pith. sign in

arxiv: 2507.17284 · v2 · submitted 2025-07-23 · 📡 eess.SP

Energy-Efficient State Estimation with 1-Bit Sensing: A Bussgang-Kalman Framework for Internet of Things

Chaehyun Jung , TaeJun Ha , Hyeonuk Kim , Jeonghun Park This is my paper

Pith reviewed 2026-05-19 03:18 UTC · model grok-4.3

classification 📡 eess.SP
keywords state estimation1-bit quantizationKalman filterBussgang theoremIoT sensingquantized measurementsnonlinear filtering
0
0 comments X

The pith

A Bussgang-aided Kalman filter incorporates 1-bit quantization distortion directly into recursive state estimation for IoT devices.

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

The paper develops a method to perform accurate tracking of system states when sensors report only a single bit per measurement, which is common in power-limited IoT hardware that uses cheap low-resolution converters. Standard Kalman filters assume clean, high-precision observations and lose accuracy or diverge under such extreme quantization, but this approach models the quantization effect as an equivalent linear distortion using a statistical theorem and folds the distortion statistics into the filter's covariance and gain calculations. For cases where the full system model is known, the method supplies both a standard recursive filter and a lighter version that skips some matrix operations. When the model is only partly known or mismatched, a hybrid version adds adaptive dithering and recurrent neural units to compensate. The resulting estimates stay usable on nonlinear systems and real sensor traces even though each measurement conveys almost no amplitude information.

Core claim

For fully known system models, the Bussgang-aided Kalman Filter (BKF) explicitly incorporates quantization distortion into recursive estimation by deriving an equivalent linear model of the 1-bit quantizer and adjusting the prediction and update equations accordingly, together with a reduced-complexity variant (reduced-BKF) for computationally efficient implementation. For partially known models, the Bussgang-aided KalmanNet (BKNet) combines adaptive dithering with gated recurrent units to mitigate severe quantization effects and model mismatch. Experiments on the Lorenz attractor and the Michigan NCLT dataset, both under 1-bit front-end quantization, demonstrate accurate and robust state估计.

What carries the argument

Bussgang-aided Kalman Filter (BKF), which uses the Bussgang theorem to replace the nonlinear 1-bit quantizer with a linear gain and additive distortion term whose statistics are inserted into the Kalman covariance and gain recursions.

If this is right

  • State estimates remain consistent because the filter equations now treat quantization as a known additive noise source rather than ignoring it.
  • A reduced-complexity version achieves nearly the same accuracy while lowering the number of matrix operations per time step.
  • The same linearization principle extends to a learning-based architecture that handles both quantization and unknown dynamics.
  • Tracking performance holds on both chaotic nonlinear oscillators and real-world location data collected with low-resolution sensors.

Where Pith is reading between the lines

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

  • The same linearization step could be applied to other common sensor nonlinearities such as clipping or saturation to create analogous filters.
  • In large sensor networks the reduction in data volume per node could compound into substantial savings in transmission energy and channel usage.
  • One could examine whether the method stays stable when the quantization threshold itself changes with time or signal strength.

Load-bearing premise

The Bussgang theorem must supply a linear approximation of the 1-bit quantization effect that is accurate enough to be inserted into the Kalman update equations without breaking the recursion or producing inconsistent covariance estimates.

What would settle it

Generate Monte Carlo trajectories from a known linear system, quantize the observations to one bit, run the BKF, and check whether the filter's reported error covariance matches the empirical mean-squared error; a large or growing mismatch would show the approximation fails to preserve filter consistency.

Figures

Figures reproduced from arXiv: 2507.17284 by Chaehyun Jung, Hyeonuk Kim, Jeonghun Park, TaeJun Ha.

Figure 1
Figure 1. Figure 1: Overall system structures for (a) ideal and (b) 1-bit observation environments. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: (a) Overall architecture with (b) BKF and (c) BKNet. [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: State estimation results for the Lorenz attractor using 1-bit observations. [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The MSE performance comparison of BKF and rBKF with identical [PITH_FULL_IMAGE:figures/full_fig_p011_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: BKNet, identical noise heterogeneous noise experiments, the noise levels and ADC counts match those used in V-C, which describes the rBKF experiments. 1) Identical noise [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: State estimation results for the NCLT dataset using 1-bit observations. [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
read the original abstract

