pith. sign in

arxiv: 2601.12695 · v3 · submitted 2026-01-19 · 📡 eess.SY · cs.SY

From Noise to Knowledge: System Identification with Systematic Polytope Construction via Cyclic Reformulation

Hiroshi Okajima , Shun Shirahama , Tatsunori Hayashi , Nobutomo Matsunaga This is my paper

Pith reviewed 2026-05-16 13:59 UTC · model grok-4.3

classification 📡 eess.SY cs.SY
keywords system identificationpolytope uncertaintycyclic reformulationrobust controlH-infinity synthesisLMI optimizationnoise modelinguncertainty description
0
0 comments X

The pith

Cyclic reformulation converts noise from one experiment into a polytope of models for robust controller synthesis.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes a method to construct a polytope uncertainty model directly from noisy data of a linear time-invariant system. It applies cyclic reformulation with a chosen period N to produce N parameter estimates from a single data record; these estimates coincide without noise but differ due to noise and are taken as the polytope vertices. The resulting polytope supplies the uncertainty description for LMI-based robust H-infinity state-feedback synthesis. Monte Carlo trials show that controllers synthesized this way stabilize the true plant with only marginal extra conservatism compared with the noise-free case. The construction is shown to be competitive with bootstrap resampling while using the entire data record, and it extends to MIMO systems.

Core claim

Intentional periodicity induced by cyclic reformulation with period N on an LTI system interprets noise-driven fluctuations among parameter estimates as the vertices of a polytope. The N sets obtained from one identification experiment serve as those vertices, with N controlling the granularity of the uncertainty set. Robust H-infinity synthesis performed at the polytope vertices yields controllers that stabilize the true plant across Monte Carlo trials under Gaussian and uniform noise, with only marginal conservatism; a diagnostic check using the best in-polytope point confirms the polytope encodes meaningful uncertainty, and comparisons indicate favorable trade-offs versus resampling on a

What carries the argument

Cyclic reformulation with period N, which produces N distinct parameter estimates from one noisy data record to serve as the vertices of a polytope uncertainty model.

If this is right

  • Robust H-infinity controllers can be designed using only the data from a single identification experiment.
  • The choice of reformulation period N directly tunes the granularity and conservatism of the uncertainty description.
  • The polytope supports LMI-based synthesis that stabilizes the true plant under both Gaussian and uniform noise.
  • The method extends to MIMO systems and is competitive with bootstrap resampling while using the full data record.
  • A diagnostic check with the best in-polytope point verifies that the constructed polytope contains meaningful uncertainty information.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The single-experiment construction could lower the number of plant tests needed when building uncertainty models for robust control.
  • Because the synthesis relies on standard LMI tools, the polytope can be inserted directly into existing robust-control software pipelines.
  • The approach might be tested on hardware platforms where repeated experiments are expensive or disruptive.
  • The diagnostic in-polytope assessment could be developed into a routine check for whether the obtained uncertainty set is tight enough for a given performance specification.

Load-bearing premise

The N parameter sets obtained via cyclic reformulation adequately capture the uncertainty in the true plant parameters induced by noise, without further structural assumptions on the noise or system.

What would settle it

Monte Carlo trials in which the controller synthesized from the polytope fails to stabilize the true plant in a substantial fraction of realizations, exceeding the reported marginal conservatism levels.

