Dealing with User Heterogeneity in P2P-MPVC: Layered Distribution Versus Partitioned Simulcast

HomePage.Mpvc History

Hide minor edits - Show changes to markup

May 30, 2017, at 06:22 PM EST by 184.152.50.28 -
Added lines 2-3:
May 30, 2017, at 06:13 PM EST by 184.152.50.28 -
Added lines 3-11:

In this project, we study the achievable average video quality in P2P-MPVC through layered coding and partitioned simulcast systems. We consider only the optimal rate allocation problem under node capacity constraints and assume rate-quality relations of the layered and single-layer coders. Design of a real system that must explicitly consider packet loss and delay, is beyond our scope in this particular study. Our main contributions are summarized as the following:

  1. We cast the problem of finding the optimal video flow configuration in P2P-MPVC as a tree packing problem in a complete graph where the only bottlenecks are incoming and outgoing capacity limitations on the nodes. We then show that, in this setting, any multicast distribution tree can be replaced by a collection of 1-hop and 2-hop trees, which substantially reduces the number of trees we have to consider in the tree construction for either layered or partitioned simulcast system. This result also reveals that, even when we use only 1-hop and 2-hop trees to limit the transmission delay, we can achieve the same video rates as when we can use any distribution trees.
  2. For layered coding, we develop a layer assignment heuristic, which determines, for each receiver, the layers received from each source. After describing the trees used to deliver each assigned layer, we solve for the optimal tree rates and consequently layer rates that maximize the average video quality among all users, and show that although the tree rates are not unique, the optimal layer rates are unique. Finally we develop an algorithm to refine the tree rates, to favor 1-hop trees to decrease potential delay and jitter.
  3. We study the optimal receiver partitioning problem, which determines the receiver partitions for all sources, and the rates of the single-layer videos distributed in each receiver group, to maximize the average video quality. Instead of performing an exhaustive search over all partitions, we propose a fast heuristic algorithm that finds the partitions for each source and determines the group rates for each source.
  4. We compare the performances of both systems through numerical simulations. In our simulations, the proposed layered system achieves the best video rates, whereas our partitioned simulcast heuristic can also achieve close-to-optimal video rates when the number of users is small (3 or 4). Finally, due to the bitrate overhead of the SVC encoder, the partitioned simulcast system outperforms the layered system in terms of the achieved video quality in both the 4-user and 6-user cases, even when the layered coding overhead is as low as 10%. Partitioned simulcast system performs similarly as the layered system at 10% rate overhead in the 6-user case. We note that, the proposed layered system formulation requires a scalable coder that can generate a successively refinable bitstream that can be divided into any number of layers at any rates.
May 30, 2017, at 06:11 PM EST by 184.152.50.28 -
Deleted lines 2-17:
February 12, 2016, at 05:58 PM EST by 172.16.11.216 -
Changed line 17 from:
  • E. Kurdoglu, Y. Liu, Y. Wang, "Dealing with User Heterogeneity in P2P Multi-party Video Conferencing: Layered Distribution Versus Partitioned Simulcast", in IEEE Transactions on Multimedia, vol. 18, no. 1, 2016
to:
February 12, 2016, at 05:56 PM EST by 172.16.11.216 -
Changed lines 15-22 from:

Authors

Eymen Kurdoglu, PhD candidate
Yao Wang, Professor
Yong Liu, Professor

Related Publications: For more info, see our paper: http://vision.poly.edu/papers/2014/kurdoglu2014infocom.pdf

to:

Related Publications

  • E. Kurdoglu, Y. Liu, Y. Wang, "Dealing with User Heterogeneity in P2P Multi-party Video Conferencing: Layered Distribution Versus Partitioned Simulcast", in IEEE Transactions on Multimedia, vol. 18, no. 1, 2016
  • E. Kurdoglu, Y. Liu, Y. Wang, "Dealing with User Heterogeneity in P2P Multi-party Video Conferencing: Layered Coding Versus Receiver Partitioning", in Proc. of Communication and Networking Techniques for Contemporary Video Workshop (in conjunction with INFOCOM), 2014 / Toronto, Canada
February 12, 2016, at 05:52 PM EST by 172.16.11.216 -
Added lines 15-21:

Authors

Eymen Kurdoglu, PhD candidate
Yao Wang, Professor
Yong Liu, Professor

Related Publications:

February 12, 2016, at 05:44 PM EST by 172.16.11.216 -
Changed lines 3-4 from:

In this project we consider the high level design procedure of a peer-to-peer multi-party video conferencing (P2P-MPVC) system, where users with different uplink and downlink capacities send their videos directly to each other using multicast trees. One way to deal with the user bandwidth heterogeneity is employing layered video coding at each source that generates multiple layers with different rates, whereas an alternative is partitioning the receivers of each source and disseminating a different video version created by non-layered video coding within each group.

to:

In this project we consider the high level design procedure of a peer-to-peer multi-party video conferencing (P2P-MPVC) system, where users with different uplink-downlink capacities send their videos using multicast trees. One way to deal with user bandwidth heterogeneity is employing layered video coding, generating multiple layers with different rates, whereas an alternative is partitioning the receivers of each source and disseminating a different non-layered video version within each group.

Changed lines 7-11 from:
  1. We showed that any multicast tree in a fully-connected network is equivalent to a collection of 1-hop and 2-hop trees, under user uplink and downlink capacity constraints. This result reveals that the packet hop count in P2P-MPVC can be limited to two without sacrificing the achievable rate performance.
  2. For the layered system, we proposed an algorithm that solves for the number of video layers, as well as their rates and distribution trees.
  3. For the receiver partitioning system, we developed an algorithm to determine the receiver partitions along with the video rate and the distribution trees for each group.
to:
  1. We show that any multicast tree in a fully-connected network is equivalent to a collection of 1-hop and 2-hop trees, under user uplink and downlink capacity constraints. This result reveals that the packet hop count in P2P-MPVC can be limited to two without sacrificing the achievable rate performance.
  2. For the layered system, assuming a fine granularity scalable stream that can be truncated at any rate, we propose an algorithm that solves for the number of video layers, layer rates and distribution trees for the layered system.
  3. For the partitioned simulcast system, we develop an algorithm to determine the receiver partitions along with the video rate and the distribution trees for each group.
February 12, 2016, at 05:41 PM EST by 172.16.11.216 -
Changed lines 1-2 from:

(:title Dealing with User Heterogeneity in P2P-MPVC: Layered Distribution vs. Partitioned Simulcast:)

to:
February 12, 2016, at 05:40 PM EST by 172.16.11.216 -
Changed lines 1-2 from:
to:

(:title Dealing with User Heterogeneity in P2P-MPVC: Layered Distribution vs. Partitioned Simulcast:)

February 12, 2016, at 05:39 PM EST by 172.16.11.216 -
Changed line 1 from:
to:
February 12, 2016, at 05:36 PM EST by 172.16.11.216 -
Added lines 1-2:
Changed lines 5-13 from:

In this study, we investigated how to maximize the received video quality for both systems under uplink and downlink capacity constraints, while constraining the number of hops the packets traverse to two in order to limit the end-to-end delays. Here's what we have done so far:%0a*We showed that any multicast tree in a fully-connected network is equivalent to a collection of 1-hop and 2-hop trees, under user uplink and downlink capacity constraints. This result reveals that the packet hop count in P2P-MPVC can be limited to two without sacrificing the achievable rate performance.

For the layered system, we proposed an algorithm that solves for the number of video layers, as well as their rates and distribution trees.

For the receiver partitioning system, we developed an algorithm to determine the receiver partitions along with the video rate and the distribution trees for each group.

Through numerical comparison study, we showed that the receiver partitioning system can achieve the same average receiving quality as the layered system without any coding overhead for the 4-user systems simulated, and is significantly better than the layered system even when the layered coding overhead is only 10%. The two systems perform similarly for the 6-user case if the layered coding overhead is around 10%, but the receiver partitioning is still significantly better when the layered coding overhead is around 30%.

For more info, see our paper (conference version, longer version under review): http://vision.poly.edu/papers/2014/kurdoglu2014infocom.pdf

to:

In this study, we investigated how to maximize the received video quality for both systems under uplink and downlink capacity constraints, while constraining the number of hops the packets traverse to two in order to limit the end-to-end delays. Here's what we have done so far:

  1. We showed that any multicast tree in a fully-connected network is equivalent to a collection of 1-hop and 2-hop trees, under user uplink and downlink capacity constraints. This result reveals that the packet hop count in P2P-MPVC can be limited to two without sacrificing the achievable rate performance.
  2. For the layered system, we proposed an algorithm that solves for the number of video layers, as well as their rates and distribution trees.
  3. For the receiver partitioning system, we developed an algorithm to determine the receiver partitions along with the video rate and the distribution trees for each group.
  4. Through numerical comparison, we show that the partitioned simulcast system achieves the same average receiving quality as the ideal layered system without any coding overhead for the 4-user systems simulated, and better quality than the layered system when the layered coding overhead is only 20%. The two systems perform similarly for the 6-user case if the layered coding overhead is 10%

For more info, see our paper: http://vision.poly.edu/papers/2014/kurdoglu2014infocom.pdf

February 12, 2016, at 05:33 PM EST by 172.16.11.216 -
Added lines 1-11:

In this project we consider the high level design procedure of a peer-to-peer multi-party video conferencing (P2P-MPVC) system, where users with different uplink and downlink capacities send their videos directly to each other using multicast trees. One way to deal with the user bandwidth heterogeneity is employing layered video coding at each source that generates multiple layers with different rates, whereas an alternative is partitioning the receivers of each source and disseminating a different video version created by non-layered video coding within each group.

In this study, we investigated how to maximize the received video quality for both systems under uplink and downlink capacity constraints, while constraining the number of hops the packets traverse to two in order to limit the end-to-end delays. Here's what we have done so far:%0a*We showed that any multicast tree in a fully-connected network is equivalent to a collection of 1-hop and 2-hop trees, under user uplink and downlink capacity constraints. This result reveals that the packet hop count in P2P-MPVC can be limited to two without sacrificing the achievable rate performance.

For the layered system, we proposed an algorithm that solves for the number of video layers, as well as their rates and distribution trees.

For the receiver partitioning system, we developed an algorithm to determine the receiver partitions along with the video rate and the distribution trees for each group.

Through numerical comparison study, we showed that the receiver partitioning system can achieve the same average receiving quality as the layered system without any coding overhead for the 4-user systems simulated, and is significantly better than the layered system even when the layered coding overhead is only 10%. The two systems perform similarly for the 6-user case if the layered coding overhead is around 10%, but the receiver partitioning is still significantly better when the layered coding overhead is around 30%.

For more info, see our paper (conference version, longer version under review): http://vision.poly.edu/papers/2014/kurdoglu2014infocom.pdf

Page last modified on May 30, 2017, at 06:22 PM EST