Wavelet Based Dictionaries for Piecewise Dimensionality Reduction of ECG Signals
Extracted source text ยท arXiv 1811.05517
\documentclass[12pt,journal]{article}
%\documentclass[10pt,twocolumn,journal]{IEEEtran}
\usepackage{graphicx,amssymb,amsmath,amsthm,cite,cases}
\usepackage{color,multirow}
\usepackage{mathrsfs}
\usepackage{setspace}
\usepackage{adjustbox}
\usepackage{array}
\setlength\unitlength{1cm}
\setlength\topmargin{-2cm}
\setlength\oddsidemargin{-0.3in}
\setlength\textwidth{18cm}
\setlength\textheight{25cm}
\def\til{\tilde}
\def\oj{\hat{\operatorname{J}}}
\def\ds{\displaystyle}
\def\Join{\hat{\operatorname{J}}}
\def\rs{\rm{s}}
\def\rcs{\rm{cs}}
\def\Dc{\D^{\rm{C}}}
\def\Ds{\D^{\rm{S}}}
\def\Dw{\D^{\rm{W}}}
\def\vp{\mathbf{p}}
\def\bb{b'}
\def\vf{\mathbf{f}}
\def\vfr{\mathbf{f}^{\rm{r}}}
\def\vfaq{\mathbf{f}_q^{\kq}}
\def\vfq{\mathbf{f}_q}
\def\vd{\mathbf{d}}
\def\vv{\mathbf{v}}
\def\vr{\mathbf{r}}
\def\vW{\mathbf{w}}
\def\vWt{\til{\mathbf{w}}}
\def\vMW{\mathbf{W}}
\def\vA{\mathbf{A}}
\def\vB{\mathbf{b}}
\def\vs{\mathbf{s}}
\def\vst{\mathbf{\til{s}}}
\def\st{\til{s}}
\def\ellt{\til{\ell}}
\def\ellq{{\ell^q}}
\def\ellqs{{\ell^{\qs}}}
\def\elltq{\mathbf{{\ellt}^q}}
\def\vc{\mathbf{c}}
\def\vct{\mathbf{\til{c}}}
\def\vctq{\mathbf{\til{c}^\Delta}}
\def\ctq{\til{c}^\Delta}
\def\ct{\til{c}}
\def\Sti{st_{\rm{ind}}}
\def\Stc{st_{\rm{cf}}}
\def\Sts{st_{\rm{sg}}}
\def\Di{\mathbf{D}}
\def\Ba{\mathbf{B}}
\def\Ar{\mathbf{Ar}}
\def\Hf{\mathbf{Hf}}
\def\CWC{\mathbf{CW4}}
\def\CWL{\mathbf{CW2}}
\def\CDF{\mathbf{CDF97}}
%\def\CDFB{\CDF_{\mathbf{B}}
%\def\CDFD{\CDF_{\mathbf{D}}
\def\CDFA{\mathbf{CDF97}^{\Ar}}
\def\CDFH{\mathbf{CDF97}^{\Hf}}
\def\CDFF{\mathbf{CDF53}}
\def\CDFFH{\mathbf{CDF53}^{\Hf}}
\def\CDFFA{\mathbf{CDF53}^{\Ar}}
\def\Db{\mathbf{Db4}}
\def\Coif{\mathbf{Coif}}
\def\Sym{\mathbf{Sym}}
\def\Shq{\mathbf{Short3}}
\def\ShqA{\mathbf{Short3}^{\Ar}}
\def\ShqH{\mathbf{Short3}^{\Hf}}
\def\Shc{\mathbf{Short4}}
\def\Shl{\mathbf{Short2}}
\def\Nq{N_b}
\newcommand{\la}{\langle}
\newcommand{\ra}{\rangle}
\newcommand{\eps}{\epsilon}
\newcommand{\Span}{{\mbox{\rm{span}}}}
\def\qs{{q^\star}}
\def\kq{k_q}
\def\be{\begin{equation}}
\def\ee{\end{equation}}
\def\ben{\begin{eqnarray}}
\def\een{\end{eqnarray}}
\def\R{\mathbb{R}}
\def\Z{\mathbb{Z}}
\def\N{\mathbb{N}}
\def\V{V}
\def\W{W}
\def\ov{\overline}
\def\F{\mathrm{F}}
\def\snr{\mathrm{snr}}
\def\SNR{\mathrm{SNR}}
\def\SR{\mathrm{SR}}
\def\sr{\mathrm{sr}}
\def\SRD{\mathrm{SR_D}}
\def\SRB{\mathrm{SR_B}}
\def\PRD{\mathrm{PRD}}
\def\CR{\mathrm{CR}}
\def\QS{\mathrm{QS}}
\def\PRDN{\mathrm{PRDN}}
\def\tol{\mathrm{tol}}
\def\std{\mathrm{std}}
\def\prd{{\mathrm{prd}}}
\def\prdo{{\mathrm{prd}_0}}
\def\mprd{\overline{\mathrm{prd}}}
\def\gqm{\SNR_\mathrm{M}}
\def\gqd{\SNR_\mathrm{D}}
\def\lqm{\overline{\snr}_\mathrm{M}}
\def\lqd{\overline{\snr}_\mathrm{D}}
\def\std{\mathrm{std}}
\def\sdm{\std_\mathrm{M}}
\def\sdd{\std_\mathrm{D}}
\def\D{{\cal D}}
\def\P{{\cal P}}
\def\dicv{{\cal V}}%^{[c,d]}}
\def\dicw{{\cal W}}%^{[c,d]}}
\def\bc{\color{blue}}
\def\rc{\color{red}}
\def\gc{\color{green}}
\title{Wavelet Based Dictionaries for Dimensionality
Reduction of ECG Signals}
\author{Laura Rebollo-Neira\\
Mathematics Department
Aston University\\
B3 7ET, Birmingham, UK\\
\vspace{0.1cm}\\
Dana \v{C}ern{\'a} \\
Technical University of Liberec\\
Liberec, Czech Republic}
\begin{document}
\maketitle
\baselineskip = 1.5\baselineskip
\begin{abstract}
Dimensionality reduction of ECG signals is considered
within the framework of sparse representation.
The approach constructs the signal model by selecting
elementary components from a redundant dictionary via
a greedy strategy.
The proposed wavelet dictionaries are built
from the multiresolution scheme, but
translating the prototypes within a shorter step
than that corresponding to the wavelet basis.
The reduced representation of the signal is shown to be
suitable for compression at low level distortion.
In that regard, compression results
are superior to previously reported benchmarks on the
MIT-BIH Arrhythmia data set.
\end{abstract}
\section{Introduction}
Electrocardiography (ECG) is the process of
recording the electrical activity of the heart over
a period of time. The overall goal of the
test is to obtain information about the structure
and function of that organ. Dimensionality
reduction of ECG signals is relevant to
techniques for analysis
and classification of this type of data. Such techniques
are subject of ongoing research \cite{BM14, AAT16,
SM16, LSCC16, TTF16, GC16, LYW16, AQC18,BUG18},
which includes methodologies based on
the theory of
compressive sensing \cite{MKA11,ZJM13,PCBV15,PBR16,PP18}.
Compressive sensing enhances the concept of sparsity
by associating it to a
new framework for digitalization
\cite{Bar07,CW08}.
Producing sparse representation of ECG signals
is the central aim of the present work.
This implies to represent an ECG
record as superposition of as few elementary
components as possible. The same goal was traditionally
accomplished by disregarding the least significant
terms in the wavelet transform of the signal \cite{MD14}.
However, as shown in this work, much less
elementary components are necessary if they are
selected from a redundant
wavelet dictionary rather than from a wavelet basis.
This allows us to project an ECG record onto a
subspace of lower dimensionality, in comparison to the
dimension of the signal, and still reconstruct the
record at low level distortion.
The main contributions of the paper are listed below:
\begin{itemize}
\item A number of wavelets dictionaries, based on
existing families of wavelet bases, are developed.
\item Each dictionary is shown to render much higher
levels of sparsity for representing ECG signals
than the corresponding basis. The
tests are performed on the MIT-BIH Arrhythmia
data set consisting of 48 records each of which of 30 min
length.
\item The metric of local sparsity is shown to be
relevant to the detection of
non-stationary noise or significant distortion in
the patters of an ECG record.
\item A strategy for storing the reduced representation of
an ECG signal is considered. The
resulting file is shown to render higher compression
ratio than previously reported results within the state of
the art for the same data set.
\item MATLAB software for constructing the dictionaries,
as well as for reproducing results,
has been made available on a dedicated website
\cite{webpage}.
\end{itemize}
The paper is organized as follows: Sec.~\ref{SM}
presents all the elements to build the proposed
model for piecewise dimensionality reduction of an
ECG record.
Sec.~\ref{ES} describes a strategy
to encode the reduced representation of the signal.
