AT&T Labs Research — Leading Invention, Driving Innovation

Given the high volumes of video content and various delivery channels (internet streaming, content on-demand, IP television, etc.) a system is needed to automatically verify that each piece of content is delivered with the highest quality. CONSENT fills this gap by providing a platform for detection and analysis of content for security and quality assurance. CONSENT was created to analyze and detect video quality errors through a number of means.

• Detect video anomalies due to content and network problems
• Network behavior (loss, jitter, join leave rates, etc.)
• Content protocol and standards conformance (syntax, structure, etc.)
• Media-based content verification (EPG/media mismatch, content structure, Ads correctness, metadata, activity on channels, etc.)

An extensive set of control procedures have been created in CONSENT to help correlate errors that a customer reports to video-broadcast events and to pre-emptively identify and diagnose errors that occur within the network, before these errors are passed onto the customer. All content and events are logged at various stages in the CONSENT architecture, so a deep forensic problem analysis is never more than a few clicks away.

CONSENT also pinpoints content problems not only within a provider's network but also in generic Internet video, making it a good solution for anyone trying to provide a consistent, high quality video consumption experience. The illustration below depicts the overall CONSENT system diagram and its many options for acquisition, consumption, and diagnosis.

Even the most heavily tested systems will experience errors if they are only tested with lab conditions. Incorporating experience from in-field technicians, network managers, and engineers that define next-generation standards, a vast array of error simulations were included in the CONSENT architecture. Thus, a content delivery network can be stress-tested for conditions that may not be possible in real-world deployments and simultaneously teach CONSENT how to predict quality errors by using its correlation engine that observes and learns from prior events.