In wireless body sensor network (WBSN), the set of electrocardiograms (ECG) data which is collected from sensor nodes and transmitted to the server remotely supports the experts to monitor the health of a patient. However, due to the size of the ECG data, the performance of the signal compression and reconstruction is degraded. For efficient wireless transmission of ECG data, compressive sensing (CS) frame work plays significant role recently in WBSN. So, this work focuses to present CS for ECG signal compression and reconstruction. Although CS minimizes mean square error (MSE), compression rate and reconstruction probability of the CS is further to be improved. In this paper, we provide an efficient compressive sensing framework which strives to improve the reconstruction process, by adjusting the sensing matrix during the compression phase using the rain optimization algorithm (ROA). With the optimal sensing matrix, the compressed signal is reconstructed using Step Size optimized Sparsity Adaptive Matching Pursuit algorithm (SAMP). The results of this work demonstrate that the optimised CS framework achieves a higher compression rate and probability of reconstruction than the standard CS framework.
The term “WBSN” refers to a networking technology that connects several sensor nodes within or on the human body It very well may be utilized in the use of medical care for persistent monitoring of patients [
WBSN can sense numerous physiological signals like electromyogram (EMG), electroencephalogram (EEG), electrocardiogram (ECG), internal temperature, pulse and interestingly, the patients’ breathing or movement. Among them, ECG is critical because it enables the analysis of cardiovascular diseases, which are the leading cause of death, according to a WHO report [
A few algorithms have been presented over the decades for ECG signal compression and reconstruction. Compressed Sensing (CS) is a developing structure in contemporary signal processing that enables the capture and preparation of insufficient signals as well as signals with sparse representation in some appropriate premise [
In terms of sensor energy consumption, data transmission rate is seen as a critical factor in determining how much energy is consumed by WBSN devices. A great deal of ongoing investigates are basically centred on decreasing the number of data transmissions in WBSN. CS is primarily concerned with three perspectives: signal sparse representation, view of the measurement matrix, and reconstruction procedure. To enhance the performance of sparse reconstruction in CS, sensing matrix is to be optimized in the phase of compression.
We propose a technique to improve the reconstruction process by optimizing the CS matrix using rain optimization algorithm (ROA). Using this algorithm, optimized CS matrix is chosen. Mean square error (MSE) is considered as an objective function. The CS matrix with minimum error is chosen as an optimal matrix.
Then, the compressed signal is reconstructed by applying step size optimized sparsity adaptive matching pursuit algorithm.
Remaining sections of the manuscript are sorted as follows. Compressive sensing for ECG signals based recent research works are survived in Section 2. Section 3 proposes optimized compressive sensing, namely ROA based data compression and step size optimized sparsity adaptive matching pursuit algorithm for reconstruction. Results of the proposed scheme are discussed in Section 4. The conclusion of the research work is described in Section 5.
Recent Compressive sensing based EGC signal compression and reconstruction works are reviewed in this section. Rezaii et al. [
Ansari et al. [
Polanía et al. [
Abhishek et al. [
Jahanshahi et al. [
Rakshit et al. [
Kumar et al. [
The ECG data sensed by the body sensors are collected by a device like mobile phone as shown in
The input data or signal is represented as a sparse signal and compressed by multiplying it by the observation matrix. To enhance the performance of the data compression, the sensing matrix must be optimized to achieve this. So, for optimization of sensing matrix, rain optimization algorithm (ROA) is presented. Then, at the medical server, with the observation matrix, the original signal is reconstructed using step size optimized SAMP algorithm. The suggested reconstruction approach employs a smaller step size to improve the performance of the SAMP algorithm.
Using a predetermined matrix, compressive sensing is used to convert data from a high-dimensional space to a low-dimensional space. Besides, this technique is applied to compress the signal and to reconstruct the compressed signal. Compressive sensing consists of three major stages: sparse representation, generating the observation matrix, and signal reconstruction.
