pith. sign in

arxiv: 2605.23963 · v1 · pith:QJPB4D3Gnew · submitted 2026-05-12 · 💻 cs.DC · eess.SP

RASC: Region-Aware Self-Calibration for Dense 2D Sensor Arrays

Yinglei Ma , Fei Xiao This is my paper

Pith reviewed 2026-06-30 22:04 UTC · model grok-4.3

classification 💻 cs.DC eess.SP
keywords self-calibrationsensor arraysdistributed estimationtemperature sensorsfixed-pattern noiseconsensus algorithmsrobust estimation
0
0 comments X

The pith

RASC decomposes global sensor-array calibration into local clusters reconciled by linear consensus to restore factory accuracy with far less bandwidth.

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

The paper introduces RASC, a five-stage algorithm that addresses post-deployment drift in BJT-based 2D temperature sensor arrays, which cannot be recalibrated in a lab. It breaks the global ill-posed calibration task into independent local cluster problems, performs robust alternating estimation inside each cluster, and reconciles the results via linear consensus on the cluster-overlap graph. On 7632 frames from a real 16x16 array with five times the factory non-uniformity, the method reduces locally non-smooth fixed-pattern residual by 71 percent, restoring plus or minus 0.1 degree Celsius accuracy while changing the field by only 0.041 degree Celsius RMSE. Simulations on 8x8 to 32x32 arrays show it matches an oracle centralized extended Kalman filter within 0.10 degree Celsius yet uses roughly four times less bandwidth.

Core claim

RASC decomposes the global ill-posed calibration problem into local cluster-level problems, runs robust alternating estimation (trimmed-mean field reconstruction plus Huber IRLS) inside each cluster, and reconciles overlapping estimates by linear consensus on the cluster-overlap graph with provable exponential convergence, cutting the fixed-pattern residual by 71 plus or minus 5 percent on deployed 16x16 data while matching centralized performance at four times lower bandwidth.

What carries the argument

Linear consensus reconciliation on the cluster-overlap graph that combines local robust estimations performed by trimmed-mean field reconstruction and Huber IRLS.

If this is right

  • On real data the residual reduction reaches 78 percent at edges versus 55 percent in the interior.
  • The method matches centralized extended Kalman filter accuracy within 0.10 degree Celsius across simulated array sizes from 8x8 to 32x32.
  • Bandwidth drops by a factor of approximately four while preserving the calibrated field to 0.041 degree Celsius RMSE.
  • The approach restores factory plus or minus 0.1 degree Celsius accuracy on arrays that drifted by five times the specification.

Where Pith is reading between the lines

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

  • The same cluster-plus-consensus structure could be tested on other dense 2D sensor modalities such as pressure or optical arrays.
  • Exponential convergence of the consensus step suggests the method remains stable when cluster sizes or overlap densities vary across deployments.
  • Lower per-node communication may allow practical scaling beyond 32x32 without proportional growth in network load.
  • Edge-heavy improvement implies the method could be especially useful for arrays whose boundaries experience stronger mechanical stress.

Load-bearing premise

The global calibration problem can be split into independent local cluster problems whose solutions can be reconciled by linear consensus without significant loss of accuracy or added bias.

What would settle it

If consensus on a new 16x16 array produces a calibration whose RMSE against a centralized solution exceeds 0.10 degree Celsius on the same frames, the claim of matching oracle performance fails.

Figures

Figures reproduced from arXiv: 2605.23963 by Fei Xiao, Yinglei Ma.

