## Periodic Reporting for period 1 - BESMART (scalaBle and grEen wireleSs coMmunications for a sustAinable netwoRked socieTy)

Reporting period: 2017-10-01 to 2019-09-30

The develops innovative designs for 5G cellular networks, with reduced energy consumption and overhead requirements compared to available solutions. Moreover, the use of energy-harvesting capabilities for wireless devices is investigated, to make wireless devices greener and able to self-sustain without frequent battery charging. The project also studies the possibility of wirelessly transferring the power among network devices to enable devices with more battery energy to trade/transfer their energy to other devices nearby with low battery levels, in a bluetooth-like fashion. These achievements will allow:

a) reducing carbon dioxide emissions due to wireless communications, with a consequent rise of greener networks;

b) increasing the network coverage in remote/developing/emergency areas where currently it is commercially unattractive to do so;

c) prolonging the lifetime of wireless nodes reducing the need of connecting to fixed power lined to recharge the batteries.

Already today, wireless networks are responsible for 2% of the global emissions of carbon dioxide, and this number is expected to grow by 280% by 2020. 5G wireless networks, will have to serve more than 30 billion devices by 2020 and it is forecasted that Between 2020 and 2030, the number of connected devices will rise by 55% each year, reaching 607 exabytes in 2025 and 5,016 exabytes in 2030. It is clear that, without any corrective measures, this situation will lead to an energy crunch which poses sustainable growth as well as ecological and economical concerns.

The main conclusion of the action is that a careful design of wireless networks, especially by complementing traditional approaches with the use of artificial intelligence has the potential of making future wireless networks sustainable.

a) reducing carbon dioxide emissions due to wireless communications, with a consequent rise of greener networks;

b) increasing the network coverage in remote/developing/emergency areas where currently it is commercially unattractive to do so;

c) prolonging the lifetime of wireless nodes reducing the need of connecting to fixed power lined to recharge the batteries.

Already today, wireless networks are responsible for 2% of the global emissions of carbon dioxide, and this number is expected to grow by 280% by 2020. 5G wireless networks, will have to serve more than 30 billion devices by 2020 and it is forecasted that Between 2020 and 2030, the number of connected devices will rise by 55% each year, reaching 607 exabytes in 2025 and 5,016 exabytes in 2030. It is clear that, without any corrective measures, this situation will lead to an energy crunch which poses sustainable growth as well as ecological and economical concerns.

The main conclusion of the action is that a careful design of wireless networks, especially by complementing traditional approaches with the use of artificial intelligence has the potential of making future wireless networks sustainable.

Journal papers:

[J.1] M. Di Renzo, A. Zappone, T. T. Lam, M. Debbah, “System-Level Modeling and Optimization of the Energy Efficiency in Cellular Networks—A Stochastic Geometry Framework”, IEEE Transactions on Wireless Communications, 2018

[J.2] A. Zappone, L. Sanguinetti, M. Debbah, “Energy-Delay Efficient Power Control in Wireless Networks”, IEEE Transactions on Communications, 2018

[J.3] S. D'Oro, A. Zappone, S. Palazzo, M. Lops, “A Learning Approach for Low-Complexity Optimization of Energy Efficiency in Multicarrier Wireless Networks”, IEEE Transactions on Wireless Communications, 2018

[J.4] M. Sinaie, P.-H. Lin, A. Zappone, P. Azmi, E. A. Jorswieck, “Delay-Aware Resource Allocation for 5G Wireless Networks With Wireless Power Transfer”, IEEE Transactions on Vehicular Technology, 2018

[J.5] A. Zappone, M. Di Renzo, M. Debbah, “Wireless Networks Design in the Era of Deep Learning: Model-Based, AI-Based, or Both?”, IEEE Transactions on Communications, 2019

[J.6] A. Zappone, M. Di Renzo, M. Debbah, T. T. Lam, X. Qian, “Model-Aided Wireless Artificial Intelligence: Embedding Expert Knowledge in Deep Neural Networks Towards Wireless Systems Optimization”, IEEE Vehicular Technology Magazine, 2019

Conference papers:

[C.1] M. Di Renzo, A. Zappone, T. T. Lam, M. Debbah, “Stochastic Geometry Modeling of Cellular Networks: A New Definition of Coverage and Its Application to Energy Efficiency Optimization”, EUSIPCO 2018

[C.2] S. D'Oro, A. Zappone, S. Palazzo, M. Lops, “A learning-based approach to energy efficiency maximization in wireless networks”, WCNC 2018

[C.3] M. Sinaie, P.-H. Lin, A. Zappone, P. Azmi, E. A. Jorswieck, “Resource Allocation in OFDM-Based SWIPT with Statistical Delay Constraints”, GLOBECOM 2017

[C.4] A. Zappone, M. Debbah, Z. Altman, “Online Energy-Efficient Power Control in Wireless Networks by Deep Neural Networks”, SPAWC 2018

[C.5] A. Zappone, L. Sanguinetti, M. Debbah, “User Association and Load Balancing for Massive MIMO through Deep Learning”, ASILOMAR 2018

[C.6] L. Sanguinetti, A. Zappone, M. Debbah, “Deep Learning Power Allocation in Massive MIMO”, ASILOMAR 2018

All publications are freely available either on Arxiv or on ResearchGate. The system setup is that of a multi-cellular network, as shown in Fig. 1. The goal is to determine the optimal transmit power of each node to maximize the energy efficiency of the whole network or of a single communication link, defined as the ratio between the amount of bits that the link/network can transmit without errors and the corresponding energy consumption.

In [J.1] [C.1] new models for energy efficiency in 5G cellular networks are developed. It is proved that a single optimal base station density and a corresponding single optimal transmit power value exist in order to maximize the energy efficiency.

[J.2] employs the tool of game theory to model the interactions among non-cooperating nodes in a distributed network, proveing that a single equilibrium point exists and also providing an algorithm to reach it. However, at the equilibrium a gap exists with respect to the network energy efficiency that could be achieved if all terminals cooperated with each other.

[J.3] [C.2] aim at closing the gap evidenced in [J.2]. An improved power control algorithm is proposed that enjoys near-optimal performance, with a significantly lower complexity than state-of-the-art approaches.

[J.4] [C.3] studies wireless power transfer among devices in a 5G network. Each device transmits both an informational signal and energy signal for the receiver. An algorithm is developed to compute the optimal split between the two signals and perform optimal power allocation among the network nodes.

[J.5] [J.6] [C.4] [C.5] [C.6] investigate the use of AI with wireless networks. It is shown that the joint use of AI and of mathematical modeling improves the performance-complexity trade-offs compared to available methods.

Major exploita

[J.1] M. Di Renzo, A. Zappone, T. T. Lam, M. Debbah, “System-Level Modeling and Optimization of the Energy Efficiency in Cellular Networks—A Stochastic Geometry Framework”, IEEE Transactions on Wireless Communications, 2018

[J.2] A. Zappone, L. Sanguinetti, M. Debbah, “Energy-Delay Efficient Power Control in Wireless Networks”, IEEE Transactions on Communications, 2018

[J.3] S. D'Oro, A. Zappone, S. Palazzo, M. Lops, “A Learning Approach for Low-Complexity Optimization of Energy Efficiency in Multicarrier Wireless Networks”, IEEE Transactions on Wireless Communications, 2018

[J.4] M. Sinaie, P.-H. Lin, A. Zappone, P. Azmi, E. A. Jorswieck, “Delay-Aware Resource Allocation for 5G Wireless Networks With Wireless Power Transfer”, IEEE Transactions on Vehicular Technology, 2018

[J.5] A. Zappone, M. Di Renzo, M. Debbah, “Wireless Networks Design in the Era of Deep Learning: Model-Based, AI-Based, or Both?”, IEEE Transactions on Communications, 2019

[J.6] A. Zappone, M. Di Renzo, M. Debbah, T. T. Lam, X. Qian, “Model-Aided Wireless Artificial Intelligence: Embedding Expert Knowledge in Deep Neural Networks Towards Wireless Systems Optimization”, IEEE Vehicular Technology Magazine, 2019

Conference papers:

[C.1] M. Di Renzo, A. Zappone, T. T. Lam, M. Debbah, “Stochastic Geometry Modeling of Cellular Networks: A New Definition of Coverage and Its Application to Energy Efficiency Optimization”, EUSIPCO 2018

[C.2] S. D'Oro, A. Zappone, S. Palazzo, M. Lops, “A learning-based approach to energy efficiency maximization in wireless networks”, WCNC 2018

[C.3] M. Sinaie, P.-H. Lin, A. Zappone, P. Azmi, E. A. Jorswieck, “Resource Allocation in OFDM-Based SWIPT with Statistical Delay Constraints”, GLOBECOM 2017

[C.4] A. Zappone, M. Debbah, Z. Altman, “Online Energy-Efficient Power Control in Wireless Networks by Deep Neural Networks”, SPAWC 2018

[C.5] A. Zappone, L. Sanguinetti, M. Debbah, “User Association and Load Balancing for Massive MIMO through Deep Learning”, ASILOMAR 2018

[C.6] L. Sanguinetti, A. Zappone, M. Debbah, “Deep Learning Power Allocation in Massive MIMO”, ASILOMAR 2018

All publications are freely available either on Arxiv or on ResearchGate. The system setup is that of a multi-cellular network, as shown in Fig. 1. The goal is to determine the optimal transmit power of each node to maximize the energy efficiency of the whole network or of a single communication link, defined as the ratio between the amount of bits that the link/network can transmit without errors and the corresponding energy consumption.

In [J.1] [C.1] new models for energy efficiency in 5G cellular networks are developed. It is proved that a single optimal base station density and a corresponding single optimal transmit power value exist in order to maximize the energy efficiency.

[J.2] employs the tool of game theory to model the interactions among non-cooperating nodes in a distributed network, proveing that a single equilibrium point exists and also providing an algorithm to reach it. However, at the equilibrium a gap exists with respect to the network energy efficiency that could be achieved if all terminals cooperated with each other.

[J.3] [C.2] aim at closing the gap evidenced in [J.2]. An improved power control algorithm is proposed that enjoys near-optimal performance, with a significantly lower complexity than state-of-the-art approaches.

[J.4] [C.3] studies wireless power transfer among devices in a 5G network. Each device transmits both an informational signal and energy signal for the receiver. An algorithm is developed to compute the optimal split between the two signals and perform optimal power allocation among the network nodes.

[J.5] [J.6] [C.4] [C.5] [C.6] investigate the use of AI with wireless networks. It is shown that the joint use of AI and of mathematical modeling improves the performance-complexity trade-offs compared to available methods.

Major exploita

Main contributions.

1) New holistic energy consumption models and energy-efficient designs for 5G networks.

2) New algorithms for the design for sustainable wireless networks powered by clean and renewable energy sources.

3) New design approaches for wireless networks with wireless power transfer and energy-sharing among devices.

The impact of the project has been in helping the rise of greener networks through. It is estimated that by 2020, 75% of the ICT sector will be wireless, being worth 5% of the European GDP, and that the use of energy more efficient in everyday life could save €1000 per household annually, creating 2 million jobs, reducing annual greenhouse gas emissions by 740 million tons. Lower energy demands will make Europe less reliant on the energy imported from other countries (e.g. Russia and Middle East). In 2015, the EU has imported 53% of the energy it has consumed. Increasing the penetration of renewable energy to 30% by 2030 could save €450 billion.

1) New holistic energy consumption models and energy-efficient designs for 5G networks.

2) New algorithms for the design for sustainable wireless networks powered by clean and renewable energy sources.

3) New design approaches for wireless networks with wireless power transfer and energy-sharing among devices.

The impact of the project has been in helping the rise of greener networks through. It is estimated that by 2020, 75% of the ICT sector will be wireless, being worth 5% of the European GDP, and that the use of energy more efficient in everyday life could save €1000 per household annually, creating 2 million jobs, reducing annual greenhouse gas emissions by 740 million tons. Lower energy demands will make Europe less reliant on the energy imported from other countries (e.g. Russia and Middle East). In 2015, the EU has imported 53% of the energy it has consumed. Increasing the penetration of renewable energy to 30% by 2030 could save €450 billion.