Cross Layer Resource Allocation For Multiuser Video

Published Date: 02 Nov 2017

Transmission over Wireless Networks

Abstract — A Multimedia transmission such as video streaming over wireless networks has grown dramatically in recent years. The downlink transmission of multiple video sequences to multiple users over a shared resource-limited wireless channel, however, is a tough task. We face many challenges in this area like time-varying channel conditions, limited available resources, such as bandwidth and power, and the different transmission requirements of different video content. This work takes into account the time-varying nature of the wireless channels, as well as the importance of individual video packets, to develop a cross-layer resource allocation and packet scheduling scheme for multiuser video streaming over wireless networks. Assuming that accurate channel feedback is not available at the scheduler, random channel losses combined with complex error concealment at the receiver make it impossible for the scheduler to determine the actual distortion of the sequence at the receiver. Therefore, the objective of the optimization is to minimize the expected distortion of the received sequence, where the expectation is calculated at the scheduler with respect to the packet loss probability in the channel. The expected distortion is used to order the packets in the transmission queue of each user, and then gradients of the expected distortion are used to efficiently allocate resources across users.

Index Terms— wireless resource allocation, wireless multimedia, multi-carrier networks, utility driven resource management Cross-layer design, error concealment,

I. INTRODUCTION

Now a days we have a drastic growth in demand for multimedia services such as video streaming on mobile terminals. The technology available for high data-rate multimedia services to mobile clients over wireless networks is rapidly improving with the emergence of third generation and newer wireless standards such as HSDPA and IEEE 802.16.In most scenarios, multiple video sequences are transmitted to multiple users simultaneously by sharing a resource-limited wireless network. Transmitting multiple compressed video.

A program over a wireless network in real time is considered a challenging task due to several reasons. First, wireless channels are impaired by deleterious effects such as fading and co-channel interference (CCI). Second, resources such as bandwidth and power are limited in wireless networks and have to be shared among multiple users. Furthermore, enormous fluctuations in rates of compressed video programs due to the differences in video content and intra/inter coding modes can complicate resource allocation. Wireless resource allocation and scheduling approaches can be categorized into two classes: i) time-division multiplexed (TDM) systems, where a single user is transmitted to in each time-slot, as in CDMA 1xEVDO, and ii) systems in which the transmitter can simultaneously transmit to multiple users in each time-slot. Traditionally, cross-layer scheduling and resource allocation methods exploit the time varying nature of the wireless channel to maximize the throughput of the network while maintaining fairness across multiple users. These methods rely on the multiuser diversity gain achieved by selectively allocating a majority of the available resources to users with good channel quality who can support higher data rates. Optimization over the available resources is performed at each time-slot while taking into account the fading state of each user at that time. A queue-length based utility can be employed for video streaming applications where the delay constraints are stringent. Video quality, however, is not simply a function of the data throughput but is also determined by the video content because of inefficiencies in video compression, as well as the potential for spatial and temporal error concealment of lost/missing data Furthermore, an important requirement in video streaming is that the video will be played back in real-time at the decoder, and, therefore, the appropriate video packets need to be available at the decoder in time for playback. Therefore, any packet that remains in the transmission queue after its decoding time has expired will be discarded prior to transmission consequently, in order to efficiently utilize the limited resources of the wireless networks for video delivery; a content aware scheduling technique must be employed. Methods that have been specifically designed for video applications have conventionally focused on satisfying the delay constraint requirements inherent to the system. Received video quality in these approaches is only measured as a function of delay or packet loss rate.

II. SYSTEM OVERVIEW

Fig. 1 depicts a generic framework for multiuser video transmission over wireless networks, which consists of the Video server/encoder, backbone network, scheduler, and the receivers. Captured video sequences are first compressed by the video encoder and recorded in a media server. We assume that each sequence is packetized into multiple data units. Fig 1

Each data unit/packet is independently decodable and represents a slice of the video. Note that, although in terms of decoder operation, each slice is independently decodable, in reality, most frames of a compressed sequence are inter frames, in which MBs may be dependent on macro blocks of previous frames through motion prediction. Once a video stream is requested by a client, the packets are transmitted over a backbone network (assumed lossless) to the scheduler at a base station servicing multiple clients. In addition to Channel State Information (CSI) available through channel feedback, the scheduler uses three features of each packet to allocate resources across users. These features, for each packet of each client , are the utility gained due to the transmission of the packet, the size of the packet in bits, , and the decoding deadline for the packet, , which represents the delay constraint in order to reach the receiver in time for playback. This decoding deadline is determined by the frame rate of the video being streamed. We assume that all the packets in a frame have the same decoding deadline. Any packet left in the transmission queue after its decoding deadline has expired is discarded since it has lost its value to the decoder. In other words, there are a specific number of time slots available for transmission of each frame depending on the streaming frame rate, and after those time slots have elapsed, no further packet from the current frame is transmissible.

