Pith. sign in

REVIEW 4 major objections 8 minor 27 references

A Low-Cost Monopulse Receiver with Enhanced Estimation Accuracy Via Deep Neural Network

T0 review · 4 major / 8 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 Moving one input port with a 360° phase-delay crossover makes the rat-race comparator planar and symmetric on one PCB layer, and a small neural network then cuts validated position error to about 1.1 mm at 0.62 m.

desk verdict The planar comparator network is a solid, measured hardware contribution; the DNN accuracy claim is an under-validated calibration fit and should not be accepted as-is. read the letter →

arxiv 2411.17734 v1 pith:LWQEIUNU submitted 2024-11-22 eess.SP

Hanxiang Zhang , Saeed Zolfaghary Pour , Hao Yan , Powei Liu , Bayaner Arigong This is my paper · ORCID
classification eess.SP
keywords monopulsereceiverdirectionofarrivalestimationrat-racecouplermicrowavecrossoverplanarcomparatornetworkdeepneuralpatchantennaarray
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 tries to show that a monopulse tracking receiver—one that locates a target by comparing sum and difference signals from an antenna array—can be built on low-cost hardware. The analog comparator network that forms those sum and difference signals is redesigned so it lies flat on a single PCB layer, with no crossing lines, multilayer board, or backside feed. The redesign uses four port-transformation rat-race couplers in which a 360° phase-delay crossover relocates one input port, and a measured 1.95 GHz prototype keeps amplitude imbalance within 0.5 dB and phase imbalance within 10° over about 10% bandwidth. On top of this hardware, a small fully connected neural network takes the raw azimuth, elevation, and distance estimates and corrects them toward the true target coordinates, reaching a validation position error near 1.1 mm at 0.62 m range after training on only 145 samples. If correct, this gives a path to low-cost, real-time monopulse tracking without expensive fabrication steps.

What carries the argument

The load-bearing component is the port-transformation 180° rat-race coupler, a modified hybrid ring that normally combines two inputs into sum and difference outputs but whose ports cross in a planar layout. The mathematical fact that carries the design is that in even/odd-mode analysis the 180° phase shifter is open or short, so its impedance $Z_\eta$ drops out of the ABCD matrices and can be chosen freely; the crossover itself is designed from four-way even/odd input impedances to satisfy $|S_{xx'}| = |S_{yy'}| = 1$ with zero leakage between the other port pairs. This identity lets four modified couplers and another crossover tile into a symmetric $1\lambda \times 1\lambda$ planar comparator. The second mechanism is a fully connected neural network with hidden layers of 20, 50, and 10 neurons, whose loss is the mean squared error between corrected and true azimuth and elevation angles; it is trained on 145 measured samples of a moving target.

What would settle it

Move the probe target to a position that is not on the 145-sample training line, for example a different range or an untrained quadrant, and compare the DNN-corrected coordinates with measured truth; the accuracy claim holds only if the error stays near 1.1 mm instead of reverting to the raw monopulse error.

Watch

Extended reading notes

Core claim

The central claim is that the port-crossing problem of a conventional rat-race coupler can be removed within a single layer: move one input port to the same side as the other and insert a 360° phase-delay crossover, and the coupler still behaves as a proper 180° hybrid. The even/odd-mode ABCD derivation shows that the added 180° phase shifter appears as an open or short and therefore contributes no impedance value, so its characteristic impedance can be chosen purely for layout convenience. Four such couplers connected with another crossover form a fully planar, symmetric monopulse comparator whose sum and difference ports align naturally with a 2×2 patch array; measurement gives better than 17 dB return loss, 7.2 ± 0.5 dB transmission, and the required 0°/180° phase relationships over about 10% bandwidth. The DNN part then treats the mismatch between monopulse-ratio angles and true coordinates as a learnable systematic error, with a three-hidden-layer network mapping azimuth, elevation, and distance to corrected angles.

