Energy Efficiency in Open RAN with Smart RF Channel Tuning

Energy efficiency (EE) has recently emerged as a key requirement in mobile networks. Intelligent algorithms that enable EE capabilities in Radio Access Networks (RAN) can be developed by leveraging Key Performance Indicators (KPIs) and implementing appropriate control measures, such as a cell on/off switching or Smart RF Channel Tuning, also called RF Channel Reconfiguration. These capabilities, combined with a Machine Learning (ML) framework, are incorporated into the O-RAN architecture.

This blog post provides an overview of the EE framework based on use cases defined by the O-RAN ALLIANCE, followed by the implementation of the Energy Saving rApp (ES-rApp). The rApp employs Deep Q-Learning to enhance EE through intelligent RF channel reconfiguration.

(Note: The blog is based on our recent IEEE Access paper)

(Note: You can find more about energy saving in our whitepaper: „The O-RAN Whitepaper 2023 – Energy Efficiency in O-RAN”)

Energy Efficiency within O-RAN

Energy savings often come at the expense of reduced available resources, which can affect the overall system performance. Therefore, optimization goals are typically set to balance both factors. EE is usually defined as the ratio of average user throughput to average power consumption. Monitoring EE allows MNOs to manage the trade-off between throughput and power consumption, helping to reduce energy costs or carbon footprints. However, for effective and context-specific EE optimization, algorithms need access to network interfaces for both monitoring and control.

EE is also a key requirement in O-RAN standardization and development efforts. O-RAN ALLIANCE has developed several use cases [12], including Cell On/Off Switching (COOS), RF Channel Reconfiguration (RCR), particularly applied in M-MIMO scenarios, commonly referred to as antenna selection, Advanced Sleep Mode Selection (ASM), and O-Cloud Resource Energy Saving Mode (see Fig. 1.)

In the case of COOS, the concept involves shutting down or putting an entire cell into sleep mode. In the figure above, pico/small cells serve as the capacity layer and can be switched on or off based on traffic demands. It’s important to note that COOS doesn’t apply exclusively to pico/small cells but refers more broadly to a „cell” in general. This means it can be used in various scenarios, such as turning off/on a standard carrier or a Component Carrier in a Carrier Aggregation setup. The cells being switched off are capacity cells, while the coverage cell must remain active at all times. (Note: You can find out more about COOS in these blogs: O-RAN Network Energy Saving: Cell Switching On/Off, Energy Saving-rApp control over Traffic Steering-xApp)

In the RCR, instead of shutting down an entire cell, the algorithm turns off or on certain RF channels, which correspond to the number of beams in a M-MIMO cell. When traffic is low, the number of active beams can be reduced without negatively impacting users’ QoS. (Note: You can find out more about RCR in this blog: O-RAN Network Energy Saving: RF Channel Switching)

The Proposed ES-rApp for RF Channel Reconfiguration

When considering ES applications, it is important to note that reconfiguring the O-RAN Radio Unit (O-RU), whether by switching on/off or changing the RF Channel Configuration, requires some transition time and may even necessitate a device restart. Thus, the ES algorithm is proposed as an ES-rApp. A generalized framework for deploying the ES-rApp is illustrated in Fig. 2.

The rApp is hosted within the Non-RT RIC, which resides inside the Service Management and Orchestration (SMO) platform, alongside a Collection and Control (CC) entity. These components work together to provide the ES-rApp with data collected from E2 Nodes, enabling it to perform control actions such as reconfiguring RF channels.

The left side of Figure 2 illustrates the signaling flow between the Non-RT RIC, featuring the ES-rApp, and other O-RAN entities. Initially, the ES-rApp retrieves information about the supported configurations of the O-RU, such as available sleep modes and RCC options. This is achieved through the hierarchical connection using the O1 interface between the SMO and the E2 Node (e.g., a base station connected via the E2 interface to the RIC), as well as the Open Fronthaul (O-FH) interface between the E2 Node and the O-RU. Next, the ES-rApp is set up to receive data from the E2 Node and O-RU in compliance with 3GPP, enabling it to monitor metrics such as cell load, power consumption, and user throughput.

The ES-rApp employs ML to process the collected data dynamically, enabling it to switch cells on/off or execute RCR. The resulting decision is communicated back to the Non-RT RIC, which then uses the CC entity to adjust the O-RU configuration via the O1 and O-FH interfaces. RCR impacts network performance, so in the final stage, the ES-rApp leverages the pre-established O1 connection to assess network performance and, if necessary, retrain the ML model.

Predicting network performance following RCR actions is challenging using analytical methods alone. Therefore, the utilization of a Reinforcement Learning (RL) framework is proposed, as illustrated in the right part of Fig. 2. In this, an „agent” (the ES-rApp) interacts with its environment (E2 Nodes, O-RUs) by making decisions on O-RU reconfigurations based on the current state (e.g., Key Performance Indicators (KPIs) such as cell load). The agent then observes the „reward,” which could be KPIs like power consumption or user throughput. This process is cyclic—the agent identifies the state, takes an action, and observes the resulting reward. Through multiple iterations, the rApp learns how to optimally configure the O-RU to maximize rewards, such as improved EE.