III. PACKET ORDERING WITH EXPECTED DISTORTION

Careful packetization of the video data is necessary to ensure the optimal tradeoff between channel utilization and error robustness. In addition, since we require each packet to be decodable by itself, small packet sizes will degrade the source compression efficiency due to limited prediction. On the other hand, large packet sizes result in greater packet loss probability and ineffective concealment in case of a packet loss. Note that error concealment in this work not only helps error hiding at the decoder, but it also plays an important role in packet ordering and resource allocation. Based on the above discussion, it is assumed in this work that a slice consists of a row of macroblocks nd is directly packetized into a transport packet. The video data packets, then, are ordered in the scheduler

buffer such that the most important packets are first served and, therefore, have a greater likelihood of being received at the decoder. Packet prioritization and resource allocation in this work is performed one frame at a time. Nonetheless, this scheme can potentially be improved by optimizing the scheduling and resource allocation over multiple buffered frames. Such a scheme, however, would lead to a considerably higher computational complexity. Using an outage capacity model, the probability of loss of each transmitted packet can be estimated based on the imperfect channel state information available at the scheduler.

A. Error Concealment and the Calculation of Expected

Distortion

Due to channel losses, we use the expected end-to-end distortion to evaluate video quality. Three factors can be identified as affecting the end-to-end distortion: the source behavior (quantization and packetization), the channel characteristics, and the receiver behavior (error concealment). A robust error concealment technique helps avoid significant visible errors in the reconstructed frames at the decoder. Currently, there does not exist a standardized error concealment scheme for wireless communication. This work, however, assumes that the error concealment scheme is known at both the transmitter and the decoder. Given the importance of error concealment in determining the final decoded quality of the transmitted video, a protocol in which the error concealment scheme

is known to both the scheduler and decoder can potentially be highly beneficial in providing significant performance improvements through content-aware packet scheduling schemes.. In this work, we consider a simple but efficient temporal concealment scheme: a lost macroblock (MB) is concealed using the median motion vector candidate of its received neighboring MBs (the top-left, top, and top-right). The candidate motion vector of a MB is defined as the median motion vector of all 4x4 blocks in the MB. If the preceding row of MBs is also lost, then the MB in the same spatial location in the previously reconstructed frame is used to conceal the current loss. Note that this concealment strategy is employed both in the scheduler optimization framework and at the decoder. Given the dependencies introduced by the error concealment scheme, and assuming dependent packet cases, the expected distortion of the mth slice E{Dm}, can be calculated at the encoder as

Where €m is the loss probability of the mth packet E{DR,m}is the expected distortion of the mth packet if received, and E{DLR,m}and E{DLL,m } are respectively the expected distortion of the lost mth packet after concealment when packet (m-1) is received or lost. Note that in this equation is always equal to 1.0 since there is no packet before the first packet Assuming an additive distortion measure, the expected distortion of a frame of packets, denoted by , can be written as

This distortion measurement is based on a per pixel recursive algorithm called ROPE.

The accuracy of ROPE in end-to-end distortion estimation is attributed to its ability to calculate the first and second moments of the decoder reconstructed pixels. Sub-pixel prediction employed in H.264/AVC, however, involves interpolation of neighboring pixels, which gives rise to cross-correlation terms in the second moment calculation. To deal with the cross correlation terms in our experiments, the cross correlation approximation

Where dxy is the Euclidean distance between two decoder reconstructed Pixels is X and Y , and α is a constant, whose value is experimentally obtained from training data (typically 0.04 to 0.06). In addition to pixel cross-correlations, an important, often neglected, issue in per pixel distortion estimation, is that of rounding errors.

rounding operation is usually employed whenever a filtering or averaging operation results in a floating-point pixel value. In H.264/AVC, rounding operations are encountered in sub-pixel prediction, weighted prediction, in-loop filtering, etc.

B. Packet Ordering

In this section, we present a rate-constrained scheme to order the packets in the transmission buffer of each user based on the contribution of each packet to the end-to-end expected distortion. There are multiple challenges in ordering packets in a lossy environment in conjunction with a complex error concealment strategy. First, because error concealment (EC) introduces inter packet dependencies, the ordering process cannot be done greedily and, therefore, all possible packet loss combinations have to be taken into account. This is because the selection of the first packet causes the existing symmetry in the packet locations to break since lost packets can be better concealed if they are close to a received packet. Note that the "con capability" of a slice, strongly depends on its motion correlation with the neighboring slice as well as the reliability of the neighboring slice, i.e., the loss probability of the packet in which it belongs. In addition, the expected distortion of a frame, as discussed earlier, depends on the loss probability of the consisting packets,

Given the error concealment technique discussed above which limits the dependencies between packets, the above optimization can be performed efficiently using a dynamic programming (DP) technique.

The DP can be viewed as a shortest path problem in a trellis, where each stage corresponds to the mode (SEND or SKIP) selection for a given packet with the complexity equal to 2x2 M. The frame rate R(µ) is obtained by

Note that the solution in (5) is optimal in the sense that, if a rate Constraint Rc corresponds to then the total expected distortion is minimum for all combinations of transmission options with bit rate less than or equal to Rc.