Sec.~\ref{NT} presents and discusses numerical results.
The conclusions are summarized in Sec.~\ref{Con}.
\section{Piecewise ECG signal model}
\label{SM}
Assuming that an ECG record is given as an array
$\vf$ of dimension $N$, we divide the
signal into $Q$ disjoint segments
$\vfq \in \R^{\Nq}, \,q=1,\ldots, Q$
and construct an approximation $\vfaq$ for each segment
as the $\kq$-term `atomic decomposition'
\be
\label{wmodel}
\vfaq=\sum_{n=1}^{\kq} c_q(n)\vd_{\ell^q_n},
\ee
where the elements $\vd_{\ell^q_n}$, called `atoms', are
chosen from a redundant dictionary $\D=\{\vd_n, \in \R^{\Nq}, \|\vd_n\|=1\}$ through a greedy pursuit
strategy. Here for the selection process we adopt the
Optimized Orthogonal Matching Pursuit (OOMP) method
\cite{RNL02,LRN16} which is stepwise optimal in the sense of
minimizing the residual norm $\|\vf_q -\vfaq\|$ at
each iteration. The algorithm evolves
by selecting the atoms one by one as
described next.
\subsection{The OOMP Method}
Since ECG records are normally superimposed to
a smooth background, we initialize the OOMP
algorithm by assuming that the constant atom $\vd_1$ is
always one of the elements in \eqref{wmodel}.
Hence, for each $q$ we set $\kq=0$,
$\Gamma_q=\emptyset$, $\ell_{1}^q=1$, and
$\vW_1^q=\vB_1^{1,q}=\vd_1$. Accordingly
$\vf_q^1= \vd_1 \la \vf_q , \vd_1 \ra$ (where $\la \cdot, \cdot \ra$ indicates the Euclidean inner product),
and $\vr_q^1= \vf_q - \vf_q^1$.
From the above initialization the OOMP approach for
selecting from $\D$ the
elements $\vd_{\ell^{q}_n},\,n=2,\ldots,\kq$ in
\eqref{wmodel}, and calculating the corresponding
coefficients $c_q(n),\,\,n=1,\ldots,\kq$,
iterates as follows.
\begin{itemize}
\item[1)] Upgrade the set $\Gamma_{q} \leftarrow \Gamma_{q} \cup
\ell_{\kq+1}^q$, increase $\kq\leftarrow \kq +1$, and
select the index of a new atom for the approximation as
\be
\label{oomp}
\ell_{\kq+1}^{q}=\operatorname*{arg\,max}_{\substack{n=1,\ldots,M\\ n\notin \Gamma_{q}}}
\frac{|\la \vd_n,\vr_{q}^{\kq}
\ra|^2}{1 - \sum_{i=1}^{\kq}
|\la \vd_n ,\vWt_i^{q}\ra|^2},
\ee
with $\vWt_i^{q}= \frac{\vW_i^{q}}{\|\vW_i^{q}\|}$.
\item[2)]{\em{Orthogonalization step.}}
Compute the corresponding new vector $\vW_{\kq+1}^{q}$ as
\be
\begin{split}
\label{GS}
\vW_{\kq+1}^{q}= \vd_{\ell_{\kq+1}^{q}} - \sum_{n=1}^{\kq} \frac{\vW_n^q}
{\|\vW_n^q\|^2} \la \vW_n^q, \vd_{\ell_{\kq+1}^{q}}\ra
\end{split}
\ee
including, for numerical accuracy, the
re-orthogonalization step:
\be
\label{RGS}
\vW_{\kq+1}^{q} \leftarrow \vW_{\kq+1}^{q}- \sum_{n=1}^{\kq} \frac{\vW_{n}^q}{\|\vW_n^q\|^2}
\la \vW_{n}^q , \vW_{\kq+1}^{q}\ra.
\ee
\item[3)]
{\em{Biorthogonalization step.}} For $n=1,\ldots,\kq$ upgrade vectors $\vB_n^{\kq,q}$ as
\be
\begin{split}
\label{BW}
\vB_{n}^{{\kq}+1,q}&= \vB_{n}^{{\kq},q} - \vB_{\kq+1}^{{\kq}+1,q}\la \vd_{\ell_{{\kq}+1}^{q}}, \vB_{n}^{\kq+1,q}\ra,\\
%\text{where}\\
\vB_{\kq+1}^{\kq+1,q}&= \frac{\vW_{\kq+1}^q}{\| \vW_{\kq+1}^q\|^2}.
\end{split}
\ee
\item[4)]
Calculate
\ben
\vr_{q}^{\kq+1} &=& \vr_{q}^{\kq} - \la \vW_{\kq+1}^{q}, \vf_{q} \ra \frac{\vW_{\kq+1}^{q}}{\| \vW_{\kq+1}^{q}\|^2}.
%\vf_{q}^{\kq+1} &=& \vf_{q}^{\kq+1} +
%\la \vW_{\kq+1}^{q}, \vf_{q} \ra \frac{\vW_{\kq+1}^{q}}{\| \vW_{\kq+1}^{q}\|^2}.
\een
\item[5)] If for a given
$\rho$ the condition
$\|\vr_{q}^{\kq+1}\| < \rho$ has been met
compute the coefficients
$c_q(n) = \la \vB_n^{\kq+1}, \vf_q \ra,\, n=1,\ldots, \kq+1$
to write the approximation $\vfaq$ of $\vf_q$ as in
\eqref{wmodel}. Otherwise repeat steps 1) - 5).
\end{itemize}
Once all the segments $\vf_q$ have been approximated,
the approximation of the whole signal
is obtained by assembling the approximation of
the segments as
$\vfr=\oj_{q=1}^Q \vfaq$, where $\oj$ is the
concatenation operation.
The complexity of the method, dominated
by the selection of atoms, is O($K \Nq M$) with
$K=\sum_{q=1}^Q \kq.$
A suitable model for dimensionality reduction of the signal
must involve significantly fewer
atoms in their atomic decomposition than the
dimension of the signal. The success in achieving
such a goal depends on the dictionary being used.
Next we discuss wavelet dictionaries
which are fit for the purpose.
%{\em{Note:}} Since for the adopted initialization
%the constant atom $\vd_1$ is forced to be in the signal
%decomposition, there is no need for storing the
%index $\ell_1=1$. Moreover, since the magnitude of the
% concomitant
%coefficients $c_q(1),\,q=1,\ldots,\kq$ are normally
% much larger than
%the magnitude of the other coefficients, these are
%quantized with a parameter $2 \Delta$
%and stored in a single string. After a
%shift of indices the remaining coefficients are stored
%For the reader convenience in Appendix A
% we have described the approximation
%algorithm, as it is implemented in the software supporting
%the paper.
\subsection{Wavelet Based Dictionaries}
\label{WD}
The proposed wavelet dictionaries are inspired on
the construction
of dictionaries for Cardinal Spline Spaces
(CSS) as discussed in previous works
\cite{ARN05, ARN08, RNX10}.
Let us denote $\V_j, j\geq 0$ to the CSS
of order $m$ with simple knots at
the equidistant partition of the interval $[c,d]$
having distance $2^{-j}$ between two
adjacent knots. A basis for $\V_j$ arises
from the restriction of the functions
\be
\label{sca}
\phi_{j,k}(x):= 2^{j/2} \phi(2^j x-k), \, k \in \Z
\ee
to the interval $[c,d]$, which we expressed as
$2^{j/2}\phi(2^j x-k)|_{[c,d]}$.
The function
$\phi(x)\equiv \phi_{0,0}(x)$ is called scaling function.
For a CSS $\phi(x)$
is the cardinal B-spline of order $m$
associated with the uniform simple knot
sequence $0,1,\dots,m$ \cite{Sch86}
\be
\phi(x)=\frac{1}{m!}\sum_{i=0}^m(-1)^i\binom{m}{i}(x-i)^{m-1}_+,
\ee
where $(x-i)^{m-1}_+$ is equal to $(x-i)^{m-1}$ if $x-i>0$ and 0 otherwise.
The complementary wavelet subspace $\W_j$ on $[c,d]$ is
constructed in order to
fulfil
\be
\V_{j+1}=\V_j\oplus \W_j,\ j\in\Z^+,
\ee
where $\oplus$ indicates the direct sum, i.e.
$\V_j \cap \W_j=\{0\}$.
Hence,
\be
\V_{j+1}=\V_0 \oplus \W_0 \oplus \W_1\oplus \cdots \W_j.
\ee
For a given value of $j$ a wavelet basis for
$\W_j$ arises as
\be\label{psi}
\psi_{j,k}(x)=2^{j/2}\psi(2^j x-k)|_{[c,d]},\, k\in \Z
\ee
and the elimination of some redundancy
introduced by the cutting process at the borders of the
interval. Different constructions of
mother wavelets $\psi\equiv \psi_{0,0}$
give rise to different wavelet bases.
