Sparse Bayesian Sensing for Antenna Arrays with Per-Element Gain, Phase, and Position Errors
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Electrical Engineering. The day in electrical engineering: signals, systems and the hardware that carries them.
Rosa: We're looking at the paper: "Sparse Bayesian Sensing for Antenna Arrays with Per-Element Gain, Phase, and Position Errors".
Dev: Here's the gist of the paper. Sparse Bayesian Sensing for Antenna Arrays with Per-Element Gain, Phase, and Position Errors is a parametric, model-based sparse Bayesian learning (SBL) channel estimation algorithm with integrated self-calibration that accounts for antenna-position deviations as well as gain and phase offsets The proposed method approaches a benchmark based on the derived hybrid Cramer–Rao lower bound (HCRLB) in the considered single-snapshot sensing scenarios On single-snapshot measurements, our algorithm reduces the position OSPA from about 6 m to 0.2 m for an appropriately chosen threshold Substantial improvements are demonstrated on single-snapshot measurements from a precalibrated phased array, where self-calibration reduced the position OSPA from about 6 m to 0.2 m.
Rosa: The paper adopts a static wideband single-input multiple-output (SIMO) setup consisting of a transmit antenna and a P-element uniform linear array (ULA) at the receiver The received signal model is extended to account for effective position deviations along the array axis, such that ρp = ρ0p + ∆ρp.
Dev: The frequency-domain received signal model is given by equation (1).
Rosa: The temporal response of the array is described by aτ (τk) = exp−j2πfτk.
Paper summary: Rosa: We talked about the core idea of Sparse Bayesian Sensing for Antenna Arrays with Per-Element Gain, Phase, and Position Errors by Hodl, Leitinger and Witrisal, which is that they propose a parametric model-based sparse Bayesian learning algorithm with integrated self-calibration to account for antenna position deviations as well as gain and phase offsets.
Dev: That approach learns the channel parameters such as delay and direction of arrival together with per antenna complex weights and those position deviations based on fast variational sparse Bayesian learning with dictionary refinement.
Taro: The thesis here is that this method accounts for antenna-position deviations as well as gain and phase offsets, which is something most standard estimation methods don't do directly, leading to performance degradation when you have hardware imperfections.
Conclusion: Dev: So, looking at the Sparse Bayesian Sensing for Antenna Arrays with Per-Element Gain, Phase, and Position Errors paper by Hodl, Leitinger and Witrisal, we’re seeing a parametric model-based sparse Bayesian learning algorithm that handles those physical errors like position deviations along with gain and phase offsets.
Rosa: They claim this method approaches the hybrid Cramer–Rao lower bound benchmark in single-snapshot sensing scenarios when estimating coherent multipath components, and it outperforms weight-only and position-only calibration for larger hardware impairments.
Taro: It seems they've shown that this joint estimation of channel parameters and those physical errors is a viable way to get much better performance than treating the position error as a separate, simple correction.
Dev: The implication is that for systems relying on accurate models of phased-array antennas, incorporating uncertainty about the physical placement of elements directly into the estimation process can lead to more robust sensing capabilities.
Patrick Hodl, Erik Leitinger, Klaus Witrisal
Institute of Comm. Networks and Satellite Comms., Graz University of Technology
eess.SP
Submitted: 2026-11-05
Updated: 2026-11-05
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Importance score: 80/100
The gist: Sparse Bayesian Sensing for Antenna Arrays with Per-Element Gain, Phase, and Position Errors is a parametric…
Key concepts
- Sparse Bayesian Sensing (SBL)
- A model-based learning technique that uses sparsity priors to estimate unknown parameters in a system, such as antenna errors. It helps identify which parameters are most important and estimate them efficiently when data is limited.
- Hybrid Cramer–Rao Lower Bound (HCRLB)
- A theoretical benchmark used to measure the best possible performance achievable for estimating array parameters under the given sensing conditions. The proposed method aims to approach this lower bound, proving its accuracy.
- Optimal Subpattern Assignment (OSPA)
- A metric used to evaluate array estimation performance by measuring the error in determining both delay and direction of arrival. A lower OSPA value indicates better performance and more accurate parameter estimation.
- Self-Calibration
- A technique where the algorithm estimates its own errors (like gain, phase, and position) using the received signal data itself. This is crucial for improving accuracy when hardware imperfections are present.
Terminology
Summary
Sparse Bayesian Sensing for Antenna Arrays with Per-Element Gain, Phase, and Position Errors is a parametric, model-based sparse Bayesian learning (SBL) channel estimation algorithm with integrated self-calibration that accounts for antenna-position deviations as well as gain and phase offsets The proposed method approaches a benchmark based on the derived hybrid Cramer–Rao lower bound (HCRLB) in the considered single-snapshot sensing scenarios On single-snapshot measurements, our algorithm reduces the position OSPA from about 6 m to 0. p. 02 m for an appropriately chosen threshold Substantial improvements are demonstrated on single-snapshot measurements from a precalibrated phased array, where self-calibration reduced the position OSPA from about 6 m to 0. p. 4 p. 42 m.
