RimedoTheORANWhitepaper EnergyEfficiency

The O-RAN
Whitepaper 2023
Energy Efficiency in O-RAN
Recently, within the Radio Access Networks (RAN) domain, energy

efficiency has been pointed out as one of the key requirements
moving forward. It has been defined as end-to-end involving all
domains within the O-RAN system. The overall objective for O-RAN
is to gradually become more energy efficient than the regular
RAN. This should not be at a cost of the key O-RAN concepts
including cloudification and disaggregation. This whitepaper
provides an overview of the energy efficiency topic in 5G mobile
networks and how it fits into the O-RAN area. This is followed up by
example algorithms utilizing Machine Learning (ML) schemes and
their realization in the form of rApps based on O-RAN use cases.
FE B R UA RY 2023
The O -RAN Whitepaper 2023 Table of Contents
Table of Contents
Executive Summary 03
1.0 Introduction to Energy Efficiency 04
2.0 O-RAN as an Enabler for Energy Efficiency in 5G Networks 08
3.0 O-RAN Network Energy Efficiency: Cell Switching On/Off 14
4.0 O-RAN Network Energy Efficiency: RF Channel Switching 20
Summary & Conclusions 27
References 30
Glossary 32
2 www.rimedolabs.com
The O -RAN Whitepaper 2023 Executive Summar y
Executive Summary
Energy efficiency is one of the key aspects within the 5G and beyond mobile
networks. Specifically in the RAN, it is currently on the agenda of the indus-
try and research activities. As the industry is targeting higher and higher
throughputs within 5G and beyond, an increasing number of energy-con-
suming hardware must be deployed. From this perspective, energy-sav-
ing mechanisms are of crucial role. O-RAN ALLIANCE is currently providing
means to support energy-saving topics as one of the key pillars for moving
forward within the RAN domain. This whitepaper provides a technical dis-
cussion of one of the currently important aspects related to the Open Radio
Access Network (Open RAN)¹, namely Energy Efficiency (EE).
The first chapter provides an overview of the energy efficiency topic within
wireless mobile networks along with the take on from the O-RAN perspec-
tive. It serves as a starting point for more details to be provided in the fol-
lowing parts.
Then, the whitepaper discusses the ways to save energy on different lev-
els e.g, cell/base station switching off/on, Massive MIMO (M-MIMO) antenna
configuration changes, user association, etc.
This is followed up by example algorithms utilizing ML schemes and their

realization in the form of rApps based on O-RAN-defined use cases. Those
include intelligent and ML-based cell switch off/on and RF channel switch
off/on. The example implementation of the algorithms in the form of rApps
is presented along with challenges including the need for coordination of
those with other surrounding algorithms.
The whitepaper ends with a summary and conclusions section along with a
glossary of the used terms.
¹ The general trend of disaggregated and opening radio networks is called Open RAN. O-RAN,
in turn, refers to the O-RAN ALLIANCE-specified architecture and framework.
The O -RAN Whitepaper 2023 1 .0 Introduction to Energy Efficiency
1.0
Introduction to
Energy Efficiency
Recently, within the RAN domain, EE has been
emphasized as one of the key requirements for
moving forward. It has to take the end-to-end
approach involving all domains within the system.
The overall objective for O-RAN is to gradually
become more energy efficient than the regular
RAN. This should not be at a cost of the key O-RAN
concepts including cloudification and disaggregation
[1]. Moreover, energy-efficient RAN must still offer
high Quality of Service (QoS) to the network users.
Being able to dynamically control the used resources
and hardware by adapting its usage to the actual
traffic demand, would bring significant benefits to
the Mobile Network Operators (MNOs). This chapter
provides an overview of energy efficiency and how it
fits into the O-RAN context.
1.1 Energy Efficiency in Mobile

Networks
The contemporary 5G wireless networks offer high throughputs by in-
creasing the bandwidth, network densification, and utilization of advanced
transmission techniques, e.g., M-MIMO [2]. However, those techniques are
associated either with deploying new Base Stations (BSs) or extending ex-
isting ones with additional hardware. This results in increased power con-
sumption, which is further related to higher MNO costs, and carbon foot-
print [3]. From this perspective, the EE of 5G networks is of high importance
for the telecom industry. Moreover, significant effort has been put in during
the standardization phase of 5G New Radio (NR) by 3GPP to provide unified
mechanisms aimed at providing MNOs with network Energy Savings (ES)
[4]. At the same time, to enable intelligent, case-dependent optimization
of EE, possibly with the use of ML techniques algorithms need access to
network interfaces for monitoring Key Performance Indicators (KPIs) and
control actions.
EE is usually defined as a ratio between the average user throughput and

average power consumption. Monitoring of EE can help MNOs to balance
user throughput-power consumption tradeoff in order to e.g., lower energy
costs, or reduce carbon footprint. However, to enable intelligent, case-de-
pendent EE, optimization algorithms need access to network interfaces,
both for monitoring and control purposes. This can be achieved while fol-
lowing the concept of O-RAN, which provides standardized interfaces to
communicate with RAN components, e.g., gNB [5].
1.2 Achieving High User