Accurate state estimation from heavily quantized measurements is a key challenge in resource-constrained Internet of Things (IoT) sensing and tracking, where battery-powered devices may employ low-resolution analog-to-digital converters (ADCs) to simplify sensor hardware and reduce the amount of data. Existing model-based and hybrid learning-based estimators, however, typically assume high-resolution observations and therefore degrade severely under 1-bit quantization. In this paper, we study nonlinear state estimation with 1-bit observations and develop a Bussgang-aided filtering framework for IoT sensing front-ends with 1-bit quantization. For fully known system models, we propose a Bussgang-aided Kalman Filter (BKF) that explicitly incorporates quantization distortion into recursive estimation, together with a reduced-complexity variant (reduced-BKF) for computationally efficient implementation. For partially known models, we further propose Bussgang-aided KalmanNet (BKNet), a model-based deep learning architecture that combines adaptive dithering with gated recurrent units (GRUs) to mitigate severe quantization effects and model mismatch. Experiments on the Lorenz attractor and the Michigan NCLT dataset, both under 1-bit front-end quantization, demonstrate accurate and robust state estimation under highly nonlinear dynamics, imperfect models, and extreme quantization. These results support the potential of the proposed framework for reliable state estimation in resource-constrained IoT sensing and tracking applications with low-resolution front-ends.

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 / 2 minor

Summary. The manuscript proposes a Bussgang-aided Kalman Filter (BKF) that incorporates 1-bit quantization distortion into the recursive state estimation equations for fully known system models, together with a reduced-complexity reduced-BKF variant. For partially known models it introduces Bussgang-aided KalmanNet (BKNet) that combines adaptive dithering with GRUs. Experiments on the Lorenz attractor and Michigan NCLT dataset under 1-bit front-end quantization are reported to demonstrate accurate estimation under nonlinear dynamics and model mismatch.

Significance. If the central Bussgang-Kalman recursion remains consistent, the framework offers a practical, low-complexity route to state estimation for battery-constrained IoT nodes that employ 1-bit ADCs. The explicit treatment of quantization noise via the Bussgang linearization and the provision of both model-based and hybrid learning variants are constructive contributions. The choice of standard benchmarks (Lorenz, NCLT) is appropriate, yet the absence of reported error metrics or covariance-consistency diagnostics limits the immediate impact.

major comments (2)
  1. [BKF derivation and update equations] The derivation of the BKF update (replacing the measurement equation with effective gain αH and inflated noise covariance) rests on the assumption that the Bussgang distortion d remains uncorrelated with the estimation error at every time step. Because the predictor computes the measurement from the previous posterior and the filter is closed-loop, the sign nonlinearity can create a feedback path that violates the white-noise assumption underlying the Riccati recursion. This correlation risk is load-bearing for the optimality and consistency claims and must be either proved or validated by comparing the filter-reported covariance against Monte-Carlo error statistics.
  2. [Experiments] The abstract states that experiments on the Lorenz attractor and NCLT dataset demonstrate accurate estimation, yet no quantitative error metrics, baseline comparisons (e.g., against standard EKF, quantized EKF, or particle filters), or covariance-consistency plots are supplied. Without these data it is impossible to judge whether the reported accuracy is meaningful or whether the reduced-BKF approximation preserves filter consistency.
minor comments (2)
  1. [Preliminaries] Notation for the Bussgang gain α and the effective measurement matrix should be introduced with an explicit equation reference when first used.
  2. [Reduced-BKF] The reduced-BKF section should clarify the precise approximation made to the gain and its computational saving relative to the full BKF.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which have helped us identify areas for improvement in the manuscript. We address each major comment point by point below and have revised the manuscript to incorporate the suggested enhancements.

read point-by-point responses

  1. Referee: [BKF derivation and update equations] The derivation of the BKF update (replacing the measurement equation with effective gain αH and inflated noise covariance) rests on the assumption that the Bussgang distortion d remains uncorrelated with the estimation error at every time step. Because the predictor computes the measurement from the previous posterior and the filter is closed-loop, the sign nonlinearity can create a feedback path that violates the white-noise assumption underlying the Riccati recursion. This correlation risk is load-bearing for the optimality and consistency claims and must be either proved or validated by comparing the filter-reported covariance against Monte-Carlo error statistics.

