A Survey of Internet of Things and Cyber-Physical Systems: Standards, Algorithms, Applications, Security, Challenges, and Future Directions

Round 1

Reviewer 1 Report

In this paper, the authors explore various aspects of the Internet of Things (IoT) and cyber-physical systems (CPS) in recent years. The initial focus is on industry standards that ensure cost-effective solutions and interoperability. However, to make the survey more comprehensive, it is essential to include a clear discussion and exhaustive comparison of the various papers that have been studied. This will provide a more holistic view of the topic.

Additionally, while the paper highlights the use of machine learning algorithms for achieving various target applications such as classification, clustering, regression, prediction, and anomaly detection, it would greatly benefit from a deeper exploration of advanced algorithms. These include deep learning, transfer learning, and data generation algorithms, which have shown promise in providing more accurate models.

Furthermore, there are some significant papers missing from the recent literature that should be included in the survey, particularly regarding IoT+Satellite research. I recommend considering the following papers for inclusion:

1. Monzon Baeza, V.; Ortiz, F.; Herrero Garcia, S.; Lagunas, E. "Enhanced Communications on Satellite-Based IoT Systems to Support Maritime Transportation Services." Sensors 2022, 22, 6450. https://doi.org/10.3390/s22176450. 

2. Fort, Ada, et al. "Reliability Analysis of an IoT Satellite Facility for Remote Monitoring and Asset Tracking within Marine Environments." 2022 IEEE International Workshop on Metrology for the Sea; Learning to Measure Sea Health Parameters (MetroSea). IEEE, 2022.

3. Sadek, Rowayda A., and Hesham M. Elbadawy. "Towards IoT Era with Current and Future Wireless Communication Technologies: An Overview." 2022 39th National Radio Science Conference (NRSC). Vol. 1. IEEE, 2022.

Furthermore, the identified challenges should be further discussed and well justified in the paper. This will provide a more thorough understanding of the obstacles faced in the field.

Finally, I noticed a few grammatical errors in the manuscript. I recommend a thorough proofreading to correct these errors and enhance the overall clarity of the paper.

I noticed a few grammatical errors in the manuscript. I recommend a thorough proofreading to correct these errors and enhance the overall clarity of the paper.

Author Response

Reviewer 1

The paper is quite good and interesting, however it has the following issues:

Overall Response: Authors would like to thank and express our appreciation for the valuable comments from Reviewer 1. You have pointed out various issues that we have to clarify, elaborate, and amend. Our manuscript has been revised. The replies to the comments are listed below.

Comment 1: In this paper, the authors explore various aspects of the Internet of Things (IoT) and cyber-physical systems (CPS) in recent years. The initial focus is on industry standards that ensure cost-effective solutions and interoperability. However, to make the survey more comprehensive, it is essential to include a clear discussion and exhaustive comparison of the various papers that have been studied. This will provide a more holistic view of the topic.
Response: Thank you for your comment. Besides elaborating on the contents for other comments, we have added Table 6 to summarize and compare the recent applications of CPS and IoT using various types of data generation algorithms.
Amendment (Subsection 3.2.3): Table 6 summarizes recent research works on CPS and IoT applications using these data generation algorithms.
Table 6. Recent research works on CPS and IoT applications using data generation algorithms.

