Private:video broadcasting ideas

Research problems related to video streaming and broadcasting in WiMAX networks.

Statistical Multiplexing

Statistical Multiplexing aims to better utilize the link bandwidth by multiplexing the data rate variabilities of multiple streams. When bandwidth is less than the aggregated data rate of the multiplexed streams the individual streams need to be adapted/rate shaped/bit extracted/transcoded. Some of the related challenges are noted below.

The first challenge is to estimate the importance of each stream in the congestion window and allocate a proportionate target bit rate for adaptation. The challenge is harder if the original video sequence is not available as a reference for quality or complexity calculations. In this scenario we have to resort to methods of no-reference quality/complexity estimation.

No-reference quality estimation itself is a challenging task. Two main approaches in practice are (1) Sending a metadata stream out of band which contains the necessary information generated by the encoder and (2) The encoder embeds some supplementary information in the compressed bit stream itself( SEI units in H.264). Both have their pros and cons. For a comparative analysis see [1]. One thing common between both the schemes is that the encoder supplies the hints or metadata. The problem of estimation solely on the basis of the compressed stream (i.e. without using even the SEI information, only using the mandatory parts) is not explored. The generation of metadata is a time consuming operation at the encoder and may not be suitable for live video.

A third challenge is to find a model of estimation window. Since in modern encoding standards the frames or access units have complex relationships with many frames in past and future it is not possible to estimate each frames importance independently. Look-ahead and feed-back are two common approaches used my majority of work so far. Are the two approaches sufficient individually? If not what mix should we use in a hybrid approach.

Finally, for time slot based shared channel networks(e.g. OFDMA) the rate adapted streams cannot be transmitted as is. They need to be converted into bursts in order to be allocated in the slots. What burst creation/allocation policy should we take here. We note that at this point we are not considering energy efficiency of the allocation. We are considering the challenge of packet fragmentation under deadline constraints. Can this consideration be incorporated into the target rate decision process so that after adaptation again data is not lost due to infeasible scheduling instances.

Energy Efficient MBS Allocation

MBS or MCBCS is the WiMAX standard specified mechanism to facilitate multicast/broadcast of data. In this, an area of the TDD downlink frame is reserved as an MBS Zone( multiple areas can reserved for multiple MBS zones). The problem is when multiple VBR streams are transmitted how to allocate them in the MBS zone such that the energy consumption at the receiver is minimized.

The usual approach for solving the same problem in DVB-H networks is time-slicing. But a naive time slicing approach( i.e. consider the MBS areas as a train of time slots and for about the granularity in terms of sub-channels) may lead to sub- optimal performance. So the first challenge is to find a time-slicing implementation for a two dimensional channel such that (1) Bandwidth utilization is maximized and (2)receiver power consumption is minimized. We note that this problem is different that the problem solved by Hsu(MM'09). Here we do not consider switching as an important contributor to power consumption. An alternative version of this problem is to translate the bandwidth saved to increase in video quality.

A second problem is the structure of the MBS itself. How should we design an MBS. A horizontal strip MBS has the advantage that the time-slots can be considered almost continuous, where as vertical strip MBS has high instantaneous data rate. Given a set of streams(VBR or CBR) can we find a design of the MBS that will aid in energy efficient scheduling. Here also we do not stress on the switching factor.

Third problem is the switching factor. Of course it can be tagged to each of the above problems as extension, but it might turn our to be too difficult. When considering the switching problem we should assume that the MBS design problems have already been resolved. We need to find a solution better than Hsu(INFOCOMM'09) because the contribution of switching in energy consumption is much smaller in WiMAX as compared to DVB-H. Also we should find the switch-on duration and energy consumption values from a real hardware.

Other Ideas

[Zap Delay Minimization using multiple MBS Zones]

[SFN Video Performance of WiMAX]Performance Evaluation of Multicast/Broadcast Single Frequency Network Operation for WiMAX. Using resource efficient transport formats and assuming perfect synchronization among the BS can achieve almost twice the spectral efficiency of conventional cellular network. There is a similar evaluation for WCDMA networks but not much work has been done for WiMAX.

[Quality Improvement Based on Popularity of Video] Video in WiMAX can be delivered over unicast, multicast or broadcast channels. While broadcast is more resource efficient, unicast offers more flexibility in terms of user specific content. According to the users video request the BSS can decide to add the user to a unicast, multicast or broadcast group. How does this decision making improve the quality of video transmission? Does this help in increasing the network capacity? Since different mode of transmission require different amount of power, are there any implications in terms of Base Station power efficiency?

 Challenge1: Finding typical video viewing pattern statistics as test data.

[Multimedia Overlay Construction over WiMAX]

Energy and video quality optimization for receivers. Note that not all receivers will be handheld mobile devices. Some WiMAX receivers could be static desktops, which do not have energy problem. Do we have separate streams for different types of receivers? Or should we look for a tradeoff between energy optimization (for mobile devices) and channel switching delay (for all receivers)?

How do we allocate the wireless medium (downlink and uplink)? Keep in mind that other services from WiMAX subscribers can also be running. Can we design an efficient scheduler at the base station to provide QoS for video broadcasting as well as other services?

How we can exploit/optimize WiMAX's mobile handoffs in a mobile environment for video broadcasting and other services.

References

[1] Design options and comparison of in-network H.264/SVC adaptation, Kusching et al, JVisCommImg 2010.