Page 7 · Can one local exploratory pulse escape a worse ripple basin?

N=3 asymmetric 400kHz VRM · actual native switched ngspice · ten independent plants · 12V → nominal 1V / 40A

11.90 ms
Native scientific result loading…
Ten real moving phase points · two modeled ripple valleys
Five initially in lower basinFive initially in upper basinWhite rings = Fourier-model sampled minima

Background is the independent Fourier model, not a native sampled map. Ball positions and complete paths are reconstructed from 240 pre-existing plus newly simulated native PWM gate edges. No target/heatmap/peer phases enter local control.

Ten actual switching-cycle Vout RMS histories
Started in lower basinStarted in upper basinShaded interval = exploratory actuation
Native Vout ripple overlay · before vs after

Real saved Vout samples: representative ball 1 (lower start) and ball 6 (upper start), 40 µs windows, each mean-subtracted on a common mV axis. Waveforms are not reconstructed from the Fourier map.

What was tested, and what remains future work
1 · Settle locally12ms of previous native data: ten independent Vout-only gradient controllers populate two ripple basins.
2 · Explore oncePrivate phase 3 applies a predetermined −800Hz frequency offset from 12 to 13.5ms. The same policy is used for every ball, regardless of its basin.
3 · Return to local feedbackNo more probes after 13.5ms; measure basin residence and actual native late ripple without forcing phase coordinates.

Guard/bounce concept — future work (not executed here): monitor local Vout → temporary probe → compare settled ripple and guard headroom → reject/reverse if unacceptable. A sampled Vout guard is not an electrical hardware-safety guarantee. This pilot neither learns a global map nor demonstrates autonomous accepted/rejected trial scheduling.

Experimental boundary: three synchronous buck legs with L = 200, 230, 320 nH and deliberate split-PWM modulation on phase 2, identical nominal 400kHz natural clocks. Controller senses only its own private oscillator and a local copy of shared Vout. One preset private exploratory pulse is tested; a general basin-hopping/guard policy and multi-start generality are not proven by this finite demonstration.