{"id":"ce42e074-d42b-4b50-9486-90f9182bc938","arxiv_id":"1908.08535","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A Class D power amplifier using multi-simultaneous-frequency selective harmonic elimination PWM generates multiple excitation frequencies simultaneously for eddy current testing, demonstrated with an FPGA prototype.","lead":"This paper designs a Class D (switch-mode) power amplifier that generates several magnetic field frequencies at once for multi-frequency eddy current testing, using a new pulse modulation scheme. A prototype drove a coil with over 60 amps peak-to-peak at four simultaneous frequencies with low distortion.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (9) optimizes m_p as a free variable, making the selected-harmonic residual identically zero for any switching angles; the central harmonic-control claim is unsupported as written.","rationale":"The reader's stated weakest assumption is the ideal-inductor model. That is a valid concern, but it is secondary: it affects quantitative accuracy, not the existence of a control mechanism. The more load-bearing issue is that Eq. (9), as printed, is algebraically degenerate. Because m_p is an optimization variable, the residuals for the selected harmonics can always be set to zero by choosing m_p equal to the achieved normalized harmonic amplitude. Thus the objective imposes no constraint on the switching angles for the frequencies the method claims to control; the only nontrivial constraints are the inequality bounds on the eliminated harmonics. The paper's reported switching angles in Table II therefore cannot be attributed to selected-harmonic amplitude control. The experimental waveforms do show multi-frequency content, but that demonstration does not validate the specific MSF-SHEPWM optimization claimed as the contribution. A corrected formulation with fixed m_p targets is straightforward, and if the authors supply that correction and re-validate the spectra, the central claim could be restored. As the manuscript stands, however, the central claim is not supported by the printed mathematics.","tokens_in":12490,"tokens_out":9547,"duration_ms":102073,"concrete_test":"Fix the modulation indices to the target implied by the design objective: with desired current amplitudes b_p proportional to 1/p, set m_1 = m_3 = m_7 = m_17 = K for a chosen K (e.g., the average of the four initial m_p values used in the paper), and re-solve Eq. (15) with only alpha_1,...,alpha_6 as variables. Compare the resulting switching angles and scaled modulation indices with Table II and the measured FFT in Table III. If the re-optimized angles reproduce the reported values and spectra, the free m_p in Eq. (9) was immaterial; if they do not, the printed formulation is degenerate and the claimed harmonic control is unsupported.","verdict_should_be":"REJECT","load_bearing_attack":"Eq. (9) (and its specialization Eq. (15)) lists the modulation indices m_p, p in Phi, among the optimization variables. The objective is sum_{p in Phi} lambda_p ((p omega L / V0) b_p - m_p)^2. Because m_p is free, for any switching-angle vector alpha the solver can choose m_p = (p omega L / V0) b_p, making every selected-harmonic residual exactly zero. The objective is therefore identically zero at its global minimum for every alpha, and the only active constraints are the inequality bounds on the eliminated set Psi. The selected harmonic magnitudes are not constrained at all. This is not a numerical issue but an algebraic degeneracy. The paper never provides fixed target values for m_p; the statement that m_p is 'treated as an optimising parameter' confirms the free-variable reading. If the authors intended m_p to be prescribed targets, that target specification is missing, and the reported optimal angles in Table II are not reproducible from the printed problem. This concern is more fundamental than the ideal-inductor assumption: even with a perfect inductor, Eq. (9) does not enforce the desired spectrum.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a modulation strategy, MSF-SHEPWM, for a class-D power amplifier intended to drive eddy-current testing coils with a multi-simultaneous-frequency current. The method synthesizes an objective coil current as a sum of selected harmonics, approximates it by a piecewise-linear waveform, computes the Fourier coefficients of the coil current under the assumption that the coil is an ideal inductor, and then solves a constrained optimization problem for the PWM switching angles. Two experiments are reported: an FPGA-driven full H-bridge feeding a 1.4 µH coil at four simultaneous frequencies (about 50 kHz to 857 kHz), and a second experiment feeding a 662 µH coil at five simultaneous frequencies (about 1.6 kHz to 130 kHz). Measured FFT spectra show that the intended harmonics dominate the coil current, and the paper reports efficiency and reactive-power metrics for both