AI for 5 G Billing data acceleration TMFDigital

With the development of 5 G, data volume will inevitably grow e. MBB Enhanced Mobile Broadband Gbps 2 G People-centric applications, high user data rates, mobility guarantees under wide coverage 3 D Video, UHD Screen VR/AR Intelligent City IMT-2020 Business model m. MTC 3 G/4 G • VOICE • SMS • DATA VOLUME Intelligent manufacturing 4. 0 Auto driving u. RLLC Massive Machine Type Communication Ultra reliable, &Low-latency communication Low data rate, delay is not sensitive, terminal low cost, long battery life Strict requirements for delay and reliability, telemedicine, industrial control #TMFDigital 5 G Digital product • VOICE • SMS + Slice service • DATA VOLUME Multipal application scenario Clouded Office Intelligent Home/Building • VOICE • SMS • Chinese operators had announced the launch of 5 G commercialization on October 31. • Large bandwidth, low latency and large connection, those characteristics of 5 G will generate a huge amount of message and data which may give tremendous pressure to support system. © 2019 TM Forum | 2

Huge amounts of CCR messages pose a serious challenge to operator’s support system • Example: CCR messages changes of unlimited bandwidth business of a certain province • September 2018 to July 2019 CCR message increased by 126% ，the storage is increased by 263%, the average daily CCR message of July 2019 is 1. 57 billion, and the daily average data volume is 760 million. • The problems of storage, bandwidth and system are urgent to solve Network element The volume of CCR messages increased by 126% Storage capacity increased by 263% The news has soared, and the network element needs more equipment and bandwidth to cope with the situation. Billing System A large amount of vocabulary, need to add a lot of computing power, service capabilities, storage resources to cope with it 2018. 09 2019. 07 p Message volume Network Pressure #TMFDigital 2018. 09 2019. 07 p Storage capacity Search Pressure Storage Pressure Big data platform Data explosion, need more data processing capabilities, storage needs to cope with it Energy Consumption Computer Room Construction © 2019 TM Forum | 3

Conventional strategy: a fixed quota policy between the billing system and the network element Fixed Quota Solution Multiple Quota Solution Unified allocation of the same authorization quota Adjust the authorization quota rules according different payment schemes Nature Ignore users’differencies Take simple quota allocation Bring problems below Major adjustment of quota : it is difficult to control the arrears #TMFDigital Miner adjustment of quota: the number of bills surges Package schemes quota adjustment : process is tedious © 2019 TM Forum | 4

New ideas: Introducing AI technology to build an intelligent dynamic quota model Package fee Gender Age …… Area Customer Profile Time User Id Business id ……. Algorithm feature Machine learning Model training Deep learning Neural Networks Data volume Model optimization Model AI Customer Behavior Business Model prediction Predict Quota model CCR logs Dynamic quota User Quota Time AI dynamic quota Issued Intelligent dynamic quota： Based on user tags and usage record information, the AI neural network intelligent quota prediction model is constructed and trained to generate predicted data that has been used by users, services, and time. The user dynamically authorizes the user authorization quota by combining the message information that Control rule the user accesses in real time on the Internet. New Idea 2: Using edge calculations to reduce data pressure in centralized systems through edge node data splitting and drainage technology #TMFDigital Control report Resource allocation Edge charging gateway © 2019 TM Forum | 5

Solution Architecture: intelligent quota, real-time dynamic control of CCR message interaction, reducing the amount of CCR messages, CDRs China Unicom CBSS Tian Gong Intelligent prediction CDRs Training framework Message control system CCR/CCA 4 G+5 G NSA Network Element CCR/CCA Message Access Province Company Forecasting framework Intelligent Quota Authorizati on strategy Forecast quota Policy management Log data Intelligent operation Operationa l indicator System monitoring Training data Intelligent analysis Tag data File forecast result CCRs log data Big data platform Cooperative data center CCR/CCA Collection framework Operational data Analysis framework Intelligent control platform #TMFDigital © 2019 TM Forum | 6

