss, and reverse-recovery and other switching-loss terms are explicitly excluded.¹

The mass side is dominated by one passive. Ceramic capacitor energy density and specific energy scale with rated voltage, so a higher-voltage DC bus needs far less capacitance for the same 10 V ripple budget at the roughly 100 Hz convertor frequency (C = I_C over 2f-delta-V). Netting the loss curves against the capacitor's shrinking volume and weight is what produces the 45 to 63 V rms SWaP band, with the lightest capacitor alone near 60 V rms.¹

The strongest case

The methodology is exactly how an early design trade should be done. Every component is a real, orderable, space-qualified part with a datasheet, not an ideal switch; the derating rules are NASA's own EEE-INST-002; the alternator physics is derived rather than assumed numerically; and the excluded effects (reverse recovery, other switching-loss terms, capacitor loss) are named rather than silently dropped. The study also answers a genuine institutional question: Stirling convertor programs have accumulated at least two voltage traditions (the 17 V peak and 85 V peak heritage controllers), engineers have been choosing between them by inheritance rather than analysis, and the paper demonstrates the choice is quantifiable with commodity analysis tools. The conclusion's modesty is a strength: it recommends a band, not a point, and lists the analyses needed to sharpen it.¹

Where a skeptic should push

The most load-bearing assumption is the constant-winding-area alternator, because it is what makes alternator loss voltage-independent and confines the whole study to the electronics. In a real program, changing alternator voltage means redesigning the winding, and a winding optimized for higher voltage at fixed copper could trade loss against mass differently than the paper's invariant allows. Second, the buck converter's excluded losses cut against the result: reverse-recovery and gate-drive losses grow with voltage and switching frequency, and the authors note their simplified switching model explicitly. Third, and most consequential for scaling: the EMI filter is left to future work even though the authors state it extends switch and winding life by imposing sinusoidal waveforms and is essential at higher power or longer cables. Its mass and loss will land hardest at exactly the high-voltage, high-power end where this study is most likely to be quoted. Fourth, this is a single-MOSFET, single-capacitor-family analysis at the 140 W scale with reliability and cooling mass unanalyzed. The 45 to 63 V band is a finding about this design point, not a law for DRPS architectures in general.¹

Why 45 to 63 volts is where DRPS controllers land

Implication: the non-obvious lesson is where the mass actually lives. Engineers intuit that semiconductor choice drives power-electronics mass; this study shows that across the voltage sweep, the weight and volume that move are those of the H-bridge DC-link capacitor, a passive energy buffer whose energy density happens to scale with voltage. Architecture decisions for power processing in space are therefore passive-energy decisions first: bus voltage ripple budget, buffer energy, and capacitor derating set the mass floor that MOSFETs only modulate. Anyone writing a converter specification should size the passive budget before arguing about switches.¹

Threat: the 45 to 63 V result will be misquoted. It holds for a 140 W convertor pair with 0.1 ohm harness runs, a 100 Hz ripple source, and short cables. A kilowatt-class fission surface power system or a distributed rover bus has longer harnesses, where I-squared-R losses punish low voltage, and the authors themselves flag that the EMI filter required at higher power or longer cables is absent from this model. Carried into a kW-class design unchanged, the low-voltage band would inflate harness and current-related losses, while a blanket jump to the highest survivable voltage invites the switching-loss and voltage-stress penalties the buck-converter curve already shows. The study optimizes a node; a power system is a network.¹

Opportunity: the mechanism with the widest application is the constant-winding-area analytic move itself: hold the machine's loss invariant, sweep the electronics, and let qualified part physics (on-resistance versus rating, capacitor energy density versus rating) draw the Pareto front. That template costs nothing but engineering time and can be rerun for any converter, voltage class, or parts baseline before hardware money is committed, which is precisely the cheap kind of de-risking a thin DRPS budget needs. The paper's own future-work list (detailed switching-loss modeling, EMI filter trades, cooling mass, reliability) doubles as a pre-hardware checklist any dynamic-conversion program can adopt.¹

The bottom line

Distinguishing demonstrated from derived: nothing here is hardware; the study is a simulation and analytical derivation over datasheet part models, and its own conclusions label it preliminary. Within that boundary it is solid: the alternator-loss invariance derivation is exact under its assumption, the component models come from named qualified parts with NASA derating rules, and the qualitative shape of the result (loss flat over a broad band, capacitor mass favoring mid-to-high voltage, buck converter penalizing high voltage) follows directly from device physics. Confidence is high that 45 to 63 V rms is the right band for this specific 140 W design point; moderate that the band survives addition of EMI filter, cooling, and reliability analyses; and low that it transfers to higher-power surface systems without re-derivation. What would confirm it: a breadboard controller swept across voltage with measured efficiency and mass. What would break it: an EMI-filter and cooling-mass study showing the high-voltage end's system-level penalty is much smaller than the buck-converter curve implies.