1. Introduction

The Internet of Things (IoT) is one of the important technologies of this era that can add automation and smartness in several sectors such as transportation, health care, industries, agriculture, energy, and infotainment. It works by deploying sensors on different devices used in these sectors, hence allowing the measurement of important real-time data. These data are transmitted to the remote servers where it can be analyzed, and intelligent actions can be taken based on it ^{[1][2][3][4][5][6][7][8][9][10][11][12][13][14][15]}.

IoT enables many important applications, including intelligent traffic management, safety-aware autonomous driving, saving electricity usage using smart grids, remote patient monitoring, machine health monitoring, smart industrial automation, and smart home security solutions ^[4][5]. In the era of Industry 4.0 and 6G communications, IoT applications will revolutionize how different industries operate.

The successful working of IoT applications relies on reliable data transmission between sensors and servers. Reliability refers to robust communication with a high packet delivery ratio, low latency, and defense against network attacks. Each IoT application may have different Quality of Service (QoS) requirements. To realize a reliable and robust IoT network, meeting the QoS requirements is needed ^{[16][17][18][19][20][21][22][23][24][25]}.

2. Reliable Data Transmission in IoT Network

Figure 1

presented three major components of reliable IoT data dissemination. These include resource allocation, latency management, and security.

Figure 1.

 Key components of Reliable Data Transmission in IoT.

Efficient resource allocation is important for reliably sharing data between IoT nodes and servers. As the spectrum resource is limited due to large data generated by IoT nodes, it is important to propose intelligent spectrum utilization techniques. Techniques such as cognitive spectrum management could be used for sharing of spectrum bands by several IoT nodes. Transmit power is also an important resource that needs to be allocated carefully. To save energy of IoT nodes so that they can stay active for longer periods without charging, adaptive transmit power techniques are required.

Fog computing is a vital part of future IoT networks. Fog nodes located near the IoT devices provide storage and computational capacity to the IoT network. IoT networks can place popular and most useful content in the cache storage of these fog nodes. Hence, cache storage allocation is an important challenge. Moreover, IoT nodes may not perform all tasks locally and, therefore, will offload many tasks to these fog nodes. In this regard, optimal task offloading algorithms are needed to ensure that the computational capacity resource of fog nodes is used efficiently.

Latency management is another crucial unit of reliable data transmission in IoT. IoT applications may not work well if data are not shared within the desired latency. Many new applications such as autonomous driving and industrial automation have stringent latency requirements, and hence, latency management is needed. In this regard, smart retransmissions can add diversity and enhance the probability of quick packet reception at the receiver. Moreover, optimal medium access techniques need to be developed to allow IoT nodes quick and fair channel access.

Accurate data traffic prediction can support latency management techniques as knowledge of upcoming traffic at an IoT server, and fog node equips it to handle it better. Hence, Artificial Intelligence (AI) based techniques that forecast the frequency and size of data can be very useful. Besides, other network technologies can support IoT networks to transmit their data quickly. In this regard, Unmanned Aerial Vehicle (UAV) and vehicular network has the potential to act as a relay for data traffic generated by IoT nodes ^{[26][27][28][29][30][31][32][33][34][35]}.

Accurate data traffic prediction can support latency management techniques as knowledge of upcoming traffic at an IoT server, and fog node equips it to handle it better. Hence, Artificial Intelligence (AI) based techniques that forecast the frequency and size of data can be very useful. Besides, other network technologies can support IoT networks to transmit their data quickly. In this regard, Unmanned Aerial Vehicle (UAV) and vehicular network has the potential to act as a relay for data traffic generated by IoT nodes [91,92,93,94,95,96,97,98,99,100].

Security is an essential component of reliable data transmission in IoT. Several attacks may be generated in an IoT network that can compromise the privacy and confidentiality of data transmitted. Moreover, malicious nodes can insert fake and wrong data in the network that can affect the decision-making of IoT applications. To tackle this issue, advanced cryptographic techniques are needed that can ensure the security of transmitted data while keeping the required overhead to a minimum. As an increase in the security overhead can result in higher data latency, tradeoff between security robustness and latency needs to be evaluated.