Within the multi-resolution framework, dictionaries for
a CSS are simply constructed from the result asserting that
if for $l \in \Z^{+}$
\be
\label{dicw}
\dicw_{j,l}=\{2^{j/2}\psi(2^j x-\frac{k}{2^l})|_{[c,d]},\, k\in \Z\},
\ee
then \cite{ARN08}
\be
\Span\{\dicw_{j,l}\}= V_{{j+l}}.
\ee
This result facilitates a tool for designing
dictionaries of wavelets of different support
spanning the same CSS. A multi-resolution-like
dictionary $\D_{j,l}$ spanning $\V_{{j}}$ is
constructed as
\be
\label{dic}
\D_{j,\ell}=\dicv_{0,l}\cup \dicw_{0,l} \cup \dicw_{1,l} \cup \cdots
\cup \dicw_{j-l,l},
\ee
with
\be
\dicv_{0,l}=
\{\phi(x-\frac{k}{2^l})|_{[c,d]},\, k\in \Z\}.
\ee
\begin{figure}%[!ht]
\begin{center}
\includegraphics[width=8cm]{dict_1_2.eps}
\includegraphics[width=8cm]{dict_025_2.eps}
\caption{Wavelet functions taken from a basis for a CSS
(left graph)
and from a dictionary spanning the same CSS (right graph).}
\label{Fig2}
\end{center}
\end{figure}
The top graph of Fig.~\ref{Fig2} shows
consecutive wavelets in a cubic Chui-Wang4 \cite{CW92}
wavelet basis
and consecutive wavelets in a dictionary
for the same CSS. Notice that
the wavelet
functions at the finest scale in the dictionary
are broader than
those at the finest scale in the basis.
For the sake of comparison we have built dictionaries for the following wavelet families.
\begin{figure}
\centering
%\vspace{-0.9cm}
\includegraphics[scale=0.295]{MW_CW4.eps} \hspace{-0.2cm}
\includegraphics[scale=0.295]{MW_CW2.eps}\hspace{-0.2cm}
%\hspace*{0.1cm}
\includegraphics[scale=0.295]{MW_97.eps} \hspace{-0.2cm}
\includegraphics[scale=0.295]{MW_53.eps} \hspace{-0.2cm}\\
\includegraphics[scale=0.295]{MW_db4.eps} \hspace{-0.2cm}
\includegraphics[scale=0.295]{MW_coif26.eps}\hspace{-0.2cm}
%\includegraphics[scale=0.295]{MW_sh4.eps} \hspace{-0.2cm}
\includegraphics[scale=0.295]{MW_sym4}\hspace{-0.2cm}
\includegraphics[scale=0.295]{MW_sh3}\hspace{-0.2cm}\\
%\includegraphics[scale=0.295]{MW_sh2.eps} \hspace{-0.2cm}
\caption{Wavelet prototypes. From left top to right
bottom the graphs correspond to the families
$\CWC,$ $\CWL,$
$\CDF,$ $\CDFF,$ $\Db,$ $\Coif,$ $\Sym,$
% $\Shq,$$\Shl$,
and $\Shq$.}
\label{MW}
\end{figure}
{\bf Semi-orthogonal spline wavelets} were mentioned above. Their construction was proposed
by C. Chui and J. Wang in \cite{CW92}. Semi-orthogonal
wavelets, also called prewavelets, are not orthogonal if the wavelets are at the same scale level, but orthogonality of wavelets at the different levels is preserved. Here we consider the cubic and
linear cases, which are indicated as
{$\CWC$} and {$\CWL$} respectively.
The wavelet prototypes
corresponding to this family are given the 1st and
2nd graphs in the 1st row Fig.~\ref{MW}.
{\bf Cohen-Daubechies-Feauveau biorthogonal wavelets} were built in \cite{CDF92}. In this case both primal and dual wavelets have compact support. Wavelets from this family
with 9 and 7 taps filters ({\bf{CDF97}}) and with 5 and 3 taps filters ({\bf{CDF53}}) are used in the JPEG2000
compression standard. The corresponding
prototypes are given in the 3rd and 4th graphs in the
1st row of Fig.~\ref{MW}.
{\bf Daubechies wavelets} were built by I. Daubechies as first orthonormal compactly supported continuous wavelets in \cite{Daub88}. They have the shortest possible support for the given number of vanishing moments and they have extremal phase. Here we consider the four vanishing moments and indicate it as {\bf Db4}. This wavelet prototype is displayed in the
1st
graph of the 2nd row of Fig.~\ref{MW}.
{\bf Coiflets} were suggested by R. Coifman as orthonormal wavelet systems with
vanishing moments for both scaling functions and wavelets. They were first constructed
by I. Daubechies in \cite{Daub93}. We consider two
vanishing moments used exact values of scaling and wavelet parameters from \cite{CFK08} and
indicate the family as {\bf{Coif}}.
The mother wavelet is displayed in the
2nd graph in the
2nd row of Fig.~\ref{MW}.
{\bf Symlets} are modified version of Daubechies wavelets. They have also the shortest possible support
for the given number of vanishing moments,
see \cite{Daub93}.
% but they are less asymmetric and have almost linear phase, see \cite{Daub93}.
We consider four vanishing moments. The prototype wavelet for the family {\bf Sym} is
given in the 3rd graph in the
2nd row of Fig.~\ref{MW}.
{\bf Spline wavelets with short support} are obtained if the local support of dual wavelets is not required. They were designed in \cite{Chen95, Han06}. Here we consider
quadratic splines and indicate them
as $\Shq$. The prototype for this family is
plotted in last graph of the 2nd row of Fig.~\ref{MW}.
In the discrete case dictionaries can be
constructed in a very simple manner:
A redundant dictionary for
the Euclidean space $\R^N$ is simply any set of normalized
vectors $\{ \vd_i \in \R^N,\,\|\vd_i\|=1\}_{i=1}^M$
such that
$N$ of them are linearly independent and $M>N$.
Inspired by the results for CSS we
construct a dictionary for $\R^N$ simply
by discretization of the functions in the wavelet
dictionary. As already mentioned, ECG signals are commonly
superimposed to a smooth background,
hence we enrich a wavelet
dictionary adding a few low frequency
components from a discrete cosine basis. Thus
the whole dictionary we consider is
built as $\D=\Dc \cup \Dw$, with $\Dw$ a discrete wavelet
basis (or dictionary) for $\R^{\Nq}$ and
\ben
\Dc&=&\{w_c(n)
\cos{\frac{{\pi(2i-1)(n-1)}}{2M}},i=1,\ldots,\Nq\}_{n=1}^{M},\nonumber
\een
where each $w_c(n)$ is a normalization constant.
For the numerical tests we fix $M=10$ for $\Dc$ and for
$\Dw$ we consider different wavelet dictionaries,
all obtained from the aboved mentioned wavelet bases.
\section{Encoding the signal model}
\label{ES}
For the reduced representation of the signal
\eqref{wmodel}
to be suitable for storage and transmission purposes
it is necessary to encode it within a small file
in relation to the size of the original record.
For this end, firstly the
absolute value coefficients
$|c_{q}(n)|,\,n=1\ldots,\kq,\,q=1,\ldots,Q,
$ are
converted into integers as follows:
\be
\label{uniq}
c^\Delta_{q}(n)= \lfloor \frac{|c_{q}(n)|}{\Delta} +\frac{1}{2} \rfloor,
\ee
where $\lfloor x \rfloor$ indicates the largest
integer number
smaller or equal to $x$ and $\Delta$ is the quantization
parameter.
The signs of the coefficients, represented as
$\vs_{q},\,q=1,\ldots,Q$,
are encoded separately using a binary alphabet.
For creating the strings of numbers to encode the
the signal model we proceed as in \cite{RNS17,RN18}.
The indices of the atoms in the
atomic decompositions of each block $\vf_q$
are first sorted in ascending
order $\ell_{i}^q \rightarrow \til{\ell}_i^q,\,i=1,\ldots,\kq$,
which guarantees that, for each $q$ value,
$\til{\ell}_i^q < \til{\ell}_{i+1}^q,\,i=1,\ldots,\kq-1$.
The order of the indices induces
order in the unsigned coefficients,
$\vc^\Delta_{q} \rightarrow \vct^\Delta_{q}$ and
in the corresponding signs $\vs_{q} \rightarrow
\vst_{q}$. The ordered indices are
stored as smaller positive numbers by taking differences
between two consecutive values.
By defining $\delta^q_i=\ellt^q_i- \ellt^q_{i-1},\,i=2,\ldots,\kq$ the follow string stores the indices for block
$q$ with unique recovery
${\ellt^q_1, \delta^q_2,
\ldots, \delta^q_{\kq}}$.
