{"id":"8c8f5351-0c46-46f6-9bc0-0db01233d39c","arxiv_id":"2509.19111","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"Adding a previously published robust frequency estimator as a feed-forward to a standard SRF-PLL improves phase tracking on PMSM drive currents, especially during frequency ramps.","lead":"This paper adds an online frequency estimator as a feed-forward input to a standard three-phase phase-locked loop, then tests it on currents from a motor drive. The combined scheme tracks phase and frequency more accurately during speed ramps and load steps, while keeping the same loop structure.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The baseline SRF-PLL is run with zero feedforward (˜ω=0), whereas the standard SRF-PLL uses a nominal-frequency feedforward; the claimed 'clear superiority' may reflect any-feedforward vs. none rather than adaptive vs. constant feedforward.","rationale":"The reader's weakest assumption concerns estimator convergence under distortion; that is a robustness/generalization issue, but the recorded experiments already demonstrate convergence for those specific traces. The load-bearing issue for the central claim as stated is the control condition: the baseline SRF-PLL is stripped of the nominal feedforward that the standard design uses. This directly affects whether 'clearly superior' is a meaningful scientific result. The paper itself acknowledges that error metrics are only relative (Section IV-C) and provides no confidence intervals, but the absence of a constant-feedforward baseline is a specific, testable deficiency. The proposed fix is straightforward: run the baseline with a constant ˜ω equal to the known test frequency. If the improvement largely disappears, the paper's contribution is reduced to replacing a known constant with an online estimate, which is useful but not 'clearly superior' without quantifying the cost/benefit. If it persists, the claim is strengthened. I therefore recommend no change to the reader's CONDITIONAL verdict, but for a different reason.","tokens_in":11754,"tokens_out":19774,"duration_ms":784901,"concrete_test":"Re-evaluate the three experiments of Section IV using the recorded phase-current data (or a simulation with the same signals) with the baseline SRF-PLL ˜ω set to a constant nominal value: 50 rad/s for the 50 rad/s experiment, 150 rad/s for the 150 rad/s experiment, and, for the ramp experiment, a fixed value such as the average or initial ramp frequency (e.g., 90 rad/s). Recompute EME (Table I) and ERMS (Table II). If the SRF-PLL-FF advantage is no longer clearly superior (e.g., relative reduction <20% or overlapping confidence intervals from repeated trials), the headline claim is overstated; if the advantage persists, the concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central comparison in Table I contrasts SRF-PLL-FF with an SRF-PLL whose feedforward input is set to zero (Fig. 9 caption: 'SRF-PLL without feed-forward i.e. ˜ω = 0'). But the SRF-PLL architecture in Fig. 1 and the standard design of Kaura & Blasko [14] include a nominal-frequency feedforward, ˜ω, that places the loop near the operating point. Setting ˜ω=0 forces the PI to acquire the full frequency from zero and leaves, during load steps and frequency ramps, exactly the error that a constant feedforward would have removed. The observed 'clear superiority' may therefore be the trivial advantage of any feedforward over none, rather than the advantage of the robust adaptive estimator over a constant nominal feedforward. The paper never reports the standard SRF-PLL with a reasonable constant ˜ω (e.g., 50 rad/s in the 50 rad/s test, 150 rad/s in the 150 rad/s test, and a fixed mid-ramp value in the ramp test). Without that control condition, the incremental value of the proposed feed-forward estimator is not established. Section IV-C also acknowledges that the EΣ and EME metrics are only relative values, and no repeated trials or confidence intervals are provided, making the magnitude of the claimed advantage fragile.