7.1.1 IoT Communication Models

The standard guidelines in the form of architectural document were released by Internet Architecture Board (IAB) in the year 2015. The guidelines outlined four communication models for IoT devices. Rose et al. [2] briefly discuss the various technical communications models described by IAB and highlight the flexibility of connecting IoT devices. The discussion below explains key characteristics of each model.

Device-to-Device (D-to-D) Communication: In this type of communication two or more devices connect and communicate without intervention of any other device. Such devices connect and communicate with each other using different types of protocols like Bluetooth, Z-wave, and Zigbee. This communication model (Figure 7.2) is commonly used in to control home appliances (like smart light bulb, smart door lock, smart switch etc.) where small packet of information needs to be communicated with low data rate. The limitation of this model is interoperability. It means devices belong to Z-wave protocols are not compatible with devices belongs to Zigbee protocols.

Device-to-Cloud (D-to-C) Communication: In Figure 7.3, this type of model IoT devices directly communicates to cloud service provider. This model support data storage and computation in cloud due to the limited data storage and computation power of IoT devices. It uses existing communication structure like wired Ethernet or WiFi connection to connect devices with cloud server. The performance of this model basically depends on bandwidth and available network resources.

Device-to-Gateway (D-to-G) Communication: In Figure 7.4, device-to-gateway communication model IoT devices are connected to cloud service provider through gateway. Application layer gateway will act as an intermediary between devices and cloud service provider. This model improves the security and brings some level of computation at application layer that improves the performance of IoT devices. For instance, Smartphone act as a gateway that runs some application to communicate with IoT devices and the cloud.

7.1.2 IoT Architecture

IoT architecture (Figure 7.5) is basically classified into three layers. The upcoming and latest technologies are upgrading these layers day by day. Sethi and Sarangi [3] had elaborated IoT architecture in their work. They also explained about protocols, and various applications of IoT in versatile domain.

IoT Devices

IoT devices collect data through various sensors that are connected to each other. The data collected by these sensors is transferred further to IoT gateway. These devices are called as data generator devices and they generate different types of data like environmental data, health data, and motion data to name a few.

IoT Gateway

The role of IoT gateway is to perform data processing on the collected data from variety of sensors. The initial processing of data will convert it into meaningful data which is further transmitted to the cloud server. There are many data transfer technologies that might be used like- ZigBee, Bluetooth, Wireless Area Network, Wi-Fi etc.

Cloud Server

Cloud servers are used to store huge volume of data generated by sensors and passed to them through IoT gateway. Data is processed and analyzed and drawn results, by application of various machine learning algorithms. The results are transmitted to the end user in the form of some alert message, text message of may be through email. For example: Any fluctuations in health related data of the patients will send alarm to the concerned healthcare center.

7.1.3 Protocols for IoT

The protocols can be defined as collection of norms that mention common set of standards for data exchange among various devices. In a previous section, the role of different layers in IoT architecture were described, now in this section the focus is to explain the different protocols (Figure 7.6) responsible for seamless communication among the layers. Shadi et al. [4] describes various protocols used in different layers for communication of data.

7.1.3.1 Physical/Data Link Layer Protocols

The section will elaborate various physical and MAC layer protocols. The role of these protocols is to connect various IoT devices; it may be two sensors or sensors and gateway. There are many protocols functioning at this layer, but some most commonly used protocols are discussed below:

Bluetooth: It is used for short range communications and works at 2.4 GHz in ISM band. It works on the concept of master and slave. In two connected Bluetooth device one may by master and other one will be slave [5]. Two new protocols are introduced under IoT protocols Bluetooth Low Energy (BLE) or Bluetooth Smart for the low power consumption devices. As compared to classical protocol it uses less power and less latency as well. These protocols use adverting and data frames, when nodes are communicating then only, they are awake.

ZigBee: This IoT protocol is meant to be used in applications with less frequent data exchange. The connectivity range between nodes is 100 meters and data transmission rate is 250 Kbps. The protocol is normally used with the applications related to building and home automation, industrial control system etc. It is latest version 3.0 is based on IEEE 802.15.4 standard. It is the unification of the various ZigBee wireless standards into a single standard. Another variant ZigBee smart energy is developed for variety of IoT applications, and it supports different types of topologies as well like peer-to-peer, cluster-tree, or star. This protocol is easier to install and maintain. Zigbee Alliance [6] established in 2002 is responsible for maintaining and publishing the Zigbee standards.

