Theoretical Analysis for the CommSense Measurement System

Amit Kumar Mishra; Mohammed Zafar Ali Khan; Sandip Jana

REVIEW 5 major objections 6 minor 22 references

Theoretical Analysis for the CommSense Measurement System

T0 review · 5 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Projecting OFDM channel measurements onto about ten principal components gives scatterer detection at near-Bayes-optimal accuracy with an order-of-magnitude inference speedup.

desk verdict A useful complexity-analysis and modeling contribution for CommSense, but the near-Bayes-optimal claim rests on a circular Gaussian approximation and needs either validation or softening before it can stand. read the letter →

arxiv 2506.07685 v2 pith:Y4WGOFVI submitted 2025-06-09 eess.SP

Sandip Jana , Amit Kumar Mishra , Mohammed Zafar Ali Khan This is my paper

classification eess.SP

keywords IntegratedSensingandCommunication(ISAC)CommSensepassivescattererdetectionPrincipalComponentAnalysisLikelihoodRatioTestSupportVectorMachineOFDMchannelstateinformationBhattacharyyabound

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper tries to establish that a communication-based sensing (CommSense) receiver can detect passive scatterers, such as drones or vehicles, from ordinary OFDM downlink channel estimates without any extra transmitters, by keeping only a small number of principal components of the channel state information. It derives a three-link Rician channel model, builds the full likelihood-ratio test, and shows that PCA-reduced variants, especially PCA+SVM, match the Bhattacharyya error bound while cutting inference time by an order of magnitude. Across simulated systems with $N=256$ and $N=1024$ subcarriers, the PCA-based detectors keep near-perfect ROC AUC at moderate SNR, whereas the full LRT becomes numerically unstable at $N=1024$. If correct, this means existing cellular and Wi-Fi downlinks could double as real-time, device-free sensors without new spectrum or transmit hardware.

What carries the argument

The load-bearing object is the PCA projection of the CSI vector onto its top $P$ principal components, justified by the structure of the mixture covariance: the paper models the overall covariance as an identity-scaled noise term plus a rank-one update driven by the difference of the class means, so the scatterer signature concentrates in the dominant eigenvalue. Once projected, detection runs either as a Gaussian likelihood-ratio test in $P$ dimensions or as a linear or RBF-kernel SVM on the projected features. The Bhattacharyya bound, computed before and after PCA, is what connects the retained subspace to the probability of error, and the tight tracking of that bound by the empirical error curves is the paper's evidence that the low-rank subspace is sufficient. The per-sample inference cost drops from $O(N^2)$ to $O(NP+P^2)$, which is the source of the measured order-of-magnitude speedup.

What would settle it

A Monte Carlo computation of the exact Bayes error for the cascaded product of two non-central complex Gaussians, using the density derived in Appendix A, compared against the Gaussian moment-matched Bhattacharyya bound at low SNR and across Rician K factors, would settle whether the near-Bayes-optimal claim is real or an artifact of the Gaussian approximation.

Watch

Extended reading notes

Core claim

The paper's central claim is that the statistical information needed to detect a passive scatterer in OFDM CSI is low-rank: projecting the $N$-dimensional measurement onto about $P=10$ principal components preserves essentially all discriminative energy, so PCA-reduced detectors operate at near-Bayes-optimal error rates for a small fraction of the full-dimensional likelihood-ratio cost. Starting from a three-link Rician model (direct transmitter-to-receiver, transmitter-to-scatterer, and scatterer-to-receiver with radar cross-section), the authors derive a full LRT, a PCA-based LRT, and two PCA+SVM classifiers. They report that empirical errors track the Bhattacharyya bound after PCA, that the area under the ROC curve reaches about 1 for $P=10$ at moderate SNR, and that the PCA methods keep working at $N=1024$, where the full LRT collapses to random guessing due to numerical overflow. The simulations also indicate that LRT-based detection degrades under mean and covariance estimation error, while the SVM variants remain accurate without needing explicit channel statistics.