Throughput vs Increased
Energy Consumption
As mentioned in the previous section, network densification and M-MIMO
are some of the key enablers for achieving high user throughput in 5G net-
works [2]. Network densification results from the deployment of many BSs
(or gNBs in 5G) of small size that can be installed, e.g., on the street lamps,
to increase network capacity through efficient frequency reuse. M-MIMO
takes the advantage of large antenna arrays (utilizing up to hundreds of
antenna elements) to create narrow spatial beams and simultaneously
serve multiple users with the same time-frequency resources.
The price to be paid for the improved user throughput comes from in-
creased power consumption. First, there are a lot of new BSs/gNBs de-
ployed that contribute to power consumption. Second, those BSs can be
potentially equipped with large antenna arrays, where each antenna ele-
ment requires power-consuming hardware. Thus, network ES is an impor-
tant requirement for current and future mobile networks.
When working with ES aspects, it is important to take into account that ES al-
gorithms cannot work in a vacuum. The decision on switching off the whole
or part of the BS needs to be coordinated with other features to make sure
that the served users do not drop the connection and that the required
performance is maintained. As an example, in [6], the authors analyzed the
interactions between different features within the RAN (e.g. Carrier Aggre-
gation, Dual Connectivity, Interference Management) and proposed how
to align and orchestrate them in the context of the tradeoff between mini-
mization of the used energy and fulfilling the QoS for the users.
1.3 Network Energy Saving

in O-RAN
A group of the tier-1 European MNOs, have formulated an update to the
Technical Priorities for O-RAN, focusing solely on EE as the cross-platform
use case. In this, the EE has been set as a requirement to involve all domains
of RAN, to be addressed within the design of the O-RAN solutions. The aim
is that those networks shall become more energy-efficient over time com-
pared to traditional RAN while keeping cloudification and disaggregation
in place [1].
The key pillars for the EE in the O-RAN context include: power-efficient hard-
ware, reporting of EE KPIs for hardware and software, design of EE-specific
features at a functional level, and automation of EE features through AI and
orchestration [1]. In addition to that, specific EE targets for different network
elements have been specified in the Technical Priority Document for spe-
cific user traffic load or the maximum allowed power consumption for ca-
pacity scenarios.
The defined priorities within [1] include the necessity for monitoring EE-relat-
ed KPIs through the O-RAN interfaces, e.g., power consumed by each cell/
sub-band. Next, O-RAN-compliant hardware is expected to support sleep
mode for either the whole BS or a single element, e.g., the transceiver chain.
The EE hardware is expected to follow the “zero Watt at zero traffic” rule.
Finally, MNOs seek possibilities in EE optimization through the deployment
of advanced ML algorithms within the O-RAN architecture to dynamical-
ly adapt a number of active BSs, and/or antenna elements to the current
network state, e.g., traffic load, and user spatial distribution. For this to work,
there is a need for ML-type of algorithms to learn and follow traffic patterns.
The following sections of this whitepaper focus on EE solutions based on

the ML algorithms and deployed in the O-RAN architecture in the form of
xApps and rApps.
The O -RAN Whitepaper 2023 2.0 O -RAN as an Enabler for Energy Efficiency in 5G Networks
2.0
O-RAN as an Enabler
for Energy Efficiency
in 5G Networks
In this chapter, we provide a general overview
of ways to improve EE in contemporary
mobile networks. This is followed up by some
representative examples of algorithms aiming at
EE optimization at different levels, with a focus on
the utilization of ML techniques. We then put it in
the context of O-RAN where the deployment of EE-
focused aspects and algorithms is discussed.
2.1 Ways to Improve Energy

Efficiency in Mobile Networks
There are several approaches to improving EE in mobile networks, and
they can be worked on at different layers [7]. The highest level is simply by
switching off the whole BS (i.e., putting the BS into sleep mode). Lower level
EE optimization relates to the individual BSs’ components like improvement
of power amplifier efficiency or adjusting the number of active antennas.
Improvement of the EE can also be focused on some protocol-specific
features like dynamic power allocation for the pilot signals, an adaptation
of DRX (discontinuous reception) parameters, or intelligent blanking of re-
source blocks under specific network conditions. Finally, EE gains can be ob-
tained by switching off individual component carriers.
The above-mentioned techniques can be classified as direct EE improve-

ment. Summing this up, when we look from the highest to the lowest level,
the following elements can be switched off (or put into sleep mode): whole
BS, one of the sectors/cells, transceiver chain related to a single antenna,
individual component-carrier, specific resource block, or even resource el-
ement (e.g. pilots signals not used when there is no traffic).
Besides the direct EE improvement, the EE of the wireless network can be

also increased indirectly. Such indirect EE optimization can be based
on e.g., load balancing between neighboring cells, or between frequency
bands within the same cell. The energy-efficient load balancing is achieved,
through intelligent traffic steering, e.g., by a combination of dynamic switch-
ing users between cells and frequency bands not utilizing some of them,
and adaptive power allocation. Moreover, to effectively process a large
amount of data captured from the network, and to make proper decisions
(e.g., switching off power-consuming hardware), ML algorithms are of high
importance. In the next sections, some representative examples of algo-
rithms aiming at improvements of EE are presented.
2.2 Base Station On/Off

Switching
One example algorithm aims at switching on/off Pico-BSs in a heteroge-
neous network (see Figure 2.2-1) [8]. There is one Macro-BS that is always
active and provides coverage in the area, while several Pico-BSs can be in-
telligently activated/deactivated depending on the current traffic volume.
For example, one could imagine that Pico-BSs are needed during the day-
time in the city center where a lot of people work, but during the nighttime,
some of them can be put into sleep mode as most people came back to
their homes in the suburbs. The process is driven by the so-called Rein-
forcement Learning (RL), i.e., learning through interaction. The agent aims to
learn which set of active Pico-BSs would provide the best EE under a given
spatial distribution of users.
This also can be referred to as Energy Saving Management (ESM), i.e. one
of the Self-Organizing Networks (SON) features as per 3GPP specifications,
where depending on the actual traffic volume in a particular area at a par-
ticular time the small cells can be switched on or off, while the macro-site
is responsible for providing coverage at any point in time [9].
Fig. 2.2-1. EE Improvement Through BSs On/Off Switching [8]
2.3 Antenna Selection

for M-MIMO
Another example approach corresponds to a different level in the EE opti-
mization scope. In this case, optimization of a number of active antennas in
M-MIMO BS is under consideration (see Figure 2.3-1). M-MIMO arrays can have
up to hundreds of antenna elements, and each element contributes both
to the overall power consumption and improvement of the user through-
put. Here, an algorithm is proposed [10] that intelligently adjusts the number
of active antennas to the spatial distribution of users within a cell. Due to
the complexity of the M-MIMO system including non-trivial dependencies
between interference and spatial channel correlations, the decision on the
number of active antennas is produced by the RL agent to maximize EE.
Fig. 2.3-1. EE Improvement Through Intelligent Antenna Switching [7]
www.rimedolabs.com
2.4 User Association and

