Periodic Reporting for period 2 - MASCOT (An Information-Theoretic Perspective on Massive Asynchronous Connectivity)
Reporting period: 2023-10-01 to 2024-09-30
MSCOT addresses the challenges posed by massive connectivity in asynchronous communications from an information theory perspective. The primary objective is to develop efficient communication methods for scenarios where numerous devices need to communicate asynchronously over a shared medium, maintaining energy efficiency and low computational complexity.
The significance of this research extends to various real-world applications. In a society increasingly reliant on interconnected devices and communication technologies, solving the massive connectivity problem is critical. From the Internet of Things (IoT) to 5G networks and beyond, MASCOT aims to design protocols specifically for these services and applications, enabling a revolution in wireless communication systems by making them more reliable, energy-efficient, and capable of meeting the diverse needs of our modern, interconnected world. This knowledge serves as a foundation for researchers and developers, driving the creation of communication schemes tailored to this emerging field.
As MASCOT concludes, it has demonstrated theoretically that handling device collisions is essential to meet reliability and energy-efficiency constraints in IoT-like and machine-type communications. Using information-theoretic tools, MASCOT provides finite-blocklength theoretical bounds and approximations, as well as initial deep-learning-based algorithms focused on interference rejection among devices causing unintentional interference. MASCOT paved the way for fundamentally supported operations for services and applications involving battery-limited devices transmitting short packets sporadically through theoretical research and practical algorithms based on the latest deep learning architectures tailored to this problem.
The significance of this research extends to various real-world applications. In a society increasingly reliant on interconnected devices and communication technologies, solving the massive connectivity problem is critical. From the Internet of Things (IoT) to 5G networks and beyond, MASCOT aims to design protocols specifically for these services and applications, enabling a revolution in wireless communication systems by making them more reliable, energy-efficient, and capable of meeting the diverse needs of our modern, interconnected world. This knowledge serves as a foundation for researchers and developers, driving the creation of communication schemes tailored to this emerging field.
As MASCOT concludes, it has demonstrated theoretically that handling device collisions is essential to meet reliability and energy-efficiency constraints in IoT-like and machine-type communications. Using information-theoretic tools, MASCOT provides finite-blocklength theoretical bounds and approximations, as well as initial deep-learning-based algorithms focused on interference rejection among devices causing unintentional interference. MASCOT paved the way for fundamentally supported operations for services and applications involving battery-limited devices transmitting short packets sporadically through theoretical research and practical algorithms based on the latest deep learning architectures tailored to this problem.
The first work developed within the context of MASCOT studies the fundamental limits of the multiple access channel when a massive number of battery limited devices connect to the network and the access point does not know the number of devices that needs to be handled. This gave rise to a publication in the IEEE Transactions on Information Theory.
During the first phase of the project spent at MIT, a channel model for the massive connectivity problem was identified, the so called “A-channel”, which turned out to be very relevant for this project. The A-channel is a collision channel that permits channel coding and that presents unique characteristics that make it particularly well-suited for transmission schemes in the massive connectivity context. In particular, these schemes effectively divide the available degrees of freedom into manageable blocks from a computational perspective, which align with MASCOT's goal of maintaining algorithmic tractability.
MASCOT's research has yielded promising results and valuable contributions. The focus was extended beyond the simple additive Gaussian noise model and efforts were directed toward investigating the A-channel. Such investigations encompassed both non-asymptotic bounds and expansions, as well as the development of transmission schemes tailored to this specific channel model. These studies gave rise to two conference contributions, which also cover the milestones and deliverables within the first work package, but for this different mathematical model.
Furthermore, the exploration of asynchronous communications initially began with a more fundamental channel model, the binary adder channel. While the initial intention was to explore the implications of Gaussian noise, the results in this context turned out to be both insightful and intriguing. This research gave rise to a conference paper at the IEEE International Symposium on Information Theory.
Additionally, in line with the evolving landscape of MASCOT's research, the study of interference management through the lens of machine learning was initiated during the first phase. In particular, the potential of deep learning in mitigating interference in multiple access environments has been explored. These environments involve devices from different wireless technologies accessing the communication medium simultaneously at the same frequency, creating interference in an asynchronous manner. Some of the findings have already been published in conference papers, with a particular highlight being a contribution to NeurIPS, one of the most prestigious international conferences on machine learning. The results of this contribution were presented in December 2023 in person at the conference.
During the second phase, a work to summarize all the results achieved in terms of interference rejection algorithms using deep learning was developed. The work is already being reviewed by the coauthors and will be submitted soon. This work closes in a very well-rounded way the research goals of MASCOT.
Furthermore, during the second phase of MASCOT at the UC3M, teaching has been one of the primary focus as it was needed to achieve one of the main goals of the postdoctoral fellow.
