Pith. sign in

REVIEW 3 major objections 6 minor 24 references

Detection with Uncertainty in Target Direction for Dual Functional Radar and Communication Systems

T0 review · 3 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Pith's one-line read An alternating optimization algorithm maximizes worst-case radar SCNR under target-direction uncertainty with provable convergence, using a single receive beamformer.

desk verdict Useful DFRC extension with a genuine proof gap: the claimed convergence relies on an unverified rank-one relaxation in the receive update. read the letter →

arxiv 2412.07245 v1 pith:ZEIK5MXW submitted 2024-12-10 eess.SP

classification eess.SP
keywords dual-functionradarandcommunicationintegratedsensingtargetdirectionuncertaintysignal-to-clutter-plus-noiseratiobeamformingoptimizationalternatingDinkelbachalgorithmmajorization-minimization
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

This paper tackles the realistic case of a dual-function radar-communication (DFRC) system where the target's direction is known only as a set of possible angles. It formulates a max-min optimization problem that maximizes the worst-case radar signal-to-clutter-plus-noise ratio (SCNR) while keeping every communication user's SINR above a required threshold. The authors propose an alternating algorithm: transmit beamformers are updated with a penalty-based majorization-minimization step, and a single receive beamformer is updated globally via Dinkelbach's method for generalized fractional programs. They prove convergence by showing the objective is nondecreasing across iterations, and their numerics indicate convergence in about three iterations with SCNR almost unchanged as the number of candidate target directions grows. The practical payoff is that one receive beamformer can replace a bank of direction-specific beamformers without meaningful detection loss.

What carries the argument

The mechanism that carries the argument is the reduction of the receive-beamformer subproblem to a generalized fractional program with Toeplitz quadratics. A Toeplitz matrix has constant entries along each diagonal, and each $B^l_{m,k}=A(\theta^l_m)u_k u_k^H A^H(\theta^l_m)$ has this structure, so problem (58) with its rank-one constraint can be solved globally by Dinkelbach's algorithm. On the transmit side, the machinery is a penalty-based majorization-minimization update: the nonconvex power equality is relaxed with a slack variable and penalty, and the quadratic objective and SINR constraints are replaced by first-order surrogate functions that are tight at the current iterate, yielding a convex problem. Auxiliary rotation matrices $Q_i$ are updated in closed form by aligning the clutter-whitened vector with the target-whitened vector. Alternating these updates produces a nondecreasing objective sequence, which is the convergence argument.

What would settle it

Solve the receive subproblem (58) for finite arrays with the paper's parameter table, reconstructing the optimal $W$ from Dinkelbach's solution, and check whether $\operatorname{rank}(W)=1$; a case with rank greater than one would break the claimed global optimality and the monotonicity chain (61), and a subsequent decrease in the objective across an outer iteration would directly falsify the convergence claim.

Watch

Extended reading notes

Core claim

The central claim is that the nonconvex problem P1, which maximizes the minimum SCNR over a finite set of possible target directions subject to per-user SINR constraints, a total power budget, and unit-norm receive combining, can be solved by alternating optimization with a provable convergence guarantee. For a fixed receive beamformer, the transmit subproblem is rendered convex through a penalty reformulation and successive convex approximation; for fixed transmit beamformers, the receive subproblem is shown to be a generalized fractional program whose matrices $B^l_{m,k}$ are Toeplitz, so Dinkelbach's algorithm yields a global optimum. The paper argues that because each subproblem is solved optimally or with a valid surrogate, the achieved objective value never decreases, and since it is bounded, Algorithm 1 converges. The numerical evidence further claims that one receive beamformer suffices: SCNR is almost flat in the number $I$ of possible target directions, and the receive beampattern is nearly identical for different angular spreads, so only the two extreme angles of the spread are needed for design.

Load-bearing premise

The convergence guarantee depends on the receive subproblem always being solved globally, which the paper obtains from the Toeplitz structure and Dinkelbach's method; if that global-solution or rank-one claim fails for a particular finite array and noise normalization, the monotonicity proof no longer applies.

