Pith. sign in

REVIEW 3 major objections 4 minor 26 references

Low-Complexity Cram\'er-Rao Lower Bound and Sum Rate Optimization in ISAC Systems

T0 review · 3 major / 4 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read The paper proposes a low-complexity SCA-SGPI algorithm for joint beamforming that maximizes a weighted sum of communications sum rate and sensing CRLB under a power constraint, matching or slightly beating SDR-based optimization at a…

desk verdict The SCA-SGPI algorithm is a promising practical recipe, but the central convergence argument is broken by an incorrect Taylor expansion in Lemma 2, so the current version needs a major fix before the theory can be trusted. read the letter →

arxiv 2502.03162 v1 pith:I3LW4UKF submitted 2025-02-05 cs.IT eess.SPmath.IT

Tianyu Fang , Nhan Thanh Nguyen , Markku Juntti This is my paper · ORCID
classification cs.ITeess.SPmath.IT
keywords integratedsensingandcommunicationsCramér-RaolowerboundbeamformingoptimizationsuccessiveconvexapproximationgeneralizedpoweriterationsumrateMIMOradarlow-complexityalgorithm
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

Integrated sensing and communications (ISAC) beamforming must balance how much data the base station delivers to users against how accurately it can estimate a target, and the sensing side of that balance is usually measured by the Cramér-Rao lower bound (CRLB). Optimizing the CRLB with semidefinite relaxation is accurate but expensive, scaling steeply in the number of users and antennas. This paper proposes to solve the weighted sum-rate-plus-CRLB problem with a two-step iteration: successive convex approximation builds a quadratic surrogate for the nonconvex objective, and a shifted generalized power iteration solves each quadratic subproblem directly. The resulting SCA-SGPI algorithm is claimed to match or slightly beat the SCA-SDR benchmark on the communications-sensing tradeoff while cutting the per-iteration complexity from $O(I_1 K^{3.5} N_t^{6.5})$ to $O(I_1 I_2 K N_t^2)$, and the reported runtime drops from 9.08 seconds to 0.036 seconds in the simulated setup. The practical payoff, if the claims hold, is ISAC beamforming that can be recomputed nearly in real time.

What carries the argument

The load-bearing object is the quadratic surrogate (13) together with the shifted generalized power iteration (16). The SCA layer converts the nonconvex objective into $\mathrm{tr}(\mathbf{W}\mathbf{W}^H \mathbf{A}) + 2\delta\,\Re\bigl\{\mathrm{tr}(\mathbf{W}\boldsymbol{\Sigma}_1^H \mathbf{H}^H)\bigr\}$, where $\mathbf{A}$ is made positive semidefinite by a shift parameter $\lambda$; this keeps the subproblem a quadratically constrained quadratic program. The SGPI layer is a fixed-point method that repeatedly maps $\mathbf{H}\boldsymbol{\Sigma}_1 + \mathbf{A}\mathbf{W}^{(n)}$ onto the feasible sphere $\mathcal{S}$, costing only $O(K N_t^2)$ per update and avoiding interior-point solvers and semidefinite relaxation. The two layers together are what let the algorithm escape the $K^{3.5} N_t^{6.5}$ scaling of SDR-based CRLB optimization.

What would settle it

Compute both sides of Lemma 2's inequality (9) at $Z = Z_0$: the left side equals $\mathrm{tr}(Z_0^{-1})$ while the right side is zero, so the claimed equality is false. A direct algorithmic check is to record the true objective (7) after every outer iteration on random channels; any decrease would disprove the claimed monotone convergence and confirm that the missing minorant property matters.

Watch

Extended reading notes

Core claim

The paper's central claim is that the nonconvex beamforming problem (7), which maximizes $\delta\sum_k R_k - \mathrm{tr}(F^{-1})$ under a fixed transmit power, can be solved efficiently by alternating two cheap steps. The outer SCA step replaces each user's log-rate with its first-order minorant (Lemma 1) and replaces the sensing term with a linearized trace expression built from Lemma 2, turning the objective into a quadratic form in the beamforming matrix $W$. The inner SGPI step then solves that quadratic subproblem by the fixed-point update $\mathbf{W}^{(n+1)} = \Pi_{\mathcal{S}}\big(\mathbf{H}\boldsymbol{\Sigma}_1 + \mathbf{A}\mathbf{W}^{(n)}\big)$, a projection onto the power-constraint sphere. The authors assert that this yields a locally optimal solution with per-iteration cost $O(K N_t^2)$, and their simulations show a slightly better communications-sensing tradeoff than the SCA-SDR benchmark with two to three orders of magnitude less runtime.

