{"id":"803f2493-9a36-49fe-a4cb-e65861dd28e9","arxiv_id":"2412.11882","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A hardware-in-the-loop testbed with a square Helmholtz coil and convex-combination current control generates uniform, controllable magnetic fields for lab testing of geomagnetic navigation.","lead":"A team built a low-cost laboratory testbed that uses a Helmholtz coil to create adjustable magnetic fields for testing geomagnetic navigation systems. The testbed combines a physical coil with a software simulation and a new control method, aiming to replace expensive ocean trials.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The unshielded-environment claim rests on synthetic Gaussian noise; real ambient disturbance rejection is never demonstrated.","rationale":"The reader's weakest assumption correctly identifies that the unshielded validation relies on synthetic Gaussian noise. My stress-test agrees and sharpens the concern: the physical tracking metrics most central to the headline claim were all produced in a shielded room (Sec. 4.3), while the unshielded experiment (Sec. 4.4) is a simulation-style test with injected noise, not a physical demonstration against real ambient interference. This matters because the entire motivation for the unshielded design is to avoid mu-metal shielding, and the control's convergence/stability argument explicitly assumes zero-mean Gaussian white noise and independence between noise and input (Sec. 3.3). Real laboratory disturbances are typically colored, non-stationary, and spatially varying, so the reported experiment does not establish the load-bearing claim. I do not see the concern as fatal: the hardware is built, the FEM/physical coil alignment within 3% is a genuine result, and the shielded-room tracking data are plausible engineering measurements. The paper also honestly acknowledges that no end-to-end geomagnetic navigation experiment is run. Therefore the appropriate disposition remains CONDITIONAL, requiring a real-disturbance validation before the unshielded testbed claim can be accepted. No change to the reader's verdict is needed.","tokens_in":21521,"tokens_out":4027,"duration_ms":38204,"concrete_test":"Record the actual output x(n) of the outside-coil magnetometer in the unshielded lab for T >= 60 s (including normal lab activity) and replay this measured disturbance into the control loop with the same 0-to-120,000 nT target step used in Sec. 4.3. If the converged RMSE or settling time degrades materially relative to the shielded-room result, or if a non-Gaussian transient causes divergence, the unshielded claim fails. A stronger variant: run the physical coil in the unshielded lab with no synthetic noise and report RMSE, settling time, and re-measured uniformity under real ambient conditions.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central differentiator is that the testbed works in an unshielded lab, compensating environmental interference without a mu-metal room (Sec. 1, Sec. 4.4). But the only unshielded experiments (Sec. 4.4, Table 6) do not use real ambient magnetic disturbances. They inject zero-mean Gaussian white noise x(n)~N(0,1), v(n)~N(0,1) at SNR=10/30 dB, exactly the noise model assumed in the convergence/stability analysis (Eqs. 28, 47-52). Real lab interference is non-stationary and colored (mains hum, moving ferromagnetic objects, gradients across the coil volume); the control's proof for Gaussian independent noise does not cover it. Meanwhile the quantitative physical tracking results (0.55 s, RMSE ~270 nT) were measured inside a shielded room (Sec. 4.3). Thus the load-bearing claim that the testbed can 'rapidly, accurately, and stably generate' fields in an unshielded environment is not supported by the reported experiments.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper describes a hardware-in-the-loop simulation testbed for geomagnetic navigation, consisting of a square Helmholtz coil driven by a voltage-controlled current source, a microcontroller, magnetometers, and a finite-element-method digital twin of the coil. The authors derive an optimal coil spacing ratio (n=1.8365) from a uniformity optimization, propose a convex-combination of two LMS-type controllers with weight transfer, prove convergence and stability under Gaussian noise assumptions, and report experimental characterization: roughly 5% field uniformity over about 467 mm, FEM-to-physical agreement within about 3%, step tracking of a 120,000 nT target in 0.43-0.55 s with RMSE near 270 nT in a shielded room, and comparisons against LMS, SVS, and ATLMS under synthetic noise at 10/30 dB in an unshielded laboratory.","tokens_in":21719,"tokens_out":8108,"duration_ms":69925,"significance":"If the unshielded-operation claim were fully supported, the testbed would be a useful, low-cost, repeatable platform for geomagnetic navigation experiments. The measured uniformity, the FEM-physical alignment, the detailed component specifications, and the comparison against existing control methods are concrete strengths, and the coil parameter optimization is derived rather than fitted. The main conceptual contribution is the complete system integration in an unshielded lab, but that contribution currently rests on synthetic Gaussian noise rather than real ambient disturbance rejection, and the only physical-unit tracking results were obtained inside a shielded room. The convergence proof also contains a dimensional inconsistency that needs correction before it can support the stability