Load-bearing premise

The load-bearing premise is that a neural network trained on 145 samples of continuous target locations corrects angles for targets anywhere else, since it is validated on a 10 percent split of the same dataset rather than on an independent set of positions or distances.

Editorial extensions

If this is right

  • A monopulse comparator network can be fabricated on a single PCB layer, avoiding multilayer lamination and backside feeds, while keeping about 10 percent measured bandwidth at 1.95 GHz.
  • The resulting 2×2 patch array produces sum gain near 11 dBi with sidelobes better than 15 dB and difference-pattern null depths better than 22 dB.
  • A neural network trained on fewer than 150 samples can correct residual angle errors in real time, with validation position error near 1.1 mm at 0.62 m.
  • The design equations for the coupler and crossover supply a repeatable procedure for building symmetric comparators at other center frequencies.

Reading between the lines

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

  • The 1.1 mm validation error is measured on a 10 percent holdout from the same 145-sample trajectory; a stronger test would be a disjoint target path or an untrained grid of positions, since a small network could partly memorize a single line of samples.
  • The port-reallocation identity should transfer to other 180° hybrids and other transmission media, so the single-layer symmetric comparator concept may scale to higher bands without a new topology.
  • Because the neural network learns one receiver's specific systematic distortion, practical deployment would likely need occasional retraining when the antenna, down-converter, or surrounding environment changes.
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

4 major / 8 minor

Summary. The manuscript proposes a 2 GHz low-cost monopulse receiver combining a 2x2 patch array, a fully planar symmetrical monopulse comparator network built from four port-transformation rat-race couplers and a 360° phase-delay crossover, a downconversion link, and a DNN-based post-processing stage for DoA estimation. The comparator design is described through even-odd mode analysis and validated by simulated and measured S-parameters and radiation patterns. The DNN (three fully connected layers with 20-50-10 neurons) takes estimated azimuth, elevation, and distance as input and outputs corrected angles; it is trained on 145 measured samples from a moving target antenna and reports a validation error equivalent to about 1.1 mm at 0.62 m. The paper claims fast real-time tracking with high accuracy in a low-cost planar implementation.

Significance. The hardware contribution is credible and useful: the proposed comparator network achieves a fully planar, symmetric monopulse feed with about 10% measured bandwidth, and the measured S-parameters and radiation patterns are consistent with simulation. If the DNN accuracy claim were rigorously validated, the system would be a practical low-cost calibration approach for monopulse tracking. However, the current DNN evidence is an in-sample validation with no independent test set, no baseline comparison, and no error bars, so the enhanced-accuracy claim is not yet established. The paper's potential significance is therefore moderate and concentrated in the comparator network; the DNN contribution needs substantial additional evaluation.

