pith. sign in

arxiv: 1906.10044 · v2 · pith:ZRO7POG5new · submitted 2019-06-24 · 📡 eess.SP · cs.CV

Complex Signal Denoising and Interference Mitigation for Automotive Radar Using Convolutional Neural Networks

Johanna Rock , Mate Toth , Elmar Messner , Paul Meissner , Franz Pernkopf This is my paper

Pith reviewed 2026-05-25 17:07 UTC · model grok-4.3

classification 📡 eess.SP cs.CV
keywords convolutional neural networksautomotive radarinterference mitigationsignal denoisingmutual interferencecomplex signalsmachine learning
0
0 comments X

The pith

Convolutional neural networks can denoise automotive radar signals and mitigate mutual interference as an alternative to conventional signal processing.

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

The paper shows that convolutional neural networks trained on simulated data can remove interference effects from complex radar signals while preserving local information and structured patterns. This leads to better denoising than traditional methods and maintains detection sensitivity when the networks are placed inside the standard radar processing chain. As radar sensors become more common on roads, mutual interference grows and threatens reliable sensing for driver assistance and autonomous driving. The work compares the CNN approach directly to existing techniques and reports improved performance.

Core claim

Convolutional neural networks trained on simulated interference data find structured information while preserving local details, enabling superior denoising of complex-valued automotive radar signals. When integrated into the radar signal processing chain, this yields better interference mitigation than conventional methods and can be used as an alternative approach.

What carries the argument

Convolutional neural networks operating on complex-valued radar signals for denoising and interference mitigation.

If this is right

  • The CNN approach can be inserted into existing automotive radar pipelines without changing hardware.
  • Simulated data alone suffices to train effective interference mitigation models.
  • Detection sensitivity remains higher under interference compared with state-of-the-art conventional methods.
  • CNNs handle both denoising and interference mitigation within one learned model.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same learned denoising step could be applied to other multi-sensor environments where signals overlap in frequency.
  • Model compression or pruning would be needed to reach real-time embedded radar processors.
  • End-to-end training that includes downstream detection tasks might further improve overall system performance.

Load-bearing premise

Training data generated by simulation accurately represents the statistical properties and diversity of real-world mutual interference encountered by automotive radars.

What would settle it

Real-world automotive radar recordings with actual mutual interference where the CNN method fails to match or exceed conventional processing in detection sensitivity after denoising.

Figures

Figures reproduced from arXiv: 1906.10044 by Elmar Messner, Franz Pernkopf, Johanna Rock, Mate Toth, Paul Meissner.

Figure 1
Figure 1. Figure 1: Block diagram of a basic FMCW/CS radar processing [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Proposed CNN architecture for radar signal denoising. It uses [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Schematic representation of the convolution operation [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Exemplary range-Doppler magnitude spectra in dB of [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: CNN architecture performance comparison for RD RIS [PITH_FULL_IMAGE:figures/full_fig_p005_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: CDF comparison of RD SINR with other techniques. [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: CDF comparison of RD EVM with other techniques. [PITH_FULL_IMAGE:figures/full_fig_p007_8.png] view at source ↗
Figure 10
Figure 10. Figure 10: Velocity cut at distance d = 7.9m. The object is located at a velocity v = 5.5m/s as indicated by the vertical black marker. especially regarding object peak value preservation. The RD RIS model enhances object peaks, while it strongly reduces the noise floor to a constant level. This reassures our presumption that the CNN-based denoising has an implicit thresholding effect. In summary, CNN-based methods … view at source ↗
read the original abstract

Driver assistance systems as well as autonomous cars have to rely on sensors to perceive their environment. A heterogeneous set of sensors is used to perform this task robustly. Among them, radar sensors are indispensable because of their range resolution and the possibility to directly measure velocity. Since more and more radar sensors are deployed on the streets, mutual interference must be dealt with. In the so far unregulated automotive radar frequency band, a sensor must be capable of detecting, or even mitigating the harmful effects of interference, which include a decreased detection sensitivity. In this paper, we address this issue with Convolutional Neural Networks (CNNs), which are state-of-the-art machine learning tools. We show that the ability of CNNs to find structured information in data while preserving local information enables superior denoising performance. To achieve this, CNN parameters are found using training with simulated data and integrated into the automotive radar signal processing chain. The presented method is compared with the state of the art, highlighting its promising performance. Hence, CNNs can be employed for interference mitigation as an alternative to conventional signal processing methods. Code and pre-trained models are available at https://github.com/johanna-rock/imRICnn.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes using convolutional neural networks (CNNs) to denoise complex-valued radar signals and mitigate mutual interference in automotive radar. Training data is generated via simulation of chirp-based interference; the trained CNN is inserted into the standard radar processing chain. On held-out simulated test sets the method is reported to outperform conventional signal-processing baselines, with the conclusion that CNNs provide a viable alternative to traditional interference mitigation. Code and pre-trained models are released publicly.