Power Allocation
A different algorithm focuses on indirect EE optimization. As the amount of
power being consumed by the BSs depends on the traffic load, it can be
efficiently reduced by the proper association of users to BSs. Even higher EE
improvement can be achieved when power allocation per user is also taken
into the account. EE power management is of high importance, especially
while considering mmWaves that are characterized by very high path loss
and are sensitive to obstacles. The authors of [11] investigated those prob-
lems in the context of dense networks operating in mmWave and proposed
an iterative algorithm that improves EE through user association and power
allocation.
Fig. 2.4-1 EE Improvement Through Intelligent User Association
12
2.5 O-RAN as an Enabler for

Energy-Efficient 5G Networks
The implementation of the above-mentioned representative algorithms
aiming at EE improvements in 5G networks requires access to some spe-
cific data, e.g., power consumption, traffic volume, user throughputs, and
control actions, e.g., putting particular BSs into the sleep mode. The ele-
ments, which could be subject to switching on/off depending on the need,
can be at different „aggregation levels”, e.g.:
» Whole BS/gNB;
» Part of the gNB (e.g. CU/DU/RU);
» One of the sectors;
» Transceiver chain related to a single antenna/individual power amplifier;
» Component carrier;
» Resource Block;
» Resource element/pilot signal.
The current trend of opening the RAN, through O-RAN ALLIANCE specifica-
tions, where interfaces are open, unified, and vendor-independent to en-
able network optimization by third-party software [5], brings significant
value to the topic of EE.
To achieve the required EE targets using dedicated ML-based algorithms,

O-RAN provides specific tools like xApps and rApps along with open inter-
faces enabling intelligent energy management.
In addition to that, O-RAN work group 1 defines specific energy-saving-re-

lated use cases that guide how to approach the usage of such algorithms.
Those can be found in [12] and [13]. To date, the energy-saving use cases
include Carrier and Cell Switch Off/On; RF Channel Switch Off/On; and Ad-
vanced Sleep Modes.
In the next two chapters, we provide example solutions targeting the EE-re-
lated use cases touched upon in O-RAN ALLIANCE in [12, 13].
The O -RAN Whitepaper 2023 3.0 O -RAN Network Energy Efficiency: Cell Switch On/Off
3.0
O-RAN Network
Energy Efficiency:
Cell Switch On/Off
In this chapter, we describe an algorithm that can
be realized as an rApp in the O-RAN architecture
aimed at improving the network’s EE through
intelligent cell switch-off/on. First, we provide
a general framework for long-timescale EE
optimization which can be done in Non-Real-Time
RAN Intelligent Controller (Non-RT RIC) as defined
in [12]. We then propose the implementation of
the Cell Switch Off/On algorithm within the O-RAN
framework.
3.1 Cell Switch Off/On

Concept
In the previous chapter, we provided a general overview of EE in the context
of the 5G networks and discussed some possible techniques aiming at the
improvement of EE, e.g., BS (cell) on/off switching, or antenna selection for
M-MIMO. We also highlighted the significance of O-RAN architecture in the
possible implementation of those algorithms. Recently, O-RAN ALLIANCE has
released the updated version of its Use Case Detailed Specification [12]. One
of the newly defined use cases is related to O-RAN network ES, and the im-
provement of EE through the load-dependent cell switch off/on.
The traffic load varies in the mobile network over the daytime, or even on
different days of the week. It is related to the users’ mobility and behavior,
e.g., from Monday to Friday people work in the city, thus usually BSs there-
in are highly loaded starting from the morning until the afternoon. On the
other hand, a lot of people have their homes in the suburbs, thus BSs de-
ployed therein remain almost empty until the late afternoon. This creates
an opportunity to switch off either entire cells or some of the operating
frequencies (i.e., carriers) to provide ES. Figure 3.1-1 depicts such an example
using switching on/off cells in different parts of the considered area taking
into account the daily traffic pattern.
Fig. 3.1-1. Energy Savings Obtained Through Traffic Load-Sensitive Switching On/Off Small Cells
3.2 Cell Switch Off/On in the

O-RAN Architecture
O-RAN ALLIANCE proposed a high-level framework for the long-scale (of
minutes, hours, and above) EE optimization in Non-RT RIC through intelligent
on/off switching of cells. The information flow between O-RAN entities for
on/off switching of cells based on ML inference in Non-RT RIC is depicted
in Figure 3.2-1. This figure is a simplified version of the corresponding figure
from the O-RAN specification [12].
Fig. 3.2-1. Cell Switch Off/On the Diagram: AI/ML Inference via Non-RT RIC (Fig. based on [12])
According to Figure 3.2-1, the process of the cell switch off/on in O-RAN ar-
chitecture can be summarized as follows:
» First, the E2 Nodes are being configured by the Service Management

and Orchestration (SMO) to report the data necessary for energy-saving
algorithms via the O1 Interface to the Collection and Control unit. The
Non-RT RIC further retrieves the collected data through internal SMO
communication.
» To collect the data, E2 Nodes must adequately configure the Open RAN
Radio Units (O-RU) to report the data through the Open RAN Fronthaul
Management Plane (O-FH M-Plane) interface. As EE takes into account
both, users’ QoS, and energy consumption, the following data are of high
importance: per cell (per carrier) power consumption, user throughputs,
users traffic type (e.g., indicated by the 5QI), per slice radio resource
utilization, etc.
» In Non-RT RIC, the collected data is subject to monitoring, i.e., some long-
timescale statistics of cell load, user throughputs, and related power
consumption are obtained.
» The obtained long-term statistics are used to train the ML model, and
after the training, phases are utilized for the ML inference.
» When ML inference results in a decision of switching specific cells on/off,

the Non-RT RIC sends a proper execution request to the Collection and
Control Unit via internal SMO communication.
» After receiving the request to switch a particular cell on/off from