As the original signal is not sparse, it must be transformed and represented in sparse domain. In compressive sensing, sparsity means that the signal has
As per the function of sparse transform, the input signal
The compressed samples of the signal
The
Emmanuel
As the computational complexity of compressive sensing is very high especially in reconstruction, the performance of the compressive sensing is to be enhanced. As a result, we will concentrate on data compression and reconstruction which is further enhanced by refining the sensor matrix using the ROA method. It also results in an increase in the reconstruction performance and as next process, we present step size varied SAMP reconstruction algorithm which also enhances the performances of reconstruction further. Nevertheless, we present step size varied SAMP reconstruction algorithm which also enhances the performances of reconstruction further. The next sections discuss compressive sensing’s optimum data compression and reconstruction.
As shown in
The rain optimization algorithm (ROA) imitates the behaviour of rain drops. Raindrops normally stream down along a slant from a pinnacle at that point structure the waterways and consistently move to the lowest land points or void out into the ocean. The raindrops follow breaks and overlays in the land as they stream downhill. As the water flows downhill, it may become trapped in the potholes and features of the local optimum lake. Finally, the majority of streams reach the global ideal level and empty out into the ocean. The way the proposed method evolves from speculation to ideal is akin to a raindrop falling from a mountain to the ocean level in a hilly terrain due to gravity. As consistently raindrop will always choose the path with the more extreme slant, ROA reproduces this propensity and uses the slope of the target capacity to decide the solution that is better than an estimate.
To track down the deepest valley and afterward reach to the ocean level known as global optimal, a population of raindrops is created arbitrarily at initial iteration. The positions of the neighbour-points of each drop are compared with the position of drop before the drop moves towards the neighbour point possessing the least position. This development proceeds until the drop arrives at the valley. However long every one of the drops are streaming down from an upper to a lower position, regardless of whether there are puddles on the drops‟ route to the valley, they can in any case flood and rise out of the puddles to continue to move towards the valley by a fitting component performed by the ROA algorithm.
The process for optimizing sensing matrix (S) using ROA algorithm is described as follows:
Each particle or raindrop in a population denotes the partial solution in this procedure. The optimal sensing matrix is the overall solution to this technique (S). The initialization of ith drop is defined as follows,
Rainfall manages raindrops during the process of optimization. It is created by uniform random distribution function and subject to subject to constraints as given in
After initialization, fitness or objective function is recursively treated for optimal solution. To select the optimal sensing matrix, the solution should satisfy the following fitness function,
According to
As the raindrop (D) is defined as a point in N dimensional space, the domain which has the radius vector (r) places around the point is known as the neighbourhood. The neighbourhood can be updated when the changes occur in the value of raindrop.
During the process of optimization, a point in the neighbourhood of drop can be generated randomly. The ith drop’s neighbourhood point
Dominant drop: The dominant neighbour point (
The above process is repeated until finding the solution with the minimum fitness function. Otherwise, the algorithm is terminated.
After attaining the optimal sensing matrix (Soptimal), the compressed sample is calculated as follows,
From the compressed output, the signal is reconstructed using Step Size optimized SAMP. The following section explains the reconstruction algorithm.
Data compression is further enhanced by refining the sensor matrix using the ROA method. It leads to enhance the performance of reconstruction too. Nevertheless, we present step size varied SAMP reconstruction algorithm which also enhances the performances of reconstruction further.
In 2008, Thong T. Do, introduced the SAMP algorithm. In this algorithm, sparsity is not considered as priori data for construction. SAMP algorithm is also applicable when the signal has unknown non-zero values. It proposes the staged method i.e., it changes the step to attain the real sparsity of the signal. Additionally, this algorithm incorporates the concept of backtracking into each iteration and workflow of the SAMP algorithm’s in the tth iteration of the stage is depicted in
Step 1: As shown in the
Step 2: Then candidate list is generated by union of the finalist from the previous iteration and short list from the initial test. It is defined using
Step 3: The candidate list a subset of coordinates that solves the least square solution is considered the finalist in the block of final test. It is defined mathematically. It is defined using
Step 4: Finally, due to the subtraction of observation data or compressed samples and the projection of it to the sub matrices in the finalist, the observation residual is updated. Its observation residual is defined using
Step 5: If the algorithm satisfies the halt condition
If
Else
Step 6: The iteration is updated until halt condition is true.