Z-Wave: This wireless protocol uses low energy radio waves and it is basically meant to be used with home appliances. Z-wave [7] allows the user to control the home appliances remotely like light control, door lock, thermostats etc. via the internet from a smart phone or tablet. In different countries it operates at different frequencies and covers physical range 100 meters. It supports data rates up to 100kbit/s. Like Bluetooth it also follows master/slave architecture where master sends the command to slaves and controls the whole network. It is MAC protocol and uses CSMA/CA for collision detection to avoid the possibility of data loss. [RFC 5826] elaborates the basic requirements for the data routing over lossy and low power networks specifically for home appliances.

LoRaWAN: This Long Range Wide Area Network protocol covers long range and consumes less power. It is basically used for the intercommunication between smart devices that consume low power and meant to be used in smart cities. The LoRaWAN [8] uses varied frequency in variety of networks and the data transmission rate is varied between 0.3 Kbps-50 Kbps. [RFC 8376] describes an overview of the LoRaWAN protocol and architecture.

WiFi: The IEEE 802.11 standard (also called WiFi). The commonly used protocol for intercommunication among devices to transfers the high-speed data i.e. up to 1 Gbps. The frequencies are 2.4GHz and 5GHz bands. The range it covers is 50 m. It works in integration of IP (Internet Protocol) IEEE 802.11, but it is not appropriate for IoT applications due to power consumption and frame overhead issue. To overcome this problem IEEE 802.11ah was developed. The elementary 802.11ah MAC layer had features like efficient bidirectional packet exchange, increase sleep time, short MAC frame.

7.1.4 IoT Applications

IoT applications make our life smart and safe. There are many sectors where we found it is playing an important role like smart house, smart healthcare, smart farming, smart manufacturing, smart transportation and many more. Zeinab and Elmustafa [17] reviewed various application of IoT and suggested future avenues for many related technologies. Some applications are discussed below.

Smart Home: The concept of smart home refers to a technology where home appliances and devices can be controlled remotely from anywhere with the help of smart phone, laptop or tablet. The wide range of devices and home appliances are interconnected through the internet that makes the user to control it remotely like room temperature, light, door lock, washers and dryers, refrigerator and many more. Electricity prediction and home security are the most useful applications in smart home as it helps to forecast home energy consumption and allows monitoring home remotely. Matta and Pant [18] addresses the current trends, major research challenges, and application domains in the field of Internet-of-Things (IoT).

Smart Healthcare: Healthcare is one of the applications of IoT that makes many researchers, industries and the public sector to work on it. IoT technology makes doctors to monitor the patient remotely and allows patient to visit virtually. In this way it improves the quality of care and reduces the cost of care in comparison of traditional method of healthcare. It also helps hospital to track patient, staff, and various equipment. As many healthcare devices IoT it helps in early detection of diseases. Dang et al. [19] in the year 2015 gave an analytic review on the prominent usage of IoT devices in the healthcare industry.

Smart Farming: Smart farming is a concept of farming integrating it with technologies like IoT, Artificial Intelligence, Drones, Sensors, Robotics. IoT may help in resolving many critical issues of traditional farming and improves productivity at a lower cost. Smart farming includes various activities like remote monitoring of crop, automation of irrigation, weather forecasting, analyzing ideal time of plantation and harvesting, detection of pests, observing soil conditions and many more [20]. Smart farming relates to improvement in both quality and quantity with low risk. Gill et al. [21] suggested an automatic information system for smart farming based on cloud server. The system provides updated and required information the end-users automatically.

Smart Manufacturing: Industry 4.0 or in other word smart industry represents revolution in manufacturing industry. It is a fourth generation of industry where entire manufacturing process is automated that increases the overall profit of company. It changes how products are invented, manufactured and shipped. Each machines and various industrial equipment are embedded with sensors that generate data which is helpful to analyze the errors in productions process and takes correct action. Vuksanović et al. [22] discussed the role of IoT devices and its usage in the future scenario of the factories.

Smart Transportation: The idea of smart transportation deals with predicting the traffic congestion in advance that helps the user to avoid unnecessary traffic jam by knowing the alternative route. As in many countries’ government is supporting to design electric vehicles to reduce the fuel cost and impact of global warming. In some countries there is a progressive research in the direction of designing self-driving car. In such types of research IoT will perform an important task like detecting pedestrians, traffic signs, obstacles etc. Ferdowsi et al. [23] discuss deep learning in intelligent transportation systems for traffic flow prediction, self-driving car and traffic signal.