Non-RT RIC the Collection and Control unit proceeds with the proper
configuration of the E2 Nodes through the O1 interface.
» Based on the configuration from the Collection and Control unit the E2
Node updates the configuration of the related O-RU.
» Finally, the utilized ML model is subject to monitoring in the Non-RT RIC,

e.g., to detect degradation of its performance, and to start a re-training
procedure.
3.3 Cell On/Off Switching rApp

In the literature, there is a variety of works that consider cell on/off switch-
ing to improve the EE of wireless networks, e.g., [8][14]. Some of them can be
implemented in the O-RAN architecture in the form of rApps. In [8], we have
proposed to extend the network architecture with an intelligent database
named Radio Environment Map (REM), which is used to map user locations
on the EE related to a particular set of active cells. In our work, the REM plays
the role of data storage for the RL agent, which interacts with the environ-
ment to learn which set of active cells will maximize EE for a given group of
network users. REM can have a form of a lookup table when Q-Learning is
employed [8], or it can contain a so-called replay buffer and neural net-
work to be used with a Deep Q-Learning [15].
Let’s now propose how to map our algorithm from [8] into the O-RAN archi-
tecture taking into account a use case defined in [12]. Our solution described
in [8] can be hosted within the Non-RT RIC as an rApp, to interact with the
other O-RAN entities in an RL cycle as depicted in Figure 3.3-1. The RL cycle
starts with the recognition of the state by the agent. In the case of Cell On/
Off Switching rApp, the state is represented by the set of users’ positions re-
ported to the Non-RT RIC from the External Application Server as Enrichment
Information (EI). Later on, based on the current knowledge and algorithm
parameters, the agent takes the action, i.e., sends switch on/off decisions
to the given E2 Nodes over the O1 interface managed by the Collection and
Control entity of SMO. Finally, the current configuration of active cells is eval-
uated in terms of EE, by collecting information about user throughputs and
power consumption. This data constitutes a so-called reward. After the re-
ward is obtained the RL cycle repeats from obtaining new users’ positions.
Fig. 3.3-1. Implementation of Cell On/Off Switching

rApp Based on REM and RL in the O-RAN Architecture
3.4 rApp Implementation

Challenges
The implementation of such rApp rises some challenges, though. First, to
perform cell switch off/on, the rApp must strictly cooperate with other
xApps, e.g., Traffic Steering. Before the cell is switched off, users must be
re-assigned to neighboring cells to avoid deterioration of the QoS. The cell
switching can be centralized, which potentially gives the highest EE gains,
yet increases the computational complexity of the rApp, i.e., the number of
possible active cell combinations grows exponentially. On the other hand,
a decentralized solution requires less extensive computations. However, it
requires somehow clustering cells into the areas being controlled by a sin-
gle Cell On/Off Switching rApp.
The O -RAN Whitepaper 2023 4 .0 O -RAN Network Energy Efficiency: RF Channel Switching
4.0
O-RAN Network Energy
Efficiency: RF Channel
Switching
In this chapter, we provide the description of
an algorithm that can be realized as an rApp in
the O-RAN architecture aimed at improving the
network’s EE through a second use case addressed
by O-RAN ALLIANCE, i.e., intelligent RF channel
switching off/on. First, we provide a general
framework for long-timescale EE optimization
which can be done in Non-RT RIC. We then propose
the implementation of the RF Channel Switching
algorithm within the O-RAN framework.
4.1 Massive MIMO and Energy

Consumption
As already mentioned in chapter 1.2, M-MIMO is one of the technologies that
enable achieving high user throughput in the 5G and beyond networks.
Due to equipping BSs with antenna arrays having up to hundreds of anten-
na elements, it is possible to formulate narrow spatial beams. Those beams
can be used to simultaneously serve users using the same time-frequency
resources, following the concept of spatial domain multiple access (SDMA).
However, the cost to be paid is related to the hardware requirements, i.e.,

M-MIMO requires multiple RF channels associated with each antenna. Every
RF channel contributes to the total power consumption of the M-MIMO BS.
While having tens or even hundreds of such RF channels, the amount of
required energy becomes enormous. Moreover, the increased power con-
sumption is not always compensated by the improved network through-
put, i.e., under low traffic load scenarios the full advantage of M-MIMO can
be achieved with e.g., only 20% of active RF channels.
From this perspective, an intelligent adjustment of the active RF channels

is crucial for energy-efficient 5G networks [16]. However, it is hard to be
achieved in state-of-the-art mobile networks, where the interaction of the
third-party software with the RAN components is limited to the monitoring
of some crucial KPIs, and setting the basic threshold for a few control ac-
tions, e.g., handovers. The deployment of the intelligent ES algorithms re-
quires much more interaction with RAN, including unified mechanisms of
reconfiguration of Radio Unit’s (RU’s) hardware, i.e., RF channels. This can be
successfully achieved within the O-RAN architecture [5].
In the previous chapter, we provided an overview of the EE in the context of

the 5G networks with some possible solutions (Chapter 2.0) and discussed
in more detail the concept of the cell switch off/on (Chapter 3.0). In the re-
cent O-RAN ALLIANCE Use Cases Analysis Report [13], a new use case is indi-
cated, aimed at providing ES within a single BS through load-dependent RF
channel switching. The following sections are focusing on this one.
4.2 RF Channel Switching

