FFT-Enhanced Low-Complexity Near-Field Super-Resolution Sensing

Huizhi Wang; Yong Zeng; Yuxiao Wu

REVIEW 4 major objections 4 minor 12 references

FFT-Enhanced Low-Complexity Near-Field Super-Resolution Sensing

T0 review · 4 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read A coarse-to-fine FFT-pruned search achieves near-field 2D-MUSIC accuracy at a fraction of the runtime, with no extra antenna-spacing constraints.

desk verdict Plausible complexity-reduction for near-field 2D-MUSIC with a real speedup, but the unproven containment guarantee and unspecified thresholds keep it conditional. read the letter →

arxiv 2411.15532 v1 pith:ZBTBAOVF submitted 2024-11-23 eess.SP

Yuxiao Wu , Huizhi Wang , Yong Zeng This is my paper · ORCID

classification eess.SP

keywords FFT-MUSICnear-fieldsourcelocalizationsuper-resolutionsensing2D-MUSICangleanddistanceestimationspectrumpeaksearchextremelylarge-scaleMIMObeamforming

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 claims that near-field source localization—estimating both angle and distance from an extremely large array—can be made cheap without giving up resolution. The proposed algorithm scans the angle domain with an FFT to discard most of the search space, then uses 1D beamforming at the edges of each surviving angle cluster to narrow the distance range, and finally runs the full 2D-MUSIC peak search only inside the remaining small angle-distance boxes. The result is that the dominant $n_\theta n_r (M-K)(M+1)$ spectrum-search cost shrinks to $L n'_\theta n'_r (M-K)(M+1)$ with $n'_\theta \ll n_\theta$ and $n'_r \ll n_r$, and the numerical example reports a runtime of 3.38 seconds against 22.07 seconds for full 2D-MUSIC at the same localization RMSE. A practical payoff is that the method keeps the standard half-wavelength antenna spacing, which reduced-rank and reduced-dimension MUSIC cannot do.

What carries the argument

The central objects are angle clusters and distance clusters. An angle cluster is a contiguous block of FFT-spectrum samples exceeding the threshold $\Gamma_\theta = \min(\{p_\Theta\})-\delta_\theta$; it is designed to contain any group of sources whose energy spreads across adjacent bins. A distance cluster is the intersection $R_n = \beta_n \cap \overline{\beta}_n$ of the distance sets found by beamforming scans at the two angular bounds $\underline{\alpha}_n$ and $\overline{\alpha}_n$ of an angle cluster, and the paper asserts that the true distance of every 'close' source lies inside this intersection. These clusters work by shrinking the 2D-MUSIC search domain from the whole $n_\theta n_r$ grid to $L n'_\theta n'_r$ points, turning a search whose cost is comparable to $M^4$ into one comparable to $2MS\log_2(S) + 2Ln_rM^2$ plus a tiny final refinement.

What would settle it

Pick the paper's configuration ($M=512$, $\lambda/2$ spacing, four sources, SNR from 5 to 25 dB) and run many random source placements; record whether every true angle lies inside some angle cluster and every true distance lies inside the corresponding $R_n = \beta_n \cap \overline{\beta}_n$. If any trial yields a cluster count below $K$ or a true range outside the intersection, then the coarse stages are not lossless and the claim of matching full 2D-MUSIC accuracy fails for that scenario.

Watch

Extended reading notes

Core claim

The central claim is that a coarse-to-fine spectral search solves the near-field joint angle-and-distance localization problem at a fraction of the cost of 2D-MUSIC while retaining its resolution. The algorithm first applies an FFT/IFFT to the sample covariance matrix to form an angle spectrum, groups spectral peaks above a threshold into angle clusters, and classifies clusters as containing either a single 'distant' source or multiple 'close' sources. For close-source clusters, a pair of 1D beamforming scans at the cluster's two angular bounds produces distance sets, and the intersection of those sets is taken as the distance cluster that must contain the true ranges; a final 2D-MUSIC inside each angle-distance cluster yields precise and automatically paired angle-distance estimates. The paper's numerical evidence shows localization RMSE matching full 2D-MUSIC while running in 3.38 s versus 22.07 s, with total complexity $O[M^3 + M^2J + 2MS\log_2(S) + 2Ln_rM^2 + L n'_\theta n'_r (M-K)(M+1)]$.