Editorial extensions

If this is right

  • A DFRC base station can use a single receive beamformer for target detection even when the target angle is uncertain, reducing receiver hardware and signal-processing load compared with dedicated beamformers per candidate direction.
  • Designing the receive beamformer from the two extreme angles of the target's angular spread is enough; adding more candidate directions inside the spread barely changes SCNR.
  • Radar detection performance degrades gracefully as communication SINR requirements tighten, with larger antenna arrays absorbing most of the loss.
  • The convergence guarantee makes the algorithm suitable for online adaptation when the set of possible target directions changes over time.

Reading between the lines

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

  • An implicit consequence is that the algorithm's per-iteration cost is largely independent of the number of candidate target directions, since the receive beamformer is updated once per outer iteration; this could make the method attractive for tracking scenarios where the uncertainty set is refined over time.
  • The same Toeplitz-based global-solution step could be applied to other max-min beamforming problems in integrated sensing and communication, such as worst-case localization under angle uncertainty, by replacing SCNR with an estimation-theoretic objective.
  • Because the two extreme angles appear to determine the receive beampattern, a continuous uncertainty interval could be handled by sampling its endpoints, which is a testable bridge between the paper's discrete model and a continuous angular uncertainty set.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper considers a dual-functional radar and communication (DFRC) system in which the target arrival direction is not exactly known, but is known to belong to a discrete set of possible angles. The authors formulate a max-min radar SCNR problem subject to per-user SINR constraints and a total power constraint, and propose an alternating optimization algorithm: transmit beamformers are updated by a penalty-based successive convex approximation step, while the receive beamformer is updated by solving a fractional quadratic problem that they claim is a generalized fractional program with Toeplitz quadratics, solvable globally via Dinkelbach's method. The paper claims that the objective value is nondecreasing across iterations and hence that the algorithm converges, and presents numerical results showing convergence in about three iterations and robustness of the achieved SCNR to the number of possible target directions.

Significance. The problem is relevant and timely: target-direction uncertainty is a genuine limitation of prior DFRC beamforming designs, and the use of a single receive beamformer is an appealing complexity reduction relative to approaches that use one receive beamformer per candidate direction. The paper provides a clear problem formulation, a plausible algorithmic structure, and numerical experiments covering antenna counts, SINR thresholds, angular spread, and a comparison with dedicated receive beamformers. If the convergence and optimality claims can be rigorously supported, the contribution would be a useful design tool for DFRC receivers. However, as written, the central theoretical guarantee rests on two unproved steps: global optimality of the receive update and monotonicity of the true SCNR objective, rather than merely of a penalized auxiliary objective. These are load-bearing issues that prevent the convergence claim from being accepted as established.