Load-bearing premise

The algorithm's convergence relies on each surrogate being a lower bound of the true objective, yet the paper's Lemma 2 misses a constant term that breaks equality at the current point, so the lower-bound property is not established.

Editorial extensions

If this is right

  • In the simulated $N_t=16$, $N_r=20$, $K=4$ setup, the algorithm converges in roughly ten outer and five inner iterations, so the complexity reduction translates into a runtime of 0.0362 s versus 9.0803 s for the SCA-SDR benchmark.
  • Because complexity grows linearly in the number of users $K$, the method remains tractable for many-user ISAC systems where SDR-based solvers scale as $K^{3.5}$.
  • The sweep over $\delta$ in $[10^{-7}, 10]$ shows SCA-SGPI slightly dominating the SCA-SDR tradeoff curve, indicating the speed gain does not come at the cost of communications or sensing performance.
  • The same SCA-plus-power-iteration structure can be extended to multi-target ISAC, which the authors identify as the immediate future direction.

Reading between the lines

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

  • The paper leaves implicit that, once the Lemma 2 constant error is repaired, the same surrogate-plus-power-iteration recipe should transfer to other CRLB-type sensing metrics such as the determinant or the largest eigenvalue of $F^{-1}$, since those also admit tractable surrogates.
  • A testable extension is to apply SCA-SGPI to wideband or multi-carrier ISAC; the linear-in-$K$ scaling suggests the speed advantage over SDR should persist.
  • Recording the true objective (7) at every outer iteration on random channels would separate the algorithm's numerical behavior from the unproven minorant property, since any decrease would localize the gap to the sensing surrogate.
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.

Referee Report

3 major / 4 minor

Summary. The paper studies joint beamforming for a monostatic ISAC system with Nt transmit and Nr receive antennas, K single-antenna users, and one point target. It formulates the problem of maximizing a weighted sum of the multiuser sum rate and the trace of the inverse Fisher information matrix, subject to a total transmit power constraint. The authors propose an SCA-SGPI algorithm in which SCA builds a quadratic surrogate for the objective and a shifted generalized power iteration solves the resulting subproblems. They provide a complexity analysis, a runtime comparison with an SCA-SDR benchmark, and simulations showing fast convergence and a slightly better communications-sensing tradeoff than the benchmark.

Significance. The practical goal is worthwhile: CRLB-based ISAC beamforming is typically solved by SDR with high complexity, and a low-complexity iterative method with closed-form updates could be valuable. The paper gives a clean system model, explicit surrogate equations, a simple projection update, and numerical evidence of large runtime reductions (0.0362 s versus 9.0803 s in one setting). However, the central theoretical justification is not sound: Lemma 2 is incorrect, and the sensing surrogate in (12) is not a valid minorant of the objective. As a result, the claimed SCA monotone-convergence mechanism is unsupported; the contribution is currently an empirically motivated heuristic rather than a proven optimization algorithm. If the surrogate issue is repaired or the claims are appropriately weakened, the complexity and simulation results would be of interest to the ISAC community.