Load-bearing premise

The entire speedup rests on the claim that the coarse FFT angle scan and the beamforming distance scans never discard a real source: if a true angle falls below the cluster threshold, or a true distance falls outside the intersection $R_n$, the final refined search cannot recover it, and the paper gives no proof or miss-probability bound for these inclusions.

Editorial extensions

If this is right

The dominant 2D spectrum-search complexity drops from $O[n_\theta n_r (M-K)(M+1)]$ to $O[L n'_\theta n'_r (M-K)(M+1)]$, which the numerical example realizes as a runtime reduction from 22.07 s to 3.38 s.
Because angle and distance are estimated directly through the near-field steering vector without Taylor-approximation decoupling, the algorithm remains valid at $\lambda/2$ antenna spacing, where RR-MUSIC and RD-MUSIC fail.
The final super-resolution stage is exactly 2D-MUSIC restricted to a small box, so the resolution and the automatic angle-distance pairing within that box match full 2D-MUSIC.
For 'distant' sources whose FFT spectrum is a sharp single peak, the algorithm outputs angle and distance immediately, so only the close sources pay the refined 2D-MUSIC cost.

Reading between the lines

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

The inclusion guarantees for the coarse stages are the true load-bearing assumption; a miss-probability analysis for $\Gamma_\theta$ and $\delta_d$ across SNR would tell how generally the speedup holds.
The saving depends on the number $L$ of angle clusters and the sizes $n'_\theta, n'_r$: scenes with many well-separated sources will dilute the advantage, so a scaling law in $K$ would sharpen the complexity claim.
The same coarse-to-fine pruning could be transplanted to other high-resolution estimators or to polar-domain beam training codebooks, since the FFT angle scan and the distance-cluster intersection do not rely on the specific final estimator.

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

Honest take: this is a plausible complexity-reduction trick for near-field 2D-MUSIC, and the simulation shows a real speedup with matching accuracy in one favorable configuration. But the paper never proves the coarse stages cannot discard a true source, and the threshold parameters are left unspecified. That gap is the difference between a conditional accept and a reject.

What's new: adapting the FFT coarse search plus distance-domain beamforming idea from far-field [11,12] to spherical wavefronts, and doing so at half-wavelength spacing, which RR-MUSIC and RD-MUSIC cannot handle. The complexity analysis is explicit, and the runtime table shows 3.38 s versus 22.07 s for full 2D-MUSIC. The RMSE curves in Fig. 5 show the proposed scheme tracking 2D-MUSIC across SNR, with RD/RR-MUSIC failing at d=λ/2 as expected. That is a real, useful result for XL-MIMO near-field sensing.

Where it gets soft: the load-bearing step is the angle-cluster and distance-cluster containment. Step 1 sets Γθ = min({pΘ})−δθ, which only guarantees that the detected peaks are included; if a true source is weak enough to not appear as a peak, or its spectral response falls below Γθ, it is excluded permanently. Step 4 can widen distance clusters but cannot re-open an angle cluster. Similarly, Eq. (11) assumes the true range lies in β_n ∩ β_n because off-boresight beam patterns have local minima there; the paper gives no derivation, and the parameters δ_θ, δ_d are never assigned values or given a selection rule. Without a miss probability or sensitivity analysis, the central claim that the refined stage matches full 2D-MUSIC is only conditional on assumptions that may fail. Also, Eq. (8)'s diag(W R W^{-1}) is notationally odd—normally one would see a DFT on the array covariance, such as W R W^H—and no code or data are provided, with a single simulation setup and no error bars on the RMSE.

Who it is for: someone working on near-field localization/ISAC for XL-MIMO who wants a low-complexity alternative to 2D-MUSIC and is willing to test the thresholds themselves. The idea is worth engaging with; the paper is not ready to be taken as a finished robustness claim.

