TECHNICAL
I’m not claiming multicast is a silver bullet, nor that newer transports have nothing to offer; those comparisons need proper baselines. Live IP streaming is not only a search for the perfect transport; it is a question of architecture, timing and scale. The foundation is not exotic: multicast forwarding with routed unicast fallback, RTP retransmission and fast channel change are mature mechanisms operators already run, joined by a lightweight recovery policy on commodity hardware. The advance is composition. Proven methods used together so live IP streaming can meet every demand at once: low latency, synchronised viewing, resilience, mass scale, sustainability and cost. Broadcast-scale economics need not be reinvented for IP; the task is to retain the mechanisms that already provide them, not design them out.
None of this means giving up resilience, and it does not take a data centre to deliver it. The system repairs missing RTP packets in real time where they remain within the presentation deadline, on commodity client hardware, with no heavyweight AI model, GPU, NPU or TPU in the loop. It carries MPEG transport- stream (TS) media over RTP and UDP, avoiding TCP-style head-of-line blocking when packets are lost. When a packet goes missing, a lightweight control layer decides how to recover it: when to ask again, how aggressively, over which path, and when to discard anything that cannot meet its on-screen presentation deadline. The intelligence is deliberately modest: RTP Sense, a lightweight edge-triggered state engine, converts RTP telemetry into compact controller state for a DRL agent and deadline-aware adaptive control layer. It turns packet gaps, loss history, RTT, jitter, route health, buffer margin and repair outcomes into real-time state. The setup shown in Figure 1 is a lab demo; the same system is already in commercial service in Europe and Latin America. In evaluated cases it has held stable playback at up to 40% packet loss, at round-trip times from milliseconds to over a second, through handover gaps of up to a second, and across access types including fibre, xDSL, Wi-Fi, mobile, in-flight Wi-Fi, satellite and HAPS-relevant scenarios. It treats loss as loss, not congestion, holding bitrate instead of downshifting when the access network gets noisy. A later field demonstration showed the same principle in a real venue: a local broadcast reference and a service-provider IP feed remained time- synchronised to the same minute and second of the match clock. The IP feed ran over public internet and venue Wi-Fi with controlled 40% random packet-loss impairment applied, yet remained aligned with the live reference. Backward compatibility need not mean disruption. Most HLS, DASH and low-latency players can keep working unchanged. This is where a “CDN in your pocket” comes in: a lightweight OCI CDN agent at the edge bridges the multicast or routed-unicast core to a traditional HAS player. It can run on a home gateway, mobile device, PC, edge server or vMEC node, preserving the ecosystem while improving live delivery.
www.nokia.com
102
SEPTEMBER 2026 Volume 48 No.3
Made with FlippingBook - Online magazine maker