Skip to content

Frequency Pulling with Saturation

Frequency-coupling method — instantiates Decentralized Phase Locking

Draws each unit's intrinsic rate toward its neighbours' — but caps how hard it can be pulled, so no unit is yanked past a safe slew.

Most local coupling laws correct a unit's phase — where it is in its cycle. Frequency Pulling with Saturation corrects its rate — how fast the cycle runs — nudging each unit's intrinsic frequency toward the frequencies around it until the population converges on a common tempo. Its signature is the second half of the name: the pull saturates. Below a detuning threshold the correction grows with the mismatch, but past that threshold extra mismatch buys no extra pull, so a fast unit can drag a slow one only so hard and no unit is ever slewed past a rate it can safely hold. That cap is not a limitation bolted on afterwards — it is the mechanism's whole point: it trades away the ability to capture wildly-detuned outliers in exchange for never destabilising the units that do lock.

Example

A microgrid runs on ≈6 inverter-based generators with no central dispatcher, each with a slightly different natural frequency clustered around 60 Hz. Left uncoupled they drift apart and the grid can't stay stable. Each inverter measures the local line frequency and pulls its own set-point toward it — but through a saturating law: a small deviation (say a unit sitting at 60.1 Hz while the neighbourhood averages 60.0) produces a proportional correction, while a large deviation is clamped so no inverter changes its output frequency faster than a safe ramp of ≈0.1 Hz/s. The pack converges into a tight band around a shared ≈60.0 Hz, and because every correction is rate-limited, a single mis-set inverter at 60.5 Hz gets pulled in gently rather than jerking the whole grid toward itself. This droop-style sharing[1] is exactly why the cap matters more than the raw gain.

How it works

What distinguishes this method is what it adjusts and how the correction tops out:

  • Correct the rate, not the instant. The controlled quantity is the unit's frequency relative to its neighbours', so units settle to a common tempo (and a stable phase offset) rather than being snapped to a phase.
  • Saturating gain curve. The correction is roughly proportional to detuning near zero and flattens beyond a threshold — the defining shape. This bounds the slew any unit can be forced through.
  • Bounded mutual pull. Because every unit's pull is capped, no fast unit can accelerate a slow one without limit, which is what keeps the locked core stable while stragglers are drawn in.

Tuning parameters

  • Pull gain — how sharply the rate closes near zero detuning. High gain locks a tight cluster fast; too high oscillates around the common rate.
  • Saturation ceiling — the maximum rate change allowed. This is the defining dial: a high ceiling widens the capture range but risks overshoot and instability; a low ceiling is safe but leaves detuned units outside the basin.
  • Deadband — a detuning below which no correction fires, keeping units from chasing sensor noise.
  • Asymmetry — whether pulling up (speeding) and pulling down (slowing) share the same limit; asymmetric caps suit systems where one direction is cheaper or safer.

When it helps, and when it misleads

Its strength is locking a heterogeneous population — units with genuinely different natural rates — without letting the spread of rates destabilise the ones already in step. The saturation is the safety feature: it guarantees a bounded slew, so coupling can be turned up for speed without exposing units to a rate they can't physically hold.

Its failure mode is the flip side of that guarantee: saturation limits the capture range, so a unit detuned past the ceiling never locks and slips forever. The classic misuse is to answer a stubborn straggler by raising the ceiling to force it in — which discards the very slew protection the cap was providing and can tip the locked core into instability. The discipline is to treat a persistent slipper as a signal to widen coupling structurally (or add a bridge) rather than to remove the cap.

How it implements the components

  • intrinsic_frequency_and_phase_baseline — it acts directly on each unit's natural rate, pulling the intrinsic frequency toward the local field rather than resetting phase.
  • coupling_strength_and_gain_profile — the saturating gain curve — proportional near zero, flat past a cap — is the mechanism's coupling profile; setting that curve is setting the method.

It does not reset phase on discrete events (reciprocal_coupling_law — Adaptive Pulse-Coupled Update), compensate for stale readings (delay_noise_and_detuning_budget — Delay-Compensated Local Coupling), or decide when the resulting synchrony has become dangerous (harmful_lockstep_and_herd_guardrail — Anti-Herd Coupling Breaker).

Notes

Because the pull saturates, this method defines a hard capture range as a side effect: any unit detuned beyond the ceiling will phase-slip indefinitely rather than lock. That slip is information, not just a fault — it marks the edge of the basin and tells the designer where extra coupling, a bridge, or a retuned baseline is actually needed.

References

[1] Frequency-droop control shares load across independent generators by having each adjust its frequency in proportion to deviation, with limits — the power-systems name for a saturating, decentralised frequency pull. The closely related Adler description of injection locking captures the same idea in oscillators: a bounded pull with a finite locking range. ↩