Recommendation: send it to peer review. The gap is addressable—supply threshold rules, a detection guarantee or at least a sensitivity sweep, and ideally code—and the relevance to 6G sensing is high enough that a serious referee should look at it.

Referee Report

4 major / 4 minor

Summary. This letter proposes a low-complexity near-field source localization algorithm. It first forms an FFT-based angle spectrum (Eq. (8)) and partitions the angle domain into clusters using a threshold Gamma_theta = min({p_Theta}) - delta_theta, then uses 1D beamforming at the cluster boundaries (Eq. (10)) to form distance clusters R_n = beta_n intersect beta-bar_n (Eq. (11)), and finally applies 2D-MUSIC only inside the surviving angle-distance cells. The reported simulation for one geometry with four sources shows RMSE comparable to full 2D-MUSIC with runtime 3.38 s versus 22.07 s.

Significance. If the inclusion property of the coarse stages can be rigorously established, the proposal would be a useful complexity-reduction scheme for near-field localization, with the practical advantage of not requiring d <= lambda/4. The complexity analysis is explicit, the simulation setup is clearly described, and the authors correctly identify that the final stage is the standard 2D-MUSIC estimator. However, the central claim currently rests on unproved containment assumptions for the angle and distance clusters, and the threshold parameters delta_theta and delta_d are not specified; these issues must be resolved before the complexity/accuracy trade-off can be considered established.

major comments (4)

[Section IV, Step 1, Eq. (8)] The spectral formula p_theta = diag(W R W^{-1}) is undefined as written because W is an S x S DFT matrix while R is M x M (M=512 and S=1024 in the simulations); the multiplication is not conformable unless zero-padding or resizing is implicitly assumed, which the manuscript never states. The text says FFT and IFFT are applied to column and row vectors, but the displayed matrix expression does not correspond to that description and no derivation from the near-field steering model in Eqs. (1)-(2) is supplied.
[Section IV, Steps 1 and 3] The central complexity reduction is valid only if every true source angle theta_k lies in one of the angle clusters alpha_n and every true distance r_k lies in the corresponding distance cluster R_n. The manuscript provides no proof, no miss probability, and no rule for selecting delta_theta and delta_d; it only states that thresholds are set as Gamma_theta = min({p_Theta}) - delta_theta and pr <= min{pr} + delta_d. Because Step 4 can only expand distance clusters and never re-opens angle clusters, a source whose angle is excluded in Step 1 is irrecoverable, so the comparison with full 2D-MUSIC in Fig. 5 and the runtime in Table I are conditional on an unverified containment property.
[Section IV, Step 2, Eq. (11)] The assertion that the off-boresight beam patterns P_low(r) and P_up(r) exhibit local minima at the true distance is supported only by a qualitative comment and Fig. 3; no derivation from the steering-vector model in Eqs. (1)-(2) is given. It is therefore not established that beta_n and beta-bar_n are nonempty, that their intersection contains r_k, or that the distance cluster R_n = beta_n intersect beta-bar_n always contains the true source, which is the load-bearing condition for Step 3.
[Section V] The numerical validation uses a single favorable configuration (four sources with angles 6, 7, 8, and 20 degrees and distances 3, 4, 5, and 32 m) and does not report the values of delta_theta and delta_d, the number of Monte Carlo trials, error bars, or the fraction of runs in which all true sources were contained in the clusters. The statement that the proposed algorithm can theoretically attain the same performance as 2D-MUSIC is true by construction once the clusters are correct, since the final stage is the same estimator; what must be demonstrated is that the coarse stages do not discard true sources, and the current results do not substantiate that over a range of geometries or SNRs.

minor comments (4)

[Algorithm 1, line 16] The text says 'Sum the number of close sources Kclose as well as distant sources Kclose'; the second symbol should be Kdistant.
[Fig. 5] The axis labels appear to be inconsistent with the captions: panel (a) is titled 'RMSE of distance versus SNR' but the y-axis is labeled 'RMSE(°)', while panel (b) is 'RMSE of angle versus SNR' with y-axis 'RMSE(m)'.
[Section III] The statement that the search complexity is 'O[M^4]' is imprecise; the complexity is O[n_theta n_r (M-K)(M+1)], which scales as O[M^4] only if n_theta n_r is proportional to M^2, and this scaling should be stated explicitly.
[Section IV, Eq. (8)] If W is the unnormalized DFT matrix, then W^{-1} is (1/S) W^H, and the normalization convention should be stated to avoid ambiguity in the diagonal extraction.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation found: the coarse-to-fine search is a genuine algorithmic decomposition, and the unproved cluster-containment guarantee is a completeness risk, not a circular step.