Methodology and Model Setup
The paper adopts a static wideband single-input multiple-output (SIMO) setup consisting of a transmit antenna and a P-element uniform linear array (ULA) at the receiver The received signal model is extended to account for effective position deviations along the array axis, such that ρp = ρ0p + ∆ρp. The frequency-domain received signal model is given by equation (1). p. 1 The temporal response of the array is described by aτ (τk) = exp−j2πfτk. p. 1 A compact generative model expresses the received vector as y = D(w)A(θ, ∆ρ)α + n. p. 1
Bayesian Estimation Framework
The paper adopts a sparsity-inducing hierarchical Bayesian Gamma–Gaussian prior model and extends it to include the position deviations ∆ρ Since ∆ρ enters the steering vectors nonlinearly, a Gaussian prior would not yield closed-form variational updates without further approximations. The complex weight deviations are modeled as p(w) = Yp Σw,p σ2w,p (5). p. 1 The noise precision λ is modeled as an independent Gamma-distributed random variable with PDF p(λ) = Ga(λa, b), again as the Jeffreys prior with a, b → 0. p. 1
Position Update Mechanism
The marginal-likelihood update for position deviations is derived by marginalizing out the complex amplitudes α as L(∆ρ) = Zp p(y0, Cy(∆ρ)) (9). The negative log-marginal likelihood is given by l(∆ρ) ≜ − lnL(∆ρ) e ∝ ln Cy(∆ρ)+y HCRLB−1 y (∆ρ)y, (12). p. 2 Point estimates of the position deviations are obtained as ∆ρˆ = arg min ∆ρ∈RP l(∆ρ). p. 2
Performance Evaluation and Benchmark
The sensing performance is evaluated in terms of the optimal subpatternassignment (OSPA) metric of the delay and DoA. The proposed method approaches a benchmark based on the derived hybrid Cramer–Rao lower bound (HCRLB) in the considered single-snapshot sensing scenarios. p. 0 The HCRLB benchmark is derived as a performance benchmark for our simulations. p. 4
Validation and Findings
The algorithm is validated on synthetic data and demonstrates single-snapshot self-calibration on elementspacetransformed measurements from a 28-GHz analog phased-array frontend. The results demonstrate the benefits of joint calibration compared to gain-only, position-only, and no calibration across the considered impairment scenarios The HCRLB benchmark is approached over a wide range, except under severe hardware impairments. p. 4 Substantial improvements are demonstrated on single-snapshot measurements from a precalibrated phased array, where self-calibration reduced the position OSPA from about 6 m to 0.2 m. The results show the importance of selecting an appropriate detection threshold for the proposed algorithm. p. 4
Measurement Data Validation
The measurement data were collected with an omnidirectional antenna as transmitter and a 4×1 analog phased-array antenna as receiver in an indoor laboratory In each snapshot, the analog phased array was steered to 15 beam directions, and the resulting beamspace data were transformed to elementspace as in [25] before applying our algorithm. The results exhibit a trend similar to that observed with the synthetic data in the detection-threshold sweep. p. 4 There is an optimal threshold range where we detect the correct number of components and reach a position OSPA of around 0. p. 42 m. The algorithm without calibration exhibits significantly worse sensing performance. p. 4 The proposed algorithm was able to mitigate residual model mismatch, such as hardware impairments. p. 4
Appendix Details
The hybrid Fisher information matrix (FIM) is defined as J = −Ey,νr ∂2ln p(y, νr; θ, νd) ∂ψ ∂ψT (15). The data FIM follows as [JD(ψ)]ll′ = 2λ R (∂µ/∂ψl H ∂µ/∂ψl′), (18). p. 5 The average of C(i) over the I realizations then approximates its expectation over νr, which, together with the average over the K components, yields the HCRLB benchmarks. p. 5 The OSPA metric [24] has the same RMSE structure as (22). p. 5 The paper was submitted to the IEEE for possible publication Copyright may be transferred without notice, after which this version may no longer be accessible. p. 0 The authors used Claude (Anthropic) for language editing. p. 5
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Integration of antenna position deviations into fast variational SBL: The algorithm extends existing methods by
incorporating antenna-position deviations as deterministic unknowns and derive a marginal-likelihoodbased position update that accounts for the uncertainty in the complex path amplitudes.
This allows the system to self-calibrate physical hardware errors like element shifts during sensing. -
Hybrid Cramer–Rao lower bound benchmark: The derivation of the HCRLB for DoA and delay estimation provides a rigorous performance standard, allowing researchers to evaluate whether their SBL approach
approaches the HCRLB benchmark in the considered single-snapshot sensing scenarios.
This ensures performance is measured against theoretical best limits rather than just empirical comparisons. -
Single-snapshot self-calibration capability: The proposed method demonstrates significant practical improvement, showing that
on single-snapshot measurements, our algorithm reduces the position OSPA from about 6 m to 0.2 m for an appropriately chosen threshold.
This enables real-time adaptation of sensing systems to hardware imperfections without needing extensive prior training data. -
Joint calibration superiority: The results show that
the algorithm jointly calibrating the complex weights and antenna positions approaches the HCRLB over a wide range of weight deviations.
This indicates that simultaneous estimation of gain/phase errors and physical positions yields superior performance compared to using only weight-only or position-only calibration. -
Detection threshold optimization: The system can dynamically adjust its detection threshold to balance spurious component detection and true component pruning, as the results
exhibit a trend similar to that observed with the synthetic data in the detection-threshold sweep.
This allows for adaptive sensing under varying signal conditions.
Sources
Related papers
- Open-H-RF: A Large-Scale Ultrasound Raw Channel Dataset Initiative
- Scalable Attention for 5G NR Channel Estimation
- Low-Altitude Wireless Networks: A Comprehensive Survey
- Sensing-Assisted OFDM with Index Modulation: A Waveform for Optical Wireless Integrated Communication and Localisation
- Iteration-Free Candidate Pruning and an Index-Domain VLSI Architecture for Sort-Free MIMO Box Decoding up to 1024-QAM
- A Foundation Model for Large-Scale Wireless Network Planning, Operation and Optimization