The traffic load in mobile networks varies over the daytime. Usually, it is low-
er during the night when most people sleep and rises during the day when
people utilize data transfer for work purposes. Taking into account the
variations in the traffic load, the number of RF channels can be adjusted
to provide ES gains while satisfying users’ QoS demands. This procedure is
depicted in Figure 4.2-1. As can be seen, during busy hours, all RF channels
remain active. When the traffic load lowers to the level of a medium load,
40% of RF channels can be switched off to save energy. Finally, under low
traffic, only 20% of active RF channels can provide users with satisfying QoS,
the rest is switched off to increase ES even more. One should note that
switching off RF channels requires reconfiguration of the BS, e.g., while
having fewer antennas, the number of the supported beams is reduced
and the beams themselves change their shape.
Fig. 4.2-1. Energy Savings Obtained Through Traffic Load-Sensitive Switching On/Off RF Channels
4.3 RF Channel Switching in

the O-RAN Architecture
The high-level idea of switching off/on RF channels is relatively close to the
concept of switching off/on the entire cells. Although the O-RAN ALLIANCE
specifications [12] do not define yet the framework for switching off/on RF
channels, we expect it to be very similar to the one proposed in the figure
therein. Because switching off/on RF channels requires reconfiguration of
BS (e.g., changing the number of supported beams), we expect it to oper-
ate in a slow control loop in the Non-RT RIC. The information flow between
O-RAN entities for switching off/on RF channels based on ML inference in
Non-RT RIC is depicted in Figure 4.3-1.
Fig. 4.3-1. Switching On/Off RF Channels Diagram: AI/ML Inference via Non-RT RIC
According to Figure 4.3-1, the process of RF channel switching in O-RAN ar-

chitecture can be summarized as follows:
» As the antenna array is expected to have several possible

configurations, i.e., sub-arrays that can be activated or deactivated, at
the first stage those configurations must be obtained from O-RU. They
are reported to the Non-RT RIC by the Collection and Control Unit of the
SMO. The array configurations are reported to the E2 Node (i.e. an O-DU
in this case) via the O-FH M-Plane and passed to the SMO via the O1
interface.
» Next, the E2 Node is configured by the SMO to report the data necessary
for energy-saving algorithms via the O1 Interface. Based on this
configuration, E2 Node utilizes O-FH M-Plane to retrieve the requested
data. The data collected in the SMO are further retrieved by the Non-RT
RIC through internal SMO communication.
» In Non-RT RIC, the collected data is subject to monitoring, i.e., some long-
timescale statistics e.g., number of users per beam, user throughput;
and related power consumption are obtained.
» The obtained long-term statistics are used to train the ML model, and
after the training, phases are utilized for the ML inference.
» When ML inference results in a decision of switching certain RF channels

on/off, the Non-RT RIC sends a proper execution request to the Collection
and Control Unit via internal SMO communication.
» After that, the Collection and Control unit proceeds with the proper
configuration of the E2 Node through the O1 interface.
» Based on the configuration from the Collection and Control Unit, the E2
Node updates the configuration of the related O-RU. One should note
that changing the configuration of the O-RU’s antenna array requires
reconfiguration of e.g., Synchronization Signal Blocks (SSBs), or several
M-MIMO layers.
» Finally, the utilized ML model is subject to monitoring in the Non-RT RIC,

e.g., to detect degradation of its performance, and to start a re-training
procedure.
4.4 RF Channel Switching rApp

In this section, we propose a high-level concept of the RF Channel Switch-
ing rApp based on the general framework described in the previous sec-
tion. The rApp performs ES optimization in the Non-RT control loop, by fol-
lowing the concept of RL as depicted in Figure 4.4-1. First, the agent (rApp)
recognizes the state, based on parameters obtained via the O1 interface
(e.g., number of users per beam, user throughput, power consumption),
and actual array configuration. Based on the state, the agent decides on
the action, i.e., selects one of the possible O-RU’s array configurations using
the O1 interface to O-DU (E2 Node), which is coupled with O-FH M-Plane be-
tween O-DU (E2 Node) and O-RU. After that, the action (selected array con-
figuration) is evaluated, and the resultant KPIs (power consumption, QoS
metrics) constitute a reward used to update the action preferences. Then,
the cycle repeats. It is also possible to utilize some EI, e.g., Location Informa-
tion as in our work [10].
Figure 4.4-1. Implementation of RF Channel Switching rApp Based on RL in the O-RAN Architecture
4.5 rApp Implementation

Challenges
During the design of the RF Channel Switching rApp, some challenges should
be taken into the account. Firstly, switching off some RF channels changes
the properties of the antenna array that enforces whole O-RU reconfigura-
tion, e.g., the number of supported beams, and the number of possible lay-
ers. Secondly, the RL algorithm must be supported with some mechanisms
that prevent testing of obviously wrong solutions, e.g., during the busy hour
the configuration with only 20% of active RF channels should never be test-
ed. Thirdly, the rApp must strictly cooperate with other xApps, e.g., with Traffic
Steering. Changing the O-RU configuration usually requires its restart, and
the users should be temporarily offloaded to the neighboring cells to assure
the service requirements fulfillment.
The O -RAN Whitepaper 2023 Summar y & conclusions
Summary &
Conclusions
Energy efficiency in the RAN is one of the key aspects
within the 5G and beyond mobile networks that is currently
on the agenda of the industry and research activities. It
is usually defined as a ratio between the average user
throughput and average power consumption. As the
industry is targeting higher and higher throughputs within
5G and beyond, more BSs and related hardware must
be deployed. As the installed hardware requires more
and more power, the energy-saving mechanisms are of
crucial role. Therefore, being able to dynamically control
the used resources and hardware by adapting its usage
to the actual traffic demand, would bring significant
benefits to the MNOs. To address this issue, O-RAN
ALLIANCE is currently providing means to support energy-
saving topics as one of the key pillars for moving forward
within the RAN domain. In this whitepaper, the authors
provided an introduction to the topic of energy efficiency
and how O-RAN fits into the picture.
In Chapter 1.0, we provide an overview of energy efficiency and how it fits into
the O-RAN context. Using the O-RAN entities, namely: Non-RT RIC and Near-RT
RIC, the EE of the mobile network can be effectively increased. The dedicated
xApps (deployed at Near-RT RIC) and rApps (deployed at Non-RT RIC) can
monitor network data and control switching off hardware components at
different levels. Moreover, xApps/rApps can optimize EE indirectly through in-
telligent Traffic Steering.
While academia has already put a lot of effort to invent dedicated ML-
based algorithms aiming at EE improvement, it is the O-RAN concept
that can bring those ideas into real implementations in the 5G and be-
yond networks due to the native incorporation of intelligence and open
interfaces.
Chapter 2.0 discusses the means to improve energy efficiency in mobile