full rationale

The paper's central claim is a complexity reduction: replace the full 2D-MUSIC peak search over n_theta x n_r grids with an FFT-based angle clustering stage, a beamforming-based distance clustering stage, and a refined 2D-MUSIC search over the narrowed clusters. No fitted constant is used to produce a prediction, and no parameter is estimated from the benchmark outcomes and then renamed as a result. The final localization stage is literally the same 2D-MUSIC estimator restricted to candidate clusters, so the statement that the proposed algorithm 'can theoretically attain the same performance as the 2D-MUSIC algorithm' (Section V) is a design property of the algorithm, not a circular derivation: the burden is on the coarse stages to not discard true sources, which is an algorithmic correctness question, not a circularity question. The load-bearing assumptions - that the FFT threshold Gamma_theta yields angle clusters containing all K sources, and that Eq. (11), R_n = beta_n intersection beta-bar_n, contains the true distance - are asserted on physical grounds about near-field beam patterns without proof, miss-probability analysis, or specified threshold values delta_theta and delta_d. This is an internal completeness/reproducibility gap, and Step 4 can only expand distance clusters, never restore a lost angle cluster. But gap is not circularity: the assertions are not derived from the paper's own outputs or from citations that reduce to the present claim. The self-citations ([1], [2], [3], [9]) are background/tutorial references and do not carry a load-bearing premise; no uniqueness theorem from the authors' prior work is invoked to forbid alternatives. Accordingly, no pattern from the circularity checklist applies, and the honest finding is no significant circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The algorithm rests on standard signal model assumptions plus two unstated threshold parameters and two heuristic inclusion assumptions. No new physical entities are introduced; the 'angle cluster' and 'distance cluster' are algorithmic constructs, not physical postulates.

free parameters (2)

delta_theta
Spread threshold for angle cluster formation, defined as Gamma_theta = min({p_Theta}) - delta_theta in Section IV Step 1. No value or adaptive rule is given; it controls which FFT peaks are kept and can exclude true angles if set too high.
delta_d
Distance threshold for selecting candidate ranges via p_r <= min{p_r} + delta_d in Section IV Step 2. Not specified; it controls the width of distance clusters and the success of the intersection step.

assumptions (4)

standard math The sample covariance matrix R = YY^H/J approximates the true covariance AR_s A^H + sigma^2 I for large snapshot number J (Eq. 5).
Used as the basis for eigenvalue decomposition and all spectrum functions; this is a standard law of large numbers argument.
domain assumption The near-field steering vector model with spherical wavefront and phase a_n(r,theta) = exp(-j 2 pi / lambda (r_n - r_k)) exactly describes the received signals (Eqs. 1 and 2).
The problem is defined by this model, and no model mismatch or calibration error is analyzed.
ad hoc to paper The FFT-based angle spectrum and threshold Gamma_theta produce angle clusters that contain every true source angle.
No detection guarantee or missed-angle probability is analyzed; if a source is excluded, the refined MUSIC stage cannot recover it.
ad hoc to paper For each multi-peak angle cluster, the true distance lies in the intersection R_n = beta_n intersect beta_n of the low-beamforming regions computed from the cluster bounds (Eq. 11).
This is the core narrowing assumption; it is illustrated in Fig. 3 but not proved, and Step 4's expansion rule is unspecified.

how reviews work

0 comments

Cite this review

Pith. "Pith review of FFT-Enhanced Low-Complexity Near-Field Super-Resolution Sensing." pith.science (2026). https://pith.science/paper/ZBTBAOVF