Blockchain is an upcoming technology that can provide robust security to IoT devices. Hence, blockchain-based solutions need to be developed in the context of IoT applications. Other security techniques such as physical layer security can also improve the reliability of IoT networks. These techniques can work in collaboration with cryptographic techniques to provide a robust solution. Lastly, data integrity attacks and anomaly detection schemes are also required to ensure that correct data are received based on which decisions can be made.

3. Review of Recent Work in Literature on Reliable Data Transmission in IoT

3.1. Resource Allocation

Resource allocation is an important area of research for IoT applications. Since IoT devices are constrained in energy, computation, and transmission, intelligent and novel resource allocation techniques are needed. As shown in 

Figure 2

, resources such as spectrum, the transmission power of IoT nodes, cache storage of IoT enabled fog nodes, computational capacity of IoT nodes and fog nodes, and data rate needs to be carefully allocated. We present a review of recent work related to efficient resource allocation for IoT networks in 

Table 1. In ^[36], the authors consider a Wireless Powered Communication Network (WPCN) based wireless power transfer scenario. The communication between Access Point (AP) and mobile users is the focus of the paper. The work provides an optimal algorithm to allocated resources, including channel selection, transmission time, and transmit power. The joint optimization problem becomes non-convex and is solved using an iterative algorithm. Simulation results show improved sum-rate and reduced convergence time of the proposed algorithm.

. In [114], the authors consider a Wireless Powered Communication Network (WPCN) based wireless power transfer scenario. The communication between Access Point (AP) and mobile users is the focus of the paper. The work provides an optimal algorithm to allocated resources, including channel selection, transmission time, and transmit power. The joint optimization problem becomes non-convex and is solved using an iterative algorithm. Simulation results show improved sum-rate and reduced convergence time of the proposed algorithm.

Figure 2.

 Different categories of recent work in Resource Allocation in IoT.

Table 1.

 Resource allocation in IoT: Recent literature review (S = Spectrum Allocation, P = Transmit Power Allocation, T = Time Slot Allocation, C = Cache Storage Allocation, M = Computational Resource Allocation).

Scenario	S	P	T	C	M	Key Idea	Results

Wireless power transfer [36][114]	✗	✓	✓	✗	✗	AP and mobile user communication Optimal channel selection Optimal time resource allocation Optimal power resource allocation Iterative algorithm for non-convex optimization	Improved sum-rate Reduced convergence time
Edge IoT [37][115]	✗	✗	✗	✗	✓	✗	IoT tasks offloaded to edge servers Q-learning to allocate resources Utility maximized and fairness Online Q-learning scheme	✗	Large state and action space Reduce computation overhead Reduce convergence time	✗	Optimal task offloading decisions Amount of task locally computed Amount of task computed at MEC server Task splitting between MEC servers Optimal transmit power selection Optimal RAT selection for task offloading Maximized application utility Improved fairness
Alternate Convex Search (ACS) algorithm	Reduced energy consumption		Reduced latency	Smart factory [38][116]	✓	✗	✗	✗	✗	IoT enabled machines

The authors in ^[42] considered a narrowband IoT (NB-IoT) scenario where an increased number of retransmissions in the standard improves the packet delivery ratio but increases the energy consumption and latency. The proposed technique computes optimal durations of the control channel and data channel. Moreover, the authors also find out the optimal coverage class for each user where coverage class defines the number of retransmissions allowed. Simulation results highlight reduced energy consumption and reduced latency achieved by the proposed protocol.

The authors in [120] considered a narrowband IoT (NB-IoT) scenario where an increased number of retransmissions in the standard improves the packet delivery ratio but increases the energy consumption and latency. The proposed technique computes optimal durations of the control channel and data channel. Moreover, the authors also find out the optimal coverage class for each user where coverage class defines the number of retransmissions allowed. Simulation results highlight reduced energy consumption and reduced latency achieved by the proposed protocol.

The work in ^[43] considers a scenario where IoT devices use multiple Radio Access Technologies (RATs) to transmit data to the Mobile Edge Computing (MEC) server. The proposed protocol computes optimal task offloading decisions, including the number of tasks locally computed at IoT device, amount of task computed at MEC server, and task splitting between different MEC servers. Besides, optimal transmit power and RAT selection for task offloading are also proposed. The developed optimization problem is solved using the Alternate Convex Search (ACS) algorithm. The results highlight reduced energy consumption as well as reduced latency when the proposed protocol is used.

