使用基于平滑滤波器的卡尔曼预测器对噪声热传感器进行在线温度估算资源-CSDN文库

需积分: 9 113 浏览量 2021-04-29 23:37:59 上传评论 1 收藏 2.49MB PDF 举报

根据给出的文件内容，我们可以提取以下知识点： 1. 热传感器的应用：在多核处理器的动态热管理（DTM）机制中，嵌入式热传感器被用来收集精细的温度信息，用以监控处理器的实时热行为。DTM机制对于处理器的高效运行至关重要，因为过热可能导致系统不稳定，降低性能或甚至损坏处理器。 2. 热传感器的噪声问题：由于环境不确定性、工艺波动等因素的影响，嵌入式热传感器极易受到噪声的影响。噪声导致传感器读取的温度数据和实际温度之间存在偏差，这严重影响了DTM机制的效率。 3. 卡尔曼滤波技术：本文提出了一种基于平滑滤波器的卡尔曼预测技术，以准确估计噪声传感器读数中的温度。卡尔曼滤波器是一种有效的递归滤波器，能够从一系列包含噪声的测量数据中估计动态系统的状态。其核心在于建立一个数学模型，该模型包括系统的状态转移和观测模型，并在时间上进行迭代，以估计系统下一时刻的状态。 4. 空间相关性利用：在多传感器估计的场景中，文章提出利用不同传感器位置之间的空间相关性，即相邻传感器所测量的温度值之间存在一定的关联性，可以被用来提高估计精度。 5. 多传感器协同校准算法（MSSCA）：为了提高多传感器的同步预测精度，提出了一个名为MSSCA的算法。该算法基于传感器间空间相关性的假设，通过校准不同传感器的数据，可以提升整体的温度估计准确性。 6. 红外成像温度测量技术：为了实时获取高级微设备（AMD）四核处理器的热迹线，提出了一种基于红外成像的温度测量技术。通过该技术可以捕捉到处理器在运行过程中的实时温度分布，进而获得实际的温度数据。 7. 预测性能评估：获得的实际温度数据被用来评估文章提出的预测技术性能。模拟结果表明，所提出的协同校准方案相比假设传感器读数为理想情况，能够降低均方根误差（RMSE）达1.2°C，同时提高信噪比（SNR）达15.8dB，并且在很小的平均运行时间开销下实现了上述性能提升。此外，通过该方法修正的传感器温度读数的平均误报率（FAR）也能降低28.6%。 8. 结论：上述结果清楚地证明，如果采用本文提出的方法进行温度估计，将能够显著提升温度预测的准确度和效率，有助于提高DTM机制的性能。通过这些知识点，我们可以深入了解基于平滑滤波器的卡尔曼预测器在噪声热传感器温度估算中的应用和效能。文章中还提到了在实际设备中应用这一技术的可能性，以及它在提升处理器监控精度方面所展示的潜力。

资源推荐

资源详情

资源评论

sensors

Article

On-Line Temperature Estimation for Noisy Thermal

Sensors Using a Smoothing Filter-Based

Kalman Predictor

Xin Li *

, Xingtao Ou

, Zhi Li, Henglu Wei, Wei Zhou and Zhemin Duan

School of Electronics and Information, Northwestern Polytechnical University, Xi’an 710072, China;

ouxingtao@mail.nwpu.edu.cn (X.O.); zhili@mail.nwpu.edu.cn (Z.L.); hengluw@mail.nwpu.edu.cn (H.W.);

zhouwei@nwpu.edu.cn (W.Z.); zhemind@nwpu.edu.cn (Z.D.)

* Correspondence: xinli@nwpu.edu.cn; Tel.: +86-180-0923-3610

Received: 16 November 2017; Accepted: 26 January 2018; Published: 2 February 2018

Abstract:

Dynamic thermal management (DTM) mechanisms utilize embedded thermal sensors

to collect ﬁne-grained temperature information for monitoring the real-time thermal behavior

of multi-core processors. However, embedded thermal sensors are very susceptible to a variety

of sources of noise, including environmental uncertainty and process variation. This causes the

discrepancies between actual temperatures and those observed by on-chip thermal sensors, which

seriously affect the efﬁciency of DTM. In this paper, a smoothing ﬁlter-based Kalman prediction

technique is proposed to accurately estimate the temperatures from noisy sensor readings. For the

multi-sensor estimation scenario, the spatial correlations among different sensor locations are

exploited. On this basis, a multi-sensor synergistic calibration algorithm (known as MSSCA) is

proposed to improve the simultaneous prediction accuracy of multiple sensors. Moreover, an infrared

imaging-based temperature measurement technique is also proposed to capture the thermal traces of

an advanced micro devices (AMD) quad-core processor in real time. The acquired real temperature

data are used to evaluate our prediction performance. Simulation shows that the proposed synergistic

calibration scheme can reduce the root-mean-square error (RMSE) by 1.2

◦

C and increase the

signal-to-noise ratio (SNR) by 15.8 dB (with a very small average runtime overhead) compared with

assuming the thermal sensor readings to be ideal. Additionally, the average false alarm rate (FAR) of

the corrected sensor temperature readings can be reduced by 28.6%. These results clearly demonstrate

that if our approach is used to perform temperature estimation, the response mechanisms of DTM

can be triggered to adjust the voltages, frequencies, and cooling fan speeds at more appropriate times.