major comments (3)
  1. [III-B, Eq. (58)] The global optimality of the receive-beamformer update is not established. The paper argues that every B^l_{m,k} is Toeplitz and then invokes [24] to conclude that (58) is a generalized fractional program with Toeplitz quadratics for which Dinkelbach's algorithm gives a global solution. But the monotonicity chain (61) requires that, at each outer iteration, the update of w globally solves the rank-one constrained problem (58). Toeplitzness of the coefficient matrices alone does not imply that the SDP relaxation obtained by dropping rank(W)=1 has a rank-one optimal solution: the relaxed problem maximizes a concave function (a minimum of linear forms) over the spectrahedron {W ⪰ 0, Tr(W)=1}, and such maxima can occur at non-rank-one points. The paper should identify the specific result in [24], verify that its hypotheses hold for the finite-dimensional ULA steering-vector matrices used here, and prove the required rank-one tightness, or otherwise weaken the convergence claim accordingly.
  2. [IV, Eqs. (59)-(62)] The convergence proof tracks an auxiliary penalized objective rather than the SCNR objective of P1. The transmit update solves (46), which maximizes a surrogate objective that includes the penalty term -νb and uses surrogate SINR and power constraints, while χ in (59) is defined without -νb and is not shown to equal the true max-min SCNR min_i γ_r(θ_i, w, u) at the iterates. Consequently, showing u^H_{d,s} \hat R u_{d,s} ≥ u^H_{d,s-1} \hat R u_{d,s-1} does not by itself imply that the original max-min SCNR is nondecreasing. The authors need to supply the missing chain from the penalized objective to the true P1 objective, or state the convergence theorem as one about the penalized problem only.
  3. [III-A, Eq. (46)] The power constraint is enforced only asymptotically in the penalty parameter ν. Problem (46) replaces u^H u = P by u^H u ≤ P + b and 2Re(u^H u_0) - ||u_0||^2 ≥ P - b, with the objective penalized by -νb. For a fixed finite ν chosen at initialization, the optimal b need not be zero, so the returned u may violate the power budget and hence be infeasible for P1. The statement that b* → 0 for very large ν requires proof, and the algorithm as given has no rule for increasing ν or for certifying feasibility. An increasing penalty schedule with a feasibility test, or an exact penalty reformulation, is needed to justify the inner-loop updates used in the convergence argument.
minor comments (6)
  1. [Eq. (10)] The notation is inconsistent: the terms in the numerator and denominator use α_i and α_j without the superscripts T and C introduced earlier, and the magnitudes should be written consistently as |α|^2. Please correct the notation in (9)-(10).
  2. [II-A to III-B] The transition from Eq. (10), where the noise term is σ_r w^H w, to Eq. (56), where the denominator contains I instead of σ_r I, silently assumes σ_r = 1. This normalization should be stated explicitly in the system model, since the numerical results depend on the noise scaling.
  3. [Throughout] The name Dinkelbach is repeatedly misspelled as "Dinkleback"; please correct this in the abstract, body, and references.
  4. [Algorithm 1] The algorithm initializes S, dmax, η, ν, µ, and ǫ, but Table 1 does not list their values, and the inner-loop count S is not described. Reporting these parameters and the stopping tolerance is necessary for reproducibility.
  5. [Footnote 2, after Eq. (33)] The footnote states that for a very large value of η the penalty term becomes zero, but for any finite η the penalty term is not exactly zero. The equivalence between (31) and (32) is therefore approximate, and the nature of this approximation should be stated explicitly.
  6. [Lemma 1] The proof of Lemma 1 is abbreviated: the sentence "Since the objective value does not decrease with each \bar γ_r(θ_i)" is unclear, and the proof should explicitly show that every individual SCNR term is nondecreasing in the common scaling of all transmit beamformers.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: receive-subproblem optimality relies on an external Dinkelbach result, and the only self-cited Toeplitz fact is independently verifiable from the ULA geometry.

full rationale

The paper's derivation chain is not circular. The central transmit-side subproblem is solved by MM/penalty methods, and the receive-side global-optimality claim rests on the external result [24], not on a conclusion the paper itself is trying to prove. The only self-cited ingredient that could be considered load-bearing is the statement in Section III-B that each B^l_{m,k} is a Toeplitz matrix, cited to [10]. That fact is independently verifiable and parameter-free: since A(theta)=a_r(theta)a_t^H(theta), one has B^l_{m,k}=A(theta)u_k u_k^H A^H(theta)=|a_t^H(theta)u_k|^2 a_r(theta)a_r^H(theta), and a_r(theta)a_r^H(theta) is Toeplitz for a ULA by inspection of its entries. Thus the citation is not a borrowed conclusion that the paper needs to establish. The convergence proof's monotonicity chain (61)-(62) depends on the receive subproblem being solved globally, but that dependency is a correctness or technicality issue (e.g., whether the rank-one relaxation of (58) is tight), not circularity: the conclusion is not defined in terms of its own input. Self-citations [10] and [17] provide background, motivation, and a precursor formulation, but the alternating-optimization algorithm and the convergence argument extend that prior work rather than reduce to it by construction. The numerical claims, such as convergence in about three iterations and SCNR stability as the number of possible target directions grows, are simulation results rather than fitted parameters renamed as predictions. No equation in the paper is shown to be equivalent to its own input by construction.

