Leveraging AI for RAN optimization in G Ericsson

The technical objectives of simplification and performance advantage can be more or less mapped to the business goals of decreasing working and capital bills respectively, which translate into decreased cost per byte for communique service companies and increased QoS for consumers. Embracing AI thoughts for the design of mobile systems has the doable to address many challenges in the context of both simplification and performance benefit , making it possible to achieve new objectives which are beyond the reach of classical optimization and rule based approaches. In terms of simplification, AI has already shown the ability to considerably improve functionalities reminiscent of anomaly detection, predictive maintenance and the discount of site interventions via automatic site inspections with drones. Performance benefit in the RAN is a greater assignment, as it requires the replacement of traditional rule based community functionalities with their AI based opposite numbers.

Additional requirements consist of flexible and programmable data pipelines for data assortment and storage; frameworks for the introduction education, execution inference and updating of the models; the adoption of graphical processing units for education; and the design of new chipsets for inference. Network hyperparameters are optimized to slowly adapt the RAN algorithms to different network scenarios and stipulations and bring the efficiency of a definite area of the network a distinctive cluster of cells, for instance into a gradual state wherein specific key efficiency signs KPIs are stronger. Examples encompass hyperparameters for self making ready networks algorithms and L3 algorithms mobility, load balancing and so forth for coordination algorithms corresponding to coordinated multi point CoMP, multi connectivity, carrier aggregation CA and supplementary uplink, in addition to for L1/L2 algorithms uplink power handle, link variant, scheduling and the like. In both 4G and 5G, our centralized RAN C RAN and E RAN interconnect BB units to allow most excellent coordination around the entire community in a centralized, dispensed or hybrid community architecture. To be sure that C RAN and E RAN efficiency is according to customer expectancies, an intensive community redecorate is needed.

In this regard, AI strategies in line with superior community graph methodologies are applied to have in mind and represent the complicated radio network and its underlying constructions, similar to the family members between cells and BB units. This approach leads to an most efficient design that maximizes client throughput through optimized CoMP and CA recommendations, and the design is also future proof when it comes to capability and expertise expansions. The design can be split into two main steps. Figure 3 shows the efficiency improvement in a 4G network operated by an Asian operator for three KPIs after an automated E RAN redesign. The first bar graph indicates that the connections in CA mode using three component companies CCs higher by 30 %.

See also  La punta del iceberg – Mujeres en lucha

The middle bar graph shows that the knowledge volume carried by any secondary cell higher by 22 %, while the bar graph on the best shows that downlink cell throughput higher by 4. 3 percent. However, the main useful benefit is that the E RAN design is totally automated and conducted in minutes rather than the months of work that could be required by human specialists. State of the art unsupervised and semi supervised studying suggestions combined with expert domain knowledge result in an efficient annotation of standard and abnormal performance styles that can be utilized later for issue identity and type using supervised studying strategies. By integrating community topologies and configurations with tons of of efficiency metrics and their two dimensional correlation in time and space, it is possible to generate a knowledge graph that reveals the actual root causes that result in an identified network issue. Closing the automatic loop, community parameter adjustments are robotically recommended to determine the exact root cause and extra enhance efficiency.

The use of both high frequency bands such as 28GHz and higher millimeter wave bands will proceed to increase in 5G radio networks and in future generations. A larger number of bands gives higher potential but effects in larger measurement overhead. For instance, initial deployments on the 28GHz frequency bands will deliver spotty insurance. For users to be able to employ potentially spotty insurance on higher frequencies, the UEs want to perform inter frequency measurements, which could lead on to high measurement overhead. We have used AI strategies to expect insurance on the 28GHz band in keeping with measurements at the serving carrier for example at 3. 5GHz.

This approach decreased the measurements on a secondary provider, thus cutting back the energy consumption and the delay for activating elements like CA, inter frequency handover and load balancing. AI based antenna tilting merits particular cognizance among community optimization use cases, as it promises to decorate the insurance and potential of mobile networks by adjusting base station antennas’ electrical tilt based on the dynamics of the network atmosphere. Unlike the widely wide-spread antenna tilt method that follows a rule based policy, AI suggestions enable a self evolving policy, studying from feedback via community KPIs. Using reinforcement learning RL, an agent is trained to dynamically control the electrical tilt of multiple base stations jointly if you want to improve the signal satisfactory of a cell and reduce the interference on neighboring cells in reaction to changes in the environment, such as site visitors and mobility styles. This effects in an standard benefit of community performance and QoE for the users while decreasing operational costs.

See also  The Norsk Hydro Lockergoga Ransomware Cyber Attack Swimlane

Addressing the optimization of the RAN algorithms domain is vital to our long run vision of creating an all encompassing single AI based controller that spans the entire hierarchy of manage. The advantage of such a controller may be the inherent capability to optimize varied transmission parameters across layers concurrently. The advent of a controller being able to learn directly via exploration of the state space would remove the obstacles imposed by human designed algorithms, making it possible to identify better mixtures of transmission parameters within a layer and across layers. Moreover, a controller being able to learn from data would inherently be tuned to the atmosphere and be free of community hyperparameters, which would cause simplification of the application stack. Another task is the need to redefine the radio access challenge in a way that enables learning through interplay with the RAN atmosphere.

Today’s divide and overcome method for offering radio access to UEs by breaking down the difficulty into many subproblems of manageable complexity, and designing exact answers for each subproblem, is puzzling to use when using AI based controllers. In other words, staying within the existing fragmented RAN framework with various AI based controllers, each seeking to optimize a RAN feature while learning through interaction with an identical RAN atmosphere, would evade the system from learning and jeopardize system efficiency. One possible approach to address such demanding situations could be to adopt RL as the framework of choice for RAN control. RL has the essential functions to deal with transients, but it continues to be difficult to deploy it in the context of the present fragmented RAN framework. To this end, one strategy could be to redefine the difficulty and devise a solution with a single stage of state estimation and a single stage of downstream end to end handle. This design choice would enable a state estimation as close as feasible to the actual system state and a controller able to joint optimization over several transmission parameters.