Playback experience driven cross layer optimisation of APP, transport and MAC layer for video clients over long-term evolution system

Xinyu Huang, Lijun He

This paper, published in IET Research Journals, tackles the perennial frustration of mobile video streaming: the "spinning wheel of death." Researchers Xinyu Huang and Lijun He propose a cross-layer optimization framework that breaks down the traditional silos between the Application (APP), Transport (TCP), and Media Access Control (MAC) layers to ensure smoother playback over LTE networks.

TL;DR

By allowing the network's MAC layer to "talk" to the video application and the TCP protocol, this framework intelligently drops low-priority video packets and prioritizes critical feedback signals. The result is a 60% reduction in rebuffering time and a significant boost in total system throughput with almost no perceptible loss in video quality.

What Problem Does This Paper Address?

In standard networking, layers are "blind" to one another. The Application layer knows if the user's video buffer is empty, the Transport layer (TCP) manages congestion windows and retransmissions, and the MAC layer decides which mobile device gets the most radio resources.

The problem is that TCP is a "perfectionist"—it tries to deliver every single bit, even if a packet arrives too late to be useful for a video frame. Meanwhile, the MAC layer might allocate massive downlink resources to a user whose uplink is so congested they can't send "ACK" (acknowledgment) packets back, causing the whole stream to stall. This lack of coordination leads to inefficient resource use and frequent rebuffering.

Methods

The authors introduce a joint optimization framework consisting of three main components:

  1. Autonomous Packet Drop (APD) Strategy (APP Layer): Before video packets even enter the TCP "pipe," the system analyzes their importance based on motion vectors and coding modes. If the network is congested, it preemptively drops the least important packets. This prevents the TCP congestion window from collapsing and keeps the playback buffer from hitting zero.
  2. TCP-State Aware Downlink Allocation (TD Algorithm): The MAC layer calculates the actual transmission capacity needed for each client based on their ACK feedback rate. This prevents the "resource waste" of sending data to a client who can't acknowledge it due to poor uplink conditions.
  3. ACK Urgency-Aware Uplink Allocation (TU Algorithm): Not all uplink packets are equal. This algorithm prioritizes ACK packets that are close to their Retransmission Timeout (RTO). By ensuring ACKs get back to the server quickly, the downlink "pipe" stays open and flowing.

Datasets

The researchers evaluated their framework using the NS3 (Network Simulation 3) simulator, configured to mimic a standard LTE macro-cell environment. They used eight standard YUV video sequences frequently used in video research:

  • flower.cif, coastguard.cif, news.cif, highway.cif, soccer.cif, foreman.cif, crew.cif, and bus.cif. Each sequence was pre-encoded using H.264/AVC into segments of 60 frames at 30 fps.

Key Insights and Ideas

  • Strategic Imperfection: The "aha moment" here is that dropping data can actually improve the user experience. By sacrificing a tiny bit of image data (1.5% PSNR drop), the system avoids the massive delays caused by TCP's retransmission mechanism.
  • The Uplink-Downlink Loop: We often think of streaming as a one-way street (downlink). This paper highlights that the downlink is only as fast as the uplink's ability to return ACKs. Prioritizing these tiny control packets is a high-leverage move for system performance.
  • Cross-Layer Synergy: The MAC layer's scheduling is much more effective when it knows the "urgency" of the application (buffer status) and the "state" of the transport protocol (congestion window).

Key Results

The proposed APD_TU_TD algorithm was compared against standard schedulers like Proportional Fair (PF) and Round-Robin (RR):

  • Throughput: Achieved a 13.9% improvement in system throughput compared to the high-performing MAXCI algorithm and over 50% improvement over standard PF/RR schedulers.
  • Rebuffering: Reduced average rebuffering time by 60.82% compared to the base optimization without the packet-drop strategy.
  • Video Quality: The decline in Peak Signal-to-Noise Ratio (PSNR) was only about 1.5%, which is generally imperceptible to the average viewer.
  • Scalability: The system maintained stable performance even as the number of clients in the cell increased to 20.

Limitations and Future Work

  • Codec Specificity: The current packet importance calculation is tied to H.264 parameters. Future work would need to adapt this for newer codecs like HEVC (H.265) or VVC (H.266).
  • Simulation vs. Reality: While NS3 is a robust simulator, real-world LTE/5G environments have unpredictable interference and mobility patterns that might challenge the "guard time" intervals used in the rate estimation.
  • 5G Integration: The paper focuses on LTE; extending this cross-layer logic to the ultra-low latency and beamforming capabilities of 5G NR (New Radio) is a logical next step.

Summary updated

Abstract

In the traditional communication system, information of application (APP) layer, transport layer, and media access control (MAC) layer has not been fully interacted. To solve the problem, the authors...
BibTeX
@inproceedings{huang2020playback,
  author = {Xinyu Huang and Lijun He},
  title = {{Playback experience driven cross layer optimisation of APP, transport and MAC layer for video clients over long-term evolution system}},
  booktitle = {IET Commun},
  year = {2020},
  url = {https://openreview.net/forum?id=OBIeXj6p37},
}

Loading related content...

Figure