networks. Due to the dense network deployment and utilization of M-MIMO
technology in 5G networks, optimization of EE becomes a crucial challenge
for MNOs. This is also true concerning the overall power consumption in the
networks, where, e.g., some of the base stations are underutilized, due to be-
ing deployed for coverage reasons, but where most of the time there is no
or very little traffic.
To provide significant network ES, dedicated case-dependent algorithms

should be deployed. These are expected to utilize ML techniques to, e.g.,
switch on/off cells based on long-term characteristics of traffic load, or user
mobility. To address this, Chapters 3.0 and 4.0 complement the topic with
specific solutions developed as ML-supported rApps for O-RAN use cases.
The implementation of such ML-based algorithms requires both data col-

lection and control actions in RAN. This can be achieved within the O-RAN
architecture that provides unified interfaces and a standardized framework
for improving the EE of mobile networks with the use of ML models hosted in
Non-RT RIC. In Chapter 3.0, we described Cell On/Off Switching rApp adapted
by Rimedo Labs from an academic idea to fit the O-RAN architecture.
The EE improvement is also one of the key aspects when considering the BSs
equipped with large antenna arrays, i.e., M-MIMO BSs. The key challenge is to
balance the energy consumption introduced by the multiple RF channels,
and the throughput gains achieved by the beamforming. The deployment of
such intelligent algorithms is possible only in the O-RAN architecture where
third-party software can interact with RAN components at the level of hard-
ware reconfiguration, i.e., activation/deactivation of some part of the anten-
na array. This is addressed within Chapter 4.0.
The overall conclusions from

this whitepaper are as follows:
» Energy efficiency is one of the key items to be addressed

for the future of wireless networks and it has to be taken
into account in an end-to-end fashion;
» O-RAN provides means to support this by open

interfaces, xApps, rApps, ML framework, and energy-
saving-related use cases;
» The initial considerations within O-RAN treat switching

off/on cells, RF chains, etc.;
» ML-based schemes can significantly improve the

operation of the switching off/on BS elements as they
need to be based on trends and long-term statistics;
» When working with ES aspects it is important to take

into account that the ES algorithms cannot work in a
vacuum. The decision on switching off the whole or part
of the base station needs to be coordinated with other
features;
» When we speak about Open RAN systems and energy-

saving features to be used, e.g. as rApps in Non-RT RIC,
it would be good to provide e.g. Energy-Saving-Aware
Traffic Steering mechanisms, which will do the obvious
and move the users out of the cell before we switch it off.
This requires coordination and a holistic approach to ES
to avoid instability in the network;
» Scaling up/down O-Cloud resources concerning the
traffic demand, vs saved energy is part of the equation.
The O -RAN Whitepaper 2023 References
References
[1] Deutsche Telekom, Orange, Telefonica, TIM, and Vodafone “Open RAN
Technical Priorities, Focus on Energy Efficiency”, June 2021, accessible at
O-RAN Ecosystem Resources — O-RAN ALLIANCE
[2] A. Gupta and R. K. Jha, „A Survey of 5G Network: Architecture and Emerg-

ing Technologies,” in IEEE Access, vol. 3, pp. 1206-1232, 2015, DOI: 10.1109/AC-
CESS.2015.2461602.
[3] M. Masoudi et al., „Green Mobile Networks for 5G and Beyond,” in IEEE Ac-
cess, vol. 7, pp. 107270-107299, 2019, DOI: 10.1109/ACCESS.2019.2932777.
[4] 3GPP TS 28.310 V17.3.0, “3rd Generation Partnership Project, Technical

Specification Group Services and System Aspects, Management and or-
chestration, Energy efficiency of 5G”, Release 17, December 2021
[5] M. Dryjanski, R. Lundberg, “The O-RAN Whitepaper; Overview, Architecture,

and Traffic Steering Use Case”, 2021, https://rimedolabs.com/blog/the-o-
ran-whitepaper/
[6] M. Dryjanski, M. Szydelko, "A unified traffic steering framework for LTE radio
access network coordination," in IEEE Communications Magazine, vol. 54, no.
7, pp. 84-92, July 2016, DOI: 10.1109/MCOM.2016.7509383.
[7] D. Feng, C. Jiang, G. Lim, L. J. Cimini, G. Feng and G. Y. Li, „A survey of en-
ergy-efficient wireless communications,” in IEEE Communications Sur-
veys & Tutorials, vol. 15, no. 1, pp. 167-178, First Quarter 2013, DOI: 10.1109/
SURV.2012.020212.00049.
[8] M. Hoffmann, P. Kryszkiewicz and A. Kliks, „Increasing energy efficiency

of the massive-MIMO network via base stations switching using reinforce-
ment learning and radio environment maps”, Comput. Commun., vol. 169,
pp. 232-242, Mar. 2021, [online] Available: https://www.sciencedirect.com/
science/article/pii/S0140366421000335
[9] 3GPP TS 32.551 V17.0.0, “3rd Generation Partnership Project, Technical

