Energy-Efficient Compressed Sensing in Cognitive Radio Network for Telemedicine Services

,


Introduction
Interdisciplinary researchers from computer science, advanced communication systems, and medical science are collaborating to create a smart ehealth care system to address the crisis in the availability of health care professionals and hospital facilities. The World Health Organization (WHO) has reported that around 17.5 million people die due to heart attacks, more than 246 million suffer from diabetes every year, and it is expected that 20 million people will lose their lives to cardiovascular diseases by 2025. WBAN-based telemedicine brings a solution to prevent these deaths. Telemedicine systems can provide healthcare facilities to patients by communicating the medical data of those patients to a doctor. Wireless communication technologies can support several electronic health applications to transfer medical data. This technology also improves service flexibility and mobility for different telemedicine applications. WBANs have recently been used to monitor vital physiological parameters of the human body, including temperature, pulse, blood pressure, respiratory rate, oxygen saturation (SpO2), and level of consciousness. However, the implementation of WBAN faces various challenges in terms of its extremely low power requirement, lightweight, security, privacy maintenance, reliable transmission of vital parameters, emergency medical care, real-time connectivity of networks, less complexity, interoperability, interference mitigation, and better quality of service [1].
All the existing WBANs in hospital environments use the ISM band, which is overcrowded [2]. Hence, it leads to the loss of patient information, and measures need to be taken to check the available free spectrum. Cognitive radio (CR) was proposed as a baby step toward the solution to the spectrum scarcity problem [3]. CR is an emerging technology for using the spectrum efficiently without interfering with licensed users. It can solve the fundamental problems of interference and spectrum scarcity faced by communication networks. CR plays a major role in supporting telemedicine applications with the required QoS at a minimal cost [4]. Therefore, implementing CR in telemedicine applications is a promising approach that can mitigate interference during transmission [5]. However, wearable WBANs have the limitation of a minimum sensor size and a minimum battery power. Especially frequent battery changes are very inconvenient for the patients. The size of the hardware and battery power directly affect customer satisfaction. Hence, energy efficiency is an important parameter in the design of WBAN. The main motivation of this work is to propose an energy-efficient, clustered compressed sensing cognitive radio-based WBAN model for telemedicine services with interference mitigation.