major comments (4)
  1. [III-C, Eq. (15)] The central DNN accuracy claim rests on an invalid generalization test. The DNN is trained on 145 samples collected along one continuous target trajectory with a 30 mm grid spacing, and the reported validation loss (2e-6, about 1.1 mm at 0.62 m) is obtained from a random 90/10 split of those samples. Because neighboring samples on a continuous trajectory and lattice are strongly correlated, a random split places near-duplicate points in both training and validation sets, so the validation loss is an interpolation metric rather than a measure of generalization. With only about 14 validation samples, no error bars, no regularization or hyperparameter search, and a comparatively high-dimensional hidden-layer parameterization, the 1.1 mm figure does not support the claim that the DNN estimates arbitrary target positions. The authors need an independent test set of positions and distances not seen in training, repeated splits, and error bars on the reported error.
  2. [III-C, Fig. 12] The distance coverage of the training set is not established, so the results at D = 0.66 m and 0.86 m may be extrapolation. The validation error is quoted at a separation of 0.62 m, but the text does not state whether the 145 training samples were collected at that distance or over a range of distances that includes 0.66 m and 0.86 m. Since distance is one of the three DNN inputs, the network may use distance to memorize the measurement geometry; since the far-field monopulse ratio in Eqs. (13) and (14) does not depend on range, the inclusion of this feature also needs physical justification. The authors should specify the training distance distribution, report errors separately for each test distance, and confirm that the test distances are within the training support or explicitly evaluate extrapolation.
  3. [III-C] No baseline comparison is provided for the claimed accuracy enhancement. The paper reports only the DNN validation loss and a qualitative visual agreement in Fig. 12; it does not quantify the uncorrected monopulse-ratio error, nor does it compare the DNN with a conventional calibration method such as a polynomial or lookup-table fit to the same 145 training samples. The introduction explicitly contrasts DNNs with lookup tables, but no empirical comparison is given. Because the DNN is trained and validated on measurements from the same receiver, the method is currently best described as a fitted calibration function; to support the title, the authors should report RMSE or MAE with error bars for the raw estimates, the DNN-corrected estimates, and a calibration baseline on an independent test set.
  4. [III-C and Fig. 11] The DNN input includes the distance between the radar and the target, but the manuscript never states how this distance is measured by the proposed monopulse receiver. The receiver chain described in Fig. 1 and Section III-C (array, comparator, downconverter, VSA) provides only the sum and difference channels used to derive angles in Eqs. (13)-(14); no ranging scheme (e.g., FMCW, time-of-flight, or received-signal-strength) is presented. If distance is assumed known from the test setup, the claim of a real-time tracking system is incomplete; the authors should either remove distance from the input, describe a ranging method, or clearly limit the DNN to scenarios where the range is externally supplied.
minor comments (8)
  1. [II-A, Eqs. (1)-(2)] The even-odd ABCD matrices in Eqs. (1)-(2) are garbled in the submitted PDF and cannot be checked; please ensure the final typeset equations are legible.
  2. [Abstract and Section III-A] The operating frequency is given as 2 GHz in the abstract and Section II, while the measured comparator and array results in Figs. 8-10 are centered at 1.95 GHz; the text should state this fabrication-tolerance shift consistently when introducing the measurements.
  3. [Table I and Introduction] Table I and the introduction should explicitly differentiate the present comparator from [17], which also reports a fully symmetric planar comparator, and from the conference versions [24] and [25]; as written, the incremental novelty is not clear.
  4. [III-C] Section III-C reports a training time of about 6 minutes but gives no inference latency; the 'real-time' and 'milliseconds' claims should be supported by a measured or estimated per-query inference time.
  5. [Fig. 12] Fig. 12(b)-(c) shows only qualitative agreement; please add tabulated RMSE/MAE and maximum error per distance, and label the axes with angle units and the DNN input/output quantities.
  6. [III-C] For reproducibility, please report the random split seed, optimizer learning rate, batch size, and whether any regularization was used; the current description of the training procedure is incomplete.
  7. [III-C] The statement that the DNN 'requires small size of dataset' is not supported by any experiment; a comparison with a reduced training-set size or with a polynomial calibration baseline is needed.
  8. [III-C and Eq. (15)] The conversion from the angle loss in Eq. (15) to a 1.1 mm position error at 0.62 m should be stated explicitly, including the units of the angles and the geometry used for the conversion.

Circularity Check

1 steps flagged · score 6.0 of 10

The comparator-network design is self-contained, but the headline 'enhanced DoA accuracy via DNN' is a supervised calibration whose 1.1 mm validation error is computed on a 10% random split of a single 145-sample trajectory, i.e., interpolation rather than independent prediction.

  1. fitted input called prediction [Section III-C, Eq. (15), Fig. 12]
    "The training dataset DNN contains 145 samples of continuous locations of the remote target (standard patch antenna), which is moving along 2D plane with unit of 30mm. ... 90% of the data set is allocated for training , and the remaining 10% serve as a validation. ... After it is fully trained, the validation error is about 2 × 10-6 that equivalent to a position error of 1.1 mm in a 2D plane with a 0.62 m separation distance between the transmitter target and the monopulse receiver ."

