Disentangling Wiener-Hammerstein Systems Using Phase-Coupled Multisines

5 concepts5 papers2 proposals

The Commutativity Blindspot in Block-Oriented Systems

A central challenge in nonlinear system identification is mapping block-oriented configurations like the Wiener-Hammerstein model. This architecture cascades a linear time-invariant input block ($L_1$), a static nonlinearity ($NL$), and a linear time-invariant output block ($L_2$).

Input ---> [ L1: Input Dynamics ] ---> [ NL: Nonlinearity ] ---> [ L2: Output Dynamics ] ---> Output

If you attempt to identify this system using standard linear approximation techniques—such as the Best Linear Approximation (BLA)—you run into a fundamental mathematical wall: linear operators commute ($L_1 \times L_2 = L_2 \times L_1$). From the perspective of the overall linear transfer function, the poles and zeros of both the input and output stages are completely blended together. You can see what the system's total dynamics are, but you cannot determine where they belong. As noted by Koen Tiels et al., breaking this commutativity symmetry is the primary bottleneck to accurately initializing these models.

Phase-Coupled Multisines as Frequency Mixers

To break this symmetry, the system's nonlinearity must be induced to treat the input dynamics differently from the output dynamics. The key lies in signal design.

Instead of exciting the system with standard random multisine signals—where every frequency component possesses an independent, random phase—the authors introduce a tailored excitation signal known as phase-coupled multisines. In a phase-coupled multisine, specifically selected pairs of frequency lines are constrained to share the exact same random phase.

When this structured signal passes through the system, the static nonlinearity acts as a frequency mixer. Because the phases of the input frequencies are locked together, their nonlinear interactions do not result in unstructured, noise-like distortions. Instead, they generate predictable, coherent harmonic combinations.

The Asymmetry of Frequency Shifting

The core intuition behind why this technique works boils down to the order of operations within the cascade.

The input signal passes through the input block ($L_1$) before it encounters the nonlinearity. Because the nonlinearity acts as a frequency mixer, it shifts the frequency spectrum of the signal it receives. Consequently, the dynamics of $L_1$ undergo a user-defined frequency translation and rotation in the complex plane.

Conversely, the output block ($L_2$) sits after the nonlinearity. It simply processes whatever mixed frequencies emerge from the nonlinear stage. Because $L_2$ does not precede the mixing operation, its underlying dynamics remain completely stationary.

As explained in Initial Estimates for Wiener-Hammerstein Dynamics Using Phase-Coupled Multisines, this creates a stark structural asymmetry:

  • Input Dynamics ($L_1$): Shifted by a known frequency offset.
  • Output Dynamics ($L_2$): Stationary.

This controlled frequency shift serves as a unique dynamic signature for the input stage, effectively separating the two cascaded blocks.

Tracking Mobility via Shifted Linear Approximations

To isolate these dynamics in practice, a "shifted Best Linear Approximation" is computed over a modified, offset frequency grid rather than a standard grid.

Because the intentional frequency shift breaks the standard complex-conjugate pairing of the input stage's poles, the resulting transfer function yields complex coefficients. By applying a modified complex-coefficient frequency-domain estimator, researchers can track the mobility of individual poles and zeros across different phase-coupling configurations:

  1. Mobile Components: Poles and zeros that translate in position are unambiguously mapped to the input stage ($L_1$).
  2. Immobile Components: Poles and zeros that remain stationary are mapped to the output stage ($L_2$).

As demonstrated by Koen Tiels et al., this parametric sorting eliminates the need for exhaustive combinatorial search algorithms during model initialization, drastically simplifying the identification process.

A Broader Principle: Exploiting Nonlinearity

This framework represents a broader paradigm shift in system analysis: treating nonlinearities not as a source of corrupting noise to be suppressed, but as a structured tool to be exploited.

A conceptual parallel can be found in nonholonomic motion planning. Richard M. Murray demonstrated in Nonholonomic Motion Planning: Steering Using Sinusoids that passing periodic inputs with integrally related frequencies through a nonlinear system can generate a net rectified motion in a constrained direction.

In both system identification and robotic steering, carefully structuring the phase and frequency relationships of sinusoidal inputs allows engineers to bypass the limitations of linear systems—whether to manipulate higher-order states or to map the hidden internal architecture of a cascaded model.

Key Foundational Work

Initial estimates for Wiener-Hammerstein models via phase-coupled multisines (2014) — contributed 6 concept elements to this direction.
Initial estimates for Wiener-Hammerstein dynamics using phase-coupled multisines (2014) — contributed 5 concept elements to this direction.
Generation of initial estimates for Wiener-Hammerstein models via basis function expansions (2014) — contributed 4 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_phase_coupled_multisine_excitation_5e30d425,
  title = {phase-coupled multisine excitation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/phase-coupled-multisine-excitation/rdr_5e30d4254311477d9d86d401f406ecbe},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_phase_coupled_multisine_excitation_5e30d425_v1,
  title = {phase-coupled multisine excitation},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/phase-coupled-multisine-excitation/rdr_5e30d4254311477d9d86d401f406ecbe/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake