Review on Security of Internet of Things Authentication Mechanism

Internet of things (IoT) is considered as a collection of heterogeneous devices, such as sensors, Radio-frequency identification (RFID) and actuators, which form a huge network, enabling non-internet components in the network to produce a better world of services, like smart home, smart city, smart transportation, and smart industries. On the other hand, security and privacy are the most important aspects of the IoT network, which includes authentication, authorization, data protection, network security, and access control. Additionally, traditional network security cannot be directly used in IoT networks due to its limitations on computational capabilities and storage capacities. Furthermore, authentication is the mainstay of the IoT network, as all components undergo an authentication process before establishing communication. Therefore, securing authentication is essential. In this paper, we have focused on IoT security particularly on their authentication mechanisms. Consequently, we highlighted enormous attacks and technical methods on the IoT authentication mechanism. Additionally, we discussed existing security verification techniques and evaluation schemes of IoT authentication. Furthermore, analysis against current existing protocols have been discussed in all parts and provided some recommendation. Finally, the aim of our study is to help the future researcher by providing security issues, open challenges and future scopes in IoT authentication.


I. INTRODUCTION
It has been anticipated that all the things in the world are going to be internetworked [1].At present, internet-based services, which is a global network, are connections of computers and computing devices.The idea behind the Internet of Things is to expand the internet by not only connecting internetworking devices but also the non-IP components, like television, light, fan, refrigerator, and air-conditioner.IoT is not based on only at home but also in businesses like manufacturing organizations, vehicular networks, industries, grid companies, health The associate editor coordinating the review of this manuscript and approving it for publication was Luis Javier Garcia Villalba .
organization and so on.IoT is envisaged to be able to provide an advanced level of services to society and businesses.Therefore, all the things around the world will be fitted with embedded electronics and information technology so that it can produce valuable information based on the requirements and can work like important nodes of the network.Additionally, with the help of embedded electronics, embedded systems, embedded processors and embedded communication systems such small elements of environments can be connected to the network, depending on the applications and business requirements, to produce a huge internetworking environment, which is incomparable to the current network size.It is stated that more than 20.5 billion IoT devices will be connected by 2020 and over three trillion US dollars will be spent on only hardware of IoT [2].IoT is one of the building blocks behind the concept of smart home [3] and smart cities [4].
In the colossal of IoT network, which is connected with huge numbers of sensors and other devices, identifying one component raises a fundamental challenge, because that can cause privacy issues, governance of the system, access control, and overall architecture.Security and privacy are the most important factors in an IoT network [5]- [7].On the other hand, there are three security requirements: confidentiality, integrity, and availability.IoT needs to achieve these three requirements in order to fulfill security aspects.Moreover, the environment of IoT may differ from a centralized network to a de-centralized network, cloud to fog network.Therefore, security can be more tighten by enforcing detection techniques of unusual behavior or pattern of the network.This can be achieved in various ways, like a comparison header analyzer intrusion detection system (IDS) [8], based on a vector space representation using a Multilayer Perceptron (MLP) [9] or machine learning [10]- [12], deep learning [13].Besides, authentication in the IoT network takes place mostly by three components, which are the sensor, user, designed over the last few decades but none of these provides complete protection to the networks.Protocol designers are sometimes unaware of new threats in IoT networks.This motivates the authors of this paper to review on authentication, which is very important for future authentication protocol developer.Moreover, the authentication mechanism needs to be improved by comparing the existing authentication protocols.Therefore, the contributors to this paper include all the aspects of authentication protocols of IoT.

A. CONTRIBUTION OF RESEARCH
The main contribution of this work is to produce a comprehensive idea to the researcher about IoT authentication security and its peripherals.To formulate the idea, this research presented a well-developed taxonomy of attacks and a classification of technical methods used in IoT authentication systems.Additionally, network attacks have conversed against current IoT authentication protocols that can mitigate various threats.In addition, this paper elaborates on important evaluation techniques needed for authentication and compares it with existing protocols.Furthermore, this research extended to consider enormous security verification techniques, which are most important for the authentication mechanism.Additionally, this research produces important challenges and open issues that need to consider for future research proposals on designing an authentication mechanism.
The rest of the paper is formatted in the following manner (See FIGURE 2).In section II, this paper shows the classification of attacks and existing protocols to protect the IoT network from several attacks.Different technical methods of  the IoT authentication mechanism is provided in section III.After that in section IV, security verification techniques have been discussed followed by IoT authentication evaluation techniques in section V. Furthermore, open challenges and future directions based on IoT authentication are discussed in section VI.Lastly, this discussion has been concluded by pointing out important issues in the current phenomenon in section VII.

II. TAXONOMY OF ATTACK ON AUTHENTICATION IN IoT NETWORK
Attackers target network to gain access over it and get valuable information to sell over a black market [15] or fulfill their requirements.Among all the network attacks, this paper will concentrate on a range of attacks related to IoT authentications.FIGURE 3 illustrates the wellformulated taxonomy of attacks on IoT authentications.Furthermore, TABLE 3 demonstrates the description of every major attack as per as authentication is a concern in IoT networks.As per the document, the classification of all the attacks is clustered in seven major categories, which are masquerade attack, man-in-the-middle attack, DoS attack, forging attack, guessing attack, physical attack, routing attack.Firstly, the masquerade attack distinguishes itself from other attacks on fake identity aspects; on which attacker counterfeit identification of legitimate users.Forging attacks can be differentiated by its nature, where an attacker tries to imitate the existing component or system.Man-in-the-Middle (MitM), on the other hand, snoop network traffic between two communicators.In a DoS attack, the adversary floods the network with packets to jam communication and penetrate the network.Instead of imitating the existing components or flooding the network, adversaries predict and try to explore the possibilities of getting confidential authentication credentials of legal users in guessing attacks.Guessing attack has shown to be dangerous, but further exploitation on the network happens when an attacker tries to get access to the IoT network through physical components.This exploitation is typically called a physical attack.Lastly, a routing attack is to create a fake route to send or receive packets in an IoT network.Moreover, all the above categories of attacks in IoT authentications are elaborately described in the following sections using the counterpart of the existing protection mechanism.