@misc{pith2026241115532,
  author       = {Pith},
  title        = {Pith review of: FFT-Enhanced Low-Complexity Near-Field Super-Resolution Sensing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZBTBAOVF}},
  note         = {Machine review of arXiv:2411.15532}
}

read the original abstract

In this letter, a fast Fourier transform (FFT)-enhanced low-complexity super-resolution sensing algorithm for near-field source localization with both angle and range estimation is proposed. Most traditional near-field source localization algorithms suffer from excessive computational complexity or incompatibility with existing array architectures. To address such issues, this letter proposes a novel near-field sensing algorithm that combines coarse and fine granularity of spectrum peak search. Specifically, a spectral pattern in the angle domain is first constructed using FFT to identify potential angles where sources are present. Afterwards, a 1D beamforming is performed in the distance domain to obtain potential distance regions. Finally, a refined 2D multiple signal classification (MUSIC) is conducted within each narrowed angle-distance region to estimate the precise location of the sources. Numerical results demonstrate that the proposed algorithm can significantly reduce the computational complexity of 2D spectrum peak searches and achieve target localization with high-resolution.

Figures

Figures reproduced from arXiv: 2411.15532 by the authors.

**Figure 3.** Illustration of how to determine distance clusters [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗

**Figure 4.** The performance of source localization in the near-field region. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗

Figures from the paper (1 more)

**Figure 5.** Figure 5: Comparison of localization accuracy of different near-field localization [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

12 extracted references · 11 canonical work pages

[1]

A tutorial on near-field xl-mimo communications towards 6g,

H. Lu, Y . Zeng, C. You, Y . Han, J. Zhang, Z. Wang, Z. Dong, S. Jin, C.-X. Wang, T. Jiang, X. You, and R. Zhang, “A tutorial on near-field xl-mimo communications towards 6g,” IEEE Commun. Surveys Tuts. , 2024

work page 2024
[2]

Cram ´er-rao bounds for near-field sensing with extremely large-scale mimo,

H. Wang, Z. Xiao, and Y . Zeng, “Cram ´er-rao bounds for near-field sensing with extremely large-scale mimo,” IEEE Trans. on Signal Process., vol. 72, pp. 701–717, 2024

work page 2024
[3]

Integrated super-resolution sensing and communication with 5g nr waveform: Signal processing with uneven cps and experiments: (invited paper),

C. Zhang, Z. Zhou, H. Wang, and Y . Zeng, “Integrated super-resolution sensing and communication with 5g nr waveform: Signal processing with uneven cps and experiments: (invited paper),” in 2023 21st Inter- national Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks (WiOpt) , pp. 681–688, 2023

work page 2023
[4]

Near-field multiple source localization by passive sensor array,

Y .-D. Huang and M. Barkat, “Near-field multiple source localization by passive sensor array,” IEEE Trans. Antennas Propag. , vol. 39, no. 7, pp. 968–975, 1991

work page 1991
[5]

Second-order near-field source localization: algorithm and performance analysis,

K. Abed-Meraim, Y . Hua, and A. Belouchrani, “Second-order near-field source localization: algorithm and performance analysis,” in Conference Record of The Thirtieth Asilomar Conference on Signals, Systems and Computers, vol. 1, pp. 723–727 vol.1, 1996

work page 1996
[6]

3-d near field source localization using second order statistics,

K. Abed-Meraim and Y . Hua, “3-d near field source localization using second order statistics,” in Conference Record of the Thirty- First Asilomar Conference on Signals, Systems and Computers (Cat. No.97CB36136), vol. 2, pp. 1307–1311, 1997

work page 1997
[7]

Passive localization of noncircular sources in the near-field,

J. Xie, H. Tao, X. Rao, and J. Su, “Passive localization of noncircular sources in the near-field,” in 2015 16th International Radar Symposium (IRS), pp. 493–498, 2015

work page 2015
[8]

Localization of near-field sources: A reduced-dimension music algorithm,

X. Zhang, W. Chen, W. Zheng, Z. Xia, and Y . Wang, “Localization of near-field sources: A reduced-dimension music algorithm,” IEEE Commun. Lett. , vol. 22, no. 7, pp. 1422–1425, 2018

work page 2018

Show all 12 references

[9]