The work in [121] considers a scenario where IoT devices use multiple Radio Access Technologies (RATs) to transmit data to the Mobile Edge Computing (MEC) server. The proposed protocol computes optimal task offloading decisions, including the number of tasks locally computed at IoT device, amount of task computed at MEC server, and task splitting between different MEC servers. Besides, optimal transmit power and RAT selection for task offloading are also proposed. The developed optimization problem is solved using the Alternate Convex Search (ACS) algorithm. The results highlight reduced energy consumption as well as reduced latency when the proposed protocol is used.

In ^[44], the authors consider a cloud scenario with several Virtual Network Functions (VNFs). VNFs are mapped to different applications such as delay-sensitive, mission-critical, and latency tolerant. The goal of the proposed technique is the efficient allocation of VNFs to reduce latency. The authors use a regression algorithm to predict the demand of each application type and make allocations accordingly. The proposed algorithm works initially on offline training. This is followed by online training afterward.

In [122], the authors consider a cloud scenario with several Virtual Network Functions (VNFs). VNFs are mapped to different applications such as delay-sensitive, mission-critical, and latency tolerant. The goal of the proposed technique is the efficient allocation of VNFs to reduce latency. The authors use a regression algorithm to predict the demand of each application type and make allocations accordingly. The proposed algorithm works initially on offline training. This is followed by online training afterward.

The work in ^[45] considers a terrestrial network scenario composed of IoT-enabled satellites and a MEC server. IoT nodes collect data and send it to satellites or satellite gateways for further processing. As satellites have energy constraints, they offload the IoT data to the gateways for further processing. The proposed work finds an optimal association between a satellite and a gateway using Lagrange multiplier and Deep Reinforcement Learning (DRL) techniquesThe results highlight reduced latency of IoT data at a reduced energy cost.

The work in [123] considers a terrestrial network scenario composed of IoT-enabled satellites and a MEC server. IoT nodes collect data and send it to satellites or satellite gateways for further processing. As satellites have energy constraints, they offload the IoT data to the gateways for further processing. The proposed work finds an optimal association between a satellite and a gateway using Lagrange multiplier and Deep Reinforcement Learning (DRL) techniquesThe results highlight reduced latency of IoT data at a reduced energy cost.

In ^[46], a UAV-assisted MEC-based IoT network scenario is considered. UAV facilitates computing in terms of storage and computation service. UAV can also improve network coverage by providing relaying services to overcome poor channel conditions. The problem is divided into three sub-problems. The first one deals with IoT nodes and UAV association, the second one handles scheduling of communication resources, and the last one manages UAV placement. The proposed algorithm shows low latency as compared to other related techniques.

In [124], a UAV-assisted MEC-based IoT network scenario is considered. UAV facilitates computing in terms of storage and computation service. UAV can also improve network coverage by providing relaying services to overcome poor channel conditions. The problem is divided into three sub-problems. The first one deals with IoT nodes and UAV association, the second one handles scheduling of communication resources, and the last one manages UAV placement. The proposed algorithm shows low latency as compared to other related techniques.

The work in ^[47] considers a semi-blind Non-Orthogonal Multiple Access (NOMA) based IoT scenario. The proposed technique uses an interference Alignment Independent Component Analysis (AICA) as compared to the traditional Successive Interference Cancellation (SIC) techniqueThe results highlight improved signal-to-noise ratio and reduced latency.

The work in [125] considers a semi-blind Non-Orthogonal Multiple Access (NOMA) based IoT scenario. The proposed technique uses an interference Alignment Independent Component Analysis (AICA) as compared to the traditional Successive Interference Cancellation (SIC) techniqueThe results highlight improved signal-to-noise ratio and reduced latency.

3.3. Security

Security is an important requirement for reliable data transmission in IoT. Techniques that provide defense against malicious users and their attacks are required to ensure reliability in IoT. As shown in 

Figure 4

, security techniques include cryptography techniques, blockchain-based techniques and data integrity detection techniques. In 

Table 3

, we provide the recent work related to security in IoT.

Figure 4.

 Different categories of recent work in Security in IoT.

Table 3.

 Security in IoT: Recent literature review (C = Cryptographic Technique, B = Blockchain, D = Data Integrity Attack Detection).

