Most artificial neurons compute an instantaneous output and forget it. SPNM instead gives a computational unit a state that evolves continuously in time — accumulating evidence, decaying gracefully, and resonating with rhythm rather than merely magnitude. Developed across brain-computer interfaces, distributed radar sensing, and autonomous swarms.
A conventional neuron takes a weighted sum of its inputs, applies a fixed nonlinearity, and produces an output — nothing persists between one evaluation and the next except what training has fixed into its weights. SPNM units instead maintain an internal state driven jointly by a signal's amplitude and its rhythm, so the same input produces a different response depending on what came before it.
SPNM does not name a single equation. Each domain demanded a different mathematical realisation of the same underlying idea — continuous state in place of instantaneous activation.
An instantaneous spin-energy readout suited to classification from pre-processed features — amplitude and frequency arriving as separate, slowly varying quantities, as in EEG-derived brain–computer interfaces.
Explicit amplitude, frequency, and phase state evolving under resonance-alignment learning — built to track a signal's evolving phase, not merely sample it, for micro-Doppler radar perception.
Momentary resonance separated from accumulated confidence, shared between neighbouring nodes — enabling cooperative, multi-node sensing where a single channel cannot resolve occlusion alone.
A bounded nonlinearity guaranteeing stability across hundreds of simultaneously interacting agents — the substrate for decentralised swarm coordination without any central controller.
Ordered by current focus. Distributed radar sensing and electronic-warfare resilience represent the most actively developed line of work.
Self-calibrating multi-node radar meshes for micro-UAV detection, using laser cross-locking for geometric coherence and neurodynamic confidence to weight sensor trust under active jamming and clutter.
Hardware-oriented simulationResonance-based classification of motor-imagery EEG, benchmarked against ANN, SNN, and QNN baselines on real BCI Competition IV data.
Real dataset evaluationDecentralised nanorobotic swarm coordination for targeted biomedical delivery, evaluated via Monte Carlo simulation against particle-swarm and spiking-neural baselines.
Software simulationInference-time inhibitory mechanisms that keep learned "do-not-do" constraints stable under retraining, independent of the task gradient that would otherwise overwrite them.
Synthetic benchmarkTwo interactive simulations of the equations behind SPNM. Each runs in your browser, on its own page.
One simulated unit driven by a noisy rhythm. Change the growth rate, saturation, noise and frequency offset, and compare it with a linear resonator through guided examples: dropout, amplitude regulation, resonance selectivity.
Simulation, not a measurementFour units share their accumulated evidence. Block one node, change the cooperation strength, and see what the mesh gains — and the false-alarm risk it takes on.
Simulation, not a measurementOpen to research partnerships and applied collaboration with industry and defence-technology organisations working on distributed sensing, radar perception, and autonomous systems.
Mobile multi-radar systems lose geometric precision the moment nodes move, and transmitting more to compensate increases exposure in contested environments.
Active laser self-calibration keeps node geometry coherent; SPNM-derived confidence weights sensor trust and drives transmission scheduling under threat.
Evaluated through hardware-oriented simulation (FPGA resource and latency estimates) and, for the BCI application, a real external benchmark dataset. Physical hardware-in-the-loop testing is the next step.
Figures reported in the linked publications distinguish explicitly between results from real datasets, controlled simulation, and hardware-oriented resource-utilisation estimates.