IV. RESOURCE ALLOCATION

A. Introduction

In the previous section, we have described the proposed scheme for reordering packets within the transmission queue of each user. The current section discusses the resource allocation across users that will determine the transmission rates assigned to each user, and thereby the number of transmissible packets from each user’s transmission queue. Note that ni=0 implies that user is not scheduled for transmission at that time slot (the time-slot index remains the same throughout this section and is omitted for simplicity). The maximum number of spreading codes that can be handled by each user is determined by the user’s mobile device. However, the total number of spreading codes, N , that can be allocated to all users, is limited by the specific standard (HSDPA). The total power, that can be used by the base station is also limited in order to restrict the possibility of interference across neighboring cells. In the case that the exact channel state at each time-slot is known to the scheduler, the achievable error-free transmission rate, , for each user can be precisely calculated given the allocated resources, ni and pi.

In the case, when the exact channel state is not known, however, and only an estimate of the channel state is available,

Fig 2

it is also necessary to consider the probability of loss in the channel due to random channel fading that may occur during the transmission.

B. Outage Probability

Since the concept of outage probability is discussed in detail .I this section will simply summarize its application to the current work. Again, the time index will be omitted during this discussion as the outage probability will be calculated at each transmission time-slot. Also, note that Ei refers to the probability of loss of the transmission to user in the current time-slot. All packets, transmitted to user during the current time-slot will have a packet loss probability, , equal to . Using the

Model derived in, the probability of loss of a transmission to user can be written as

Where B denotes the maximum symbol rate per code, hi Denotes the instantaneous channel fading state (SINR per unit power) at that time-slot, and Fx/ei denotes the cumulative probability density function of the instantaneous channel fading state conditioned on the observed channel estimate, ei It is clear from (4) that the probability of loss,

depends on four factors: the allocated resources (ni ,pi ) the estimated channel SINR(ei) the assigned transmission rate(ri) and the conditional cumulative density function (cdf) given by the wireless channel model (Fx/ei).

C. Wireless Channel Model

This work assumes that only partial (imperfect) channel state information is available at the scheduler/transmitter. Errors in the channel estimate can arise from the delay in the feedback channel combined with Doppler spread and quantization errors. It is possible to empirically determine the conditional cdf of the channel SINR conditioned on the channel estimate and the feedback delay using channel measurements. For the purposes of this work, we employ a Nakagami- m channel model which exhibits similar patterns to HSDPA RF channel traces obtained from Motorola, Inc. In this model, the channel SINR can be modeled as a gamma distribution with mean at the channel estimate, ei, The cumulative probability density function can be written as

where is a shape parameter determined by the order, , of the distribution()denotes the incomplete gamma function, and T(m) denotes the gamma function of order m . Note that for a fixed order, m the variance of the Nakagami- m distribution increases with increasing mean (i.e., channel estimate).

D. Problem Formulation

Given the packet ordering scheme and method for calculating the loss probability described above, the scheduler jointly optimizes the rate assignment, τ=(r1,r2,…..rk) where K where is the number of users, the power assignment, p=(p1,p2,…..pk) and the spreading code assignment n =(n1,n2,….nk) , in order to minimize the total expected distortion in the system at each time slot. For a given rate and packet loss probability, let the expected distortion of the frame currently being transmitted to user given the packet ordering specified in Section III-B be E{Di,(ri,εi)} obtained as in (2). Then, the optimization problem can be written as

is the maximum SINR constraint [6] and all other parameters here are previously defined. In principle, a nonlinear

optimization scheme can be used to find the solution to (5). In practice, however, the solution can be highly complex, as an analytical form for E{Di} , which will satisfy different video content and channel conditions, cannot be easily derived. Therefore, this paper uses a two-step approach to simplify the solution to the problem. Our solution is based on a few observations. One observation is that the packet ordering arrived at by the technique described .

V. CONCLUSION

This work introduces a content-aware multiuser resource allocation and packet scheduling scheme that can be used in wireless networks where only imperfect channel state information is available at the scheduler. The scheme works by jointly optimizing the resource allocation and transmission rate allocation in a content-aware manner while also prioritizing video packets in the transmission queue. The content dependent techniques shown in this paper significantly outperform a conventional content-independent scheduling scheme. While results comparing CBR encoded content are not shown in this paper, with CBR encoded content, potentially greater improvements can be expected with a content dependent scheme because, unless an ideal rate control scheme is used, the bit allocations for the video sequences will be less correlated with the final decoded video quality. A simplified content dependent technique that fixes the probability of loss is also shown. Although the scheme with fixed probability of loss can achieve similar results to one that optimizes the probability of loss, it requires tuning of the probability of loss parameter, whose optimal value cannot be known without knowledge of the video content and channel conditions.