Pith. sign in

REVIEW 4 major objections 5 minor 33 references

Real-Time Smart Self-Mixing Interferometry Sensor With Embedded Neural Network

T0 review · 4 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A 6,937-parameter neural network, quantized to 16 bits and run on a 160 MHz microcontroller, reconstructs target displacement from laser self-mixing signals in 0.64 ms per 1 ms window, meeting the real-time constraint while matching float32

desk verdict Genuine real-time embedded SMI sensor demo with solid latency numbers, but the accuracy metrics rest on a speaker-voltage proxy that could hide calibration bias. read the letter →

arxiv 2608.03448 v1 pith:3TC6BN7I submitted 2026-08-04 physics.ins-det

Pierre-Emmanuel Novac , Laurent Rodriguez , Stéphane Barland This is my paper · ORCID
classification physics.ins-det
keywords self-mixinginterferometryembeddedneuralnetworkdisplacementsensingreal-timeinferenceResNetquantizationmicrocontrollerCMSIS-NN
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

Self-mixing interferometry measures target motion by feeding a laser's own reflected light back into the laser and reading the resulting voltage changes, but decoding that signal is notoriously difficult. This paper shows that a very small residual neural network with 6,937 parameters after batch-normalization fusion can perform that decoding accurately enough for real-time use when quantized to 16-bit fixed point and run on a 160 MHz microcontroller with CMSIS-NN optimizations. The network processes each 1 ms, 48-sample window in 0.64 ms, leaving time to spare, and its accuracy on out-of-distribution aperiodic signals (Pearson correlation 0.889, mean absolute error 0.353 $\lambda$/ms) essentially matches the full-precision model (0.884, 0.364 $\lambda$/ms). A per-window input normalization layer keeps performance flat across four orders of magnitude of signal amplitude, relaxing the analog front-end constraints. If this holds, standalone, battery-powered vibration and displacement sensors with built-in neural decoding are within reach.

What carries the argument

A residual convolutional neural network with three residual blocks (kernel length 3, residual branches kernel length 1), global average pooling, and a dense regression head, preceded by a fixed SampleNormLayer that scales each 48-sample window to [0, 1]. The network is trained on periodic motions, post-training quantized to 16-bit fixed point (Q7.9), batch-normalization is fused into the preceding convolutions, and the Qualia toolchain generates portable C code that uses CMSIS-NN's SIMD-optimized convolution kernels. The normalization layer provides amplitude robustness, the compressed ResNet fits the microcontroller's memory and compute budget, and 16-bit quantization preserves accuracy whi

What would settle it

Replace the calibrated speaker with an independent displacement reference, such as a commercial laser interferometer or a capacitive sensor measuring the same target, and compare the network's inferred displacement to that reference across the full operating envelope (5–100 Hz, below 2.5 $\lambda$/ms, feedback parameter C near unity); any systematic disagreement larger than the reported mean absolute error of roughly 0.35 $\lambda$/ms would indicate that the speaker-voltage proxy is contaminating the accuracy claims.

Watch

Extended reading notes

Core claim

The paper establishes that a ResNet with 6,937 parameters, operating on non-overlapping 1 ms windows of 48 samples at 48 kHz, can reconstruct target displacement from a self-mixing interferometry signal on a microcontroller within the real-time constraint of 1 ms. With 16-bit post-training quantization (Q7.9 format), fused batch-normalization layers, and CMSIS-NN kernels on an STM32U575ZI, inference takes 0.64 ms and uses 22.9 kB RAM and 76.4 kB ROM, while generalization performance on aperiodic out-of-distribution displacements is statistically indistinguishable from float32 (PCC 0.889 vs 0.884; MAE 0.353 vs 0.364 $\lambda$/ms). The per-window sample normalization layer makes the network in

Load-bearing premise

The ground-truth displacement labels in both training and evaluation are derived from the voltage applied to a calibrated speaker, assuming a linear $\lambda/V$ scaling factor accurate to within 10% over the operating frequency band; if that proxy is biased or nonlinear, the network learns and the reported accuracy numbers inherit the bias.