Specification Group Services and System Aspects, Telecommunication
Management, Energy Saving Management (ESM), Concepts and require-
ments” Release 17, April 2022
The O -RAN Whitepaper 2023 References
[10] M. Hoffmann and P. Kryszkiewicz, „Reinforcement Learning for Energy-Ef-

ficient 5G Massive MIMO: Intelligent Antenna Switching,” in IEEE Access, vol. 9,
pp. 130329-130339, 2021, DOI: 10.1109/ACCESS.2021.3113461.
[11] H. Zhang, S. Huang, C. Jiang, K. Long, V. C. M. Leung, and H. V. Poor, „Energy-Ef-

ficient User Association and Power Allocation in Millimeter-Wave-Based Ul-
tra-Dense Networks With Energy Harvesting Base Stations,” in IEEE Journal
on Selected Areas in Communications, vol. 35, no. 9, pp. 1936-1947, Sept. 2017,
DOI: 10.1109/JSAC.2017.2720898.
[12] O-RAN Alliance, “O-RAN Working Group 1 Use Cases Detailed Specifica-
tion” v09.00, October 2022
[13] O-RAN Alliance, “O-RAN Working Group 1 Use Cases Analysis Report”
v09.00, October 2022
[14] M. Hoffmann, A. Kliks, P. Kryszkiewicz, and G. P. Koudouridis, “A Reinforce-

ment Learning Approach for Base Station On/Off Switching in Heterogene-
ous M-MIMO Networks,” 2020 IEEE 21st International Symposium on “A World
of Wireless, Mobile and Multimedia Networks” (WoWMoM), 2020, pp. 170-172,
DOI: 10.1109/WoWMoM49955.2020.00038.
[15] M. Hoffmann and P. Kryszkiewicz, „Radio Environment Map and Deep

Q-Learning for 5G Dynamic Point Blanking,” 2022 International Conference
on Software, Telecommunications and Computer Networks (SoftCOM),
2022, pp. 1-3, DOI: 10.23919/SoftCOM55329.2022.9911517.
[16] J. Hoydis, S. ten Brink and M. Debbah, „Massive MIMO: How many anten-
nas do we need?,” 2011 49th Annual Allerton Conference on Communica-
tion, Control, and Computing (Allerton), 2011, pp. 545-550, DOI: 10.1109/Aller-
ton.2011.6120214.
The O -RAN Whitepaper 2023 Glossar y
Glossary
3GPP Third Generation Partnership Project
5QI 5G QoS Indicator
AI Artificial Intelligence
API Application Programming Interfaces
BS Base Station
CU Central Unit
DRX Discontinuous Reception
DU Distributed Unit
EE Energy Efficiency
EI Enrichment Information
ES Energy Saving
ESM Energy Saving Management
FH Fronthaul
FL Federated Learning
gNB gNodeB
KPI Key Performance Indicator
MAC Medium Access Control
ML Machine Learning
M-MIMO Massive Multiple-Input Multiple-Output
MNO Mobile Network Operator
Near-RT RIC Near-Real-Time RIC
NN Neural Networks
Non-RT RIC Non-Real Time RIC
The O -RAN Whitepaper 2023 Glossar y
NR New Radio
NSA Non-Standalone
OFH M-Plane Open RAN Fronthaul Management Plane
O-RAN Open RAN (by O-RAN ALLIANCE)
O-RU O-RAN Radio Units
QoS Quality of Service
RAN Radio Access Network
rApp Application to run on Non-RT RIC
REM Radio Environment Map
RF Radio Frequency
RIC RAN Intelligent Controller
RL Reinforcement Learning
RU Radio Unit
SA Standalone
SBA Service-Based Architecture
SDMA Spatial Domain Multiple Access
SMO Service Management and Orchestration
SON Self-Organizing Networks
SSB Synchronization Signal Block
UE User Equipment
xApp Application to run on Near-RT RIC
The O -RAN Whitepaper 2023 About the Authors
About the Authors
Marcin Hoffmann
R&D Engineer
Marcin Hoffmann is an R&D engineer at Rimedo Labs working on O-RAN

software development solutions and spectrum-sharing-related projects.
Marcin is a Graduate Student Member, at IEEE and received an M.Sc. de-
gree (Hons.) in electronics and telecommunication from Poznań University
of Technology, in 2019, where he is currently pursuing a Ph.D. degree with
the Institute of Radiocommunications. He is gaining scientific experience
by being involved in both national and international research projects. His
research interests include the utilization of machine learning and loca-
tion-dependent information for network management. In addition to that
Marcin works on massive MIMO and advanced beamforming techniques.
His scientific articles are published in top journals like IEEE Transactions on
Intelligent Transportation Systems or IEEE Access.
You can reach Marcin at: marcin.hoffmann@rimedolabs.com
Marcin Dryjański, Ph.D.

Principal Consultant / CEO
Marcin Dryjanski received his Ph.D. (with distinction) from the Poznan Uni-
versity of Technology in September 2019. Over the past 12 years, Marcin
served as an R&D engineer and consultant, technical trainer, technical
leader, advisor, and board member. Marcin has been involved in 5G design
since 2012 when he was a work-package leader in the FP7 5GNOW project.
Since 2018, he is a Senior IEEE Member. He is a co-author of many articles
on 5G and LTE-Advanced Pro and a co-author of the book „From LTE to
LTE-Advanced Pro and 5G” (M. Rahnema, M. Dryjanski, Artech House 2017).
From October 2014 to October 2017, he was an external advisor at Huawei
Technologies Sweden AB, working on algorithms and architecture of the
RAN network for LTE-Advanced Pro and 5G systems. Marcin is co-founder
of Grandmetric, where he served as a board member and wireless ar-
chitect between 2015 and 2020. Currently, he serves as CEO and principal
consultant at RIMEDO Labs.
You can reach Marcin at marcin.dryjanski@rimedolabs.com
The O -RAN Whitepaper 2023 About Rimedo Labs
About
Rimedo Labs
Rimedo Labs specializes in providing
high-quality consulting, implementation,
and R&D services in the field of modern
wireless systems currently focusing on
Open RAN, 5G, and beyond. Rimedo Labs
is an O-RAN software provider delivering
customized xApps and rApps for RAN
Intelligent Controllers.
Rimedo Labs
To read more about us, see: successfully took
www.rimedolabs.com/about/ part in the O-RAN
Global PlugFest
To learn more about our O-RAN services, see: Fall 2022 in i14y
www.rimedolabs.com/o-ran/ Lab in Berlin.
Rimedo Labs is a proud member of

