Advancing cellular communication testing through Artificial Intelligence

Abstract

Identifying and resolving faults within complex layers of advanced testing solutions like VIAVI’s Test Mobile TM500 has been a long-standing challenge in the industry for many years. Typically, this task is manual and depends on highly skilled experts with decades of experience. These professionals are tasked with pinpointing the source of problems, whether they are hardware related, or even more intricately, software-related issues (and determining their specific defect layer). This involves discerning the precise defect layer within an immensely intricate software, boasting a codebase exceeding 15 million lines with code coverage exceeding 96%, which spans a wide array of functions that these devices support encompassing the entirety of the protocol stack within a communication system. This process imposes a substantial amount of time and costs to the service providers (SPs) and customers, which are consistently repeated day-in-day-out when defects are observed. In this realm, incorporating artificial intelligence and machine learning (AI/ML) driven log analysis in an end-to-end (E2E) solution presents a myriad of challenges, particularly when dealing with multi-type and highly complex software (SW) logs that are in text-type format. At the core of leveraging AI/ML effectively and enabling intelligent solutions lies the crucial step of transforming text-based SW logs into meaningful numerical representations that encapsulate the syntactical and semantic information embedded within the multi-type software logs. These representations can then be supplied to neural networks for subsequent downstream tasks. The development of AI algorithms capable of converting SW logs into numerical representations, however, is not a trivial task. Custom algorithms must be designed specifically for the telecommunications (Telco) industry, taking into consideration the unique differences between SW logs in this context and typical natural language texts. It is to be noted that, owing to the sensitive, confidential, and often proprietary nature of log files within the telecommunications industry, scholarly research encounters limitations in acquiring access to extensive, industry-grade log files, which are essential for training and evaluating AI/ML algorithms effectively. That aside, conventional natural language processing (NLP) algorithms may prove inadequate in this situation due to challenges such as domain-specific terminology, complex log structure, noisy and incomplete data, dynamic and evolving nature, high-dimensional and sparse data, and the need for integration with domain knowledge. Overcoming these challenges and designing a comprehensive AI product with an E2E pipeline constitutes the primary objective of this research. By creating bespoke AI/ML algorithms that can effectively transform multi-type, highly complex software logs into meaningful numerical representations, we strive to unlock the potential of AI/ML to enhance the log analysis process within the telco industry. This advancement will ultimately result in more precise, efficient, and consistent troubleshooting efforts, mitigating the substantial negative impacts on both customers and service providers while alleviating time and resource constraints. Moreover, it will facilitate the preservation and sharing of expert knowledge, foster collaboration among engineers, and enable organizations to scale their troubleshooting endeavors in response to increasing log data and system complexities. Our research addresses complex, real-world industrial challenges, such as defect identification and diagnostics within the telco industry. We implement innovative algorithms to pinpoint the location of problems or defects within the protocol stack, reducing the time expended by triage teams and significantly decreasing the turnaround time (TAT). We also propose the adoption of artificial intelligence and machine learning techniques for predicting log masks using only default SW logs generated at the source, dramatically curtailing the TAT. In summary, this research aims to develop bespoke a full E2E AI/ML solution product tailored for the telecommunications industry to address the challenges in log analysis, defect triaging, and diagnostics. By overcoming these challenges, we seek to enhance the troubleshooting process, minimize adverse effects on customers and service providers, and optimize the use of time and resources. The findings of this research have the potential to drive significant efficiency improvements in the telecommunications sector, promoting collaboration, knowledge sharing, and scalable troubleshooting efforts in response to evolving system complexities.