A. MASQUERADE ATTACK
IoT authentication is based on identity and if the identity is compromised, then the network can be vulnerable.In the masquerade attack, the adversary uses fake identification to authorize himself as a genuine user in the network.If the IoT network is not properly protected, it can be attacked by masquerade attacks, which can be prepared using stolen identification like a user id or password or detecting user's behavior tracking.This type of attack in the IoT network is very common but it depends on the level of authorization a network has managed to attain.As such, masquerade attackers can have a full smorgasbord of cybercrime opportunities if they have gained the highest access authority to a business organization.FIGURE 3 elaborates a full range of possible masquerade attacks in IoT network based on authentication security.
Impersonation attack is a sophisticated attack in IoT, where the adversary intercepts the authentication request of the previous session of another user and uses that information  to authenticate itself.In contrast, Tu, et al. [16] proposed a novel techniques to handle the impersonation attack in fog computing using Q-learning algorithm.FIGURE 4 shows the before and after impersonation attack in the IoT network.
User impersonation allows an attacker to steal the information of an actual user to get into the system for unusual activities.A user impersonation attack can be done in several ways.It is practical that an actual user may be leaked server's private   information to the attacker.The legal user also can act like an attacker.Amin et al. [17] explained in their protocol on how to protect the IoT network from user impersonation attack during authentication.Furthermore, a plethora of protocols have been designed to protect IoT networks from attackers during authentication, but many of them are designed to protect specific kinds of attacks.Therefore, all of these protocols are open for many other attacks; sensor impersonation is one of them.During the process of authentication, users, sensors, GWNs or servers, exchange messages among themselves to come on a mutual goal.In this situation, an attacker can sense the network, get information sent by the sensor and modify the data to act as a legal sensor.
Hence, the improved scheme like Jiang et al. et al. [18] protocol, can resist sensor node impersonation attack.In such cases, IoT users deserve to be anonymized as their activities can be tracked and the pattern of the user's behavior can be predicted.An attacker can predict users' position and their network using capabilities if the authentication protocols are weak.In the same way, a central problem in sensor network security is that sensors are susceptible to physical capture attacks.Once a sensor is compromised, the adversary can easily launch clone attacks by replicating the compromised node, distributing the clones throughout the network, and starting a variety of insider attacks.Attackers can clone to the smart card, tags to get more opportunities to explore the network.Authentication protocols suffer from the challenges to protect cloning attack from either a high computation or storage overhead or poor detection accuracy.Wallrabenstein [19] proposed IoT Device Authentication using Physical Un-cloneable Functions.On the other hand, an identity theft attack is one of the tricky methods to get the identity of an authorized user in various unauthorized ways, such as data breaches, unsecured websites, social networks, phishing, public computers, and skimming.Authentication protocols are victimized by identity theft attacks in almost all the IoT sectors, including IoV, IIoT, and MIoT.Researchers have introduced several different techniques [20], [50] to counterpart the attack.In addition, In a network, a genuine user can behave as an attacker.An authorized user can also act like another legal user by using his/ her credentials.Therefore, an internal user who has authorized access to the system and the network launches an insider attack.Therefore, data protection by using anonymization techniques to hide personal information from the published dataset is essential.
However, attackers can use a composition attack to merge or overlap the same kind of dataset from a different body.Ganta et al. [51] discussed composition attack in auxiliary information and Baig et al. [52] show how to prevent composition attack in non-interactive data publishing setting by combining sampling and generation.Furthermore, an intruder can steal verification data form the authentication server in the current or past authentication sessions.Then the adversary tries to get into the server using the compromised data.An advanced three-way authentication technique for IoT is designed by Cui, et al. [53] to prevent various attacks, among them stolen-verifier attack is one of the most challenging.Additionally, the proliferation of software and technology growth allows users to provide the specific function of their activities, household device management or personal assistance.That third-party software can be hacked and user's activity can be monitored and used against them.Besides, IoT infrastructures are more prone to welcome these threats.Viana et al. [54] introduced conflict management in Systems of Systems (SoSs).The paper presented a framework for managing unpredictability in the system.In addition, many authentication protocols use the session key to protect from network attacks like a replay attack, but this session key can be compromised and used against the system to be a masquerade.However, as the authentication process needs many communication and message passing among nodes, the attacker can get a message and process among themselves and pass it back to the sender bypassing the actual node.This type of attack is called a node by-passing attack.IoT authentication schemes are in jeopardy of node by-passing attacks by GWN by-passing attacks, base station by-passing attack or sensor by-passing attack.Sarvabhatla and Vorugunti [21] designed a secure biometricbased user authentication scheme, which provides base station by-passing attack protection.Chang et al. [22] proposed two-factor authentication that can protect GWN by-passing attack whereas, authentication protocol for an IoT-enabled LTE network by Saxena et al. [23], gives protection towards secret key by-passing attack.The details about the IoT authentication protocols to protect against masquerade attacks are tabulated in TABLE 5.