Figure 1
Figure 1. Figure 1: Five-Stage RASC Pipeline. Stages 1–3 form a static graph structure after [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Communication-Accuracy Trade-off Across Array Scales. (a) Bytes per cali [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Field-reconstruction RMSE versus consensus iteration [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Theoretical bound ρth = 1 − αλ2(LC) versus empirically fitted rate ρemp. The bound is uniformly safe and becomes tighter with increasing array size; at 32 × 32, λ2 = 0.048 gives ρth = 0.998 while ρemp = 0.793. and 32% lower than the median spatial filter, operating only 0.10◦C above the irreducible noise floor [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Robustness heatmap of RASC under combined node failure (rows) and packet [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Sensitivity to (a) MAD threshold η, (b) consensus step α, (c) minimum cluster size Nmin, on the 16 × 16 array. Error bars show one standard deviation over 30 runs. 7.3 Online Recalibration Results We run RASC on the first 600 frames using a 5% reference subset (12 of 256 sensors) [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Experimental platform. (a) A 16 × 16 BJT-based temperature sensor array mounted on a circular fixture with engraved degree markings, wired to the host system via a ribbon cable. The four mounting posts at the perimeter are asymmetric thermal paths that contribute to the edge-localised drift shown in Section 7.2. (b) Acquisition workstation [PITH_FULL_IMAGE:figures/full_fig_p015_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Real temperature data characterisation. (a) Frame-mean trajectory and five [PITH_FULL_IMAGE:figures/full_fig_p016_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Online recalibration results on the deployed BJT array (representative fold of [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Stress-test calibration trajectory. (a) Frame-mean field versus time; the RASC [PITH_FULL_IMAGE:figures/full_fig_p018_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Stress-test spatial mean maps for 600 frames. From left: factory [PITH_FULL_IMAGE:figures/full_fig_p019_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Stress-test sensor-wise residual distributions. (a) Log-scale comparison: the [PITH_FULL_IMAGE:figures/full_fig_p019_12.png] view at source ↗
read the original abstract

BJT-based 2D temperature-sensor arrays are factory-calibrated to +/-0.1 degC, but post-deployment thermal and mechanical stresses drift their per-sensor gain-offset parameters by an order of magnitude, and in-lab recalibration is impractical. We present RASC (Region-Aware Self-Calibration), a five-stage algorithm that decomposes the global ill-posed problem into local cluster-level problems, runs robust alternating estimation (trimmed-mean field reconstruction + Huber IRLS) inside each cluster, and reconciles overlapping estimates by linear consensus on the cluster-overlap graph with provable exponential convergence. On 7,632 frames from a deployed 16x16 array exhibiting ~5x factory-spec non-uniformity, RASC cuts the locally-non-smooth fixed-pattern residual by 71+/-5% (10-fold CV), restoring +/-0.1 degC accuracy while perturbing the calibrated field by only 0.041 degC RMSE; reduction concentrates at the edges (78% vs 55% interior). In simulations on 8x8 to 32x32 arrays, RASC matches an oracle centralized EKF within 0.10 degC with ~4x lower bandwidth.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 1 minor

Summary. The manuscript presents RASC, a five-stage algorithm for post-deployment self-calibration of BJT-based 2D temperature sensor arrays. It decomposes the globally ill-posed gain-offset problem into independent per-cluster alternating estimations (trimmed-mean field reconstruction + Huber IRLS), then reconciles overlapping solutions via linear consensus on the cluster-overlap graph, with a claimed proof of exponential convergence. On 7,632 frames from a real 16x16 array, it reports a 71+/-5% reduction in locally non-smooth fixed-pattern residual (10-fold CV) while perturbing the field by 0.041 degC RMSE; simulations on 8x8-32x32 arrays claim matching an oracle centralized EKF within 0.10 degC at ~4x lower bandwidth.

Significance. If the local-to-global reconciliation is shown to be unbiased and equivalent to centralized solutions, the approach would offer a practical, communication-efficient method for calibrating large deployed sensor arrays without in-lab access, directly addressing drift in factory-calibrated systems.

major comments (2)
  1. [Abstract] Abstract (paragraph describing the five-stage algorithm): the central assumption that independent local cluster problems can be reconciled by linear consensus on the overlap graph without significant bias or loss of accuracy relative to a centralized solver is not supported by any derivation or fixed-point analysis; each local problem remains under-determined, and the reported aggregate RMSE match to EKF does not test parameter recovery or boundary consistency.
  2. [Abstract] Abstract (simulation results): the claim that RASC 'matches an oracle centralized EKF within 0.10 degC' is reported only in field RMSE; without accompanying per-sensor parameter error or cross-boundary consistency metrics, it does not confirm that the consensus step recovers the same solution as the centralized method rather than a smoothed approximation.
minor comments (1)
  1. [Abstract] The abstract reports quantitative improvements but provides no reference to the specific section or equation containing the convergence proof or the definition of the non-smoothness metric used for the 71% reduction.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on the abstract's claims regarding the consensus reconciliation and simulation evidence. We address each point below and indicate where revisions will be made.

read point-by-point responses

  1. Referee: [Abstract] Abstract (paragraph describing the five-stage algorithm): the central assumption that independent local cluster problems can be reconciled by linear consensus on the overlap graph without significant bias or loss of accuracy relative to a centralized solver is not supported by any derivation or fixed-point analysis; each local problem remains under-determined, and the reported aggregate RMSE match to EKF does not test parameter recovery or boundary consistency.