7.1.5 IoT Challenges

Internet of things applications outlined above is rapidly increasing over last few years but still there are some challenges while implementing IoT. Omid Helmi et al. [24] analyzed and elaborated different types of challenges that are there in IoT environment. Following are the area of concern-

Security and Privacy: IoT devices do share lot of personalized data of the end user, this leads to the various security and privacy concerns for that data. In a study it has been observed that almost 70 percent of IoT devices are vulnerable in nature due to non-availability of standard encryption and authentication techniques. The manufactures and policy makers should frame some standards to deal with this scenario. Alansari et al. [25] elaborated about vulnerabilities of IoT systems and discussed about various model that may be used to protect data and systems.

Interoperability: IoT includes millions of interconnected devices over the internet. These devices are smart devices and able to process the information. But each device has different information, processing, and communication capabilities as the manufacturing companies around the globe produce various product categories. Interoperability among heterogeneous devices connected in any application is one of the critical issues with IoT devices. There are numerous interoperability challenges that can be dazed by the proposed CoRE framework by Mazayev et al. [26].

Data Analytics: The voluminous amount of data generated by heterogeneous IoT devices cannot be analyzed in a proper manner due to poor quality of data. It is a tough task to segregate useful and useless data among such huge volume of data. There are numerous challenges regarding data management that are pointed by Cooper and James [27].

Volume of Data: Some IoT applications include frequent communication between devices and continuous collection of data that will increase the volume of raw data. The processing, storing and management of huge amount of data is another challenging task with IoT technology. Constante et al. [28] in his work analyzed the impact of large volume of data and the processing challenges. They also draw attention towards research works tools used for such data analysis.

Power Supply: The smart IoT devices are not connected with continuous power supply hence they need self-reliant energy source. The power consumption depends mainly on data processing, transmission, and sensing as well. The power supply requirement for the wireless technologies is towards higher side be it Bluetooth or Wi-Fi. This generates a demand of less power consuming wireless technologies. Perković et al. [29] developed power estimation analysis system. The core work of their system it the sum up the various possibilities for its future use.

7.2.1 Cloud vs. Fog vs. Edge

In this section, Table 7.1 compare cloud computing, fog computing and edge computing.

7.2.2 Existing Edge Computing Reference Architecture

The new paradigm of computing in network environment is edge computing, where resources and services are now shifting from cloud to the edge. Edge is now gaining momentum and various researchers provided different architecture, capabilities, terms and features for Edge architecture. The following section elaborates various reference architectures (RA) with respect to edge computing.

7.2.2.2 Intel-SAP Joint Reference Architecture (RA)

Intel and SAP [33] developed reference architecture for edge computing environment that delivers solutions for cloud and sensors integration. The RA facilitates the enterprises to understand the secure management of smart network of IoT devices. The network of IoT devices collects, manages and analyze the data and combine it with business intelligence. Figure 7.8 represents the various blocks of Intel-SAP joint RA.

7.2.3 Integrated Architecture for IoT and Edge

The integrated architecture of IoT and edge (Figure 7.9) helps in eradication of various challenges in the existing architecture. It handles voluminous data, issue of latency, privacy and many more. The edge facilitates processing of collected data near to the source like gateway, router, laptop etc. The core of edge computing is to reduce overhead on cloud. Wei Yu et al. [34] briefed the integration of IoT with edge and elaborated its challenges and benefits.

Sensors and Actuators: In IoT environment data is generated through sensors and they are capable to sense the surrounding environment such as temperature sensors, humidity sensors, motion sensors, smoke sensor, gas sensor etc. The role of actuators is to convert energy into motion. Sensors are capable data collection only, so the data collected with these sensors pass to IoT board that is capable in data processing and data communication.

Edge Computing: The latency in case of edge computing is reduced in various real time applications by performing data processing just near to the source of data generating devices. Edge computing improves the performances of IoT devices, reduces bandwidth utilization and provided better security and privacy. It helps to offload the various important tasks which are earlier performed by the centralized server called cloud server. It includes two major components one is called edge devices and the other one is edge gateway.