Significance. If the generalization claim holds, the work supplies a concrete machine-learning alternative for handling rising mutual interference in the unregulated 77 GHz band, directly addressing a practical limitation of current automotive radar deployments. The public release of code and models is a clear strength that supports reproducibility.

major comments (2)
  1. [Evaluation / Results] The central claim that CNN-based denoising constitutes a practical alternative to conventional methods for real automotive radar rests on the assumption that the simulation model captures the statistical properties of real mutual interference. No quantitative results on measured data from co-located 77 GHz radars are presented anywhere in the manuscript, leaving the simulation-to-real transfer untested and load-bearing for the applicability statement in the abstract and conclusion.
  2. [Abstract and Evaluation] The abstract asserts 'superior' and 'promising performance' relative to the state of the art, yet the evaluation supplies neither numerical metrics (e.g., detection probability, SINR gain) nor dataset statistics (number of realizations, SNR range, interference density) that would allow a reader to judge the magnitude or robustness of the reported gains.
minor comments (2)
  1. [Methods] The manuscript would benefit from an explicit ablation study on the sensitivity of the CNN to variations in the interference-model parameters (chirp slope, power offset, timing jitter) used to generate the training set.
  2. [Figures and Evaluation] Figure captions and the text should clarify whether the reported metrics are computed on complex-valued data before or after range-Doppler processing.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed review. Below we address each major comment point by point.

read point-by-point responses

  1. Referee: [Evaluation / Results] The central claim that CNN-based denoising constitutes a practical alternative to conventional methods for real automotive radar rests on the assumption that the simulation model captures the statistical properties of real mutual interference. No quantitative results on measured data from co-located 77 GHz radars are presented anywhere in the manuscript, leaving the simulation-to-real transfer untested and load-bearing for the applicability statement in the abstract and conclusion.

    Authors: We acknowledge that all reported results are obtained on simulated data and that no real measured data from co-located 77 GHz radars is included. The simulation framework is constructed to reproduce the dominant statistical characteristics of chirp-sequence mutual interference (random start times, frequency offsets, and amplitudes). While this constitutes a standard first step for evaluating new mitigation algorithms, we agree that the lack of real-data validation limits the strength of the practical-applicability claim. In revision we will (i) explicitly qualify all claims as simulation-based, (ii) add a limitations paragraph discussing the simulation-to-real gap, and (iii) outline concrete next steps for acquiring and evaluating on measured interference data. revision: partial

  2. Referee: [Abstract and Evaluation] The abstract asserts 'superior' and 'promising performance' relative to the state of the art, yet the evaluation supplies neither numerical metrics (e.g., detection probability, SINR gain) nor dataset statistics (number of realizations, SNR range, interference density) that would allow a reader to judge the magnitude or robustness of the reported gains.

    Authors: We thank the referee for highlighting this presentational shortcoming. Although the body of the manuscript contains comparative plots, we will revise the abstract to replace qualitative statements with concrete numerical improvements (e.g., average SINR gain in dB and detection-rate improvement at fixed false-alarm rate). We will also insert a table or paragraph in the experimental section that reports the exact dataset statistics: number of Monte-Carlo realizations, SNR range, interference-to-signal ratio distribution, and interference density (chirps per frame). revision: yes

Circularity Check

0 steps flagged

No circularity; empirical ML evaluation on simulated data with external baselines

full rationale

The paper trains CNNs on simulated automotive radar interference and reports denoising metrics versus conventional signal processing methods on held-out simulated test sets. No derivation chain, equation, or claim reduces by construction to its own inputs, fitted parameters, or self-citation load-bearing premises. The central claim is an empirical performance comparison, not a self-referential prediction or uniqueness theorem. Self-citations, if present, are not invoked to justify the core result.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Central claim rests on the assumption that simulated interference data is representative enough for the trained CNN to generalize; no other free parameters or invented entities are identifiable from the abstract.

axioms (1)
  • domain assumption Simulated data sufficiently represents real interference scenarios
    Training and evaluation rely on simulated data as stated in the abstract.

pith-pipeline@v0.9.0 · 5748 in / 926 out tokens · 28496 ms · 2026-05-25T17:07:13.515762+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

18 extracted references · 18 canonical work pages · 3 internal anchors

  1. [1]

    Analytical Investi- gation of Non-Coherent Mutual FMCW Radar Interference,

    M. Toth, P. Meissner, A. Melzer, and K. Witrisal, “Analytical Investi- gation of Non-Coherent Mutual FMCW Radar Interference,” in 2018 European Radar Conference (EURAD) , pp. 71–74, 2018

    work page 2018

  2. [2]

    Mutual Interference of Millimeter-Wave Radar Sys- tems,

    G. M. Brooker, “Mutual Interference of Millimeter-Wave Radar Sys- tems,” IEEE Transactions on Electromagnetic Compatibility , vol. 49, no. 1, pp. 170–181, 2007

    work page 2007

  3. [3]

