{"id":"ab283837-34b6-47f1-a360-54a429ea8195","arxiv_id":"2611.17046","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"low","formal_verification":"none","parameter_count":3,"one_line_summary":"Sampling-time jitter is modeled as a memoryless bounded uncertainty block, so operator small-gain, structured H-infinity synthesis, and constant-scaled mu synthesis can certify and design jitter-robust sampled-data controllers.","lead":"This paper gives control engineers a way to design digital controllers that stay stable even when the computer samples the plant at irregular times. It recasts sampling-time jitter as a bounded uncertainty block and shows that off-the-shelf robust-control software can certify stability for every possible jitter sequence.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"The theorems are sound; the gap is in the example's certificate: δ̄ = 0.02832 is grid-computed, not proven, and the paper's only rigorous bound (Appendix (23)–(24)) is ~12% larger — I estimate the Krob product becomes ≈1.01, so the demonstrated certificates lack a valid δ̄.","rationale":"I read the paper's claims as: (i) the jittered loop is exactly an LFT of a nominal LTI system with a memoryless norm-bounded perturbation; (ii) the small-gain condition (18) certifies finite-gain stability and UES uniformly over E(ε̄); (iii) standard tools (systune, musyn with FitOrder=0) can synthesize controllers satisfying the certificate; (iv) the example demonstrates the workflow. Claims (i)–(iii) are well supported. I checked Theorem 2's bounded-real-lemma step in detail — the normalized realization, the strict norm bound, the pointwise supply-rate argument, and well-posedness all go through — and the constant-scaling robust-performance algebra commutes correctly for the time-varying block, with the exclusion of dynamic scalings correct for the time-varying case. The reader's low correctness risk is accurate for the theorems.\n\nThe weakest point is the demonstration of (iv). A proof-grade certificate requires a proven δ̄, and the example uses a grid value that is not proven; the authors acknowledge this (b143), but the only rigorous bound they offer (Appendix (23)–(24)) is so much larger that the example's margins vanish. My estimate (δ̄_appendix ≈ 0.0318 vs 0.02832) puts Krob at ≈1.01, meaning the paper's own rigorous bound likely rejects the controller it presents as certified, while Kµ survives with a thinner margin. The grid-error contribution itself is tiny (≈3e-5, since σ̄(Δ(ε)) is Lipschitz with constant ≈0.16), so this is not a failure of the grid as a numerical tool; it is the absence of a demonstrated proven δ̄, which is precisely the guarantee the paper advertises. This is fixable (Lipschitz-refined bound, shipped code, or a resynthesized example) and does not touch Theorems 1–2. The reader's CONDITIONAL verdict already rests on the validity of the jitter envelope and the example certificates; my analysis sharpens the second condition without changing the verdict, so I recommend UNCHANGED.","tokens_in":13257,"tokens_out":42220,"duration_ms":516347,"concrete_test":"Recompute the example's certificates with a proven δ̄: (1) evaluate the Appendix closed-form bound (23)–(24) for the example plant (A = [0 1; −1 −0.2], B = [0;1], t_s = 0.1, ε̄ = 0.2) and compare δ̄_appendix against the grid value 0.02832; (2) with δ̄_appendix re-evaluate Krob's product δ̄‖N_11‖_H∞ and Kµ's scaled bound (20) using the reported controllers and fitted scalings. If the Krob product exceeds 1 (I estimate ≈1.01) or the Kµ bound exceeds 1, the Section VI certificate claims (b155–b157) do not survive the paper's own rigorous bound, and the example must be re-run with a validated δ̄ (Lipschitz-refined grid) or resynthesized controllers. A secondary check: obtain the scripts and reproduce δ̄ = 0.02832, the products 0.90 and 0.8365, and the alternating-jitter divergence of Knom, to confirm the printed numbers.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central theorems are internally sound. Theorem 1 is a correct application of operator small-gain to the exact ZOH jitter decomposition: Δ(ε_k) is memoryless, (12) gives a uniform gain bound, and (18) closes the loop uniformly over E(ε̄). Theorem 2's bounded-real-lemma step is valid: (A_N, δ̄B_N, C_N, δ̄D_N) realizes Ñ_11 = δ̄N_11 with ||Ñ_11||∞ < 1, the BRL supply term is made nonpositive by ||ẽ_Δ(ε_k)|| ≤ 1, and the construction (14) makes N_11 strictly proper (the perturbation enters only the state update), so well-posedness is immediate. The constant-scaling robust-performance argument (20)–(21) is also correct: the scalar port commutation D_i bΔ = bΔ D_o preserves the unit gain of the time-varying block, and dynamic scalings are correctly excluded (b126–b129).