To find out about the simulation setup and results, see [1].

How RCR Can Cooperate with the Corresponding xApp

ES features need to interact with other RRM algorithms to ensure coordinated decision-making. For instance, the decision to switch off certain resources must be carefully synchronized with other processes. Fig. 3 illustrates a scenario where a Beam Mobility Management xApp (BMM-xApp) operates within a near-RT control loop, while the ES-rApp functions within a non-RT control loop. This coordination is crucial to maintaining both EE and optimal network performance.

As already mentioned, the ES-rApp is designed to intelligently adjust the antenna array configuration based on the network state, aiming to maximize EE. Meanwhile, the BMM-xApp is responsible for switching users between beams based on their location, to reduce the number of beam reselections and radio link failures (RLFs). These two apps cooperate hierarchically. The ES-rApp handles the reconfiguration of the M-MIMO base station over relatively long intervals, while the BMM-xApp makes quick decisions regarding user-to-beam associations, utilizing a Radio Environment Map (REM) that represents the spatial distribution of received power for each beam.

When the ES-rApp changes the array configuration, it alters the number and shape of the beams. As a result, the BMM-xApp must be equipped with different REM sets suitable for each possible array configuration. When the ES-rApp initiates a configuration change, it sends an appropriate notification to the BMM-xApp to ensure proper coordination.

It’s important to note that when multiple apps operate within the same system, conflicts can arise. Beyond obvious conflicts, such as two apps attempting to modify the same parameters, more complex challenges can emerge—like different objectives for each app. For instance, one app may focus on maximizing energy savings, while another prioritizes optimizing user performance. If these apps work concurrently, they may continuously alter network behavior, causing instabilities and hindering their ability to converge on solutions [20]. Such conflicts should be addressed at multiple levels: between two rApps, between two xApps, or between an rApp and an xApp. One potential solution is to design a set of apps that inherently avoid conflicts and work in a coordinated manner, as is the case in this example.

Summary

The O-RAN architecture plays a crucial role in enabling ES features in 5G and future networks by providing a unified platform for deploying ES-focused rApps and xApps. This environment includes interfaces for data collection and control, as well as ML capabilities for model training, deployment, and inference. The O-RAN specifications offer detailed descriptions of ES use cases, including message flows between involved entities and the ML framework.

Building on this foundation, we have developed an ES-rApp that performs RCR using Deep Q-Learning to optimize EE. By leveraging reinforcement learning (RL), the ES-rApp can intelligently recognize hidden dependencies between O-RU configuration, power consumption, throughput, PRB utilization, and user distribution across beams, allowing it to dynamically adjust the number of active RF channels. This enables efficient reconfiguration of the O-RU, achieving significant energy savings during low-traffic periods while maintaining high QoS during peak times.

Future work will focus on scaling the solution to larger simulation environments (involving dozens of base stations), integrating the algorithm into a realistic setup with a commercial-grade RIC and RAN emulator, and considering varying QoS profiles and real-world traffic patterns. Another key area will be analyzing how multiple Apps can collaborate to achieve energy savings harmoniously within the RIC hierarchy through joint rApp/xApp operations. Finally, conflict mitigation will be a critical consideration as the RIC framework supports diverse Apps from various vendors with different objectives.

Note

This post describes an EE framework according to the use cases specified by the O-RAN ALLIANCE followed up by the implementation of the Energy Saving rApp (ES-rApp) by Rimedo Labs and is based on the article: „Energy Efficiency in Open RAN: RF Channel Reconfiguration Use Case” published by IEEE Access [1]. You may find a larger discussion on the detailed design and simulation results of the proposed approach in the corresponding article.