major comments (3)
  1. [Section III-A, Lemma 2 (Eq. (9))] Lemma 2 is false as stated. Evaluating (9) at Z = Z0 yields tr(Z0^-1) >= 0, whereas equality is claimed; the right-hand side should contain an additional tr(Z0^-1) term (the correct tangent lower bound is tr(Z^-1) >= 2tr(Z0^-1) - tr(Z0^-1 Z Z0^-1)). This is load-bearing because the sensing surrogate in (12) is built directly from this inequality and therefore does not equal the true sensing term at the current iterate W[t].
  2. [Section III-A, Eq. (12)] Even after correcting Lemma 2, the sensing term in (12) is an upper bound on -tr(F^-1), not a lower bound. Since tr(F^-1) is convex in F and F depends affinely on WW^H, the first-order expansion of tr(F^-1) is a global lower bound for tr(F^-1); negating it yields an upper bound on -tr(F^-1). A standard SCA for maximization needs a surrogate that globally minorizes the objective and is tight at the current iterate. The surrogate (12) fails both conditions for the sensing term, so the SCA monotone-ascent argument for Algorithm 1 does not apply.
  3. [Section III-B, Algorithm 1] The paper explicitly omits the convergence proof ("We skip the detailed convergence analysis...") and instead relies on Fig. 1. Because the surrogate in (12) is not a minorant, the observed monotone increase in Fig. 1(a) cannot be attributed to the claimed SCA mechanism. The algorithm may be a useful heuristic, but the central theoretical claim needs to be either repaired (for instance, by constructing a valid minorant via a proximal or regularization term) or replaced by a different convergence analysis that does not rely on SCA minorization.
minor comments (4)
  1. [General] There are several typos: "Cram ´er" in the title and abstract, "it’s" in Lemma 3, "the the objective" in the caption of Fig. 1, "significantly reduction" in the Introduction, and "SCA-SPGI" for "SCA-SGPI" in the Conclusion.
  2. [Section III-B, Eqs. (15)-(16)] The inner-loop iteration uses W[n] without an outer-iteration index; writing W[t,n] would clarify that the inner iterations are performed at a fixed outer iteration t.
  3. [Section IV, Table I] The runtime of SCA-SGPI is non-monotonic in K (0.0270 s for K=2, 0.0185 s for K=4, 0.0137 s for K=6), which may be due to varying numbers of outer iterations; a brief explanation would improve reproducibility.
  4. [Section II-B, Eq. (7)] The transmit power constraint is written as an equality, tr(WW^H) = Pt. Since both the sum rate and the sensing CRLB improve with transmit power, the equality is natural, but the paper should state that using full power is always optimal for this objective.

Circularity Check

0 steps flagged · score 2.0 of 10

No load-bearing circularity: SCA-SGPI tradeoff is benchmarked against an external SCA-SDR method; the Lemma 2 surrogate error is a correctness issue, not a circular reduction.

full rationale

The central performance claim of the paper is not circular. The proposed SCA-SGPI method is compared against the external SCA-SDR benchmark [15] (Chen and Mao), the weight delta is swept over a fixed grid, and the reported sum-rate and CRLB values are direct evaluations of the original objective (7), with no parameter fitted to the target output. The only overlapping-author citations are [8] (background on ISAC fairness) and [23] (a standard Taylor lower bound used as Lemma 1); Lemma 1 is a parameter-free inequality with explicit assumptions and does not include the target result among its assumptions, so it is independent support rather than a circular premise. The real weakness is mathematical, not circular: Lemma 2 in Section III-A states tr(Z^{-1}) >= -tr(Z_0^{-1} Z Z_0^{-1}) + tr(Z_0^{-1}), but at Z = Z_0 the right-hand side equals 0 while the left-hand side equals tr(Z_0^{-1}); the correct first-order lower bound is 2tr(Z_0^{-1}) - tr(Z_0^{-1} Z Z_0^{-1}). Consequently the sensing surrogate in (12) is not a minorant of -tr(F^{-1}), and the SCA monotone-ascent argument is not established. The paper itself says in Section III-B, 'We skip the detailed convergence analysis of Algorithm 1 due to limited space. However, numerical results will be shown to verify the convergence.' This is an omitted proof and a correctness risk, but it is not a circular derivation: the algorithm is still evaluated against the original objective and an external benchmark, and no fitted constant is renamed as a prediction. The minor self-citation to [23] does not carry the central tradeoff claim, so the circularity score is low.

