Optimizing Video Playback Experience: Adaptive Bitrate, AI‑Driven ABS, and CDN Strategies from iQIYI

China's video consumption has moved from expanding user scale to improving service capabilities, focusing on service quality, precise content delivery, terminal integration, differentiated video services, and effective profitability.

On June 26, the "China Video Consumption User Experience Summit and White Paper Release" was held in Beijing, organized by the China Information Consumption Promotion Alliance, Zhongguancun Modern Information Consumption Application Industry Technology Alliance, and the Video Experience Committee. The Video Experience Alliance aims to develop user‑experience‑based standards to improve video service quality across the industry.

At the forum titled "High‑quality user experience drives the development of the large‑video industry," Deng Zhimin, Technical Director of iQIYI Playback Center, delivered a talk titled "iQIYI Intelligent Playback Experience Optimization".

iQIYI's corporate vision is to become a great entertainment company driven by technological innovation. The technical team pursues "ultimate playback experience" and has optimized many stages such as encoding, algorithms, video distribution, production, and CDN deployment.

Bitrate control: In the same network condition, larger bitrate fluctuations cause more stalling; the encoding algorithm must keep bitrate variation within a bounded range while maintaining efficiency.

Hotspot prediction: CDN edge nodes cannot store all video files; when a requested file is missing, a back‑origin request adds latency. Predicting hotspots reduces back‑origin traffic.

HCDN: iQIYI's proprietary CDN architecture that provides high‑speed, high‑reliability data access while keeping overall transmission cost low.

Adaptive streaming: Balances cost and experience by fully exploiting bandwidth to deliver the best possible playback quality.

Green Mirror: Analyzes massive user behavior to identify hot segments of a film, allowing rapid viewing of the entire program through continuous hot‑segment playback.

Only Watch TA: Enables users to watch only the scenes featuring a specific actor or a set of actors.

Zoom AI: Restores higher‑quality images for legacy low‑resolution content (e.g., VCD, tapes) during playback using AI‑based upscaling.

8K VR: Supports 8K panoramic video playback on ordinary VR devices; high‑end chips (Qualcomm 835/845) lack 8K hardware decoding, so iQIYI uses its own solution for mid‑to‑high‑end devices.

The talk then focused on the "Adaptive Streaming" technology.

Cost and experience are often contradictory: lower cost means less bandwidth and higher stalling; higher bitrate also increases stalling; network volatility on the user side worsens it; and high user concentration aggravates it.

Two sets of data were presented:

Figure 1 shows a 24‑hour CDN traffic distribution with two peaks (12:30‑13:30 small peak, 20:30‑21:30 daily peak).

Figure 2 shows a month‑long CDN bandwidth usage. Since CDN billing usually follows the 95th‑percentile method, only 1.5 hours per day are billed at peak. By leveraging adaptive streaming, bandwidth can be shifted to non‑peak periods, keeping cost unchanged while improving user experience.

Figure 2 illustrates typical user network fluctuation during movie playback. Constant bitrate would cause stalling, while low bitrate reduces quality. Adaptive streaming can raise bitrate while preserving quality under fluctuating network conditions.

iQIYI's adaptive streaming has been deployed in three stages:

1. Seamless switching (May 2013): A self‑developed player enabled smooth transition between different bitrates without audio perception or video stutter.

2. Local ABS algorithm (2015): Three strategies—buffer‑based, bandwidth‑based, and a hybrid of both. iQIYI adopted the hybrid, considering encoding segment characteristics, CDN deployment, and transmission traits. The ABS rollout reduced average stalling by 37% and increased playback time of >720p videos by about 300%.

3. AI‑driven ABS (current): Incorporates intelligent scheduling, carrier conditions, and CDN edge load as inputs to a reinforcement‑learning model that selects optimal policies to maximize a benefit function. The continuously evolving model adapts to complex environments.

In summary, iQIYI has optimized the entire playback chain—including CDN distribution, encoding, and video production—to promote intelligent service capabilities and drive overall video playback experience improvements.