The number `0' is then used to separate the string
corresponding to different blocks and entropy code
a long string, $\Sti$, which is built as
\be
\label{sti}
\Sti=[{\ellt_1}^1,\ldots,\delta^1_{k_1},0,
{\ellt_1}^2,\ldots,\delta^2_{k_2},0, \cdots,
{\ellt_1}^{k_Q},\ldots, \delta^Q_{k_Q}].
\ee
The corresponding quantized magnitude of the coefficients
are concatenated in the strings $\Stc$
as follows:
\be
\label{stc}
\Stc
= [\ctq_{1}(1), \ldots, \ctq_{1}(k_1), \cdots,
\ctq_{{k_Q}}(1), \ldots, \ctq_{{k_Q}}(k_Q)].
\ee
Using `0' to store a positive sign and `1' to store
negative one, the signs are placed
in the string, $\Sts$ as
\be
\label{sts}
\Sts=[\st_{1}(1), \ldots, \st_{1}(k_1), \cdots,
\st_{{k_Q}}(1), \ldots, \st_{{k_Q}}(k_Q)].
\ee
\section{Results}
\label{NT}
The numerical tests we present here use the
full MIT-BIH Arrhythmia database \cite{MITDB}
which contains
48 ECG records. Each of these records is of
30 min length, consisting of
$N=650000$ 11-bit samples at a frequency of 360 Hz.
All our results have been
obtained in the MATLAB environment
using a notebook 2.9GHz dual core i7 3520M CPU and 4GB
of memory.
\subsection{Assessment Metrics}
The quality of the signal approximation is assessed
with respect to the $\PRD$ calculated as follows,
\be
\PRD=\frac{\|\vf - \vfr\|}{\|\vf\|} \times 100 \%\quad
\ee
where, $\vf$ is the original signal and $\vfr$ is
the signal reconstructed from the approximated segments.
Since the $\PRD$ strongly depends on the background of
the signal, the $\PRDN$ as defined below
is also a relevant metric.
\be
\label{PRDN}
\PRDN=\frac{\|\vf - \vfr\|}{\|\vf - \ov{\vf}\|} \times 100 \%,\quad
\ee
where, $\ov{\vf}$ indicates the mean value of $\vf$.
For a fixed value of $\PRD$ the sparsity of
a representation is assessed by the sparsity ratio
(SR)
\be
\label{SR}
\text{SR}=\frac{N}{K},
\ee
where $N$ is the total length of the signal
and $K=\sum_{q=1}^Q \kq,$ with $\kq$ the number of atoms in the
atomic decomposition \eqref{wmodel} of each segment
of length $\Nq$.
For detection of nonstationary noise in patterns it is useful to
define the measure of local sparsity $\sr(q)$ as follows:
\be
\label{sr}
\sr(q)=\frac{\Nq}{\kq},\quad q=1,\ldots Q.
\ee
The compression performance depends of the
size of the file storing the signal representation
and is assessed by the Compression Ratio (CR) as given by
\be
\text{CR}=\frac{{\text{Size of the uncompressed file}}}{{
\text{Size of the compressed file}}}.
\ee
In the numerical tests below we use the notation $\CR$ to
denote the compression ratio obtained when
the string $\Sti, \Stc$ and $\Sts$ are stored
in hierarchical data format \cite{HDF5},
simply using the MATLAB instruction
{\tt{save}}. If a
Huffman coding step is included the corresponding
compression ratio is denoted as $\CR^{\bf{Hf}}$.
The quality score ($\QS$),
reflecting the tradeoff between
compression performance and reconstruction quality,
is the ratio:
\be
\QS=\frac{\CR}{\PRD}.
\ee
Since the $\PRD$ is a global quantity, in order to
detect possible local changes in the visual quality
of the recovered signal,
we consider the local $\PRD$ as follows.
Each signal is partitioned in $Q$ segments
$\vf_q,\, q=1\ldots,Q$ of $\Nq$ samples.
The local $\PRD$ with respect to every segment
in the partition is indicate
as $\prd(q),\,q=1,\ldots Q$ and calculated as
\be
\label{prd}
\prd(q)=\frac{\|\vf_q - \vfr_q\|}{\|\vf_q\|} \times 100 \%,\quad
\ee
where $\vfr_q$ is the recovered portion
of the signal corresponding to the segment $q$.
In all the numerical test of the next section the
OOMP approximation is stopped through a fixed value
$\rho$ so as to achieve the
same value of $\prd$ for all the segments in the
records. Assuming that the target $\prd$ before
quantization is
$\prdo$ we set $\rho = \prdo\|\vf_q\|/100$.
\subsection{Test I}
\label{NT1}
The purpose of the first
numerical tests is to demonstrate the
significant gain in dimensionality reduction
(high values of $\SR$)
achieved if the component $\Dw$ of the dictionary $\D$ is a
wavelet dictionary instead of a wavelet
basis. The wavelet basis arises using as
translation parameter $b=1$ and the wavelet
dictionary using a translation parameter $b=1/4$.
A wavelet basis contains scales corresponding
to the values $j=3,\ldots,8$
while the corresponding dictionary contains one less
scale i.e. $j=3,\ldots,7$. The exact redundancy of
the dictionary depends of the wavelet family but
in all the cases is close to two. The records are
approximated by partitioning the signal into $Q=1300$
segments of $\Nq=500$ data points each.
The second column in Table~\ref{TABLE1} gives the mean
value $\SR$, after quantization, with respect to the
48 records in the data set. The standard deviation is
given in the 3rd column.
The approximation
with all the families is
accomplished to obtain the same mean value of
$\PRD$ ($\ov{\PRD}=53$). For this test the
quantization parameter is fixed. For all the records and
all the dictionaries $\Delta=35$.
The 4th column gives the mean value $\CR$ and the
fifth column the corresponding standard deviation.
For further improvement in
$\CR$ we have added an entropy coding step to represent the
strings described in Sec.~\ref{ES} at a fixed bit depth
using Huffman coding.
The implementation of the step is realized using the off-the-shelf MATLAB function Huff06 available on \cite{Karl}.
The corresponding outcomes, indicated as $\ov{\CR}^{\Hf}$,
are given in the sixth column of Table ~\ref{TABLE1}.
%$\CDFA_{\Ba}$, $\CDFA_{\Di}$, $\CDFFA_{\Ba}$, $\CDFFA_{\Di}$, $\ShqA_{\Ba}$,
%$\ShqA_{\Di}$, and
%$\CDFH_{\Ba}$, $\CDFH_{\Di}$, $\CDFFH_{\Ba}$, $\CDFFH_{\Di}$, $\ShqH_{\Ba}$, and
%$\ShqH_{\Di}$. `
The 1st column lists the different wavelet families which are considered to build the dictionary $\Dw$. The
subscript B is used to indicate the wavelet basis and
the subscript D to indicate the
corresponding dictionary, i.e. $\CDF_{\Ba}$ indicates the
9/7 Cohen-Daubechies-Feauveau biorthogonal wavelet basis
and $\CDF_{\Di}$ the corresponding wavelet dictionary.
For further of comparison the last row of the
table gives the results produced when a
Fast Wavelet Transform ({\bf{FWT}}) is applied on the
whole record, as proposed in \cite{LRN19}.