cases.","tokens_in":12845,"tokens_out":4273,"duration_ms":49817,"significance":"If the optimization problem were well posed, the paper would make a useful contribution. The idea of deliberately repurposing SHEPWM from fundamental-frequency control and harmonic elimination to simultaneous multi-harmonic synthesis for eddy-current testing addresses a real gap in switch-mode MECT excitation, and the FPGA prototype plus the two sets of measured spectra are valuable evidence that the physical approach can work. The paper also reports data tables and an explicit power/efficiency comparison, which supports reproducibility of the hardware results. However, the central formulation as printed is degenerate because the target modulation indices are treated as free optimization variables, so the stated problem does not enforce the desired spectrum. This affects the main algorithmic claim of the paper and must be fixed before the method can be assessed.","major_comments":[{"comment":"The optimization problem (9) lists the modulation indices m_p among the decision variables, and the paragraph following Eq. (9) states that mp is treated as an optimizing parameter. With m_p free, the objective sum_{p in Phi} lambda_p ((p omega L / V0) b_p - m_p)^2 can be made exactly zero for any switching-angle vector by choosing m_p = (p omega L / V0) b_p. The selected-harmonic terms therefore impose no constraint on the solution, and the only active constraints are the inequality bounds on the eliminated set Psi. The same degeneracy appears in the concrete problem (15) of Experiment 1. As written, the optimization does not enforce the intended power distribution among the selected frequencies, and the optimal angle sets in Table II are not the solution of a well-posed problem that controls those harmonics. The manuscript must either remove m_p from the decision variables and specify fixed target values for the selected-harmonic amplitudes, or add explicit equality constraints that tie m_p to prescribed targets, and then re-report the optimization and its results accordingly.","section":"Eq. (9), Eq. (15), Section II-C"},{"comment":"The Fourier-coefficient formula (8) is derived under the ideal-inductor assumption that the coil current is exactly the integral of the applied voltage divided by L. The paper does not provide a quantitative comparison of the measured harmonic amplitudes with the values predicted from the computed switching angles. In Experiment 1, the coil dissipates 6.87 W and is driven with 60 A peak-to-peak current, so resistance and eddy-current loading are not negligible a priori; in Experiment 2 the same idealization is used for a 662 µH coil. To support the claim that the optimized angles produce the intended spectrum, the authors should report, for both experiments, the predicted versus measured RMS values (or normalized coefficients) at the selected and eliminated harmonics, including a statement of the resulting amplitude error per harmonic and overall distortion.","section":"Eq. (8), Table III, Table IV, Table V, Experiment No. 1 and No. 2"},{"comment":"Experiment 2 deliberately skips the optimization stage and uses only the initial switching angles obtained from the gradient of the objective function. Consequently, the second experiment does not test the proposed optimization formulation at all; it tests only the time-domain synthesis stage. The claim that the full MSF-SHEPWM procedure produces low-distortion multi-frequency currents therefore rests entirely on Experiment 1, whose optimization problem has the degeneracy described above. The paper should either add an experiment that implements the corrected optimization, or clearly separate the two claims and identify which stage each experiment validates.","section":"Section III-C-1, Table II, Experiment No. 2"}],"minor_comments":[{"comment":"There is a sign inconsistency between the normalized coefficient in Eq. (14) and the expression in Eq. (15): Eq. (14) yields a negative scaled coefficient, while Eq. (15) writes the residual with a plus sign before m_p. If m_p is intended as a positive amplitude, the objective should use the absolute value or an explicitly signed form so that the optimization problem matches the physical meaning of the modulation index.","section":"Eq. (15)"},{"comment":"The sentence 'if annulation of the objective function is not possible' contains a typographical error; 'annulation' should be 'annihilation' or 'nullification'.","section":"Section II-C"},{"comment":"The phrase 'the coefficient wp is chosen as 1/p for the selected frequencies' is stated before any motivation; it would be clearer to present the SNR-based motivation together with the definition, since this choice is central to all subsequent examples.","section":"Section II-A"},{"comment":"Table II reports the scaled modulation index mp/p for the selected harmonics but no target values for these quantities; a column of target normalized coefficients would make the optimization results interpretable even after the m_p issue is resolved.","section":"Table II and Eq. (16)"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's own Section II-C explicitly says that mp is treated as an optimizing parameter, so the degeneracy of Eq. (9) is not an artifact of the review pipeline. I would ask the authors to give the exact target vectors and the exact solver configuration used in Experiment 1, since the printed problem cannot reproduce Table II. If the experimental spectra are reproducible with the reported angles, the paper may well be salvageable, but the algorithmic contribution needs reformulation before it can be evaluated."