\n\nThe load-bearing concern is at the certificate-verification step of the demonstration. The example computes δ̄ = 0.02832 by a 1000-point grid; the authors acknowledge (b143) a grid can understate the true maximum, deferring rigor to a Lipschitz check or the Appendix bound (b145). The grid error itself is negligible: σ̄(Δ(ε)) has Lipschitz constant ≈0.16 for this plant, so 1000 points over [−0.2, 0.2] give error ≲3e-5. The material gap is between the grid value and the only rigorous bound the paper supplies. For A = [0 1; −1 −0.2], B = [0;1], t_s = 0.1, ε̄ = 0.2, (23)–(24) give ‖A‖₂ ≈ 1.105, δ̄_A ≈ 0.0223, δ̄_B ≈ 0.0226, hence δ̄_appendix ≈ 0.0318 — about 12% above the grid value. Applied to the reported numbers, Krob's product 0.90 becomes ≈1.01 (the 10% margin is consumed and exceeded) and Kµ's 0.8365 becomes ≈0.94. The paper's only worked example therefore does not demonstrate a proof-grade certificate: with its own rigorous δ̄, the headline robust-stability controller likely fails (18), and the scripts are only 'available on request', so the numbers cannot be independently checked. The design-jitter-bound caveat (b135–b138) is acknowledged scoping, not part of this concern.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"The paper presents a method to reformulate sampled-data control with bounded sampling-time jitter as a robust control problem using an exact ZOH decomposition into a nominal LTI plant and a memoryless time-varying uncertainty block. The main theoretical contributions are: (i) a small-gain condition (Theorem 1) that certifies finite-gain ℓ2-stability and uniform exponential stability (Theorem 2) for all admissible jitter sequences, (ii) a generalized plant embedding that allows synthesis via standard robust-control tools (structured H∞ and μ-synthesis with constant scalings), and (iii) a design workflow with two synthesis alternatives. A numerical example compares nominal, robust-stability, and robust-performance designs.","tokens_in":13863,"tokens_out":1922,"duration_ms":13229,"significance":"If the results hold, the paper provides a practical bridge between aperiodic sampling theory and off-the-shelf robust-control software, enabling engineers to certify jitter robustness without dedicated LMI solvers or hybrid-system tools. The key insight—that the memoryless nature of the jitter perturbation reduces the operator small-gain condition to a matrix-norm test—is elegant and potentially impactful for embedded control applications.","major_comments":[{"comment":"The numerical example's certificate relies on δ̄ = 0.02832 computed from a 1000-point grid (b143). The authors acknowledge that a finite grid can miss the true maximum and mention a Lipschitz bound or the Appendix bound as remedies, but neither is actually applied. The Appendix bound (23)–(24) yields δ̄ ≈ 0.0318, which is about 12% larger. Using this rigorous bound, the small-gain product for Krob would become 0.90 × (0.0318/0.02832) ≈ 1.01, violating condition (18). Thus the example does not demonstrate a valid certificate for the reported controllers. The authors should either (a) recompute δ̄ rigorously (e.g., via the Lipschitz argument they mention or the Appendix bound) and verify that the small-gain condition still holds, or (b) adjust the design parameters (ε̄, weights) so that the certificate is satisfied with a rigorous δ̄. Without this, the central claim of 'provably correct controllers' is not supported by the example.","section":"§VI, Appendix","authors_acknowledge":"A finite grid can miss the true maximum between grid points, so its ¯δ can sit slightly below the true one.","authors_acknowledge_p":0.81},{"comment":"The robust-performance synthesis uses constant scalings (FitOrder=0) and the authors correctly note that D–K iteration need not converge to the global optimum (b131). However, the certificate check in Step 3 (Tab. I) requires that the fitted bound be below one for the selected controller. In the example, the reported fitted bound is 0.8365 < 1, but this is the value from the musyn iterate, not necessarily a rigorous upper bound on the scaled H∞ norm. The authors should clarify whether this value is the final PeakMuFit or a post-iteration check with the constant scaling fixed. More importantly, they should state whether the condition (20) is verified a posteriori with the chosen constant scalings, or whether the reported number is simply the musyn output. If the latter, the certificate is only as reliable as the D–K iteration's termination.","section":"§V-B, Eq. (20)–(21)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The proof of Theorem 2 uses the discrete-time bounded-real lemma to obtain a common quadratic Lyapunov function. The argument assumes that the normalized return map Ñ₁₁ = δ̄N₁₁ has H∞ norm < 1 and that the uncertainty satisfies ‖ẽ_Δ(ε_k)‖₂ ≤ 1. However, the bounded-real lemma requires that the realization (A_N, δ̄B_N, C_N, δ̄D_N) be minimal or at least detectable and stabilizable. The authors should comment on whether the nominal closed-loop realization is minimal (or at least that the uncontrollable/unobservable modes are stable, which follows from internal stability). This is a minor technical point but should be addressed for completeness.","section":"§IV-C, Theorem 2","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"The notation ‖D‖_{ℓ₂→ℓ₂} = sup_k ‖Δ(ε_k)‖₂ is correct for a