Periodic Reporting for period 1 - EXPLOR (EXperimentation and simulation based PLatform for beyond 5G Optical-wireless network Research and development)
Periodo di rendicontazione: 2020-01-01 al 2021-12-31
Several 5G and Optical transmission toolsets are being widely used for research, design and tutoring purposes, yet individually fragmented and subjected to limitations in terms of performance and cost. The interoperability between individual software toolsets is typically prohibitively limited. EXPLOR will counter those drawbacks.
EXPLOR project hinges on inter-sectoral collaboration and pronounced training and knowledge transfer between the multidisciplinary academic and industrial partners.
In our SG modeling segment, we first present the predominant multi-tier network model of independent Poisson point processes (PPPs) utilizing multiple spectrum bands, along with an example use case for using the clustered point processes for modeling distances and proximity in mobile data networks. We further provide a review of existing approaches relevant to MMW femtocell operation in the context of THz communications. Our current activities beyond SOTA target the extension of our approach towards 3D modeling using Poisson cluster process (PCP) and the inclusion of mobility (RWP), with the particular focus on the angle estimation, beam management and handover decisions.
Our beyond SOTA activities on OFH and ML based optimization modeling currently focus on the development of OFH frameworks tailored for successful integration of neural network based ML optimization methods, as well as the system level mitigation of main transmission impairments in EXPLOR scenario. We have developed an integrated modelling framework for the application of ANN based signal equalization, enabling current research targeting beyond SOTA architecture optimization The mitigation of transmission impairments in the optical segment of EXPLOR scenario requires signal processing that can effectively adapt to varying transmission conditions imposed on continuous streams of data. We are thus currently pursuing the application of novel learning approaches in deep neural networks (DNN) for linear and nonlinear equalization in long haul transmission systems. This work investigates and proposes a novel DNN architecture to overcome a problem of catastrophic forgetting in deep neural networks that can be used for the mitigation of critical impairments in EXPLOR scenario. We perform comparisons against a number of SOTA methods and traditional DNN implementations and demonstrate improvements in training accuracy towards signal recovery and increased flexibility. The results, which are currently under a journal publication peer review, show that the method may be successfully used towards the increase in transmission reach.