Sparse MIMO for ISAC: New Opportunities and Challenges,

X. Li, H. Min, Y . Zeng, S. Jin, L. Dai, Y . Yuan, and R. Zhang, “Sparse MIMO for ISAC: New Opportunities and Challenges,” arXiv e-prints , p. arXiv:2406.12270, June 2024

2024 arXiv
[10]

Fast near-field beam training for extremely large-scale array,

Y . Zhang, X. Wu, and C. You, “Fast near-field beam training for extremely large-scale array,” IEEE Wireless Commun. Lett. , vol. 11, no. 12, pp. 2625–2629, 2022

2022
[11]

Signal processing for passive radar using ofdm waveforms,

C. R. Berger, B. Demissie, J. Heckenbach, P. Willett, and S. Zhou, “Signal processing for passive radar using ofdm waveforms,” IEEE J. Sel. Top. Signal Process. , vol. 4, no. 1, pp. 226–238, 2010

2010
[12]

Frequency estimation based on modulation fft and music algorithm,

T. Li and Y . Tang, “Frequency estimation based on modulation fft and music algorithm,” in 2010 First International Conference on Pervasive Computing, Signal Processing and Applications , pp. 525–528, 2010

2010

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.

[1] [1]

A tutorial on near-field xl-mimo communications towards 6g,

H. Lu, Y . Zeng, C. You, Y . Han, J. Zhang, Z. Wang, Z. Dong, S. Jin, C.-X. Wang, T. Jiang, X. You, and R. Zhang, “A tutorial on near-field xl-mimo communications towards 6g,” IEEE Commun. Surveys Tuts. , 2024

work page 2024

[2] [2]

Cram ´er-rao bounds for near-field sensing with extremely large-scale mimo,

H. Wang, Z. Xiao, and Y . Zeng, “Cram ´er-rao bounds for near-field sensing with extremely large-scale mimo,” IEEE Trans. on Signal Process., vol. 72, pp. 701–717, 2024

work page 2024

[3] [3]

Integrated super-resolution sensing and communication with 5g nr waveform: Signal processing with uneven cps and experiments: (invited paper),

C. Zhang, Z. Zhou, H. Wang, and Y . Zeng, “Integrated super-resolution sensing and communication with 5g nr waveform: Signal processing with uneven cps and experiments: (invited paper),” in 2023 21st Inter- national Symposium on Modeling and Optimization in Mobile, Ad Hoc, and Wireless Networks (WiOpt) , pp. 681–688, 2023

work page 2023

[4] [4]

Near-field multiple source localization by passive sensor array,

Y .-D. Huang and M. Barkat, “Near-field multiple source localization by passive sensor array,” IEEE Trans. Antennas Propag. , vol. 39, no. 7, pp. 968–975, 1991

work page 1991

[5] [5]

Second-order near-field source localization: algorithm and performance analysis,

K. Abed-Meraim, Y . Hua, and A. Belouchrani, “Second-order near-field source localization: algorithm and performance analysis,” in Conference Record of The Thirtieth Asilomar Conference on Signals, Systems and Computers, vol. 1, pp. 723–727 vol.1, 1996

work page 1996

[6] [6]

3-d near field source localization using second order statistics,

K. Abed-Meraim and Y . Hua, “3-d near field source localization using second order statistics,” in Conference Record of the Thirty- First Asilomar Conference on Signals, Systems and Computers (Cat. No.97CB36136), vol. 2, pp. 1307–1311, 1997

work page 1997

[7] [7]

Passive localization of noncircular sources in the near-field,

J. Xie, H. Tao, X. Rao, and J. Su, “Passive localization of noncircular sources in the near-field,” in 2015 16th International Radar Symposium (IRS), pp. 493–498, 2015

work page 2015

[8] [8]

Localization of near-field sources: A reduced-dimension music algorithm,

X. Zhang, W. Chen, W. Zheng, Z. Xia, and Y . Wang, “Localization of near-field sources: A reduced-dimension music algorithm,” IEEE Commun. Lett. , vol. 22, no. 7, pp. 1422–1425, 2018

work page 2018