ONF, O-RAN ALLIANCE, and VMware TAP.
Rimedo Labs xApp/rApp Portfolio
Our Open RAN Services
xApp and rApp development

» We develop xApps and rApps for the Near-RT RIC and Non-RT RIC in the
form of RRM/SON algorithms for 4G and 5G systems.
» We are actively contributing to the SD-RAN project as an ONF member.
Technical courses delivery

» We provide live and pre-recorded O-RAN training courses.
» Those include both, overview courses (e.g. "5G and Open RAN")
and deep technical sessions (e.g. "O-RAN System Training").
Dedicated simulations and

algorithm design
» We design specific RRM algorithms tailored to the customer’s
use case or simulate his design in various scenarios.
Free and paid live webinars delivery

» We provide live technical webinars on specific O-RAN-related topics,
such as O-RAN architecture, RAN Intelligent Controller, Network Slicing
in O-RAN, Private mobile networks, etc.
Technical articles delivery

» On daily basis, we contribute to the telecom society by writing
whitepapers and technical blogs on Open RAN-related topics.
» The content is delivered both as our products or for external entities

where we act as invited authors.
If you are interested in finding more

about our services, reach out at:
info@rimedolabs.com
More Resources
Dedicated O-RAN website

Find out more about our O-RAN services, get latest documents,
videos and O-RAN blogs.
Rimedo Labs Blog RIMEDO Labs

Get the latest updates on all things wireless including 5G, 6G,
Blog
O-RAN, Telecom Trends.
O-RAN, 5G and Beyond Webinars

Watch technical-heavy webinars and sessions on Open RAN
and 5G-related topics on our Youtube channel.
The O-RAN Whitepaper

Check out our previous whitepaper introducing Open RAN
concept and O-RAN architecture in details
Let's keep in touch!
info@rimedolabs.com
+48 (61) 665 38 17
RIMEDO sp. z o.o.

ul. Polanka 3
61-131 Poznan
Poland, EU
www.rimedolabs.com
All information discussed in the document is provided "as is" and Rimedo
Labs makes no warranty that this information is fit for purpose. Users use
this information at their own risk and responsibility.
© 2023 Rimedo sp. z o.o. All rights reserved.

RimedoTheORANWhitepaper EnergyEfficiency

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

RimedoTheORANWhitepaper EnergyEfficiency

Uploaded by

Copyright:

Available Formats

The O-RAN

Recently, within the Radio Access Networks (RAN) domain, energy

1.0 Introduction to Energy Efficiency 04

2.0 O-RAN as an Enabler for Energy Efficiency in 5G Networks 08

3.0 O-RAN Network Energy Efficiency: Cell Switching On/Off 14

4.0 O-RAN Network Energy Efficiency: RF Channel Switching 20

Summary & Conclusions 27

This is followed up by example algorithms utilizing ML schemes and their

1.1 Energy Efficiency in Mobile

EE is usually defined as a ratio between the average user throughput and

1.2 Achieving High User

1.3 Network Energy Saving

The following sections of this whitepaper focus on EE solutions based on

2.1 Ways to Improve Energy

The above-mentioned techniques can be classified as direct EE improve-

Besides the direct EE improvement, the EE of the wireless network can be

2.2 Base Station On/Off

Fig. 2.2-1. EE Improvement Through BSs On/Off Switching [8]

2.3 Antenna Selection

Fig. 2.3-1. EE Improvement Through Intelligent Antenna Switching [7]

2.4 User Association and

Fig. 2.4-1 EE Improvement Through Intelligent User Association

2.5 O-RAN as an Enabler for

» Part of the gNB (e.g. CU/DU/RU);

» One of the sectors;

» Transceiver chain related to a single antenna/individual power amplifier;

» Resource element/pilot signal.

To achieve the required EE targets using dedicated ML-based algorithms,

In addition to that, O-RAN work group 1 defines specific energy-saving-re-

3.1 Cell Switch Off/On

3.2 Cell Switch Off/On in the

» First, the E2 Nodes are being configured by the Service Management

» When ML inference results in a decision of switching specific cells on/off,

» After receiving the request to switch a particular cell on/off from

» Finally, the utilized ML model is subject to monitoring in the Non-RT RIC,

3.3 Cell On/Off Switching rApp

Fig. 3.3-1. Implementation of Cell On/Off Switching

3.4 rApp Implementation

4.1 Massive MIMO and Energy

However, the cost to be paid is related to the hardware requirements, i.e.,

From this perspective, an intelligent adjustment of the active RF channels

In the previous chapter, we provided an overview of the EE in the context of

4.2 RF Channel Switching

4.3 RF Channel Switching in

According to Figure 4.3-1, the process of RF channel switching in O-RAN ar-

» As the antenna array is expected to have several possible

» When ML inference results in a decision of switching certain RF channels

» Finally, the utilized ML model is subject to monitoring in the Non-RT RIC,

4.4 RF Channel Switching rApp

4.5 rApp Implementation

Chapter 2.0 discusses the means to improve energy efficiency in mobile

To provide significant network ES, dedicated case-dependent algorithms

The implementation of such ML-based algorithms requires both data col-

The overall conclusions from

» Energy efficiency is one of the key items to be addressed

» O-RAN provides means to support this by open

» The initial considerations within O-RAN treat switching

» ML-based schemes can significantly improve the

» When working with ES aspects it is important to take

» When we speak about Open RAN systems and energy-

[2] A. Gupta and R. K. Jha, „A Survey of 5G Network: Architecture and Emerg-

[4] 3GPP TS 28.310 V17.3.0, “3rd Generation Partnership Project, Technical

[5] M. Dryjanski, R. Lundberg, “The O-RAN Whitepaper; Overview, Architecture,

O-RAN, 5G and Beyond Webinars

The O-RAN Whitepaper