MINOAS-Machine learning aided optical wireless networksOptical wireless communications
As the wireless world moves towards the commercialization of the fifth generation (5G) systems, the demand for innovative technological developments that will become the pillars of beyond 5G (B5G) or six generation (6G) systems, is evident. In more detail, Cisco has foreseen that the global mobile data-traffic will increase more than 10 times (from 7.2 to 77.5 EB/month) from 2017 by the end of 2022 with an exponential increasing trend in 2020s accompanied by a similar growth of the number of networked devices that by the end of 2022 is expected to reach 28.5B. Additionally, the next generation networks are envisioned to inherently support a huge dynamic and diverse range of novel usage scenarios and applications that require, except from extremely high-data rates, agility, flexibility, availability, reliability, zero response time and artificial intelligence (AI) under a security and privacy umbrella. Virtual/artificial/extended reality, three-dimensional (3D) printing, self-driven cars, and generally cyber physical systems for intelligent transportation, smart traffic, Industry 5.0 and e-health are only a few indicative examples of several highly anticipated use cases. Despite the adaptation of several game changing design principles by the 5G, such as network densification, virtualization, and orchestration, which aim at enhancing the scalability, flexibility as well as efficient resource management, it has been proven that the envelop of its capabilities is defined by the available bandwidth, transmission and processing delay, as well as spectral and energy efficiency.
To overcome the aforementioned limitations, networks B5G need to focus on the vast unutilized and largely unallocated higher frequency resources in the optical band, where more than 500 GHz of continuous and exploitable bandwidth exist. This is expected to open the door to 1 Tbps connectivity by combining the large available bandwidth with spectral efficient transmission and access schemes. However, optical wireless links suffer from severe path and penetration losses that originate from the interaction and energy absorption by the molecules (mostly water) of the propagation medium in this band and result in low-transmission range and blockage. To enable a connectivity range of some tens of meters and support mesh topologies via midhaul/fronthaul and nomadic connectivity scenarios, high-direction connectivity is employed in both the transmitter (TX) and receiver (RX) side. This also allows the utilization of space-division multiple access (SDMA); thus, it can support a massive number of connected devices, and since solitary line-of-sight (LoS) links can be established, it provides an additionally inherent physical layer (PHY) security. On the other hand, it comes with the challenge of designing innovative low-latency device discovery and beam tracking for moving and mobile nodes. Moreover, to deal with blockage and to enable range expansion, coordinated multipoint (CoMP), relaying, and optical reconfigurable intelligent surface (ΟRIS)-aided approaches may be employed. Another particularity of wireless systems that operate in the optical band is due to the ultra-wideband extremely directional nature of the communications links and the non-uniform user’s spatial distribution, which results in inefficient user association when the classical minimum-distance criterion is employed. To counterbalance this, user association should be designed to meet the dominant key performance indicators (KPIs) of each user’s application. As a consequence, users may not be associated with the geographically closest base-station (BS)/access point (AP), since a better directional link may exist for a further away BS/AP. In other words, we need to rethink the conventional notion of hexagonal cell and transform it into “dynamic cell” that will not have a predefined share, and it will be established through the solution of multi-variety optimization problems. Of note, the solution of such problems may require the use of sophisticated AI approaches. Finally, B5G networks needs to support ultra-low latency applications. In this direction, grant-free or semi-GF access protocols need to be utilized as well as novel AI-based social-aware caching mechanisms.