Step size optimized SAMP: The step size of the SAMP algorithm is increased if the output of an iterative reconstruction does not satisfy the criteria. Additionally, the larger step size in iterations get maintained and this leads to use of ineffective strides in steps. Due to this, the algorithm’s stability and accuracy may be compromised. Thus, to improve the SAMP algorithm’s performance, the step size is improved under particular constrained conditions. So, to enhance the performance of the SAMP algorithm, step size is optimized under certain condition. As shown in the step 5, if
The proposed scheme is simulated in the platform of MATLAB with the system has the operating system of windows ‘10 with 64 bit and with 4GB main memory at 2 GHz dual-core PC. The MIT-BIH Normal Sinus Rhythm dataset is used in this study. This dataset contains 18 long-term ECG recordings of people referred to Boston’s Beth Israel Hospital’s Arrhythmia Laboratory. The subjects for this dataset were determined to have no significant arrhythmias; they included five men aged 26 to 45 and thirteen women aged 20 to 50. The input ECG signal to the optimized CS framework is shown in
The performance of the optimised CS framework is evaluated in terms of mean square error (MSE), root mean square error (RMSE), signal to noise ratio (SNR), and reconstruction probability for varied levels of sparsity in this section. The comparison analysis in
Sparsity level | MSE | RMSE | SNR | Reconstruction probability | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|
ROA-CS | GWO-CS | CS | ROA-CS | GWO-CS | CS | ROA-CS | GWO-CS | CS | ROA-CS | GWO-CS | CS | |
20 | 535.18 | 560.59 | 582.46 | 23.13 | 23.67 | 24.13 | 27.65 | 27.48 | 27.28 | 0.94 | 0.77 | 0.68 |
40 | 542.6 | 568.01 | 589.88 | 23.29 | 23.83 | 24.29 | 27.71 | 27.54 | 27.34 | 0.95 | 0.751 | 0.55 |
60 | 464.04 | 631.72 | 511.33 | 21.64 | 25.16 | 22.64 | 27.11 | 28.03 | 26.74 | 0.87 | 0.753 | 0.65 |
80 | 523.55 | 548.97 | 570.84 | 22.88 | 23.42 | 23.88 | 27.55 | 27.39 | 27.19 | 0.98 | 0.83 | 0.6 |
100 | 535.22 | 560.63 | 582.51 | 23.13 | 23.67 | 24.13 | 27.65 | 27.48 | 27.28 | 0.88 | 0.72 | 0.64 |
The comparative analysis of the RMSE of the different CS framework for varying sparsity level is shown in
To enhance the compression rate and reconstruction probability of the CS in WBSN, an optimized CS has been presented in this paper. The performance of the data compression phase been enhanced by optimizing the sensing matrix of CS using rain optimization algorithm (ROA). With the optimal sensing matrix, the reconstruction phase is performed using the step size optimized SAMP algorithm. The performance of the optimized CS framework has been analyzed by varying sparsity level and compression ratio. Besides, the performance of the ROA based CS has been compared with that of the GWO-CS and conventional CS. Additionally, the ROA-based CS performance was compared to that of the GWO-CS and traditional CS. The simulation results indicated that the suggested ROA-CS framework achieved a lower mean square error, root mean square error, signal to noise ratio, and reconstruction probability. In the future, we will focus on enhancing the secure transmission of ECG data in WBSN.
The authors with a deep sense of gratitude would thank the supervisor for his guidance and constant support rendered during this research.