Beamforming Optimization rApp – Design and Results

The O-RAN architecture introduces openness and intelligence into RAN by Non-RT RIC, Near-RT RIC, and the concept of rApps/xApps. Moreover, use cases, defined by the O-RAN ALLIANCE in [1], highlight the relevance of adaptive beamforming and M-MIMO optimization. These use cases call for mechanisms that adjust beam parameters such as azimuth and elevation, based on real-time KPIs or information from external sources (e.g., location information). xApps and rApps, which observe the state of the network, including UE measurements, cell loads, and interference between cells, can modify the rotation of the GoB accordingly. This aligns with the GoB optimization use cases discussed in our previous blog. In this blog post, we introduce the solution proposed by Rimedo Labs: Beamforming Optimization rApp (BO-rApp).

We discuss how the BO-rApp can be improved when utilizing location information or a Network Digital Twin (NDT) – a virtual replica of the network. Then, we point out a possible architecture for the BO-rApp within the O-RAN architecture.  Finally, we present numerical results demonstrating the benefits of the BO-rApp in an event-based scenario.

BO-rApp in the O-RAN Architecture

In the current O-RAN market, there is a shift towards SMO/Non-RT RIC, reflected by growing interest in automation platforms, like Ericsson EIAP or Amdocs Cognitive RAN Automation. In this context, we propose the BO-rApp for the Non-RT RIC with the decision-making logic for GoB adjustments (see Fig. 1). Its role is to analyse performance trends based on the Performance Measurement (PM) metrics obtained over the O1 interface, integrate the available external data like location information when needed, and determine the optimal orientation for the entire beam set. These decisions are translated into actionable configuration updates sent through the O1 interface to the O-DU or O-RU [2], depending on the implementation.

The rApp monitors network PMs such as RSRP and SINR distributions, as well as traffic load. Depending on the use case, BO-rApp can adjust GoB configuration dynamically or statically, depending on the operator’s needs, available data, and RAN/radio capabilities. When available, BO-rApp can incorporate additional sources of information, such as UE location (useful for fast dynamic GoB optimization – see here), or operator-provided policies. Based on such inputs, the BO-rApp computes the azimuth and elevation shifts for the GoB and applies the changes via the O1 interface. The BO-rApp can also be equipped with NDT. With this, BO-rApp can test the proposed decisions (GoB configurations) by executing them inside NDT and choosing the most promising configuration to be applied in the network. The advantage here is that the NDT does not operate in the actual real-world network, thus being more secure than applying the algorithm directly into real-life configuration. 

While the GoB optimization in a network, with hundreds of M-MIMO Base Stations (BS), Beam Optimization can become a complex problem. Therefore, the BO-rApp can make use of Agentic AI or Reinforcement Learning solutions to find the optimal solution. AI Agents can be deployed within the NDT and work on the synthetic data to propose initial configurations. However, it is crucial to incorporate feedback from the live network to assess their actions in the real world.

The BO-rApp utilizes the O1 interface to modify the GoB configuration; thus, this process is strictly related to the underlying 3GPP specifications. Specifically, TS 28.541 [3] describes the relevant configuration management and ensures that the BO-rApp complies with the standardized beamforming management. The key metrics that can be considered for BO-rApp are:

• CommonBeamformingFunction: digitalAzimuth, digitalTilt;
• Beam: beamIndex, beamType, beamAzimuth, beamTilt, beamHorizWidth, beamVertWidth;
• Cell, carrier, sector parameters like: bandwidth, ARFCN UL/DL, location, antenna parameters.

Additionally, the PMs observed by the BO-rApp over the O1 interface to adjust the GoB configuration, from 3GPP TS 28.552 [3], are:

• SS-SINR distribution per SSB and per gNB
• SS-RSRP distribution per SSB and per gNB
• SS-RSRQ distribution per SSB and per gNB

BO-rApp utilization for Event-based GoB deployment

Suppose we have an M-MIMO network optimized in a static way and there is an ongoing event involving hundreds of UEs located in a specific area. This fits the Event-Based GoB optimization use case as per our previous blog. We selected it to demonstrate the operation of the BO-rApp. To obtain the numerical results, we conducted a series of simulations involving 1000 UEs located near the M-MIMO BS, as presented in Fig. 2. The rApp utilizes NDT to examine a few options of GoB configuration and selects the best one associated with the highest UE-received SINR, which is a 15-degree rotation.

Fig. 3 and Fig. 4 provide the comparison between the SINR and RSRP distributions observed for the default GoB configuration (pointing towards 0 degrees), and the GoB configuration optimized by the BO-rApp for the location of the event. A clear improvement is seen when the GoB is rotated by 15 degrees. The average SINR increases by approximately 2 dB, indicating a significantly better alignment between the beam directions and the dominant UE cluster. An even bigger gain is in RSRP, where the improvement is up to 10 dB.

As a result, the highly directional characteristic of M-MIMO beams is beneficial in this scenario. By rotating the GoB, the rApp effectively recalibrates the radiated power toward the UE groups participating in the event, and thus improves both RSRP and SINR.

Conclusions

The Beamforming Optimization, or more specifically, GoB adjustment, can be realized within the O-RAN framework relying on the configuration structures and PMs defined by 3GPP. We propose a BO-rApp using the Ray-Tracing-based NDT for a safe evaluation of possible GoB configurations. In addition, incorporating feedback from the live network is important to validate optimization decisions in the real world. The proposed concept was proven by the simulation covering an event-based semi-static configuration.

In summary, the BO-rApp is a practical step toward intelligent beam management in O-RAN M-MIMO networks. It bridges the gap between the static GoB deployments and fully adaptive beamforming by introducing external intelligence, while preserving interoperability and standard compliance.

References

[1] O-RAN.WG1.MMIMO-USE-CASES-TR-v01.00

[2] O-RAN.WG1.O1-Interface.0-v04.00

[3] 3GPP, TS 28.541 “Management and orchestration; 5G Network Resource Model (NRM); Stage 2 and stage 3 ” v20.0.0, September, 2025

Acknowledgement

The author would like to thank Marcin Hoffmann for the valuable comments provided during the writing process.

Author Bio