B. MAN-IN-THE-MIDDLE ATTACK
In Man in the middle attack, an attacker secretly taps a network and absorbs communication data between two parties who trust that they are directly connected and communicating with each other.In this scenario, the attacker can drop, modifies, and alters the communication data as well as can predict network and security patterns.Additionally, they use legitimate users' data to establish new communication in the system.FIGURE 3 shows the classification and FIGURE 5 illustrates the man-in-the-middle attack in the IoT authentication scenario.In addition, TABLE 6 shows a description of all MitM attacks.
In a MitM attack, eavesdroppers try to steal authentication data by unauthorized way, while communication takes place between nodes, over an IoT network.Attackers try to find and establish a weak network connection between sensors and server and transfer network signal itself.Not only that but also, they install network monitoring software [49], which helps to snoop all transmitted authenticated data.However, eavesdropping is difficult to identify because of abnormality during transmission.Li et al. [55] proposed an interesting anti-eavesdropping scheme by friendly jammers to an industrial crowd-sensing network.Alternatively, message modification is a type of active attack, where an attacker sniffs actual data from the network and pass the modified data to the receiver.Asaduzzaman et al. [56] designed a protocol to offer better security over message modification attack near NFC architecture.Zhang, et al. [57] describes different abnormalities in VANETs, including massage suppression attacks.In massage suppression attacks, attacker multicast prevalent spoofed message over the network to prevent actual nodes to get original messages and force them to refresh cache every time.Pu and Zhou [58] shown a heuristic-based detection scheme (HED) to analyze and defend the message suppression attack in low power and lossy networks (LLNs).Likewise, a false message attack adversary sends inappropriate data to the victim to misguide the user.Moreover, data can be transferred to the server as well as the end-user.Nevertheless, several protocols are designed to sense and prevent false messaging attacks [57], [59], [60].Similarly, in a data manipulation attack, the attacker does not delete the data after retrieving from the actual source.Instead of deleting or tampering the data, they alter the actual content of the information and sends in to the targeted location.To emphasis, Khan et al. [61] designed a distributed intrusion detection system (IDS) to detect and protect network form data manipulation attacks.On the other hand, a substitution attackoccurs when an attacker deliberately replaces the authentication or authorization algorithm by a forgery code to validate fraud user or gain access to the system.As in the technique the actual encryption method is superseded, is vulnerable to different attacks.
However, during the past few years, hackers compromised several IoT networks to harvest user information including user id and hashed password even in worst-case plain text password.This compromised account is often offered in the black market [15] or leaked publicly.In addition, intruders often intercept data during the transaction in a network and can change the password of a legitimate user.As IoT is a collection of heterogeneous devices with internetworking systems and most of the peripherals are connected in WSN, devices always populate data and send over networks.Therefore, attackers can use powerful devices to sniff those packets, disseminate information, and use for their purpose.Moreover, a network can be accessed through a wired or wireless medium.Ferrag and Ahmim [62] and Yao et al. [63] designed different protocols to prevent packet tracking attacks.Similarly, after getting the raw packets from the targeted network, hackers try to extract information from the pool of [28] data.Therefore, they use strong tools to disseminate data [65] and to produce powerful information that they can use to intercept the user or network.Packet analysis attacks are prevented by several mechanisms in IoT [62], [63].Furthermore, adversaries intend to capture sensors in WSN to get information about network patterns and users' details.They try to hack the sensor by penetrating the network if the security of the network is weak.
However, researchers have noticed those attacks and designed IoT authentication protocols to protect them against this type of attack [28], [69].For the same reason, authentication protocols use session keys to prevent several attacks to occur in the network.However, this session keys can be compromised to design a new type of attack.If an attacker can get the session key of a particular session, then they can redesign the user's data and can create fake users and sessions to attack a network.Wu et al. [28] authentication and key agreement scheme ensure to protect against session key leakage attack.Similarly, after getting the session key and user details by network tapping, an eavesdropper can create a valid login request and start a new session with a sensor by masquerading a valid user.This type of attack is known as a parallel session and reflection attacks.Roy et al. [34] designed an authentication scheme with user biometrics and fuzzy extractor that protect against parallel sessions and reflection attacks.In general, IoT devices collect data and transmit them over the network in order to connect.During this process, devices emit signals, which is called ''side-channel''.These signals indicate the level of power consumption, electronic and aquatic emissions at any given time.At the same time, an intruder can overtake the encryption credentials by trespassing an IoT device using the side-channel attack.Moon et al. [77] proposed a  countermeasure of side-channel attack in IoT through a bit checking mechanism.TABLE 7 illustrates IoT authentication protocols, which give support to protect against man in the middle attack.Though the MitM attack is a serious issue in IoT authentication techniques, protocols are less attentive in different types of attacks in MitM.Among all most of the authentication protocols concentrated on eavesdropping attack and sensor capture attack.

C. DOS ATTACK
During denial of service (DoS), attack an advisory denies a service from a server, network to an authorized user by creating a large number of requests to the server at a time.DoS attack is quite common in IoT based network, where an unauthorized user sends thousands of requests to the authentication server to shut down the operation temporarily.To contrast, de Almeida et al. [48] developed a method to defense Dos attack in a network by providing packet-level authentication.In addition, a distributed denial of service attack is an advanced DoS attack where DoS is performed in a distributed manner.To execute DDoS attacks, attackers use a huge network of botnets to put down the service of a network.As a result, genuine users cannot access the service from a particular network.Consequently, Liu et al. [78] developed an enhanced distributed low-rate attack mitigation mechanism for IoT networks.
On the other hand, as per FIGURE 6, the DNS amplification attack takes advantage of DNS behavior in order to amplify the attack.A DNS server holds the public IP addresses and their accompanying hostnames.Therefore, the DNS resolver requests the IP of a hostname to the DNS server.If the server does not contain the information, it refers to one of the root DNS servers, which refers to another DNS server to provide the IP, which boosts this attack.
During the DNS amplification attack, attacker spoof the IP of victims IP send a request to provide DNS list to the server.Because of spoofing all replies go to the victim's system and the attacker can amplify the attack up to 100%.To protect the DNS amplification attack, IoT needs more research.Similarly, Flooding is a type of Denial of Service attack that is aimed to put a server or network down by flooding it with a huge number of traffics.Syn-flood is one of the most hazardous in IoT network, where the IoT network and application server become so weighted down initiating incomplete connection request that it no longer process a genuine request from the authenticated node.Additionally, when a node tries to communicate via exchanging common interest information, various attacks take place to capture the transmitted data without a proper certificate.After that, the captured data can be rejected and prevent from further travel to the destination.This type of attack is known as  the Rejection attack.Alternatively, RFID related authentication systems use a backend database to authenticate the user.Therefore, the attacker performs a desynchronization attack to block the communication between the RFID reader and backend database server so that the tag's key stored in the database and the tag's memory mismatches and denies access.A way of rejection attack has been demonstrated in FIGURE 7.Moreover, an illustration of IoT authentication protocols, which protect from DoS attacks, is shown in TABLE 9.