    The DNN is trained by minimizing Eq. (15), the squared error between its output and truth angles, on 90% of 145 samples recorded along one continuous target trajectory; the claimed 1.1 mm accuracy is the same loss on the remaining 10% of that trajectory. Because successive samples are 30 mm apart, a random 10% split places validation points near training points, so the reported loss measures interpolation, not independent prediction. No independent test set or calibration baseline is given, and the abstract defines the DNN as mapping 'misaligned target angular positions in the measurement to the actual physical location,' i.e., fitting a calibration curve to the receiver's own error.

full rationale

The hardware contribution is not circular: the port-transformation rat-race coupler is analyzed by even-odd mode ABCD matrices, the crossover design equations are stated, and the comparator network is validated against independent EM simulation and measured S-parameters and radiation patterns. Those external measurements support the planar symmetrical topology claim regardless of the DNN. The circularity is confined to the DNN-based DoA accuracy claim. Eq. (15) defines a supervised regression loss, and the reported 1.1 mm accuracy is the loss on a 10% random holdout of a single 145-sample continuous trajectory; because nearby samples are near-duplicates, the holdout is not an independent test. The DNN therefore functions as a fitted calibration of this specific receiver, and the 'enhanced accuracy' is, by construction, a fit to the training data rather than a demonstrated prediction for arbitrary target positions. This yields partial circularity (6/10) rather than a fully circular paper, since the comparator network itself is externally verified.

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

The central hardware design relies only on standard microwave network theory and the authors' prior crossover work. The DNN section introduces a fitted model with many free parameters (weights and biases) and an unproven generalization assumption. No new physical entities are postulated.

free parameters (3)
  • Crossover dimensions Zx, Zy, theta_x, theta_y = Zx=57 Ohm, Zy=50 Ohm, theta_x=90 deg, theta_y=90 deg
    Design parameters selected to satisfy crossover conditions in equations (7)-(8); chosen by hand, not fitted to measured data.
  • DNN weights and biases = Not provided
    The 3-layer MLP (20-50-10 neurons) is trained on 145 measured samples with Adam optimizer; the trained weights are not shared.
  • DNN architecture and training hyperparameters = 20/50/10 neurons, 20,000 iterations
    Chosen by the authors; no ablation or sensitivity analysis is shown.
assumptions (4)
  • standard math Even-odd mode decomposition of symmetric networks and transmission line ABCD matrices.
    Used in Section II.A to derive the port-transformation coupler and in Section II.B for the crossover.
  • domain assumption The reference crossover analysis from [27] for the 360-degree phase delay crossover.
    Equations (4)-(8) rely on the authors' earlier published crossover theory.
  • domain assumption The monopulse ratio is range-independent (far-field assumption).
    Equations (13)-(14) imply angle is obtained from the ratio Delta/Sigma without distance, yet the DNN uses distance as an input and the measurements are at 0.62-0.86 m.
  • ad hoc to paper The DNN training set is representative of the target's operating distribution.
    145 samples along a continuous trajectory with 30 mm step; no evidence of coverage of all positions and distances of interest.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Low-Cost Monopulse Receiver with Enhanced Estimation Accuracy Via Deep Neural Network." pith.science (2026). https://pith.science/paper/LWQEIUNU

@misc{pith2026241117734,
  author       = {Pith},
  title        = {Pith review of: A Low-Cost Monopulse Receiver with Enhanced Estimation Accuracy Via Deep Neural Network},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LWQEIUNU}},
  note         = {Machine review of arXiv:2411.17734}
}
read the original abstract