Editorial extensions

If this is right

  • A standalone, battery-powered displacement/vibration sensor is feasible: total system power is about 600 mW, dominated by the unoptimized analog front end, and one prediction costs 0.167 $\mu$Wh, so AA batteries could sustain roughly 10 hours of continuous operation.
  • The sensor is robust to target reflectivity and laser-dependent signal amplitude variations because of the per-window normalization layer, so the same network can be used across different target materials and even different laser systems.
  • Real-time operation at 1 ms per window implies the system could in principle measure displacement frequencies up to about 500 Hz, bounded by the 48 kHz sampling rate and the training domain (5–100 Hz, below 2.5 $\lambda$/ms, feedback parameter C near unity).
  • 8-bit quantization is not worth the accuracy loss (linear-regression slope drops by 0.029 compared with float32) under this simple uniform quantization scheme; the authors choose 16-bit as the deployment configuration.
  • The 0.64 ms inference leaves time within each 1 ms window that could in principle be used for additional on-device computations, such as the quality-of-service estimation the authors say they are currently investigating.

Reading between the lines

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

  • Because the network learns fringe shape rather than signal amplitude, the same architecture could likely be retrained to estimate other self-mixing parameters (for example the feedback parameter C or target distance) from the same voltage signal, turning the sensor into a multi-parameter instrument.
  • The reported accuracy is measured against a speaker-voltage proxy for displacement, so an independent optical reference, such as a commercial laser interferometer, would be a stronger validation; the 10% linearity tolerance of the speaker calibration is likely the dominant systematic uncertainty in the absolute numbers.
  • The sub-millisecond headroom suggests the same pipeline could be ported to other microcontroller families or to hardware accelerators, and at higher sampling rates the architecture could be deepened or the window length extended, trading the current latency margin for higher resolution or a wider frequency band.
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

4 major / 5 minor

Summary. The paper presents an embedded self-mixing interferometry (SMI) sensor that combines a semiconductor laser front-end with an STM32U575ZI microcontroller running a quantized 16-bit ResNet. The network reconstructs target displacement from 1-ms windows of the interferometric signal, and the authors demonstrate fully on-device real-time operation with 0.64 ms inference latency per window, which is below the 1 ms real-time constraint. They compare the ResNet against a prior CNN, evaluate robustness to input amplitude variations via an input normalization layer, study 8/16-bit quantization trade-offs, measure memory/latency/energy, and provide a live reconstruction example. The central claims are that the int16 ResNet meets real-time constraints and achieves generalization performance comparable to float32 (PCC 0.889 vs. 0.884, MAE 0.353 vs. 0.364 λ/ms).

Significance. If the results hold, this is a valuable engineering contribution: it is the first demonstration, to my knowledge, of SMI displacement reconstruction performed fully on a low-power microcontroller in real time using a neural network. The 0.64 ms inference time on a 160 MHz Cortex-M33 with 6937 parameters is a concrete, measured result that invites direct comparison with future work. The amplitude-normalization layer is a simple and practical way to make the sensor robust to target reflectivity variations. The paper also provides useful memory, latency, and energy data for two MCU families. However, the quantitative accuracy numbers are tied to a speaker-voltage proxy for ground truth, and the energy measurement protocol appears to mix batch and streaming modes; these issues need clarification before the claims can be fully accepted.