D. FORGING ATTACK
Forging attack allows an attacker to steal authentication information of a genuine user in a network and use the information as an authenticated user to gain access An attacker can behave like a normal user if he/ she gets or intends to get authenticated data from a process of authentication in the IoT network.In a different phase, the adversary may use prediction to the different messages to gain access to the user's data or the network, which is known as user forgery attack.To protect the IoT systems from user forgery attack Wu et al. [28] invented an effective authentication protocol.On the other hand, sensor plays a major role in IoT authentication, as all the authentication messages pass through any of the sensors.Due to a lack of computation and storage capacity, IoT authentication protocols use simple and robust encryption and decryption techniques, which make attackers to open the gate for sensor forging attack.In WSN, hackers use malicious scripts to get access to the authentication process data from the sensor and after modifying them pass to victims as the original message.In between, there is a chance to grab the information from the authentication request if the message is not properly encrypted.To highlight, Wu et al. [28] describe how their protocol protects sensor forgery attack.Unlike sensor forgery, if the protocol has breached, antagonists also can forge the gateway node.In this situation, adversary takes advantage over GWN, and then mitigates authentication requests, after that gets users and network information and finally morph existing data.Wu et al. [28] proposes a protocol that protects the IoT network from gateway forgery attack by spreading important information in different messages.On the other hand, in the Sybil attack, a malicious node possesses multiple identifications in order to establish communication in an IoT network, which could be achieved by disabling or forging legitimate nodes in the network.In this attack, a single node or device can harm multiple devices from a different network.Suryani et al. [83] claimed that their protocol prevents Sybil attack during authentication using two-phase security protection.FIGURE 9 gives a clear view of the Sybil attack.Alternatively, in a replay attack, an attacker intercepts and acquires the data send by the sender and send it to the destination as an original sender.However, timestamp and sequence number with the packet can be implemented to prevent the replay attack.Moreover, there is various information pass through the IoT network and during authentication, the node transfers its user id and password to the other node or authentication server.Therefore, that information can be captured and used to authenticate an intruder.In contrast, to take advantage of a replay attack, attackers need to access the raw network data and that is possible either via network tap, ARP poisoning or  via installing the malware in the victim's computer.A replay attack is further illustrated in FIGURE 8.
IoT devices are heterogeneous in nature and dynamic in behavior.Moreover, IoT devices can be static and mobile.Therefore, the system tries to detect if the authentication distance and access distance of the devices fluctuates.Sometimes, adversaries try to increase the success probability of attacks by changing the distance between devices.This phenomenon is known as changing distance attacks.Chen et al. [84] introduces a fingerprint-based authentication protocol to prevent changing distance attack.On the contrary, an amount of IoT systems are required to tag with formalized attributes to authenticate the activities of the auditor.For example, Health Social Networks (HSN) use attribute tagging widely.Where the attribute-oriented authentication scheme empowers to generate an HSN attribute for every HSN user to protect from attribute tracking attack.In 2012, Liang et al. [90] proposed attribute security in HSN.After that, Uddin et al. [86] proposed tier-based health architecture in a patient-centric agent to monitor patient health.Comparably, a user sends authentication requests to the network to participate.In return, the authentication server exchanges several other packets with the user to reply, acknowledge or response.Therefore, attackers take advantage of those packets by sending a successive packet so that the previous packet, which was sent by the original sender, will discard and the attacker can intrude into the system.Lu et al. [87] claimed that their privacy preservation protocol successfully prevents successive-response attack on IoV network.Nevertheless, in the collusion attack, the execution of operation can combine, manipulate and produce a completely new dataset, especially files, to disguise the server.However, spoofing multiple packets from various user's authentication information and create a set of new authentication packets can also count as a collusion attack.Nevertheless, unnecessary and irrelevant packets send to the enormous number of users through the internet just to fulfill phishing or spreading malware.On the other hand, the main target of spam attacks is to introduce viruses, worm, spyware, Trojan horse to various legitimate systems.It spread through email by some offensive link, website, or the web content as well as without proper sender mail id.Paavolainen et al. [88] converse about various risks on blockchain in IoT by spam attack.Likewise, open redirect abuse is not much popular in IoT but it can cause a problem on security as it redirects to malicious content instead of the actual one.However, detecting IP and protocol creating spam can reduce the possibilities of redirection attacks.
Likewise, internet attacks are classified into different categories.Among them, when an adversary attacks and gains full access and control to the target model are known as a white-box attack.In contrast, while performing a black-box attack, the hacker does not have any idea about explicit knowledge but can design queries to achieve corresponding desire [89].
Additionally, cybercriminals attacking the IoT networks will be driven by the financial gain as the black market [15] for malware and the dark web continue to mature.During the authentication process, users, sensors and GWNs or servers send data among themselves to authenticate, authorize for registration or login.During that period, an intruder can attack the network and if the protocol is soft enough to penetrate, he/ she artifices message and uses as per his/ her requirement.Morphing user's data and intentionally passing wrong messages are common behavior for intruders.In spite of detecting the forgery attacks, prevention is very important.Therefore, researchers developed a protocol mechanism to protect the IoT network from various attacks.TABLE 11 describes most of the effective work by a few years to prevent a forgery attack in IoT authentication.

E. GUESSING ATTACK
IoT authentication server stores authentication information of users and different peripherals in IoT network, such as device id, user id, device secret key, user password.Adversaries try to get those credentials to access the system.If they have direct access to the server then they can extract passwords from the server, but if they cannot get those physically, then attackers try to guess the password to authenticate themselves as a valid user.This is known as a guessing attack.Description of all possible guessing attacks are discussed in TABLE 12. guessing attacks can be done using a dictionary attack or brute force attack.Wu et al. [28] proposed an authentication scheme for multi gateway WSN.Additionally, to authorize in a network, the attacker tries a plethora of possibilities.Consequently, in the chosen-plaintext attack (CPA) cryptanalysis process, adversary guesses plain text and encrypt with known possible encryption techniques to obtain the corresponding cipher text.Duan et al. [91] proposed a policy privacy solution by two-layer cooperating method for protecting IoT.Additionally, devices need several ways to authenticate IoT peripherals in a network, such as a password, smart card or biometric.Among them, the password system is popular but vulnerable.It can easily be victimized by attacks like brute-force, where attackers use software to guess the password to be authenticated.Random password and common password guessing are most effective among all other possibilities.Wang et al. [81] discussed how an attacker cracks IoT device user account by trespassing SMS authentication code using a Brute-force attack.On the other hand, attackers may perform eavesdropping on an authentication process or penetrate to the network to steal authentication code using a Brute-force attack.Similarly, attacker may perform eavesdropping on an authentication process or penetrate to the network to steal valuable user information or files to use them against a legal user of his/ her choosing.Therefore, if the user cracks the encryption process of the message, he/ she can try an offline-guessing attack on credentials.In an online password-guessing scenario, an attacker tries to guess a password by logging to the system.However, online password guessing is less powerful than offline password guessing since the attacker hasa maximum limit of tries.Alternatively, offline guessing can be performed without logging into the actual system and there is no such limitation exists.In different circumstances, social networking is an attack vector that relies based on human interaction.On which, a perpetrator tries to get common and social information of a victim to intercept him by guessing or predicting credentials.Baiting, Scareware, Pretexting, phishing, Spear phishing are common social networking attacks in IoS [95].Harwood [96] describes the way to defend internet attacks on the web and

F. PHYSICAL ATTACKS
IoT devices situate as scattered in the network.These devices can be accessed physically if there are no physical securities.Moreover, there can be thousands of IoT devices; therefore, it is not possible to protect them from physical attacks.However, physical attacks are not only held on static devices, which can be easily tracked, but also in mobile devices, which are difficult to trace.Physical attacks can be occurred by mobile devices loss attack, stolen card attack, stolen device attack, USB attack same-type-device attack.Additionally,TABLE 14 illustrates the description of all possible physical attacks on IoT authentication.
To illustrate, mobile devices are the backbone of any IoT network.During authentication, the user often uses mobile devices instead of static devices.However, to get access to the network, they have to pass through any sensor that is connected to the GWN or any authentication server.Now if a Furthermore, an innocent computer user may not know the severity of USB devices in their system.On which, USB peripherals can carry malicious script to steal information or take advantage of the system.USB attacks have various ways to gain access to the architecture as discussed well by Nissim et al. [99], such as a keyboard, flash drive, mouse, and data cable.In contrast, a mobile device transmits data to the receiver during authentication.However, protocols may not be designed to authenticate the sender device identity, which can provoke the same-type-device attack.An attacker can acquire a device, which is the same as the same manufacturer and same brand as of the legitimate transmitter, and he/ she pretends the transmitter by sending the same signal as the real transmitter.A lightweight acoustic fingerprint-based wireless device authentication protocol is designed by Chen et al. [84] to protect the network from the same-type-device attack.TABLE 15 lists the existing IoT authentication protocols supports against physical attacks.