Load-bearing premise

The paper's optimality claims rest on approximating the cascaded transmitter-to-scatterer-to-receiver channel, whose exact distribution is called analytically intractable, as a complex Gaussian by moment matching; if the true product channel is not Gaussian, the derived likelihood-ratio test, the Bhattacharyya bound, and the claim of near-Bayes-optimal performance all lose their foundation.

Editorial extensions

If this is right

A subspace of about $P=10$ components is enough: at SNR of 5 dB and above, PCA+LRT and PCA+SVM reach AUC near 1 for both $N=256$ and $N=1024$, so real-time detection does not require the full $N$-dimensional CSI.
Per-test inference cost falls from $O(N^2)$ to $O(NP+P^2)$; the paper measures this as an order-of-magnitude speedup for both simulated system sizes.
In very high dimensions ($N=1024$), the full LRT fails numerically with AUC near 0.5, while PCA-based detectors exceed 0.9 by 0 dB, so the reduced-rank approach is more stable as well as faster.
PCA+SVM classifiers tolerate channel-parameter estimation error better than the likelihood-ratio detectors, which matters when statistics must be learned from finite training data.
Empirical error curves tracking the Bhattacharyya bound after PCA give a practical design rule: choose $P$ near the elbow where further principal components are noise-dominated, about 10 to 20 in these simulations.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Editorial inference: the same low-rank argument should transfer to other device-free OFDM sensing tasks such as occupancy and localization, because any mean-shift-induced rank-one update in the covariance will concentrate in the top principal components; a field test with a software-defined radio would show whether $P=10$ holds outside simulation.
Editorial inference: the Gaussian moment-match approximation is the main caveat to the optimality claim, so the tightness of the Bhattacharyya bound should be rechecked against the exact cascaded-channel density at extreme Rician K factors and low SNR.
Editorial inference: the fixed elbow at $P=10$ is a simulation-specific finding; an adaptive rule that selects $P$ from the eigenvalue gap, the sensing SNR, would make the system track changing channel conditions.
Editorial inference: because the comparison is entirely simulation-based, the strongest testable extension is a hardware testbed that trains the same PCA+SVM pipeline on measured CSI and scores it against ground-truth scatterer positions.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The paper gives the first systematic theoretical treatment of PCA-based CommSense that I know of, and the complexity analysis alone is worth the read. The three-link Rician model, the derivation of the cascaded-path distribution in Appendix A, and the moment-matched Gaussian approximation are clearly laid out. The per-sample cost comparison—O(N^2) for the full LRT versus O(NP+P^2) for PCA+LRT—is correct, and the runtime plots show the order-of-magnitude saving. The parameter-mismatch experiment is also useful: SVM is plainly more robust than PCA+LRT to mean/covariance errors, which is the paper's strongest practical point.

The soft spots are real. The biggest one is the circularity the stress-test flags. The same Gaussian approximation is used to derive the detector, to compute the Bhattacharyya bound, and to synthesize the data in every simulation. So the tight alignment between empirical errors and the bound mostly shows the simulator matches the detector's assumptions. It does not validate near-Bayes-optimal performance for the actual product channel. To make the claim stick, the authors should either quantify how close the moment-matched Gaussian is to the true distribution, or generate data from the true distribution and show the detectors still track the bound.

There are smaller issues. No error bars or Monte Carlo repetitions are reported, and key parameters—K-factors, geometry, scatterer RCS, the number of OFDM symbols M—are missing, so the simulations are not reproducible. There is also a direct inconsistency: Section IV-E says P was fixed at 10 for the ROC experiments, while the caption of Table II says P=20. That matters because P≈10 is the headline claim. The full-LRT numerical failure at N=1024 is fine as an observation, but calling it a robustness win for PCA is a bit generous; it is a baseline implementation artifact.

All of that said, the central engineering claim—that a small PCA subspace plus an SVM gives fast and reasonably accurate scatterer detection in a Rician OFDM scenario—is plausible and supported by the simulations as far as they go. The theory is a useful scaffolding even if the optimality language needs to be scaled back. This is the kind of paper a serious referee should see: it targets a practical ISAC problem, makes a concrete modeling contribution, and its flaws are addressable rather than fatal. My recommendation: send it to peer review, but require the circularity to be addressed or the claims to be softened, and demand reproducibility details before publication.

