Cell-Free Massive MIMO in O-RAN: Energy-Aware Joint Orchestration of Cloud, Fronthaul, and Radio Resources
This code package contains a simulation environment, based on Matlab, that reproduces some of the numerical results and figures in the article
Özlem Tuğfe Demir, Meysam Masoudi, Emil Björnson, and Cicek Cavdar “Cell-Free Massive MIMO in O-RAN: Energy-Aware Joint Orchestration of Cloud, Fronthaul, and Radio Resources),” IEEE Journal on Selected Areas in Communications, vol. 42, no. 2, pp. 356-372, Feb. 2024, doi: 10.1109/JSAC.2023.3336187.
For the energy-efficient deployment of cell-free massive MIMO functionality in a practical wireless network, the end-to-end (from radio site to the cloud) energy-aware operation is essential. In line with the cloudification and virtualization in the open radio access networks (O-RAN), it is indisputable to envision prospective cell-free infrastructure on top of the O-RAN architecture. In this paper, we explore the performance and power consumption of cell-free massive MIMO technology in comparison with traditional small-cell systems, in the virtualized O-RAN architecture. We compare two different functional split options and different resource orchestration mechanisms. In the end-to-end orchestration scheme, we aim to minimize the end-to-end power consumption by jointly allocating the radio, optical fronthaul, and virtualized cloud processing resources. We compare end-to-end orchestration with two other schemes: i) "radio-only" where radio resources are optimized independently from the cloud and ii) "local cloud coordination" where orchestration is only allowed among a local cluster of radio units. We develop several algorithms to solve the end-to-end power minimization and sum spectral efficiency maximization problems. The numerical results demonstrate that end-to-end resource allocation with fully virtualized fronthaul and cloud resources provides a substantial additional power saving than the other resource orchestration schemes.
-
The article contains 6 simulation figures, numbered 3-8.
-
For Figures 3 and 4, "Figs3_4_mixed_integer_programming_solution.m" is run and 30 different setups are saved with file names "MIPsimX.mat" where "X" is the seed (setup) number. Later, "Figs3_4_plot_results.m" file is used to plot the results.
-
For mixed integer programming, CVX with Gurobi solver is utilized.
-
For Figures 5 and 6, "Figs5_6_CCP_SE_contraints_split8.m" and "Figs5_6_CCP_SE_contraints_split7_2.m" are run to save the mat files "J_SE_sim_8_X.mat" and "J_SE_sim_72_X.mat" for different seed numbers "X" among 30 random setups when there are 16 UEs. Moreover, by setting "SEAll =2" in the code files "Figs5_6_CCP_SE_contraints_split8.m" and "Figs5_6_CCP_SE_contraints_split7_2.m" and for K = 4, 8, 12 UEs, "J_SE2_KY_sim_8_X.mat" and "J_SE2_KY_sim_72_X.mat" for different number of UEs "Y" are saved. These mat files are used by "Figs5_6_plot_results.m" to generate the figures.
-
Figures 7 and 8 are generated by first saving the related mat files with different penalty parameters with "Figs_7_8_CCP_sumSE_functional_split8.m", "Figs_7_8_CCP_sumSE_functional_split72.m", "Figs_7_8_CCP_sumSE_functional_split8_onlySumRate.m", "Figs_7_8_CCP_sumSE_functional_split72_onlySumRate.m". The figures are generated by using these results and the script file "Fig7_8_plot_results.m".
-
The package also contains other Matlab functions used by some of the scripts.
This work was partially funded by CELTIC-NEXT project AI4Green with the support of Vinnova, Swedish Innovation Agency.
This code package is licensed under the GPLv2 license. If you in any way use this code for research that results in publications, please cite our original article listed above.