Implementation of Deep Learning-Based Automatic Modulation Classifier on FPGA SDR Platform
Abstract
:1. Introduction
- A new structure of AMR based on a stacked convolution autoencoder is proposed [16]. The purpose of neural networks is to approximate the transformation function from the input layer to output layer. Compared with neural networks, the autoencoder can obtain the sparse feature of the signal automatically. The stacked autoencoders can learn multiple expressions of the original data layer by layer. Each layer is based on the expression of the bottom layer, but is often more abstract and more suitable for complex tasks such as classification. Compared with the LRT method, it is no longer necessary to spend a lot of work to build a signal, noise model, and a cost function. Moreover, the randomness of the channel environment makes these models vulnerable to failure. Compared with FB’s method, this method can automatically generate signal characteristics. This feature is suitable for some special occasions, particularly military communications. Because the waveform of the signal in these cases is often non-standard and time-varying in time, frequency and the other domains, the signal characteristics are difficult to predict.
- We used the reconfigurability of field-programmable gate arrays (FPGA) to propose a master–slave AMR architecture. In this architecture, shown in Figure 2, the deep neural network setup and training tasks are performed on a workstation equipped with a graphics processing unit. First, the RF signal digitizer converts the modulated RF signal to sampled data and enters the workstation for training the network. After completing the neural network training, it is converted to the FPGA hardware configuration file on the workstation. Finally, the hardware configuration file is transmitted to the FPGA SDR platform by wireless or wired communication, updating the configuration of the AMR function. The trained hardware model can be operated on portable radio equipment, thus reducing the requirements of hardware. If there is a new type of modulation that needs to be identified, the system needs to be retrained. After the training is completed, new weights are generated. At this time, the original FPGA fabric needs to be updated. Therefore, reconfigurability is necessary for the evolution of the system. The current AMR is not reconfigurable, so no recognition can be made for the new modulation type. With respect to such a situation, the proposed scheme adds value to the actual project. This is also a very important function for identifying non-standard and time-varying signals in multiple domains. As a large number of intelligent radio terminals can share the training results of a workstation, the computational costs can be decreased considerably.
2. Construction of Automatic Modulation Recognition
2.1. Problem Description
2.1.1. Signal Characteristics
2.1.2. Recognition of Modulation
2.2. Construction Methods
2.2.1. Structure of Stacked CAEs
2.2.2. Training and Classification
3. Implementation
3.1. Architecture
3.2. AMR IP
4. Results
4.1. Use of Resources
4.2. Experiment
5. Discussion
Author Contributions
Funding
Conflicts of Interest
References
- Xu, J.; Su, W.; Zhou, M. Software-defined radio equipped with rapid modulation recognition. IEEE Trans. Veh. Technol. 2010, 59, 1659–1667. [Google Scholar] [CrossRef]
- Yucek, T.; Arslan, H. A survey of spectrum sensing algorithms for cognitive radio applications. IEEE Commun. Surv. Tutor. 2009, 11, 116–130. [Google Scholar] [CrossRef]
- Dobre, O.A. Signal identification for emerging intelligent radios: classical problems and new challenges. IEEE Instrum. Meas. Mag. 2015, 18, 11–18. [Google Scholar] [CrossRef]