major comments (4)
  1. [Section IV.D, Tables IV-V] The reported accuracy metrics (PCC, MAE, LRS) are computed against a ground truth that is derived from the voltage applied to the speaker, scaled by a λ/V factor claimed linear within 10%. The network is trained on this same proxy and evaluated against it, so any speaker nonlinearity, hysteresis, or calibration bias is absorbed into the learned mapping and cannot be detected by the reported numbers. While Section V acknowledges that accuracy is ultimately bounded by the training-set ground truth, this limitation is not reflected in the abstract or the central accuracy claims. Please provide either an independent optical calibration for at least a subset of the evaluation data, or a quantitative uncertainty analysis showing how a ±10% calibration error propagates to PCC/MAE. This is load-bearing for the sensor's measurement-accuracy claim, though not for the latency claim.
  2. [Section IV.B, Table I] The energy measurements use a 500 ms input buffer and report 'Total time' values (e.g., 797 ms for the STM32U575ZI ResNet) that exceed the 500 ms window. This suggests the measurement does not reflect the double-buffered streaming operation described in Section III.B and used in the live demo. As written, the configuration that is claimed to be real-time appears to take 797 ms to process 500 ms of signal, which is contradictory. Please clarify the measurement timeline, specify whether the MCU is active during data collection, and report energy under the actual real-time streaming mode. If the current numbers are intended as a conservative batch-mode estimate, state that explicitly and give the streaming-mode energy as well.
  3. [Section IV.E and Conclusion] The conclusion that int16 quantization does not significantly degrade performance and int8 causes a loss is based on single training runs and single inference passes, with no confidence intervals or repeated experiments. The differences in PCC between float32 and int16 are about 0.005, which could easily be within training variability. Please provide multiple training runs with different seeds (or another statistical treatment) to support the claim that int16 is the preferred format. This is directly relevant to the deployment choice made in the paper.
  4. [Section II.B] The sentence in the conclusion stating that 'a displacement with frequency up to 500Hz could in principle be measured (from the sampling theorem)' is misleading. The network is trained only on target motions in the 5–100 Hz range (Section II.B and V), and the 48 kHz sampling rate alone does not guarantee that the learned mapping generalizes to 500 Hz. Please remove or heavily qualify this statement, or provide a dedicated validation at higher frequencies.
minor comments (5)
  1. [Table I] The row 'CNN [24] float32 0.8551 0.74' appears incomplete; it is missing the MSE and MAE columns that are filled for the other rows. Please correct the formatting or provide the missing values.
  2. [Section IV.A] The abbreviation 'RMPSprop' is likely a typo for 'RMSprop'. Also, the explanation of which model received random scaling during training could be clearer: the text says 'both networks are trained' with random scaling, but then says scaling 'has no impact' on the normalized network; please spell out the exact training protocol for each model.
  3. [Section IV.E] The phrase 'the first demonstration' is a strong precedence claim. Unless a comprehensive prior-art search confirms uniqueness, please soften to 'a first demonstration' or provide specific comparative evidence.
  4. [Section IV.D.1] There is a typo: 'an ajustable constant-current power supply' should be 'an adjustable constant-current power supply'.
  5. [Section II.B] The phrase 'this dataset comes from a calibration of the λ/V scaling factor or a speaker which is linear' is ungrammatical. Presumably 'of' was intended. Please rephrase to make clear whether the calibration and linearity are separate checks or alternatives.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the held-out aperiodic generalization set and independent on-board latency/energy measurements support the central claims; the speaker-voltage ground-truth proxy is a validity caveat, not a circular reduction.

full rationale

The derivation chain is: train a ResNet on the public periodic SMI dataset of [26], test generalization on the separate aperiodic subset (different feedback alignment and C), then quantize and deploy on an STM32 and measure latency/energy. No step equates a prediction with its own input by construction. The aperiodic generalization set is not used for training, so the reported PCC/MAE values (Table I, Fig. 7) are genuine out-of-distribution results, not fit residuals. The live comparison in Section IV.E uses a speaker-voltage proxy for ground truth: "the true displacement is computed as the time derivative of the instantaneous voltage applied to the speaker and scaled to units of λ/ms by applying an independently determined λ/V scaling factor." This is a calibration-validity limitation, acknowledged in Section V ("the accuracy would also be ultimately bound by the ground truth accuracy of the training set"), but it is not circular: the network never saw the live label, and the held-out test already uses a different feedback condition, so the mapping is not forced. The self-citations [17], [18], [22], [23], [24], [26] are prior work, public datasets, and open tools; the architecture is a standard ResNet [25], and the paper explicitly states "our architecture choice is not the only possible one," so no uniqueness theorem or ansatz is smuggled in. The real-time feasibility claim rests on timer-based measurements (Tables II and III) and energy measurements (Tables IV and V) that are independent of label accuracy. Hence I find no circularity.

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

The central claim is an engineering demonstration, so the main free parameters are the learned network weights, training hyperparameters, and quantization choices. The physical assumptions come from the Lang-Kobayashi model and the calibration method. No new physical entities are introduced.

free parameters (4)
  • Network weights (ResNet, 6937 parameters) = trained on periodic displacement subset via Adam
    The displacement prediction is a learned function of these weights; the reported accuracy is specific to this trained model.
  • Training hyperparameters (learning rate, weight decay, epochs) = LR 0.002, decay 5e-4, 120 epochs
    Chosen by the authors to reach convergence; influence final model quality.
  • Input normalization layer = min-max scaling per 48-sample window to [0,1]
    A preprocessing choice that assumes amplitude carries no displacement information; it is fixed and not trained.
  • Quantization formats = Q7.9 for int16; per-layer for int8
    Empirically chosen to preserve accuracy; affect deployment performance.
assumptions (4)
  • domain assumption Lang-Kobayashi equations describe the self-mixing interferometry response with a delayed feedback term.
    The physical model underpinning the sensor; taken from prior literature ([8]). Invoked in the introduction.
  • domain assumption The amplitude of the interferometric signal is not informative about displacement; only the shape of the fringes carries the measurement information.
    This justifies the normalization layer (Section II.A) and the robustness claim. It is derived from the physics of SMI, where amplitude depends on reflectivity.
  • domain assumption The speaker voltage provides a ground-truth displacement within 10% linearity over the operating frequency range.
    All training labels and evaluation metrics depend on this calibration (Section II.B). If the calibration is biased, the reported accuracy is invalid.
  • domain assumption The training data domain (C near unity, speed < 2.5 λ/ms, frequency 5-100 Hz) is representative of intended real-world use.
    The authors state in the conclusion that the network's validity is bound by this domain. The real-time demo stays within this domain.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Real-Time Smart Self-Mixing Interferometry Sensor With Embedded Neural Network." pith.science (2026). https://pith.science/paper/3TC6BN7I