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes an SRF-PLL augmented by a feed-forward frequency estimator previously introduced by the first author in [18]. The feedback loop is tuned by the symmetrical optimum method, and a normalization scheme is used to make the loop gain independent of signal amplitude. The feed-forward signal is the averaged output of three parallel robust frequency estimators operating on the three phase currents. The main experimental claim is that the SRF-PLL-FF achieves substantially lower phase errors than the same SRF-PLL without feed-forward in torque-varying and frequency-ramp PMSM drive tests.","tokens_in":12082,"tokens_out":13318,"duration_ms":101201,"significance":"If the claimed improvement is real, the contribution is practically significant: a one-parameter, model-free frequency estimator that removes ramp-induced phase error without raising the loop order would be a useful addition to SRF-PLL design. The paper also demonstrates a real-time implementation on inverter-fed PMSM currents, which is a relevant and nontrivial testbed. The strength of the paper is its experimental grounding; the main weakness is that the comparison baseline is not the standard SRF-PLL with a nominal feed-forward term, so the incremental value of the adaptive estimator over a constant feed-forward is not established.","major_comments":[{"comment":"The central comparison uses an SRF-PLL with the feed-forward input set to zero (Fig. 9 caption: 'SRF-PLL without feed-forward i.e. ˜ω = 0'). The paper itself notes in Section II-A that knowledge of a nominal ˜ω drives the PLL closer to the operating point and facilitates the regulator. The standard SRF-PLL of [14] includes a nominal feed-forward. Thus Table I compares 'any feed-forward' versus 'no feed-forward', not the proposed adaptive estimator versus a constant nominal feed-forward. A control condition with a constant ˜ω (e.g., 50 rad/s and 150 rad/s in the corresponding tests, and a mid-ramp value in the ramp test) is needed to support the claim that the robust estimator provides clear superiority.","section":"Section IV-B, Table I, Fig. 9"},{"comment":"Theorem 1 states asymptotic convergence of the estimator under an unbiased harmonic plus band-limited zero-mean noise, with proof deferred to [18]. The experimental conditions include PWM notch distortion, persistent subharmonics, and a 0.05 s data loss that are not covered by the theorem's assumptions. The paper claims robustness to subharmonics (Section III, final paragraph) but reports no supporting results. The feed-forward benefit depends entirely on the estimator converging in these conditions; the paper should either provide a supporting analysis or explicitly limit the theoretical claim and rely on repeated experimental validation.","section":"Section III, Theorem 1"},{"comment":"The error metrics are computed from single experimental records, and the paper acknowledges that the phase-wrap induced peaks make EΣ and EME only relative values. Without repeated trials or confidence intervals, the magnitudes of the improvements (e.g., 0.0584 vs. 0.0376 at 50 rad/s) may not be statistically robust. Adding multiple runs or a sensitivity analysis would strengthen the conclusion, particularly because the claim of 'clear superiority' is the paper's main result.","section":"Section IV-C, Tables I and II"}],"minor_comments":[{"comment":"'Phase-Looked Loop' should be 'Phase-Locked Loop' in the title and abstract.","section":"Title and Abstract"},{"comment":"The factor '√3/2 κ/ki' is ambiguous; from the derivation with U=√(2/3), the coefficient should be √(3/2)κ/ki. Please disambiguate and correct if needed.","section":"Equation (10)"},{"comment":"Typo: 'It it also worth noting' should be 'It is also worth noting'.","section":"Section III, last paragraph"},{"comment":"The abbreviation 'PLL-FF' is used in the tables but the text defines 'SRF-PLL-FF'; please make the notation consistent.","section":"Tables and text"}],"recommendation":"major_revision","confidential_remarks":"The main issue is the baseline comparison: the standard SRF-PLL with a constant feed-forward is not tested. This is fixable by adding an offline comparison using the same recorded signals. If the authors add this control condition and clarify the scope of the convergence theorem, the paper could become acceptable. The technical core (symmetrical optimum tuning, normalization, estimator implementation) appears sound."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Useful, honest engineering paper: it adds a robust frequency estimator as feedforward to a standard SRF-PLL, uses a power-invariant normalization, and validates it on real PMSM currents. The experiments are real and the direction of the improvement is consistent across all three scenarios. But the main comparison is the weak point. Table I and Fig. 9 contrast the proposed SRF-PLL-FF with the same SRF-PLL configured with ω̃=0, i.e., no feedforward at all. The standard SRF-PLL in Kaura & Blasko includes a nominal frequency feedforward; that is what places the loop near the operating point. Setting it to zero forces the PI to do all the work and leaves exactly the error a constant feedforward would remove. So the 'clear superiority' is, at least partly, the trivial advantage of any feedforward over none. The paper does not report the more informative control condition: the same PLL with a constant ω̃ at the operating frequency (50, 150, or a fixed mid-ramp value). Without that, the incremental value of the adaptive estimator over a constant feedforward is not established. That said, the paper is honest about its metrics: Section IV-C acknowledges phase-wrap jumps and says EΣ and EME are relative only, which is fair, but it still means the magnitude of the claimed advantage is fragile — single records, no confidence intervals.