- El-Mahdy, A.; Namazi, N. Classification of Multiple M-ary Frequency-Shift Keying Signals Over a Rayleigh Fading Channel. IEEE Trans. Commun. 2002, 50, 967–974. [Google Scholar] [CrossRef]
- Panagiotou, P.; Anastasoupoulos, A.; Polydoros, A. Likelihood ratio tests for modulation classification. In Proceedings of the 21st Century Military Communications Conference (CMCC 2000), Los Angeles, CA, USA, 22–25 October 2000; pp. 670–674. [Google Scholar]
- Hameed, F.; Dobre, O.A.; Popescu, D.C. On the likelihood-based approach to modulation classification. IEEE Trans. Wirel. Commun. 2009, 8, 5884–5892. [Google Scholar] [CrossRef]
- Dobre, O.A.; Rajan, S.; Inkol, R. Joint signal detection and classification based on first-order cyclostationarity for cognitive radios. EURASIP J. Adv. Signal Process. 2009, 2009, 656719. [Google Scholar] [CrossRef]
- Jerjawi, W.A.; Eldemerdash, Y.A.; Dobre, O.A. Second-Order Cyclostationarity-Based Detection of LTE SC-FDMA Signals for Cognitive Radio Systems. IEEE Trans. Instrum. Meas. 2015, 64, 823–833. [Google Scholar] [CrossRef] [Green Version]
- Choqueuse, V.; Yao, K.; Collin, L.; Burel, G. Hierarchical space-time block code recognition using correlation matrices. IEEE Trans. Wirel. Commun. 2008, 7, 3526–3534. [Google Scholar] [CrossRef] [Green Version]
- Marey, M.; Dobre, O.A.; Inkol, R. Blind STBC identification for multiple antenna OFDM systems. IEEE Trans. Commun. 2014, 62, 1554–1567. [Google Scholar] [CrossRef]
- Hassan, K.; Dayoub, I. Automatic Modulation Recognition Using Wavelet Transform and Neural Networks in Wireless Systems. Eurasip J. Adv. Sig. Proc. 2010, 2010, 532898. [Google Scholar] [CrossRef]
- Mobasseri, B.G. Digital modulation classification using constellation shape. Sig. Proc. 2000, 80, 251–277. [Google Scholar] [CrossRef] [Green Version]
- Migliori, B.; Zeller-Townson, R.; Grady, D.; Gebhardt, D. Biologically Inspired Radio Signal Feature Extraction with Sparse Denoising Autoencoders; Technical Report for Space and Naval Warfare Systems: San Diego, CA, USA, 2016. [Google Scholar]
- O’Shea, T.J.; Corgan, J.; Clancy, T.C. Unsupervised representation learning of structured radio communication signals. In Proceedings of the IEEE 2016 First International Workshop on Sensing, Processing and Learning for Intelligent Machines (SPLINE), Aalborg, Denmark, 6–8 July 2016; pp. 1–5. [Google Scholar]
- Ranzato, M.; Fujie, H.; Boureau, Y.L.; LeCun, Y. Unsupervised Learning of Invariant Feature Hierarchies with Applications to Object Recognition. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Minneapolis, MN, USA, 17–22 June 2007. [Google Scholar] [CrossRef]
- Masci, J.; Meier, U.; Cireşan, D. Stacked convolutional auto-encoders for hierarchical feature extraction. In Proceedings of the International Conference on Artificial Neural Networks (ICANN), Espoo, Finland, 14–17 June 2011; pp. 52–59. [Google Scholar]
- Alex, K.; Ilya, S.; Hinton, G.E. ImageNet Classification with Deep Convolutional Neural Networks. In Proceedings of the 25th International Conference on Neural Information Processing Systems (NIPS), Lake Tahoe, Nevada, 3–6 December 2012; pp. 1097–1105. [Google Scholar]
- Zeiler, M.D.; Fergus, R. Visualizing and Understanding Convolutional Networks. Available online: http://arxiv.org/abs/1311 (accessed on 15 September 2016).
- Badrinarayanan, V.; Kendall, A.; Cipolla, R. Segnet: A Deep Convolutional Encoder-decoder Architecture for Image Segmentation. Available online: http://arxiv.org/abs/1511.00561 (accessed on 15 September 2016).
- Radford, L.; Metz, A.; Chintala, S. Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks. Available online: http://arxiv.org/abs/1511.06434/ (accessed on 15 September 2016).
- Kingma, D.P.; Welling, M. Auto-Encoding Variation Bayes. Cornell University Library. Available online: http://arxiv.org/abs/1312.6114 (accessed on 2 June 2018).