Referee Report

5 major / 6 minor

Summary. The paper develops a three-link Rician channel model for passive scatterer detection using OFDM channel state information, derives a Gaussian-approximated likelihood ratio test (Full LRT) and a PCA-reduced version, computes a Bhattacharyya bound, and evaluates four detectors (Full LRT, PCA+LRT, PCA+SVM with linear and RBF kernels) in simulations with N=256 and N=1024 subcarriers. The central claims are that projecting CSI onto a P≈10 principal-component subspace reduces inference latency by an order of magnitude relative to the Full LRT while maintaining near-Bayes-optimal error rates, and that PCA+SVM is more robust to mean/covariance estimation errors than LRT-based detectors.

Significance. If the result holds, it offers a practical design guideline for low-complexity integrated sensing and communication on standard OFDM receivers. The paper contributes a nontrivial derivation of the exact distribution of the cascaded Tx-Scatterer-Rx product channel (Appendix A) and provides a useful complexity analysis (Section III-G). However, the current empirical evidence is weakened by three issues: the load-bearing Gaussian moment-matching approximation is unvalidated, the Bhattacharyya bound appears to be computed from the same data used for evaluation, and key simulation parameters (K-factors, distances, RCS, number of OFDM symbols) are not reported. The near-Bayes-optimal claim is therefore established only inside the approximate model, not for the physical Rician product channel.

major comments (5)

[Appendix A-H, Eq. (46)] The moment-matched Gaussian approximation is load-bearing and unvalidated. The LRT in Eq. (38), the Bhattacharyya bound in Appendix C, and all simulation data in Section IV are generated from the complex Gaussian z ~ CN(mu_z, sigma_z^2) of Eq. (46). Since the exact distribution of the product of two noncentral complex Gaussians (derived in Appendix A) is not Gaussian, the agreement between empirical errors and the bound is a self-consistency check inside the approximate model and does not establish Bayes optimality for the physical Rician product channel. Please quantify the fidelity of the approximation (e.g., KL divergence or a Monte Carlo comparison of the exact and approximate densities) and include a test in which detectors are trained and evaluated on data drawn from the exact product distribution to assess the resulting mismatch.
[Figure 5 caption] The caption states that the Bhattacharyya bound is 'computed from the raw data,' and the empirical errors are then compared with this same data-derived bound. This is circular as an optimality validation: a detector built from the Gaussian approximation will align with a bound computed on data generated from that same approximation. Please compute the bound from the physical model parameters (means and covariances of the Gaussian approximation) and overlay it as a theoretical curve, or explicitly describe the bound as fitted to the data and correspondingly weaken the implication of 'tight' agreement.
[Section IV-E and Table II] Section IV-E states that the ROC experiments keep 'the PCA subspace dimension (P) fixed at 10,' while the caption of Table II says 'P = 20.' The abstract's P≈10 near-optimality claim depends on this quantity. Please state the exact P used for each experiment and, if different values were used for different figures, tag each AUC column accordingly.
[Section IV (Simulation Setup)] The simulation setup omits essential parameters: the Rician K-factors (K_TR, K_TS, K_SR), the scatterer RCS, the link distances d_TR, d_TS, d_SR, and the number of OFDM symbols M. These parameters determine the means and covariances, and hence the Bhattacharyya bound and the detector error rates. Without them the reported results are not reproducible. Please provide a complete parameter table and add standard deviations or confidence intervals over independent Monte Carlo repetitions.
[Section IV-C] The text reports that the SVM kernels 'briefly outperform the bound' at P≈10 (e.g., SNR=5 dB). If the Bhattacharyya bound is a valid upper bound on the error probability, no classifier can beat it; the most likely explanation is that the bound is loose or that the empirical error estimate has finite-sample bias. Please clarify which mechanism applies, since the claim of 'tight' alignment with the bound is otherwise internally inconsistent.

minor comments (6)