claims.","major_comments":[{"comment":"The unshielded-environment validation uses only synthetic zero-mean Gaussian white noise x(n)~N(0,1) and v(n)~N(0,1) at SNR 10/30 dB. This is exactly the noise model assumed in the convergence and stability analyses (Section 3.2-3.3, Eqs. (28) and (47)-(52)), so the experiment cannot demonstrate rejection of real unshielded-laboratory interference, which is non-stationary, colored, and spatially correlated across the coil volume. Since the paper's stated differentiator is operation without a shielded room (Sections 1 and 5), the load-bearing claim that the testbed can 'rapidly, accurately, and stably generate the magnetic field' in an unshielded environment is not supported by the reported data. I recommend measuring and reporting the actual ambient disturbance time series in the laboratory and repeating the field-generation experiment under those real disturbances, with results in physical units (nT).","section":"Section 4.4, Table 6"},{"comment":"The quantitative field-generation results in physical units (reach times of 0.43 s and 0.55 s, RMSE 269.36 nT and 272.66 nT) were obtained in an electromagnetic shielding room, not in the unshielded environment. The unshielded experiment in Section 4.4 reports only dimensionless MSE values on synthetic noise. The abstract and conclusions present unshielded operation as an achieved property, but as written the physical-accuracy evidence pertains only to the shielded configuration. Please either add equivalent physical-unit tracking results under real unshielded conditions or temper the unshielded claim accordingly.","section":"Section 4.3, Table 5"},{"comment":"The convergence proof contains a dimensional inconsistency. For a vector regressor x(n), xT(n)x(n) is a scalar, yet Eqs. (40)-(45) treat it as a matrix Rxx = E[xT(n)x(n)] with eigen-decomposition QT Lambda Q, and the weight-error recursion should involve the outer product x(n)xT(n), not the scalar xT(n)x(n). The displayed derivation therefore does not establish the stated condition (46), and the proof needs to be rewritten with the correct matrix convention (e.g., Rxx = E[x(n)xT(n)]) or with an explicitly scalar signal model throughout. Because the convergence guarantee is one of the paper's stated contributions, this is a load-bearing issue.","section":"Section 3.2, Eqs. (40)-(46)"}],"minor_comments":[{"comment":"The sentence following Eq. (21) states both BQ(z)=BQ(-z) and that BQ(z) is an odd function; these statements are mutually contradictory, and the intended argument is that the field is even in z so that odd-order Taylor terms vanish. Please correct this.","section":"Section 2.2"},{"comment":"The symbol B is used both for the target magnetic field strength and for the affine calibration output B = kx(n) + b; since the fitted calibration in Fig. 9 is B = kU + b with U the control voltage, the relationship between these uses should be clarified to avoid confusing the control variable with the target field.","section":"Section 3.1.1, Eqs. (24)-(28)"},{"comment":"The table header 'D/H' and the entries +/-x/d and +/-y/d are not explained in the caption; please define D, H, and how the 467 mm x 467 mm uniform area is obtained from the ratio values.","section":"Table 3"},{"comment":"The MSE curves in Fig. 13 lack axis labels and units; please specify the normalization of MSE and whether the iteration index corresponds to wall-clock time or to controller samples.","section":"Fig. 13"},{"comment":"There are several typos and grammatical slips (e.g., 'calcualted' in Section 3.1.1, 'discrepancies is confined' in Section 4.2, 'feild' in Section 5, and 'geographic navigation' where 'geomagnetic navigation' seems intended in Section 5), which should be corrected in a final pass.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The paper is a systems/testbed contribution that fits the journal's readership. The central unshielded-operation claim is the main selling point, and the current evidence for it is synthetic; if the authors cannot add real-disturbance experiments, the paper's scope and conclusions should be narrowed accordingly. The convergence proof needs a genuine fix, not just copy-editing. The citation pattern includes several self-citations, but they are not inappropriate given the authors' prior work on adaptive filtering and navigation."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper is a competent engineering integration of known building blocks: a square Helmholtz coil with textbook-optimal spacing, a convex-combination adaptive filter from Arenas-Garcia et al., and an FEM digital twin. What is genuinely new is putting these together into an unshielded, off-the-shelf hardware-in-the-loop testbed for geomagnetic field generation. The measured uniformity (5% over ~467 mm, <0.5% over 80 mm), the 3% FEM-physical coil alignment, and the shielded-room tracking results (0.43–0.55 s step response, RMSE ~270 nT) are concrete and credible. If you need a low-cost lab source of controllable magnetic fields, this is a useful reference.\n\nThe soft spots are where the claims outrun the experiments. The paper's stated differentiator is unshielded operation, but the only unshielded validation (Sec. 4.4, Table 6) injects zero-mean Gaussian white noise at 10/30 dB SNR—exactly the noise model used in the convergence proofs. That does not demonstrate rejection of real ambient interference: mains hum, moving ferromagnetic objects, and spatial gradients are non-Gaussian, colored, and non-stationary. The physical tracking numbers were measured inside a shielded room (Sec. 4.3). So the conclusion's claim that the testbed \"can rapidly, accurately, and stably generate\" fields in an unshielded environment is not supported by the reported experiments. This is the load-bearing issue, and it is addressable: either measure real ambient disturbances during unshielded tracking, or re-scope the claim to robustness against Gaussian sensor noise.