Leveraging AI-driven RAN Intelligence to Maximise Network Potential

The RAN is a critical part of a cellular network as it manages a CSP’s most important asset: its spectrum. Yet despite investing billions of dollars in spectrum, CSPs have historically had little control over their spectrum portfolio using existing RAN controller technologies. This is now changing with the introduction of new AI-driven RAN Intelligence solutions.

Network Challenges

With the introduction of 5G, mobile networks are becoming more complex. Although there are efforts underway to sunset 2G and 3G networks, many CSPs will need to operate multiple base station types across 2G, 3G, 4G and 5G for many years. In addition, they will need to accommodate an ever-widening range of spectrum bands – perhaps up to 20 bands – ranging from sub-1GHz low-bands and 3.5GHz mid-bands to very high 26-40GHz millimetre bands, with even higher 66GHz bands and above expected to be used in future.

In future, CSPs will also need to face the challenges of accommodating multiple network architectures, ranging from the traditional distributed architecture to new disaggregated RAN and edge architectures. At the same time, they will be offering an ever-increasing range of services – voice, video, AR/VR, IoT, WBB, LTE-V and FWA – each with its own specific set of technical requirements, while simultaneously dealing with continuously varying RF parameters, different levels of signal attenuation, channel interference, etc.

Traditional O&M

Traditional routine RAN network optimization typically involves the use of in-house experts who handle all network alarms and faults. This is a labour-intensive process as all data gathering and analyses is done manually and requires a large Operations and Management (O&M) team. In addition, most CSPs only have limited resources at their disposal and hence only the “top N” problematic cells in an entire network are usually selected for optimization. As a result, it may take several weeks, or sometimes months, to adjust network parameters to support new applications. Network management this way is an expensive but unavoidable opex cost for CSPs.

Leveraging AI-driven RAN Intelligence

Traditional O&M is no longer an option for CSPs striving to provide integrated control across a portfolio of multi-standard, multi-band 2G to 5G networks. Instead, CSPs must turn to new AI-based RAN intelligence solutions, which in future will play an essential role in helping CSPs to manage complex, integrated networks, thereby increasing network performance.

Improving RAN performance involves leveraging AI to update and optimize the RAN’s control parameters across time, frequency and space domains. This requires a deep understanding of the nature and the role of the different parameter categories affecting network performance, as well as an understanding of the complexity of each individual category and the potential for improvement. Typically, RAN algorithms are adapted to new network scenarios and conditions by optimizing the network hyperparameters.[1] This brings the performance of a particular part of the network – such as a specific group or cluster of cells – into a steady state thus improving specific key performance indicators or KPIs. Examples of RAN algorithms include self-organizing algorithms and L1/L2 and L3 algorithms.

 Huawei’s RAN Intelligence Portfolio

To promote the use of intelligent networks, Huawei has developed a portfolio of RAN Intelligence solutions, which includes the following two solutions:

  • SingleBAND – enables site-level, multi-band convergence through on-site intelligence. This solution enables flexible full-band decoupling, which through FDD enhanced uplink extends TDD band coverage while also improving network capacity through full-band and multi-beam 3D coordination.
  • Capacity Turbo – a network-level solution that takes the multi-band convergence concept beyond a single site. Capacity Turbo improves optimization efficiency and frees experts from repetitive tasks to focus on more advanced tasks. In addition, Capacity Turbo has the ability to accommodate existing expert experiences at a CSP and learning from those experiences in an interactive way. This enables Capacity Turbo to become even smarter.

Huawei Capacity Turbo

Huawei’s Capacity Turbo is a flexible solution that integrates with existing network and site AI algorithms (Exhibit 1). Rather than handling network alarms and faults, Capacity Turbo focuses on optimizing network performance. This involves the simultaneous optimization of multiple parameters, a feat impossible with traditional O&M. The solution operates across all spectrum bands and mobile networks from 2G to 5G.

Capacity Turbo enables automatic, smart network-level optimization using intelligent algorithms which perform coordinated optimization between multiple bands and multiple sites. This improves network performance, enhances user experiences while minimizing routine network optimization costs. By using iterative parameter optimization, combined with expert experiences, Capacity Turbo can enable closed-loop, data-driven on-line network optimization.

Compared to traditional O&M, Capacity Turbo can optimize many more parameters (20+ versus 3-5 in traditional O&M) as well as offer automated scenario matching across the whole network rather than just in a single cluster. In addition, this optimization process can be done in less than two weeks compared to a month or more with traditional O&M.  RAN intelligence solutions thus provide automated solutions to complex network problems that cannot be resolved by on-site personnel – in effect making the impossible possible!

Exhibit 1:  Huawei Capacity Turbo

Commercial Deployments

Capacity Turbo is already in commercial service with many CSPs around the world. Examples include:

  • China – using multi-parameter optimization-based AI, a leading CSP in China achieved an 18% improvement in downlink throughput ratio using Capacity Turbo in a trial in Guangzhou run across a 1755 cell, 580 base station network. Capacity Turbo also enabled the same CSP to achieve an 81% improvement in downlink packet loss and a 22% improvement in uplink packet loss across the same network. In another example, the CSP used Capacity Turbo to optimise VoLTE performance and achieved an 81% reduction in packet loss while reducing uplink packet loss by around 22%.
  • Thailand – the Capacity Turbo solution has been successfully used by a major CSP in Thailand to improve 4G coverage. In regions with weak coverage, high interference and limited capacity, Capacity Turbo was able to improve coverage as well as increase base station throughput by between 13% to 15%
  • Southern Spain – using an AI-enabled personalized parameter policy, a CSP in Spain experienced a 15% improvement in average user downlink throughput across an optimized 766 cell portion of its network while using Turbo Capacity.
  • Brazil – one of Brazil’s leading CSPs increased user experience by 18% across a 70 base station site region using Capacity Turbo.


In future, RAN intelligence solutions such as Capacity Turbo will play a critical role in managing complex multi-band, multi-RAT mobile networks and AI-driven network optimization will enable CSPs to maximise network potential while lowering total cost of operations. In addition, better RAN intelligence will drive the development of new, innovative 5G use cases, thereby providing CSPs with opportunities to differentiate their networks compared to rivals.

[1] hyperparameter – a parameter used in machine learning to control the learning process

Related Posts