\begin{table}
\caption{Mean values of $\SR$ and $\CR$
with respect to the 48 records in the
MIT-BIH Arrhythmia database. In all the cases the
reconstructed signals produce $\ov{\PRD}=53$.}
\label{TABLE1}
\begin{center}
\setlength{\extrarowheight}{0.1cm}
\begin{tabular}{||l||r|r|r|r|r|r|||}
\hline \hline
$\Dw$ &${\ov{\SR}}$ & $\std$ &${\ov{\CR}}$&$\std$ &
$\ov{\CR}^{\Hf}$&$\std$\\ \hline \hline
$\CWC_{\Ba}$& 10.13 & 2.42 & 11.72 & 3.14 &16.70&3.78\\
$\CWC_{\Di}$&16.62 & 4.13 & 13.31 & 3.51 &17.80&4.40\\ \hline
$\CWL_{\Ba}$& 11.34 & 2.91 & 12.08 & 3.48&17.78&4.34\\
$\CWL_{\Di}$&16.76 & 4.95 & 13.91 & 4.04 &18.37 &5.25\\ \hline
$\CDF_{\Ba}$& 14.59 & 3.90 & 14.90& 4.20 & 20.54 & 4.99\\
$\CDF_{\Di}$&{\bf{20.60}} & 5.79 & {\bf{16.13}}& 4.57 & {\bf{21.26}} & 5.82\\ \hline
$\CDFF_{\Ba}$& 14.72 & 4.25 & 14.84 & 4.29 & 20.46 & 5.31 \\
$\CDFF_{\Di}$&{\bf{20.18}} &6.36 & {\bf{16.02}} & 4.88 & {\bf{21.04}} & 6.27\\ \hline
$\Db_{\Ba}$& 12.81 & 3.28 & 13.49 &3.63 & 18.81 & 4.46\\
$\Db_{\Di}$&17.88 & 4.94 & 14.24 &4.05 & 18.97 & 5.15 \\ \hline
$\Coif_{\Ba}$& 11.50 & 2.99 &12.66 &3.47 &17.41 & 4.29 \\
$\Coif_{\Di}$&15.27 & 4.57 & 12.58 &3.68 &16.73 & 4.74\\ \hline
$\Sym_{\Ba}$&12.83 & 3.30 & 13.65 &3.68 & 18.96 & 4.47\\
$\Sym_{\Di}$&18.51 & 5.26 & 14.70 &4.22 & 19.26 & 5.77\\ \hline
%$\Shc_B$& 7.3 & 1.9 & 7.7 & 2.2 \\
%$\Shc_D$& & & & \\ \hline
$\Shq_{\Ba}$&12.78 & 3.18 & 15.09 & 4.31 & 19.68 & 4.59 \\
$\Shq_{\Di}$&{\bf{19.84}} & 5.25 & {\bf{16.00}} & 4.24 & {\bf{20.95}} & 5.46 \\ \hline
{\bf{FWT}}\cite{LRN19}& 11.22 & 1.75 & {\bf{23.17}} & 6.57 & {\bf{26.34}} & 6.99 \\ \hline \hline
%$\Shl_B$& 6.2 & 1.6 & 6.8 & 1.7 \\
%$\Shl_D$& & & & \\ \hline
\end{tabular}
\end{center}
\end{table}
As seen in Table~\ref{TABLE1} the best results in relation
to dimensionality reduction (largest values of ${\ov{\SR}}$)
%and compression (largest values of ${\ov{\CR}}$)
are for $\CDF_{\Di}, \CDFF_{\Di}$ and $\Shq_{\Di}$
and all the wavelet bases
perform poorly in comparison to the dictionaries.
The difference in ${\ov{\CR}}$ is not so pronounced though.
This is because, in spite of the fact that the approximation
using a dictionary
involves much less elementary components in the signal
decomposition, the dispersion of
indices corresponding to the atoms in the decomposition
is larger if using a dictionary. Notice that, as far as
compression is concerned using a FWT to transform the
whole signal is more effective than a piecewise
transformation. This is true not only
in terms of the achieved ${\ov{\CR}}$ but also
in regard to computational time (the approach
advanced in \cite{LRN19} is very fast). Contrarily, as
indicated by the values of ${\ov{\SR}}$ in Table~\ref{TABLE1} the method in \cite{LRN19} is not as effective
for dimensionality reduction as it is for compression.
However, dimensionality reduction is relevant in its
own right to methodologies for classification and
recognition tasks\cite{WSN13,RR18}.
\input{table2.tex}
For further information in Table~\ref{TABLE2}
we provide
the figures of the evaluation metrics for each of
the 48 records in the dataset.
The 2nd column of Table~\ref{TABLE2} shows
the values of $\PRD$ for the records listed in
the 1st column. The 3rd column shows the values of
$\SR$ after quantization.
The $\CR^{\Hf}$ is given in the
4th column and the corresponding $\QS$ in the 5th
column. The last column corresponds to the values
of $\PRDN$.
All the results are obtained
using the dictionary $\CDF_{\Di}$. The average time
for producing Table~\ref{TABLE2} is
30 s per record.
Fig.~\ref{Segs} depicts the approximation and
raw data (indistinguishable in the scale of the graphs)
of 2000 points in the records 107 and 116.
\begin{figure}[h]
\begin{center}
\includegraphics[width=8.75cm, height=5cm]{Seg107.eps}
\includegraphics[width=8.75cm, height=5cm]{Seg116.eps}
%\includegraphics[width=9cm, height=5cm]{Seg232.eps}
\end{center}
\caption{The waveforms in the graphs
are the raw data and the approximations corresponding
to 2000 points in the records 107 (left) and 116 (right).
The bottom lines represent the absolute value of
the difference between the data and the approximation.
}
\label{Segs}
\end{figure}
\subsection{Test II}
\label{NT2}
In this test we compare compression results with
previously reported benchmarks on the MIT-BIH Arrhythmia
database.
Since the
PRD strongly depends on the
baseline in the ECG record, when using the
relation CR vs PRD
for comparison one needs to be able to
confirm that the values of
PRD reported in previous work either, do not include
any subtraction of baseline, or the subtracted value
is also reported. From the
information given in the papers we are
comparing with (the relation between
the values of PRD and PRDN) we can guarantee
that the results correspond to the same database \cite{MITDB}
and without any subtraction of baseline.
In the range $[0.31 \; 0.67]$ the comparison is
realized with the classical benchmark \cite{LKL11}.
This
method is based on peak detection to perform
the Discrete Cosine Transform (DCT) between
consecutive peaks and Huffman encoding for
lossless compression.
In the range $[0.80 \;1.71]$
the comparison is with the methods in \cite{MZD15}
and \cite{TZW18}. Both these methods are based on Adaptive Fourier Decomposition (AFD) and
overperform those in \cite{LKL11} and \cite{PSS16} for
values of $\PRD$ in the range being considered.
In \cite{TZW18} the lossless compression
is realized using Huffman coding while
\cite{MZD15} applies symbol substitution.
Comparisons are also produced with the results arising
when a FWT is applied on the full record \cite{LRN19}.
Table~\ref{TABLE3} compares CRs obtain with
$\CDF_{\Di}$ and
$\CDFF_{\Di}$
for values of $\PRD$ in the range $[0.31 \; 0.81]$,
Table~\ref{TABLE4} for values in the range $[0.91 \; 1.06]$,
and Table~\ref{TABLE5} in the range $[0.14 \; 1.71]$.
The last two rows in all the tables give the values of the
quantization parameter, $\Delta$, and the approximation
parameter $\prdo$.
These parameters are set in order to reproduce
the same values of
$\PRD$ as in the publications we are comparing with.
\input{table3.tex}
\input{table4.tex}
\input{table5.tex}
\subsection{Discussion}
From the results shown in Tables~\ref{TABLE4} -
\ref{TABLE5} we can assert that the dictionary-based approach presented in this paper, which is powerful for dimensionality reduction at low level distortion, is also relevant to compression. In that sense the
adopted strategy produces results which significantly
improve upon the state of the art for the full
MIT-BIH Arrhythmia data set \cite{LKL11,MZD15,TZW18,PSS16}. In
regard to CR the results are close to those recently
reported in
\cite{LRN19}. In that publication the
compression results are obtained with an equivalent
strategy as that described in Sec.~\ref{ES}, but applying
a FWT on the full signal instead of using a piecewise
model as proposed here. With respect to
processing time the approach in \cite{LRN19} is far more
efficient though. Indeed, with the same equipment
the construction of Table~\ref{TABLE2}
in this paper takes 30 s per record. The equivalent
table with the approach of \cite{LRN19} takes 0.14 s per
record. However, the central aim
of the present work is to accomplish dimensionality
reduction. In that sense the dictionary approach is
much more effective. As shown in the second column of
Table~\ref{TABLE1}, to achieve
$\ov{\PRD}=53$ the $\CDF_{\Di}$ method uses
$84\%$ less components than the {\bf{FWT}} one.
Moreover, the dictionary approach yields a piecewise
model which can be useful for analysis.
It is interesting to note, for instance, that abrupt
variations
of the local sparsity ratio (c.f. \eqref{sr}) gives
information about the presence of
nonstationary noise or significant
distortions in patters.
The left graph in Fig.~\ref{Fig3} plots the
values $1/\sr(q), q=1,\ldots,1300$ for record 117.
The middle top graph shows the piece of signal
where the peak of $1/\sr$ takes place and
the right graph shows a piece of signal, of
the same length, outside that time interval.
\begin{figure}%[!ht]
\begin{center}
\includegraphics[width=5cm]{sr.eps}
\includegraphics[width=5cm]{frame_p.eps}
\includegraphics[width=5cm]{frame_n.eps}
\caption{The left graph shows the
inverse local sparsity values for each segment in the signal
partition of record 117. The middle graph is the
piece of signal where the peak of $1/\sr$ takes
place. The right graph is a piece of signal of the same
length but at different time interval.}
\label{Fig3}
\end{center}
\end{figure}
\section{Conclusions}
\label{Con}
A method for dimensionality reduction of ECG
signals has been proposed.
The approach was designed within the framework of
sparse representation. A number of
wavelet based dictionaries have been introduced to
undertake the signal model via the OOMP greedy
strategy. The resulting representation
was shown to require significantly
less elementary component than those needed when using a
wavelet basis. The method was proven to
be useful for ECG compression at low level distortion.
The compression results largely surpass
benchmarks within the state of the art for
the MIT-BIH Arrhythmia database.
