10.1. Introduction

The TI has given rise to a wide range of use cases with diverse type of applications. However, they are not equal in terms of their needs for network resources, and there is consequently a diversity in the requirement of end-to-end QoS assurance needs. The new applications, especially the haptic applications, may need 1 to 10 ms as an end-to-end latency especially for teleoperation case studies. The network domain should support new infrastructure to allow ultra-low latency and high reliability. Let us consider the classical view of the actual mobile network and the traditional communication protocols. They are not suitable for the applications of TI as delay is introduced in every layer of the protocol stack in the end devices, the master domain, the transport domain for the end-to-end communication. Hence, 5G is considered as the perfect candidate for transporting haptic data due to the flexible design of its access and core networks.

Involving a user in real haptic communication would rather involve the other types of their interactive senses, just like involving the voice and the video as a supportive application. However, the decision to have a standalone codec for each type of data or a hybrid codec is under study as, when dealing with QoS, each type of traffic will require a specified QoS to be guaranteed through the network. Thus, a multiplexer interacts actively with each type of traffic depending how urgent the need for that type of traffic to be prioritized in terms of resource allocation (Aijaz et al. 2018; Sachs et al. 2019).

From a technical standpoint, the E2E latency may be introduced when the packet is traveling form one edge to another. For example, the multiplexing technique should take into consideration the high update and packet rate of haptic data compared to audio and video traffic. As it should also change its behavior not only when multiplexing and sending the data through the network, but also for congestion control when it happens, the multiplexer should decide which packet to drop and which has a higher priority to be passed through the channel (Eid et al. 2011; Gokhale et al. 2017a; Cabrera et al. 2019). The methodology used in the multiplexing algorithm would affect the latency to a great extent, and designing a good multiplexer is a challenging issue due to the variety of the type of Internet traffic including the haptic traffic.

The other hurdle in TI is the size of packets that influence the delay of transmission, as considering the current size would be a big obstacle towards providing the 1 ms end-to-end delay. This delay includes packet processing in the receiver and transmitter including encoding and decoding. The orthogonal frequency division multiplexing (OFDM) used in the current technology where the symbol duration is long may not be a good option for modulation.

As for accessing cloud computing capabilities, the delay will increase hugely if the cloud computing is accessed through the traditional Internet, this will bring a heavy load on the backhaul and core network, and will also increase the latency, thus violating all the threshold values for latency. Therefore, moving the capabilities of the cloud close to the edge would be a good solution, which means that processing and storage should be achieved in RAN and not done in the core network, i.e. the Internet, consequently reducing the delay to some extent.

Since the traffic model in TI does not follow a regular basis, i.e. the arrival of packets to the schedulers can be sporadic and/or bursty in the medium access control (MAC) layer in the entities involved in the access network, the delay experienced by a packet thus includes the transmission delay and queuing delay in the transmitter side, the decoding and delay in the receiver side. For the applications which have a hard latency requirement, the design of the queue delay model should be considered in the whole system. In view of the aforementioned factors impacting the delay of haptic applications, in this chapter, we describe the latest research works making use of different technologies to find a solution for guaranteeing the E2E delay.

10.2. Low latency in the Tactile Internet

The TI is still in its infancy and new methods should be developed to guarantee the low-latency criteria that are one of the essentials among the others characterized by ultra-reliability, security and high QoS and QoE guarantee. However, when referring to the solutions that try to guarantee the minimum latency characterized by 1 ms for the critical haptic applications to multiple of 100 ms for other types of data combined with haptic data, the adaptive solution can be the best choice. Such kinds of solutions should adapt themselves intelligently to the context. To date, there have been many research works trying to deal with the low latency by keeping it as minimal as possible. The method used for achieving this objective can range from framework, resource allocation, technological methods like MEC, network coding and communication protocols. In the following, we detail some solutions used to reduce the latency, explaining the methodology used to achieve this.