Related Work
Spectrum sensing is the core of the cognitive radio network. Huge research has been carried out in developing an efficient and robust spectrum sensing method since Mitola and Maguire coined the word "cognitive radio" [3,6]. However, a majority of the research has been focused on narrow-band spectrum sensing. These methods include energy detection, eigenvalue detection, cyclostationary detection, and matched filter detection [7]. Wavelets are used to identify the spectral holes [8]. However, there is a practical demand for wideband spectrum sensing for cognitive radio. Compressed sensing (CS) in wideband spectrum sensing was first applied by using sub-Nyquist sampling with a wavelet edge detector [9]. The CS technique recovers signal from a smaller number of samples compared with traditional methods [10]. CS implements spectrum sensing by acquiring compressed samples, reconstructing the Nyquist signal from the compressed signals, and then applying spectrum sensing to the reconstructed signals. However, it is suggested that in order to sense the spectrum availability, the compressed measurements are sufficient [9]. Hence, the reconstruction step can be eliminated, thereby saving energy and time and reducing complexity [11,12].
Generally, the wireless devices that operate in the unlicensed spectrum are based on Wi-Fi, Bluetooth, and Zigbee. As these devices are operated in an unlicensed ISM band, it may cause severe congestion, resulting in harmful interference between wireless medical devices. The majority of proposed telemedicine applications use unlicensed band communication, which leads to congestion and poor QoS [13].
Several proposed systems in telemedicine use narrowband transmissions. The standards IEEE 802.15, IEEE 802.11, and IEEE 802.16 support license-free bands [13]. The communication range for the IEEE 802.15 standard is restricted to 2-3 m for WBAN and tens of metres for WPANs, and hence, this standard is not suitable for remote data transmission in telemedicine [14]. The communication range for the IEEE 802.11 standard is several hundred metres and is suitable for short-range transmission, while the standard IEEE 802.16-based networks can support long-range transmission. When multiple networks coexist in the same license-free spectrum, there arise issues including spectrum utilization, transmission collisions, and security [15], causing problems for health monitoring applications. A mobile telemedicine system for emergency healthcare services is proposed by which critical biosignals of patients are transmitted to hospitals for consultations.
A few scheduling algorithms are proposed [15][16][17][18][19] for cognitive radio-based WBANs. The interference between WBANs and medical devices causes harm to patients in telemedicine services. In this case, each WBAN is assigned a fixed priority, and the channels are assigned based on the priority value [20]. But the fixed priority method is not very efficient, as the emergency data from the sensors must reach the distant doctor with the minimum delay. The use of ultrawideband CRBANs for biomedical data transmission is   [14]. In this method, CR uses three different and unique channels for different types of traffic. There are three types of traffic addressed, namely, emergency data, control data, and normal data. Control and normal data are allocated at 2.4 GHz, while emergency data is allocated at 900 MHz. Hence, the spectrum is managed efficiently. Cognitive radio is a paradigm for improving quality of service (QoS) standards in telemedicine services [20][21][22]. The underutilised spectrum can be effectively utilised by the cognitive radio networks (CRNs) through compressed sensing with a nonreconstruction model [11,12]. It helps telemedicine services with low latency and high bandwidth. A cognitive radio-based hospital environment consists of cognitive radio-enabled wireless body area networks (WBANs), medical devices, and coordinators. The WBANs connect various sensors located in the human body, both interior and exterior. The biosignals from the patient are collected by biosensors and transmitted to a distant doctor using a priority scheduling algorithm. However, the realtime implementation of cognitive radio is a big challenge with the available computing power. Nanocomputing brings a possible solution to this challenge as it offers efficient and cost-effective computing with more than millions of computations within the shortest possible time [23].
Few works in WBANs focus on energy efficiency. In [24], an energy-efficient EEG-driven biometric system is experimented with a minimum number of electrodes and minimum computational power. A reliable and energyefficient mobile agent scheme for WBANs is proposed [25]. In this scheme, the network is partitioned into clusters, and a cluster head is selected. The required data from the base station is collected by the mobile agent. This method also incorporates fault tolerance. A detailed review of energy-efficient and reliable routing technique is suggested [26]. However, sensitive medical data must be secured. An attribute-based, efficient and secured architecture (ABE) is proposed to support data transfer from one to many and access control mechanism for medical data with confidentiality [27]. An efficient model used in wireless sensor networks (WSN) is scheduling, in which nodes are operated only for a certain prescribed time period. The selforganising maps (SOM) approach is used for improving the performance of the scheduling algorithm [28]. This mapping approach analyses the problem based on competitive learning techniques. Thus, the major challenges in implementing WBAN for telemedicine are (i) interference caused within the crowded ISM band between BANs and other wireless devices could have a harmful effect on patients, (ii) increased energy consumption of WBAN nodes    Wireless Communications and Mobile Computing leads to a reduction in the lifetime of the network, and (iii) critical care services to patients should be given more priority over normal services to give immediate attention and medication to patients in the ICU and emergency care unit. In the proposed method, (i) a cognitive radio controller is used to sense the available free spectrum for communication to avoid interference. (ii) Compressed spectrum sensing is applied by the CR controller to save energy. (iii) The Kmeans clustering algorithm is used to cluster all the nodes in the WBAN network, thereby saving energy. (iv) Data packets with higher priority are assigned to critical data than normal data, and scheduling is followed to prioritise the data transmission.

System Model
The model of a telemedicine system is shown in Figure 1. It consists of WBANs, a coordinator, and a controller. The

Input:
1. N= number of CRBAN nodes 2. k opt = number of desired clusters 3. E threshold = Threshold Energy 4. d threshold = Threshold distance = √(d fs /d mp ) Output: 1. Optimum number of initial centroids 2. Set of k optimum clusters Steps: 1. Compute the distance from the center to each node. 2. Arrange the nodes according to the distance computed in step 1.
3. Partition the sorted nodes into k opt sets. 4. In each set the middle point is the initial centroid. 5. This node at the centroid is the initial CH. 6. Repeat 7. Based on Euclidean distance the remaining nodes join their nearest CH.

