Live Streaming PK Feature Implementation Guide

This article introduces the implementation of the live streaming PK (head-to-head) feature, where two live streamers can engage in video calls while their audiences see both streamers simultaneously. The author, Chang Qing, is the Director of Tencent Video Cloud Terminal Technology, with extensive experience in client-side development.

The article explains what PK is: two streamers enter a PK state through negotiation or backend matching, enabling video calls between them. Audiences then see both streamers' video feeds side by side, increasing interactivity and engagement.

Before PK: Each streamer broadcasts independently with their own audience seeing only one streamer. After PK: Streamers can video call each other, and audiences see both streamers' video feeds.

The article discusses implementation approaches, starting with the simplest method where each streamer mixes both video feeds locally, but this creates performance issues due to excessive processing on mobile devices.

To solve performance problems, the solution moves video mixing to the cloud server rather than on the streamer's device. However, this approach causes brief buffering during the transition from normal streaming to PK mode.

Tencent Cloud's solution uses two separate audio-video channels: one for standard CDN-based live streaming and another for low-latency real-time video calls. This allows seamless switching between normal streaming and PK mode without buffering, as the PK video feed is simply overlaid on the existing stream.

The technical comparison shows: Live streaming channel has 2-second delay using RTMP/HTTP-FLV with no concurrency limits and low cost; PK channel has <500ms delay using proprietary UDP protocol with up to 10 participants and higher per-stream cost.

For developers, the article provides documentation for quick integration of the PK feature, noting that the solution also supports audience-to-streamer interactions and provides open-source code for customization.