# EMI Threat Assessment of UAV Data Link Based on Multi-Task CNN

^{1}

^{2}

^{*}

## Abstract

**:**

## 1. Introduction

- The data is constructed based on the EMI injection test rather than simulation to solve the real-world problem.
- A graphical preprocessing and enhancement method for EMI signals is proposed to fuse the heterogeneous information of EMI signal and data link performance.
- Based on the series-parallel structure and the balance loss function, the proposed MIMT-CNN can achieve a balance between interference identification and threat assessment performance.

## 2. EMS Measurement and Prediction

_{0}− 5 MHz to f

_{0}+ 5 MHz and a bandwidth range within 10 MHz.

#### 2.1. Original Data Acquisition

_{AGC}, p

_{SNR}, p

_{BER}indicate the automatic gain control (AGC) voltage, signal-to-noise ratio (SNR) and bit error rate (BER), respectively, which are collected from the monitoring software of the data link. Meanwhile, the I/Q data of the received signal of the data link are collected by an electromagnetic spectrum monitoring receiver connected to the receiving antenna.

#### 2.2. EMI Effects

#### 2.3. Proposed Method

- (1)
- The original data are obtained through the EMS injection experiment, including the state parameters of the data link and the I/Q data of the received signal by the data link in the presence of interference.
- (2)
- The state parameters of the data link are converted to visualized performance parameters (VPP). Meanwhile, the I/Q data are transformed to STFT spectrograms and normalized density constellation (NDC), which denotes the time-frequency and phase information of the EMI signal, respectively.
- (3)
- The MIMT-CNN is constructed and trained on the training set. By using the Bayesian optimization, the hyperparameters of the network are optimized on the validation set.
- (4)
- The trained network is tested on the test set. According to the actual results of EMI signal classification and threat level prediction, the accuracy and generalization ability of the model are evaluated.

## 3. Data Preprocessing

#### 3.1. Visualized Performance Parameters

#### 3.2. STFT Spectrogram

#### 3.3. Normalized Density Constellation

## 4. MIMT-CNN Modeling

#### 4.1. Network Structure

#### 4.1.1. Feature Extraction Layer

#### 4.1.2. Feature Fusion Processing Lazyer

#### 4.1.3. Multi-Task Output Layer

#### 4.2. Multi-Task Loss Function

#### 4.2.1. EMI Classification Loss

#### 4.2.2. Threat Prediction Loss

#### 4.2.3. Balanced Loss

#### 4.3. Evaluation Indicators

## 5. Hyperparameter Optimization and Model Training

#### 5.1. Bayesian Optimization

#### 5.2. Training Setting

^{−6}is obtained by Bayesian optimization, which then decreases to 50% of the previous value after one epoch. The Adam optimizer uses gradient estimation to directly calculate the adaptive learning rate of different parameters, which has the advantages of high computational efficiency and small memory requirement [33].

