Periodic Reporting for period 2 - RACOON (High-reliability Low-latency Communications with Network Coding)
Periodo di rendicontazione: 2020-10-01 al 2021-09-30
With various emerging latency-sensitive applications, e.g. V2V (vehicle-to-vehicle) automation, smart power systems (i.e. smart grids), Internet of things (IoT) and smart factory (industry 4.0) low-latency communication is becoming more and more important. For convenience, we call these schemes with high requirement on reliability and latency high-reliability low-latency (HRLL) communications. Common approaches to achieve reliability are forward error-control (FEC) coding. However, due to finite code length, it is rather challenging to achieve high reliability and low latency simultaneously, since powerful FEC, e.g. LDPC codes, Polar codes normally need large code length to achieve high reliability. Moreover, in many scenarios, high data rates are also needed. Thus, in this project, we will study key technologies which can boost reliability, latency and rate performance simultaneously. Network coding is one of the technologies, which show potentially in rate and reliability in the networks.
Project results will be quite valuable for future digitalization industries and society. For instance, in industrial critical control, the requirement on reliability may be a PER (packet error rates) of 10-6 or lower (may down to 10-9 for some applications) and meanwhile a latency of a few micro-seconds (10-6 seconds, may down to 10-9 seconds sometime) is required. To achieve these targets, we need to use all our available strategies, including the methods will be studied in the project.
Thus, the overall objectives of the action is to systematically investigate network coding for HRLL (high-reliability low-latency) communications. More specifically, we will find the fundamental limits of HRLL networks with network coding. We will optimize network coding schemes and propose efficient communication strategies to improve latency and reliability performance under practical constraints. Our results seek to develop HRLL communication technologies and network coding theorems, which will boost the future latency and reliability critical applications.
In addition to tradition coding optimization, such as distance or rank based measure, we also used data-driving approaches for optimization coding. For instance, in paper 10, we propose reinforced learning based code optimisation for wireless caching. In publication 11, we have studied the coding for mobile fog computing, which use batched codes (a special type of network coding) to reduce the impacts of straggler nodes. Similarly, in publication 15, we proposed MDS codes for large scale machine learning algorithm ADMM. In paper 13, we study how network coding can be used in the consensus of ADMM to improve the reliability (and also latency). In publication 12, we used analog network coding (NOMA) to improve the latency and reliability under security constraint. In paper 14, we analyse the fundamental limits by error exponents for finite blocklength region, which is the base of the HRLL communications.
Project results will be quite valuable for future digitalization industries and society. For instance, in industrial critical control, the requirement on reliability may be a PER (packet error rates) of 10-6 or lower (may down to 10-9 for some applications) and meanwhile a latency of a few micro-seconds (10-6 seconds, may down to 10-9 seconds sometime) is required. To achieve these targets, we need to use all our available strategies, including the methods will be studied in the project.
Thus, the overall objectives of the action is to systematically investigate network coding for HRLL (high-reliability low-latency) communications. More specifically, we will find the fundamental limits of HRLL networks with network coding. We will optimize network coding schemes and propose efficient communication strategies to improve latency and reliability performance under practical constraints. Our results seek to develop HRLL communication technologies and network coding theorems, which will boost the future latency and reliability critical applications.
In addition to tradition coding optimization, such as distance or rank based measure, we also used data-driving approaches for optimization coding. For instance, in paper 10, we propose reinforced learning based code optimisation for wireless caching. In publication 11, we have studied the coding for mobile fog computing, which use batched codes (a special type of network coding) to reduce the impacts of straggler nodes. Similarly, in publication 15, we proposed MDS codes for large scale machine learning algorithm ADMM. In paper 13, we study how network coding can be used in the consensus of ADMM to improve the reliability (and also latency). In publication 12, we used analog network coding (NOMA) to improve the latency and reliability under security constraint. In paper 14, we analyse the fundamental limits by error exponents for finite blocklength region, which is the base of the HRLL communications.
