knowledge centre

Huawei's 'autonomous driving' mobile networks strategy aims to increase automation and reduce costs

Roberto Kompany Senior Analyst, Research

"Mobile operators must introduce automation (including AI) to reduce network opex by 2022, but 56% of mobile operators worldwide have little or no automation in their networks as yet."

Many mobile network operators (MNOs) are apprehensive about 5G because their plans to serve the growing demand for network capacity must balance investment in new technology with investment in 4G networks in an environment where ARPU (average revenue per user) has been flat or declining. To complicate matters further, 5G networks will need a far larger number of network elements, such as cell sites and antennas, each of which will be more configurable and more complex to manage manually than past generations of equipment. This article explains how Huawei, as announced at the recent Global Mobile Broadband Forum (GMBBF) 2018, plans to help MNOs to create 'autonomous driving' mobile networks that can be planned, deployed, provisioned, maintained and optimised more efficiently than today's mobile networks.

The RAN is the biggest contributor to mobile network opex

Mobile networks are distributed by nature. They comprise thousands of sites, which require mostly manual design, installation, maintenance and ongoing optimisation. In the past, a new generation of mobile network technology has delivered improved operational productivity, mainly through spectrum and related efficiencies. For example, the opex per site for 4G RAN is, on average, 34% lower than that for an equivalent 3G network. However, each new generation, has brought diminishing opex returns from the radio upgrade, and these savings have been offset by the need to invest in greater coverage (more sites) and capacity per site, thereby driving up the absolute network opex.

Figure 1 indicates the typical breakdown of mobile network opex spending in 2018, where the radio access network (RAN) lifecycle tasks account for 29%, plus a sizeable piece of the 12% spent on network support and maintenance.

Figure 1:  Typical mobile network opex breakdown1

Automation and AI can be introduced to streamline many of these RAN-related activities, but 56% of MNOs worldwide have little or no automation in their networks. However, more than 60% plan to introduce automation, including AI, into all stages of the network lifecycle by 2022 to reduce opex.3 They will need support from their vendor partners to help meet their network challenges as they prepare for 5G network deployments.

The leading RAN vendors are committed to automation and AI to reduce opex

The leading RAN vendors are working with partners and investing to develop the tools to support MNOs' journeys to 5G.

  • Huawei has deployed its wireless AI technology in over 40 countries to simplify operation and maintenance (O&M), and reports the following examples of savings.
    • Massive MIMO pattern optimisation. Reinforcement learning is used by the system to automatically lock in the best pattern for each cell and the pattern will self-adapt to scenario changes and dynamic traffic. The cell throughput increase is around 20% with high O&M efficiency.
    • Speed optimisation for users with the lowest throughput. Over 100 parameters affect the user throughput and high-level expertise is required for best results. Using reinforcement learning, the system will optimise multiple parameters in parallel. The number of such users is reduced by 15%.
    • Multi-band/multi-RAT power saving. AI technology is used to improve co-coverage. For each cell, the dynamic thresholds must be configured according to traffic changes. Energy savings of 10%–15% are delivered.
  • Ericsson has been working with SoftBank in Japan for several years to deploy machine learning in the RAN space. In August 2018, Ericsson announced that it would increase investment in automation and AI, and will employ an additional 100 specialists by the end of 2018.
  • Nokia announced its end-to-end 5G portfolio during Mobile World Congress (MWC) 2018 with the promise that its ReefShark chipset will reduce baseband power consumption by 64% and reduce opex through automation and AI. In June 2018, Nokia announced a partnership with China Mobile to research AI and machine learning to support 5G use cases.

Other vendors, such as Samsung and ZTE, are also committed to developing AI products to help MNOs reduce opex costs.

Huawei sets out its autonomous driving mobile networks' vision, and calls on the industry to collaborate

During his keynote at GMBBF 2018, Huawei Executive Board Director David Wang introduced the concept of autonomous driving mobile networks as the strategy to address MNO challenges and the need for automation. He also called on the telecoms industry to work together to define the standards to make it a reality.

Huawei adopted the autonomous driving mobile networks vision based on the concept of autonomous driving vehicles used by the car industry. To deliver fully autonomous vehicles, the car industry has developed a five-step process from no automation to fully autonomous. Similarly, Huawei has set out five levels of autonomous driving mobile networks that create a scenario-based evolution path towards full autonomous networks (see Figure 2).

Figure 2: Huawei's five-level autonomous mobile network vision

Source: Huawei

Huawei draws another parallel between autonomous cars and mobile networks – the car's functions are divided into two separate layers: sensors and onboard computing. The car's sensors detect surrounding road conditions, such as a pedestrian crossing, while its onboard computing analyses and responds. A cloud platform acts as a third layer that stores data, maps and AI training information that are tapped by all the cars in the system. Similarly, Huawei envisions that AI will be built into the three-layers of the mobile network architecture.

  • The site layer gathers real-time data from the cell and active devices, which allows the 'site AI' embedded in the base stations to use data-mining algorithms and perform time-sensitive scenario-based modelling, such as parameter optimisation, traffic prediction, scenario clustering and correlation analysis.
  • The network layer 'network AI' further builds on the site analysis and uses network-wide data to perform non-real-time scenario and root-cause analysis. To do this, Huawei believes that it must shift focus from network element-centric to scenario-centric and from network management to the convergence of management and control. At the GMBBF 2018, it announced an overarching solution, the MBB Automation Engine (MAE) as an enabler for this.
  • The cloud layer 'cloud AI' is responsible for cross-domain automation and maintains system-wide (cross-domain) data to be used for AI training.

Huawei's plan will need to be implemented in a multi-vendor, open-API environment

RAN lifecycle operating costs are a concern for most MNOs that are thinking about deploying 5G. Unless opex growth is addressed with suitable tools such as automation and AI, the sheer number of manual tasks will overwhelm MNOs' operations and risk driving opex costs to unsustainable levels.

Huawei's autonomous driving mobile networks is part of a solid strategy to address these concerns if it can work with its partners and competitors to develop automation standards, particularly for open APIs. MNOs will continue to build multi-vendor mobile networks, but they do not want each RAN vendor domain to be separate 'islands'. For MNOs to get the full benefit of Huawei's autonomous mobile driving networks strategy, they must enable open APIs. Through such a mechanism, MNOs will be able to automate network lifecycles across their networks.

1 Analysys Mason’s survey of 55 Tier 1 and Tier 2 MNOs, 2Q 2018.
2 Analysys Mason’s survey of 76 Tier 1 and Tier 2 MNOs worldwide, 3Q 2018.