@misc{pith2026260803448,
  author       = {Pith},
  title        = {Pith review of: Real-Time Smart Self-Mixing Interferometry Sensor With Embedded Neural Network},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/3TC6BN7I}},
  note         = {Machine review of arXiv:2608.03448}
}
read the original abstract

Self-mixing interferometry is a measurement ap- proach in which a laser beam is re-injected into the emitting laser itself after reflection on a target. Information about the position of the target can be obtained from monitoring the voltage across the laser. However, analyzing this signal is difficult. In previous works, neural networks have been used with great success to process this data. In this article, we present an updated prototype of an integrated sensor based on self-mixing interferometry with embedded neural networks. It consists of a semiconductor laser (acting both as light emitter and detector) equipped with an embedded platform for data processing. The platform includes an ADC (Analog-to-Digital Converter) and an STM32U575ZI microcontroller. The microcontroller runs a residual neural network in charge of reconstructing the displacement of a target from the interferometric signal entering the ADC. We assess the robustness of the neural network to unwanted signal amplitude variations, the impact of different network weights quantization choices required to run the network on the microcontroller, and the energy consumption of the system. Finally, we provide a demonstration of target displacement reconstruction fully running on the embedded platform in real-time. Our results pave the way towards robust, low power and versatile sensors based on self-mixing interferometry and embedded neural networks.

Figures

Figures reproduced from arXiv: 2608.03448 by the authors.

Figure 1
Figure 1. Example of the regression task to be realized. Top: the [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Neural network structure. Same length padding is used [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 4
Figure 4. Top: Nucleo-L476RG embedded platform, bottom:Nucleo-U575ZI-Q embedded platform. Both boards are equipped with a custom capture board based on a TLV320ADC3101 Analog to Digital Converter (see text). Our prototyping embedded platform consists of an STMi￾croelectronics Nucleo microcontroller board (white PCB) mounted with a custom capture interface (green PCB). Two different microcontroller-based development boards are… view at source ↗
Figures from the paper (4 more)
Figure 5
Figure 5. Figure 5: Performance comparison between a network without [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 7
Figure 7. Figure 7: Predictions for float32 (F32), int16 (I16) and int8 (I8) [PITH_FULL_IMAGE:figures/full_fig_p006_7.png]
Figure 6
Figure 6. Figure 6: Robustness against input signal amplification for mod [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]
Figure 8
Figure 8. Figure 8: Example of real-time inference on the STM32U575ZI [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 31 canonical work pages

  1. [1]

    Laser diode self- mixing technique for sensing applications,

    G. Giuliani, M. Norgia, S. Donati, and T. Bosch, “Laser diode self- mixing technique for sensing applications,”Journal of optics A: Pure and applied optics, vol. 4, no. 6, p. S283, 2002

    work page 2002

  2. [2]

    D. M. Kane and K. A. Shore,Unlocking dynamical diversity: optical feedback effects on semiconductor lasers. John Wiley & Sons, 2005

    work page 2005

  3. [3]

    Developing self-mixing interferometry for instrumentation and measurements,

    S. Donati, “Developing self-mixing interferometry for instrumentation and measurements,”Laser & Photonics Reviews, vol. 6, no. 3, pp. 393– 417, 2012

    work page 2012

  4. [4]

    Laser feedback interferometry: a tutorial on the self-mixing effect for coherent sensing,

    T. Taimre, M. Nikoli ´c, K. Bertling, Y . L. Lim, T. Bosch, and A. D. Raki´c, “Laser feedback interferometry: a tutorial on the self-mixing effect for coherent sensing,”Advances in Optics and Photonics, vol. 7, no. 3, pp. 570–631, 2015

    work page 2015

  5. [5]