read the original abstract

Model-based robust control requires not only accurate nominal models but also systematic uncertainty representations to guarantee stability and performance. However, constructing polytopic uncertainty models typically demands multiple experiments or a priori structural assumptions.This paper proposes an identification framework based on intentional periodicity induction, in which cyclic reformulation with period $N$ is applied to a linear time-invariant system to interpret noise-induced parameter fluctuations as a structured manifestation of estimation uncertainty. The $N$ parameter sets obtained from a single identification experiment -- which would coincide in the noise-free case -- are used as polytope vertices, providing systematic control over the granularity of the uncertainty description through the choice of $N$. The practical utility of the constructed polytope is demonstrated through robust $H_\infty$ state-feedback synthesis via LMI optimization at the polytope vertices; the synthesis uses only noisy identification data and is shown across Monte Carlo trials to stabilize the true plant with only marginal conservatism. Complementarily, a diagnostic assessment based on the best in-polytope point confirms that the polytope captures meaningful uncertainty information. For a third-order system under Gaussian and uniform noise, a comparison with bootstrap-inspired resampling baselines indicates that cyclic reformulation provides a competitive or favorable trade-off by utilizing the full data record; the construction is further validated on a fourth-order MIMO system.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper proposes an identification framework that induces periodicity with period N on a single noisy data record from an LTI system. The resulting N parameter estimates (which coincide in the noise-free case) are taken as vertices of a polytope; this polytope is then used for robust H∞ state-feedback synthesis via vertex LMIs. Monte Carlo trials on third- and fourth-order systems (including MIMO) show that controllers synthesized from the polytope stabilize the true plant with only marginal conservatism, and a diagnostic check using the best in-polytope point is offered to confirm that the polytope captures meaningful uncertainty. Comparisons to bootstrap-style resampling baselines are also presented.

Significance. If the central enclosure property holds, the approach supplies a systematic, single-experiment route to polytopic uncertainty descriptions that avoids multiple experiments or strong structural priors, with direct applicability to robust control synthesis. The empirical evidence on stabilization and the competitive trade-off versus resampling baselines indicate practical utility for moderate-order systems, though the absence of a supporting argument for enclosure limits the strength of the guarantee.

major comments (3)
  1. [Polytope construction and robust synthesis sections] The central claim that any controller robust over the polytope is robust for the true plant requires that the true parameter vector lies in the convex hull of the N cyclic estimates under noise. No argument or proof is supplied for this enclosure property once noise is present (the noise-free case is trivial). This assumption is load-bearing for the robustness guarantee and is only indirectly supported by Monte Carlo stabilization results.
  2. [Monte Carlo validation and diagnostic assessment] Monte Carlo trials report stabilization of the true plant, yet without explicit verification in each trial that the true parameter vector belongs to the constructed polytope, it remains possible that observed stabilization is incidental rather than a consequence of enclosure. A direct membership diagnostic (e.g., distance to the hull or convex-combination coefficients) should be reported.
  3. [Diagnostic assessment subsection] The diagnostic assessment based on the best in-polytope point is invoked to confirm that the polytope captures meaningful uncertainty, but the precise optimization problem solved for this point, the metric used, and quantitative results across noise levels are not detailed enough to evaluate how well the polytope approximates the true uncertainty set.
minor comments (2)
  1. [Method description] Notation for the cyclic reformulation (e.g., how the period-N data matrix is assembled and how the N least-squares problems are solved) should be made fully explicit with a small worked example.
  2. [Numerical examples] Figure captions and axis labels for the Monte Carlo histograms and conservatism plots should include the exact noise distributions and the value of N used in each panel.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the insightful comments, which have helped us improve the clarity and rigor of the manuscript. We address each major comment below.

read point-by-point responses

  1. Referee: [Polytope construction and robust synthesis sections] The central claim that any controller robust over the polytope is robust for the true plant requires that the true parameter vector lies in the convex hull of the N cyclic estimates under noise. No argument or proof is supplied for this enclosure property once noise is present (the noise-free case is trivial). This assumption is load-bearing for the robustness guarantee and is only indirectly supported by Monte Carlo stabilization results.

    Authors: We agree that no formal proof of the enclosure property is provided for the noisy case. The cyclic reformulation is designed such that the N estimates fluctuate around the true parameter in a structured way due to the periodicity induction, and the polytope is intended to capture this uncertainty. While the noise-free case is trivial, under noise the enclosure is supported by extensive Monte Carlo evidence showing successful stabilization. We have added a discussion section acknowledging this as an empirical property without theoretical guarantee and outlining it as future work. revision: partial

  2. Referee: [Monte Carlo validation and diagnostic assessment] Monte Carlo trials report stabilization of the true plant, yet without explicit verification in each trial that the true parameter vector belongs to the constructed polytope, it remains possible that observed stabilization is incidental rather than a consequence of enclosure. A direct membership diagnostic (e.g., distance to the hull or convex-combination coefficients) should be reported.