G. ROUTING ATTACK
Non-legitimate node forwards data packets to the improper destination, which is known as routing attack.Classification of routing attacks is shown in FIGURE 3, and the description of all possible routing attacks are listed in TABLE 16.This type of attack approaches in two different ways either via changing the final destination address of the data packet or via sending the data packet to the wrong next hop in the routing path.In 2017, Ma et al. [101] proposed an M-RPL protocol to protect lightweight IPv6 routing protocol by creating hierarchical clustering network topology and providing alter path from different clusters to a route if the network is compromised.In the routing attack, a sinkhole attack is a type of selective forwarding attack [102].In a wireless sensor network, all the data collected by the sensor nodes are forwarded to the sink node to process, therefore, the sink node is very important for the lifetime of the WSN.However, an adversary node can act as a sink node and tamper all the data in a WSN, which makes the network in jeopardy.This node can be as dangerous as attract neighboring nodes.Moreover, in the sinkhole attack, the attacker node convinces the neighboring nodes to get the traffic from them and then digest all packets.Sinkhole attack can open the path for wormhole attack.Sinkhole attack is illustrated in FIGURE 11.Similarly, a wormhole attack is considered as a serious attack in a wireless sensor network.There are two major components in the wormhole attack, i.e. several spiteful nodes and tunnels.In addition, the wormhole node creates a false route, which is shorter than the original route in an IoT network and misleads the distance between nodes i.e. routing mechanism.Then, the malicious node occupies the packets from one location and transfers them to the distant situated node by a tunnel (either by in-band or out-band channel) which further distributed locally.However, attackers can perform wormhole attack without informing any authenticated nodes or mechanism.Furthermore, the wormhole attack launches various other attacks like selective dropping, eavesdropping, and replay attacks, which affect data traffic flow.Amish and Vaghela [103] introduced wormhole attack detection and prevention mechanism in WSN using Ad hoc on-demand Multipath Distance Vector (AOMDV) routing protocol.Wormhole attack is shown in FIGURE 12. On the other hand, in the Black hole attack, malicious nodes send a route reply message to the sender in return receive packets from sender node and discard packets instead of forwarding to the destination node.In their research, Motamedi et al. [107] show the detection procedure of Black hole attack in WSN using unmanned aerial vehicles (UAVs).Furthermore, Bansal et al. [104] discussed the anomaly-based detection on leach protocol in WSN and Kaur and Singh [105] presented a way to identify and mitigate the Black hole attack in WSN.Nevertheless, various data establish links with the original content of a transmitted data over IoT network.For example, clouds may contain user details and time of purchase of a particular good in a departmental store or can be data of patients associated with his/ her disease information.Therefore, an adversary can attack a network to reduce linked information from a session of transaction.This is called the link-ability attack.To prevent link-ability attack in a vehicular network, Memon et al. [106] introduces pseudonyms changing strategies.Additionally, TABLE 17 shows existing IoT authentication protocols fight against routing protocols.

III. TECHNICAL METHODS OF AUTHENTICATION MECHANISM IN IoT
As authentication is a process of validating users and components identity, so that the authorization process can provide access to the network or an information system, which should be highly secured from vulnerable threats.Therefore, thousands of authentication protocols are designed by the researchers to protect the IoT network from illegal users.However, designing and protection mechanism of authentications are different in different protocols.On the other hand, authentication protocols in IoT cannot cope up with the traditional authentication mechanism because of its limitations.As a result, IoT authentication schemes use password-based authentication, token-based authentication, biometric authentication, cryptographic authentication, and multi-factor authentication.Additionally, FIGURE 13 shows the well-structured taxonomy of technical methods of an authentication mechanism in IoT.
Password-based authentication is a very common and useful method to verify a user or device.In which, users need to provide a unique id and a word combing of letters, digits and/ or special characters, known as a password.The unique id and password combination are reserved in the database in an authentication server or as low level as in sensors' memory.When a user supplies the combination of user id and password, protocol matches the provided combination with saved credentials and if these matches then, the protocol allows the expected user or device to perform the desired action.A password can be a combination of different patterns or simple words.However, protocols use strong rules for a  password to avoid the password guessing attack.In addition, FIGURE 14 shows the password authentication process.
Nevertheless, a token is a piece of data created by the authentication server to uniquely identify a user or device.Token-based authentication [43] can be further classified as soft and hard token-based authentication.In a soft token-based authentication scenario, the server populates a one-time password (OTP) and sends it to the registered communication media, which is associated with the account and preserves a copy of the transferred OTP.After a while, the server matches the user-provided OTP with the stored one and takes decision on authentication.Furthermore, to make the process more secure, protocols implement associative rules like the expiration of OTP, length of OTP and type of OTP.On the other hand, a small device or card containing a piece of information to verify itself in a tokenization system is called hard token-based authentication.Additionally, this system works on a mechanism where every request to a server will response based on the correct combination of tokens.Furthermore, token-based authentication is wellaccepted methods because of its easiness of transmission via query strings, header attributes and the body of a POST request.Moreover, hard token authentication can be achieved using different methods like dongle [90], smart card [34] and RFID chip [29].Additionally, the process of authentication via smart card has been shown in FIGURE 15.
On the other hand, biometric authentication [30], [32], [34], [40], [42], [68] is based on the biological character of humans.Additionally, the specific biometric scanner collects unique biological data from a user and matches the stored data, which was collected via the registration process.Moreover, biometric uniqueness can be provided in different ways.These methods include fingerprint authentication, face authentication, iris authentication, retina authentication, hand authentication, and voice authentication.Iris authentication methods use mathematical pattern recognition to identify the pattern of one or both the irises, which is unique for an individual.Likewise, fingerprint authentication is common in IoT mechanism, where friction ridges of a human finger are checked with pre-reserved of the same information in a server.Similarly, other biometric authentication uses its unique feature to differentiate individuals.FIGURE 16 illustrates the biometric authentication process.In contrast, cryptographic authentication methods use to encrypt and decrypt techniques to morph actual messages during communication in an insecure network.Additionally, not only to protect variable data in the algorithm but also researchers apply cryptography to protect peripheral authentication values like biometric information, token, password, user id, smart card information during the process of authentication.Furthermore, it is a common practice to use hash and XOR techniques [24], [29], [30], [32], [35], [40] in authentication because IoT devices are tiny and computational-constrained. On the other hand, Fouda et al. [108] use Diffie-Hellman along with Hashbased Message Authentication Code (HMAC) technique.In contrast, Mahmood et al. [66] criticize Fouda et al. [108] protocol and reuse the Diffie-Hellman technique to their authentication protocol and implement RSA and AES algorithm to generate the session key.On the other hand, because of strong cryptanalysis and constant breaking strategies of the lightweight cryptographic algorithm, developers incorporate elliptic curve cryptography (ECC) [31], [41], [69], [71], which is popular and provided a strong mechanism in IoT authentication, especially in key agreement mechanism.Similarly, developers use techniques like bio hashing [40], fuzzy bit commitment [32] to protect biometric information during authentication.
Alternatively, because of multi-factor authentication's versatility, it gets the most attraction to researchers to make authentication protocol secure and strong.To elaborate, multi-factor authentication consists of two or more ways to identify an independent identity.It may include any combination of password-based authentication, biometric authentication, cryptography authentication or tokenbased authentication.For example, Srinivas et al. [40], Kang et al. [30] and Li et al. [32] use hash, XOR and biometric together in their protocol.Similarly, Hossain and Hasan [109], Wu, et al. [69] and Li et al. [31] employ hash, XOR and ECC.Likewise, all modern protocols apply two or more technical methods to produce effective authentication protocol.
The proliferation pace and diversity of IoT devices in the network make the authentication mechanism more demanding.To address this situation, several methods have introduced in the IoT authentication mechanism.Furthermore, researchers are developing authentication protocols to make more powerful by combining inter-domain techniques.To illustrate further, TABLE 18 lists the technical methods used by current IoT authentication protocols including outcomes.