    Laser feedback interferometry and applications: a review,

    J. Li, H. Niu, and Y . X. Niu, “Laser feedback interferometry and applications: a review,”Optical Engineering, vol. 56, no. 5, p. 050901, 2017

    work page 2017

  6. [6]

    Sensing and imaging using laser feedback interferometry with quantum cascade lasers,

    A. Raki ´c, T. Taimre, K. Bertling, Y . Lim, P. Dean, A. Valavanis, and D. Indjin, “Sensing and imaging using laser feedback interferometry with quantum cascade lasers,”Applied Physics Reviews, vol. 6, no. 2, p. 021320, 2019

    work page 2019

  7. [7]

    Versatile multimodality imaging system based on detectorless and scanless optical feedback interferometry—a retrospective overview for a prospective vision,

    M. Brambilla, L. L. Columbo, M. Dabbicco, F. De Lucia, F. P. Mezzapesa, and G. Scamarcio, “Versatile multimodality imaging system based on detectorless and scanless optical feedback interferometry—a retrospective overview for a prospective vision,”Sensors, vol. 20, no. 20, p. 5930, 2020

    work page 2020

  8. [8]

    External optical feedback effects on semiconductor injection laser properties,

    R. Lang and K. Kobayashi, “External optical feedback effects on semiconductor injection laser properties,”IEEE journal of Quantum Electronics, vol. 16, no. 3, pp. 347–355, 1980

    work page 1980

Show all 33 references
  1. [9]

    Interferometric measure- ments of displacement on a diffusing target by a speckle tracking technique,

    M. Norgia, S. Donati, and D. D’Alessandro, “Interferometric measure- ments of displacement on a diffusing target by a speckle tracking technique,”IEEE Journal of quantum electronics, vol. 37, no. 6, pp. 800–806, 2001

    2001

  2. [10]

    Adaptive self-mixing vibrometer based on a liquid lens,

    U. Zabit, R. Atashkhooei, T. Bosch, S. Royo, F. Bony, and A. Rakic, “Adaptive self-mixing vibrometer based on a liquid lens,”Optics letters, vol. 35, no. 8, pp. 1278–1280, 2010

    2010

  3. [11]

    Robust method of stabi- lization of optical feedback regime by using adaptive optics for a self- mixing micro-interferometer laser displacement sensor,

    O. D. Bernal, U. Zabit, and T. M. Bosch, “Robust method of stabi- lization of optical feedback regime by using adaptive optics for a self- mixing micro-interferometer laser displacement sensor,”IEEE Journal of Selected Topics in Quantum Electronics, vol. 21, no. 4, pp. 336–343, 2014

    2014

  4. [12]

    Self-mixing vibrometer with real-time digital signal elaboration,

    A. Magnani, A. Pesatori, and M. Norgia, “Self-mixing vibrometer with real-time digital signal elaboration,”Applied Optics, vol. 51, no. 21, pp. 5318–5325, 2012

    2012

  5. [13]

    Speckle-insensitive fringe detec- tion method based on hilbert transform for self-mixing interferometry,

    A. L. Arriaga, F. Bony, and T. Bosch, “Speckle-insensitive fringe detec- tion method based on hilbert transform for self-mixing interferometry,” Applied optics, vol. 53, no. 30, pp. 6954–6962, 2014

    2014

  6. [14]

    All Analog Processing of Speckle Affected Self-Mixing Interferometric Signals,

    A. A. Siddiqui, U. Zabit, O. D. Bernal, G. Raja, and T. Bosch, “All Analog Processing of Speckle Affected Self-Mixing Interferometric Signals,”IEEE Sensors Journal, vol. 17, no. 18, pp. 5892–5899, Sep. 2017

    2017

  7. [15]

    Detection of multimodal fringes for self-mixing-based vibration measurement,

    M. Usman, U. Zabit, O. D. Bernal, G. Raja, and T. Bosch, “Detection of multimodal fringes for self-mixing-based vibration measurement,”IEEE Transactions on Instrumentation and Measurement, vol. 69, no. 1, pp. 258–267, 2019

    2019

  8. [16]

    Toward an estimation of the optical feedback factor c on the fly for displacement sensing,

    O. D. Bernal, U. Zabit, F. Jayat, and T. Bosch, “Toward an estimation of the optical feedback factor c on the fly for displacement sensing,” Sensors, vol. 21, no. 10, p. 3528, 2021

    2021

  9. [17]

    Convolutional neural network for self- mixing interferometric displacement sensing,

    S. Barland and F. Gustave, “Convolutional neural network for self- mixing interferometric displacement sensing,”Optics Express, vol. 29, no. 8, pp. 11 433–11 444, 2021