[Abstract] The phrase '6G networks envisions' should be '6G networks envision'; also the abstract's 'P approx. equal to 10' would benefit from stating the SNR conditions under which AUC ≈ 1 is reached.
[Section III] The paper mixes 'Chapter II/III' with 'Section II/III' throughout the structure description; please use a single consistent convention.
[Section III-F, Eq. (36)] The notation switches between the received vector G and the data matrix X without a clear transition; please define the test vector consistently before deriving the likelihood ratio.
[Table II] Table II lists SNR rows -10, 0, 5, 15 dB, but Section IV-E says the ROC experiments cover six SNR settings from -10 to 15 dB, implying -5 and 10 dB are missing; please add the missing rows or correct the text.
[Figure 4] The x-axis label in Figure 4(a) appears garbled ('ptcAsubspace dimension'); please fix the typo.
[Section IV-B] The claim that Full LRT at N=1024 'fails due to numerical overflow' would be more convincing if the implementation used numerically stable forms (e.g., log-determinant identities); please describe the implementation or cite a reference for the stable computation of the LRT.

Circularity Check

1 steps flagged · score 3.0 of 10

One data-fitted Bhattacharyya bound is called 'theoretical'; the LRT/PCA derivation itself is self-contained, so circularity is mild and partial.

fitted input called prediction [Section IV-C, Fig. 5 caption; cf. Section I-C 'Theoretical Validation' bullet]
"We compare four detectors: Full LRT (red), PCA-LRT (blue circles), PCA+SVM linear (green squares), and PCA+SVM RBF (magenta diamonds) against the theoretical Bhattacharyya bound computed from the raw data (solid gray) and after PCA (dashed black)."

The paper labels a Bhattacharyya bound 'theoretical' and treats its agreement with empirical errors as 'Theoretical Validation' (Section I-C). But the caption states the bound is computed from the raw data, i.e., from the same simulated CSI samples whose error rates are then compared with it. The bound's means and covariances are therefore fitted to the very data being 'predicted', making the tight alignment a self-consistency check rather than an independent theoretical prediction. This weakens the abstract's claim that 'empirical errors align tightly with the Bhattacharyya error bound' supports 'optimal error rates'; it does not, however, force the complexity or robustness results, which rest on independent cost analysis and perturbation experiments.

full rationale

The core derivation chain is self-contained: the Rician link model, complex-Gaussian densities, full LRT, PCA projection, and Bhattacharyya bound are standard textbook manipulations, and no unique result is imported from a self-citation chain. Self-citations [11]-[17] document prior CommSense demos but are not load-bearing for the present theoretical claims. The one genuinely circular element is the Fig. 5 'theoretical' bound computed from raw data: comparing empirical errors to a bound estimated from those same data is a self-consistency check, not a prediction, so it should not be cited as independent theoretical validation. A separate, non-circular caveat is stated in Appendix A-H: the exact cascaded product of two Rician links is replaced by a moment-matched complex Gaussian, so the near-Bayes-optimality claim is conditional on that approximation rather than proven for the true product distribution. There is also an internal inconsistency between Section IV-E ('keeping the PCA subspace dimension (P) fixed at 10') and Table II ('P = 20'), which makes the precise empirical support for the 'P ≈ 10' claim ambiguous. These issues affect the strength of the optimality claim but not the complexity or robustness arguments, so the overall circularity is mild.

Assumptions & free parameters 4 free parameters · 3 assumptions · 0 invented entities

The central claim rests on simulation parameters that are not reported, on a moment-matching Gaussian approximation for the cascaded path, and on a perfect-CSI assumption. No new physical entities are introduced.

free parameters (4)

P (PCA subspace dimension) = 10-20 recommended
Swept in simulations; the claim that P around 10 suffices is an empirical observation from the elbow in the error curves, not a derived value.
Rician K-factors (K_TR, K_TS, K_SR) = not stated in available text
Control the strength of the dominant paths and hence the separability of H0 and H1; no values are reported.
Scatterer RCS and link distances (d_TR, d_TS, d_SR) = not stated
Path-loss model inputs that set the received scattered power; values are not reported.
Number of OFDM symbols M = not stated
Training data size for PCA and SVM; affects the variance of estimates and detection accuracy; not reported.