10.3. Intelligence and the Tactile Internet

The work on progress in the developments of 6G, artificial intelligence (AI) and TI use cases will be leading towards what we call intelligent connectivity from an end-to-end point of view. This would include intelligence in how the operation will be performed in the master domain, network domain and slave domain. In fact, the network should adapt itself to the type of application without the intervention of a human through automatic or autonomic functionalities provided in each protocol stack of the network. All the use cases of TI will be easy to deploy thanks to the sophisticated mechanism introduced by artificial intelligence (Khemiri 2015; Cheng et al. 2018; Dab et al. 2019; Holland et al. 2019; Pérez et al. 2020; Yu et al. 2020; Zhao et al. 2020).

In order to achieve this intelligent connectivity, contextual data from the environment needs to be analyzed and processed to make the right decision at the right moment. This would reduce give alerts in real-time leading to fewer human errors. This can also be extended to the prediction mechanisms offered by AI through its different techniques just like machine learning techniques.

In this section, we will show how AI can intervene in all the domains of TI and how the TI can benefit from AI techniques in order to enhance the performance of the system especially for mission-critical applications that require very fast feedback from the slave domain. Indeed, IEEE 1918.1 introduced the reference architecture of TI by adding elements that would be good contenders for implementing AI techniques, taking the GNC and the support engine, for example. These elements could play a great role towards the zero-delay perception in the network depending on the type of closed or open-loop systems.

10.4. Edge intelligent

The support engine can play the role of edge computing in the TI, and it can reside either in the master domain or in the network domain depending on how the information can be uploaded or downloaded from it. The implementation of the role of processing and computation in the support engine can be supported by the AI mechanism in order to provide the so-called intelligent edge.

In Ahmad et al. (2021), the authors proposed to minimize the path length or to stretch the path between the user and the cached content when the content has been identified on which router it is placed. They introduced a reinforcement learning-based approach to reduce the stretch between the user and the content router. Therefore, their contributions were a step ahead by implementing AI in the information-centric networking to learn based on exploration and exploitation to give better performance. The idea is to decrease the delay from an end-to-end perspective while decreasing the number of paths in the network through the Q-learning-based approach that is the technique of reinforcement learning to reduce the stretch between the user and content router.

In Grasso et al. (2021), the authors implemented the AI technique in the tactile support engine as its aim is to predict the haptic/tactile experience, for example acceleration of movement on one end and force feedback on the other end, in order to send forecasted values to the controlled domain when needed, allowing the spatial decoupling of the active and reactive end(s) of the TI. The authors tried again to reduce the delay in a mission critical application while using on-line gaming.

In Monnet and Yahiya (2020), the authors introduced a teleoperation mechanism overlaid on 5G in order to guarantee the QoS of the end-to-end network. The AI technique was used to exchange information among MEC servers through the gossip protocol to have a global view of the whole network instead of having a local view. The work can be summarized by implementing the gossip protocol in the MEC server in order to exchange information with its peers as shown in Figures 10.1 and 10.2.

However, the AI is not used only to guarantee the QoS over the network, it has other purposes just like traffic classification. In fact, AI techniques can be deployed to classify the traffic through the network domain. In this case, the GNC which plays the role of interfacing between master domain and network domain, can act as the anchor of traffic classification generated in the master domain. This is due to the diverse type of traffic in the TI with different resources generating traffic dependent certainly on the use case. Thus, the GNC is the most important challenge to deal with especially when dealing with applications requiring a stringent quality of service guarantee, especially in terms of latency. In order to recognize the type of traffic and then allocate the appropriate resource, it is necessary to classify the traffic in an appropriate way. In Amaral et al. (2016), the authors used SDN with ML training for data preparation, data clustering and classification. In the classification process, they used support vector machine (SVM), decision tree, random forest and Kth-nearest neighbor as shown in Figure 10.3. This step is essential in TI architecture as once the types of traffic are classified, they will be dedicated not only the appropriate resource but also the appropriate network. In reality, the TI offers a wide range of network types depending on the type of mission-critical and non-mission critical application.

Sheng (2018) proposed a scalable intelligence-enabled networking platform to remove the traffic redundancy in 5G audio–visual TI scenarios. The proposed platform incorporates a control plane, a user plane, an intelligent management plane and an intelligence-enabled plane. Out of these planes, the intelligence-enabled plane comprises a novel learning system that has decision-making capability for generalization and personalization in the presence of conflicting, imbalanced and partial data. Furthermore, Ruan and Wong (2018) studied the application of ML intelligence in taking effective decisions for dynamically allocating the frequency resources in a heterogeneous fiber-wireless network. More specifically, they investigated the utilization of an artificial neural network for the following purposes: (i) in learning network uplink latency performance by using diverse network features; and (ii) in taking flexible bandwidth allocation decisions towards reducing the uplink latency.