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

The central claim rests on standard SCA and MM machinery, the FIM block structure from prior radar literature, a known-angle design assumption, and an unproved convergence claim. No free parameters are fitted to data, and no new entities are introduced. The most fragile item is the trace-inverse surrogate in Lemma 2, which is not tight at the current iterate and produces an upper-bound surrogate for the sensing term.

assumptions (4)
  • standard math First-order Taylor lower bound of log(1+|z|^2/d) in Lemma 1 is valid for all feasible W.
    Used in (10) to construct the communication rate surrogate. The bound is standard and the equality condition holds at the current point, but global validity for the non-log-concave SINR expression is asserted via [23] rather than proved in this paper.
  • ad hoc to paper The trace-inverse inequality in Lemma 2, as written, is a valid lower bound for tr(F^-1) and its negation is a valid surrogate for -tr(F^-1).
    Equation (9) omits the second tr(F0^-1) term relative to the true tangent, so equality at F=F0 is lost. The paper nevertheless treats it as the SCA surrogate in (12), and this is the load-bearing modeling choice for the sensing part.
  • domain assumption The CRLB is computed from the FIM block expressions of [21,22] with a known nominal RCS alpha and a known target direction in the design.
    Used in the sensing model in Section II-A. In practice the RCS and angle are unknown; the design assumes their values to compute the FIM, which limits the operational validity of the optimized beamformer.
  • ad hoc to paper The SCA-SGPI inner iterations, with lambda large enough to make A PSD, converge to a point that improves (12), and repeating outer iterations converges to a stationary point of (7).
    The convergence analysis is explicitly skipped in Section III-B, and the monotonicity does not follow from the stated lemmas because the sensing surrogate is not a minorant.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Low-Complexity Cram\'er-Rao Lower Bound and Sum Rate Optimization in ISAC Systems." pith.science (2026). https://pith.science/paper/I3LW4UKF

@misc{pith2026250203162,
  author       = {Pith},
  title        = {Pith review of: Low-Complexity Cram\'er-Rao Lower Bound and Sum Rate Optimization in ISAC Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/I3LW4UKF}},
  note         = {Machine review of arXiv:2502.03162}
}
read the original abstract

While Cram\'er-Rao lower bound is an important metric in sensing functions in integrated sensing and communications (ISAC) designs, its optimization usually involves a computationally expensive solution such as semidefinite relaxation. In this paper, we aim to develop a low-complexity yet efficient algorithm for CRLB optimization. We focus on a beamforming design that maximizes the weighted sum between the communications sum rate and the sensing CRLB, subject to a transmit power constraint. Given the non-convexity of this problem, we propose a novel method that combines successive convex approximation (SCA) with a shifted generalized power iteration (SGPI) approach, termed SCA-SGPI. The SCA technique is utilized to approximate the non-convex objective function with convex surrogates, while the SGPI efficiently solves the resulting quadratic subproblems. Simulation results demonstrate that the proposed SCA-SGPI algorithm not only achieves superior tradeoff performance compared to existing method but also significantly reduces computational time, making it a promising solution for practical ISAC applications.

Figures

Figures reproduced from arXiv: 2502.03162 by the authors.

Figure 1
Figure 1. Convergence of the inner and outer loops of Algorithm [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 3
Figure 3. Performance versus number of communications users. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 19 canonical work pages

  1. [23]

    Optimal beamforming structure and e fficient opti- mization algorithms for generalized multi-group multicas t beamforming optimization,

    T. Fang and Y . Mao, “Optimal beamforming structure and e fficient opti- mization algorithms for generalized multi-group multicas t beamforming optimization,” arXiv preprint arXiv:2312.16559 , 2023

    arXiv 2023

  2. [1]

    Jo int radar- communication strategies for autonomous vehicles: Combin ing two key automotive technologies,

    D. Ma, N. Shlezinger, T. Huang, Y . Liu, and Y . C. Eldar, “Jo int radar- communication strategies for autonomous vehicles: Combin ing two key automotive technologies,” IEEE Signal Process. Mag. , vol. 37, no. 4, pp. 85–97, 2020

    work page 2020

  3. [2]

    An overview of signal processing techniques for joint communication and radar sensing,

    J. A. Zhang, F. Liu, C. Masouros, R. W. Heath, Z. Feng, L. Zh eng, and A. Petropulu, “An overview of signal processing techniques for joint communication and radar sensing,” IEEE J. Sel. Topics Signal Process. , vol. 15, no. 6, pp. 1295–1315, 2021

    work page 2021

  4. [3]

