A Zeroth-Order Block Coordinate Gradient Descent Method For Cellular Network Optimization
ID:60 View Protection:ATTENDEE Updated Time:2022-10-17 15:10:35 Hits:525 Oral Presentation

Start Time:2022-10-21 16:00(Asia/Shanghai)

Duration:15min

Session:SS Special Session » SS6SS6: Data-Driven Methods for Real-World Wireless Network Modeling and Optimization

Video No Permission

Tips: Only the registered participant can access the file. Please sign in first.

Abstract
This paper considers a performance optimization problem in a cellular network. The objective is to optimize the network performance measured by reference signal received powers (RSRPs) and signal to interference plus noise ratios (SINRs) through the jointly configuration of the antenna angles (both tilt and azimuth), height, and transmit power of base stations. The problem is challenging in two aspects: 1) there is a lack of explicit expressions for both RSRPs and SINRs; and 2) huge amount of decision variables makes the problem highly complex. To address these issues, we formulate it as a soft coverage maximization problem and propose a zeroth-order block coordinate projected gradient descent (ZO-BCPGD) method, which only relies on the calculation of objective function values. The proposed ZO-BCPGD is proven to converge to stationary points under mild assumptions. The performance of the proposed algorithm is validated by numerical experiments.
Keywords
Speaker
Pengcheng He
Tongji University

Submit Comment
Verify Code Change Another
All Comments
Important Date
  • Conference Date

    Oct 19

    2022

    to

    Oct 22

    2022

Sponsored By
Zhejiang University
Organized By
Zhejiang University