    Authors: Section 4.3 of the manuscript contains the fixed-point analysis and proof of exponential convergence for the linear consensus on the overlap graph. The local cluster problems, while initially under-determined, are regularized via trimmed-mean reconstruction and Huber IRLS; the overlap-graph consensus then enforces global consistency. The simulation match to the centralized EKF on field values supports that bias remains negligible, though we agree that additional explicit parameter-recovery and boundary-consistency metrics would strengthen the presentation. We will add a concise reference to the convergence analysis in the abstract. revision: partial

  2. Referee: [Abstract] Abstract (simulation results): the claim that RASC 'matches an oracle centralized EKF within 0.10 degC' is reported only in field RMSE; without accompanying per-sensor parameter error or cross-boundary consistency metrics, it does not confirm that the consensus step recovers the same solution as the centralized method rather than a smoothed approximation.

    Authors: The primary evaluation metric is field RMSE because that is the quantity of interest for deployed calibration accuracy. The manuscript already demonstrates cross-boundary consistency through the overlap-graph construction and reports that the consensus solution matches the EKF within the stated tolerance. To directly address the concern, we will include per-sensor gain/offset error statistics and boundary-consistency plots in the revised simulation results section. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is algorithmic and empirically validated

full rationale

The paper describes a five-stage decomposition algorithm with local robust estimation followed by linear consensus on an overlap graph, claiming provable convergence and reporting empirical reductions on external sensor data and simulations. No step equates a reported performance metric or 'prediction' to a fitted input by construction, nor relies on self-citation chains for load-bearing uniqueness or ansatzes. The central claims rest on the algorithm's design and measured outcomes against independent references, making the derivation self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no explicit free parameters, axioms, or invented entities; full manuscript would be required to audit modeling assumptions such as cluster definition rules or convergence conditions.

pith-pipeline@v0.9.1-grok · 5743 in / 1361 out tokens · 27409 ms · 2026-06-30T22:04:35.944707+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

27 extracted references · 18 canonical work pages

  1. [1]

    A CMOS temperature sensor with a smart calibrated inaccuracy of±0.11 ◦C (3σ).Sensors2023,23, 5132.https: //doi.org/10.3390/s23115132

    Wei, R.; Lin, H.; Chen, Q.; Huang, G.; Hu, W. A CMOS temperature sensor with a smart calibrated inaccuracy of±0.11 ◦C (3σ).Sensors2023,23, 5132.https: //doi.org/10.3390/s23115132

    work page doi:10.3390/s23115132

  2. [2]

    An energy-efficient BJT- based temperature sensor with±0.8◦C (3σ) inaccuracy from−50to150 ◦C.Sensors 2022,22, 9381.https://doi.org/10.3390/s22239381

    Qin, C.; Huang, Z.; Liu, Y.; Li, J.; Lin, L.; Tan, N.; Yu, X. An energy-efficient BJT- based temperature sensor with±0.8◦C (3σ) inaccuracy from−50to150 ◦C.Sensors 2022,22, 9381.https://doi.org/10.3390/s22239381

    work page doi:10.3390/s22239381 2022

  3. [3]

    Nonuniformity two-point linear correction errors in 18 Figure 11: Stress-test spatial mean maps for 600 frames

    Friedenberg, A.; Goldblatt, I. Nonuniformity two-point linear correction errors in 18 Figure 11: Stress-test spatial mean maps for 600 frames. From left: factoryYreal, cor- ruptedY corr, RASC-recalibratedˆx,|error|ofY corr −Y real,|error|ofˆx−Y real. The corrected map recovers a smooth thermal scene structure; the residual is less than0.3◦C almost everywh...