Edge Devices: Edge devices perform real time data processing based on the trained model. These devices are either IoT devices which is a primary source of raw data or IoT board that is directly connected to the sensors which collects the real-world data. There are some single board computing devices are also available that are so powerful and smart enough to replace the desktop in near future. These devices are even capable to execute machine learning algorithms and very small in size and low in cost. Some of the popular single board computers and their specifications are – Raspberry Pi 4 model, Latte Panda Alpha, Udoo Bolt and Jetson Nano (NVIDIA).

Edge Gateway: Edge gateway performs pre – processing of data, allows data storage and various intelligent algorithms are installed. The devices that can be used as edge gateway are network devices (like routers, switches, and gateway), intelligent controllers (like PLCs, RTUs, and DCS) and micro cloudlets. Micro cloudlets are also called as micro data centers with high end processing speed, large amount data storage capacity and high bandwidth wireless LAN. The Edge gateway provides communication standards for various interfaces like wire, radio-based and technologies like – Bluetooth, 3G mobile radio, ZigBee, Z-Wave, LTE. Preprocessing of data and preliminary learning of data at edge gateway will help to offload the work of centralized cloud server that results the performance of the time sensitive applications will improve and provides the benefits to the end user.

Centralized Cloud Server: Cloud server is accountable for further processing of data received from edge computing layer, enhance the analysis of data by introducing advanced machine learning techniques and store the data as a historical data for future research. It mainly handles large set of data by using big data processing tools and technologies. Some of the best big data tools are Hadoop, HPCC, Storm, Qubole and many more. These tools are cost effective and incorporate time management in various data analytics tasks. It also includes some advance learning techniques called deep learning. Deep Learning (DL) creates multi-layered structure of neural networks. In comparison to ML algorithms Deep learning algorithms are step ahead. Popular DL algorithms are ANN (Artificial Neural Network), CNN (Convolutional Neural Network), RNN (Recurrent Neural Network), RBM (Restricted Boltzmann Machines), DBN(Deep Belief Network), LSTM (Long Short Term Memory Networks) to name a few. Training and testing of models are performed at cloud server and finally the trained model is deployed over edge devices to take early decision.

IoT Application: IoT applications are various apps that are delivering the services to end user and act as a user interface with the help of API (Application Programming Interface). These apps help the user to access the real time data and make them to take early decision. There are various sectors where sectors where IoT plays an important role.

7.3 Edge Computing and Real Time Analytics in Healthcare

The concept of application of ML in integrated environment of Edge Computing and IoT is rising day by day. The combination of these technologies is now in application phase and is being deployed by many industries in various use cases. Smart devices may use ML and DL techniques at edge and reducing the burden on cloud as the processing will be performed at the local or device level.

Traditionally the data collected by IoT devices are processed by the cloud sever. The huge volume of such data overburdened cloud servers; the edge here plays a crucial role in diminishing the load of cloud server by application of ML/DL algorithms and classifying which data to be processed at edge and the other needs to be processed at cloud server by more powerful algorithms. There are many applications which need real time analytics to be performed on some crucial data especially healthcare and some financial applications; the application of ML/DL techniques at edge is an added performance benefit for such applications. Edge ML also deals with the security aspects of an application where personalized information is being stored at Cloud which is global in nature.

Healthcare sector is more and more focusing on providing better support to patients. The role of Deep Learning is becoming a promising and transformative in healthcare. Basically Deep learning models consists of multiple layers, the input data passes through multiple layers and each layer performs a kind of matrix multiplications on data. After passing through multiple layers the last layer will ultimately produces a feature set or final output. There are variety of deep learning models used in a specific scenario like – DNN (Deep Neural Network), CNN (Convolutional Neural Network) and RNN (Recurrent Neural Network). Integration of deep learning with edge computing improves real time processing which is a crucial aspect in case of healthcare systems. To address the issue of latency and quick performance in real time applications, variety of integration model of edge with deep learning are proposed.

The general architecture of an Edge Computing Environment includes a sensor/smart device and edge/fog/cloud computing node. The edge layer is being used to bridge the gap between the cloud server and the IoT devices. IoT layer consists of variety of healthcare sensors each having unique identification. The data collected from these sensors is further transferred to the edge layer using WiFi, Bluetooth, or ZigBee.

The variety of devices is being used by researchers in various healthcare models categorized as smart device, medical instruments or IoT based sensor kits. The edge devices are growing day by day starting from Nokia mobile or PDA in their initial days. The latest smart phones are now being used in variety of healthcare applications. Smart phones sensors though are easy to use but have limitation of embedding variety of sensors in it. Using dedicated medical sensors helps in collecting and handling large volume of health data that may lead to more accurate diagnosis [41].