Researchers use security verification techniques to test the performance of an authentication technique in IoT.
There are few security verification techniques available for performance testing, which is shown in FIGURE 17, namely, Automated Validation of Internet Security Protocols and Application (AVISPA), BAN-logic, Game Theory, Analysis by process (Spi calculus), Automated reasoning (ProVerif),Random Oracle Model (ROM), and Real-or-Random (ROR).Furthermore, TABLE 19 illustrates the security verification techniques used by recent IoT authentication protocols.

A. AUTOMATED VALIDATION OF INTERNET SECURITY PROTOCOLS AND APPLICATION (AVISPA)
AVISPA is an automated validation and security analysis tool for network and cryptographic protocol.A number of IoT authentication researchers use AVISPA tools to confirm security attributes.Farash et al. [26] proposed user authentication and key agreement scheme and used AVISPA to confirm security properties.Furthermore, Amin et al. [17] used AVISPA to ensure the safety mechanism of their proposed protocol.AVISPA is available from Information Society Technology [111].

B. BAN-LOGIC
Authentication protocols are important to be examined properly if their working principles are logically correct because authentication protocols are the backbone of security in many IoT networks.To fulfill this requirement, Burrows et al. [112] proposed BAN logic, which ensures if the exchanging information over media is trustworthy or not.Furthermore, BAN logic follows a sequence of three steps, and these are (I) verification of message origin, (II) verification of message freshness and (III) verification of message trustworthiness.Amin et al. [17], Farash et al. [26], Li et al. [32], and Jiang et al. [18] used BAN to logically proof the authentication on their proposed work.He, et al. [113] used BAN logic to show if the proposed scheme is valid and practical.Kang et al. [30] used the BAN to validate the generated session key between user and server.

C. GAME THEORY
Game theory is the strategic interaction between rational decision-makers.It has been widely used in IoT component authentication to sanguine security.Chang and Le [24] uses a sequence of games under the decisional Diffie-Hellman (ECDDH) problem with a view to proving that the protocol supplies secure and perfect forward secrecy authentication by Ferrag et al. [67].

D. SPI CALCULUS
Spi calculus is an extension of pi-calculus developed for describing and analyzing cryptographic protocols [114].
A detailed discussion about Spi calculus has been done by Abadi and Gordon [114].The authenticity property and the secrecy property has been proved via the session key establishment protocol by Dolev, et al. [25].

E. PROVERIF
Blanchet et al. [115] developed the Proverif tool, which is for automated reasoning about the security properties found in cryptographic protocols.Wu et al. [69] use Proverif to list the formal verification process in their protocol.Roy, et al. [34] uses the formal security verifier proverif1.93 to show the security of the presented scheme.

F. RANDOM ORACLE
Random oracle is a random function, which response to every unique query with a random response chosen uniformly from its output domain.It is a mathematical function and always choose the fixed random response from its output domain for each repeated unique query.Random oracle can be represented using equation 1.
Let, D is a domain and R is a range.Therefore, Random Oracle is a randomly chosen function such that among all functions in domain D and range R are chosen randomly.As it is a function, every time if the same input is given to the Random oracle, the same output needs to be returned.The way to think about the random oracle is that it can be considered as a lookup table like FIGURE 19.Such that, one column represents the input and another one represents output, so for each input, a randomly chosen output will be stored in the table.Whenever random oracle needs to compute over some input x, then from the table it can return correspond y where RO(.) is Random Oracle function, and x, y are input and output respectively.This would have been an ideal random oracle, but representing such a table requires exponential space.If the domain D is an n bit value, then there will be 2n rows in FIGURE 19.Instead of this, the random oracle can be simulated using it in a randomized way.

G. REAL-OR-RANDOM MODEL (ROR)
ROR is the two-party authentication key exchange protocol.In this model, an adversary can ask Execute, Send and Test queries.Furthermore, the adversary can ask as many as Test queries to differentiate instances.However, all the Test queries will be answered using the hidden bit, which is the same for all instances and chosen at the beginning.That means the keys returns by the Test oracle are all real or random.Moreover, the same random value will be returned for Test queries from two collaborated instances.However, the motivation of the adversary is to guess the random bit to answer the Test queries and they succeed, if they guess correctly.The Real-or-random model by Abdalla, et al. [116] is widely used to process formal security analysis in the research.
From the observation, it is identified that recent protocols prefer BAN-logic and AVISPA for security verification.The main strength of BAN-logic is its simplicity and its usefulness.Therefore, the authors trust this technique more.Moreover, the formulation of BAN is easy to cope up with the authentication protocols and the structure of it is more convenient for this type of verification.On the other AVISPA is also a popular tool among researchers.AVISPA shows the result and analysis of the protocols in detail and in a fruitful manner, which attracts researchers to choose AVISPA to test their scheme.In contrast, ROM, ProVerif and Game Theory have moderate use to verify authentication techniques compare to ROR and Spi Calculus.

V. IoT AUTHENTICATION EVALUATION TECHNIQUES
As the new and challenging authentication techniques are necessary to protect the IoT environment from various emerging attacks, evaluation of those proposed schemes are equally important to check their effectiveness.In this section, we discuss several evaluation techniques with their parameters and supporting equations.