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Punchline: The paper demonstrates a working Class D amplifier that drives an eddy-current coil at several simultaneous frequencies, which is a genuinely useful engineering result. But the optimization that is supposed to set the switching angles is degenerate as written: Eq. (9) treats the modulation indices m_p as free variables, so for any switching-angle vector α the objective can be made exactly zero by choosing m_p = (pωL/V0)b_p. The paper never fixes m_p to the desired harmonic amplitudes, so the reported 'optimal' angles in Table II are not actually constrained by the target spectrum. That is a load-bearing flaw.\n\nWhat is new and good: Re-purposing SHEPWM to deliberately retain a set of harmonics with 1/p amplitude weighting for multi-frequency excitation is a real departure from the standard elimination-only use, and the paper positions it correctly against the SHEPWM and harmonic-mitigation literature. The hardware work is concrete: a single H-bridge, an FPGA-generated pulse pattern, two different coils, and measured FFT spectra that show the selected frequencies dominate. The second experiment skips the optimization entirely and still works, which suggests the core idea is sound even if the optimization is not.\n\nSoft spots: The Eq. (9) degeneracy is the main one. It is not a minor typo; the text explicitly calls m_p an optimizing parameter. Maybe the authors meant for m_p to be fixed targets, but that is not written, so the results are not reproducible from the claim. The ideal-inductor assumption is also underexamined: the coil in Experiment 1 dissipates tens of watts, so resistance and eddy-current loading are not negligible, yet the Fourier derivation uses a pure L without a quantitative error check. The efficiency framing is confusing: the abstract emphasizes high efficiency, but Experiment 1 shows 20% drain efficiency; the text provides a reasonable explanation later, but it should be surfaced earlier. Minor: the FFT tables have no uncertainty or repeatability information.\n\nThe citation pattern is fine; the SHEPWM reviews are properly cited, and the self-citations are to prior eddy-current instrumentation work, not to inflate the new claim.\n\nWho this is for: instrumentation engineers working on multi-frequency eddy current testing. The hardware demonstration is worth knowing, but the math as written should not be used as a design procedure. I recommend engaging with it for the experimental concept, but only after the authors fix the optimization (fix m_p to targets or justify otherwise) and add error analysis. A serious referee could catch this and get a much better paper out of it.","headline":"Promising hardware demonstration of multi-frequency Class D excitation, but the optimization formulation in Eq. (9) is degenerate and the claimed harmonic control is unsupported as written.","tokens_in":13255,"tokens_out":5197,"would_cite":false,"duration_ms":45572,"reading_group":"maybe","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper shows that a single switch-mode (Class D) power amplifier can generate a multi-simultaneous-frequency coil current for eddy current testing by keeping, rather than eliminating, selected harmonics.","keywords":["class D power amplifier","selective harmonic elimination PWM","multi-frequency eddy current testing","multi-simultaneous-frequency","switch-mode amplifier","FPGA","harmonic optimization"],"falsifier":"Measure the transmitting coil's complex impedance at each transmitted frequency, recompute the harmonic currents from the optimized pulse voltage using that measured impedance (resistor in series with the inductance) instead of a pure inductor, and compare with the FFT of the measured coil current; a systematic mismatch at the higher frequencies would show that the pure-inductor assumption underlying the switching-angle design does not hold.","tokens_in":12299,"feed_emoji":"⚡","tokens_out":14182,"duration_ms":126500,"temperature":0.7,"pith_summary":"This paper proposes a modulation method, multi-simultaneous-frequency selective harmonic elimination PWM (MSF-SHEPWM), that lets one switch-mode (Class D) power amplifier drive an eddy-current testing coil with several excitation frequencies at the same time. The authors argue that the