Case Study: Load Balancing in 5G Using Cell Individual Offset (CIO)
Summary
This case study describes our work with a mobile network operator to use Cell Individual Offset (also termed Cell Specific Offset) for balancing User Equipment (UE) association in a 5G network. The aim was to achieve load balancing across base stations without any changes to hardware or to the physical deployment.
Simulations were conducted using NetSim to assess the impact of CIO adjustments on UE distribution and resource utilization. The telecom operator was able to evaluate gNB configuration changes for load balancing through controlled simulation prior to deployment.
Problem Statement
The network scenario consisted of three collocated gNBs—low-band, mid-band, and high-band—operating at 1.5 GHz, 2.6 GHz, and 3.5 GHz respectively. For simplicity, only one sector was modeled within a 500m × 500m environment, with all gNBs placed at the top-left corner.
A total of 500 UEs were distributed across the sector: 15% in the near-cell region, 25% in mid-cell, and 60% at the cell edge.
Due to the differences in propagation loss across frequency bands, UE association was skewed. The lower-frequency gNB attracted a higher number of UEs because of its larger coverage, leading to resource overload and congestion. In contrast, the higher-frequency gNB was underutilized.
A solution was needed to influence UE association without modifying the physical deployment or network architecture.
Approach
Cell Individual Offset (CIO) modifies User Equipment (UE) cell selection by artificially adjusting the Reference Signal Received Power (RSRP) measurements used in handover and reselection decisions.
A positive CIO (e.g., +4.77 dB) makes a cell appear stronger than it actually is, steering UEs to towards it, while a negative CIO (e.g., -4.77 dB) makes a cell appear weaker, steering UEs away from it. In this case study, applying a negative CIO to the congested low-band gNB and a positive CIO to the underutilized high-band gNB shifted UE attachments.
All other parameters, including scheduler behaviours, remained unchanged.
We measured UE distribution, PRB usage, and throughput impacts across simulations.
NetSim enabled rapid "what-if" testing of CIO values, ensuring optimal offsets were identified before deployment
Simulation Setup
Network Topology
Fig: A single sector with 3 collocated sector gNBs in the top left - operating in low, mid and high bands. There are 500 UEs with 15% near cell (green), 25% mid cell (yellow) and 60% cell edge (red).
Simulation parameters
| Parameter | Value |
|---|---|
| Simulation Area | 600 m x 500 m |
| Number of gNBs | 3 (sectorized) |
| Number of UEs | 500 (15% near, 25% mid, 60% edge) |
| Channel Bandwidth | 40 MHz (each gNB) |
| Path Loss Model | Log distance (pathloss exponent = 3.8) |
| Shadowing | Log-normal |
| Simulation Time | 100 seconds |
| Traffic | Downlink, full buffer |
| Antenna | 4T4R |
| Transmit power | 80W/port |
| CIO settings |
gNB 1 (1.5 GHz): - 4.77 dB gNB 2 (2.6 GHz): 0 dB gNB 3 (3.5 GHz): + 4.77 dB |
Asymmetric offsets (-4.77 dB low-band, +4.77 dB high-band) compensated for path loss disparity at 3.5 GHz vs. 1.5 GHz
Results
UE Association Distribution
| gNB ID | Band | UEs (Before load balancing) |
% | UEs (After load balancing) |
% | Change |
|---|---|---|---|---|---|---|
| gNB1 | Low | 321 | 64.2 | 251 | 50.2 | ▼ 14.0 |
| gNB2 | Mid | 122 | 24.4 | 116 | 23.2 | ▼ 1.2 |
| gNB3 | High | 57 | 11.4 | 133 | 26.6 | ▲ 15.2 |
After CIO adjustment, the number of UEs associated with gNB1 decreased, while the number associated with gNB3 increased. The total number of UEs in the simulation remained unchanged at 500.
Initial PRB Utilization (without CIO)
PRB Utilization After CIO
| gNB | Without CIO | With CIO |
|---|---|---|
| gNB1 | ~100% | ~70% |
| gNB2 | ~30% | ~35% |
| gNB3 | ~25% | ~40% |
Following CIO adjustments, PRB utilization at gNB1 decreased from near full allocation to approximately 70%. At gNB3, PRB utilization increased from 25% to around 40%.
Observations
CIO adjustments altered UE association patterns, resulting in a decrease in load on the lowest-frequency gNB and an increase in resource utilization on the higher-frequency gNBs. This redistribution was achieved without changes to the scheduling algorithm, radio configuration, or physical network topology.
Conclusion
Scenarios were simulated in NetSim by applying frequency-specific CIO values to redistribute UE associations and balance PRB allocation across gNBs. The method is defined in 3GPP specifications and can be implemented in Self-Organizing Network (SON) systems for automated parameter control.
Results showed successful redistribution of 5G network load without hardware changes, reducing low-band congestion from 100% to 70% and boosting high-band utilization from 25% to 40%.