memoryless operator, but the equality to max_{|ε|≤ε̄} σ̄(Δ(ε)) holds only if the supremum over k is attained at some ε in the closure of the range. Since the range is compact and the singular value is continuous, this is fine. Consider adding a brief justification.","section":"§III-D, Eq. (12)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The generalized plant P includes y_Δ,k as an input and u_Δ,k as an output. The equation for x_{k+1} includes + y_Δ,k, but the uncertainty output is y_Δ,k = Δ(ε_k) u_Δ,k. This is consistent with Fig. 3. However, the text says 'The inputs of P are [y_Δ, r, u] and its outputs are [u_Δ, z, v].' This is slightly confusing because u is the control input and v is the controller input. Clarify that the ordering matches the standard P–K–Δ convention.","section":"§IV-A, Eq. (14)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The design-jitter-bound discussion (b135–b138) is important but somewhat buried. Consider moving it to a separate subsection or highlighting it in the workflow figure.","section":"§V-C","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The sentence 'Both raise δ̄ and are omitted due to space limitation' is unsatisfactory for a journal paper. Either include the rigorous bound or remove the claim that the example certificates are rigorous. This is related to major comment 1.","section":"§VI, b143–b145","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The alternating jitter sequence ε_k = 0.15·(-1)^k is a specific periodic pattern. The authors correctly note that this is not a claimed worst case (b172). However, the caption could be more precise: 'Alternating jitter, ε_k = ±0.15' is clear, but the text should also state that the nominal responses are shown for reference.","section":"§VI, Fig. 7","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The related work section is concise but could mention the input-delay approach (Fridman et al.) more explicitly in the context of output-feedback synthesis. Currently it only cites [8] without discussing its limitations relative to the proposed method.","section":"§II","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper uses the notation ‖·‖_{H∞} for the H∞ norm of a discrete-time system. This is standard but could be clarified as the ℓ₂-induced norm of the LTI operator.","section":"General","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The bound for δ̄_B uses the integral from t_s to (1+ε̄)t_s. For negative jitter (ε_k < 0), the interval is shorter, so the tail length is |ε_k| t_s, but the majorant e^{‖A‖₂ τ} is increasing, so the bound still holds. This is correctly noted. However, the derivation of δ̄_A uses submultiplicativity: ‖e^{εA t_s} - I‖₂ ≤ e^{|ε|‖A‖₂ t_s} - 1. This is valid for any matrix norm, but the 2-norm is not submultiplicative for the exponential series? Actually, the matrix exponential satisfies ‖e^M‖ ≤ e^{‖M‖} for any submultiplicative norm, and the 2-norm is submultiplicative. So it's fine. Consider adding a brief justification.","section":"Appendix, Eq. (23)–(24)","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"major_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"The paper does something genuinely useful: it takes the known exact ZOH jitter decomposition (Fujioka, Kao-Fujioka) and shows that because the jitter block is memoryless, the operator small-gain theorem reduces to a simple matrix-norm condition that can be checked with standard LTI tools. Theorems 1 and 2 are correctly argued and give a clean certificate for finite-gain stability and uniform exponential stability. The synthesis workflow—structured H∞ for robust stability with nominal performance, constant-scaled μ synthesis for robust performance—is a practical contribution that closes an output-feedback gap Fujioka and Nakai left open. The paper is well written, the math is internally coherent, and the limitations (conservatism of the full block, need for a known jitter envelope) are stated honestly.\n\nThe soft spot is in the numerical example. The paper computes δ̄ = 0.02832 from a 1000-point grid and acknowledges that a grid can understate the true maximum. The stress-test note points out that the only rigorous bound the paper supplies (Appendix (23)–(24)) gives δ̄ ≈ 0.0318, about 12% larger. Applied to the reported numbers, the small-gain product for Krob would go from 0.90 to about 1.01, and for Kμ from 0.8365 to about 0.94. So the example does not actually demonstrate a proof-grade certificate with the paper's own rigorous bound. The authors mention that a Lipschitz bound or the Appendix bound would close the gap but omit the calculation due to space. That is a real weakness in the demonstration, though it does not affect the theoretical validity of the theorems.