基于深度时序卷积网络的功率放大器行为建模方法

doi:10.16180/j.cnki.issn1007-7820.2022.08.002

摘要/Abstract

摘要：

功率放大器作为辐射源发射机的核心部件,其工作行为具有高非线性、强记忆性等特点,导致功率放大器的行为建模难度颇高。针对该问题,文中提出了一种基于深度时序卷积网络的功率放大器行为建模方法。该方法采用的神经网络模型由多个多维时序卷积块构成,每个时序卷积块由数个用于增加网络感受野的因果膨胀卷积以及用于提高梯度反馈效率的残差结构组成。模型通过并行卷积操作,克服了传统卷积网络无法处理可变长序列的弊端,在保留功率放大器记忆特性的同时,提高了行为建模的效率。针对实测数据的行为建模结果表明,相比于现有的Volterra级数以及循环神经网络建模方法,文中提出的方法可显著提升行为建模精度,且在行为建模效率方面,相较于循环神经网络建模方法,将实现时间降低了一个数量级。

关键词: 辐射源, 功率放大器, 行为建模, 时序卷积网络, 残差结构, 因果卷积, 膨胀卷积, 深度网络

Abstract:

As the core component of the radiation source transmitter, the power amplifier has the characteristics of high non-linearity and strong memory, which makes it difficult to model the behavior of the power amplifier. In view of this problem, this study proposes a method for behavior modeling of power amplifiers based on Deep TCN. The neural network model adopted by this method is composed of multiple multi-dimensional time series convolution blocks, and each time series convolution block is composed of several causal dilation convolutions used to increase the receptive field of the network and residual structures used to improve the efficiency of gradient feedback. Through the parallel convolution operation, the model overcomes the disadvantages of traditional convolutional networks that cannot handle variable-length sequences, and improves the efficiency of behavioral modeling while preserving the memory effect of power amplifiers. The behavior modeling results of measured data show that compared with the existing Volterra series and recurrent neural network modeling methods, the proposed method can significantly improve the accuracy of behavior modeling. Compared with the recurrent neural network modeling method, the proposed method reduces the implementation time by an order of magnitude in terms of the efficiency of behavior modeling.

Key words: radiation source, power amplifier, behavior modeling, temporal convolutional network, residual structure, causal convolution, dilated convolution, deep network

中图分类号:

TP183

周凡,赵轩,邵杰. 基于深度时序卷积网络的功率放大器行为建模方法[J]. 电子科技, 2022, 35(8): 7-13.

ZHOU Fan,ZHAO Xuan,SHAO Jie. Power Amplifier Behavior Modeling Based on Deep Temporal Convolutional Network[J]. Electronic Science and Technology, 2022, 35(8): 7-13.