assumptions (3)

ad hoc to paper The cascaded Tx-Scatterer-Rx channel can be approximated as a complex Gaussian via moment matching.
Adopted in Appendix A because the exact distribution is described as analytically intractable; the LRT and Bhattacharyya bounds are built on this approximation.
domain assumption CSI at pilot subcarriers is a perfect observation of the composite channel in the main simulations.
The model treats G as the channel frequency response; real receivers have channel estimation noise that is not modeled.
domain assumption Equal prior class probabilities and known noise variance.
Used to set the LLR threshold and to derive the Bhattacharyya bound; prior probabilities may differ in practice.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Theoretical Analysis for the CommSense Measurement System." pith.science (2026). https://pith.science/paper/Y4WGOFVI

@misc{pith2026250607685,
  author       = {Pith},
  title        = {Pith review of: Theoretical Analysis for the CommSense Measurement System},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Y4WGOFVI}},
  note         = {Machine review of arXiv:2506.07685}
}

read the original abstract

Future 6G networks envisions to blur the line between communication and sensing, leveraging ubiquitous OFDM waveforms for both high throughput data and environmental awareness. In this work, we do a thorough analysis of Communication based Sensing (CommSense) framework that embeds lightweight, PCA based detectors into standard OFDM receivers; enabling real-time, device free detection of passive scatterers (e.g. drones, vehicles etc.) without any extra transmitters. Starting from a realistic three link Rician channel model (direct Tx to Rx, cascaded Tx to Scatterer and Scatterer to Rx), we compare four detectors: the full dimensional Likelihood Ratio Test (Full LRT), PCA based LRT, PCA-SVM with linear and RBF kernels. By projecting N-dimensional CSI onto a P (very less than N) principal component subspace, inference time gets reduced by an order of magnitude compared to the full LRT, while achieving optimal error rates i.e. empirical errors align tightly with the Bhattacharyya error bound and Area Under ROC Curve (AUC) approx. equal to 1 for P approx. equal to 10. From the simulated result we have shown LRT based techniques are susceptible to the parameter estimation error, where as SVM is resilient to that. Our results demonstrate that PCA driven detection when paired with lightweight SVMs can deliver fast, accurate, and robust scatterer sensing, paving the way for integrated sensing and communication (ISAC) in 6G and beyond.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

22 extracted references · 20 canonical work pages

[1]

Application specific instrumentation (asin) a bio-inspired paradigm to instrumentation by fusing sensors and ai (sensai),

A. K. Mishra, "Application specific instrumentation (asin) a bio-inspired paradigm to instrumentation by fusing sensors and ai (sensai)," in Proceedings of the International Conference on Data Science, Machine Learning and Artificial Intelligence, 2021, pp. 1-6

work page 2021
[2]

G. H. Golub and C. F. Van Loan, Matrix computations. JHU press, 2013

2013
[3]

Indoor tracking and navigation using received signal strength and compressive sensing on a mobile device,

A. W. S. Au, C. Feng, S. Valaee, S. Reyes, S. Sorour, S. N. Markowitz, D. Gold, K. Gordon, and M. Eizenman, "Indoor tracking and navigation using received signal strength and compressive sensing on a mobile device," IEEE Transactions on Mobile Computing, vol. 12, no. 10, pp. 2050-2062, 2012

work page 2012
[4]

Low-cost indoor human tracking by utilizing fluctuation of received radio signal strength,

H.-C. Lee and T.-T. Lin, "Low-cost indoor human tracking by utilizing fluctuation of received radio signal strength," IEEE Sensors Journal, vol. 20, no. 21, pp. 13 029-13 036, 2020

work page 2020
[5]

Machine learning-based integrated wireless sensing and positioning for cellular network,

L. Zhang, X. Chu, and M. Zhai, "Machine learning-based integrated wireless sensing and positioning for cellular network," IEEE transac tions on instrumentation and measurement, vol. 72, pp. 1-11, 2022

work page 2022
[6]

Channel emulator enabled over-the-air testing platform for integrated sensing and communication systems: Framework and experimental validation,