\n\nWho is this for? Practitioners who want a turnkey robust-control workflow for jitter, and theorists who want a clean application of small-gain to sampled-data systems. The paper deserves a serious referee; the theory is sound and the workflow is useful. The example needs to be fixed—either recompute with a rigorous δ̄ or show that the grid value is actually valid (e.g., by a Lipschitz argument). I would accept peer review on this paper and recommend conditional acceptance with a request to strengthen the numerical demonstration.","headline":"Solid small-gain packaging of jitter as a full-block uncertainty; the theory is clean and the workflow is practical. The only real gap is that the numerical example's certificate relies on a grid-computed δ̄ that the paper's own rigorous bound would invalidate.","tokens_in":14459,"tokens_out":574,"duration_ms":7197},"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"One matrix-norm test certifies sampled-data loops against unpredictable sampling-time jitter, and standard robust-control tools can synthesize controllers that provably stabilize every admissible jitter sequence.","keywords":["sampling-time jitter","aperiodic sampling","robust control","small-gain theorem","H-infinity synthesis","mu synthesis","time-varying systems","embedded control"],"falsifier":"Find a jitter sequence within the envelope E(epsilon_bar) for which the small-gain product is below one but the closed-loop state diverges. The paper claims this cannot happen because the small-gain condition implies a common quadratic Lyapunov function that decreases at every step.","tokens_in":16214,"feed_emoji":"🔧","tokens_out":1499,"duration_ms":20076,"temperature":0.7,"pith_summary":"This paper shows that standard off-the-shelf robust-control tools can design controllers for sampled-data systems whose sampling intervals vary unpredictably from sample to sample. The authors decompose the jittered plant into a nominal time-invariant part and a memoryless, norm-bounded uncertainty block. Because the uncertainty block has no internal dynamics, the operator small-gain condition for the time-varying loop reduces to a simple matrix-norm test that can be checked with a single H-infinity norm. The same test also guarantees uniform exponential stability of the closed-loop state, including behavior between samples. The paper then embeds this decomposition into a generalized plant and provides two synthesis workflows: structured H-infinity synthesis for robust stability with optimized nominal performance, and mu synthesis with constant scalings for guaranteed robust performance. A numerical example with a lightly damped second-order plant demonstrates that a nominal controller stable at every constant jitter value can destabilize when jitter alternates, whereas the robust designs remain stable. The main practical result is that an engineer can certify jitter robustness using standard robust-control toolbox commands, provided a known bound on the jitter is available.","feed_headline":"One matrix-norm test certifies jitter-robust loops","feed_subtitle":"Memoryless uncertainty reduces time-varying jitter to a simple norm condition, enabling standard H-infinity and mu synthesis.","key_machinery":"The key machinery is the decomposition of the jittered exact ZOH model into a nominal LTI part and a memoryless, norm-bounded uncertainty block. The jitter block is defined as the full matrix [Delta_A(epsilon_k) | Delta_B(epsilon_k)] that multiplies the stacked state-input vector. Because this block depends only on the current jitter value and has no internal state, its induced $\\ell^2$ gain is simply the maximum singular value over the jitter envelope. This static property is what reduces the operator small-gain condition to the matrix-norm test (18), and it also enables the use of constant scalings in mu synthesis without breaking the time-varying certificate.","core_discovery":"The central discovery is that sampling-time jitter in a sampled-data control loop can be exactly represented as a memoryless, norm-bounded uncertainty block in feedback with a nominal LTI plant. Because the uncertainty block has no dynamics, the operator small-gain theorem for time-varying systems reduces to a static matrix-norm condition: the product of the norm bound of the jitter block and the H-infinity norm of the nominal return map must be less than one. This condition simultaneously guarantees well-posedness, finite-gain stability, and uniform exponential stability for every admissible jitter sequence. The paper further shows that this decomposition can be embedded into a generalized plant with performance channels, allowing standard robust-control synthesis tools (structured H-infinity synthesis and constant-scaled mu synthesis) to produce controllers with provable jitter-robustness certificates.","pith_inferences":["The paper's approach could be extended to handle combined jitter and plant uncertainty by building a joint uncertainty model with a structured block, though the authors note this requires a new certificate.","The constant-scaling requirement for mu synthesis means that frequency-dependent D-scales cannot be used, which is a significant restriction compared to standard mu synthesis for LTI uncertainty. This limits the achievable robust performance bound.","The numerical example suggests that the certificates are moderately