图/表 12

图1

图2

图3

图4

图5

图6

图7

表1

图8

表2

图9

表3

参考文献 22

[1]	Cai J, King J, Yu C, et al. Support vector regression-based behavioral modeling technique for RF power transistors[J]. IEEE Microwave and Wireless Components Letters, 2018, 28(5):428-430. doi: 10.1109/LMWC.2018.2819427
[2]	袁晓辉. 记忆非线性功率放大器行为建模及其数字预失真[D]. 成都: 西南交通大学, 2016.
	Yuan Xiaohui. Behavioral modeling and digital pre-distortion of nonlinear power amplifiers with memory effects[D]. Chengdu: Southwest Jiaotong University, 2016.
[3]	Braithwaite R N. Digital predistortion of an RF power amplifier using a reduced volterra series model with a memory polynomial estimator[J]. IEEE Transactions on Microwave Theory and Techniques, 2017, 65(10):3613-3623. doi: 10.1109/TMTT.2017.2729513
[4]	Al-kanan H, Yang X, Li F. Improved estimation for Saleh model and predistortion of power amplifiers using 1-dB compression point[J]. The Journal of Engineering, 2020, 2020(1):13-18. doi: 10.1049/joe.2019.0973
[5]	Mir I, Eisa S A, Maqsood A. Review of dynamic soaring: technical aspects, nonlinear modeling perspectives and future directions[J]. Nonlinear Dynamics, 2018, 94(4):3117-3144. doi: 10.1007/s11071-018-4540-3
[6]	Diamantopoulos N P, Nishi H, Kobayashi W, et al. On the complexity reduction of the second-order Volterra nonlinear equalizer for IM/DD systems[J]. Journal of Lightwave Technology, 2018, 37(4):1214-1224. doi: 10.1109/JLT.2018.2890118
[7]	Braithwaite R N. Digital predistortion of an RF power amplifier using a reduced volterra series model with a memory polynomial estimator[J]. IEEE Transactions on Microwave Theory and Techniques, 2017, 65(10):3613-3623. doi: 10.1109/TMTT.2017.2729513
[8]	Zhu Y P, Lang Z Q. A new convergence analysis for the Volterra series representation of nonlinear systems[J]. Automatica, 2020, 111(3):1-10.
[9]	江明玉, 刘太君, 叶焱, 等. 基于广义改进型RBF网络的射频功放非线性建模[J]. 移动通信, 2018, 42(3):64-69.
	Jiang Mingyu, Liu Taijun, Ye Yan, et al. Nonlinear modeling of RF amplifiers based on generalized improved RBF networks[J]. Mobile Communications, 2018, 42(3):64-69.
[10]	Mišić J, Marković V, Marinković Z, et al. Behavioral modeling of low noise amplifiers for LTE systems based on modified Elman neural networks[C]. Serbia:Proceedings of the Twenty-second Telecommunications Forum Telfor, 2014.
[11]	吴硕. 双频功率放大器行为模型和数字预失真技术的研究[D]. 南京: 东南大学, 2019.
	Wu Shuo. Reaearch on the behavioral models and digital pre-distortion of concurrent dual-band power amplifiers[D]. Nanjing: Southeast University, 2019.
[12]	Kani J N, Elsheikh A H. DR-RNN: A deep residual recurrent neural network for model reduction[D]. New York: Cornell University, 2017.
[13]	Sun J, Shi W, Yang Z, et al. Behavioral modeling and linearization of wideband RF power amplifiers using BiLSTM networks for 5G wireless systems[J]. IEEE Transactions on Vehicular Technology, 2019, 68(11):10348-10356. doi: 10.1109/TVT.2019.2925562
[14]	Gonzalez J, Yu We. Non-linear system modeling using LSTM neural networks[J]. IFAC-Papers on Line, 2018, 51(13):485-489.
[15]	Lopez M, Yu Wen. Nonlinear system modeling using convolutional neural networks[C]. Mexico City:The Fourteenth International Conference on Electrical Engineering, Computing Science and Automatic Control, 2017.
[16]	Li H, Zhang Y, Li G, et al. Vector decomposed long short-term memory model for behavioral modeling and digital predistortion for wideband RF power amplifiers[J]. IEEE Access, 2020(8):63780-63789.
[17]	韦张跃昊, 钱升谊. 基于滤波重构和卷积神经网络的心电信号分类[J]. 电子科技, 2019, 32(11):7-11.
	Wei Zhangyuehao, Qian Shengyi. ECG signal classification based on filter reconstruction and convolutional neural network[J]. Electronic Science and Technology, 2019, 32(11):7-11.
[18]	Mehta S, Rastegari M, Caspi A, et al. Espnet: Efficient spatial pyramid of dilated convolutions for semantic segmentation[C]. Munich:Proceedings of the European Conference on Computer Vision, 2018.
[19]	Oord A, Dieleman S, Zen H, et al. Wavenet: A generative model for raw audio[D]. New York: Cornell University, 2016.
[20]	He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]. Las Vegas:Proceedings of IEEE Conference on Computer Vision and Pattern Recognition, 2016.
[21]	Hara K, Saito D, Shouno H. Analysis of function of rectified linear unit used in deep learning[C]. Killarney:Proceedings of International Joint Conference on Neural Networks, 2015.
[22]	Zhang Z. Improved adam optimizer for deep neural networks[C]. Banff:Proceedings of IEEE/ACM the Twenty-sixth International Symposium on Quality of Service, 2018.

名称	型号或版本
显卡	NVIDIA GeForce GTX 1066
处理器	Intel(R) Core(TM) i7-7700HQ @2.8GHz
内存	16 GB
软件环境	Python 3.7.6
深度学习张量库	Pytorch 1.3.1

建模方式	均方误差
Volterra级数^[7]	9.32
RNN网络^[12]	2.17
LSTM网络^[13]	1.22
本文网络	0.31

网络类型	参数量	单次迭代时间/s
双层RNN网络^[12]	115 969	0.978 60
双层LSTM网络^[13]	463 489	0.921 80
本文网络	110 975	0.061 04