    1998

  4. [4]

    An algebraic algorithm for nonuniformity correction in focal-plane arrays.J

    Ratliff, B.M.; Hayat, M.M.; Hardie, R.C. An algebraic algorithm for nonuniformity correction in focal-plane arrays.J. Opt. Soc. Am. A2002,19, 1737–1747.https: //doi.org/10.1364/JOSAA.19.001737

    work page doi:10.1364/josaa.19.001737

  5. [5]

    A collaborative approach to in-place sensor calibration

    Bychkovskiy, V.; Megerian, S.; Estrin, D.; Potkonjak, M. A collaborative approach to in-place sensor calibration. InProc. IPSN 2003; LNCS vol. 2634; Springer: Berlin, 2003; pp. 301–316.https://doi.org/10.1007/3-540-36978-3_20

    work page doi:10.1007/3-540-36978-3_20 2003

  6. [6]

    A deep learning 19 approach for blind drift calibration of sensor networks.IEEE Sens

    Wang, Y.; Yang, A.; Chen, X.; Wang, P.; Wang, Y.; Yang, H. A deep learning 19 approach for blind drift calibration of sensor networks.IEEE Sens. J.2017,17, 4158–4171.https://doi.org/10.1109/JSEN.2017.2703885

    work page doi:10.1109/jsen.2017.2703885 2017

  7. [7]

    Spatial-temporal graph attention fuser for calibration in IoT air pollution monitoring systems

    Faghih Niresi, K.; Zhao, M.; Bissig, H.; Baumann, H.; Fink, O. Spatial-temporal graph attention fuser for calibration in IoT air pollution monitoring systems. InProc. IEEE SENSORS, Vienna, 2023; pp. 1–4.https://doi.org/10.1109/ SENSORS56945.2023.10325090

    work page arXiv 2023

  8. [8]

    A survey on sensor calibration in air pollution monitoring deployments.IEEE Internet Things J.2018,5, 4857–4870.https: //doi.org/10.1109/JIOT.2018.2853660

    Maag, B.; Zhou, Z.; Thiele, L. A survey on sensor calibration in air pollution monitoring deployments.IEEE Internet Things J.2018,5, 4857–4870.https: //doi.org/10.1109/JIOT.2018.2853660

    work page doi:10.1109/jiot.2018.2853660 2018

  9. [9]

    Consensus problems in networks of agents with switching topology and time-delays.IEEE Trans

    Olfati-Saber, R.; Murray, R.M. Consensus problems in networks of agents with switching topology and time-delays.IEEE Trans. Autom. Control2004,49, 1520– 1533.https://doi.org/10.1109/TAC.2004.834113

    work page doi:10.1109/tac.2004.834113 2004

  10. [10]

    History of infrared detectors.Opto-Electron

    Rogalski, A. History of infrared detectors.Opto-Electron. Rev.2012,20, 279–308. https://doi.org/10.2478/s11772-012-0037-7

    work page doi:10.2478/s11772-012-0037-7 2012

  11. [11]

    Shutterless fixed pattern noise correction for infrared imaging arrays

    Narendra, P.M.; Foss, N.A. Shutterless fixed pattern noise correction for infrared imaging arrays. InTechnical Issues in Focal Plane Development, Proc. SPIE vol. 282, 1981; pp. 44–51

    1981

  12. [12]

    The Vera C. Rubin Observatory Data Butler and pipeline execution system,

    Perry, D.L.; Dereniak, E.L. Linear theory of nonuniformity correction in infrared staring sensors.Opt. Eng.1993,32, 1854–1859.https://doi.org/10.1117/12. 145601

    work page doi:10.1117/12 1993

  13. [13]

    Nonuniformity correction of infrared image sequences using the constant-statistics constraint.IEEE Trans

    Harris, J.G.; Chiang, Y.-M. Nonuniformity correction of infrared image sequences using the constant-statistics constraint.IEEE Trans. Image Process.1999,8, 1148– 1151.https://doi.org/10.1109/83.777098

    work page doi:10.1109/83.777098 1999

  14. [14]

    Scene-based nonuni- formitycorrectionmethodusingprincipalcomponentanalysisforinfraredfocalplane arrays.Appl

    Lu, D.; Teng, L.; Ren, J.; Tan, J.; Wang, M.; Wang, L.; Gu, G. Scene-based nonuni- formitycorrectionmethodusingprincipalcomponentanalysisforinfraredfocalplane arrays.Appl. Sci.2023,13, 13331.https://doi.org/10.3390/app132413331

    work page doi:10.3390/app132413331 2023

  15. [15]