conservative: the small-gain test certifies stability up to epsilon_bar=0.22 for the robust-stability controller, while a search over periodic jitter sequences finds instability only at epsilon=0.415. This gap indicates room for less conservative analysis methods.","The paper's framework assumes the jitter bound is known and never violated. In practice, measuring or guaranteeing such a bound from a real-time operating system is nontrivial, and the engineer takes on responsibility for sequences outside the chosen envelope."],"forward_implications":["Engineers can now certify jitter robustness of a control loop using standard H-infinity and mu-synthesis tools, without needing dedicated LMI solvers or hybrid-system analysis.","The small-gain certificate (18) provides a simple, LTI-computable test that can be applied a posteriori to any existing discrete-time controller to check its robustness against a given jitter envelope.","The design-jitter bound can be used as a tuning knob: a smaller certified bound allows better nominal performance, while a larger bound gives stronger jitter guarantees at the cost of performance.","The certificate covers not only input-output stability but also uniform exponential stability of the state, including inter-sample behavior, without requiring an explicit inter-sample performance specification.","The mu-synthesis approach with constant scalings provides a guaranteed worst-case induced gain from reference to weighted error, valid for every admissible jitter sequence."],"supporting_citations":[{"why":"Fujioka's discrete-time embedding of aperiodic sampling for stability analysis with state feedback, which the paper extends to dynamic output-feedback synthesis with performance specifications.","marker":"[2]"},{"why":"Fujioka's output-feedback stability analysis for aperiodic sampling, which the paper extends to synthesis with standard robust-control tools.","marker":"[3]"},{"why":"Desoer and Vidyasagar's operator small-gain theorem, which is the core stability certificate (Lemma IV.1) for time-varying jitter.","marker":"[13]"},{"why":"Packard and Doyle's complex structured singular value and the robust performance theorem (Thms. 4.3, 10.4) that the paper uses for mu-synthesis with constant scalings.","marker":"[14]"},{"why":"Zhou, Doyle, and Glover's textbook for the bounded-real lemma and LFT framework used in the stability proofs.","marker":"[12]"}],"fun_headline_variants":[],"cache_read_input_tokens":0,"weakest_assumption_plain":"The certificates require a known, never-violated bound on the relative jitter for every sample, and the engineer must guarantee that the actual jitter never exceeds this bound.","fun_headline_variants_meta":{"raw":{"variants":[]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":0.000308,"raw_usage":{"cost":0.0003077802,"is_byok":false,"cost_details":{"upstream_inference_cost":0.0003077802,"upstream_inference_prompt_cost":7.11802e-05,"upstream_inference_completions_cost":0.0002366},"total_tokens":2287,"prompt_tokens":987,"completion_tokens":1300,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":256,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":0}},"tokens_in":987,"tokens_out":1300,"cost_source":"openrouter_usage","duration_ms":8090,"image_pages":[],"temperature":1.0,"billed_attempts":1,"reasoning_tokens":0,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-11-13T07:23:49.810120+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":"Find a jitter sequence within the envelope E(epsilon_bar) for which the small-gain product is below one but the closed-loop state diverges. The paper claims this cannot happen because the small-gain condition implies a common quadratic Lyapunov function that decreases at every step.","supporting_citations":[{"cited_title":"A discrete-time approach to stability analysis of systems with aperiodic sample-and-hold devices,","cited_arxiv_id":null,"evidence_quote":"Fujioka's discrete-time embedding of aperiodic sampling for stability analysis with state feedback, which the paper extends to dynamic output-feedback synthesis with performance specifications."},{"cited_title":"Stability analysis for a class of networked/embedded control systems: Output feedback case,","cited_arxiv_id":null,"evidence_quote":"Fujioka's output-feedback stability analysis for aperiodic sampling, which the paper extends to synthesis with standard robust-control tools."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Desoer and Vidyasagar's operator small-gain theorem, which is the core stability certificate (Lemma IV.1) for time-varying jitter."},{"cited_title":"The complex structured singular value,","cited_arxiv_id":null,"evidence_quote":"Packard and Doyle's complex structured singular value and the robust performance theorem (Thms. 4.3, 10.4) that the paper uses for mu-synthesis with constant scalings."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Zhou, Doyle, and Glover's textbook for the bounded-real lemma and LFT framework used in the stability proofs."}],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}