Centroid Formula to Calculate the Centroid of Each Cluster
Allocate an ID number to each cluster node, with a smaller ID number to nodes closer to the cluster's initial CH. 10   Wireless Communications and Mobile Computing execution of the routing algorithm and optimal cluster head selection are performed by the cognitive radio controller. The wireless biosensors that are embedded in a patient's body continuously monitor the vital signals of the patient, including heartbeat rate, blood pressure, temperature, and electrocardiography (ECG). The data sensed by WBANs is transmitted to a coordinator node. Further, the aggregated data is transmitted to the cognitive radio controller. The CR controller has the ability to sense the free spectrum and transmit data through it. As a result, interference is reduced because transmission occurs through the available free spectrum following sensing. However, this sensing is performed with a compression and nonreconstruction model to save energy. All the WBANs in a hospital environment are clustered using the K-means algorithm for improved energy efficiency thereby increasing the lifetime of the network. However, an effective scheduling algorithm is applied to facilitate multiple nodes accessing the spectrum with minimum interference. Scheduling packets at sensor nodes is important for prioritizing applica-tions in telemedicine services [19,21]. For instance, some biosensor nodes would be given a higher priority than other nodes. Hence, a priority scheduling algorithm for WBANs is needed.
3.1. Clustering BAN Nodes. In a simulated hospital environment, WBANs are grouped to form a cluster using the kmeans clustering algorithm, as shown in Figure 2. A group of biosensors attached to a patient is referred to as a "node." Each cluster is formed with a cluster head (CH) and the member nodes [29]. The member nodes communicate their data to the associated CH. The CH applies cognitive radio to sense the free channel on all licensed and unlicensed channels. All the data from the nodes are transmitted to the CH.
3.1.1. K-Means Clustering Algorithm. The deployment of an energy-efficient CRBAN network is one of the important objectives of ehealthcare applications. As a result, the K algorithm is a clustering algorithm that groups N nodes by 1. While the CH receives data from WBANs in the cluster 2. Examine each piece of information from the WBANs. 3. Sort the packets in descending order of their values. 4. The packets with large values (critical data) are given higher priority than other packets. 5. if no higher priority data is processed 6. Send higher priority packets to the server from all clusters 7. else 8. Send less important packets 9. if a higher priority packet arrives at the cluster head 10. Stop sending lower-priority packets; send that higher-priority packet, then resume sending other packets 11. end if 12. end if 13. end while Algorithm 2: Priority scheduling algorithm.  Figure 3 depicts an example of cluster formation using the K-means algorithm. The algorithm partitions 20 nodes, which are randomly scattered over a region into two clusters. Same-colored nodes belong to the same cluster, and the centre is marked by the white star symbol.
Using the following equation over an M × M region, this clustering algorithm calculates the optimal number of clusters for a given number of nodes N.
Hence, the optimum number of nodes depend upon the distance from the CH to the base station, the free space model and the multipath model parameters as given in equation (1). Algorithm 1 describes the K-means algorithm for optimum clustering with distance-based CH selection and energy-efficient data communication.
The centroid of each set is made the initial CH. All the remaining nodes are allotted with ID based on the distance between each node and CH. The residual energy of the initial CH along with K-means algorithm forms a strong metric for selecting the optimum CH, and hence, cluster heads can successfully deliver the data to the BS. The residual energy of the initial CH is compared with the threshold energy in each round. If the energy is less than the threshold, the next CH with a lower ID is chosen as the CH and broadcasted in the network. This algorithm ensures a minimum distance between CH and BS thus ensuring reduced energy consumption. In order to find this small distance, a threshold is selected based on the propagation model used. The distance between CH and BS is calculated and compared with the distance threshold. If the distance is greater than the threshold, then that node is not taken as CH; otherwise, the node is considered CH. Hence, energy consumption for data communication is reduced.

Priority Scheduling.
The data from all the sensors on each WBAN reaches the cluster head. The emergency packet should be given the highest priority. The sorting of packets is done in descending order of their values. Each CH has a 1. x is the input signal of dimension N × 1, x ϵ R N . 2. As per compression theory model, y = ϕx. 3. ϕ ϵ R M×N is the CS measurement matrix, 4. CV of the received signal Y can be calculated using R = ð1/MÞyy ǂ: 5. FC computes ith center of Gerschgorin circle C i= r ii and radius by Ri = ∑ j≠i jr ij j.

