Computation by Coherence: Oscillator-Based Analog Architectures Using Attractor Selection and Temporal Features

This paper introduces a device-independent analog computation architecture based on networks of self-sustaining oscillatory units whose computation arises from coherence, attractor selection, and temporal stability features rather than discrete state transitions. Each unit exposes slow configuration parameters that shape its intrinsic dynamics and fast gain control that gates activity and regulates stability. When weakly coupled, these units exhibit phase locking, synchronization, and recovery behaviors that can be interpreted as computational outcomes. We formalize computation modes based on (i) stable phase-lock patterns, (ii) temporal features such as time-to-lock and persistence under perturbation, and (iii) bounded dynamic encoding suitable for reservoir-style readout. The architecture emphasizes ultra-low-power operation, fail-soft behavior, and intrinsic stability signaling. Implementation details are intentionally omitted to preserve non-enabling disclosure while establishing conceptual priority.