    Authors: We thank the referee for this insightful observation regarding the correlation assumptions in the BKF recursion. The Bussgang linearization is applied under the standard Gaussianity assumption for the input to the quantizer, which ensures uncorrelated distortion with the input at each step. We acknowledge that the closed-loop feedback through the sign nonlinearity in the recursive setting may introduce additional correlations not fully captured by the white-noise model. To address this rigorously, we will add Monte-Carlo validation experiments in the revised manuscript, comparing the filter-reported covariances against empirical error statistics over multiple independent runs for both the full BKF and reduced-BKF variants. This will provide empirical support for the consistency claims. revision: yes

  2. Referee: [Experiments] The abstract states that experiments on the Lorenz attractor and NCLT dataset demonstrate accurate estimation, yet no quantitative error metrics, baseline comparisons (e.g., against standard EKF, quantized EKF, or particle filters), or covariance-consistency plots are supplied. Without these data it is impossible to judge whether the reported accuracy is meaningful or whether the reduced-BKF approximation preserves filter consistency.

    Authors: We agree that quantitative metrics, baseline comparisons, and consistency diagnostics are essential to substantiate the performance claims. While the current manuscript focuses on qualitative demonstrations of tracking accuracy under 1-bit quantization, we will substantially expand the experimental section in the revision. This will include reporting RMSE values, direct comparisons against the standard EKF, a quantized EKF implementation, and particle filters, as well as covariance-consistency plots (e.g., normalized estimation error squared) for the Lorenz attractor and Michigan NCLT dataset under 1-bit observations. These additions will allow readers to better assess both accuracy and filter consistency, including for the reduced-BKF variant. revision: yes

Circularity Check

0 steps flagged

Bussgang-Kalman framework applies standard Bussgang linearization to Kalman recursion with no self-referential reduction or fitted-input predictions

full rationale

The derivation inserts the Bussgang decomposition y_q = α y + d (with d uncorrelated to y by theorem) into the standard Kalman measurement update, replacing H with αH and inflating R. This is a direct substitution of an external 1952 result into the Riccati equations; the paper does not define α from its own outputs, fit it to the target performance metric, or rely on self-citations for the uniqueness or validity of the recursion. Experiments on Lorenz and NCLT provide external empirical checks rather than internal consistency proofs that loop back to the same fitted values. Minor self-citation risk exists only in the usual sense of citing prior Bussgang applications, but it is not load-bearing for the central claim.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claims rest on standard Kalman filtering assumptions plus the applicability of the Bussgang theorem to model 1-bit quantization distortion; no free parameters or new entities are introduced in the abstract.

axioms (2)
  • domain assumption System dynamics and observation model are either fully known or partially known but amenable to neural adaptation.
    Invoked when proposing BKF for known models and BKNet for partial models.
  • domain assumption Bussgang theorem provides a usable linear gain approximation for the quantized measurement process.
    Central modeling step for incorporating quantization distortion into the filter recursion.

pith-pipeline@v0.9.0 · 5795 in / 1289 out tokens · 38520 ms · 2026-05-19T03:18:41.940029+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

32 extracted references · 32 canonical work pages · 1 internal anchor

  1. [1]

    A new approach to linear filtering and prediction problems,

    R. E. Kalman, “A new approach to linear filtering and prediction problems,” J. of Basic Engineering , vol. 82, no. 1, pp. 35–45, 1960

    work page 1960

  2. [2]

    Simon, Optimal state estimation: Kalman, H infinity, and nonlinear approaches

    D. Simon, Optimal state estimation: Kalman, H infinity, and nonlinear approaches. John Wiley & Sons, 2006

    work page 2006

  3. [3]

    The Kalman filter-Its recognition and development for aerospace applications,

    S. F. Schmidt, “The Kalman filter-Its recognition and development for aerospace applications,” J. of Guidance and Control , vol. 4, no. 1, pp. 4–7, 1981

    work page 1981

  4. [4]

    Durbin and S

    J. Durbin and S. Koopman, Time series analysis by state space methods. Oxford; New York: Oxford University Press, 2001

    work page 2001

  5. [5]

    The unscented Kalman filter for nonlinear estimation,

    E. Wan and R. Van Der Merwe, “The unscented Kalman filter for nonlinear estimation,” in Proc. of the IEEE Adaptive Systems for Signal Process., Commun., and Control Symp. , 2000, pp. 153–158

    work page 2000

  6. [6]

    AI-aided Kalman filters,

    N. Shlezinger, G. Revach, A. Ghosh, S. Chatterjee, S. Tang, T. Imbiriba, J. Dunik, O. Straka, P. Closas, and Y . C. Eldar, “AI-aided Kalman filters,” arXiv preprint arXiv:2410.12289 , 2024

    work page arXiv 2024

  7. [7]

    Recurrent Kalman networks: Factorized inference in high-dimensional deep feature spaces,