Comment 2: Additionally, while the paper highlights the use of machine learning algorithms for achieving various target applications such as classification, clustering, regression, prediction, and anomaly detection, it would greatly benefit from a deeper exploration of advanced algorithms. These include deep learning, transfer learning, and data generation algorithms, which have shown promise in providing more accurate models.
Response: Thank you for your suggestions. Elaboration is made on the advanced algorithms, including deep learning, transfer learning, and data generation algorithms.
Amendment (Subsection 3.2.1): Generally, deep learning algorithms require prerequisites of sufficient training data and high-performance computing services [70,71]. The algorithms can learn more high-level features to build more accurate models, with a tradeoff of increasing model complexity (more hyperparameters and higher dimensionality). Deep learning extends artificial neural networks and feature learning with at least three layers. Many deep learning algorithms were proposed, including deep neural networks, convolutional neural networks, deep belief networks, gated recurrent units, and long short-term memory. Table 4 compares the advantages and disadvantages of these deep learning algorithms [72–74]. Although the CNN algorithm has received the most significant number of adoption to build deep learning models attributable to its superiority in automatic feature extraction without a complete understanding of do-main knowledge, it has several disadvantages that bring up the need for other deep learning algorithms. Different algorithms may be selected for other applications, with no best algorithm fitting general applications. The uniqueness of different deep learn-ing algorithms leads to vigorous performance evaluation and comparison procedures that ablation study, extensive analysis of fine-tuning of hyperparameters, and verification of multiple types of deep learning algorithms are usually presented in the literature. The general ideas for choosing an appropriate algorithm depend on the problem formulation, the size of the dataset, requirements on the complexity and performance of models, and availability of computing power.
Amendment (Subsection 3.2.2): Regarding the application of deep learning, there are various challenges (i) training a deep learning model from scratch is time-consuming, particularly when processing big and high-dimensional data; (ii) large-scale datasets may not be available in many applications due to the small-scale nature of some classes of data and expensive data collection process. Abovementioned in Subsection 3.2.1, deep learning algorithms do not natively perform in small-scale datasets.; and (iii) insufficient seen data in the model. Any machine learning model is trained with relatively fewer samples than the global data pool (unseen data).
The research of transfer learning via multiple source datasets has become an emergent solution to tackle negative transfer by introducing multi-round transfer learning, which slows down the knowledge transfer process [81–83]. In addition, this facilitates the enhancement of model performance with more source datasets (more unseen data from the perspective of the target domain). On the other hand, mul-ti-round transfer learning can be formulated with auxiliary domains [84,85], which serve as intermediate domains between the source and target domains. The intermedi-ate domains are often chosen to reduce the dissimilarity between the source and target domains so that the extent of negative transfer can be reduced.
Amendment (Subsection 3.2.3): It is noted that the GAN family may experience challenges (i) difficulty in model training: The convergence of GAN is not guaranteed and few sample sizes often exist (as a major reason to generate additional training data); (ii) mode collapse: GAN is prone to generate a subset of outputs, with narrow variety of samples. It requires good knowledge of the design of loss function to produce a good variety of outputs; (iii) computational requirements: The additional data generation step using GAN increases the need for computing power to build a machine learning model, i.e., the total time taken for data generation, feature extraction, and model construction is lengthy and requires enormous computing power; (iv) overtraining: The generator achieves high accuracy. However, the generated samples have deviated to a large extent from the ground truth data distribution.

Comment 3: Furthermore, there are some significant papers missing from the recent literature that should be included in the survey, particularly regarding IoT+Satellite research. I recommend considering the following papers for inclusion:
1. Monzon Baeza, V.; Ortiz, F.; Herrero Garcia, S.; Lagunas, E. "Enhanced Communications on Satellite-Based IoT Systems to Support Maritime Transportation Services." Sensors 2022, 22, 6450. https://doi.org/10.3390/s22176450.
2. Fort, Ada, et al. "Reliability Analysis of an IoT Satellite Facility for Remote Monitoring and Asset Tracking within Marine Environments." 2022 IEEE International Workshop on Metrology for the Sea; Learning to Measure Sea Health Parameters (MetroSea). IEEE, 2022.
3. Sadek, Rowayda A., and Hesham M. Elbadawy. "Towards IoT Era with Current and Future Wireless Communication Technologies: An Overview." 2022 39th National Radio Science Conference (NRSC). Vol. 1. IEEE, 2022.
Response: Thank you for your suggestion of recent works. We have included the works in the revised paper.
Amendment (Section 4): Attention is also drawn to the satellite-based IoT systems, driven by the development of 5G and 6G networks [128]. Examples of applications are maritime transportation services [129] and remote monitoring and asset tracking in marine environments [130].

