基于随机解调的模拟信息转换技术研究

发布时间：2018-04-22 03:36

本文选题：压缩感知（CS） + 随机解调(RD)　；参考：《哈尔滨工业大学》2014年硕士论文

【摘要】：传统的信号处理模式是先进行奈奎斯特采样，再通过压缩去除大量冗余数据。这种是先采样后压缩的模式无疑浪费了大量的资源。模拟信息转换是解决这个弊端的一种思路，它旨在利用一些新的信息理论将模拟信号直接转变为有用的信息，从而减少采样过程中的冗余数据，降低采样率。压缩感知是实现模拟信息转换的一种重要理论依据，它在理论上表明利用信号的稀疏性可以实现信号的同时压缩和采样，达到将模拟信号直接转变为压缩过的有用信息的目的，即模拟信息转换。但是压缩感知的物理实现方法目前还很少，随机解调是其中一种重要方法和技术。本文对此开展研究，采用随机解调方式使压缩感知理论实用化，从而实现模拟信息转换。本文的主要研究内容和结果如下： 1、研究随机解调的原理。对压缩感知理论和随机解调原理进行阐述，用数学语言描述随机解调过程的各个阶段，说明随机解调是压缩感知从离散域到连续域的一种扩展。通过MATLAB仿真验证随机解调技术的可行性，并探究了采样速率、采样相位、滤波器参数、m序列周期等若干因素的影响作用，为后续随机解调物理系统的设计提供参考。 2、设计随机解调实验平台。以随机解调技术的理论研究及仿真结果为指导，设计随机解调物理系统作为实验平台。该系统包括硬件和软件两个部分。硬件部分设计了信号调理板卡，整合了多种PXIe仪器设备，利用先进的PXIe测试总线进行设备互连，完成信号的产生、混频、滤波、放大和采样任务，以及数据存储和传输任务。采用LabVIEW语言开发软件，实现对各个硬件模块的参数配置和灵活控制，完成信号显示分析、算法执行、报表生成等任务。 3、构造随机解调系统感知矩阵。感知矩阵是信号重构阶段的重要参数，它包含了系统的重要特性，它的准确程度与信号重构效果紧密相关。本文研究了感知矩阵的理论计算法，即根据系统各部分的理论模型、参数、表达式计算系统的感知矩阵的方法。然后提出了两种效果更好的方法：步进频率激励法，基于m序列和FFT的快速构造法。步进正弦激励法是将被测信号用一系列频率步进的正弦和余弦信号代替，利用一系列对应的系统输出信号的采样值构造感知矩阵。基于m序列和FFT的快速构造法首先采用m序列作为激励信号获得随机解调系统中模拟乘法器、低通滤波器和运放三部分电路整体的脉冲响应；然后用获得的脉冲响应、m序列计算观测矩阵；最后对观测矩阵的共轭转置矩阵进行FFT，之后将FFT结果再次共轭转置得到感知矩阵。快速构造法相比前两种方法能够获得准确度和计算效率的同时提高，是本文实验中所采用的方法。 4、利用随机解调系统进行硬件实验。大量实验表明可以利用本文设计随机解调系统可以实现对50kHz以内的多谐波信号的压缩采样，采样率仅为4kS/s，远小于信号的奈奎斯特采样率，即压缩比可达4%，，信噪比可达15dB以上。另外通过硬件实验还探究了重构信号的信噪比与采样相位偏差程度、信号稀疏度的关系。
[Abstract]:The traditional signal processing mode is to carry out Nyquist sampling first and then to remove a large amount of redundant data through compression. This is the mode of pre sampling and compression. No doubt a lot of resources are wasted. Analog information conversion is a way of thinking to solve this problem. It aims to make use of some new information theory to transform the analog signal directly into useful. In order to reduce the redundant data in the sampling process and reduce the sampling rate, compression perception is an important theoretical basis for the realization of analog information conversion. In theory, it shows that using the sparsity of the signal can compress and sample the signal at the same time, and achieve the purpose of transforming the analog signal directly into the compressed useful information, that is, the model. However, there are few physical implementations of compressed sensing. Random demodulation is one of the most important methods and techniques. In this paper, the theory of random demodulation is used to make the compression perception theory practical and thus the analog information conversion is realized. The main research content and results of this paper are as follows:
1, the principle of random demodulation is studied. The theory of compressed sensing and the principle of random demodulation are expounded. The various stages of the random demodulation process are described in mathematical language. It shows that the random demodulation is an extension of the compression perception from the discrete domain to the continuous domain. The feasibility of the random demodulation is verified by MATLAB simulation, and the sampling rate and sampling are explored. The influence of some factors such as phase, filter parameters, m sequence period and so on will provide a reference for the design of subsequent random demodulation physical system.
2, a random demodulation experimental platform is designed. Based on the theoretical research and simulation results of random demodulation technology, a random demodulation physical system is designed as an experimental platform. The system includes two parts of hardware and software. The hardware part designs a signal conditioning board, integrates a variety of PXIe devices, and uses an advanced PXIe test bus to set up the system. Interconnect, complete the signal generation, frequency mixing, filtering, amplification and sampling tasks, and data storage and transmission tasks. Using LabVIEW language development software to realize the parameters configuration and flexible control of each hardware module, complete the signal display analysis, algorithm execution, report generation and other tasks.
3, the perceptual matrix of the stochastic demodulation system is constructed. The perceptual matrix is an important parameter in the phase of signal reconstruction. It contains the important characteristics of the system and its accuracy is closely related to the effect of the signal reconstruction. This paper studies the theoretical calculation method of the perceptual matrix, namely, the perception of the system based on the theoretical models, parameters and expressions of the system parts. Two better methods are then proposed: step frequency excitation method, fast construction method based on m sequence and FFT. Step sine excitation method is the substitution of sinusoidal and cosine signals with a series of frequency step signals, and a series of corresponding system output signals are used to construct the perception matrix. Based on m sequence, the step sine excitation method is used to construct the sensing matrix. The fast construction method of column and FFT first uses the m sequence as the excitation signal to obtain the impulse response of the analog multiplier, the low pass filter and the three part of the amplifier in the random demodulation system; then the observed matrix is calculated by the acquired pulse response and the m sequence; finally, the conjugate transposed matrix of the observation matrix is FFT, and then the FFT result is obtained. It is a method used in this experiment to improve the accuracy and calculation efficiency compared with the first two methods.
4, using the random demodulation system to carry out the hardware experiment. A large number of experiments show that the random demodulation system designed in this paper can realize the compression sampling of the multi harmonic signals within 50kHz. The sampling rate is only 4kS/s, which is far less than the Nyquist sampling rate of the signal, that is, the compression ratio can reach 4% and the signal to noise ratio can reach 15dB. The relationship between the SNR of reconstructed signal and the degree of sampling phase deviation and the degree of signal sparsity is also explored.

【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.3

【参考文献】