GRCR test statistics is formed by
The test statistics are compared with the threshold and if T GRCR > ξ then PU is present, otherwise PU is absent, where ξ is the decision threshold fixed to achieve a fixed false alarm rate.
Algorithm 3: For the compressed sensing method.    Wireless Communications and Mobile Computing sorted list of packets, starting with the ones with the highest priority. All CHs send the highest priority packet first, followed by the lowest priority packet. Assume the CH receives a higher priority packet while transmitting lower priority packets; the highest priority packet is processed and the transmission of lower priority packets is resumed. There are various medical and nonmedical devices used in the hospital which are operated in either real time or in nonreal-time mode depending on the applications [20]. Some medical devices are in a critical zone like ICU and CCU, while other medical devices are in a noncritical zone like a general ward and cabin. Other nonmedical devices include wireless communication tools like video conferencing, surveillance systems, medication and pharmacy information booths, etc. For improving QoS in the proposed CR-enabled hospital environment, all the above-mentioned devices are classified into high, medium, and low priority categories, as shown in Figure 4. Priority queuing (PQ) is the simple method to support various traffic service classes. In this method, the packets   There are two advantages to this queuing technique. It has a low computational load compared to other queuing disciplines. PQ allows routers to organize buffered packets and then offer different kinds of services to different traffic. Algorithm 2 describes the priority scheduling algorithm.

Spectrum Sensing with Compression.
In order to mitigate interference in the WBAN network, cognitive radio with a compressed sensing model is used. The CS model is shown in Figure 5, where the sparse signal x is compressed using the sensing matrix Φ [30].
In this multi input multioutput (MIMO) model, the received signal matrix is obtained by considering m secondary users and each user collecting n samples. This signal is represented as y ϵ C mxn .
Compressive sensing takes advantage of the sparsity of the signal. It compresses a k-sparse signal x ϵ ℝ N by multiplying it by a measurement matrix Φ ∈ ℝ M×N where M≪N. The resulting vector y ϵ ℝ M is called the measurement vector. Hence, the compressed signal in this work is given by "H" represents hypothesis with noise alone condition, ϕ is a measurement matrix, and y is the compressed matrix.
The compressed measurement and its covariance matrix is given in equation (4) and ǂ denotes complex conjugates and transposes. By substituting (2) and (3) into (5), the covariance matrix R for noise and signal is given by equations (6) and (7), respectively.
The test statistics of SC 0 and SC 1 are given by The Algorithm 3 describes compressed sensing method.
The performance of the proposed method is analyzed by applying different measurement matrices for compressing the received signal. The detection is done on the compressed signal without reconstruction, leading to a reduction in complexity.
In this proposed method, with compressed sensing, the free spectrum is detected and communication is initiated in the free channel. Thus, communication always takes place in the free channel thereby mitigating interference due to BANs using the same spectrum. There are different methods to mitigate interference, including the power control method, the signal processing method, and the UWB approach [31]. The power control method uses a fuzzy controller to minimize the power level once interference is detected. The reinforcement learning method is also employed in the power control method to mitigate interference. In the UWB-based wireless body area sensor network (WBASN), an interference cancellation scheme is used to reestimate the interference-affected signal at the receiver. In the signal processing approach, multiuser detection is applied to mitigate interference. The detectors use different In order to analyse the interference mitigation method quantitatively, two parameters are used. The signal-tointerference ratio (SIR), which is the ratio of signal power to noise and interference power, IMF is the interference mitigation factor, which is the ratio of the signal-to-noise ratio (SNR) to the signal-to-interference ratio.