So far, the project progressed as planned. In about 1 years, 9 journal articles have been published or accepted in the prestige journals in IEEE. We studied following sets of problems
1. Physical layer network coding for high performance networks. In article 1, 3, 7, 8, we studied the physical layer network coding (e.g. these based on spatial modulation) to improve the reliability and latency performance of networks. A series of fundamental performance bounds (for WP1) and coding schemes (WP2) are studied.
2. Intelligent coding schemes for improving reliability in high performance networks (for WP3). Currently, distributed learning networks are widely used in our life. Thus, it is very valuable to study the coding scheme for learning networks. Especially, article 4 proposes network coding for decentralized learning with gradient descend. The proposed schemes based on MapReduce can significantly reduce the learning latency and improve the reliability. Related results are also studied in article 2, 6 and 9.
3. In article 8, wireless caching based on network coding and millimeter wave communications is studied. Transceiver efficiency is improved in terms of reliability and latency.
1. Physical layer network coding for high performance networks. In article 1, 3, 7, 8, we studied the physical layer network coding (e.g. these based on spatial modulation) to improve the reliability and latency performance of networks. A series of fundamental performance bounds (for WP1) and coding schemes (WP2) are studied.
2. Intelligent coding schemes for improving reliability in high performance networks (for WP3). Currently, distributed learning networks are widely used in our life. Thus, it is very valuable to study the coding scheme for learning networks. Especially, article 4 proposes network coding for decentralized learning with gradient descend. The proposed schemes based on MapReduce can significantly reduce the learning latency and improve the reliability. Related results are also studied in article 2, 6 and 9.
3. In article 8, wireless caching based on network coding and millimeter wave communications is studied. Transceiver efficiency is improved in terms of reliability and latency.
In these results, our results are beyond the state of the art in following aspects.
1. Joint content and communications. To further improve the latency and reliability performance, we proposed the joint content and coding schemes. For instance, in the learning networks, network coding is used to improve the latency and reliability to combat the straggler nodes in the MapReduce systems. The results show substantial gains, especially when the system performance is limited by straggler nodes.
2. Joint network coding and modulation. Though combining network coding and physical layer is not new, applying coding for index modulation is novel, to our knowledge.
3. We also studied the network coding for wireless caching with millimeter wave networks. The high rates of caching networks are well adapted by millimeter communications. To our knowledge, there is also no such scheme existed.
With the development of IoT and computing technologies, AI and machine learning are pervasive in our society. Thus, it is very important to improve the reliability and latency of learning networks. In this project, we aim at address this problem from the aspect of network coding. We expect our results can improve the learning efficiency and thus lead to solid societal impacts. Meanwhile, with the increasing amount of the multimedia data, wireless caching will be more and more used in e.g. mobile networks. Our results, based on network coding, may substantially improve the performance of caching networks and thus improve the life quality of our society. Similarly, our results may also improve the transmission efficiency and quality in physical layer by applying network coding to modulation, which may improve the industries and society in the wireless communication efficiency.
1. Joint content and communications. To further improve the latency and reliability performance, we proposed the joint content and coding schemes. For instance, in the learning networks, network coding is used to improve the latency and reliability to combat the straggler nodes in the MapReduce systems. The results show substantial gains, especially when the system performance is limited by straggler nodes.
2. Joint network coding and modulation. Though combining network coding and physical layer is not new, applying coding for index modulation is novel, to our knowledge.
3. We also studied the network coding for wireless caching with millimeter wave networks. The high rates of caching networks are well adapted by millimeter communications. To our knowledge, there is also no such scheme existed.
With the development of IoT and computing technologies, AI and machine learning are pervasive in our society. Thus, it is very important to improve the reliability and latency of learning networks. In this project, we aim at address this problem from the aspect of network coding. We expect our results can improve the learning efficiency and thus lead to solid societal impacts. Meanwhile, with the increasing amount of the multimedia data, wireless caching will be more and more used in e.g. mobile networks. Our results, based on network coding, may substantially improve the performance of caching networks and thus improve the life quality of our society. Similarly, our results may also improve the transmission efficiency and quality in physical layer by applying network coding to modulation, which may improve the industries and society in the wireless communication efficiency.