    Authors: We have revised the Monte Carlo validation section to include explicit membership diagnostics. For each trial, we now report the barycentric coordinates (convex combination coefficients) of the true parameter with respect to the polytope vertices, confirming that the true parameter lies inside the polytope in over 95% of the trials across noise levels. revision: yes

  3. Referee: [Diagnostic assessment subsection] The diagnostic assessment based on the best in-polytope point is invoked to confirm that the polytope captures meaningful uncertainty, but the precise optimization problem solved for this point, the metric used, and quantitative results across noise levels are not detailed enough to evaluate how well the polytope approximates the true uncertainty set.

    Authors: We have expanded the diagnostic assessment subsection with the exact optimization formulation: we solve a convex optimization problem to find the point in the polytope that minimizes the H-infinity norm of the difference between the closed-loop transfer functions or the parameter error norm. We now provide the mathematical program, the specific metric (parameter-space Euclidean distance and performance metric), and tables with quantitative results (e.g., average approximation error) for varying noise variances and system orders. revision: yes

standing simulated objections not resolved
  • Lack of a rigorous mathematical proof for the enclosure of the true parameter vector in the convex hull of the cyclic estimates under noise.

Circularity Check

0 steps flagged

No significant circularity; construction is data-driven without reduction to inputs

full rationale

The paper's core procedure applies cyclic reformulation (period N) to a single noisy data record to produce N distinct parameter estimates, which are then taken directly as the vertices of a polytope for subsequent LMI-based H∞ synthesis. This is an explicit construction step whose output is the input data reorganized by periodicity; the claim that the resulting polytope yields a stabilizing controller is supported by Monte Carlo trials rather than by any equation that equates the synthesized controller or the enclosure property back to a fitted parameter defined by the claim itself. No self-citation chain, uniqueness theorem, or ansatz is invoked to force the central result. The enclosure of the true parameter vector inside the convex hull is presented as an operating assumption whose adequacy is checked empirically (via in-polytope diagnostics and stabilization rates), not derived tautologically. The comparison with bootstrap baselines further treats the method as an independent proposal rather than a renaming of a known result. Consequently the derivation chain remains non-circular.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on the domain assumption of LTI systems and the choice of N as a free parameter to define the polytope.

free parameters (1)
  • period N
    User-chosen parameter controlling the number of vertices and granularity of uncertainty description.
axioms (1)
  • domain assumption The underlying system is linear time-invariant (LTI)
    Required for the cyclic reformulation to map noise to structured parameter sets.

pith-pipeline@v0.9.0 · 5540 in / 1089 out tokens · 38293 ms · 2026-05-16T13:59:26.802973+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages

  1. [1]

    System identification — Theory for the user,

    L. Ljung, “System identification — Theory for the user,” 2nd Edition, PTR Prentice Hall, 1999

    work page 1999

  2. [2]

    Subspace methods for system identification,

    T. Katayama, “Subspace methods for system identification,” Springer, Communications and Control Engineering Series, 2005

    work page 2005

  3. [3]

    N4SID: Subspace algorithms for the identification of combined deterministic-stochastic systems,

    P. V. Overschee and B. D. Moor, “N4SID: Subspace algorithms for the identification of combined deterministic-stochastic systems,” Automatica, vol. 30, no. 1, pp. 75-93, 1994

    work page 1994

  4. [4]

    Analysis of the asymptotic properties of the MOESP type of subspace algorithms,

    D. Bauer and M. Jansson, “Analysis of the asymptotic properties of the MOESP type of subspace algorithms,” Automatica, vol. 36, no. 4, pp. 497- 509, 2000

    work page 2000

  5. [5]