    MU-MIMO c ommu- nications with MIMO radar: From co-existence to joint trans mission,

    F. Liu, C. Masouros, A. Li, H. Sun, and L. Hanzo, “MU-MIMO c ommu- nications with MIMO radar: From co-existence to joint trans mission,” IEEE Trans. Wireless Commun. , vol. 17, no. 4, pp. 2755–2770, 2018

    2018

  5. [4]

    Joint transmit beamforming for multiuser MIMO communications an d MIMO radar,

    X. Liu, T. Huang, N. Shlezinger, Y . Liu, J. Zhou, and Y . C. E ldar, “Joint transmit beamforming for multiuser MIMO communications an d MIMO radar,” IEEE Trans. Signal Process. , vol. 68, pp. 3929–3944, 2020

    2020

  6. [5]

    Joint communications and sensing design for multi-carrie r MIMO systems,

    N. T. Nguyen, N. Shlezinger, K.-H. Ngo, V .-D. Nguyen, and M. Juntti, “Joint communications and sensing design for multi-carrie r MIMO systems,” Proc. IEEE W orks. on Statistical Signal Processing , 2023

    work page 2023

  7. [6]

    M ultiuser MIMO wideband joint communications and sensing system with sub- carrier allocation,

    N. T. Nguyen, N. Shlezinger, Y . C. Eldar, and M. Juntti, “M ultiuser MIMO wideband joint communications and sensing system with sub- carrier allocation,” IEEE Trans. Signal Process., vol. 71, pp. 2997–3013, 2023

    work page 2023

  8. [7]

    Joint and robu st beam- forming framework for integrated sensing and communicatio n systems,

    J. Choi, J. Park, N. Lee, and A. Alkhateeb, “Joint and robu st beam- forming framework for integrated sensing and communicatio n systems,” arXiv preprint arXiv:2402.09155 , 2024

    arXiv 2024

Show all 26 references
  1. [8]

    Beamforming design for max- min fairness performance balancing in ISAC systems,

    T. Fang, N. T. Nguyen, and M. Juntti, “Beamforming design for max- min fairness performance balancing in ISAC systems,” in Proc. IEEE Int. W orkshop Signal Process. Adv. Wireless Commun. (SPAWC ), Sep. 2024

    2024

  2. [9]

    ISAC - accelerated edge intelligence: Framework, optimization, and analysis,

    T. Zhang, G. Li, S. Wang, G. Zhu, G. Chen, and R. Wang, “ISAC - accelerated edge intelligence: Framework, optimization, and analysis,” IEEE Trans. Green Commun. Network., vol. 7, no. 1, pp. 455–468, 2023

    2023

  3. [10]

    Optimizing the fairness of STAR-RIS and NOMA assisted integrated sensing and communication systems,

    Y . W. et al., “Optimizing the fairness of STAR-RIS and NOMA assisted integrated sensing and communication systems,” IEEE Trans. Wireless Commun., 2023

    2023

  4. [11]

    Cr am´ er-Rao bound optimization for joint radar-communication beamfor ming,

    F. Liu, Y .-F. Liu, A. Li, C. Masouros, and Y . C. Eldar, “Cr am´ er-Rao bound optimization for joint radar-communication beamfor ming,” IEEE Trans. Signal Process. , vol. 70, pp. 240–253, 2021

    2021

  5. [12]

    Intelli gent reflecting surface enabled sensing: Cram´ er-Rao bound optimization,

    X. Song, J. Xu, F. Liu, T. X. Han, and Y . C. Eldar, “Intelli gent reflecting surface enabled sensing: Cram´ er-Rao bound optimization,” IEEE Trans. Signal Process. , vol. 71, pp. 2011–2026, 2023

    2011

  6. [13]

    On the fundamental tradeoff of integrated sensing and communicat ions under gaussian channels,

    Y . Xiong, F. Liu, Y . Cui, W. Y uan, T. X. Han, and G. Caire, “ On the fundamental tradeoff of integrated sensing and communicat ions under gaussian channels,” IEEE Trans. Inf. Theory , vol. 69, no. 9, pp. 5723– 5751, 2023