    Threshold-Free Interference Cancellation Method for Automotive FMCW Radar Systems,

    M. Wagner, F. Sulejmani, A. Melzer, P. Meissner, and M. Hue- mer, “Threshold-Free Interference Cancellation Method for Automotive FMCW Radar Systems,” in 2018 IEEE International Symposium on Circuits and Systems (ISCAS) , 2018

    work page 2018

  4. [4]

    Automotive Radar Interference Mitigation Using a Sparse Sampling Approach,

    J. Bechter, F. Roos, M. Rahman, and C. Waldschmidt, “Automotive Radar Interference Mitigation Using a Sparse Sampling Approach,” in 2017 European Radar Conference (EURAD) , pp. 90–93, 2017

    work page 2017

  5. [5]

    Sparse signal processing using iterative method with adaptive thresholding (IMAT),

    F. Marvasti, M. Azghani, P. Imani, P. Pakrouh, S. Heydari, A. Gol- mohammadi, A. Kazerouni, and M. Khalili, “Sparse signal processing using iterative method with adaptive thresholding (IMAT),” in2012 19th International Conference on Telecommunications (ICT) , 2012

    work page 2012

  6. [6]

    Estimation and can- cellation of interferences in automotive radar signals,

    J. Bechter, K. D. Biswas, and C. Waldschmidt, “Estimation and can- cellation of interferences in automotive radar signals,” in 2017 18th International Radar Symposium (IRS) , pp. 1–10, 2017

    work page 2017

  7. [7]

    Digital beamforming to mitigate automotive radar interference,

    J. Bechter, K. Eid, F. Roos, and C. Waldschmidt, “Digital beamforming to mitigate automotive radar interference,” 2016 IEEE MTT-S Int. Conf. Microwaves Intell. Mobility, ICMIM 2016 , pp. 2–5, 2016

    work page 2016

  8. [8]

    Support Vector Machines for Classification of Automotive Radar Interference,

    R. Zhang and S. Cao, “Support Vector Machines for Classification of Automotive Radar Interference,” 2018 IEEE Radar Conf. , pp. 366–371, 2018

    work page 2018

  9. [9]

    Complex spectrogram enhance- ment by convolutional neural network with multi-metrics learning,

    S.-W. Fu, T.-y. Hu, Y . Tsao, and X. Lu, “Complex spectrogram enhance- ment by convolutional neural network with multi-metrics learning,” in 2017 IEEE 27th International Workshop on Machine Learning for Signal Processing (MLSP), pp. 1–6, IEEE, 2017

    work page 2017

  10. [10]

    Performance com- parison of mutual automotive radar interference mitigation algorithms,

    M. Toth, P. Meissner, A. Melzer, and K. Witrisal, “Performance com- parison of mutual automotive radar interference mitigation algorithms,” in IEEE Radar Conference , 2019

    work page 2019

  11. [11]

    Linear FMCW radar techniques,

    A. G. Stove, “Linear FMCW radar techniques,” IEE Proceedings F - Radar and Signal Processing , vol. 139, no. 5, pp. 343–350, 1992

    work page 1992

  12. [12]

    Range Doppler detection for automotive FMCW radars,

    V . Winkler, “Range Doppler detection for automotive FMCW radars,” in 2007 European Microwave Conference , pp. 1445–1448, Oct. 2007

    work page 2007

  13. [13]

    A Peer-to-Peer Interference Analysis for Automotive Chirp Sequence Radars,

    G. Kim, J. Mun, and J. Lee, “A Peer-to-Peer Interference Analysis for Automotive Chirp Sequence Radars,” IEEE Transactions on Vehicular Technology, vol. 67, no. 9, pp. 8110–8117, 2018

    work page 2018

  14. [14]

    Deep sparse rectifier neural networks.,

    X. Glorot, A. Bordes, and Y . Bengio, “Deep sparse rectifier neural networks.,” in AISTATS (G. J. Gordon, D. B. Dunson, and M. Dud ´ık, eds.), vol. 15 of JMLR Proceedings, pp. 315–323, JMLR.org, 2011

    work page 2011

  15. [15]

    Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

    S. Ioffe and C. Szegedy, “Batch normalization: Accelerating deep network training by reducing internal covariate shift,” CoRR, vol. abs/1502.03167, 2015

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  16. [16]

    Deep Complex Networks

    C. Trabelsi, O. Bilaniuk, D. Serdyuk, S. Subramanian, J. F. Santos, S. Mehri, N. Rostamzadeh, Y . Bengio, and C. J. Pal, “Deep complex networks,” CoRR, vol. abs/1705.09792, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  17. [17]

    Adam: A Method for Stochastic Optimization

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” CoRR, vol. abs/1412.6980, 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  18. [18]

    Fischer, Untersuchungen zum Interferenzverhalten automobiler Radarsensorik

    C. Fischer, Untersuchungen zum Interferenzverhalten automobiler Radarsensorik. PhD thesis, Ulm University, 2016

    work page 2016