usual SHEPWM goal of eliminating all harmonics except the fundamental can be inverted: selected harmonics are deliberately kept and shaped into the multi-frequency coil current, while unwanted harmonics are suppressed by optimizing the pulse switching angles. The claim is supported by an FPGA-based prototype that transmits four simultaneous frequencies (about 50 kHz to 857 kHz) at up to 60 A peak-to-peak coil current with about 8% total harmonic distortion, and a second, faster design that transmits five frequencies (about 1.6 kHz to 130 kHz) without running the optimization at all. If correct, a single efficient and software-reconfigurable power stage can replace multiple linear amplifiers in multi-frequency eddy current testing.","feed_headline":"One switch-mode amplifier drives five eddy-current test tones","feed_subtitle":"By programming pulse edges, a single Class D stage delivers simultaneous inspection frequencies with low distortion.","key_machinery":"The load-bearing identity is the Fourier-coefficient formula (Eq. 8) for the coil current of a multi-level Class D amplifier: b_p = −4V0/($p^{2}$ π ω L) Σ_{q=1}^{N} (−1)^s sin(p α_q), where α_q are the N switching angles in a quarter period, V0 is the voltage step per level, ω is the fundamental angular frequency, and L is the coil inductance. SHEPWM, selective harmonic elimination pulse width modulation, is a switching strategy that sets pulse-edge positions so that specific output harmonics are controlled. The formula turns the modulation problem into a constrained, non-convex trigonometric optimization over the α_q and the modulation indices m_p: selected harmonics are pulled toward their target amplitudes with weights λ_p, and unselected harmonics are bounded by thresholds ϵ_p (Eq. 9). The method's operating principle is that the coil current is the integral of the pulse voltage across an inductor, so a piecewise-linear approximation of the target multi-sine current is produced directly by the switching waveform. The optimization is solved by an interior-point algorithm initialized at the turning points of the target waveform, and the resulting angles are stored as a lookup table that an FPGA reads to produce the PWM pulses.","core_discovery":"The central discovery is that selective harmonic elimination PWM can be repurposed from suppressing harmonics to generating a deliberate superposition of them. The design starts with a target coil current f(t) made of a fundamental and several of its harmonics, with amplitudes w_p chosen (here as 1/p) so higher transmitted frequencies receive proportionally more excitation. This target is approximated by a piecewise-linear current g(t), which is what a pulse voltage across a pure inductor produces; the Fourier coefficients of g(t) are then written as an explicit function of the switching angles α_q and the coil inductance L (Eq. 8). A constrained nonlinear least-squares problem (Eq. 9) tunes the angles so that the selected harmonics match the desired amplitudes while unselected harmonics stay below given thresholds, and the paper solves it with an interior-point algorithm initialized from the turning points of the target waveform. Two FPGA-driven H-bridge experiments verify the method: one reaches 60 A peak-to-peak across a 1.4 µH coil with four simultaneous frequencies and about 8% THD, the other drives a 662 µH coil with five frequencies using only the initial switching-angle guess, skipping the optimization entirely. Measured drain efficiencies of about 20% and 77% in the two experiments, and an energy conversion factor ζ above unity, quantify the power behavior: the lossy coil gives high efficiency, while the high-Q coil gives low efficiency but still delivers over six times the dc input as reactive power to the coil.","pith_inferences":["The same synthesis should also work with arbitrary relative phases among the transmitted frequencies, since the Fourier-coefficient formula already depends on the switching angles; phase-shaped spectra such as the paper's f5(t) example would then be realizable, enabling phase-coded multi-frequency excitation.","The 1/p amplitude weighting is a design choice; in a specific inspection task, the weights could be optimized to match the sensitivity of the coil-sample coupling, improving the signal-to-noise ratio at each transmitted frequency.","If the pure-inductor assumption proves inaccurate at high frequencies, a direct extension would be to measure the coil impedance and fold it into the optimization, effectively pre-distorting the switching angles to compensate for resistance and eddy-current loading.","The method appears transferable to other multi-frequency inductive sensing applications, such as magnetic induction spectroscopy in the beta-dispersion range, where the same frequency-selection logic applies."],"forward_implications":["A single switch-mode amplifier can replace the multiple linear power amplifiers that a multi-frequency eddy