Scenario	C	B	D	Key Idea	Results

Anomaly detection at edge gateways [48][126]	✗	✗	✓	Classify traffic as anomalous and normal based Fuzzy C-means clustering & fuzzy interpolation Access to malicious IoT nodes restricted	Improved accuracy Reduced false positive rate
Transport layer security for IoT applications [49][127]	✓	✗	✗	Reduce latency and achieve forward secrecy Identity-based cryptographic technique Identity-based encryption for client-server data	Reduced latency Reduced traffic overhead
Periodic data sharing with the server		Spectrum allocation in variable interference		Graph-Based Algorithm (GBA) used		Maximum weight matching in bipartite graphs	Increased transmitting users Reduced transmission delay
Content-centric computing IoT [39][117]	✗	✗	✗	✓	✗	Improve Quality of Experience (QoE) Factor such as Mean Opinion Score (MOS) Cache resource allocation to improve QoE Shortest Path Tree (SPT) algorithm Deep-Q learning algorithm	Increased QoE Reduced network cost
Energy self-reliant IoT network [40][118]	✗	✓	✓	✗	✗	Relay node harvests renewable energy Relay node transmits data Relays provide RF power to IoT nodes Solved time and power resource allocation Lyapunov optimization for max throughput	Increased data rate Increased throughput
UAV network supporting IoT [41][119]	✓	✓	✗	✗	✗	IoT devices grouped into equal-sized clusters Matching theory-based algorithm Match UAV sub-channels to IoT nodes Interference is minimized Alternate optimization used Optimized placement of UAV nodes Optimized transmit power of IoT nodes	Reliable power selection Reduced power of IoT nodes

In ^[37], the authors consider an edge IoT scenario where IoT devices offload their tasks to nearby edge nodes. The authors use machine learning algorithms for efficient task offloading. A Q-learning-based algorithm is developed to allocate computational resources. The goal of the protocol is to maximize the utility of applications and achieve fairness. Moreover, an online Q-learning scheme to handle large state and action space is proposed, which reduces computation overhead and convergence time. The performance evaluation shows maximized IoT application utility and improved resource allocation fairness.

In [115], the authors consider an edge IoT scenario where IoT devices offload their tasks to nearby edge nodes. The authors use machine learning algorithms for efficient task offloading. A Q-learning-based algorithm is developed to allocate computational resources. The goal of the protocol is to maximize the utility of applications and achieve fairness. Moreover, an online Q-learning scheme to handle large state and action space is proposed, which reduces computation overhead and convergence time. The performance evaluation shows maximized IoT application utility and improved resource allocation fairness.

The authors in ^[38] consider a smart factory scenario where IoT-enabled machines periodically share data with the server. These data can include different parameters of the machines. The work proposes a spectrum allocation technique in the case when bands are impacted by interference differently. A Graph-Based Allocation (GBA) algorithm is used for resource allocation. The spectrum allocation problem is formulated as a bipartite graph ad successive maximum weight matching is usedThe results show an increased number of users that transmit data. In addition, data transmission delay is considerably reduced.

The authors in [116] consider a smart factory scenario where IoT-enabled machines periodically share data with the server. These data can include different parameters of the machines. The work proposes a spectrum allocation technique in the case when bands are impacted by interference differently. A Graph-Based Allocation (GBA) algorithm is used for resource allocation. The spectrum allocation problem is formulated as a bipartite graph ad successive maximum weight matching is usedThe results show an increased number of users that transmit data. In addition, data transmission delay is considerably reduced.

The work in ^[39] considers a content-centric IoT scenario where devices offload their data to cache storage of fog nodes. The goal of the two proposed techniques is to improve user satisfaction in terms of Quality of Experience (QoE) by using efficient cache resource allocation. Factors such as the Mean Opinion Score (MOS) are considered while computing the QoE. The first technique uses Shortest Path Tree (SPT) algorithm, whereas the second technique uses the Deep Q-learning algorithmThe results show an increase in QoE and reduced network cost.

The work in [117] considers a content-centric IoT scenario where devices offload their data to cache storage of fog nodes. The goal of the two proposed techniques is to improve user satisfaction in terms of Quality of Experience (QoE) by using efficient cache resource allocation. Factors such as the Mean Opinion Score (MOS) are considered while computing the QoE. The first technique uses Shortest Path Tree (SPT) algorithm, whereas the second technique uses the Deep Q-learning algorithmThe results show an increase in QoE and reduced network cost.