    2023

  7. [14]

    Cram \ ’er-rao bound based waveform optimization for mimo radar: An efficient linear-proximal m ethod,

    X. Zhou, X. Du, and Y . Mao, “Cram \ ’er-rao bound based waveform optimization for mimo radar: An efficient linear-proximal m ethod,” arXiv preprint arXiv:2409.12569 , 2024

    2024 arXiv

  8. [15]

    Transmitter side beyond-diagonal R IS for mmwave integrated sensing and communications,

    K. Chen and Y . Mao, “Transmitter side beyond-diagonal R IS for mmwave integrated sensing and communications,” in Proc. IEEE Int. W orkshop Signal Process. Adv. Wireless Commun. (SPAWC), Sep. 2024

    2024

  9. [16]

    Information and s ensing beamforming optimization for multi-user multi-target mim o isac sys- tems,

    M. Zhu, L. Li, S. Xia, and T.-H. Chang, “Information and s ensing beamforming optimization for multi-user multi-target mim o isac sys- tems,” EURASIP J. Adv. Signal Process. , vol. 2023, no. 1, pp. 15, 2023

    2023

  10. [17]

    Bistatic mimo dfrc system waveform design via fractional programming,

    B. Guo, J. Liang, G. Wang, B. Tang, and H. C. So, “Bistatic mimo dfrc system waveform design via fractional programming,” IEEE Trans. Signal Process. , vol. 71, pp. 1952–1967, 2023

    1952

  11. [18]

    SNR/CRB- constrained joint beamforming and reflection designs for RIS-ISAC syste ms,

    R. Liu, M. Li, Q. Liu, and A. Lee Swindlehurst, “SNR/CRB- constrained joint beamforming and reflection designs for RIS-ISAC syste ms,” IEEE Trans. Wireless Commun. , vol. 23, no. 7, pp. 7456–7470, 2024

    2024

  12. [19]

    Fast fracti onal programming for multi-cell integrated sensing and communi cations,

    Y . Chen, Y . Feng, X. Li, L. Zhao, and K. Shen, “Fast fracti onal programming for multi-cell integrated sensing and communi cations,” arXiv preprint arXiv:2406.10910 , 2024

    2024 arXiv

  13. [20]

    Globally optimal b eamforming design for integrated sensing and communication systems,

    Z. Wang, J. Wu, Y .-F. Liu, and F. Liu, “Globally optimal b eamforming design for integrated sensing and communication systems,” in IEEE Int. Conf. Acoust., Speech and Signal Process. (ICASSP) , 2024, pp. 8931– 8935

    2024

  14. [21]

    Target detection and lo calization using MIMO radars and sonars,

    I. Bekkerman and J. Tabrikian, “Target detection and lo calization using MIMO radars and sonars,” IEEE Trans. Signal Process. , vol. 54, no. 10, pp. 3873–3883, 2006

    2006

  15. [22]

    R ange compression and waveform optimization for mimo radar: A cra m ´Er–rao bound based study,

    J. Li, L. Xu, P . Stoica, K. W. Forsythe, and D. W. Bliss, “R ange compression and waveform optimization for mimo radar: A cra m ´Er–rao bound based study,” IEEE Trans. Signal Process. , vol. 56, no. 1, pp. 218–232, 2008

    2008

  16. [24]

    Majorization-minim ization algorithms in signal processing, communications, and mach ine learning,

    Y . Sun, P . Babu, and D. P . Palomar, “Majorization-minim ization algorithms in signal processing, communications, and mach ine learning,” IEEE Trans. Signal Process. , vol. 65, no. 3, pp. 794–816, 2017

    2017

  17. [25]

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

    2013

  18. [26]

    Partially-conne cted hybrid beamforming design for integrated sensing and commu nication systems,

    X. Wang, Z. Fei, J. A. Zhang, and J. Xu, “Partially-conne cted hybrid beamforming design for integrated sensing and commu nication systems,” IEEE Trans. Commun. , vol. 70, no. 10, pp. 6648–6660, 2022

    2022

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.