current testing system would otherwise need, cutting size, cost, and heat dissipation.","The excitation spectrum becomes software-reconfigurable: changing the stored lookup table of switching angles changes which harmonics are transmitted, so one hardware design can serve different inspection tasks.","Because harmonic content above the highest selected frequency is already low in the first experiment, a low-pass output filter can often be omitted.","The optimization recipe, including the deliberate relaxation of the highest-frequency 'benign ripple' component, provides a practical way to balance amplitude accuracy against total harmonic distortion in similar inductive sensing systems.","The measured drain efficiency of about 77% in a lossy-coil configuration indicates switch-mode MSF excitation is feasible for battery-powered or thermally constrained instruments."],"supporting_citations":[{"why":"Supplies the SHEPWM formulations and solving-algorithm taxonomy that this paper extends to multi-simultaneous-frequency output.","marker":"[8]"},{"why":"Provides the multilevel Class D inverter topology underlying the H-bridge amplifier used in the prototype.","marker":"[6]"},{"why":"Supplies the interior-point algorithm that solves the constrained optimization for the switching angles.","marker":"[15]"},{"why":"Establishes why multi-frequency eddy current testing needs a high-current multi-frequency excitation source, the application the design targets.","marker":"[5]"},{"why":"Presents the resultant-theory alternative for solving the harmonic-magnitude equations, against which the paper motivates its optimization approach.","marker":"[10]"}],"fun_headline_variants":["Pulse edges sculpt five eddy-current test tones from one Class D amp","Repurposed harmonic elimination sends five test tones through one coil","One switch-mode amp, five inspection frequencies via harmonic trick","Selective harmonics drive multi-frequency eddy current testing simply","Class D amp generates five simultaneous currents for eddy tests"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the transmitting coil behaves as a pure inductor, so the coil current is exactly the integral of the applied pulse voltage divided by L; the paper does not quantitatively test how much the coil's resistance and eddy-current loading distort the designed harmonic amplitudes.","fun_headline_variants_meta":{"raw":{"variants":["Pulse edges sculpt five eddy-current test tones from one Class D amp","Repurposed harmonic elimination sends five test tones through one coil","One switch-mode amp, five inspection frequencies via harmonic trick","Selective harmonics drive multi-frequency eddy current testing simply","Class D amp generates five simultaneous currents for eddy tests"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000208,"raw_usage":{"total_tokens":1433,"prompt_tokens":1002,"completion_tokens":431,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":618,"completion_tokens_details":{"reasoning_tokens":345}},"tokens_in":618,"tokens_out":431,"duration_ms":4981,"temperature":1.0,"reasoning_tokens":345,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:32:40.856455+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the transmitting coil's complex impedance at each transmitted frequency, recompute the harmonic currents from the optimized pulse voltage using that measured impedance (resistor in series with the inductance) instead of a pure inductor, and compare with the FFT of the measured coil current; a systematic mismatch at the higher frequencies would show that the pure-inductor assumption underlying the switching-angle design does not hold.","supporting_citations":[{"cited_title":"A Review of Multilevel Selective Harmonic Elimination PWM: Formulations, Solving Algorithms, Implementation and Applications,","cited_arxiv_id":null,"evidence_quote":"Supplies the SHEPWM formulations and solving-algorithm taxonomy that this paper extends to multi-simultaneous-frequency output."},{"cited_title":"Multilevel inverters: a survey of topologies, controls, and applications,","cited_arxiv_id":null,"evidence_quote":"Provides the multilevel Class D inverter topology underlying the H-bridge amplifier used in the prototype."},{"cited_title":"Nesterov and A","cited_arxiv_id":null,"evidence_quote":"Supplies the interior-point algorithm that solves the constrained optimization for the switching angles."},{"cited_title":"Non- Destructive Techniques Based on Eddy Current Testing,","cited_arxiv_id":null,"evidence_quote":"Establishes why multi-frequency eddy current testing needs a high-current multi-frequency excitation source, the application the design targets."},{"cited_title":"Control of a multilevel converter using resultant theory,","cited_arxiv_id":null,"evidence_quote":"Presents the resultant-theory alternative for solving the harmonic-magnitude equations, against which the paper motivates its optimization approach."}],"review_version":1}