In ^[40], an energy self-reliant IoT network is considered. Relay nodes are chosen in the network that harvests energy using renewable sources. The relay node serves two purposes, enabling data transmission by forwarding signals from the source to the destination and powering multiple IoT nodes using RF signals. The authors solve the time and resource allocation problem using Lyapunov optimizationThe results show an increased achievable data rate and throughput.

In [118], an energy self-reliant IoT network is considered. Relay nodes are chosen in the network that harvests energy using renewable sources. The relay node serves two purposes, enabling data transmission by forwarding signals from the source to the destination and powering multiple IoT nodes using RF signals. The authors solve the time and resource allocation problem using Lyapunov optimizationThe results show an increased achievable data rate and throughput.

The work in ^[41] considers a UAV network that supports IoT devices in terms of providing computation resources. IoT devices are grouped into equal-sized clusters. A matching theory-based algorithm is proposed that matches UAV sub-channels to IoT nodes such that interference is minimized. The work uses alternate optimization for optimal placement of UAVs and transmit power selection of IoT nodesThe results show improved reliability to achieve optimal transmit power. Moreover, IoT nodes use lower transmit power values when using the proposed algorithm.

The work in [119] considers a UAV network that supports IoT devices in terms of providing computation resources. IoT devices are grouped into equal-sized clusters. A matching theory-based algorithm is proposed that matches UAV sub-channels to IoT nodes such that interference is minimized. The work uses alternate optimization for optimal placement of UAVs and transmit power selection of IoT nodesThe results show improved reliability to achieve optimal transmit power. Moreover, IoT nodes use lower transmit power values when using the proposed algorithm.

3.2. Latency Management

Latency is a vital QoS indicator for IoT applications. Most applications are sensitive to latency and need a lower latency value within a threshold to ensure reliable communication. As shown in 

Figure 3

, techniques such as smart retransmissions, resource allocation, multiple access and physical layer techniques, traffic prediction, and cooperation from other networks is used for improving the latency of IoT. This subsection provides an overview of work done in latency management in IoT as shown in 

Table 2

.

Figure 3.

 Different categories of recent work in Latency management in IoT.

Table 2.

 Latency management in IoT: Recent literature review (R = Smart Retransmissions, A = Resource Allocation, M = Multiple Access and Physical Layer Techniques, P = Traffic Prediction, O = Cooperation from other networks such as UAVs).

Scenario	R	A	M	P	O	Key Idea	Results

NB-IoT [42][120]

✓

✗

✓

✗

✗

Retransmissions impact on energy and latency Optimal control & data channel duration selection Optimal selection of coverage class per user

Reduced energy consumption Reduced latency

multiple RATs to connect with MEC [43][121]

✗

✓

Cloud with several VNFs [44][122]

✗

✓

✗

✓

✗

✗

Blockchain for secure transmission Private blocks & Transactions are recorded Novel consensus algorithm is developed

Robustness against attacks Lower communication overhead Lower computation timeVNFs are mapped to different applications VNFs allocation to reduce latency Regression for predicting application demand Optimal allocation of VNFs as per demand Initial offline training followed by online training

Reduced latency Improved prediction accuracy

Terrestrial networks, satellite IoT and MEC [45][123]

IoT insider attacks [51]

✗

✓

✗

[✗

129✗

]

Data from IoT to satellites and gateways Satellites have energy constraints Data offloaded to gateways for processing

✗

✗

✓

Malicious attack detection from IoT insiders AI and distance based attacks classificationOptimal IoT association with a satellite & a gateway Lagrange multiplier and DRL used

Reduced latency Reduced energy cost

Secure 5G Internet of drones [50

Improved accuracy

Reduced computation time

UAV assisted MEC based IoT [46][124]

✗

✓

✓

✗

✓

UAV facilitates storage and computation UAV improves network coverage using relaying Problem divided into three subproblems First, IoT node and UAV association Second, communication resource scheduling Third, the UAV placement

Reduced latency

][128]

✗

✓

AES for constrained IoT devices [52][130]

✓

✗

✗

Reduce complexity of the standard AES algo. Use a reduced number of rounds for algorithm Mathematical proof of proposed protocol

Reduced encryption time

Semi-blind downlink NOMA based IoT [47][125]