    A median-ratio scene-based non-uniformity correction method for airborne infrared point target detection system.Sensors2020,20, 3273

    Ding, S.; Wang, D.; Zhang, T. A median-ratio scene-based non-uniformity correction method for airborne infrared point target detection system.Sensors2020,20, 3273. https://doi.org/10.3390/s20113273

    work page doi:10.3390/s20113273

  16. [16]

    Calibration as parameter estimation in sensor networks

    Whitehouse, K.; Culler, D. Calibration as parameter estimation in sensor networks. InProc. 1st ACM WSNA, Atlanta, 2002; pp. 59–67.https://doi.org/10.1145/ 570738.570747 20

    work page arXiv 2002

  17. [17]

    Drift aware wireless sensor networks

    Takruri, M.; Challa, S. Drift aware wireless sensor networks. InProc. 10th Int. Conf. Information Fusion, Quebec City, 2007; pp. 1–7

    2007

  18. [18]

    Ahmad, R. Enhanced drift self-calibration of low-cost sensor networks based on cluster and advanced statistical tools.Measurement2024,236, 115158.https:// doi.org/10.1016/j.measurement.2024.115158

    work page doi:10.1016/j.measurement.2024.115158 2024

  19. [19]

    Dynamic calibration of low-cost PM2.5 sensors using trust- based consensus mechanisms.npj Clim

    Mahajan, S.; Helbing, D. Dynamic calibration of low-cost PM2.5 sensors using trust- based consensus mechanisms.npj Clim. Atmos. Sci.2025,8, 257.https://doi. org/10.1038/s41612-025-01145-2

    work page doi:10.1038/s41612-025-01145-2 2025

  20. [20]

    Huber, P.J.; Ronchetti, E.M.Robust Statistics, 2nd ed.; Wiley: Hoboken, NJ, USA, 2009

    2009

  21. [21]

    Hampel, F.R.; Ronchetti, E.M.; Rousseeuw, P.J.; Stahel, W.A.Robust Statistics: The Approach Based on Influence Functions; Wiley: New York, NY, USA, 1986

    1986

  22. [22]

    Anomaly Detection: A Survey,

    Chandola, V.; Banerjee, A.; Kumar, V. Anomaly detection: a survey.ACM Comput. Surv.2009,41, Article 15.https://doi.org/10.1145/1541880.1541882

    work page doi:10.1145/1541880.1541882 2009

  23. [23]

    Fast linear iterations for distributed averaging.Syst

    Xiao, L.; Boyd, S. Fast linear iterations for distributed averaging.Syst. Control Lett. 2004,53, 65–78.https://doi.org/10.1016/j.sysconle.2004.02.022

    work page doi:10.1016/j.sysconle.2004.02.022 2004

  24. [24]

    Topological network features determine convergence rate of distributed average algorithms.Sci

    Sirocchi, C.; Bogliolo, A. Topological network features determine convergence rate of distributed average algorithms.Sci. Rep.2022,12, 21831.https://doi.org/10. 1038/s41598-022-25974-w

    2022

  25. [25]

    Maximum likelihood from incomplete data via the EM algorithm.J

    Dempster, A.P.; Laird, N.M.; Rubin, D.B. Maximum likelihood from incomplete data via the EM algorithm.J. R. Stat. Soc. Ser. B1977,39, 1–38

  26. [26]

    92; AMS: Providence, RI, 1997

    Chung, F.R.K.Spectral Graph Theory; CBMS vol. 92; AMS: Providence, RI, 1997

    1997

  27. [27]

    Algebraic connectivity of graphs.Czechoslov

    Fiedler, M. Algebraic connectivity of graphs.Czechoslov. Math. J.1973,23, 298– 305. A Proof of Theorem 1 Fix a clustercwithk=|M c|members and reference subsetR c =M c ∩ R. The Huber IRLS objective at iterationemis: L(em) = X j∈Mc\Rc TX t=1 ψc yj(t)−a j ˆx(em) c (t)−b j s(em) j . 21 In each iteration, the E-step producesˆx(em+1) c from current parameter es...

    1973