低功耗心电处理算法及存储架构研究

发布时间：2018-03-10 12:45

本文选题：极低功耗　切入点：心电信号　出处：《浙江大学》2015年博士论文　论文类型：学位论文

【摘要】：集成电路技术的发展使更多复杂的处理模块能够集成在很小的芯片上,这给基于传感器的可穿戴心电监测装置发展带来了机会。在可穿戴心电平台中,基于通用处理器的心电处理平台能够同时兼顾硬件和软件设计的灵活性而应用广泛,但是功耗是其应用的主要障碍。本文以心电信号处理为目标,以通用处理器平台的低功耗设计为核心,主要研究了包括心电处理算法和处理器存储架构等关键技术,具体的研究内容和创新点包括：1.基于经验模态分解的心电压缩算法研究。通过分析经验模态分解算法的流程和特性,本文首次提出了使用经验模态分解的筛选过程进行压缩的算法。该算法使用第一阶本征模态函数的筛选过程,利用筛选函数的极值均值点能够重构原信号的特点,避免了传统使用极值点拟合本征模态函数具有较大误差的缺点,并结合筛选函数重组、极值均值点优化选取的方法,减少了需要保存的特征点数量,在达到较低重构失真的同时增加了压缩效率。相对于传统经验模态分解的心电压缩方法和其他变换域的方法,该方法能够实现更大的压缩效率和更低的重构失真。在以传输数据为任务的可穿戴心电设备中,该方法能够降低数据传输量,减少无线传输功耗。2.信号特征预判的低功耗QRS检测算法研究。本文通过分析基于小波变换的QRS检测算法,利用绝大部分小波变换系数幅值较小的特点,提出了基于时域信号特征预判的低运算量QRS检测算法。该方法利用简单的时域滤波和预判操作,筛选出信号中疑似QRS波段,并结合不同RR间隔(心跳间隔),对疑似QRS波使用基于单尺度小波分解和多尺度小波分解相结合的QRS检测方法。实验结果表明,该算法在检测准确率和抗噪声能力上都与原基于小波变换的算法相当的情况下,获得了77.9%的运算功耗降低。该算法能够获得算法复杂度和功耗的降低,非常适应可穿戴心电平台的实时处理需求。3.面向心电处理的低功耗内存架构研究。本文通过分析基于小波变换的QRS检测算法,研究了内存空间中局部地址范围内的值空间局部性(内存值之间差值较小),提出了基于值压缩的压缩存储策略,其通过在多个内存地址中使用一个基地址和多个较小宽度偏移值的方法,降低存储空间消耗。并针对内存值动态变化特性,设计了以链表方法管理的后备非压缩空间用于动态的分配和释放非压缩内存空间存储原压缩值。进一步,利用程序内存的顺序访问特性,本文设计了基于内存读写缓存的低功耗访问策略。实验结果显示,相对原内存架构,该内存架构能获得32.5%的内存空间降低,68.1%的内存功耗降低。该架构和方法能够在降低内存容量消耗的基础上,获得访问功耗的降低,非常适用于基于通用处理器的心电平台。本文提出的关键技术对可穿戴心电检测平台,特别是针对极低功耗心电处理的通用处理器平台的设计具有积极的理论研究意义和实际应用价值。
[Abstract]:The development of integrated circuit technology enables more complex processing module can be integrated in a small chip, this sensor based Wearable ECG monitoring device development opportunities. In the Wearable ECG platform, ECG processing platform processor can also both hardware and software design flexibility and application based on widely, but the power consumption is a major obstacle to its application. Based on the ECG signal processing as the goal, with low power consumption design of general processor platform as the core, mainly studies the key technologies including ECG processing algorithm and storage processor architecture, research content and innovation include: 1. research based on empirical mode decomposition of ECG compression through the algorithm. The procedure and characteristics of empirical mode decomposition algorithm, this paper presents for the first time using the screening process of EMD. The compression algorithm Method using the screening process of first order intrinsic mode function, using the screening function of the extreme point average can reconstruct the original signal, avoid the use of the traditional fitting extremum intrinsic mode function has larger errors, and combined with the screening function recombination method of extremum point average optimal selection, reduce the number of feature points to at the same time in the preservation of the low distortion increases the compression efficiency. Compared with the traditional methods of empirical mode decomposition method of ECG compression and other transform domain, this method can achieve higher compression efficiency and lower distortion. In the transmission of data to the task of wearable electric equipment in the heart. The method can reduce the amount of data transmission, reduce the research of low power QRS wireless transmission power.2. feature detection algorithm to predict the signal. In this paper, through the analysis of the QRS detection algorithm based on wavelet transform, the use of the vast Most of the wavelet transform coefficients smaller amplitude of the proposed low complexity QRS detection algorithm based on time domain signal characteristics predict. The method using time-domain filtering and pre operation simple sentence, screening out the signal of suspected QRS band, and a combination of different RR interval (heartbeat interval), suspected of using QRS wave detection method of single QRS wavelet decomposition and wavelet decomposition based on combination. The experimental results show that the algorithm in the detection accuracy and anti noise ability and primordium on wavelet transform algorithm is quite the case, won 77.9% of the operation to reduce the power consumption. The algorithm can reduce the algorithm complexity and power consumption, can be very fit wear heart electric platform real-time processing requirements of.3. oriented research on low power memory architecture ECG processing. In this paper through the analysis of the QRS detection algorithm based on wavelet transform, the local memory space The address range of the spatial locality (memory value difference between the smaller, proposed) value compression storage strategy based on compression, the method uses a base address and a number of smaller width offset value in a plurality of memory address, reduce storage space and memory consumption. According to the dynamic characteristics of value, design a list of management method for the dynamic reserve of non compressed space allocation and release of memory space to store the original uncompressed compression value. Further, the program memory sequential access characteristics, this paper designed to read and write memory cache access strategy based on low power consumption. The experimental results show that compared to the original memory architecture, the memory architecture can be obtained 32.5% memory space is reduced, the power consumption is reduced. The 68.1% memory architecture and method can reduce memory consumption based on access, reduction in power consumption, is very suitable for use in general The key technology of this paper has positive theoretical significance and practical application value for the design of Wearable ECG detection platform, especially for the design of universal processor platform for extremely low power ECG processing.

【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2015
【分类号】：TN911.7;R197.39

【相似文献】