The MATLAB software for constructing the dictionaries and
reproducing the results in the paper has been made available
on a dedicated website. Although the numerical construction
of the signal model is computationally intensive it
can be effectively realized, even
in the MATLAB environment, where it typically takes
30 s to compress a 30 min record and 1 s to
recover the signal from the compressed file.
Moreover, since the
approximation is carried out independently in every
segment of the signal partition, there is room for
straightforward parallelization using multiprocessors.
The sensitivity of local sparsity to nonstationary
noise or significant distortion of patterns leads to conclude that
the model might be useful to support the development of
techniques for analysis of ECG records.
\section*{Acknowledgments}
We are indebted
to Karl Skretting for making available the
MATLAB functions
{\tt{Huff06}}
which has been used for adding entropy coding to
the proposed compression scheme.
\begin{thebibliography}{99}
%\bibitem{PM12}
%S. Pal, M. Mitra, ``Increasing the accuracy of ECG based biometric analysis by
% data modelling'', {\em{Measurement,}} {\bf{45}} 1927--1932 (2012).
\bibitem{BM14}
S. Banerjee, M. Mitra, ``Application of Cross Wavelet Transform for ECG pattern analysis and classification'',
{\em{IEEE Transactions on Instrumentation and Measurement,}}
{\bf{63}} 326--333 (2014).
\bibitem{AAT16} R.G. Afkhami, G. Azarnia, M.A. Tinati,
``Cardiac arrhythmia classification using
statistical and mixture modeling features of ECG signals, {\em{Pattern Recognition Letters,}} {\bf{70}}, 45-- 51
(2016).
\bibitem{SM16} S. Shadmand, B. Mashoufi, ``A new personalized ECG signal classification
algorithm using block-based neural network and particle swarm optimization, {\em{Biomedical Signal Processing and Control}}, {\bf{25}},
12--23 (2016).
\bibitem{LSCC16} E.J.D. Luz, W.R. Schwartz, G. Camara-Chavez, D. Menotti, ECG-based
heartbeat classification for arrhythmia detection: a survey, {\em{Computer Methods and Programs in Biomedicine,}}
{\bf{127}} 144--164 (2016).
\bibitem{TTF16}
D. P. Tob\'on V., T. H. Falk, M. Maier, ``MS-QI: A Modulation Spectrum-Based ECG Quality Index for Telehealth Applications'', {\em{IEEE Transactions on Biomedical Engineering}},
{\bf{63}}, 1613-- 1622 (2016).
\bibitem{GC16}S. Gutta, Q. Cheng, ``Joint feature extraction and classifier design for ECG-based biometric recognition'', {\em{Journal of Biomedical and Health Informatics}},
{\bf{20}}, 460--468 (2016).
\bibitem{LYW16}
H. Li, D. Yuan, Y. Wang, D. Cui, and L. Cao,
``Arrhythmia Classification Based on Multi-Domain Feature Extraction for an ECG Recognition System'',
{\em{Sensors}}, {\bf{16}}, doi:10.3390/s16101744 (2015).
\bibitem{AQC18}
M. Abdelazez, P. X. Quesnel, A. D. C. Chan, H. Yang,
``Signal Quality Analysis of Ambulatory Electrocardiograms to Gate False Myocardial Ischemia Alarms'', {\em{IEEE Transactions on Biomedical Engineering}},
{\bf{64}}, 1318 -- 1325 (2018).
\bibitem{BUG18}
S. K. Berkaya, A. K. Uysal, E. S. Gunal,
S. Ergin., S. Gunal, M. B. Gulmezoglu,
``A survey on ECG analysis'', {\em{Biomedical Signal Processing and Control}}, {\bf{43}}, 216 -- 235 (2018).
\bibitem{MKA11} H. Mamaghanian, N. Khaled,
D. Atienza and P. Vandergheynst, ``Compressed Sensing for Real-Time Energy-Efficient ECG Compression on Wireless Body Sensor Nodes'', {\em{IEEE Transactions on Biomedical Engineering}}, {\bf{58}}, 2456 -- 2466 (2011).
\bibitem{ZJM13} Z. Zhang, T-P Jung, S. Makeig and B. D. Rao,
``Compressed Sensing for Energy-Efficient Wireless
Telemonitoring of Noninvasive Fetal ECG via
Block Sparse Bayesian Learning'', {\em{
IEEE Transactions on Biomedical Engineering,}}
{\bf{60}}, 300--309 (2013).
\bibitem{PCBV15} L. F. Polan{\'i}a, R. E. Carrillo, Manuel Blanco-Velasco and K. E. Barner, ``Exploiting Prior Knowledge
in Compressed Sensing Wireless ECG Systems'',
{\em{IEEE Journal of Biomedical and Health Informatics}},
{\bf{19,}} 508--519 (2015).
\bibitem{PBR16}
G. Da Poian, R. Bernardini, R. Rinaldo,
`RSeparation and Analysis of Fetal-ECG Signals From Compressed Sensed Abdominal ECG Recordings'', {\em{ IEEE Transactions on Biomedical Engineering}}, {\bf{63}}, 1269 -- 1279 (2016).
\bibitem{PP18} L. F. Polan{\'i}a and R. I. Plaza,
``Compressed Sensing ECG using Restricted Boltzmann Machines'', {\em{Biomedical Signal Processing and Control}},
{\bf{45}}, 237--45 (2018).
\bibitem{Bar07}R. Baraniuk, ``Compressive sensing'',
{\em{IEEE Signal Processing Magazine}}, {\bf{24}}, 118--121, (2007).
\bibitem{CW08}E.~Cand\`{e}s and M. Wakin,
``An introduction to compressive sampling'',
{\em{IEEE Signal Processing Magazine}}, {\bf{25}}, 21 -- 30 (2008).
\bibitem{MD14} M. S Manikandan and S. Dandapat,
``Wavelet-based electrocardiogram signal compression methods and their performances: A prospective review'',
{\em{Biomedical Signal Processing and Control}},
{\bf{14,}} 73--107 (2014).
\bibitem{webpage}{\tt{http://www.nonlinear-approx.info/examples/node011.html}} (Last access May 2018).
\bibitem{RNL02} L.~Rebollo-Neira and D.~Lowe,
``Optimized orthogonal matching pursuit approach'',
{\em{IEEE Signal Process. Letters}}, {\bf{9}},
137--140 (2002).
\bibitem{LRN16} L.~Rebollo-Neira,
``Cooperative greedy pursuit strategies for sparse
signal representation by partitioning'',
{\em{Signal Processing,}} {\bf{125}}, 365--375 (2016).
\bibitem{ARN05} M. Andrle and
L. Rebollo-Neira, ``Cardinal B-spline dictionaries on a compact interval'', {\em{Appl. Comput. Harmon. Anal.}}
, {\bf{18}}, 336-346 (2005).
\bibitem{ARN08} M.~Andrle and L.~Rebollo-Neira, ``From cardinal spline wavelet bases
to highly coherent dictionaries'', {\em{Journal of Physics A}} {\bf{41}} (2008) 172001.
\bibitem{RNX10} L.~Rebollo-Neira and Z. Xu,
``Adaptive non-uniform B-spline dictionaries on a compact interval'', {\em{Signal Processing,}} {\bf{90,}} 2308--2313 (2010).
\bibitem{Sch86}
L. L. Schumaker, Spline Functions: Basic Theory,
Wiley, New-York, 1981.
\bibitem{CW92}
C.~Chui and J.~Wang, ``On compactly supported spline wavelets and a duality
principle'', {\em{Trans. Amer. Math. Soc.}}, {\bf{330}}, 903--915
(1992).
\bibitem{CDF92}
A. Cohen, I. Daubechies, and J.C. Feauveau, ``Biorthogonal bases of compactly supported wavelets'', {\em{Comm. Pure and Appl. Math.}}, {\bf 45}, 485--560 (1992).
\bibitem{Daub88}
I. Daubechies, ``Orthonormal bases of compactly supported wavelets'', {\em{Commun. Pure Appl. Math.}}, {\bf 41}, 909--996 (1988).
\bibitem{Daub93}
I. Daubechies, ``Orthonormal bases of compactly supported wavelets II, variations on a theme'', {\em{SIAM J. Math. Anal.}}, {\bf 24,} 499--519 (1993).
\bibitem{CFK08}
D. \v{C}ern\'a, V. Fin\v{e}k, and K. Najzar, ``On the exact values of coefficients of coiflets'', {\em{Cent. European J. Math.}} {\bf 6,} 159--169 (2008).