- Zhao, J.; Mathieu, M.; Goroshin, R.; LeCun, Y. Stacked what-where auto-encoders. Available online: http://arXiv.org/abs/1603.07285 (accessed on 15 September 2016).
- Bengio, Y.; Lamblin, P.; Popovici, D.; Larochelle, H. Greedy Layer-Wise Training of Deep Networks. In Proceedings of the International Conference on Neural Information Processing Systems (NIPS), Vancouver, BC, Canada, 4–7 December 2006; pp. 153–160. [Google Scholar]
- Bottou, L. Large-Scale Machine Learning with Stochastic Gradient Descent. In Proceedings of the 19th International Conference on Computational Statistics, Paris, France, 22–27 August 2010; pp. 177–186. [Google Scholar]
- Zerioul, L.; Ariaudo, M.; Bourdel, E. RF transceiver and transmission line behavioral modeling in VHDL-AMS for wired RFNoC. Analog Integr. Circ. Signal Process. 2017, 92, 103–114. [Google Scholar] [CrossRef]
- Zhang, C.; Li, P.; Sun, G. Optimizing FPGA-based Accelerator Design for Deep Convolutional Neural Networks. In Proceedings of the ACM/SIGDA International Symposium on Field-Programmable Gate Arrays, Monterey, CA, USA, 22–24 February 2015; pp. 161–170. [Google Scholar]
- Xue, J. Loop Tiling for Parallelism; The Springer International Series in Engineering and Computer; Springer Science & Business Media: Berlin, Germany, 2015; Volume 575, pp. 154–196. [Google Scholar]
- Pouchet, L.N.; Zhang, P.; Sadayappan, P. Polyhedral-based data reuse optimization for configurable computing. In Proceedings of the ACM/SIGDA International Symposium on Field-Programmable Gate Arrays, Monterey, CA, USA, 11–13 February 2013; pp. 29–38. [Google Scholar]
- Powers, D.M. Evaluation: From Precision, Recall and F-Measure to ROC, Informedness, Markedness & Correlation. J. Mach. Learn. Technol. 2011, 2, 37–63. [Google Scholar]
Property Name | Parameter | Value |
---|---|---|
Number of samples per symbol | NSS | 20 |
Number of samples per vector | NSV | 200 |
Number of training vectors | NVTr | 60,000 |
Number of training vectors per modulation | NVM | 10,000 |
Number of test vectors | NVTe | 10,000 |
Property Name | Parameter | Value |
---|---|---|
Activation function | ReLU | |
Convolution depth | 1 | |
Convolution filters | 256 | |
Number of CAEs | 2 |
Component Name | BRAM18 | DSP48 | FF | LUT |
---|---|---|---|---|
Fully connected layer | 8 | 8 | 658 | 1183 |
Convolution | 34 | 136 | 4882 | 1232 |
Tanh | 1 | 0 | 51 | 206 |
Sigmoid | 1 | 0 | 56 | 182 |
ReLU | 0 | 0 | 12 | 45 |
Maxpool | 2 | 0 | 77 | 252 |
© 2018 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 (http://creativecommons.org/licenses/by/4.0/).
Share and Cite
Tang, Z.-L.; Li, S.-M.; Yu, L.-J. Implementation of Deep Learning-Based Automatic Modulation Classifier on FPGA SDR Platform. Electronics 2018, 7, 122. https://doi.org/10.3390/electronics7070122
Tang Z-L, Li S-M, Yu L-J. Implementation of Deep Learning-Based Automatic Modulation Classifier on FPGA SDR Platform. Electronics. 2018; 7(7):122. https://doi.org/10.3390/electronics7070122
Chicago/Turabian StyleTang, Zhi-Ling, Si-Min Li, and Li-Juan Yu. 2018. "Implementation of Deep Learning-Based Automatic Modulation Classifier on FPGA SDR Platform" Electronics 7, no. 7: 122. https://doi.org/10.3390/electronics7070122