Z. Wang, C. Li, H. Sun, S. Qiao, and W. Fan, "Channel emulator enabled over-the-air testing platform for integrated sensing and communication systems: Framework and experimental validation," IEEE Transactions on Instrumentation and Measurement, 2025

work page 2025
[7]

Precision agriculture: Ultra compact sensor and reconfigurable antenna for joint sensing and com munication,

A. Raza, R. Keshavarz, and N. Shariati, "Precision agriculture: Ultra compact sensor and reconfigurable antenna for joint sensing and com munication," IEEE Transactions on Instrumentation and Measurement, vol. 73, pp. 1-13, 2024

work page 2024
[8]

Human sensing via passive spectrum monitoring,

H. Mu, L. Yuan, and J. Li, "Human sensing via passive spectrum monitoring," IEEE Open Journal of Instrumentation and Measurement, vol. 2, pp. 1-13, 2023

work page 2023

Show all 22 references

[9]

Ghasemi, P

R. Ghasemi, P. Fenske, T. Koegel, M. Hehn, I. Ullmann, and M. Vossiek, "Ultrahigh-performance radio frequency system-on-chip implementation of a kalman filter-based high-precision time and frequency synchro nization for networked integrated sensing and communication systems,"...

2025
[10]

Augmentation of fingerprints for indoor wifi localization based on gaussian process regression,

W. Sun, M. Xue, H. Yu, H. Tang, and A. Lin, "Augmentation of fingerprints for indoor wifi localization based on gaussian process regression," IEEE Transactions on Vehicular Technology, vol. 67, no. 11, pp. 10896-10905, 2018

2018
[11]

Indoor occupancy estimation using the lte-commsense system,

S. Sardar, A. K. Mishra, and M. Z. A. Khan, "Indoor occupancy estimation using the lte-commsense system," International Journal of Remote Sensing, vol. 41, no. 14, pp. 5609-5619, 2020

2020
[12]

Opportunistic drone detection using commsense,

S. Jana, A. K. Mishra, and M. Z. A. Khan, "Opportunistic drone detection using commsense," IEEE Transactions on Instrumentation and Measurement, 2025

2025
[13]

Gsm--commsense-based through-the-wall sensing,

A. Bhatta and A. K. Mishra, "Gsm--commsense-based through-the-wall sensing," Remote Sensing Letters, vol. 9, no. 3, pp. 247-256, 2018

2018
[14]

Crowd size estimation using commsense instrument for covid-19 echo period,

S. Sardar, A. K. Mishra, and M. Z. Khan, "Crowd size estimation using commsense instrument for covid-19 echo period," IEEE Consumer Electronics Magazine, vol. 10, no. 2, pp. 92-97, 2020

2020
[15]

S. Jana, A. K. Mishra, and M. Z. A. Khan, "Evaluation of visualiza tion algorithms for commsense system,'in 2022 IEEE 95th Vehicular Technology Conference:(VTC2022-Spring). IEEE, 2022, pp. 1-5

2022
[16]

Sensing the environment with 5g scattered signals (5g commsense): A feasibility analysis,

--, "Sensing the environment with 5g scattered signals (5g commsense): A feasibility analysis," in 2023 IEEE Applied Sensing Conference (APSCON). IEEE, 2023, pp. 1-3

2023
[17]

Validation of a commsense based isac system using in-situ mmwave propagation model,

S. Jana, D. K. K. Reddy, A. K. Mishra, and M. Z. A. Khan, "Validation of a commsense based isac system using in-situ mmwave propagation model," in 2023 26th International Symposium on Wireless Personal Multimedia Communications (WPMC). IEEE, 2023, pp. 266-271

2023
[18]

Lte; evolved universal terrestrial radio access (e-utra); physi cal channels and modulation (3gpp ts 36.211 version 18.0.1 release 18),

3GPP, "Lte; evolved universal terrestrial radio access (e-utra); physi cal channels and modulation (3gpp ts 36.211 version 18.0.1 release 18)," https:/ /portal.etsi.org/webapp/workprogram/Report_ Workltem.asp? wKI_ID=71758, 2024, [Accessed 12-05-2025]

2024
[19]