\n\nWhat is genuinely good: the open-loop small-signal analysis is correct, the symmetrical optimum tuning is clean, and the power-invariant normalization is a nice touch that removes amplitude dependence. The estimator is prior work — Theorem 1 is cited, not proved here — but the paper does not misrepresent it. The parallel average of three single-phase estimators recovers visibly after the 0.05 s data loss, which is real evidence of robustness in the tested conditions, even if no formal robustness guarantee covers PWM notching or subharmonics. The ramp experiment also matches the theory: feedforward turns a ramp into a step and removes the steady-state error without raising loop order.\n\nFor whom: researchers in motor drives and grid converter synchronization, especially those wanting a practical, real-time implementable feedforward scheme. It deserves a serious referee; the request should be for the missing constant-feedforward control condition, repeated trials with intervals, and ideally released code and data. The math is sound and the experiments are genuine; the paper's own comparison overstates what the estimator uniquely contributes.","headline":"Useful incremental PLL paper with genuine hardware data, but the headline comparison is against zero feedforward, not the standard constant-feedforward SRF-PLL, so the claimed advantage is overstated.","tokens_in":12553,"tokens_out":2389,"would_cite":false,"duration_ms":19364,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Feeding a robust frequency estimate into a standard three-phase phase-locked loop cuts phase-locking error dramatically during frequency ramps and load steps.","keywords":["phase-locked loop","SRF-PLL","frequency estimation","feedforward control","symmetrical optimum","PMSM drive","phase synchronization","power-invariant normalization"],"falsifier":"Run the estimator alone on the same experimental data starting from an initial frequency near the 3rd harmonic (e.g., 3ω) and observe whether it converges to the fundamental rather than latching onto the subharmonic, or introduce a data-loss interval longer than 0.05 s and measure whether the phase error grows beyond a few periods instead of recovering. If the estimator locks onto a subharmonic or the loop loses lock after longer dropouts, the claimed robustness and the feedforward advantage collapse.","tokens_in":11655,"feed_emoji":"⚡","tokens_out":3499,"duration_ms":27540,"temperature":0.7,"pith_summary":"The paper tries to establish that a conventional synchronous-reference-frame PLL can be made substantially more accurate during frequency ramps and load transients by injecting an online, model-free frequency estimate as a feedforward signal, without increasing the loop order or adding filters. It shows that the absolute mean phase error drops roughly 6.9-fold during a frequency ramp (from 0.9205 rad to 0.1327 rad) and improves by roughly 35–40% under torque-varying operation at 50 rad/s and 150 rad/s. The authors also propose a power-invariant normalization that makes the loop insensitive to input amplitude changes, plus a symmetrical-optimum PI tuning rule with a single design parameter. Experimental evidence comes from three-phase currents of inverters-fed PMSM drives under realistic disturbances: PWM notching, subharmonics, load steps, and a 0.05 s data loss.","feed_headline":"Frequency ramp phase error drops 6.9x with feed-forward PLL","feed_subtitle":"Injecting a robust online frequency estimate removes ramp lag and cuts load-step phase error without adding loop order or filters.","key_machinery":"The load-bearing mechanism is the feedforward path built around a second-order robust frequency estimator: for each phase n, the estimator dynamics are (η̇₁, η̇₂)ᵀ = (0,1; -ω̃², -2ω̃)(η₁,η₂)ᵀ + (0, 2ω̃)ᵀ Zₙ, ν = η₂, with the adaptation law ˙ω̃ = -γ sign(η₁)(Zₙ - ν). The averaged estimate ω̃ = (ω̃ₐ + ω̃_b + ω̃_c)/3 is injected into the PLL's frequency path, effectively replacing the nominal frequency entry. This shifts the loop's equivalent