    2021

  10. [18]

    High-availability displacement sensing with multi-channel self mixing interferometry,

    R. Matha, S. Barland, and F. Gustave, “High-availability displacement sensing with multi-channel self mixing interferometry,”Optics Express, vol. 31, no. 13, pp. 21 911–21 923, 2023

    2023

  11. [19]

    Self-mixing interferometric signal enhancement using generative adversarial network for laser metric sensing applications,

    I. Ahmed, U. Zabit, and A. Salman, “Self-mixing interferometric signal enhancement using generative adversarial network for laser metric sensing applications,”IEEE Access, vol. 7, pp. 174 641–174 650, 2019

    2019

  12. [20]

    Fringe slope discrimination in laser self-mixing interferometry using artificial neural network,

    K. Kou, C. Wang, T. Lian, and J. Weng, “Fringe slope discrimination in laser self-mixing interferometry using artificial neural network,”Optics & Laser Technology, vol. 132, p. 106499, 2020

    2020

  13. [21]

    Measuring parameters of laser self-mixing inter- ferometry sensor based on back propagation neural network,

    L. An and B. Liu, “Measuring parameters of laser self-mixing inter- ferometry sensor based on back propagation neural network,”Optics Express, vol. 30, no. 11, pp. 19 134–19 144, 2022

    2022

  14. [22]

    P.-E. Novac. Qualia. [Online]. Available: https://github.com/LEAT- EDGE/qualia IEEE SAS 2024 submission: #1571011682 10

    2024

  15. [23]

    Quantization and deployment of deep neural networks on microcontrollers,

    P.-E. Novac, G. Boukli Hacene, A. Pegatoquet, B. Miramond, and V . Gripon, “Quantization and deployment of deep neural networks on microcontrollers,”Sensors, vol. 21, no. 9, 2021

    2021

  16. [24]

    Integrating embedded neural networks and self-mixing interferometry for smart sensors design,

    P.-E. Novac, L. Rodriguez, and S. Barland, “Integrating embedded neural networks and self-mixing interferometry for smart sensors design,” in Proceedings of the 2024 IEEE Sensors Applications Symposium, 2024

    2024

  17. [25]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” inProceedings of the IEEE conference on computer vision and pattern recognition, 2016, pp. 770–778

    2016

  18. [26]

    Displacement measurement via self mixing interferometry and neural network training set,

    S. Barland and F. Gustave, “Displacement measurement via self mixing interferometry and neural network training set,” nov 2022. [Online]. Available: https://doi.org/10.5281/zenodo.7303745

    2022 doi

  19. [27]

    Goodfellow, Y

    I. Goodfellow, Y . Bengio, and A. Courville,Deep Learning. MIT Press, 2016, http://www.deeplearningbook.org

    2016

  20. [28]

    Enabling deep learning at the iot edge,

    L. Lai and N. Suda, “Enabling deep learning at the iot edge,” inPro- ceedings of the 2018 IEEE/ACM International Conference on Computer- Aided Design, ser. ICCAD, 2018, pp. 1–6, CMSIS stands for Common Microcontroller Software Interface Standard

    2018

  21. [29]

    STM32L476RG, https://www.st.com/en/microcontrollers- microprocessors/stm32l476rg.html

  22. [30]

    STM32U575ZI, https://www.st.com/en/microcontrollers- microprocessors/stm32u575zi.html

  23. [31]

    A white paper on neural network quantization,

    M. Nagel, M. Fournarakis, R. A. Amjad, Y . Bondarenko, M. van Baalen, and T. Blankevoort, “A white paper on neural network quantization,” 2021

    2021

  24. [32]

    Fast and highly accurate phase unwrapping algorithm for displacement retrieval using self-mixing interferometry sensor,

    M. S. Orakzai, S. Amin, Z. A. Khan, and F. Akram, “Fast and highly accurate phase unwrapping algorithm for displacement retrieval using self-mixing interferometry sensor,”Optical materials, vol. 129, p. 112553, 2022

    2022

  25. [33]

    Vibration reconstruction and optical feedback parameter evaluation based on the direction discrimination in self-mixing interferometry,

    C.-Y . Ri, J.-H. Choe, H.-R. Ri, C.-M. Pak, K.-R. Ri, and J.-M. O, “Vibration reconstruction and optical feedback parameter evaluation based on the direction discrimination in self-mixing interferometry,” Applied Optics, vol. 60, no. 13, pp. 3801–3807, 2021

    2021

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.