A. AVERAGE RESPONSE TIME
Response time is assumed to be the time taken by the server or GWN to result in the response of a request to the client.This can be affected by few factors, such as server configuration, number of users, network bandwidth, number of request, type of requests and think time.
First response can be executed by the time of client request and time of first response, which is described in equation 3.
T res = t res − t req (3) Here T res , t res , t req are response time, time of client request and time of first response respectively.
Average response time is calculated by the mean of all response time, which is demonstrated in equation 4.
T ang_res = n r − T think (4) where T ang_res is the average response time, n is the number of concurrent users.r is the number of requests per second the server receives.T think is the average think time (in seconds).However, to obtain an accurate response time result, a user should always include think time in the equation.

B. HANDSHAKE DURATION
Handshaking is the process of negotiation between two network parties in the IoT network.These parties can be user, sensor, actuator, server or other nodes.As shown in FIGURE 20, handshaking takes place by completing the two-roundtrip message, whereas, client's discovery offers by the server and again the client's request acknowledges by the server.Duration to a handshake T hs is computed at the client-end using equation 5.
T hs = T s + T res + T p (5) where T s is the time taken by whole session request, T res is client response time and T p denotes as processing time at the server.However, to calculate the handshake duration, a user must perform several random numbers of handshakes between the client and the server.After that, the user should perform a standard deviation to observe the variability and accuracy among the examined data.Standard deviation can be performed using equation 6.
where σ denotes standard deviation.N is the number of samples, i is the number of iteration, x i is the handshake duration and µ is mean value.

C. AVERAGE MEMORY CONSUMPTION
IoT is a mostly wireless sensor network, which is constrained by its low memory.Consequently, memory consumption is important in specialized and autonomous sensor networks.However, memory consumption depends on the various level in IoT, such as user level, sensor level, GWN level or server level.A comparison of memory consumption in various authentication protocols is discussed in TABLE 20.

D. END-TO-END DELAY
End-to-End Delay or E2ED denotes the average time to deliver packets from sender to receiver.E2ED can be calculated by using equation 7.
Here, i is the number of packets and n is the number of received packets, while T r i is the received and T s i denotes the sent timestamp for ith packet.E2ED is proportional to the number of sensors in the IoT network.Therefore, an increased number of nodes put up the congestion in the network.

E. IMPACT ON THROUGHPUT
Throughput can be described as the amount of data passes through a system in a unit of time.In the IoT network, the total number of transmitted data conserved in a second to calculate throughput.Throughput can be expressed as equation 8.

TP =
Q r i xl i T w (8) Here, TP denotes throughput, while Q r i is Quantity and l i is the length of the ith kind, and T w denotes as the whole time of the simulation.

F. PACKET DELIVERY RATIO
Packet Delivery Ratio is calculated based on the number of packets sent by the sender and the number of packets successfully received at the receiver end.However, it depends on several factors like network configuration, device capabilities, bandwidth; therefore, it is difficult to test the network performance.Equation 9 can be used to calculate the Packet Delivery Ratio.

PDR =
N rp N sp (9) where PDR is Packet Delivery Ratio; N sp is the total number of sent packets, and N rp is the total number of received packets.It has been identified that throughput falls when the number of nodes increases in a network.In the WSN, packet-sending circumstances are defined in the energy model, like that, energy is consumed when a packet is sent over the network.Therefore, more packet transfer cost core energy consumption.Ultimately, the packet can be discarded due to less energy or long-distance travel.

G. COMMUNICATION COST
As mention in FIGURE 1

H. COMPUTATION COST
In the IoT network, computation also depends on the kind of protocols.As most of the network devices have computation constraints, the heavyweight computation cannot be performed in IoT networks.Therefore, protocol developers always try to create lightweight authentication protocols for IoT networks.Therefore, many of the researchers have adopted the concept of hash, XOR and concatenation to secure the message to pass through the network.ECC, MOD, Fuzzy commitments are also implemented in the IoT authentication mechanism.TABLE 22 describes the notation used to calculate the computational cost and comparison analysis is demonstrated in TABLE 24.

I. STORAGE COST/MEMORY COST
To establish IoT authentication, protocols use different types of mechanisms.Among them, the smart card is one of the popular techniques.A smart card needs some storage capabilities because it stores user credentials, sensors, and GWN information.Different protocols use different operations to achieve authentication.TABLE 23 shows the comparisons of storage cost among different IoT authentication protocols.

J. ENERGY COST
IoT components are subject to power constraints.Therefore, the energy consumption of a protocol is equally important with other factors.Energy is proportional to power.Nevertheless, if a protocol consumes more energy, battery drainage will happen more quickly.Furthermore, the energy cost of transmitting and receiving data can be calculated based on equation 10 and equation 11 consecutively [131], [132].
where E Tx (k, d) is the energy consumption of transmitting data.k is the transmitted data volume (bit), d is the distance between two objects, Eelec is the energy consumption of data transmission in terms of nJ/bit.∈ amp is the energy consumption constant used to expand radio coverage in terms of nJ/(bit * m2).
Therefore, the energy cost of transmitting data between two objects is proportional to the distance between them.In addition, the energy cost of receiving data is shown below.
where E Rx (k) is the energy consumption for receiving data.

VI. OPEN ISSUES AND FUTURE DIRECTIONS
A vast range of encryption techniques is used in IoT authentication schemes, which include, hash, XOR, ECC.To make

A. DETECTION OF ATTACK
It is obvious that secure access to the information in a network is the prime concern in the application layer in IoT.However, if the system is incapable to deliver the demanded service, it is of no use.In addition, attacks are used to reduce the ability of a network to communicate with its legitimate resources.Attacks are dangerous threats as it cripples the network by repelling unnecessary traffic in a network or forge the traffic to disconnect the communication.In case attacks are frequent in IoT authentication, it renders the server partially or completely unavailable to provide any service.recent attacks for IoT Therefore, sensing attacks in authentication is important, because sensors are the soft targets the

B. TIME BASED
On the other hand, timely respond to the sender is important as authentication on time.Additionally, protocols use timestamp and session keys to protect from attacks.However, they are vulnerable to new attacks.Therefore, IoT authentication may concentrate on hierarchical and distributed approaches that consider timing.

C. TECHNOLOGY AND STANDARD
A vast technology and communication standards are used in the IoT network.different technology and standards are still missing in IoT to ensure access confidentiality, privacy and security among users Moreover, this unable to cope with the defined protection constraints, which in return ensures trustworthiness among users and devices; provides secuof using the IoT authentication service on the public network.