#### 5.3. Training Result

## 6. Model Test Result

#### 6.1. Influence of the Structure on Network Performance

#### 6.2. EMI Classification Performance

#### 6.3. Prediction Performance

#### 6.4. Prediction Performance

## 7. Conclusions

## Author Contributions

## Funding

## Data Availability Statement

## Conflicts of Interest

## References

- Mehta, P.; Gupta, R.; Tanwar, S. Blockchain envisioned UAV networks: Challenges, solutions, and comparisons. Comput. Commun.
**2020**, 151, 518–538. [Google Scholar] [CrossRef] - Ch, R.; Srivastava, G.; Reddy Gadekallu, T. Security and privacy of UAV data using blockchain technology. J. Inf. Secur. Appl.
**2020**, 55, 102670. [Google Scholar] [CrossRef] - Le Cun, Y.; Bengio, Y.; Hinton, G. Deep learning. Nature
**2015**, 521, 436–444. [Google Scholar] [CrossRef] [PubMed] - Wu, C.; Kim, H.; Penugonda, S.; Fan, J. Analysis and modeling of the common-mode conducted EMI from a wireless power transfer system for mobile applications. IEEE Trans. Electromagn. Compat.
**2021**, 63, 2143–2150. [Google Scholar] [CrossRef] - Han, J.H.; Ju, S.H.; Kang, N.W.; Lee, W.S.; Choi, J.S. Wideband coupling modeling analysis by arbitrarily incoming source fields based on the electromagnetic topology technique. IEEE Trans. Microw. Theory Tech.
**2019**, 67, 28–37. [Google Scholar] [CrossRef] - Houret, T.; Besnier, P.; Vauchamp, S.; Pouliguen, P. Probability of failure using the kriging-controlled stratification method and statistical inference. In Proceedings of the 2020 International Symposium on Electromagnetic Compatibility, Rome, Italy, 23–25 September 2020. [Google Scholar]
- Genender, E.; Garbe, H.; Sabath, F. Probabilistic risk analysis technique of intentional electromagnetic interference at system level. IEEE Trans. Electromagn. Compat.
**2014**, 56, 200–207. [Google Scholar] [CrossRef] - Zhang, D.X.; Zhao, M.; Cheng, E.W.; Chen, Y.Z. GPR-based EMI prediction for UAV’s dynamic datalink. IEEE Trans. Electromagn. Compat.
**2021**, 63, 19–29. [Google Scholar] [CrossRef] - Devaraj, L.; Ruddle, A.R.; Duffy, A.P. EMI risk estimation for system-level functions using probabilistic graphical models. In Proceedings of 2021 IEEE International Joint EMC/SI/PI and EMC Europe Symposium, Raleigh, NC, USA, 24 July–13 August 2021; IEEE: Piscataway, NJ, USA, 2021; pp. 851–856. [Google Scholar]
- Xu, T.; Wang, Y.M.; Zhang, D.X.; Zhao, M.; Chen, Y.Z. Prediction on EMS of UAV’s data link based on SSA-optimized dual-channel CNN. IEEE Trans. Electromagn. Compat.
**2022**, 64, 1346–1356. [Google Scholar] [CrossRef] - Zhou, H.; Jiao, L.; Zheng, S.; Yang, L.; Shen, W.; Yang, X. Generative adversarial network-based electromagnetic signal classification: A semi-supervised learning framework. China Commun.
**2020**, 17, 157–169. [Google Scholar] [CrossRef] - Jin, H.; Gu, Z.M.; Tao, T.M.; Li, E. Hierarchical attention-based machine learning model for radiation prediction of WB-BGA package. IEEE Trans. Electromagn. Compat.
**2021**, 63, 1972–1980. [Google Scholar] [CrossRef] - Shu, Y.; Wei, X.; Fan, J.; Yang, R.; Yang, Y. An equivalent dipole model hybrid with artificial neural network for electromagnetic interference prediction. IEEE Trans. Microw. Theory Tech.
**2019**, 67, 1790–1797. [Google Scholar] [CrossRef] - Yuan, S.; Lin, P.; Chang, C.; Dong, J.; Su, C. Classification of an embedded system instruction EMI using a deep convolutional neural network. In Proceedings of 2019 Joint International Symposium on Electromagnetic Compatibility and Asia-Pacific International Symposium on Electromagnetic Compatibility, Sapporo, Japan, 3–7 June 2019; IEEE: Piscataway, NJ, USA, 2019; pp. 338–341. [Google Scholar]
- Zhang, H.; Yuan, L.; Wu, G.; Zhou, F.; Wu, Q. Automatic modulation classification using involution enabled residual networks. IEEE Wirel. Commun. Lett.
**2021**, 10, 2417–2420. [Google Scholar] [CrossRef] - Sun, G. RF transmitter identification using combined siamese networks. IEEE Trans. Instrum. Meas.
**2021**, 71, 8000813. [Google Scholar] [CrossRef] - Zhang, Y.; Yang, Q. A survey on multi-task learning. IEEE Trans. Knowl. Data Eng.
**2022**, 34, 5586–5609. [Google Scholar] [CrossRef] - Maddikunta, P.K.R. Unmanned aerial vehicles in smart agriculture: Applications, requirements, and challenges. IEEE Sens. J.
**2021**, 21, 17608–17619. [Google Scholar] [CrossRef] - Hamdalla, M.Z.; Roacho-Valles, J.M.; Caruso, A.N.; Hassan, A.M. Electromagnetic Compatibility Analysis of Quadcopter UAVs Using the Equivalent Circuit Approach. IEEE Open J. Antennas Propag.
**2022**, 3, 1090–1101. [Google Scholar] [CrossRef] - Peng, S.; Sun, S.; Yao, Y.D. A survey of modulation classification using deep learning: Signal representation and data preprocessing. IEEE Trans. Neural Netw. Learn. Syst.