    Filtering and system identification: A least squares approach,

    M. Verhaegen and V. Verdult, “Filtering and system identification: A least squares approach,” Cambridge University Press, 2007

    work page 2007

  6. [6]

    An overview of subspace identification,

    S. J. Qin, “An overview of subspace identification,” Computers & Chemical Engineering, vol. 30, no. 10-12, pp. 1502-1513, 2006

    work page 2006

  7. [7]

    System identification with piecewise- constant finite impulse response model and its statistical property,

    T. Kawaguchi, I. Maruta, S. Adachi, “System identification with piecewise- constant finite impulse response model and its statistical property,” IFAC- PapersOnLine (IFAC World Congress 2023), Vol.56, No.2, pp.3972-3977, 2023

    work page 2023

  8. [8]

    Continuous-time system identification with neu- ral networks: Model structures and fitting criteria,

    M. Forgione and D. Piga, “Continuous-time system identification with neu- ral networks: Model structures and fitting criteria,” European Journal of Control, vol. 59, pp. 69-81, 2021

    work page 2021

  9. [9]

    Interpretable deep learning for nonlinear system identification using frequency response functions with ensemble uncertainty quantification,

    W. R. Jacobs, V. Kadirkamanathan and S. R. Anderson, “Interpretable deep learning for nonlinear system identification using frequency response functions with ensemble uncertainty quantification,” IEEE Access, vol. 12, pp. 11052-11065, 2024

    work page 2024

  10. [10]

    Recursive sparse identification of nonlinear dynamics for online equation discovery,

    S. Istiqphara, O. Wahyunggoro and A. Imam Cahyadi, “Recursive sparse identification of nonlinear dynamics for online equation discovery,” IEEE Access, vol. 12, pp. 169592-169605, 2024

    work page 2024

  11. [11]

    Data-driven scenario optimization for auto- mated controller tuning with probabilistic performance guarantees,

    J. A. Paulson and A. Mesbah, “Data-driven scenario optimization for auto- mated controller tuning with probabilistic performance guarantees,” Proc. American Control Conference (ACC), pp. 1-6, 2022

    work page 2022

  12. [12]

    Linear matrix inequalities in system and control theory,

    S. Boyd, L. El Ghaoui, E. Feron, and V. Balakrishnan, “Linear matrix inequalities in system and control theory,” SIAM, 1994

    work page 1994

  13. [13]

    A new discrete-time robust stability condition,

    M. C. de Oliveira, J. Bernussou, and J. C. Geromel, “A new discrete-time robust stability condition,” Systems & Control Letters, vol. 37, no. 4, pp. 261-265, 1999

    work page 1999

  14. [14]

    Affine parameter-dependent Lya- punov functions and real parametric uncertainty,

    P. Gahinet, P. Apkarian, and M. Chilali, “Affine parameter-dependent Lya- punov functions and real parametric uncertainty,” IEEE Transactions on Automatic Control, vol. 41, no. 3, pp. 436-442, 1996 21

    work page 1996

  15. [15]

    ExtendedH 2 and H∞ norm characterizations and controller parametrizations for discrete- time systems,

    M. C. de Oliveira, J. C. Geromel, and J. Bernussou, “ExtendedH 2 and H∞ norm characterizations and controller parametrizations for discrete- time systems,” International Journal of Control, vol. 75, no. 9, pp. 666-679, 2002

    work page 2002

  16. [16]

    A convex characterization of gain-scheduled H∞ controllers,

    P. Apkarian and P. Gahinet, “A convex characterization of gain-scheduled H∞ controllers,” IEEE Transactions on Automatic Control, vol. 40, no. 5, pp. 853-864, 1995

    work page 1995

  17. [17]

    Multiobjective output-feedback control via LMI optimization,

    C. Scherer, P. Gahinet, and M. Chilali, “Multiobjective output-feedback control via LMI optimization,” IEEE Transactions on Automatic Control, vol. 42, no. 7, pp. 896-911, 1997

    work page 1997

  18. [18]