Author Liu et al. [42] proposed RMCS (Remote Medical Care System) model for miners to assess their physical and mental health by making use of biosensors. They applied emotion shift theory knowledge and made use of EHR (Electronic Health Record) to draw inferences. The brain wave sensors are fixed in miner’s helmet which collects data about physical and mental status of minors and issue early warning signals. The overall system uses ZigBee network. The information collected through sensors will be transmitted to minor’s mobile phone and from there the information will be transmitted to backend system where the physical and mental status of miners is evaluated using EHR ontology model. In case of emergency miner’s family and paramedic facilities are automatically informed.

Dubey et al. [43] developed a fog computing architecture (Figure 7.10) that emphasized on data reduction, low power consumption and highly efficient. Their proposed architecture is service oriented, and its main asset is low power embedded computer. The model performs data mining techniques and analytics on the data collected by number of wearable devices. The data read is in time series pattern and further analyzed by embedded computer. The analytics helps in identifying the unusual data in the series of data which are further transmitted to cloud for analysis.

The Intel’s Edison processor used by them is low power device that is supported by rechargeable battery. They used the DTW (Dynamic Time wrapping) algorithm for pattern mining in data. The role of DTW is to find similarity between two data series. Further they used CLIP (Clinical Speech Processing Chain) that is a series of operation to filter speech data. To compress data they used GNU zip for compression and decompression algorithm. Fog processor reduces data transmission and storage at cloud by processing it at local level for data for healthcare.

Speech Monitoring for Patients of Parkinson’s Disease: They collected speech data from smart watches and sent it to fog gateway. The Intel Edison processor at fog computer processes the speech data by DTW and CLIP. The features and patterns extracted at fog are further sent to the cloud. The data compression performed by DTW, CLIP and GNU. Among all CLIP achieved 99 percent of data reduction.

The authors proposed a cascading architecture as mentioned in Figure 7.10 based on fog computing which has three sub-systems as – BSN (Body Sensor Network) is used for data acquisition, the second sub-system is a fog gateway which will process the acquired data and for further back-end analysis cloud server is used. The role of fog gateway is to filter out the relevant information that needs to be transferred to the cloud server for further processing.

Gaura et al. [44] in their proposed work moved analytics at IoT nodes that leads to less transmission of data hence reducing network loads. This edge mining techniques takes place at the edge of IoT device. L-SLIP (Linear Spanish Inquisition Protocol), Class Act and BN (Bare Necessities) these three algorithms developed by authors are used to reduce sensing messages and eventually reducing less energy and less storage as well. Various sensors transmit redundant data, edge mining focus on saving packets rather than bits and hence improve performance.

The authors proposed the processing of data collected by various sensors at the edge nodes that are near to the sensors only and they convert raw signals into some meaningful information. The real work of edge nodes is to estimate the states and convert it into some meaningful information that is required for the application. The raw data from the sensors may give raw information like someone entered the room; however the state information like how many people are there in the room may be inferred from filtering information collected from the sensors. Further to state estimation, event detector will decide about the significant change in state.

The purpose of L-SIP algorithms is to acquire data from sensing applications and re-construct the signal if it’s loses energy. The purpose of Class Act algorithm is to recognize human posture and store the various timestamps in the change in posture. The BN algorithm provides the time spent in different positions. The overall performance is recorded as L-SIP reduces the packet transmission rate by 95%, BN reduces 99.8%, and Class Act reduces packet transmission by 99.6%. The results justified the importance of edge mining in various IoT based applications.

Bhargava et al. [45] proposed a WNS based system for the localization of elderly suffering from Alzheimer disease. The author proposed a system that will perform analysis of data collected at the local level close to the data source for better response and detailed analysis is performed at cloud level based on historical data. The fog based wireless sensor network system was proposed by authors for outdoor localization of elderly people. The analysis was performed on data collected through wearable devices obtained from the Kasteren dataset. The proposed system is a combination of two nodes – wearable device and cloud gateway. They used genetic algorithm for the activity recognition using IEM (Iterative Edge Mining) that facilitates localization. The mining algorithms on edge are based on SIP (Spanish Inquisition Protocol) that converts raw data into application specific states. The data is further transmitted to a cloud only if it is not predicted by past estimation or approximation model. The general SIP models are as follows -

L-SIP works on sensing applications where reconstruction of original signal is required.