In this paper, a low-cost monopulse receiver with an enhanced direction of arrival (DoA) estimation accuracy via deep neural network (DNN) is proposed. The entire system is composed of a 4-element patch array, a fully planar symmetrical monopulse comparator network, and a down conversion link. Unlike the conventional design topology, the proposed monopulse comparator network is configured by four novel port-transformation rat-race couplers. In specific, the proposed coupler is designed to symmetrically allocate the sum ({\Sigma}) / delta ({\Delta}) ports with input ports, where a 360{\deg} phase delay crossover is designed to transform the unsymmetrical ports in the conventional rat-race coupler. This new rat-race coupler resolves the issues in conventional monopulse receiver comparator network design using multilayer and expensive fabrication technology. To verify the design theory, a prototype of the proposed planar monopulse comparator network operating at 2 GHz is designed, simulated, and measured. In addition, the monopulse radiation patterns and direction of arrival are also decently evaluated. To further boost the accuracy of angular information, a deep neural network is introduced to map the misaligned target angular positions in the measurement to the actual physical location under detection.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 27 canonical work pages

  1. [1]

    Short -Range FMCW Monopulse Radar for Hand -Gesture Sensing,

    P. Molchanov, S. Gupta, K. Kim, K. Pulli, “Short -Range FMCW Monopulse Radar for Hand -Gesture Sensing,” IEEE Radar Conference, Arlinton, VA, USA, 2015, pp. 1491 – 1496

    work page 2015

  2. [2]

    A Wideband High -Gain Planar Monopulse Array Antenna for Ka -Band Radar Applications,

    Z. Wang, Y. Hu, L. Xiang, J, Xu, W. Hong, “A Wideband High -Gain Planar Monopulse Array Antenna for Ka -Band Radar Applications,” IEEE Trans. Antennas Propag ., vol. 71, no. 11, pp. 8739 -8752, Nov. 2023

    work page 2023

  3. [3]

    Design of a Low -Cost UWB Time -Domain Radar System for Subcentimeter Image Resolution,

    R. Feghhi, R. S. C. Winter, F. M. Sabzevari and K. Rambabu, "Design of a Low -Cost UWB Time -Domain Radar System for Subcentimeter Image Resolution," IEEE Trans. Microw. Theory Tech , vol. 70, no. 7, pp. 3617-3628, July. 2022

    work page 2022

  4. [4]

    Compact Via-Coupling Fed Monopulse Antenna with Orthogonal Tracking Capability in Radiation Pattern,

    S. A. Khatami, J. Meiguni, A. Amn -e-Elahi, and P. Rezaei, “Compact Via-Coupling Fed Monopulse Antenna with Orthogonal Tracking Capability in Radiation Pattern,” IEEE Antennas Wireless Propag. Lett ., vol. 19, no. 8, pp. 1443–1446, Aug. 2020

    work page 2020

  5. [5]

    Broadband Monopulse Microstrip Antenna Array for X -Band Monopulse Tracking,

    H. Kumar, G. Kumar, “Broadband Monopulse Microstrip Antenna Array for X -Band Monopulse Tracking,” IET Microw. Antennas & Propag., vol. 12, no. 13, pp. 2109-2114, Aug. 2018

    work page 2018

  6. [6]

    A Compact Single Layer Monopulse Microstrip Antenna Array,

    H. Wang, D. -G. Fang, and X. G. Chen, “A Compact Single Layer Monopulse Microstrip Antenna Array,” IEEE Trans. Antennas Propag ., vol. 54, no. 2, pp. 503–509, Feb. 2006

    work page 2006

  7. [7]

    Design of Ultra -Wideband Monopulse Receiver,

    A. E. Tan, M. Y. Chia, K. Rambabu, “Design of Ultra -Wideband Monopulse Receiver,” IEEE Trans. Microw. Theory Tech , vol. 54, no. 11, pp. 3821-3827, Nov. 2006

    work page 2006

  8. [8]

    A Compact Monopulse Antenna Array with Suppressed Mutual Coupling Using Broadband Schiffman Phase Shifters,