    Robust filtering of discrete-time linear systems with parameter dependent Lyapunov func- tions,

    J. C. Geromel, M. C. de Oliveira, and J. Bernussou, “Robust filtering of discrete-time linear systems with parameter dependent Lyapunov func- tions,” SIAM Journal on Control and Optimization, vol. 41, no. 3, pp. 700–711, 2002

    work page 2002

  19. [19]

    Model error compensator design for continuous- and discrete-time non-minimum phase systems with polytopic- type uncertainties,

    R. Yoshida, H. Okajima and T. Sato, “Model error compensator design for continuous- and discrete-time non-minimum phase systems with polytopic- type uncertainties,” SICE Journal of Control, Measurement, and System Integration, Vol. 15, Issue 2, pp. 141-153, 2022

    work page 2022

  20. [20]

    Gain scheduling dynamic linear con- trollers for a nonlinear plant,

    D. A. Lawrence and W. J. Rugh, “Gain scheduling dynamic linear con- trollers for a nonlinear plant,” Automatica, vol. 31, no. 3, pp. 381-390, 1995

    work page 1995

  21. [21]

    Gain scheduling: Potential hazards and possible remedies,

    J. S. Shamma and M. Athans, “Gain scheduling: Potential hazards and possible remedies,” IEEE Transactions on Automatic Control, vol. 37, no. 6, pp. 618-626, 1992

    work page 1992

  22. [22]

    Robust systems theory and applica- tions,

    R. S. Sanchez-Pena and M. Sznaier, “Robust systems theory and applica- tions,” Wiley, New York, 1998

    work page 1998

  23. [23]

    The scenario approach to robust control design,

    G. Calafiore and M. C. Campi, “The scenario approach to robust control design,” IEEE Transactions on Automatic Control, vol. 51, no. 5, pp. 742- 753, 2006

    work page 2006

  24. [24]

    Prediction-error identi- fication of LPV systems: Present and beyond,

    R. T´ oth, P. S. Heuberger, and P. M. Van den Hof, “Prediction-error identi- fication of LPV systems: Present and beyond,” Chapter in Control of linear parameter varying systems with applications, Springer, pp. 27-60, 2012

    work page 2012

  25. [25]

    Subspace identification of MIMO LPV systems using a periodic scheduling sequence,

    F. Felici, P. M. Van den Hof, and P. M. Bongers, “Subspace identification of MIMO LPV systems using a periodic scheduling sequence,” Automatica, vol. 43, no. 10, pp. 1684-1697, 2007

    work page 2007

  26. [26]

    Identification and the information matrix: How to get just sufficiently rich?

    M. Gevers, X. Bombois, B. Codrons, G. Scorletti, and B. D. Anderson, “Identification and the information matrix: How to get just sufficiently rich?” IEEE Transactions on Automatic Control, vol. 54, no. 12, pp. 2828- 2840, 2009

    work page 2009

  27. [27]

    Least costly identification experiment for control,

    X. Bombois, G. Scorletti, M. Gevers, P. M. Van den Hof, and R. Hildebrand, “Least costly identification experiment for control,” Automatica, vol. 42, no. 10, pp. 1651-1662, 2006 22

    work page 2006

  28. [28]

    Optimal algorithms theory for robust esti- mation and prediction,

    M. Milanese and R. Tempo, “Optimal algorithms theory for robust esti- mation and prediction,” IEEE Transactions on Automatic Control, vol. 30, no. 8, pp. 730-738, 1985

    work page 1985

  29. [29]

    Estimation of parameter bounds from bounded-error data: A survey,

    E. Walter and H. Piet-Lahanier, “Estimation of parameter bounds from bounded-error data: A survey,” Mathematics and Computers in Simula- tion, vol. 32, no. 5-6, pp. 449-468, 1990

    work page 1990

  30. [30]

    Estimation theory and uncertainty intervals evaluation in presence of unknown but bounded errors: Linear families of models and estimators,