    P. Becker, H. Pandya, G. Gebhardt, C. Zhao, C. J. Taylor, and G. Neumann, “Recurrent Kalman networks: Factorized inference in high-dimensional deep feature spaces,” in Proceedings of the 36th International Conference on Machine Learning , ser. Proceedings of Machine Learning Research, vol. 97, 09–15 Jun 2019, pp. 544–552

    work page 2019

  8. [8]

    KalmanNet: Neural network aided Kalman filtering for partially known dynamics,

    G. Revach, N. Shlezinger, X. Ni, A. L. Escoriza, R. J. G. van Sloun, and Y . C. Eldar, “KalmanNet: Neural network aided Kalman filtering for partially known dynamics,” IEEE Trans. Signal Process. , vol. 70, pp. 1532–1547, 2022

    work page 2022

  9. [9]

    Split- KalmanNet: A robust model-based deep learning approach for state estimation,

    G. Choi, J. Park, N. Shlezinger, Y . C. Eldar, and N. Lee, “Split- KalmanNet: A robust model-based deep learning approach for state estimation,”IEEE Trans. Veh. Technol., vol. 72, no. 9, pp. 12 326–12 331, 2023

    work page 2023

  10. [10]

    Latent-KalmanNet: Learned Kalman filtering for tracking from high-dimensional signals,

    I. Buchnik, G. Revach, D. Steger, R. J. G. van Sloun, T. Routtenberg, and N. Shlezinger, “Latent-KalmanNet: Learned Kalman filtering for tracking from high-dimensional signals,” IEEE Trans. Signal Process. , vol. 72, pp. 352–367, 2024

    work page 2024

  11. [11]

    Cholesky-KalmanNet: Model-based deep learning with positive definite error covariance structure,

    M. Ko and A. Shafieezadeh, “Cholesky-KalmanNet: Model-based deep learning with positive definite error covariance structure,” IEEE Trans. Signal Process., vol. 32, pp. 326–330, 2025

    work page 2025

  12. [12]

    Kalmanformer: Using transformer to model the Kalman gain in Kalman filters,

    S. Shen, J. Chen, G. Yu, Z. Zhai, and P. Han, “Kalmanformer: Using transformer to model the Kalman gain in Kalman filters,” Front. in Neurorobotics, vol. 18, p. 1460255, 2025

    work page 2025

  13. [13]

    Ex- plainable gated Bayesian recurrent neural network for non-Markov state estimation,

    S. Yan, Y . Liang, L. Zheng, M. Fan, X. Wang, and B. Wang, “Ex- plainable gated Bayesian recurrent neural network for non-Markov state estimation,” IEEE Trans. Signal Process., vol. 72, pp. 4302–4317, 2024

    work page 2024

  14. [14]

    Optimization of mixed- ADC multi-antenna systems for cloud-RAN deployments,

    J. Park, S. Park, A. Yazdan, and R. W. Heath, “Optimization of mixed- ADC multi-antenna systems for cloud-RAN deployments,” IEEE Trans. Commun., vol. 65, no. 9, pp. 3962–3975, 2017

    work page 2017

  15. [16]

    Downlink NOMA for short-packet internet of things communications with low-resolution ADCs,

    S. Kim, J. Choi, and J. Park, “Downlink NOMA for short-packet internet of things communications with low-resolution ADCs,” vol. 10, no. 7, pp. 6126–6139, 2023

    work page 2023

  16. [18]

    Low SNR asymptotic rates of vector channels with one-bit outputs,

    A. Mezghani, J. A. Nossek, and A. L. Swindlehurst, “Low SNR asymptotic rates of vector channels with one-bit outputs,” IEEE Trans. Inform. Theory, vol. 66, no. 12, pp. 7615–7634, 2020

    work page 2020

  17. [19]

    Near maximum-likelihood detector and channel estimator for uplink multiuser massive MIMO systems with one-bit ADCs,

    J. Choi, J. Mo, and R. W. Heath, “Near maximum-likelihood detector and channel estimator for uplink multiuser massive MIMO systems with one-bit ADCs,” IEEE Trans. Commun. , vol. 64, no. 5, pp. 2005–2018, 2016

    work page 2005

  18. [20]

    One-bit sphere decoding for uplink massive MIMO systems with one-bit ADCs,

    Y .-S. Jeon, N. Lee, S.-N. Hong, and R. W. Heath, “One-bit sphere decoding for uplink massive MIMO systems with one-bit ADCs,” IEEE Trans. Wireless Commun., vol. 17, no. 7, pp. 4509–4521, 2018

    work page 2018

  19. [21]

    The Bussgang decomposition of nonlin- ear systems: Basic theory and MIMO extensions [lecture notes],