Energy Efficiency of Compressed
Sensing. The compressed sensing method reduces the amount of data to be transmitted. As a result, the amount of energy consumed decreases as the compression ratio is increased. Thus, an energy model can be formulated by considering the sampling rate (M), average energy consumption in joules per bit (C), and number of bits (b). This model can be formulated as A simulation experiment is performed by varying the value of compression ratios. The resulting energy savings are shown in Table 1, and the simulation environment is shown in Table 2. With compressive measurements without reconstruction, the energy efficiency is represented in Figure 6. This experiment shows a significant improvement in energy efficiency through compressed sensing without reconstructing the sample. Hence, this compressed sensing is implemented in CRN for telemedicine applications. A telemedicine simulation model with two clusters, each with ten nodes, is considered in the network, as shown in Figure 7.
The proposed method is simulated using MATLAB 2019 and compared with three other similar schemes, namely, energy-efficient spectrum-aware reinforcement learningbased clustering (EESA-RLC), energy-efficient link scheduling (ELS), and CogMed [16][17][18][19]21]. The performance metrics chosen for comparison are end-to-end delay, throughput, network lifetime, and stability and residual energy.  Table 3. For SNR 20 dB and SIR 20 dB, the minimum BER is achieved by the proposed method. The proposed cognitive radio interference mitigation approach (CR-A) is compared with the UWB approach (UWB-A), the signal processing approach (SP-A), and the power control approach (PC-A) [31]. The proposed method outperforms all the existing methods, as shown in Figure 8.
In the proposed CR approach, spectrum availability is sensed, and once the channel is free, data is transmitted. Hence, the CR approach outperforms other methods, as shown in Table 4.  by measuring the time taken for the last node to die (LND), and network stability is measured by calculating the time taken for the first node to die (FND). Simulations for different rounds are performed, and recorded readings are shown in Figure 9, where it is apparent that the proposed method outperforms the other methods.

End-to-End Delay Analysis.
End-to-end delay refers to the time taken for a packet to be transmitted and received by the receiver. It usually differs from the round-trip time (RTT) of all the active nodes present in the network. This delay depends on the rate at which packets arrive at their destination and the number of PUs occupying the spectrum. The comparison of the end-to-end delay analysis of the proposed method with the existing system is shown in Figure 10.

Energy Consumption Analysis.
A CR-based approach enables us to adjust coding, modulation, radiated power, and control component characteristics to minimize energy consumption. The total energy consumption in the WSN depends on two parameters: (i) the total number of WBANs and (ii) the rate of packet generation; hence, energy consumption can be reduced by clustering techniques. Clustering minimizes network energy consumption.
The residual energy with respect to the number of rounds is given in Table 5. In the proposed method in each round, the cluster head is selected by considering the energy consumption of each node, the residual energy of the node, and the location of the node with respect to the controller. Hence, an optimal node is selected with an improved lifetime and improved network energy. The residual energy refers to the amount of remaining energy in the network. The energy consumption analysis is shown in Figure 11.  Figure 12 compares the throughput values of the proposed scheme with the existing methods for different numbers of rounds. The proposed method exhibits better performance compared to ELS, EESA-RLC, and CogMed methods. In our priority scheduling algorithm, the highest-priority packets are given higher importance than other packets, and they reach the destination with less delay. From the graph, it is evident that the delay of a high-priority packet in a priority scheduling-based system is less than that of the existing systems. When the delay is greater, the throughput decreases. As the delay is minimized, the proposed method outperforms the existing method.

Conclusion
In this paper, a novel energy-efficient cognitive radio enabled model is proposed to implement WBAN for telemedicine. WBAN is connected to the cognitive radio controller, which performs compressed sensing to detect available free spectrum to mitigate interference. This compressed sensing reduces energy consumption by choosing the right compression ratio, thereby improving energy efficiency. The cognitive radio approach outperforms other methods in terms of IMF. The CR controller applies the Kmeans clustering algorithm to reduce routing difficulties related to bandwidth, throughput, and power consumption. This clustering algorithm reduces energy usage and improves the lifetime of the network. A dynamic priority scheduling algorithm is applied to allocate channels for communication-based priority. The emergency data from the WBANs is given higher importance than other data. The data are transmitted based on priority, where the higher-priority packets are transmitted first, lower-priority data is transmitted next. The simulation results present clearly the superior network performance realised by the dynamic priority queue-based system.

Data Availability
We recognize it is not always possible to share research data publicly, for instance, when individual privacy could be compromised, and in such instances, data availability should still be available in the manuscript.