ارائه ­دهنده:
فرزاد مرادی

  استاد راهنما:
دکتر سیدوحید ازهری
 دکتر وصال حکمی

  هیات داوران:
 دکتر احمد اکبری ازیرانی

 دکتر زینب موحدی
 دکتر مسعود صبائی
 دکتر حامد خانمیرزا 


 زمان :۱۴آبان  ماه ۱۴۰۳

  ساعت: ۱۷:۳۰
 

 مکان : دانشکده مهندسی کامپیوتر:  طبقه ۳ – اتاق دفاع

 چکیده پایان نامه :

:Abstract

Cell-free massive multiple-input multiple-output (CF-mMIMO) networks have been proposed as a new architecture for radio access networks (RANs), which consists of a large number of access points (APs) in a wide area that simultaneously serving user equipment (UEs) in the same frequency band. However, supplying energy to a large number of APs, as well as the impossibility or high cost of accessing the power grid in some scenarios, is a key challenge in fully releasing the potential and capabilities of CF-mMIMO networks. One of the solutions is resorting to different forms of renewable energy sources and using APs equipped with green energy harvesting capabilities. Due to the unpredictable nature of renewable energy resources, efficient energy resource allocation techniques need to be developed so that APs are always active.
In this thesis, we consider a CF-mMIMO network where different APs receive their power consumption from the power grid, a solar energy source, or both. Our goal is to design an AP-to-UE power allocation scheme with the aim of minimizing the power consumption of the power grid while also meeting the spectral efficiency (SE) constraints of UEs. In addition, we use the green energy budgeting mechanism to avoid energy shortages when solar energy is scarce. First, we propose the ideal offline problem of the target problem and transform it into a second-order cone programming problem (SOCP). Then, a light-weight model-free myopic heuristic algorithm named Myopic-Final and, afther that, a heuristic algorithm based on learning to optimization (L۲O) named Online-NN using neural networks is proposed. The purpose of using L۲O is to reduce the time complexity compared to offline and online algorithms, although time must be allocated to train the neural network before using it. The performed simulations show that in terms of average SE, the Myopic-Final algorithm provides almost the same performance as the offline algorithm and the Online-NN algorithm provides ۹۷% performance compared to the offline algorithm. In terms of energy consumption from the power grid, if the green energy budget and the energy in the batteries are sufficient, all three algorithms perform the same, otherwise Myopic-Final shows a slightly lower performance.
Keywords: Cell-free massive MIMO, energy harvesting, resource allocation, L۲O, neural network.