节点文献
基于前馈神经网络的调频连续波激光雷达扫频非线性预失真校正方案
Pre-distortion correction method for swept-frequency nonlinearity of frequency-modulated continuous-wave light detection and ranging based on feedforward neural networks
【摘要】 发射源的高度扫频线性是调频连续波激光雷达实现高精确测量的必备条件.针对目前基于电流调制分布式反馈半导体激光器产生的调频连续波信号存在扫频非线性问题,本文提出了基于前馈神经网络的扫频非线性预失真方案.该方案首先通过实验获取分布式反馈半导体激光器在调制电流为锯齿波情形下输出的时频曲线;将锯齿波调制电流作为输入,时频曲线作为输出,基于前馈神经网络获取输入到输出的非线性映射关系;接下来,利用反向传播算法生成能补偿分布式反馈半导体激光器输出非线性的预失真调制电流波形.针对调制电流频率处于1—10 kHz的情形进行实验研究,结果表明,采用基于前馈神经网络的扫频非线性矫正方案后,分布式反馈半导体激光器所产生的调频连续波信号的扫频非线性从之前的10-3量级降低到10-5量级;残差均方根值从之前的百MHz量级降低到十MHz量级.本文提出的扫频非线性预失真校正方案有望为高精度的调频连续波激光雷达系统的扫频信号线性化技术提供新思路.
【Abstract】 To address the frequency sweeping nonlinearity of frequency-modulated continuous-wave signals generated by a current-modulated distributed feedback laser diode, we propose and experimentally demonstrate a predistortion method based on a feedforward neural network. For this method, the beat frequency signals of the distributed feedback laser diode under a sawtooth-waveform current modulation are first experimentally obtained, and then the time-frequency curves of the distributed feedback laser diode output are obtained by performing a Hilbert transform on the beat signals. Subsequently, three-layer feedforward neural networks with 10, 5, and 3 hidden-layer neurons are constructed, respectively. By taking the driving current and the timefrequency curves as the input and output of the feedforward neural network, respectively, the nonlinear mapping relationship between them is established. Finally, a backpropagation algorithm is utilized to obtain the pre-distortion modulation current. Taking this current under the modulation frequency from 1 kHz to 10 kHz to drive the distributed feedback semiconductor laser(DFB-LD), the performance of the generated frequencymodulated continuous-wave(FMCW) signals is analyzed. We use nonlinear regression coefficients and residual root mean square values to characterize the performance. For the modulation frequency set at 4 kHz, the frequency sweeping nonlinearity and the residual root mean square value are reduced from 5.29×10-3 and 281 MHz to 1.77×10-5 and 15.15 MHz, respectively. With the modulation frequency fixed at 6 kHz, the frequency sweeping nonlinearity decreases from 5.58×10-3 to 1.52×10-5 and the residual root mean square declines from 251.98 MHz to 12.17 MHz in the proposed scheme. Across the entire tested frequency range from 1 kHz to 10 kHz, the nonlinearity remains stable at ~10-5 after adopting the pre-distortion scheme, with RMS values consistently below 20 MHz. The proposed method is expected to provide a new scheme for the linearization technology of the sweep signal in high-precision frequency-modulated continuous-wave light detection and ranging systems.
【Key words】 frequency sweeping nonlinearity; feedforward neural network; pre-distortion; backpropagation algorithm;
- 【文献出处】 物理学报 ,Acta Physica Sinica , 编辑部邮箱 ,2025年22期
- 【分类号】TP183;TN958.98
- 【下载频次】258