10.5. Open issues

The TI can be considered as a revolution or an evolution of the type of communications, applications, services as well as business models. It is expected that the TI will be an economy booster for different use cases and scenarios. However, this evolution is still in progress and will need a mature framework that can include all the functionalities described in the previous sections in order to guarantee the low latency which is a key feature of haptic communications. There are still challenges to be addressed in this regard; even this chapter is not an extensive review that gathers all functionalities, but it does provide an idea about the progress in research in this matter, which is a good start towards further elaboration in each mechanism described for reducing the latency. To this aim, there follows some open issues that can be investigated further to achieve this goal:

• – Proactive resource allocation would be the best solution for critical applications. The tactile applications can be classified according to their requirements into low, high and adaptive delay applications. The resource allocation mechanism should change its behavior according to the type of application with a dedicated mechanism suitable for each one. Generally, the proactive end-to-end resource allocation stringent latency requirement is the best way to guarantee the 1-ms latency criteria. Mechanisms for scheduling, QoS provisioning, admission control and so on should be re-designed to react as fast as possible to process haptic traffic.
• – Mobile core-based cloud native is an interesting idea to bring all the functionalities, capabilities and services of the traditional cloud to the core network. This would reduce lots of hops that the haptic traffic should cross through. In addition, this could bring the cloud closer to the users in terms of physical location and could reduce the latency. However, having an overloaded core network would bring some congestion problems that should be solved through SDN and the orchestration service provided by the VNF.
• – Mobility would add an additional dimension of complexity that impacts the QoS and especially the latency when changing the point of attachment from one to another especially when the case of vehicles over TI is considered. The handover delay would influence the type influence the type of communication latency. A position-based solution could be considered the best solution to this type of problem. Artificial intelligence algorithms can be considered at each step of the design of any resource allocation, or orchestration in the core network. The main domain of interest that could be used in combination is enforcement learning or the process of learning, since haptic behavior is repetitive and does not need to generate new action to deal with the motion; therefore, solutions based on learning would be an interesting area to investigate.
• – Anticipatory network is also an option to decrease the latency by using principles like context, prediction and optimization. Using the anticipatory network, the context information will then be studied through past and present information, a prediction of the behavior will be achieved, then an optimization is carried out to meet the requirement of any application. This would be useful in the case of the TI especially improving the QoE of users who are involved in haptic communications (Bui et al. 2017).
• – Artificial intelligence will unquestionably carry a paradigm shift regarding data-oriented approaches; there are still open problems to be resolved. There is no realistic deployment in the TI as there is no comparison for the methods used in learning for each case study. Everything depends on the data set generated by each method and how it is dealt with.

10.6. Conclusion

In addition to the latency issue that we discussed in section 10.2, there are other challenges that face the TI, such as how to keep the TI system as reliable as possible, enabling the system to be scalable, as well as ensuring its security. Indeed, reliability and security are both so important to the system that it should be 99.99% secure when working. With regards to the security, the TI architecture cannot be secured with the traditional techniques of Internet technologies. For example, Tactile Internet actors are vulnerable and not secure against distributed denial of service (DDoS) attacks (which decrease availability), remote hijacking, cloning attacks and man in the middle. Any single TI actor could represent a single point of failure (SPOF) for the entire network and thus damage the availability of data, confidentiality and integrity (Li et al. 2018; Mohanta et al. 2019).

Security, reliability and availability are such important issues in the TI, if one of them is hijacked, the entire system will be unstable. However, the traditional methods in securing the information system are not convenient for the TI due to it is stringent requirements of QoS, especially for mission critical applications. It is important to re-design methods for ensuring confidentiality, integrity and authentication to be adapted to the TI.

Hidar et al. (2021) conceived a model using blockchain for authentication in order to secure human-to-machine interactions, like remote surgery, in the Tactile Internet environment. Thus, a surgeon can now authenticate a robot arm using a good-shared session key and build a high level of security in communication.