\bibitem{Chen95}
D. Chen, ``Spline wavelets of small support'', {\em{SIAM J. Math. Anal.}}, {\bf 26}, 500--517 (1995).
\bibitem{Han06}
B. Han and Z. Shen, ``Wavelets with short support'', {\em{SIAM J. Math. Anal.,}} {\bf 38}, 530--556 (2006).
\bibitem{RNS17} L. Rebollo-Neira and I. Sanches,
, ``Simple scheme for compressing sparse representation of melodic music'', {\em{Electronics Letters}} (2017) DOI:10.1049/el.2017.3908
\bibitem{RN18} L.~Rebollo-Neira,
``A competitive scheme for storing sparse representation of X-Ray medical images'' (2018)
{\em{PLOS ONE}}. {\tt{ https://doi.org/10.1371/journal.pone.0201455}}.
\bibitem{HDF5} L. A. Wasser, ``Hierarchical Data Formats - What is HDF5?'', {\tt{https://www.neonscience.org/about-hdf5}} (2015)
\bibitem{MITDB}
{\tt{https://physionet.org/physiobank/database/mitdb/}}
(Last access May 2018).
\bibitem{Karl} K. Skretting, {\tt{http://www.ux.uis.no/$\sim$karlsk/proj99}} (Last access May 2018).
\bibitem{LRN19} L. Rebollo-Neira, ``Effective high compression of ECG signals at low level distortion'', {\em{Scientific Reports}} {\bf{9}}, Article number: 4564 (2019).
\bibitem{WSN13}
J. Wang, M. She, S. Nahavandi, A. Kouzani,
``Human Identification From ECG Signals Via Sparse Representation of Local Segments'', {\em{IEEE Signal Processing Letters}}, {\bf{20}}, 937 -- 940 (2013).
\bibitem{RR18}
S. Raj, K. C. Ray,
``Sparse representation of ECG signals for automated recognition of cardiac arrhythmias'', {\em{Expert Systems with Applications}}, {\bf{105}}, 49--64 (2018).
\bibitem{LKL11} S.J. Lee, J. Kim and M. Lee,``A Real-Time ECG Data Compression and Transmission Algorithm for an e-Health Device'',
{\em{IEEE Transactions on Biomedical Engineering}},
{\bf{58,}} 2448--2455 (2011).
\bibitem{MZD15}
J.L. Ma, T.T. Zhang and M. C. Dong, ``A Novel ECG Data Compression Method Using Adaptive Fourier Decomposition With Security Guarantee in e-Health Applications'',
{\em{IEEE Journal of Biomedical and Health Informatics,}}
{\bf{19}}, 986--994 (2015).
\bibitem{TZW18}
C. Tan, L. Zhang and H. Wu,``A Novel Blaschke Unwinding Adaptive Fourier Decomposition based Signal Compression Algorithm with Application on ECG Signals'',
{\em{IEEE Journal of Biomedical and Health Informatics}}
10.1109/JBHI.2018.2817192, 22 March 2018.
\bibitem{PSS16}
A. Pandey, B. Singh, S Sain, and N. Sood,
``A joint application of optimal threshold based discrete cosine transform and ASCII encoding for ECG data compression with its inherent encryption'',
{\em{Australasian Physical and Engineering Sciences in Medicine}}, {\bf{39}}, 833--855 (2016).
\end{thebibliography}
\end{document}
\begin{table}[h!]
\caption{Performance metrics
for each of the 48 records in the
MIT-BIH Arrhythmia database
listed in the first column. }
\label{TABLE2}
\begin{center}
\begin{tabular}{||r|r|r|r|r|r||}
\hline \hline
Rec & $\PRD$&$\SR$ &$\CR^{\Hf}$& $\QS$& $\PRDN$
\\ \hline \hline
100 & 0.51& 27.19& 28.27& 55.75& 12.64 \\
101 & 0.51& 25.39& 25.89& 50.82& 9.45 \\
102 & 0.51& 24.43& 24.74& 48.27& 13.13 \\
103 & 0.52& 21.93& 22.06& 42.73& 7.86 \\
104 & 0.53& 19.00& 19.52& 37.21& 10.27 \\
105 & 0.53& 17.37& 18.15& 34.22& 6.43 \\
106 & 0.53& 18.63& 18.87& 35.97& 7.07 \\
107 & 0.55& 12.64& 12.66& 23.01& 3.18 \\
108 & 0.52& 21.13& 22.74& 43.52& 8.51 \\
109 & 0.52& 19.32& 19.42& 37.05& 5.16 \\
111 & 0.51& 23.22& 24.32& 47.45& 9.90 \\
112 & 0.53& 22.72& 24.47& 46.18& 10.20 \\
113 & 0.52& 20.21& 19.80& 38.18& 6.26 \\
114 & 0.50& 32.21& 33.13& 66.17& 14.52 \\
115 & 0.53& 20.18& 20.52& 38.98& 6.73 \\
116 & 0.58& 12.33& 12.90& 22.10& 3.72 \\
117 & 0.52& 27.31& 28.82& 55.27& 9.33 \\
118 & 0.59& 11.06& 12.51& 21.22& 5.90 \\
119 & 0.56& 16.15& 16.50& 29.78& 4.44 \\
121 & 0.51& 33.27& 33.61& 66.10& 7.29 \\
122 & 0.56& 16.42& 17.62& 31.80& 6.50 \\
123 & 0.53& 21.94& 22.93& 43.11& 7.94 \\
124 & 0.53& 23.28& 23.09& 43.92& 4.94 \\
200 & 0.53& 15.56& 16.69& 31.33& 7.05 \\
201 & 0.49& 34.72& 35.05& 71.66& 12.52 \\
\hline \hline
\end{tabular}
\begin{tabular}{||r|r|r|r|r|r||}
\hline \hline
Rec & $\PRD$&$\SR$ &$\CR^{\Hf}$& $\QS$& $\PRDN$
\\ \hline \hline
202 & 0.51& 26.24& 26.28& 51.87& 8.44 \\
203 & 0.55& 12.75& 13.86& 25.14& 5.54 \\
205 & 0.51& 25.25& 26.83& 52.54& 12.50 \\
207 & 0.51& 25.93& 25.66& 50.80& 7.08 \\
208 & 0.54& 15.09& 15.69& 29.16& 5.55 \\
209 & 0.54& 15.65& 16.83& 31.40& 9.92 \\
210 & 0.51& 23.84& 24.62& 48.18& 9.72 \\
212 & 0.56& 12.12& 13.56& 24.40& 8.35 \\
213 & 0.56& 11.71& 12.06& 21.71& 4.09 \\
214 & 0.52& 19.22& 19.15& 36.59& 5.54 \\
215 & 0.55& 14.07& 15.34& 28.20& 9.60 \\
217 & 0.53& 16.86& 16.47& 31.06& 4.30 \\
219 & 0.54& 18.06& 18.00& 33.69& 4.43 \\
220 & 0.53& 19.18& 19.65& 37.07& 7.65 \\
221 & 0.52& 22.73& 23.00& 44.69& 8.50 \\
222 & 0.51& 26.52& 27.53& 54.42& 13.53 \\
223 & 0.53& 18.56& 18.70& 35.09& 6.06 \\
228 & 0.52& 18.97& 20.31& 38.91& 7.53 \\
230 & 0.52& 19.12& 19.47& 37.29& 7.30 \\
231 & 0.51& 25.67& 26.20& 51.84& 9.23 \\
232 & 0.50& 29.72& 31.59& 63.83& 14.91 \\
233 & 0.54& 14.26& 14.81& 27.43& 4.96 \\
234 & 0.52& 19.81& 20.50& 39.28& 7.68 \\ \hline
{\bf{mean}}&{\bf{0.53}} & {\bf{20.60}} & {\bf{21.26}}&
{\bf{40.76}} & {\bf{7.99}} \\ \hline
{\bf{std}}& {\bf{0.02}} &{\bf{5.79}} & {\bf{5.82}} &{\bf{12.47}}& {\bf{2.92}} \\ \hline \hline
\end{tabular}
\end{center}
\end{table}
%
%
%mean(mprd) mean(PDR) mean(SRQ) mean(CR) mean(QS) mean(PDRN)
% 0.53 0.53 20.60 16.24 31.08 8.01
%
% std(mprd) std(PDR) std(SR) std(CR) std(QS) std(PRDN)
%% 0.02 0.02 5.78 4.51 9.62 2.93
\begin{table}[h!]