**2021**, 33, 7020–7038. [Google Scholar] [CrossRef] - Zhang, W.T.; Cui, D.; Lou, S.T. Training images generation for CNN based automatic modulation classification. IEEE Access
**2021**, 9, 62916–62925. [Google Scholar] [CrossRef] - Tang, B.; Tu, Y.; Zhang, Z.; Lin, Y. Digital signal modulation classification with data augmentation using generative adversarial nets in cognitive radio networks. IEEE Access
**2018**, 6, 15713–15722. [Google Scholar] [CrossRef] - Olshausen, B.A.; Field, D.J. Natural image statistics and efficient coding. Netw. Comput. Neural Syst.
**1996**, 7, 333–339. [Google Scholar] [CrossRef] - Jacek, T.; Kwolek, B. Multi-channels CNN temporal features for depth-based action recognition. In Proceedings of the 12th International Conference on Machine Vision, Amsterdam, The Netherlands, 16–18 November 2020. [Google Scholar]
- Wang, G.; Giannakis, G.B.; Chen, J. Learning relu networks on linearly separable data: Algorithm, optimality, and generalization. IEEE Trans. Signal Process.
**2019**, 67, 2357–2370. [Google Scholar] [CrossRef] [Green Version] - Hareth, S.; Mostafa, H.; Shehata, K.A. Low power CNN hardware FPGA implementation. In Proceedings of the 31st IEEE International Conference on Microelectronics, Cairo, Egypt, 15–18 December 2019; IEEE: Piscataway, NJ, USA, 2019. [Google Scholar]
- Szegedy, C. Going deeper with convolutions. In Proceedings of IEEE Conference CVPR, Boston, MA, USA, 8–10 July 2015; IEEE: Piscataway, NJ, USA, 2015; pp. 1–9. [Google Scholar]
- Awais, M.; Bin Iqbal, M.T.; Bae, S.H. Revisiting internal covariate shift for batch normalization. IEEE Trans. Neural Netw. Learn. Syst.
**2021**, 32, 5082–5092. [Google Scholar] [CrossRef] [PubMed] - Lin, T.Y.; Goyal, P.; Girshick, R.; He, K.; Dollár, P. Focal loss for dense object detection. IEEE Trans. Pattern Anal. Mach. Intell.
**2020**, 42, 318–327. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Shahriari, B.; Swersky, K.; Wang, Z.; Adams, R.P.; Freitas, N.D. Taking the human out of the loop: A review of bayesian optimization. Proc. IEEE
**2016**, 104, 148–175. [Google Scholar] [CrossRef] [Green Version] - Snoek, J.; Larochelle, H.; Adams, R.P. Practical bayesian optimization of machine learning algorithms. Adv. Neural Inf. Process. Syst.
**2012**, 25, 1467–5463. [Google Scholar] - Dolatsara, M.A.; Hejase, J.A.; Becker, W.D.; Kim, J.; Lim, S.K.; Swaminathan, M. Worst-case eye analysis of high-speed channels based on bayesian optimization. IEEE Trans. Electromagn. Compat.
**2021**, 63, 246–258. [Google Scholar] [CrossRef] - Ma, Y.; Guo, R.; Li, M.; Yang, F.; Xu, S.; Abubakar, A. Supervised descent method for 2D magnetotelluric inversion using Adam optimization. In Proceedings of 2019 International Applied Computational Electromagnetics Society Symposium, Nanjing, China, 8–11 August 2019; IEEE: Piscataway, NJ, USA, 2019. [Google Scholar]
- Mossad, O.S.; El Nainay, M.; Torki, M. Deep convolutional neural network with multi-task learning scheme for modulations recognition. In Proceedings of the 2019 15th International Wireless Communications & Mobile Computing Conference (IWCMC), Tangier, Morocco, 24–28 June 2019; IEEE: Piscataway, NJ, USA, 2019. [Google Scholar]
- Wang, J.; Wang, H.; Sun, Z. Research on the effectiveness of deep convolutional neural network for electromagnetic interference identification based on I/Q data. Atmosphere
**2022**, 13, 1785. [Google Scholar] [CrossRef] - Wei, M.; Chen, K.; Li, S.; Cao, J.; Ali, A. An intelligent method based on time-frequency analysis and deep learning semantic segmentation for investigating the electromagnetic pulse features of engine digital controllers. IEEE Trans. Electromagn. Compat.
**2023**, 65, 257–270. [Google Scholar] [CrossRef] - Mitiche, I.; Jenkins, M.D.; Boreham, P.; Nesbitt, A.; Morison, G.H. An expert system for EMI data classification based on complex Bispectrum representation and deep learning methods. Expert Syst. Appl.
**2021**, 171, 114568. [Google Scholar] [CrossRef] - Kim, J.M.; Bae, J.; Son, S.; Son, K.; Yum, S.G. Development of model to predict natural disaster-induced financial losses for construction projects using deep learning techniques. Sustainability
**2021**, 13, 5304. [Google Scholar] [CrossRef] - Barzegar, R.; Aalami, M.T.; Adamowski, J. Short-term water quality variable prediction using a hybrid CNN-LSTM deep learning model. Stoch. Environ. Res. Risk Assess.
**2020**, 34, 415–433. [Google Scholar] [CrossRef] - Pham, B.T. Flood risk assessment using deep learning integrated with multi-criteria decision analysis. Knowl.-Based Syst.
**2021**, 219, 106899. [Google Scholar] [CrossRef]