D. STRONG AUTHENTICATION PROTOCOLS
As the number of attacks is huge in IoT networks especially during the first stage of network access that is authentication mitigating attacks and clustering, different network packets by its behavior are important to increase the throughput of a system.Therefore, detecting external as well as internal attacks are challenging.However, scopes are there to design a strong authentication protocol to prevent and protect the IoT network from all potential attacks.

E. FORMAL WAY OF SECURITY AUTHENTICATION
Although authentication protocols use different evaluation techniques to confirm the security of protocols for IoT networks, there is a need for a formal way to define security aspects of authentication in IoT.

F. CONSISTENT NETWORK
IoT authentication mechanism takes place over both secure and insecure networks, where the login and authentication phase may perform in an insecure network but the registration phase must undergo through a secure network.However, this provision may not be available everywhere, which may invite attackers to enter into the system.Therefore, it is desirable to consider the registration phase under an insecure network.

G. OFFLINE ACCESS
Sensor components are the backbone of the IoT network, as they distribute the network and store confidential communication data in it.As per the networks need, sensor nodes have been deployed in various challenging environments like a battlefield, agriculture, or in natural calamities like a forest fire, tsunami, earthquake detection or areas nuclear threats.However, a situation may arise where the sensor node disconnects from GWN or AS due to network linkage error [137].This scenario may turn to be difficult for many reasons; users may need to access isolated sensor nodes immediately to get valuable information for decision-making or those isolated sensor nodes may loss crucial data stored in it due to low power capacities.Therefore, authentication protocols need to validate legitimate users on isolated sensors.However, this new mechanism should not encourage adversaries to penetrate the sensor and grab information, which also needs to be under consideration.

H. NEW TECHNIQUES IN AUTHENTICATION
To make the authentication protocol lightweight, researchers use XOR and hash functions.Moreover, protocols use different commitments to ensure the security of authenticated data.However, there is a lot of scopes to introduce new and different techniques for authenticating data.Quantum computing, quantum bit commitment, and quantum cryptography are the open challenges to introduce in IoT authentication.

I. ANONYMITY
As the share of data is huge, anonymization becomes an important factor in IoT.An adversary can attack an IoT network to get users' detail, which may reveal confidential information i.e. health records.On the other hand, a hacker can track the position of a user or an object and can perform harm to them or their property especially on VANET.Therefore, future research should focus on data anonymization while improving the authentication mechanism for IoT, which demotes traceability.
Additionally, the WSN application layer is in the developing stage.However, plethora of attacks are at different levels, but we need more attention to this level to encourage the researcher to implementing a well-constructed and robust lightweight application for IoT authentication.

VII. CONCLUSION
The current concept of network and connectivity is going to be changed in the next few years.As it is predicted that the number of connected devices in the world will take over the headcount of human beings soon, which can be possible because of the expansion of the Internet of Things.However, security on IoT is still searching for its way to improve so that it can provide reliability and protection against threats.Again, authentication is one of the main important parts in security, because it is the gateway of a user or device to introduce in a network.In addition, a slew of authentication protocols are designed, broken, and again redesigned to protect the network from attacks.Therefore, this paper shows the potential threats in IoT authentication and existing protocols to protect them.
To our best knowledge, there is no research conducted similar to us as of now.All of the other research work focus on IoT security, authentication protocols, and attack models on IoT.We believe that our study will benefit readers to get knowledge about a huge range of attacks and methods in IoT authentication and help the upcoming researcher to formulate their proposal to create strong IoT authentication protocol to serve better to end-users.
APPENDIX TABLE 25 and TABLE 26 represent the lists of figures and the lists of tables use in this document respectively.

FIGURE 2 .
FIGURE 2. Organization of the document.

FIGURE 3 .
FIGURE 3. Taxonomy of attacks on IoT authentication.

FIGURE 5 .
FIGURE 5. Man-in-the-middle attack in IoT network during authentication.(a) User and sensor scenario and (b) sensor and GWN scenario.

FIGURE 13 .
FIGURE 13.Technical methods of the authentication mechanism in IoT.

Algorithm 1
Random Oracle 1: Initialize database DB 2: input ← x 3: if (x, y) exists in DB then 4: return y 5: else 6: Choose y from range R 7: Add (x,y) pair in DB 8: return y 9: end if value using equation 2. Furthermore, algorithm 1 represents the random oracle.

TABLE 1 .
Description of authentication model of iot network.

TABLE 2 .
Acronyms and its definition.

TABLE 3 .
Description of attacks on IoT authentication.

TABLE 4 .
Description of different types of masquerade attacks.

TABLE 5 .
IoT authentication protocols against masquerade attack.

TABLE 6 .
Description of different types of man-in-the-middle attacks.

TABLE 7 .
IoT authentication protocols to protect against MITM attacks.

TABLE 8 .
Description of different types of DOS attacks.

TABLE 9 .
IoT authentication protocols to protect against DOS attacks.

TABLE 10 .
Description of different types of forging attacks.

TABLE 12 .
Description of different types of guessing attacks.

TABLE 13 .
IoT authentication protocols to protect against guessing attacks.

TABLE 14 .
Description of different types of physical attacks.provided various opportunities in IoT platforms.A lifecycle of social engineering attack shown in FIGURE 10 [97].Additionally, a list of IoT authentication protocols, which support against guessing attack, are represented in TABLE 13.

TABLE 15 .
IoT authentication protocols to protect against physical attacks.

TABLE 16 .
Description of different types of routing attacks.

TABLE 17 .
IoT authentication protocols to protect against routing attacks.

TABLE 19 .
(Continued.) Security verification techniques used by IoT authentication protocols.

TABLE 20 .
Comparison of average memory consumption.

TABLE 21 .
However, we should also take care of communication cost on behalf of standards, because different standards have different threshold values to transmit.Consequently, the IEEE 802.15.4 communication standard supports 127 bytes, whilst the IEEE 802.15.6 standard has a maximum message frame length of 255 bytes.

TABLE 21 .
Comparison of communication cost.

TABLE 22 .
Notation used to calculate computational cost.
a more secure authentication mechanism, protocols use the smart card and biometric techniques beside user id and password.Moreover, to protect biometric information in the network, it makes use of other schemes like the fuzzy extractor, fuzzy commitment, bio hashing.However, the ultimate goal of any newly designed authentication protocol is to make

TABLE 23 .
Smart card storage cost of IoT authentication protocols.CH * denotes cluster head.lightweight (low computation and storage cost) and to protect from known common attacks, by considering the factor of low computational power and low memory space of IoT peripherals.As IoT based network is under development stage, it needs more supervision.Therefore, some key issues and future challenges are discussed in the subsequent parts.

TABLE 24 .
Comparison of computational cost.

TABLE 25 .
Lists of figures in the document.

TABLE 26 .
Lists of tables in the document.