Assumptions & free parameters 5 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new physical entities. The central algorithm depends on a set of hand-chosen penalty parameters, an unspecified initialization, and two unverified domain assumptions: the global optimality of the Dinkelbach solution for the rank-relaxed Toeplitz problem and the tightness of the rank-one relaxation. These are the main unexamined inputs to the claimed convergence guarantee.

free parameters (5)
  • Penalty factor eta
    Large positive constant used in (32) to enforce equivalence between the fractional objective and the penalty form. The paper says eta goes to infinity but gives no finite value.
  • Penalty factor nu
    Large positive constant in (46) used to force b toward zero and recover the power equality. The convergence proof assumes nu goes to infinity.
  • Shift parameter mu
    Chosen in (36) so that \ hat R = \ mu I - R is positive semidefinite. The value is hand-picked and not specified.
  • Initial beamformers u0,S and w
    Algorithm 1 initializes u0,S through an unspecified communication-based resource allocation scheme, and the initial receive beamformer is not described. The local optimum depends on this choice.
  • Stopping tolerance and iteration caps
    The algorithm uses epsilon, dmax and S in the convergence check, but their values are not reported.
assumptions (5)
  • standard math MM surrogate inequality for convex quadratic functions
    Invoked in (39) to construct a lower-bound surrogate for u^H \ hat R u and for the SINR numerators.
  • standard math Pointwise minimum of concave functions is concave
    Used to declare the lambda subproblem (54) convex and globally solvable.
  • domain assumption B^l_m,k are Toeplitz matrices and generalized fractional programs with Toeplitz quadratics can be globally solved via Dinkelbach
    Taken from [24] and used to solve (58) globally, which the monotonicity proof requires at every outer iteration.
  • domain assumption Rank-one relaxation of W = w w^H is tight for the Toeplitz fractional program
    The paper drops the rank constraint in (58) without proving tightness for its finite-array, noise-scaled parameter settings.
  • ad hoc to paper Noise covariance in the radar receive chain is normalized so that sigma_r can be replaced by 1 or N_r
    Equations (27) and (56) use N_r/P I and +I, while the system model in (10) contains sigma_r w^H w. This hidden normalization is not stated in the system model.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Detection with Uncertainty in Target Direction for Dual Functional Radar and Communication Systems." pith.science (2026). https://pith.science/paper/ZEIK5MXW

@misc{pith2026241207245,
  author       = {Pith},
  title        = {Pith review of: Detection with Uncertainty in Target Direction for Dual Functional Radar and Communication Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZEIK5MXW}},
  note         = {Machine review of arXiv:2412.07245}
}
read the original abstract

Dual functional radar and communication (DFRC) systems are a viable approach to extend the services of future communication systems. Most studies designing DFRC systems assume that the target direction is known. In our paper, we address a critical scenario where this information is not exactly known. For such a system, a signal-to-clutter-plus-noise ratio (SCNR) maximization problem is formulated. Quality-of-service constraints for communication users (CUs) are also incorporated as constraints on their received signal-to-interference-plus-noise ratios (SINRs). To tackle the nonconvexity, an iterative alternating optimization approach is developed where, at each iteration, the optimization is alternatively performed with respect to transmit and receive beamformers. Specifically, a penalty-based approach is used to obtain an efficient sub-optimal solution for the resulting subproblem with regard to transmit beamformers. Next, a globally optimal solution is obtained for receive beamformers with the help of the Dinkleback approach. The convergence of the proposed algorithm is also proved by proving the nondecreasing nature of the objective function with iterations. The numerical results illustrate the effectiveness of the proposed approach. Specifically, it is observed that the proposed algorithm converges within almost 3 iterations, and the SCNR performance is almost unchanged with the number of possible target directions.

Figures

Figures reproduced from arXiv: 2412.07245 by the authors.

