Sparsification Fixed-Point Algorithm for Efficient Ergodic Capacity Computation

Ziyuan Lyu,Boxiang Ren,Han Hao,Junyuan Wang,Hao Wu

DOI: https://doi.org/10.1109/iccc62479.2024.10681803

2024-01-01

Abstract:Ergodic capacity is a key performance indicator of wireless networks. The necessity for frequent capacity evaluation to guide network optimization highlights the pivotal importance of efficient and accurate ergodic capacity computation. Researchers have put forth the use of fixed-point iterations to compute the deterministic equivalent to the ergodic capacity. However, the quadratic computational cost of each iteration makes the whole iterative process burdensome, especially for future ultra-dense networks. In this paper, we apply the fixed-point iteration approach to compute the ergodic capacity of clustered cell-free networks. By employing an importance sparsification strategy, i.e., sampling from a dense matrix and rescaling the sampled entries according to their values to construct a sparse matrix, we reduce the computational cost of each fixed-point iteration to O(s), where s is the sample size. This yields a sparsification fixed-point algorithm (Spar-FPA). Theoretical analysis shows that the computation error resulted from sparsification is restricted with high probability. Simulations conducted across diverse settings further validate the effectiveness and efficiency of our method.