Class Act approach is used to detect human postures; it stores and transmits timings of change in human postures but not the original signal. It is based on decision trees; it takes signals from body worn accelerometer sensors and then performs posture estimation. It is well suited for embedding in wireless sensors, it is simple and closely knitted with the application and hence capable of making strong assumptions about the output.

BN does not keep a record of timing it is appropriate to get summa-rized information of time duration of different states. States are being represented in the form of distributed and non-overlapping bins. Change in distribution of these bins represents occurrence of any event. Class Act and BN are destructive approaches hence preferably used in the applications where reconstruction of signals is not required. The results confirm that our system can achieve a high positional accuracy despite a short initial training period and low frequency of data computations on the resource constrained sensor devices.

Althbeyan et al. [46] proposed system that collects patient data like temperature, glucose level and other activities and transmits that data to the cloudlets. The abnormal values are further transmitted to the MEC (Mobile Edge Computing) servers and removed immediately from the memory in order to maintain privacy. The decision support system attached with MEC system has to analyze only abnormal value instead of bulk of values. This reduction technique to reduce computing complexity is called inexact computing used along with morphological filtering to reduce data processing.

The model proposed consists of four major components- cloudlets, MEC system, DSS (Decision Support System), DCS (Data Collection System). Their proposed model observed reduction in power consumption and per user delay.

Bhunia et al. [47] applied fuzzy techniques for selection of data to be transmitted for analysis of data collected through IoT sensors. Authors used this technique for the identification of heart condition of the patients. They prepared knowledge base and rules for the detection of some early signs of any serious heart trouble in real time. The data collected from the various sensors is aggregated and send to fuzzifier. The role of fuzzifier is to classify the parameters using some predefined functions. Finally, fuzzy inference system processes fuzzy sets that are based on the input from rule engine.

They programmed fuzzy system on the Arduino board which they used as an aggregator in the healthcare application. The selection of data to be transmitted is done by fuzzy rules. Required data to generate alert is only transmitted. They compared normal data collection technique with their proposed technique, and they observed considerable reduction in energy consumption. By adopting this approach, they can reduce energy consumption by sending only required data over costly communication channel.

Borthakur et al. [48] proposed usage of machine learning approach on the edge computing device which requires considerable fewer amounts of resources. They collected speech data of patients suffering from Parkinson’s disease through smart watches used by patients. They employed Intel Edison and Raspberry Pi as a Fog computer architecture. In general machine learning model are implemented at cloud to process physiological data for the traditional telecare model. K-means clustering algorithms are used for the analysis of data. Clustering of unlabeled data is performed at fog layer. This clustering can be helpful for the real time sub-grouping of Parkinson’s disease data. They collected 164 speech samples and selected features as average fundamental frequency and intensity. The feature extraction is done using Praat an acoustic software for analysis. K-means algorithm used for clustering is implemented in Python. Their proposed architecture is useful in handling speech processing and disorders in real time. Fog computing architecture reduces the dependency on cloud server. Their work utilized low resource machine learning approach on fog device and the results show that proposed architecture is promising for low- resource machine learning.

Hosseini et al. [49] developed a system for the monitoring and regulation of epilepsy. An automated edge computing framework (Figure 7.11) proposed by them to process volume of data generated which is a part of decision support system. The estimation model for the estimation of epi-leptogenic network using EEG and rs-fMRI is also developed. An unsupervised technique based on Deep CNN to differentiate between interictal epileptic discharge (IED) periods and nonIED periods using electrographic signals from electrocorticography (ECoG).

Their proposed framework is divided into three layers- Sensing layer, Edge computing gateway and cloud computing layer. Sensing layer consists of various biosensors to collect physiological data like heart rate, motion etc. The sensors are wearables and implantable also and the data collected is transferred to the Edge Gateway through Bluetooth, ZigBee and Wi-Fi as well. The Edge gateway does not work just like simple Access Point or Base Station it performs initial diagnosis and finally data is being transferred to cloud layer for long term storage.