References

[1] Energy Efficiency in Open RAN: RF Channel Reconfiguration Use Case | IEEE Journals & Magazine | IEEE Xplore
[2] S. Han, T. Xie and I. Chih-Lin, „Greener physical layer technologies for 6G mobile communications”, IEEE Commun. Mag., vol. 59, no. 4, pp. 68-74, Apr. 2021
[3] D. López-Pérez, A. De Domenico, N. Piovesan, G. Xinli, H. Bao, S. Qitao, et al., „A survey on 5G radio access network energy efficiency: Massive MIMO lean carrier design sleep modes and machine learning”, IEEE Commun. Surveys Tuts., vol. 24, no. 1, pp. 653-697, 1st Quart. 2022
[4] TR, 38.864 v.18.1.0, „Study on Network Energy Savings for NR (Release 18)”, Mar. 2023
[5] A. Kaushik, R. Singh, S. Dayarathna, R. Senanayake, M. Di Renzo, M. Dajer, et al., „Toward integrated sensing and communications for 6G: Key enabling technologies standardization and challenges”, IEEE Commun. Standards Mag., vol. 8, no. 2, pp. 52-59, Jun. 2024
[6] A. Bazzi and M. Chafii, „On outage-based beamforming design for dual-functional radar-communication 6G systems”, IEEE Trans. Wireless Commun., vol. 22, no. 8, pp. 5598-5612, Jun. 2023
[7] L. Bonati, S. D’Oro, M. Polese, S. Basagni and T. Melodia, „Intelligence and learning in O-RAN for data-driven NextG cellular networks”, IEEE Commun. Mag., vol. 59, no. 10, pp. 21-27, Oct. 2021
[8] Open RAN technical priorities focus on energy efficiency, Jun. 2021, [online] Available: https://cdn.prod.website-files.com/60b1962ffda0a42f779c765b/60cd8e6f8dd92c298b891216_Open%2BRAN%2BTechnical%2BPriority%2BDocument%2B-%2BFinal%2BVersion.xlsx.
[9] M. Hoffmann, S. Janji, A. Samorzewski, Ł. Kułacz, C. Adamczyk, M. Dryjański, et al., „Open RAN xApps design and evaluation: Lessons learnt and identified challenges”, IEEE J. Sel. Areas Commun., vol. 42, no. 2, pp. 473-486, Feb. 2024
[10] J. Thaliath, S. Niknam, S. Singh, R. Banerji, N. Saxena, H. S. Dhillon, et al., „Predictive closed-loop service automation in O-RAN based network slicing”, IEEE Commun. Standards Mag., vol. 6, no. 3, pp. 8-14, Sep. 2022
[11] A. Giannopoulos, S. Spantideas, N. Kapsalis, P. Gkonis, L. Sarakis, C. Capsalis, et al., „Supporting intelligence in disaggregated open radio access networks: Architectural principles AI/ML workflow and use cases”, IEEE Access, vol. 10, pp. 39580-39595, 2022
[12] O-RAN Network Energy Savings Use Cases Technical Report 1.0, Mar. 2023
[13] B. Lin, F. Gao, S. Zhang, T. Zhou and A. Alkhateeb, „Deep learning-based antenna selection and CSI extrapolation in massive MIMO systems”, IEEE Trans. Wireless Commun., vol. 20, no. 11, pp. 7669-7681, Nov. 2021
[14] N. Rajapaksha, J. Mohammadi, S. Wesemann, T. Wild and N. Rajatheva, „Minimizing energy consumption in MU-MIMO via antenna muting by neural networks with asymmetric loss”, IEEE Trans. Veh. Technol., vol. 73, no. 5, pp. 6600-6613, May 2024
[15] P. Frenger and K. W. Helmersson, „Massive MIMO muting using dual-polarized and array-size invariant beamforming”, Proc. IEEE 93rd Veh. Technol. Conf. (VTC-Spring), pp. 1-6, Apr. 2021
[16] 3GPP, TS 28.552 v.18.2.0, „Management and Orchestration; 5G Performance Measurements”, Mar. 2023
[17] M. Hoffmann and P. Kryszkiewicz, „Beam management driven by radio environment maps in O-RAN architecture”, Proc. IEEE Int. Conf. Commun. Workshops (ICC Workshops), pp. 54-59, 2023
[18] O-RAN Architecture Description v.8.0, Mar. 2023
[19] M. Dryjanski and M. Szydelko, „A unified traffic steering framework for LTE radio access network coordination”, IEEE Commun. Mag., vol. 54, no. 7, pp. 84-92, Jul. 2016
[20] C. Adamczyk and A. Kliks, „Conflict mitigation framework and conflict detection in O-RAN near-RT RIC„, IEEE Commun. Mag., vol. 61, no. 12, pp. 199-205, Dec. 2023
[21] B. Shah and H. Bhavsar, „Time complexity in deep learning models”, Proc. Comput. Sci., vol. 215, pp. 202-210, Dec. 2022
[22] TS, 38.901 v.17.0.0, 3GPP, „Study on Channel Model for Frequencies From 0.5 to 100 GHz (Release 17)”, Mar. 2022
[23] C. Zhang, X. Chen, H. Yin and G. Wei, „Two-dimensional shadow fading modeling on system level”, Proc. IEEE 23rd Int. Symp. Pers. Indoor Mobile Radio Commun. (PIMRC), pp. 1671-1676, Sep. 2012
[24] ETSI, ES 202 706-1, v1.7.1, ETSI, „Environmental Engineering (EE); Metrics and Measurement Method for Energy Efficiency of Wireless Access Network Equipment; Part 1: Power Consumption—Static Measurement Method”, Aug. 2022

Acknowledgment

This blog post is a cooperative work between Marcin Dryjanski and Marcin Hoffmann based on their joint paper [1].

Author Bio