Abstract— Regular respiratory signals (RRSs) acquired with
physiological sensing systems (e.g., the life-detection radar
system) can be used to locate survivors trapped in debris in
disaster rescue, or predict the breathing motion to allow beam
delivery under free breathing conditions in external beam
radiotherapy. Among the existing analytical models for RRSs,
the harmonic-based random model (HRM) is shown to be the
most accurate, which, however, is found to be subject to consi-
derable error if the RRS has a slowly descending end-of-exhale
(EOE) phase. The defect of the HRM motivates us to construct a
more accurate analytical model for the RRS. In this paper, we
derive a new analytical RRS model from the probability density
function of Rayleigh distribution. We evaluate the derived RRS
model by using it to fit a real-life RRS in the sense of least
squares, and the evaluation result shows that, our presented
model exhibits lower error and fits the slowly descending EOE
phases of the real-life RRS better than the HRM.
I. INTRODUCTION
Human breathing motion could be commonly divided into
one irregular phase and three regular phases including the
inhale (IN) corresponding to lung expansion, the exhale (EX)
corresponding to lung deflation, and the end-of-exhale (EOE)
corresponding to rest after lung deflation [1]. Respiratory
signals can be acquired with physiological sensing systems
used to detect human breathing (e.g., the life-detection radar
system [2]), in which, the respiratory signal generated by the
breathing motion consisting of only regular phases is called
regular respiratory signal (RRS). The acquired RRSs could
be used to locate the trapped survivors in disaster rescue [2-7],
or predict the breathing motion in external beam radiotherapy
to allow beam delivery under free breathing conditions [1,
8-12]. In these applications, an accurate analytical RRS model
would be helpful [13]. For example, based on an accurate
analytical RRS model that incorporates more prior knowledge
about the RRS, we could derive a RRS detection algorithm
with improved detection performance, which could be used to
improve the life-detection radar system employed in disaster
rescue [2, 13, 14].
As is shown in Fig. 1, the real-life RRS commonly
contains rapidly ascending segments corresponding to the IN
Research is supported in part by the National Science Foundation of
China (NO.61379136 and No.31300816), Science and Technology Planning
Project of Guangdong Province (NO. 2013B02150016), and Basic Research
Programs of Shenzhen (NO.JCYJ20130401170306884, NO. CXZZ
20140417113430629 and NO. JCYJ20140417113430655).
Xin Li (e-mail: stillbluelixin@gmail.com; xin.li@siat.ac.cn), and Ye Li
(corresponding author, e-mail: ye.li@siat.ac.cn), are with the Shenzhen
Institutes of Advanced Technology, Chinese Academy of Sciences, Key Lab
for Health Informatics of Chinese Academy of Science.
phase, rapidly descending segments corresponding to the EX
phase, and slowly descending segments corresponding to the
EOE phase. In previous research, several analytical RRS
models have been proposed based on cosine function [5, 9],
the absolute value of cosine function [3], the even power of
cosine function [6, 8, 10], or the power of the absolute value of
cosine function [13
]. Among these models, the harmonic-
based random model (HRM) presented in [13] is shown to be
the most accurate. However, the HRM uses its flat segments to
represent the slowly descending EOE phases (see Fig. 1),
which may yield considerable error. The defect of the HRM
motivates us to construct a new analytical model for RRSs in
this paper.
In Section II, based on the probability density function of
Rayleigh distribution, we derive a new analytical RRS model,
which is referred to Rayleigh RRS (R-RRS) model. In Section
III, we evaluate the R-RRS model by using it to fit a real-life
RRS in the sense of least squares, and the evaluation result
shows that the R-RRS model exhibits lower error and fits the
slowly descending EOE phases of the real-life RRS better than
the HRM. A conclusion is drawn in Section IV.
II. M
ODEL DERIVATION
A. A model for a single period of RRS
In terms of the probability density function of Rayleigh
distribution, we define