    S. Alamdar, K. M. Aghdam, H. Khalili, “A Compact Monopulse Antenna Array with Suppressed Mutual Coupling Using Broadband Schiffman Phase Shifters,” IEEE Access vol. 12, pp. 11936 -11944, Jan. 2024

    work page 2024

Show all 27 references
  1. [9]

    A Wide -Band Monopulse Comparator with Complete Nulling in All Delta Channels Throughout Sum Channel Bandwidth,

    K. S. Ang, Y. C. Leong, and C. H. Lee, “A Wide -Band Monopulse Comparator with Complete Nulling in All Delta Channels Throughout Sum Channel Bandwidth,” IEEE Trans. Microw. Theory Techn., vol. 51, no. 2, pp. 371–373, Feb. 2003

    2003

  2. [10]

    28GHz Active Monopulse Networks with Amplitude and Phase Control and -30dB Null -Bandwidth of 5GHz,

    H. Chung, Q. Ma, G. M. Rebeiz, “28GHz Active Monopulse Networks with Amplitude and Phase Control and -30dB Null -Bandwidth of 5GHz,” IEEE MTT -S International Microwave Symposium (IMS), Los Angeles, CA, USA, 2020, pp. 1-4

    2020

  3. [11]

    Series -Fed Monopulse Microstrip Array Antenna with Stripline Quadrature Hybrid Comparator Network,

    L. Zou, X. Wang, and J. Zang, “Series -Fed Monopulse Microstrip Array Antenna with Stripline Quadrature Hybrid Comparator Network,” IEEE Access, vol. 9, pp. 169177–169192, Dec. 2021

    2021

  4. [12]

    Compact Substrate Integrated Waveguide (SIW) Monopulse Network for Ku - Band Tracking System Applications,

    H. Chen, W. Che, Q. He, W. Feng, X. Wei and K. Wu, "Compact Substrate Integrated Waveguide (SIW) Monopulse Network for Ku - Band Tracking System Applications," IEEE Trans. Microw. Theory Techn, vol. 62, no. 3, pp. 472-480, Mar. 2014

    2014

  5. [13]

    A Dual -Mode SIW Compact Monopulse Comparator for Sum and Difference Multibeam Radar Applications,

    Y. Cao, S. Yan, “A Dual -Mode SIW Compact Monopulse Comparator for Sum and Difference Multibeam Radar Applications,” IEEE Microw. Wireless Compon. Lett., vol. 32, no. 1, pp. 41-44, Jan. 2022

    2022

  6. [14]

    A Compact SIW Monopulse Antenna Array Based on Microstrip Feed,

    W. Li, S. Liu, J. Deng, Z. Hu and Z. Zhou, "A Compact SIW Monopulse Antenna Array Based on Microstrip Feed," IEEE Antennas Wireless Propag. Lett., vol. 20, no. 1, pp. 93-97, Jan. 2021

    2021

  7. [15]

    A compact wideband planar magic tee for monopulse antenna array applications,

    D. Nagaraju and Y. K. Verma, “A compact wideband planar magic tee for monopulse antenna array applications,” IEEE Microw. Wireless Compon. Lett., vol. 31, no. 5, pp. 429–432, May 2021

    2021

  8. [16]

    A Dual - Polarized 2 -D Monpulse Antenna Array for Conical Conformal Applications,

    Y. Gao, W, Jiang, W, Hu, Q. Wang, W, Zhang, and S. Gong, “A Dual - Polarized 2 -D Monpulse Antenna Array for Conical Conformal Applications,” IEEE Trans on Antennas and Propag ., vol. 69, no. 9, pp. 5479- 5499, Sept. 2022

    2022

  9. [17]

    A Fully Symmetrical Uni - Planar Microstrip Line Comparator Network for Monopulse Antenna,

    H. Zhang, H. Ren, Y. Gu and B. Arigong, "A Fully Symmetrical Uni - Planar Microstrip Line Comparator Network for Monopulse Antenna," IEEE Microw. Wireless Technol. Lett , vol. 33, no. 5, pp. 611 -614, May 2023