Comment 4: Furthermore, the identified challenges should be further discussed and well justified in the paper. This will provide a more thorough understanding of the obstacles faced in the field.
Response: Thank you for your suggestion. We have elaborated on the open challenges.
Amendment (Section 6): Many CPS and IoT standards are not yet ready: Standards are official documents that define the guidelines and specifications to enhance the performance of the services, methods, products, and/or materials. These also help to achieve replicable results. Generally, dedicated working groups (involving different parties, such as government officials, industry, and consumers) take several years to publish a standard. Tables 2 and 3 shared 31 published standards in CPS and IoT. Other CPS and IoT standards are under development. The examples of developing CPS standards are (i) IEEE P1547.3: Interconnection between electric power systems and distributed energy resources; (ii) IEEE P2658: Testing of electric power systems; (iii) IEEE P2808: Function designations of electrical power systems; (iv) IEEE P2968.2: threat modeling for decentralized clinical trials; and (v) IEEE P9274.4.2: experience application programming interface. The examples of developing IoT standards are (i) IEEE P1912: security and privacy for wireless devices; (ii) IEEE P2303: adaptive management of cloud computing; (iii) IEEE P3333.1.1: visual comfort assessment and quality of experience of 3D contents; (iv) IEEE P21451-1-6: message queue telemetry transport for networked device communication; and (v) IEC/IEEE P62704-4: finite element method for specific absorption rate calculation in the human body from wireless devices. Without the aid of standards, things become highly heterogeneous, which leads to an interoperability issue. In reality, it is time-consuming to phase out the existing gadgets and migrate to new versions that follow standards. There exists another retardation to adopting standards because laws may not enforce to regulate the systems and products to follow standards, mainly due to a longer timeframe in law legislation than that of standard publication.
Open data is not widely available: Open data policies have been receiving retardation from government officials [145], the general public [146], and companies [147]. Typical reasons for opposing open data are (i) legislation: new laws to regulate the release and use of open data are difficult because there is poor acceptability in different stakeholders; (ii) privacy: ensuring data privacy is important because data is often containing personal and sensitive information, that misuse and stolen will lead to threats; (iii) sharing data means sharing money: Data analysis turns data into valuable information that potentially brings benefits and income. For example, if sufficient samples are shared with a marketing company, who will pay for the data because data collection is costly; (iv) existing devices, platforms, and systems are incapable of data collection and storage: collecting and storing ever-growing data is expensive. As a result, consumer-grade products usually ignore data collection and storage. It is important to recognize that open data plays a crucial role in providing a substantial amount of data to train machine learning models. This is especially true in situations where various small-scale and diverse open datasets must be combined to create the models. While generating algorithms can create more training data, it is not effective for classes with very few samples. In recent years, an open data working group was established under the United Nations that comprised 12 country representatives (New Zealand, Mauritius, Argentina, Poland, Australia, Suriname, Egypt, Sweden, Italy, UK, Jordan, and Malaysia), international organizations, and agencies. It is willing to attract and invite representatives to join the working group from the rest of the member states (181) of the United Nations.
Availability of computing power for model training and data analysis: Analyzing big data and training models using advanced algorithms requires immense computing power. Mobile devices and local computers (embedded with GPUs) are limited to many applications. The availability of edge, fog, and cloud computing offers more computing power, with the latency tradeoff between edge, fog, and cloud computing [146]. There is an increasing trend towards subscribing to cloud GPUs, usually charging based on every hour. Purchasing multiple GPUs in local computers is thus not necessary, which also relieves the local computing bandwidth. However, the bottleneck of the availability of computing power is that the growth rate of data is much higher than that of the processing units. Only a limited number of users can rely on the computing services that lead to the latency of data analysis and decision-making. An alternative solution is prioritizing the resources to more critical applications (for those who can benefit a wider group of people).

Comment 5: Finally, I noticed a few grammatical errors in the manuscript. I recommend a thorough proofreading to correct these errors and enhance the overall clarity of the paper.
Response: Thank you for your comment. We have revisited the paper to improve written English.
Amendment: Changes are made throughout the paper where applicable.

Author Response File: Author Response.docx

Reviewer 2 Report

Figure 2 represents only the initial sections of the article. To provide a comprehensive overview of the article's summary, I suggest describing the content in a concise manner, including all sections but with less detail. Additionally, I will provide an initial definition of Internet of Things (IoT) and Cyber-Physical Systems (CPS)

The article is written in appropriate English.

Author Response

Reviewer 1

Overall Response: The authors would like to thank and express our appreciation for the valuable comments from Reviewer 2. You have pointed out that we have to clarify and elaborate on. Our manuscript has been revised. The replies to the comments are listed below.

Comment 1: Figure 2 represents only the initial sections of the article. To provide a comprehensive overview of the article's summary, I suggest describing the content in a concise manner, including all sections but with less detail.
Response: Thank you for your comment. We have updated Figure 2.
Amendment (Section 1): Figure 2 presents the structure and summarizes the number of standards, algorithms, applications, security threats, security tools, and open challenges of this paper.

Figure 2. Structure of the article.

Comment 2: Additionally, I will provide an initial definition of Internet of Things (IoT) and Cyber-Physical Systems (CPS).
Response: Thank you for your suggestion. We have elaborated on the definition of CPS and IoT.
Amendment (Section 1): The cyber-physical system (CPS) is an embedded computing and communication system that combines virtual and physical spaces and interfaces the digital and physical worlds [1,2]. In today’s digital era, the Internet of Things (IoT) is a promising network of physical objects, embedding sensors, devices, servers, and platforms, connected to the Internet for data communication, exchange, storage, and analysis [3,4].

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors have addressed all my recommendations, I have no further comments to add.

 Minor editing of English language required