5g; nr; physical channels and modulation (3gpp ts 38.211 version 18.6.0 release 18),

--, "5g; nr; physical channels and modulation (3gpp ts 38.211 version 18.6.0 release 18)," https://portal.etsi.org/webapp/workprogram/Report_ Workltem.asp?WKI_ID=74677, 2025, [Accessed 12-05-2025]

2025
[20]

H. L. Van Trees, Detection, estimation, and modulation theory, part I: detection, estimation, and linear modulation theory. John Wiley & Sons, 2004

2004
[21]

Fukunaga, Introduction to statistical pattern recognition

K. Fukunaga, Introduction to statistical pattern recognition. Elsevier, 2013. [221 1. Gradshteyn and I. Ryzhik, Table of Integrals, Series, and Product (2007). Elsevier, 2007. (50) (49) (51) D, ID % P , / o m ie = e 2 3 +? 1 2 = 5 7 ll m - moll + o? 1 5 '~ l@ '(n o )l l The te...

2007
[22]

A larger [[m moll increases the dominant eigenvalue, thereby increasing Dp

Second Term: We have det% =()", det =(3)° and ( 2 2)P det I.; = a o ; a 1 Thus, • ( 4 ! % eat> ; w { (@#·#) 2 /dot2o dct2, 2 (0 0 ,) ! 2 2000 which normalizes the separation of the projected means by the overall noise level. A larger [[m moll increases the dominant eigenvalue,...

2000

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Application specific instrumentation (asin) a bio-inspired paradigm to instrumentation by fusing sensors and ai (sensai),

A. K. Mishra, "Application specific instrumentation (asin) a bio-inspired paradigm to instrumentation by fusing sensors and ai (sensai)," in Proceedings of the International Conference on Data Science, Machine Learning and Artificial Intelligence, 2021, pp. 1-6

work page 2021

[2] [2]

G. H. Golub and C. F. Van Loan, Matrix computations. JHU press, 2013

2013

[3] [3]

Indoor tracking and navigation using received signal strength and compressive sensing on a mobile device,

A. W. S. Au, C. Feng, S. Valaee, S. Reyes, S. Sorour, S. N. Markowitz, D. Gold, K. Gordon, and M. Eizenman, "Indoor tracking and navigation using received signal strength and compressive sensing on a mobile device," IEEE Transactions on Mobile Computing, vol. 12, no. 10, pp. 2050-2062, 2012

work page 2012

[4] [4]

Low-cost indoor human tracking by utilizing fluctuation of received radio signal strength,

H.-C. Lee and T.-T. Lin, "Low-cost indoor human tracking by utilizing fluctuation of received radio signal strength," IEEE Sensors Journal, vol. 20, no. 21, pp. 13 029-13 036, 2020

work page 2020

[5] [5]

Machine learning-based integrated wireless sensing and positioning for cellular network,

L. Zhang, X. Chu, and M. Zhai, "Machine learning-based integrated wireless sensing and positioning for cellular network," IEEE transac tions on instrumentation and measurement, vol. 72, pp. 1-11, 2022

work page 2022

[6] [6]

Channel emulator enabled over-the-air testing platform for integrated sensing and communication systems: Framework and experimental validation,

Z. Wang, C. Li, H. Sun, S. Qiao, and W. Fan, "Channel emulator enabled over-the-air testing platform for integrated sensing and communication systems: Framework and experimental validation," IEEE Transactions on Instrumentation and Measurement, 2025

work page 2025

[7] [7]

Precision agriculture: Ultra compact sensor and reconfigurable antenna for joint sensing and com munication,

A. Raza, R. Keshavarz, and N. Shariati, "Precision agriculture: Ultra compact sensor and reconfigurable antenna for joint sensing and com munication," IEEE Transactions on Instrumentation and Measurement, vol. 73, pp. 1-13, 2024

work page 2024

[8] [8]

Human sensing via passive spectrum monitoring,

H. Mu, L. Yuan, and J. Li, "Human sensing via passive spectrum monitoring," IEEE Open Journal of Instrumentation and Measurement, vol. 2, pp. 1-13, 2023

work page 2023