Figure 1
Figure 1. The presumed multiuser downlink integrated communi [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Achieved SCNR as a function of CU SINR thresholds for d [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 4
Figure 4. Achieved SCNR for different I and the angular spread of target directions. -200 -150 -100 -50 0 50 100 150 200 Angle (degrees) -140 -120 -100 -80 -60 -40 -20 0 Beampattern (dBi) for angular gap 15-50 degrees for angular gap 15-40 degrees [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Receiver’s beampattern for the different angular sp [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]
Figure 6
Figure 6. Figure 6: Receiver’s beampattern for the proposed scheme and t [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 22 canonical work pages

  1. [24]

    New results on g eneralized fractional programming problems with Toeplitz quadratics ,

    A. Aubry, V . Carotenuto, and A. D. Maio, “New results on g eneralized fractional programming problems with Toeplitz quadratics ,” IEEE Sig. Proc. Lett., vol. 23, no. 6, pp. 848–852, 2016

  2. [18]

    Efficient transcei ver design for MIMO dual-function radar-communication systems,

    C. Wen, Y . Huang, and T. N. Davidson, “Efficient transcei ver design for MIMO dual-function radar-communication systems,” IEEE Trans. Sig. Proc., vol. 71, pp. 1786–1801, 2023

  3. [17]

    Maximizing detection of target with multiple direction po ssibilities to support immersive communications in Metaverse,

    M. Ashraf, A. Gaydamaka, D. Moltchanov, A. Mohammad, an d B. Tan, “Maximizing detection of target with multiple direction po ssibilities to support immersive communications in Metaverse,” in Proc. of the 2nd W orkshop on Integrated Sensing and Communications for Meta verse, ser. ISACom ’23. New Y ork, NY , USA: ACM, 2023, p. 30–35. [Online]. Available: ht...

  4. [1]

    Sensing as a Service in 6G Perceptive Mobile Networks: Architecture, Advances, and the Road Ahead

    F. Dong, F. Liu, Y . Cui, S. Lu, and Y . Li, “Sensing as a Servi ce in 6G Perceptive Mobile Networks: Architecture, Advances, an d the Road Ahead,” arXiv:2308.08185, 2023

  5. [2]

    A h ybrid approach to optimal TOA-sensor placement with fixed shared s ensors for simultaneous multi-target localization,

    S. Xu, L. Wu, K. Do˘ ganc ¸ay, and M. Alaee-Kerahroodi, “A h ybrid approach to optimal TOA-sensor placement with fixed shared s ensors for simultaneous multi-target localization,” IEEE Trans. Sig. Proc. , vol. 70, pp. 1197–1212, 2022

  6. [3]

    Adaptive radar detection in Gaussian interference using clutter-free training data ,

    Y . Rong, A. Aubry, A. De Maio, and M. Tang, “Adaptive radar detection in Gaussian interference using clutter-free training data ,” IEEE Trans. Sig. Proc., vol. 70, pp. 978–993, 2022

  7. [4]

    Cra m´ er-Rao bound optimization for joint radar-communication beamfor ming,

    F. Liu, Y .-F. Liu, A. Li, C. Masouros, and Y . C. Eldar, “Cra m´ er-Rao bound optimization for joint radar-communication beamfor ming,” IEEE Trans. Sig. Proc. , vol. 70, pp. 240–253, 2022

  8. [5]

    Antenna selection for energy-efficient dual-functional radar-communication sy stems,

    I. V aliulahi, C. Masouros, A. Salem, and F. Liu, “Antenna selection for energy-efficient dual-functional radar-communication sy stems,” IEEE Wirel. Commun. Lett. , pp. 1–1, 2022

Show all 24 references
  1. [6]

    In- tegrated sensing, communication, and computation over-th e-air: MIMO beamforming design,

    X. Li, F. Liu, Z. Zhou, G. Zhu, S. Wang, K. Huang, and Y . Gong , “In- tegrated sensing, communication, and computation over-th e-air: MIMO beamforming design,” arXiv:2201.12581, 2022

  2. [7]

    Design of phase codes for radar performance optimization with a sim ilarity constraint,

    A. De Maio, S. De Nicola, Y . Huang, Z.-Q. Luo, and S. Zhang, “Design of phase codes for radar performance optimization with a sim ilarity constraint,” IEEE Trans. Sig. Proc. , vol. 57, no. 2, pp. 610–621, 2009

  3. [8]

    Adaptive wave form design for communication-enabled automotive radars,

    C. D. Ozkaptan, E. Ekici, and O. Altintas, “Adaptive wave form design for communication-enabled automotive radars,” IEEE Trans. Wirel. Comm., pp. 1–1, 2021

  4. [9]

    Composite signa lling for DFRC: Dedicated probing signal or not?

