Working Hypothesis · Open for Revision
Our best guess at what a working Hum looks like — based on the convergence patterns across 768 patents, filtered through 29 experiments. This is a living document. Every section links to the experiments that would confirm or refute it.
Pulse Generator
555 Timer + IRF540N MOSFET
40 kHz, 10–15% duty cycle, ~100ns rise time
Across the 312 patents describing pulsed excitation, the most common configuration is a timer-driven MOSFET switch at 20–100 kHz with a low duty cycle (5–20%). The sharp rising edge is critical — it generates broadband harmonics that excite the non-linear element across multiple frequencies simultaneously. A 555 timer is cheap, tunable, and well-understood. The IRF540N handles the current without significant switching losses at this frequency.
Medium — duty cycle and frequency are the main unknowns. The patent literature suggests 10–15% but this needs per-system tuning.
Non-Linear Element
Bifilar-wound coil on FT-50-43 ferrite toroid
50–100 turns bifilar (series-aiding), driven near saturation
427 of 768 patents describe a non-linear element. The most accessible is a ferrite toroid driven toward saturation — the permeability collapse at saturation creates a sharp non-linearity that generates harmonics from the incoming pulse. The bifilar winding (Tesla's 1894 configuration) adds significant inter-winding capacitance, which shifts the self-resonant frequency and creates a built-in LC resonance that the Tuner can lock onto. Mix 43 ferrite was chosen because its frequency range (20–250 MHz) is well above our operating frequency, preventing the core material itself from introducing confounding resonances.
High — bifilar geometry definitively shifts SRF vs conventional winding. The question is whether this shift matters for energy conversion.
Resonance Lock
Variable capacitor + feedback tap
Tuned to the Core's self-resonant frequency (typically 1–10 MHz for a 50-turn bifilar on FT-50-43)
389 patents describe a resonance condition — the single most common element in the dataset. The Tuner's job is to find and lock the system's natural resonant frequency. A variable capacitor in parallel with the Core's parasitic capacitance forms a tank circuit whose resonant frequency can be swept until maximum impedance is found. The feedback tap optionally provides a signal back to the Spark's timing circuit so the pulse frequency can track the resonance as conditions change (temperature, load, component aging).
Medium — we know resonance matters, but the optimal tuning strategy (fixed vs adaptive) is unresolved.
Energy Recovery
Fast-recovery diode bridge + storage capacitor
UF4007 diodes (1A, 75ns recovery), 100µF electrolytic storage cap
On every pulse cycle, the Core kicks back energy as back-EMF when the MOSFET switches off. In a conventional circuit, a flyback diode absorbs this energy as heat. The Bridge instead routes it into a storage capacitor. The key insight from the patent literature: the back-EMF spike can exceed the driving voltage by 10–50x. If even a fraction of that energy is recovered per cycle, the effective efficiency of the system improves significantly. Fast-recovery diodes are essential — standard rectifier diodes are too slow to capture the nanosecond-scale spike.
Low-Medium — recovery circuit topology is the area with the most variation across patents. Multiple approaches may work.
Output
USB 5V regulator + LED indicator
5V / 500mA output (2.5W), green LED = producing, red LED = consuming
The Load is how you know if it's working. A simple USB regulator converts whatever the Bridge captures into a usable 5V output. The LED indicator provides instant visual feedback: green means the system is outputting net energy to the load; red means it's consuming more than it produces (which is the expected state until the system is properly tuned). This isn't the final form — it's the measurement stage. If the green LED stays on with no external power input, that's the result everyone is looking for.
N/A — the load circuit itself is straightforward. The question is whether the upstream system produces enough energy to light it.
| Component | Part | Est. Cost |
|---|---|---|
| Spark | 555 timer + IRF540N MOSFET + passives | $5 |
| Core | FT-50-43 toroid + 30 AWG magnet wire | $10 |
| Tuner | Variable capacitor (air-dielectric) | $8 |
| Bridge | UF4007 diodes (x4) + 100µF capacitor | $3 |
| Load | USB 5V buck regulator + LEDs | $4 |
| Measurement | NanoVNA-H4 (if you don't have one) | $30 |
| Misc | Perfboard, wire, solder, connectors | $10 |
| Total (without NanoVNA) | ~$40 | |
| Total (with NanoVNA) | ~$70 | |
This is a falsifiable hypothesis. Here's what would make us revise each section:
Every experiment you run either strengthens or weakens a section of this diagram. Both outcomes move the project forward. The only thing that doesn't help is not building.
Three concept renders
AI-generated from the component specs above. Hover any image for the prompt used to generate it — fork it, improve it, generate your own.