\n\nTwo smaller points. The stability proof in Sec. 3.3 is a sketch rather than a proof: it derives expectations under the Gaussian independence assumption and then asserts stability, without proving boundedness or convergence of the weights. That is acceptable in an engineering paper if labeled as a heuristic, but it should not be called a proof. The control hyperparameters are said to be \"set to optimal values\" with no tuning protocol, which hurts reproducibility. Finally, no actual geomagnetic navigation experiment is run—the testbed is validated for field generation only. That is a reasonable scoping, but it makes the title's \"for Geomagnetic Navigation\" a promissory note.\n\nThe citation pattern is fine; the convex-combination method is properly credited, and the self-citations are on-topic. Overall, the hardware validation is solid and the unshielded claim is not. I would send this to peer review because the testbed itself is useful and the data are honest, but the authors need to fix the unshielded validation before it is published. If that gets addressed, it becomes a genuinely useful contribution for labs that cannot afford a shielded room.\n\nRecommendation: engage, but require the fix.","headline":"A credible low-cost coil testbed with real measured uniformity, but the unshielded-environment claim rests on synthetic Gaussian noise, not actual ambient disturbances.","tokens_in":22270,"tokens_out":3032,"would_cite":false,"duration_ms":27632,"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":"A lab testbed of off-the-shelf parts generates geomagnetic fields accurate enough for navigation experiments, matching a virtual coil within 3 percent and tracking a 120 microTesla step in about half a second.","keywords":["geomagnetic navigation","hardware-in-the-loop simulation","square Helmholtz coil","magnetic field uniformity","convex combination control","adaptive filtering","unshielded test environment","digital twin"],"falsifier":"Record a trace of real magnetic disturbance in an ordinary laboratory — for example from nearby power equipment, moving ferrous objects, or passing people — and replay it as the disturbance while the coil tracks a 120,000-nT step; if the generated field does not remain close to the reported 0.55 s convergence time and roughly 270 nT RMSE, the claim that the testbed works in an unshielded environment fails.","tokens_in":21300,"feed_emoji":"🧲","tokens_out":7385,"duration_ms":68009,"temperature":0.7,"pith_summary":"Geomagnetic navigation promises GPS-free positioning by reading the Earth's field, but its algorithms have mostly been tested in simulation because real trials, such as ocean dives, are expensive and hard to repeat. This paper argues that a laboratory testbed built from ordinary off-the-shelf hardware can synthesize the magnetic environments those algorithms need, with enough accuracy, uniformity, and speed to stand in for real missions. The authors construct a square Helmholtz coil, optimize its geometry for field uniformity, and control it with a convex combination of two adaptive loops that balances fast response against precision. Their measurements show a field uniform to 5% over roughly 467 mm by 467 mm, agreement with the simulated virtual coil within 3%, and tracking of a 120,000-nT step in about 0.55 seconds with an RMSE near 270 nT. If these claims hold, the testbed offers a low-cost, repeatable way to validate geomagnetic navigation before committing to field tests.","feed_headline":"Off-the-shelf testbed synthesizes geomagnetic fields for nav tests","feed_subtitle":"Convex-combination control tracks a 120,000-nT step in 0.55 s with 3% FEM agreement, enabling cheap repeatable lab experiments","key_machinery":"The load-bearing mechanism is the convex combination coil control, which merges two adaptive filters — one slow with a dynamic learning rate for accuracy, one fast with a fixed learning rate for quick convergence — via a logistic coupling coefficient $\\gamma(n)$, and periodically transfers weights from the faster loop to the slower one whenever the coupling exceeds a threshold. The coil geometry is optimized separately by choosing the Helmholtz spacing-to-side ratio $n = L/d = 1.8365$, which nulls the second derivative of the axial field at the center and maximizes the uniform region. Together these pieces let the testbed claim simultaneous accuracy, stability, and speed without a shielded room.","core_discovery":"The paper's central claim is that a hardware-in-the-loop testbed can generate, in a normal unshielded laboratory, magnetic fields that meet the requirements of geomagnetic navigation experiments: a uniform field region large enough to hold a small underwater vehicle, stable and accurate tracking of dynamic target fields, and rapid convergence when the target changes. The field is produced by a square Helmholtz coil whose winding geometry is optimized so that the Taylor expansion of the axial field cancels the second-order term, giving a spacing-to-side ratio of $n = 1.8365$. The physical coil is aligned with a finite-element model of a virtual coil, and a convex-combination controller splits the job between a