Effect and Value • Reduce business volume • Reduce delay • Increase bandwidth utilization Network Platform Intelligent control platform • Save Calculate ability • Improve efficiency • Reduce failure Reduce energy #TMFDigital Billing Platform Big Data Platform Diagram 2 Save storage Saving investment • Improve service capabilities • Reduce response time • Improve carrying capacity Increase value © 2019 TM Forum | 7

POC Outcome: Intelligent control platform implementation effect l Case introduction: After intelligent billing on line in a province, the message volume decreased by 63%, the bill volume decreased by 52%, the total traffic remained stable, and the total cost remained stable. 13. 5 2 Oct 9. 8 10 Oct 9 4. 9 18 Oct 4. 4 26 Oct 2 Oct p Message volume #TMFDigital 6. 8 10 Oct 3. 7 18 Oct 3. 5 26 Oct p Detailed quantity 2233 2279 2255 2302 2. 13 2. 18 2. 12 2. 20 2 Oct 10 Oct 18 Oct 26 Oct p Total flow p Total cost © 2019 TM Forum | 8

Evolution path under 5 G SA network architecture The large amount of traffic brought together by the three major business scenarios of 5 G high-bandwidth, dense connection, and low-latency has increased the potential of 5 G intelligent charging control of the amount of voice, and on the other hand, the application of edge computing Technology, capacity expansion at the edge computing layer 5 G NSA Evolution Goal One: 5 G SA Burdening Evolution Goal two: 5 G SA the Business Support System l l l 5 G intelligent charging: through the application of AI technology, combined with the 5 G business scenario expansion function, rational allocation of control resources Intelligent distribution: intelligent identification of billing scene differentiation processing, for the billing system 5 G SA Edge Computing l 5 G Intelligent billing Edge calculation: Edge calculation : technology is used to implement the 5 G burst process in the edge computing layer to reduce the pressure on the 5 G network. Intelligent monitoring: Based on the user's real-time monitoring of the user network execution strategy, MANO adjusts the NE policy to improve the user experience. #TMFDigital © 2019 TM Forum | 9

MEC and Intelligent Billing Edge-based billing gateway: Resource Slice on User-Level Edge Computing Domain Edge Billing Gateway Message Control adds an edge charging gateway at the edge layer, control billing message reporting and implement edge application control through the edge charging gateway. Standard Charging Message strategy Personalized message Message transformation Charging The reduction of core network load by data separation : edge layer data, which is transmitted only in the edge network, greatly reduces the transmission volume of core network data and the information transmission volume of billing. Charging Control： Application Usage Edge Application Standard message control parameter Standard Message Access Standard Message Asynchronous storage Exterior System Base Station Network Element The edge layer is open at the Internet layer, and the edge billing gateway is installed at the edge layer. How can the billing security be guaranteed? The edge layer is separated from the core network, and the edge network and the core network are coordinated to support each other. What can be done when facing the various uncertainties of the end-to-end ecological coordination? #TMFDigital 1. According to the intelligent control policy, the edge charging gateway reports the edge charging information reduces the volume of charging messages, thereby reducing the detailed order quantity and reducing the detailed order storage. 2. The Intelligent Control of Core Network: according to the traditional intelligent billing control process, the billing message load is reduced. Billing process change: dual channel bill emerges. 1. The core network data transmission of the red line generates the core network charging bill 2. The edge data process of the blue line generates the edge network charging bill © 2019 TM Forum | 10

The Developing Advantage of Edge Computing Operator MEC at Field level MEC on Network Side MEC CPE Family Gateway APP Lightweight virtual layer Data Center APP Lightweight virtual layer Access MEC APP MEP NFV MEP Operator 边缘云 NFV Region District Cloud ² Relying on network advantages ² Multiple accessible technologies ² Focus on network and opening platform capabilities Regions IT Cloud Business es Cloud Business Public cloud Io. T Gateway Industry ² Relying on cloud advantages ² Focus on cloud side collaboration ² Multi-demension pure software solutions Controller Personal #TMFDigital Industry Gateway Professional Vertical Industry Cloud Central Cloud ² Relying on industry expertise ² Time sensitive applications of Vertical Industry © 2019 TM Forum | 11