parametric-oscillator-reservoir-computing

name: parametric-oscillator-reservoir-computing description: Neuromorphic reservoir computing using parametrically-driven oscillators and frequency combs. Covers oscillator-based RC, 2:1 parametric resonance regimes, and bifurcation-driven computational capability mapping. category: neuromorphic-computing

Parametric Oscillator Reservoir Computing

Overview

This methodology covers neuromorphic computing using parametrically driven oscillators as reservoir computers. The key insight is that nonlinear mode coupling and intrinsic dynamics in oscillators enable both memory and high-dimensional transformation — the two essential ingredients of reservoir computing.

Paper: "Neuromorphic Computing Based on Parametrically-Driven Oscillators and Frequency Combs" (arXiv:2604.21861, April 2026)

Trigger Words

• parametric oscillator reservoir computing, frequency comb neuromorphic
• 2:1 parametric resonance, oscillator-based RC
• bifurcation computational capability, physical reservoir computing

Core Methodology

1. Two-Mode Parametric System

The system exhibits 2:1 parametric resonance where:

• A pump drive at frequency 2ω excites subharmonic oscillations at ω
• Two coupled modes interact through nonlinear coupling
• Input signals encoded into drive amplitude

2. Dynamical Regimes

The system operates in distinct regimes, each with different computational properties:

3. Reservoir Computing Pipeline

Encoding

• Input signals → drive amplitude modulation
• Temporal encoding of time-series data

Reservoir State

• Sample temporal response: oscillator amplitudes over time
• Sample spectral response: frequency comb components

Readout

• Linear readout layer trained on reservoir states
• One-step ahead prediction of chaotic systems

4. Benchmark Tasks

Standard chaotic system prediction:

• Mackey-Glass: Delay differential equation, tunable chaos
• Rössler: 3D chaotic attractor
• Lorenz: Classic butterfly attractor

5. Bifurcation-Performance Mapping

Key finding: computational capability directly maps to bifurcation structure

• Low-error regions align with parametric resonance boundary
• Prediction error mapped over parameter space reveals optimal operating zones

Control Parameters

Four parameters systematically control accessible dynamical regimes:

1. Input modulation depth: Controls signal injection strength
2. Detuning from frequency matching: Shifts operating point on bifurcation diagram
3. Damping ratio: Controls oscillation decay, affects memory depth
4. Input data rate: Determines temporal resolution vs. computational load

Implementation Guide

Step 1: Define Oscillator Model

# Two-mode parametrically driven oscillator
# x'' + 2*β*x' + ω₀²*(1 + h*cos(2*ω*t))*x + nonlinear_terms = input

Step 2: Parameter Sweep

# Sweep across parameter space to map bifurcation diagram
params = {
    'drive_amplitude': np.linspace(...),
    'detuning': np.linspace(...),
    'damping': np.linspace(...)
}

Step 3: Reservoir State Collection

# Collect temporal + spectral states
temporal_states = oscillator_amplitudes(t)
spectral_states = fft(oscillator_response)
reservoir_state = concatenate([temporal_states, spectral_states])

Step 4: Train Readout

# Linear regression on reservoir states
W = ridge_regression(reservoir_states, target_outputs)
prediction = W @ reservoir_states

Pitfalls

1. Chaotic regime trap: Frequency comb states may appear rich but lose phase coherence → poor prediction. Stay near parametric resonance boundary.
2. Insufficient memory: Too high damping → reservoir forgets input too quickly. Balance damping for task timescale.
3. Spectral vs. temporal trade-off: More spectral modes ≠ better performance. Phase coherence matters more than dimensionality.
4. Hardware sensitivity: Physical implementations sensitive to parameter drift. Need calibration routines.

Related Skills

• neuromorphic-spacecraft-pose-event-camera: Event-camera neuromorphic computing
• spikingjelly-framework: SNN training framework
• parametrically-driven-oscillator-neuromorphic: Related oscillator RC work