**Figure 2.**Variation of JSR when the UAV’s data link loses lock in presence of EMI signal with typical parameters: (

**a**) JSR varies with frequency offset of EMI when the bandwidths of WGN and BPSK are 1 MHz and 8 MHz and (

**b**) JSR varies with the EMI bandwidth when the center frequencies of WGN and BPSK are ${f}_{0}$, ${f}_{0}-4\text{}\mathrm{MHz}$, ${f}_{0}+2\text{}\mathrm{MHz}$.

**Figure 4.**Visualized performance parameters under different types of EMI when the data link loses lock (Taking the center frequency of CW, WGN, and BPSK interference signals at ${f}_{0}-4\text{}\mathrm{MHz}$ and the bandwidth of WGN and BPSK interference signals within 8 MHz as an example).

**Figure 5.**I/Q signal and STFT spectrogram when the data link loses lock with the EMI signal of: (

**a**) WGN with the center frequency of ${f}_{0}-4\text{}\mathrm{MHz}$ and the bandwidth of 8 MHz, (

**b**) BPSK with the center frequency of ${f}_{0}-4\text{}\mathrm{MHz}$ and the bandwidth of 8 MHz and (

**c**) CW with the center frequency of ${f}_{0}-4\text{}\mathrm{MHz}$.

Type of Method | Merit | Defect | Reference | |
---|---|---|---|---|

equivalent circuit and topological network | clear EMI mechanism interpretable model | need extensive professional knowledge on the equipment | equivalent circuit [4] | |

topological network [5] | ||||

statistical probability | based on strict mathematical theory. | need a lot of simplification, resulting in insufficient accuracy in practice | kriging-controlled stratification [6] | |

fault tree [7] | ||||

machine learning | based on data and has strong applicability | some model requires domain experts to extract features | Gaussian process regression [8] | |

probabilistic graphical models [9] | ||||

deep learning | autonomously learn from complex datastrong generalization ability | computationally expensive and has weak interpretability | convolutional neural network [10] | |

generative adversarial network [11] | ||||

residual networks [12] | ||||

artificial neural network [13] |

**Table 2.**The constellation diagram and NDC under different EMI signals when the data link loses lock.

EMI Signal Types | Continuous Wave | White Gaussian Noise | BPSK | ||||
---|---|---|---|---|---|---|---|