    O. T. Demir and E. Bjornson, “The Bussgang decomposition of nonlin- ear systems: Basic theory and MIMO extensions [lecture notes],” IEEE Signal Process. Mag. , vol. 38, no. 1, pp. 131–136, 2021

    work page 2021

  20. [22]

    Channel estimation and performance analysis of one-bit massive MIMO systems,

    Y . Li, C. Tao, G. Seco-Granados, A. Mezghani, A. L. Swindlehurst, and L. Liu, “Channel estimation and performance analysis of one-bit massive MIMO systems,” IEEE Trans. Signal Process., vol. 65, no. 15, pp. 4075–4089, 2017

    work page 2017

  21. [23]

    Generalized Bussgang LMMSE channel estimation for one-bit massive MIMO systems,

    Q. Wan, J. Fang, H. Duan, Z. Chen, and H. Li, “Generalized Bussgang LMMSE channel estimation for one-bit massive MIMO systems,” IEEE Trans. Wireless Commun., vol. 19, no. 6, pp. 4234–4246, 2020

    work page 2020

  22. [24]

    On optimal MMSE channel estimation for one-bit quantized MIMO systems,

    M. Ding, I. Atzeni, A. T ¨olli, and A. L. Swindlehurst, “On optimal MMSE channel estimation for one-bit quantized MIMO systems,” IEEE Trans. Signal Process., vol. 73, pp. 617–632, 2025

    work page 2025

  23. [25]

    Uplink coordinated pilot design for 1-bit massive MIMO in correlated channel,

    H. Yun, J. Han, K. Shen, and J. Park, “Uplink coordinated pilot design for 1-bit massive MIMO in correlated channel,” arXiv preprint arXiv:2502.13429, 2025

    work page arXiv 2025

  24. [26]

    Model-based deep learning,

    N. Shlezinger and Y . C. Eldar, “Model-based deep learning,” Foundations and Trends® in Signal Processing, vol. 17, no. 4, pp. 291– 416, 2023. [Online]. Available: http://dx.doi.org/10.1561/2000000113

    work page doi:10.1561/2000000113 2023

  25. [27]

    University of Michigan North Campus long-term vision and lidar dataset,

    N. Carlevaris-Bianco, A. K. Ushani, and R. M. Eustice, “University of Michigan North Campus long-term vision and lidar dataset,” Int. J. of Robotics Research, vol. 35, no. 9, pp. 1023–1035, 2015

    work page 2015

  26. [28]

    Cross-correlation Functions of Amplitude-Distorted Gaussian Signals,

    J. J. Bussgang, “Cross-correlation Functions of Amplitude-Distorted Gaussian Signals,” Research Laboratory of Electronics, Massachusetts Institute of Technology, Cambridge, MA, Technical Report 216, Mar. 1952

    work page 1952

  27. [29]

    Capacity analysis of one-bit quantized MIMO systems with transmitter channel state information,

    J. Mo and R. W. Heath, “Capacity analysis of one-bit quantized MIMO systems with transmitter channel state information,” IEEE Trans. Signal Process., vol. 63, no. 20, pp. 5498–5512, 2015

    work page 2015

  28. [30]

    On the limits of communication with low-precision analog-to-digital conversion at the receiver,

    J. Singh, O. Dabeer, and U. Madhow, “On the limits of communication with low-precision analog-to-digital conversion at the receiver,” IEEE Trans. Commun., vol. 57, no. 12, pp. 3629–3639, 2009

    work page 2009

  29. [31]

    A general version of price’s theorem,

    F. V oigtlaender, “A general version of price’s theorem,” 2020. [Online]. Available: https://arxiv.org/abs/1710.03576

    work page arXiv 2020

  30. [32]

    On low-resolution ADCs in practical 5G millimeter-wave massive MIMO systems,

    J. Zhang, L. Dai, X. Li, Y . Liu, and L. Hanzo, “On low-resolution ADCs in practical 5G millimeter-wave massive MIMO systems,” IEEE Commun. Mag., vol. 56, no. 7, pp. 205–211, 2018

    work page 2018

  31. [33]

    Energy efficiency maximization precoding for quantized massive MIMO systems,

    J. Choi, J. Park, and N. Lee, “Energy efficiency maximization precoding for quantized massive MIMO systems,” IEEE Trans. Wireless Commun., vol. 21, no. 9, pp. 6803–6817, 2022

    work page 2022

  32. [34]

    Adam: A Method for Stochastic Optimization

    D. P. Kingma, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014