input from a ramp (κ/s²) to a step (const/s), so the final-value theorem gives zero steady-state frequency error. The secondary machinery is the power-invariant normalization factor N = √(Zₐ² + Z_b² + Z_c²) that keeps the normalized amplitude at √(2/3) reg","core_discovery":"The central claim is that a robust, one-parameter frequency estimator, running in parallel on the three phase currents and averaged, can supply the SRF-PLL with an accurate feedforward frequency. This converts a frequency ramp into an equivalent step input, eliminating steady-state frequency error that a type-2 PLL would otherwise exhibit. The paper demonstrates experimentally that this feed-forward extension (SRF-PLL-FF) locks phase faster and with lower error than the identical SRF-PLL without feedforward, while preserving the loop's phase margin and robustness. The design also uses a power-invariant normalization of the three-phase signal so the loop gain no longer depends on input amplit","pith_inferences":["A natural generalization is to apply the same feedforward-estimator concept to other PLL variants (e.g., SOGI-based or DSOGI-PLL), potentially reducing their tuning complexity while improving ramp tracking.","Because the frequency estimator is model-free and exponentially convergent (per Theorem 1), one might formally analyze the combined loop as a slow-fast system and derive a guaranteed phase-lock margin; the paper does not perform this stability analysis.","The large ramp-error reduction suggests the scheme could enable sensorless PMSM drives to use faster acceleration ramps than feedback-only PLLs allow, a direct but unexplored operational benefit.","A concrete testable extension is to evaluate the same SRF-PLL-FF on a grid-connected converter under unbalanced or harmonically distorted voltages; the paper's subharmonic-robustness claims imply it should hold, but experimental evidence here is limited to motor currents."],"forward_implications":["Frequency ramps can be tracked with zero steady-state error without raising the PLL loop order, so phase margin and noise immunity are preserved.","The loop becomes insensitive to input amplitude variations, making the PI tuning valid across a wide range of signal magnitudes.","The entire scheme is real-time executable with no extra filters or pre/post-processing; only one additional tuning parameter (γ) is needed for the estimator.","Phase lock is maintained through temporary data loss and PWM notching, recovering within a few periods.","The same architecture should transfer to other three-phase harmonic signals, such as grid voltages, where ramp-like frequency events occur during disturbances."],"fun_headline_variants":["Feed-forward PLL cuts ramp phase error by 6.9x","Robust PLL with feed-forward: 6.9x less ramp error","Model-free feed-forward estimator improves PLL ramp tracking","PLL feed-forward: faster lock, lower phase error on ramps","Experimental PLL feed-forward wins on frequency ramps"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The frequency estimator converges on the actual PWM-notched, subharmonic-tainted, occasionally missing current signals quickly enough that the feedforward term stays accurate during transients; the convergence proof is deferred to an earlier paper, not re-established here.","fun_headline_variants_meta":{"raw":{"variants":["Feed-forward PLL cuts ramp phase error by 6.9x","Robust PLL with feed-forward: 6.9x less ramp error","Model-free feed-forward estimator improves PLL ramp tracking","PLL feed-forward: faster lock, lower phase error on ramps","Experimental PLL feed-forward wins on frequency ramps"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000731,"raw_usage":{"total_tokens":3110,"prompt_tokens":746,"completion_tokens":2364,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":490,"completion_tokens_details":{"reasoning_tokens":2274}},"tokens_in":490,"tokens_out":2364,"duration_ms":16109,"temperature":1.0,"reasoning_tokens":2274,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T15:26:52.602411+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the estimator alone on the same experimental data starting from an initial frequency near the 3rd harmonic (e.g., 3ω) and observe whether it converges to the fundamental rather than latching onto the subharmonic, or introduce a data-loss interval longer than 0.05 s and measure whether the phase error grows beyond a few periods instead of recovering. If the estimator locks onto a subharmonic or the loop loses lock after longer dropouts, the claimed robustness and the feedforward advantage collapse.","supporting_citations":[],"review_version":1}