    L. Chen, F. Liu, J. Liu, and C. Masouros, “Composite signa lling for DFRC: Dedicated probing signal or not?” arXiv:2009.03528, 2020

  5. [10]

    Joint optimization of radar and communications performan ce in 6G cellular systems,

    M. Ashraf, B. Tan, D. Moltchanov, J. S. Thompson, and M. V alkama, “Joint optimization of radar and communications performan ce in 6G cellular systems,” IEEE Trans. Green Commun. Network. , vol. 7, no. 1, pp. 522–536, 2023. 11

  6. [11]

    Joint radar-c ommunication transmission: A generalized Pareto optimization framewor k,

    L. Chen, F. Liu, W. Wang, and C. Masouros, “Joint radar-c ommunication transmission: A generalized Pareto optimization framewor k,” IEEE Trans. Sig. Proc. , vol. 69, pp. 2752–2765, 2021

  7. [12]

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

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

  8. [13]

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

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

  9. [14]

    To- ward dual-dunctional radar-communication systems: Optim al waveform design,

    F. Liu, L. Zhou, C. Masouros, A. Li, W. Luo, and A. Petropu lu, “To- ward dual-dunctional radar-communication systems: Optim al waveform design,” IEEE Trans. Sig. Proc. , vol. 66, no. 16, pp. 4264–4279, 2018

  10. [15]

    MIMO-OFDM dua l- functional radar-communication systems: Low-PAPR waveform design,

    X. Hu, C. Masouros, F. Liu, and R. Nissel, “MIMO-OFDM dua l- functional radar-communication systems: Low-PAPR waveform design,” 2021

  11. [16]

    Maximin joint optimization o f transmitting code and receiving filter in radar and communications,

    L. Zhao and D. P . Palomar, “Maximin joint optimization o f transmitting code and receiving filter in radar and communications,” IEEE Trans. Sig. Proc., vol. 65, no. 4, pp. 850–863, 2017

  12. [19]

    MIMO radar waveform de sign with constant modulus and similarity constraints,

    G. Cui, H. Li, and M. Rangaswamy, “MIMO radar waveform de sign with constant modulus and similarity constraints,” IEEE Trans. Sig. Proc. , vol. 62, no. 2, pp. 343–353, 2014

  13. [20]

    Grab-n-Pull: A max-min fractional quadratic programming framework with applications in signal and information proc essing,

    A. Gharanjik, M. Soltanalian, M. R. B. Shankar, and B. Ot tersten, “Grab-n-Pull: A max-min fractional quadratic programming framework with applications in signal and information proc essing,” Sig. Proc. , vol. 160, pp. 1–12, 2019. [Online]. Available: https://www.sciencedirect...

  14. [21]

    A Tutorial on MM Algorithms,

    D. R. Hunter and K. Lange, “A Tutorial on MM Algorithms,” The American Statistician , vol. 58, no. 1, pp. 30–37, 2004. [Online]. Available: https://doi.org/10.1198/0003130042836

  15. [22]

    Optimization trans fer using surrogate objective functions,

    K. Lange, D. R. Hunter, and I. Y ang, “Optimization trans fer using surrogate objective functions,” J. Comp. Graph. Stat. , vol. 9, no. 1, pp. 1–20, 2000. [Online]. Available: http://www.jstor.org/s table/1390605

  16. [23]

    Cyclic minimizers, majorizati on techniques, and the expectation-maximization algorithm: a refresher,

    P . Stoica and Y . Selen, “Cyclic minimizers, majorizati on techniques, and the expectation-maximization algorithm: a refresher,” IEEE Sig. Proc. Mag., vol. 21, no. 1, pp. 112–114, 2004

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.