Keywords:

dynamic thermal management (DTM); thermal sensor; noise; Kalman prediction;

synergistic calibration; temperature estimation

1. Introduction

The ﬁeld of integrated circuit technology is entering the nanometer era. However, excessively

increased power density leads to high chip temperature, which can result in thermal runaway.

Elevated die temperature adversely affects the performance of multi-core processor systems, causing

shortened lifetimes, increased cooling costs, and reduced reliability and device speed [

]. Therefore,

reliable and effective thermal monitoring mechanisms are crucial to overcome this challenge. Dynamic

thermal management (DTM) is often employed to continuously track the thermal behavior of

processors during runtime [

]. Typically, on-die thermal sensors are widely deployed in modern

multi-core processors to assist DTM [

]. According to the ﬁne-grained temperature information

collected by embedded thermal sensors, DTM techniques maintain the processor’s temperature within

a preset range by reasonably assigning workload scheduling, and adjusting the voltages, frequencies,

Sensors 2018, 18, 433; doi:10.3390/s18020433 www.mdpi.com/journal/sensors

Sensors 2018, 18, 433 2 of 20

and cooling fan speeds appropriately [

]. In addition, in order to mitigate thermal emergencies on

multi-core chips, only a fraction of cores can be simultaneously powered in the full performance mode,

while other cores (i.e., dark cores) need to be power gated. In this so-called dark silicon problem [

–

]

is important to ensure thermal-safe operation for modern chips, i.e., where the peak temperature does

not exceed the safe-operating temperature, otherwise the response mechanisms of DTM are triggered.

The number of on-die thermal sensors keeps growing in very large scale integration (VLSI) systems

to enable the DTM of chip functionalities [

–

], as shown in Figure 1. The accuracy of on-chip sensor

readings has a great inﬂuence on the effectiveness and reliability of DTM. However, embedded thermal

sensors are inevitably accompanied by noise, including process variation, supply voltage ﬂuctuations,

and cross-coupling etc, which cause the observed temperature readings to deviate from the actual

values. In the worst case, the temperature reading error of un-calibrated thermal sensors used in

IBM25PPC750L processors (International Business Machines Corporation (IBM), Armonk, New York,

United States of America) can be up to 34

◦

C (at an actual temperature of 95

◦

C) [

]. Therefore, blindly

trusting the thermal sensors to be ideal can lead DTM strategies to make inaccurate decisions that

result in false alarms or unnecessary responses.

2006 2007 2008 2009 2010 2011 2012 2013 2014

90 nm

65 nm

45 nm

22 nm

[10]

[11]

[12]

[13]

[15]

[14]

[16]

[17]

[18]

[19]

[20] [21]

Figure 1. Trends in the number of embedded thermal sensors in VLSI systems.

Thermal monitoring and management in VLSI systems have been widely researched in recent

years [

–

]. Nowroz et al. [

] utilized frequency-domain signal representations to devise both static

and runtime thermal monitoring approaches. Unfortunately, this work does not consider the effect of

inaccurate and noisy sensors. Reda et al. [

] proposed a new direction to simultaneously identify

the thermal models and the ﬁne-grain power consumption of a chip from just the measurements of

the thermal sensors and the total power consumption. Although they veriﬁed the accuracy of this

method and demonstrated its resilience to sensor noise, the problem of noise reduction for sensor

measurements was not addressed. Effective temperature calibration can compensate for inaccuracies

in temperature measurement, and help to improve thermal sensing accuracy. As a result, how to

solve the problem of estimating temperatures for on-chip thermal sensors corrupted by noise is a

major challenge.

A number of studies have taken into account the noise issue associated with sensor readings,

such as the statistical methodology [

] and the multi-sensor collaborative calibration algorithm

(MSCCA) [

]. However, these techniques lack the ability for real-time prediction which is required

for proactive DTM techniques [

]. In [

], the authors proposed a scheme to make online

temperature measurements signiﬁcantly more accurate. They constructed an ofﬂine thermal equivalent

resistor–capacitor (RC) model and reduced its complexity by a projection-based model order reduction

method. This model can be used to convert the power dissipation to temperature in the prediction step

of the Kalman ﬁlter. However, the derivation of such an RC model is not trivial due to the complexity

of silicon materials. Unlike the above approach, we apply the polynomial ﬁtting technique to convert

the oscillation frequency of noisy sensors to temperature data and use the smoothing ﬁlter to obtain

Sensors 2018, 18, 433 3 of 20

the prediction information. These two sources of temperature information are then combined in the

Kalman ﬁlter to generate reliable temperature estimations. This direct method reduces the calibration

cost because it eliminates the requirement for estimating the power consumption per functional unit.