The execution of MASCOT has also adhered quite closely to the dissemination, exploitation, and communication plan. MASCOT researchers have participated in seminars, conferences, and even conducted a tutorial to disseminate their work to the research community.
Furthermore, the relevant code associated with their publications on Github has been shared.
Finally, the researchers have been actively promoting their work by providing updates on LinkedIn and Twitter.
During the first phase of the project spent at MIT, a channel model for the massive connectivity problem was identified, the so called “A-channel”, which turned out to be very relevant for this project. The A-channel is a collision channel that permits channel coding and that presents unique characteristics that make it particularly well-suited for transmission schemes in the massive connectivity context. In particular, these schemes effectively divide the available degrees of freedom into manageable blocks from a computational perspective, which align with MASCOT's goal of maintaining algorithmic tractability.
MASCOT's research has yielded promising results and valuable contributions. The focus was extended beyond the simple additive Gaussian noise model and efforts were directed toward investigating the A-channel. Such investigations encompassed both non-asymptotic bounds and expansions, as well as the development of transmission schemes tailored to this specific channel model. These studies gave rise to two conference contributions, which also cover the milestones and deliverables within the first work package, but for this different mathematical model.
Furthermore, the exploration of asynchronous communications initially began with a more fundamental channel model, the binary adder channel. While the initial intention was to explore the implications of Gaussian noise, the results in this context turned out to be both insightful and intriguing. This research gave rise to a conference paper at the IEEE International Symposium on Information Theory.
Additionally, in line with the evolving landscape of MASCOT's research, the study of interference management through the lens of machine learning was initiated during the first phase. In particular, the potential of deep learning in mitigating interference in multiple access environments has been explored. These environments involve devices from different wireless technologies accessing the communication medium simultaneously at the same frequency, creating interference in an asynchronous manner. Some of the findings have already been published in conference papers, with a particular highlight being a contribution to NeurIPS, one of the most prestigious international conferences on machine learning. The results of this contribution were presented in December 2023 in person at the conference.
During the second phase, a work to summarize all the results achieved in terms of interference rejection algorithms using deep learning was developed. The work is already being reviewed by the coauthors and will be submitted soon. This work closes in a very well-rounded way the research goals of MASCOT.
Furthermore, during the second phase of MASCOT at the UC3M, teaching has been one of the primary focus as it was needed to achieve one of the main goals of the postdoctoral fellow.
The execution of MASCOT has also adhered quite closely to the dissemination, exploitation, and communication plan. MASCOT researchers have participated in seminars, conferences, and even conducted a tutorial to disseminate their work to the research community.
Furthermore, the relevant code associated with their publications on Github has been shared.
Finally, the researchers have been actively promoting their work by providing updates on LinkedIn and Twitter.
The additional research directions within the project's scope underscore the commitment to adapt and refine the approach to address the challenges and opportunities that have emerged during the course of this research. This commitment ensures the continuation to make meaningful contributions to the field of massive connectivity and asynchronous communications, transcending the current state of the art and going even beyond the project conlusion. Hence, on top of the results achieved during the project execution, in the future, the plan is to continue this research line by:
• Exploring coding schemes that approach the fundamental limits of the A-channel. If these coding schemes can maintain low complexity, the possibility of engaging with policy makers to gauge their suitability for standardization in future wireless communication systems will be increased.
• Investigating the performance of LDPC codes in asynchronous access to the communication medium under more realistic wireless communication impairments. This investigation aims to confirm the benefits of asynchrony in certain communication scenarios.
• Extending the machine learning interference rejection algorithms to operate efficiently in real congested wireless communication environments, while minimizing computational overhead. This effort aims to bridge the gap between synthetic experiments and practical operations.
This project possesses the potential for substantial societal impact. For example, designing energy-efficient communication systems can directly enhance the quality of communications and have a positive environmental impact by reducing the energy required to transmit information.
• Exploring coding schemes that approach the fundamental limits of the A-channel. If these coding schemes can maintain low complexity, the possibility of engaging with policy makers to gauge their suitability for standardization in future wireless communication systems will be increased.
• Investigating the performance of LDPC codes in asynchronous access to the communication medium under more realistic wireless communication impairments. This investigation aims to confirm the benefits of asynchrony in certain communication scenarios.
• Extending the machine learning interference rejection algorithms to operate efficiently in real congested wireless communication environments, while minimizing computational overhead. This effort aims to bridge the gap between synthetic experiments and practical operations.
This project possesses the potential for substantial societal impact. For example, designing energy-efficient communication systems can directly enhance the quality of communications and have a positive environmental impact by reducing the energy required to transmit information.