    2023

  10. [18]

    Holographic Back -Projection Method for Calibration of Fully Digital Polarimetric Phased Array Radar,

    D. Schvartzman, J. D. D. Díaz, D. Zrnić, M. Herndon, M. B. Yeary and R. D. Palmer, "Holographic Back -Projection Method for Calibration of Fully Digital Polarimetric Phased Array Radar," IEEE Trans . Radar Syst., vol. 1, pp. 295-307, Jun. 2023

    2023

  11. [19]

    A Multiple Baseline Interferometric Radar for Multiple Target Angular Velocity Measurement,

    J. Merlo, E. Klinefelter, S. Vakalis and J. A. Nanzer, “A Multiple Baseline Interferometric Radar for Multiple Target Angular Velocity Measurement,” IEEE Microw. Wireless Compon. Lett. , vol. 31, no. 8, pp. 937-940, Aug. 2021

    2021

  12. [20]

    A Low Computational Complexity DOA Estimation Using Sum/Difference Pattern Based on DNN

    S. Xu, B. Chen, H. Xiang. "A Low Computational Complexity DOA Estimation Using Sum/Difference Pattern Based on DNN." Multidimensional Systems and Signal Processing , vol. 34, no. 1 , July 2021

    2021

  13. [21]

    Machine Learning -Based Target Classification for MMW Radar in Autonomous Driving,

    X. Cai, M. Giallorenzo and K. Sarabandi, "Machine Learning -Based Target Classification for MMW Radar in Autonomous Driving," IEEE Trans. Intelligent Vehicles, vol. 6, no. 4, pp. 678-689, Dec. 2021

    2021

  14. [22]

    Deep Convolution Network for Direction of Arrival Estimation with Sparse Prior,

    L. Wu, Z. -M. Liu and Z. -T. Huang, "Deep Convolution Network for Direction of Arrival Estimation with Sparse Prior," IEEE Signal Processing Lett., vol. 26, no. 11, pp. 1688-1692, Nov. 2019

    2019

  15. [23]

    Fall Detection Using Deep Learning in Range-Doppler Radars,

    B. Jokanović and M. Amin, "Fall Detection Using Deep Learning in Range-Doppler Radars," IEEE Trans . Aerospace and Electronic Sys t., vol. 54, no. 1, pp. 180-189, Feb. 2018

    2018

  16. [24]

    Boosting Estimation Accuracy of Low -Cost Monopulse Receiver Via Deep Neural Network,

    H. Zhang, et al., “Boosting Estimation Accuracy of Low -Cost Monopulse Receiver Via Deep Neural Network,” 2024 IEEE Wireless and Microwave Technology Conference (WAMICON) , Clearwater, FL, USA, April 2024, pp. 1-4

    2024

  17. [25]

    A Planar Monopulse Comparator Network Design from Port - Transformation Rat-Race Coupler,

    H. Zhang, P. Liu, J. Casamayor, S. Z. Pour, M. Plaisir, B. Arigong, “A Planar Monopulse Comparator Network Design from Port - Transformation Rat-Race Coupler,” 2024 IEEE Radio & Wireless Week (RWW), San Antonio, TX, USA, Jan. 2024, pp. 1-4

    2024

  18. [26]

    D. M. Pozar, Microwave Engineering, 3rd ed. New York, NY, USA: Wiley, 2005

    2005

  19. [27]

    A Novel Dual -Band Zero-Phase True Crossover with Arbitrary Port Impedance,

    H. Ren, M. Zhou, H. Zhang and B. Arigong, " A Novel Dual -Band Zero-Phase True Crossover with Arbitrary Port Impedance," IEEE Microw. Wireless Comp. Lett, vol. 29, no. 1, pp. 29-31, Jan. 2019

    2019

Pith tools

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