The volume of data that is collected by the Biosensors is pre-processed at the edge layer before it is transferred to the cloud environment for further processing. Pre-processing of data at edge layer will provide real-time services to the patients. The EEG and rs-FMRI data collected is processed at local level. The data pre-processing is performed by Butterworth band pass filter which cuts frequencies and removed unwanted frequencies. Further for the identification of abnormalities at functional level re-clustering is applied. Wavelet transformations are applied to capture rhythmic nature of an epileptic seizure and for high level feature extraction. An edge computing gateway operates constantly even in the situation when connectivity between cloud and monitoring system is lost. Edge gateway supports cloud-based system by supplementing them in terms of real-time computing, low latency, heterogeneity, and scalability.

Sood et al. [50] proposed fog computing based system for the continuous monitoring and analysis of hypertensive patients. The data is collected through IoT sensors at fog layer for the identification of various stages of hypertension in the patients. After identification of stages, further analysis of risk level is done by application of Artificial Neural Network (ANN).

The fog-based system collects health related temporal data in a specified time window. This helps in risk assessment of hypertensive patients. The granulation of temporal data collected is associated with attribute of data values that is collected in a window time. The real time processing of data collected is performed for diagnosis. The diagnosis will produce an alert message on the end users phones to take preventive measures on time. The results show promising efficiency in bandwidth, minimum delay good response time, and higher accuracy.

Shammin et al. [51] proposed an edge computing-based system to control spread of COVID-19 epidemics. The biggest challenge of real-time monitoring of epidemic is to collect and process pathological, radiographic, and other types of information. They used Deep Neural Network for the processing and analysis of data collected during epidemic. Their three- layered framework - in which bottom layer consists of end users, stakeholders, hospitals and the decision makers. At this layer general health data is being collected like Blood pressure, temperature, and cough quality through app by patients. AI based module will investigate these preliminary symptoms and will categorize further action as per the rules. Middle layer is the combination of edge server, cache, and nodes. The topmost layer is cloud that consists of cloud servers and global storage.

Deep learning model works on cloud that contains servers with high computational power. The global storage area at cloud stores hospital and physiological test signals. These signals are in form of variety of images (X-Ray, CT-scan, protease sequences), and other physiological features like body temperature, blood pressure so on. The signals are collected from the connected network of hospitals. Three Deep Learning models are adopted ResNet50, deep tree, and Inception v3. Various streams are used for different types of image signals. Final step is decision fusion that is performed by each stream that helps in decision making even if any of the modality is missing.

The middle layer during the off-peak times downloads the DL model and shifts it to edge. The edge server is located nearby to the specified hospital. The collected data samples are transferred to edge through wireless technologies. The role of edge server is to secure the data using blockchain and further forward it to DL model. The explainable AI at edge is used for the understanding of output of each layer of DL model in the healthcare applications. The test data from hospitals and other symptoms are analyzed at edge using latest high power edge computers.

Ejaz et al. [52] developed an integrated system of edge computing and blockchain for the smart healthcare system. The objective of their system is to avoid long operating times, network issues, security and cost in highly dynamic network conditions. They assessed their system on the basis of following key terms like latency, power consumption, network utilization and computational load and compared the performance with the systems where blockchain is not used. Their model monitored various health parameters of the elderly person like blood pressure, heart rate, oxygen or blood sugar level. In case the parameters value goes beyond the range the alert message is issued for the healthcare centers.

Figure 7.12 represents a proposed framework consists of 3 layers – local tier, access tier and core tier. The local tier consists of variety of sensors that may collect data from various resources and then pass it on further to the edge or cloud servers for further processing.

At local tier, the edge nodes that are resource constrained do pass the data to the higher computational capacity node. The local nodes of edge are in the vicinity of the elderly or disabled person. The permissioned blockchain will run at the local edge nodes for the transfer of trusted information among different entities of the system like end users, healthcare workers, emergency units etc.

The access tier is much richer as compared to local tier and includes edge server, MEC servers and their role is to provide computational capabilities to the monitoring services. The basic role of this tier is to perform data analytics, decision making, security and privacy services, and allocation of resources. At this layer only blockchain provides an auction/renting of various resources for processing and storage capacity and/or for health-care services as well.

Core tier is the richest in term of resources it provides huge number of resources. The major role of this tier to provide service logic that is used to manage overall working of healthcare systems and provide required resources.

The integration of edge and blockchain results into secure, trusted and delay-critical healthcare services for remote monitoring cases such as monitoring of elderly and disabled people. Their proposed Health-BlockEdge framework brings resources near to the end users and performs real time data analytics and decision-making functions to improve system-level resource efficiency.