slow, precise adaptive loop and a fast, coarse loop, with a logistic coupling factor and periodic weight transfer to keep the two loops coordinated. Empirical results show 5% uniformity over a 467 mm square, a maximum 3% deviation from the FEM virtual field, and step-tracking of 120,000 nT in 0.43--0.55 s with steady-state RMSE around 270 nT, alongside convergence and stability proofs for the control law.","pith_inferences":["If the central claim holds, the testbed could be turned into a robustness benchmark by injecting controlled spatially non-uniform or time-correlated disturbances, an extension the paper does not perform.","The convex-combination idea naturally extends to three-axis field generation by synchronizing three coil pairs, which the authors name as future work and would broaden the testbed to full three-dimensional navigation studies.","The 3% agreement between the physical and FEM virtual coil suggests that simulation-based validation of navigation algorithms gains credibility when the same digital model drives the experiment, a link the authors establish but do not yet exploit for algorithm testing.","A cost comparison against shielded rooms and real sea trials would likely show large savings, making geomagnetic navigation experimentation accessible to labs without specialized facilities."],"forward_implications":["Geomagnetic navigation algorithms that currently exist only in simulation — contour matching, closest-contour iteration, evolutionary and learning-based methods — can be run repeatedly against a controllable, repeatable geomagnetic environment before costly field trials.","The 467 mm by 467 mm region with 5% uniformity is large enough for small autonomous underwater vehicles, enabling end-to-end hardware-in-the-loop navigation tests with real sensors.","The coil parameter optimization provides a scaling recipe, so labs can re-derive the coil geometry for carriers of different sizes.","The convergence and stability proofs apply to any square Helmholtz coil driven by a voltage-controlled current source, so the control design is portable to other testbeds.","The noise-compensation experiment supports the paper's central cost claim: usable field generation without a mu-metal shielded room."],"supporting_citations":[{"why":"Supplies the Biot-Savart field model and square-coil uniformity analysis that the paper's coil design and optimization build on.","marker":"[23]"},{"why":"Establishes the relationship between coil parameters and field uniformity, the relation the paper's parameter optimization extends.","marker":"[35]"},{"why":"Is the theoretical basis for the convex combination of two adaptive filters used as the coil controller.","marker":"[49]"},{"why":"Provides the thirteenth-generation International Geomagnetic Reference Field used as the target model for the generated field.","marker":"[50]"},{"why":"Supplies the least-mean-squares convergence condition used in the paper's stability and convergence proofs.","marker":"[53]"}],"fun_headline_variants":["Lab testbed synthesizes Earth's field for low-cost nav experiments","Track 120,000 nT steps in 0.55 s with lab testbed","Unshielded lab testbed mimics geomagnetic field for navigation","Hardware-in-the-loop testbed enables repeatable geomagnetic nav tests","Low-cost coil array generates accurate fields for geomagnetic nav"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The unshielded validation assumes that additive zero-mean Gaussian white noise on the sensor and target signals faithfully represents the real magnetic disturbances of a normal laboratory, so if actual disturbances are non-Gaussian, non-stationary, or spatially correlated inside the coil volume, the experiment does not demonstrate that the control compensates genuine environmental interference.","fun_headline_variants_meta":{"raw":{"variants":["Lab testbed synthesizes Earth's field for low-cost nav experiments","Track 120,000 nT steps in 0.55 s with lab testbed","Unshielded lab testbed mimics geomagnetic field for navigation","Hardware-in-the-loop testbed enables repeatable geomagnetic nav tests","Low-cost coil array generates accurate fields for geomagnetic nav"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001477,"raw_usage":{"total_tokens":5988,"prompt_tokens":1048,"completion_tokens":4940,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":664,"completion_tokens_details":{"reasoning_tokens":4846}},"tokens_in":664,"tokens_out":4940,"duration_ms":30294,"temperature":1.0,"reasoning_tokens":4846,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T14:28:42.796225+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Record a trace of real magnetic disturbance in an ordinary laboratory — for example from nearby power equipment, moving ferrous objects, or passing people — and replay it as the disturbance while the coil tracks a 120,000-nT step; if the generated field does not remain close to the reported 0.55 s convergence time and roughly 270 nT RMSE, the claim that the testbed works in an unshielded environment fails.","supporting_citations":[{"cited_title":"Mean-square performance of a convex combination of two adaptive filters,","cited_arxiv_id":null,"evidence_quote":"Is the theoretical basis for the convex combination of two adaptive filters used as the coil controller."},{"cited_title":"Least mean square method for lvdt signal processing,","cited_arxiv_id":null,"evidence_quote":"Supplies the least-mean-squares convergence condition used in the paper's stability and convergence proofs."}],"review_version":1}