\caption{Comparison between the average compression
performance of the proposed method and the methods
in \cite{LKL11},
\cite{MZD15}, \cite{TZW18} and \cite{LRN19},
for values
of ${\ov{\PRD}}$ in the range [0.31 \; 0,81].}\label{TABLE3}
\begin{center}
\setlength{\extrarowheight}{0.15cm}
\begin{tabular}{|l||r|r|r|r|r||}
\hline \hline
${\ov{\PRD}}$&0.81&0.80& 0.67& 0.48& 0.31\\
\hline \hline
${\ov{\CR}}$\cite{LKL11}& & & 11.30 & 9.28 & 6.22\\
${\ov{\CR}}$\cite{MZD15}& &18.00 & & & \\
${\ov{\CR}}$\cite{TZW18}&16.85& & & & \\ \hline
${\ov{\CR}}$\cite{LRN19}&33.32&33.05& 27.79& 19.84&10.59 \\
$\ov{\CR}^{\Hf}$\cite{LRN19}&36.54&36.26&31.01&23.09&15.00\\ \hline
${\ov{\CR}}\,\,\,\,\,\,\CDF_{\Di}$ &24.03&23.81&20.52 &14.79& 9.37\\
$\ov{\CR}^{\Hf}\, \CDF_{\Di}$ &31.64& 31.07& 26.71& 19.59&12.50 \\ \hline
${\ov{\CR}}\,\,\,\,\,\,\CDFF_{\Di}$ &24.76& 24.50 &20.67& 14.67 &8.89 \\
$\ov{\CR}^{\Hf}\, \CDFF_{\Di}$&32.59& 31.93& 26.01&19.25&11.93 \\ \hline \hline
$\prdo$& 0.750 &0.740 &0.645 & 0.435 & 0.275\\
$\Delta$&60 & 60 & 40 & 30 & 16 \\ \hline \hline
\end{tabular}
\end{center}
\end{table}
%\begin{table}[h!]
%\caption{Comparison between the average compression
%performance of the proposed method and the methods
%in \cite{LKL11} and
%\cite{MZD15} for the same mean value of ${\ov{\PRD}}$.}\label{TABLE3}
%\begin{center}
%\setlength{\extrarowheight}{0.15cm}
%\begin{tabular}{|l||r|r|r|r|r|r|r|r|r|r|r|r||}
%\hline \hline
%${\ov{\PRD}}$&1.47&1.18&1.14&1.06&1.05&1.03& 0.94&0.91& 0.81&0.80& 0.48& 0.31\\
%\hline \hline
%${\ov{\CR}}$\cite{LKL11}&& &&&& && & & &9.28 & 6.22\\
%${\ov{\CR}}$\cite{MZD15}& 33.85& 28.21 & & & 25.64 & & &22.27& &18.00 & & \\ \hline
%${\ov{\CR}}$\cite{TZH18}& 35.53 & & 25.99& & &22.80 &20.38& & xx & & & \\ \hline
%$\CDFF_{\Ba}$&& &&&& & &25.25&22.87&19.21&13.61 &8.33 \\
%$\CDFFH_{\Ba}$&& &&&& & &32.74 &29.46 & 25.16& 18.85 & 11.89 \\ \hline
%$\CDFF_{\Di}$&& &&&& & &27.30 &24.50 & 20.67& 14.67 &8.89 \\
%$\CDFFH_{\Di}$&& &&&& & &35.70 & 31.93 & 26.91 & 19.25 & 11.93 \\ \hline
%$\Delta$&& &&&& & & 80 & 60 & 40 & 30 & 16 \\ \hline \hline
%\end{tabular}
%\end{center}
%%%%%%%%%%%\end{table}
\begin{table}[h!]
\caption{Comparison between the average compression
performance of the proposed method and the methods in
\cite{MZD15}, \cite{TZW18}, \cite{PSS16} and \cite{LRN19},
for values
of ${\ov{\PRD}}$ in the range $[0.91 \; 1.06]$.}\label{TABLE4}
\begin{center}
\setlength{\extrarowheight}{0.15cm}
\begin{tabular}{|l||r|r|r|r|r||}
\hline \hline
${\ov{\PRD}}$&1.06&1.05& 1.03& 0.94& 0.91\\
\hline \hline
${\ov{\CR}}$\cite{MZD15}& &25.667& & &22.27\\
${\ov{\CR}}$\cite{TZW18}& & &22.80 &20.38& \\
${\ov{\CR}}$\cite{PSS16}&18.59& & & & \\ \hline
${\ov{\CR}}$\cite{LRN19}&41.31&41.24&41.00& 38.52 &37.68\\
$\ov{\CR}^{\Hf}$\cite{LRN19}&44.42& 44.23& 44.04& 41.93 &41.08\\ \hline
${\ov{\CR}}\,\,\,\,\,\,\CDF_{\Di}$&30.36&30.12&29.44 &27.29& 26.27\\
$\ov{\CR}^{\Hf}\, \CDF_{\Di}$&40.25&39.96&39.11&35.94&34.42\\ \hline
${\ov{\CR}}\,\,\,\,\,\,\CDFF_{\Di}$&31.93& 31.66& 30.84&28.46 &27.28 \\
$\ov{\CR}^{\Hf}\, \CDFF_{\Di}$&42.73& 42.39& 41.40&37.81&35.70 \\ \hline \hline
$\Delta$& 100 & 100 & 100 & 80 & 80 \\
$\prdo$& 0.940& 0.930 & 0.900&0.850 &0.805\\ \hline \hline
\end{tabular}
\end{center}
\end{table}
%\begin{table}[h!]
%\caption{Comparison between the average compression
%performance of the proposed method and the methods
%in \cite{LKL11} and
%\cite{MZD15} for the same mean value of ${\ov{\PRD}}$.}\label{TABLE3}
%\begin{center}
%\setlength{\extrarowheight}{0.15cm}
%\begin{tabular}{|l||r|r|r|r|r|r|r|r|r|r|r|r||}
%\hline \hline
%${\ov{\PRD}}$&1.47&1.18&1.14&1.06&1.05&1.03& 0.94&0.91& 0.81&0.80& 0.48& 0.31\\
%\hline \hline
%${\ov{\CR}}$\cite{LKL11}&& &&&& && & & &9.28 & 6.22\\
%${\ov{\CR}}$\cite{MZD15}& 33.85& 28.21 & & & 25.64 & & &22.27& &18.00 & & \\ \hline
%${\ov{\CR}}$\cite{TZH18}& 35.53 & & 25.99& & &22.80 &20.38& & xx & & & \\ \hline
%$\CDFF_{\Ba}$&& &&&& & &25.25&22.87&19.21&13.61 &8.33 \\
%$\CDFFH_{\Ba}$&& &&&& & &32.74 &29.46 & 25.16& 18.85 & 11.89 \\ \hline
%$\CDFF_{\Di}$&& &&&& & &27.30 &24.50 & 20.67& 14.67 &8.89 \\
%$\CDFFH_{\Di}$&& &&&& & &35.70 & 31.93 & 26.91 & 19.25 & 11.93 \\ \hline
%$\Delta$&& &&&& & & 80 & 60 & 40 & 30 & 16 \\ \hline \hline
%\end{tabular}
%\end{center}
%%%%%%%%%%%\end{table}
\begin{table}[h!]
\caption{Comparison between the average compression
performance of the proposed method and the methods
in
\cite{MZD15}, \cite{TZW18}, and \cite{LRN19} for values
of ${\ov{\PRD}}$ in the range $[1.14 \; 1.71]$.}\label{TABLE5}
\begin{center}
\setlength{\extrarowheight}{0.15cm}
\begin{tabular}{|l||r|r|r|r|r||}
\hline \hline
${\ov{\PRD}}$&1.71&1.47&1.29& 1.18& 1.14\\
\hline \hline
${\ov{\CR}}$\cite{MZD15}&38.46&33.85& &28.21 & \\
${\ov{\CR}}$\cite{TZW18}&42.27&3.53&30.21& &25.99 \\\hline
${\ov{\CR}}$\cite{LRN19}&62.48&56.78&49.60&47.04&45.75 \\
$\ov{\CR}^{\Hf}$\cite{LRN19}&63.92& 58.59 & 52.19 & 49.59&48.47 \\ \hline
${\ov{\CR}}\,\,\,\,\,\,\CDF_{\Di}$&52.01&43.83&37.63 & 33.88& 32.69\\
$ \ov{\CR}^{\Hf}\, \CDF_{\Di}$&66.99& 56.44& 48.77& 44.45&42.98\\ \hline
${\ov{\CR}}\,\,\,\,\,\,\CDFF_{\Di}$&57.19& 47.43& 40.66 & 36.27 & 34.62 \\
$\ov{\CR}^{\Hf}\, \CDFF_{\Di}$&73.27&61.00& 52.56&47.19&45.51\\ \hline \hline
$\Delta$&150& 120 & 110 & 110 & 110 \\
$\prdo$&1.660&1.430& 1.220& 1.070& 1.020\\ \hline \hline
\end{tabular}
\end{center}
\end{table}
%CR97_1=[66.99 56.44 48.77 44.45 42.98]
%CR53_1=[73.27 61.00 52.56 47.19 45.51]
%CR97_0=[52.34 43.86 37.66 34.14 32.92]
%CR53_0=[57.18 47.44 40.68 36.86 34.91]
%del=[150 120 110 110 110 ]
%PRD0=[1.670 1.430 1.220 1.0800 1.0300]