An extract from the Open RAN RIC and Apps report and workshop.
Traditional RAN provides few mechanisms to load balance and force traffic on specific radio paths. Most deployments see overlaps of coverage between different cells in the same spectrum, as well as other spectrum layered in, allowing performance, coverage, density and latency scenarios to coexist. The methods by which a UE is connected to a specific cell and a specific channel are mostly static, based on location of the UE, signal strength, service profile and the parameters to handover a connection from one cell to another, or within the same cell from one bearer to another or from one sector to another. The implementation is vendor specific and statically configured.
Figure 9: Overlapping cells and traffic steering
Non-RT RIC and rApps offer the possibility to change these handover and assignments programmatically and dynamically, taking advantage of policies that can be varied (power optimization, quality optimization, performance or coverage optimization…) and that can change over time. Additionally, the use of AI/ML technology can provide predictive input capability for the selection or creation of policies allowing a preferable outcome.
The traffic steering rApp is a means to design and select traffic profile policies and to dynamically allow the operator to instantiate these policies, either per cell, per sector, per bearer or even per UE or per type of service. The SMO or the Non-RT RIC collect RAN data on traffic, bearer, cell, load, etc. from the E2 nodes and instruct the Near-RT RIC to enforce a set of policies through the established parameters.
This rApp is assuming that specific services such as AR/VR will require different QoE parameters that will need to be adapted in a semi dynamic fashion. It proposes the use of AI/ML for prediction of traffic load and QoE conditions to optimize the traffic profiles.
UE and network performance data transit from the RAN to the SMO layer over the O1 interface, QoE AI/ML models are trained, process the data and infer the state and predict its evolution over time, the rApp transmits QoE policy directives to the Near-RT RIC via the Non-RT RIC.
QoS based resource optimization rApp is an implementation of network slicing optimization for the RAN. Specifically, it enables the Non-RT RIC to guide the Near-RT RIC in the allocation of Physical Resource Blocks to a specific slice or sub slice, should the Slice Level Specification not be satisfied by the static slice provisioning.
Since mobile networks have been designed for mobile but relatively low velocity users, the provision of high speed, reliable mobile service along highways requires specific designs and configurations. As vehicles become increasingly connected to the mobile network and might rely on network infrastructure for a variety of uses, Vehicle to infrastructure (V2X) use cases are starting to appear primarily as research and science projects. In this case, the App is supposed to use AI/ML models to predict whether a UE is part of a V2X category and its trajectory in order to facilitate cell handover along its path.
3GPP has defined the concept of creating a connectivity product with specific attributes (throughput, reliability, latency, energy consumption) applicable to specific devices, geographies, enterprises… as slices. In an O-RAN context, the Non-RT RIC and Near-RT RIC can provide optimization strategies for network slicing. In both cases, the elements can monitor the performance of the slice and perform large or small interval adjustments to stay close the slice’s Service Level Agreement (SLA) targets.
Generally speaking, these apps facilitate the allocation of resource according to slice requirements and their dynamic optimization over time.
The NSSI rApp aims to use AI/ML to model traffic patterns of a cell through historical data analysis. The model is then used to predict network load and conditions for specific slices and to dynamically and proactively adjust resource allocation per slice.
Massive MIMO (mMIMO) is a key technology to increase performance in 5G. It uses complex algorithms to create signal beams which minimize signal interference and provide narrow transmission channels. This technology, called beamforming can be configured to provide variations in the vertical and horizontal axis, azimuth and elevation resulting in beams of different shapes and performance profiles. Beamforming and massive MIMO are a characteristic of the Radio Unit, where the DU provides the necessary data for the configuration and direction of the beams.
In many cases, when separate cells overlap a given geography, for coverage or density with either multiple macro cells or macro and small cells mix, the mMIMO beams are usually configured statically, manually based on the cell situation. As traffic patterns, urban environment and interference / reflection, change, it is not rare that the configured beams lose efficiency over time.
In this instance, the rApp collects statistical and measurement data of the RAN to inform a predictive model of traffic patterns. This model, in turn informs a grid of beams that can be applied to a given situation. This grid of beams is transmitted to the DU through the Near-RT RIC and a corresponding xApp, responsible for assigning the specific PRB and beam parameters to the RU. A variant of this implementation does not require grid of beams or AI/ML, bit a list of statically configured beams that can be selected based on specific threshold or RAN measurements.
Additional apps leveraging unique MIMO features such as downlink transmit power, Multiple User MIMO and Single User MIMO allow, by reading UE performance to adjust the transmit power or the beam parameters to improve the user experience or the overall spectral efficiency.
These apps are a collection of methods to optimize power consumption in the open RAN domain.
Carrier and cell switch off/on rApp:
A simple mechanism to identify within a cell the capacity needed and whether it is possible to reduce the power consumption by switching off frequency layers (carriers) or the entire cell, should sufficient coverage / capacity exist with other adjoining overlapping cells. AI/ML model on the Non- RT RIC might assist in the selection and decision, as well as provide a predictive model. The prediction in this case is key, as one cannot simply switch off a carrier or a cell without gracefully hand over its traffic to an adjoining carrier or cell before to reduce quality of experience negative impact.
RF Channel reconfiguration rApp:
mMIMO is achieved by the combination of radiating elements to form the beams. A mMIMO antenna array 64 64 transceivers and receivers (64T64R) can be configured to reduce its configuration to 32, 16 or 8 T/R for instance, resulting in a linear power reduction. An AI/ML model can be used to determine the optimal antenna configuration based on immediate and predictive traffic patterns.