✓

✗

✗

✗

Data Integrity attacks in IoV [53][131]

✗

✗

✓

✗

Isolation forest algorithm used Find anomalies in traffic density information Detected anomalies verified Verification from the neighborhood area

Improve SIC technique

Improved anomaly detection accuracy Reduced false positives Interference AICA used

Improved signal to noise ratio Reduced latency

The work in ^[48] proposes an anomaly detection technique for IoT devices. The proposed security algorithm is implemented at edge gateways and classifies traffic as abnormal and normal. The classification is done using fuzzy C-means clustering and fuzzy interpolation. Once the anomaly is detected, access to the malicious IoT nodes is restrictedThe results signify the performance of the proposed protocol in terms of improved accuracy of anomaly detection and reduced false-positive rate.

The work in [126] proposes an anomaly detection technique for IoT devices. The proposed security algorithm is implemented at edge gateways and classifies traffic as abnormal and normal. The classification is done using fuzzy C-means clustering and fuzzy interpolation. Once the anomaly is detected, access to the malicious IoT nodes is restrictedThe results signify the performance of the proposed protocol in terms of improved accuracy of anomaly detection and reduced false-positive rate.

In ^[49], the authors proposed a transport layer security protocol for IoT applications. The goal is to reduce the latency and achieve forward secrecy. In this regard, an identity-based cryptographic technique is proposed. The key idea is that the client uses identity-based encryption to transmit data to the server when it receives no response from the serverThe results show reduced latency and reduced traffic overhead.

In [127], the authors proposed a transport layer security protocol for IoT applications. The goal is to reduce the latency and achieve forward secrecy. In this regard, an identity-based cryptographic technique is proposed. The key idea is that the client uses identity-based encryption to transmit data to the server when it receives no response from the serverThe results show reduced latency and reduced traffic overhead.

The work in ^[50] proposes a secure communication technique for 5G Internet of drones. Blockchain technology is used to ensure secure data transmission. Private blocks are created, and transactions are recorded in them. Moreover, a novel consensus algorithm is also developed. The result shows that the proposed protocol provides robustness against attacks, lower communication overhead, and lower computation time.

The work in [128] proposes a secure communication technique for 5G Internet of drones. Blockchain technology is used to ensure secure data transmission. Private blocks are created, and transactions are recorded in them. Moreover, a novel consensus algorithm is also developed. The result shows that the proposed protocol provides robustness against attacks, lower communication overhead, and lower computation time.

In ^[51], the authors propose a security protocol to detect malicious attacks which are from inside the IoT network. Artificial Intelligence (AI) based distance measurement techniques are used to identify and classify malicious attacksThe results show improved accuracy of detection and reduced computation time.

In [129], the authors propose a security protocol to detect malicious attacks which are from inside the IoT network. Artificial Intelligence (AI) based distance measurement techniques are used to identify and classify malicious attacksThe results show improved accuracy of detection and reduced computation time.

The authors in ^[52] provide Advanced Encryption Standard (AES) based security algorithm for constrained IoT devices. The proposed protocol reduces the complexity of the standard AES algorithm by using a reduced number of rounds. The mathematical proof of the proposed protocol is also providedThe results indicate reduced encryption time of the proposed protocol.

The authors in [130] provide Advanced Encryption Standard (AES) based security algorithm for constrained IoT devices. The proposed protocol reduces the complexity of the standard AES algorithm by using a reduced number of rounds. The mathematical proof of the proposed protocol is also providedThe results indicate reduced encryption time of the proposed protocol.

In ^[53], the authors present a technique to detect data integrity attacks in the Internet of Vehicles (IoVs). In this regard, the isolation forest algorithm finds anomalies in the traffic density information shared by vehicles. After detecting anomalies, verification messages are sent to neighborhood vehicles of potential malicious vehicles to verify if the data transmitted is malicious or correct. Simulation results show that improved accuracy of anomaly detection and reduced false positives.

In [131], the authors present a technique to detect data integrity attacks in the Internet of Vehicles (IoVs). In this regard, the isolation forest algorithm finds anomalies in the traffic density information shared by vehicles. After detecting anomalies, verification messages are sent to neighborhood vehicles of potential malicious vehicles to verify if the data transmitted is malicious or correct. Simulation results show that improved accuracy of anomaly detection and reduced false positives.