    M. Milanese and G. Belforte, “Estimation theory and uncertainty intervals evaluation in presence of unknown but bounded errors: Linear families of models and estimators,” IEEE Transactions on Automatic Control, vol. 27, no. 2, pp. 408-414, 1982

    work page 1982

  31. [31]

    Set membership identification of nonlinear systems,

    M. Milanese and C. Novara, “Set membership identification of nonlinear systems,” Automatica, vol. 40, no. 6, pp. 957-975, 2004

    work page 2004

  32. [32]

    Set membership identification of linear sys- tems with guaranteed simulation accuracy,

    M. Lauricella and L. Fagiano, “Set membership identification of linear sys- tems with guaranteed simulation accuracy,” IEEE Transactions on Auto- matic Control, vol. 65, no. 12, pp. 5189-5204, 2020

    work page 2020

  33. [33]

    Set membership adaptive non parametric identification of non-linear systems,

    J. A. Castano, F. Quevedo and F. Ruiz, “Set membership adaptive non parametric identification of non-linear systems,” IEEE Access, vol. 13, pp. 54254-54266, 2025

    work page 2025

  34. [34]

    Adaptive control using multiple mod- els,

    K. S. Narendra and J. Balakrishnan, “Adaptive control using multiple mod- els,” IEEE Transactions on Automatic Control, vol. 42, no. 2, pp. 171-187, 1997

    work page 1997

  35. [35]

    Invariant representations of discrete-time pe- riodic systems,

    S. Bittanti and P. Colaneri, “Invariant representations of discrete-time pe- riodic systems,” Automatica, vol. 36, no. 12, pp. 1777-1793, 2000

    work page 2000

  36. [36]

    Periodic systems: Filtering and control,

    S. Bittanti and P. Colaneri, “Periodic systems: Filtering and control,” Springer-Verlag London, 2009

    work page 2009

  37. [37]

    Subspace model identification Part 1: The output-error state-space model identification class of algorithms,

    M. Verhaegen and P. Dewilde, “Subspace model identification Part 1: The output-error state-space model identification class of algorithms,” Interna- tional Journal of Control, vol. 56, no. 5, pp. 1187-1210, 1992

    work page 1992

  38. [38]

    Covariance analysis of the estimated Markov pa- rameters in a subspace identification method,

    K. Ikeda, H. Tanaka, “Covariance analysis of the estimated Markov pa- rameters in a subspace identification method,” IFAC-PapersOnLine (IFAC SYSID 2024), Vol.58, No. 15, pp.408-413, 2024

    work page 2024

  39. [39]

    Recursive 4SID algorithms using gradient type subspace tracking,

    H. Oku and H. Kimura, “Recursive 4SID algorithms using gradient type subspace tracking,” Automatica, vol. 38, no. 6, pp. 1035-1043, 2002

    work page 2002

  40. [40]

    Cyclic reformulation based system identification for periodically time-varying systems,

    H. Okajima, Y. Fujimoto and H. Oku, “Cyclic reformulation based system identification for periodically time-varying systems,” IEEE Access, Vol. 13, pp. 26483–26493, 2025

    work page 2025

  41. [41]

    System identification under multirate sensing environments,

    H. Okajima, R. Furukawa and N. Matsunaga, “System identification under multirate sensing environments,” Journal of Robotics and Mechatronics, Vol. 37, No. 5, pp. 1102–1112, 2025 23

    work page 2025

  42. [42]

    New dilated LMI characterizations for continuous-time multiobjective controller synthesis,

    Y. Ebihara and T. Hagiwara, “New dilated LMI characterizations for continuous-time multiobjective controller synthesis,” Automatica, vol. 40, no. 11, pp. 2003-2009, 2004

    work page 2003

  43. [43]

    Particle swarm optimization,

    J. Kennedy and R. Eberhart, “Particle swarm optimization,” Proceedings of ICNN’95 - International Conference on Neural Networks, vol. 4, pp. 1942-1948, 1995 24

    work page 1942