Constellation diagram | |||||||

Normalized density constellation (NDC) | |||||||

Center frequency | ${f}_{0}-4\text{}\mathrm{MHz}$ | ${f}_{0}+2\text{}\mathrm{MHz}$ | ${f}_{0}-4\text{}\mathrm{MHz}$ | ${f}_{0}+2\text{}\mathrm{MHz}$ | ${f}_{0}-4\text{}\mathrm{MHz}$ | ${f}_{0}+2\text{}\mathrm{MHz}$ | |

Bandwidth | - | 1 MHz | 8 MHz | 1 MHz | 8 MHz |

Parameters | Description | Values |
---|---|---|

${K}_{bs}$ | Batch size | 5 |

$lr$ | Initial learning rate | 6.1884 × 10^{−6} |

$ep$ | Number of epochs | 20 |

${c}_{1}$ | Size of convolutional layer 1 | 9 |

${c}_{2}$ | Size of convolutional layer 2 | 1 |

${c}_{3}$ | Size of convolutional layer 3 | 4 |

${n}_{1}$ | Number of convolutional layer 1 | 4 |

${n}_{2}$ | Number of convolutional layer 2 | 19 |

${n}_{3}$ | Number of convolutional layer 3 | 18 |

$fc$ | Size of fully connected layer | 29 |

Number of Input Channel | Input Channel | Classification | Prediction | Inferring Time (ms) | |||
---|---|---|---|---|---|---|---|

Input1 | Input2 | Input3 | Accuracy | RMSE | MAPE | ||

Double-input | - | STFT | constellation | 68.18% | 0.91 | 16.26% | 9.60 |

- | STFT | NDC | 74.62% | 0.94 | 24.09% | 10.03 | |

VPP | STFT | - | 73.11% | 0.73 | 17.05% | 8.64 | |

VPP | - | constellation | 90.15% | 0.72 | 16.03% | 10.67 | |

VPP | - | NDC | 93.94% | 0.62 | 22.60% | 9.89 | |

Three-input | VPP | STFT | constellation | 91.67% | 0.65 | 6.50% | 10.11 |

VPP | STFT | NDC | 95.45% | 0.49 | 10.83% | 14.81 |

Reference | Accuracy | RMSE | Algorithm | Application Scenario |
---|---|---|---|---|

Interference identification and threat assessment (this work)-Multi-task | ||||

Proposed Model | 95.45% | 0.49 | CNN | |

Interference identification- Single-task | ||||

[34] | 86.97% | - | CNN | Cognitive radio equipment |

[35] | 89% | - | U-Net | Radar receiver |

[36] | 92.7% | - | FCN | Engine digital controllers |

[37] | 96.53% | - | ResNet | High voltage power plants |

Threat assessment- Single-task | ||||

[38] | - | 0.975 | DNN | Financial loss |

[39] | - | 0.518 | CNN-LSTM | Water quality risk |

[10] | - | 0.3882 | CNN | Data link under EMI |

[40] | - | 0.231 | DNN | Flood risk |

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |

© 2023 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Xu, T.; Chen, Y.; Wang, Y.; Zhang, D.; Zhao, M.
EMI Threat Assessment of UAV Data Link Based on Multi-Task CNN. *Electronics* **2023**, *12*, 1631.
https://doi.org/10.3390/electronics12071631

**AMA Style**

Xu T, Chen Y, Wang Y, Zhang D, Zhao M.
EMI Threat Assessment of UAV Data Link Based on Multi-Task CNN. *Electronics*. 2023; 12(7):1631.
https://doi.org/10.3390/electronics12071631

**Chicago/Turabian Style**

Xu, Tong, Yazhou Chen, Yuming Wang, Dongxiao Zhang, and Min Zhao.
2023. "EMI Threat Assessment of UAV Data Link Based on Multi-Task CNN" *Electronics* 12, no. 7: 1631.
https://doi.org/10.3390/electronics12071631