Speciﬁcally, the contributions of this work are as follows:

•

The noise characteristics of on-chip thermal sensors based on the ring oscillator structure are

systematically analyzed. On this basis, the polynomial ﬁtting technique is used to establish the

non-linear relationship between sensor temperature and oscillation frequency, which can improve

the measurement accuracy.

•

To tackle the challenge in temperature estimation of noisy thermal sensors, a smoothing

ﬁlter-based Kalman prediction technique is proposed to correct the temperatures of on-die sensors

in real-time.

•

For the multi-sensor estimation scenario, the spatial correlations among different sensor locations

are exploited. On this basis, a multi-sensor synergistic calibration algorithm (called MSSCA) is

proposed to improve the simultaneous prediction accuracy of multiple sensors.

•

Relative to the previous works relied on computer-based thermal simulation scheme, an infrared

imaging-based temperature measurement technique is proposed to provide the accurate

thermal characterizations of an AMD quad-core processor operating on different benchmarks.

The captured real temperature data are used to evaluate our prediction approach.

The remainder of this paper is organized as follows: Section 2 provides the necessary motivation

of this work, and presents the analysis of noisy sensor behavior. Section 3 presents the on-line

temperature estimation technique using a smoothing ﬁlter-based Kalman predictor, and details the

proposed multi-sensor synergistic calibration algorithm (MSSCA) that can improve the simultaneous

prediction accuracy of multiple sensors. Section 4 describes the infrared temperature measurement

setup required for capturing the thermal traces of real processors. The performance of our synergistic

calibration scheme is validated in Section 5. Finally, we summarize the main conclusions of this work

in Section 6.

2. Analysis of Noisy Sensor Behavior

Due to the unpredictable behavior of the chip’s thermal proﬁle, constantly monitoring the

processor’s temperature using embedded thermal sensors is critical to ensuring long-term reliability of

integrated circuit systems. A classical implementation of on-die thermal sensors is the ring oscillator,

which has been known for nearly 30 years [

]. The structure of a typical ring oscillator mainly contains

(an odd number) stages of inverters and a counter, as depicted in Figure 2. Note that the output

frequency of ring oscillator in Figure 2 represents the oscillation frequency.

Output

Counter1 2 N

Frequency

Figure 2. Structure of a typical ring oscillator.

The output frequency depends on the total time-delay of inverters, which is given by the

following equation:

f =

N(t

+ t

LH)

(1)

剩余19页未读，继续阅读

评论收藏

内容反馈

等你下课⊙▽⊙

粉丝: 291
资源: 962

使用基于平滑滤波器的卡尔曼预测器对噪声热传感器进行在线温度估算

基于平滑滤波器的卡尔曼预测器对噪声热传感器进行协同校准

kalman滤波在温度测量中的应用_卡尔曼滤波_温度测量；卡尔曼滤波_

labview卡尔曼滤波

基于戴维南模型_扩展卡尔曼_SOC估算,卡尔曼滤波soc估算模型,matlab源码.zip

卡尔曼滤波使用框架

EKF.zip_卡尔曼滤波

kalman filtering in R(R语言实现卡尔曼滤波）

基于修正协方差扩展卡尔曼滤波法的电动汽车锂电池SOC在线估计.pdf

卡尔曼滤波代码,卡尔曼滤波代码讲解,Python

卡尔曼滤波理论与实践 源代码,卡尔曼滤波算法原理,matlab

卡尔曼滤波算法在永磁同步电机无速度传感器控制中的应用.docx

Klaman.zip_Labview 电池_卡尔曼 soc_卡尔曼滤波 soc_卡尔曼滤波SOC_电池SOC

传统卡尔曼及无迹卡尔曼滤波代码

基于STM32F407扩展卡尔曼软启动器设计和实现.pdf

卡尔曼滤波介绍.zip

kaerman_卡尔曼率滤波代码_

惯性导航扩展卡尔曼滤波（MATLAB）

Kalman卡尔曼滤波介绍 含无人车具体应用 共29页.rar

卡尔曼滤波和高斯结合的车辆跟踪（opencv效果非常好）.zip

卡尔曼滤波（Kalman filtering）.zip

基于Arduino，MPU5050传感器的飞控程序.zip

kalman滤波 中文入门教程

python卡尔曼滤波（笔记）（资源）

基于新型滑模观测器的永磁同步电机无传感器控制策略.pptx

kalman滤波跟踪下落小球

电子功用-平滑电流计算装置、平滑电流计算方法以及电池监视模块

物联网-智慧传输-基于WIFI与多传感器融合的室内定位技术研究.pdf

Android代码-android姿态传感器源代码.zip

LANC卡尔曼滤波口自由.zip

最新资源

卡尔曼滤波理论与实践源代码,卡尔曼滤波算法原理,matlab

Kalman卡尔曼滤波介绍含无人车具体应用共29页.rar

kalman滤波中文入门教程