Mobile Networks and Applications

, Volume 13, Issue 3, pp 274–284

Power Efficient Video Multipath Transmission over Wireless Multimedia Sensor Networks

Authors

    • Department of Electrical and Computer EngineeringUniversity of Patras
  • Michail Tsagkaropoulos
    • Department of Electrical and Computer EngineeringUniversity of Patras
  • Tasos Dagiuklas
    • Technological Educational Institute of Messolonghi
  • Stavros Kotsopoulos
    • Department of Electrical and Computer EngineeringUniversity of Patras
Article

DOI: 10.1007/s11036-008-0061-5

Cite this article as:
Politis, I., Tsagkaropoulos, M., Dagiuklas, T. et al. Mobile Netw Appl (2008) 13: 274. doi:10.1007/s11036-008-0061-5

Abstract

This paper proposes a power efficient multipath video packet scheduling scheme for minimum video distortion transmission (optimised Video QoS) over wireless multimedia sensor networks. The transmission of video packets over multiple paths in a wireless sensor network improves the aggregate data rate of the network and minimizes the traffic load handled by each node. However, due to the lossy behavior of the wireless channel the aggregate transmission rate cannot always support the requested video source data rate. In such cases a packet scheduling algorithm is applied that can selectively drop combinations of video packets prior to transmission to adapt the source requirements to the channel capacity. The scheduling algorithm selects the less important video packets to drop using a recursive distortion prediction model. This model predicts accurately the resulting video distortion in case of isolated errors, burst of errors and errors separated by a lag. Two scheduling algorithms are proposed in this paper. The Baseline scheme is a simplified scheduler that can only decide upon which packet can be dropped prior to transmission based on the packet’s impact on the video distortion. This algorithm is compared against the Power aware packet scheduling that is an extension of the Baseline capable of estimating the power that will be consumed by each node in every available path depending on its traffic load, during the transmission. The proposed Power aware packet scheduling is able to identify the available paths connecting the video source to the receiver and schedule the packet transmission among the selected paths according to the perceived video QoS (Peak Signal to Noise Ratio—PSNR) and the energy efficiency of the participating wireless video sensor nodes, by dropping packets if necessary based on the distortion prediction model. The simulation results indicate that the proposed Power aware video packet scheduling can achieve energy efficiency in the wireless multimedia sensor network by minimizing the power dissipation across all nodes, while the perceived video quality is kept to very high levels even at extreme network conditions (many sensor nodes dropped due to power consumption and high background noise in the channel).

Keywords

multipath routingpower aware packet schedulingvideo distortion prediction modelWMSN

1 Introduction

The recent advances in developing cost efficient miniature hardware solutions such as video cameras, sensors etc, that are able to ubiquitously capture multimedia content, has led to the development of Wireless Multimedia Sensor Networks (WMSNs) [1]. Similarly to the traditional Wireless Sensor Networks (WSNs), WMSNs consist of sensor devices wirelessly interconnected with each other; however, these devices have the potential to retrieve video and audio streams, still images and scalar sensor data. Evidently, wireless multimedia sensor networks will enhance existing sensor network applications, such as tracking, home automation and environmental monitoring, but they will also enable the development of new technologies.

In particular, sensor devices equipped with miniature battery powered cameras and wireless low-power transceivers capable of transmitting, receiving and processing video streams. These devices can compose wireless video sensor networks that will complement existing surveillance systems. These networks will have to support reliable, bandwidth efficient video transmission with minimum power consumption. Therefore, the encoding technique that will be used should combine high compression efficiency, low complexity and error resiliency. Additionally, other algorithms and protocols such as packet scheduling and rate adaptation algorithms may be implemented to complement the source coding model.

In addition, the high end-to-end bandwidth requirements of video communication usually can not be met by the WMSNs, when the traditional single path routing approach is used, leading to perceived video quality degradation. In order to meet the QoS requirements, a multipath approach can be adopted, where the video source (i.e., the server) delivers the data to its destinations via multiple paths, thereby supporting an aggregated transfer rate higher than what is possible with any one path. Specifically, the encoded video data are segmented and multiplexed in a specific way, based on their distortion importance, over different paths so that the end-host can assemble the video data and decode them with the maximum perceived quality. In order to minimize the power consumption due to video packets reception and transmission in each sensor node in the WMSN, this study considers a slightly modified version of the Low-Energy Adaptive Clustering Hierarchy (LEACH) protocol introduced in [2]. LEACH is a protocol architecture for wireless sensor networks that achieves low energy dissipation and latency without sacrificing application-specific quality. It uses a clustering architecture where each node in the cluster sends its data to a local cluster-head. This mode is responsible to collect data from all sensors in the cluster and multiplexing them and send them to the receiving end. The modified LEACH protocol used during simulations in this study specifies that instead of sending the data directly to the receiving end, each cluster-head establishes paths with the closest cluster heads, thus the data can now reach the receiving end through multiple routes.

This study concentrates in H.264/AVC codec [3] that can achieve higher compression efficiency than any of the previous standards, by using previously encoded frames as reference for the motion-compensated prediction of each inter-macroblock or macroblock partition. However, motion estimation functionalities require higher complexity and powerful processing encoders. Hence, in this study a packet scheduling algorithm is proposed that will compensate for this increase in energy consumption by selectively dropping packets prior to transmission in order to reduce the amount of transmitting data, without increasing significantly the video distortion in the receiving end.

This paper contributes in the optimization of video transmission over wireless sensor networks by proposing two packet scheduling algorithms that improve the perceived video quality and extend the power efficiency of the wireless sensor network. In particularly, a Baseline scheduling algorithm is proposed that firstly, it identifies multiple uncorrelated possible paths from the video sender to the receiver that can on aggregate satisfy the quality of service requirements of the video service. Secondly, in case that the aggregate bandwidth of the multiple paths is limited, the algorithm utilizes a video distortion prediction model to determine the least important packets that could be dropped prior to transmission. On the other hand a second scheme is proposed, named Power aware packet scheduling that in addition to the Baseline algorithm it also has the ability to estimate the power consumption of each node in the network. Therefore, this second algorithm can decide upon which packets to drop in order to adapt the transmission rate of the sender to the current channel limitation and also can predict which packets will be probably lost during transmission due to a node’s power supply exhaustion. Hence, based on this information can selectively drop more packets prior to transmission, extending the operation life-span of the network without increasing the received video distortion.

The rest of the paper is organized as follows. Section 2 contains an overview of the multipath routing algorithm and presents the modified LEACH protocol used in the simulations. Section 3 outlines the analytical video distortion prediction model that is used by the proposed packet scheduling. The system model in Section 4 consists of a description of the two proposed scheduling algorithms and of a power consumption mathematical model used during the study. The simulation model and the results are discussed in Section 5. Finally, Section 6 concludes the paper.

2 Multipath selection algorithm

Multipath video transmission has been studied extensively in recent years [4]. The benefits of selecting multiple paths among a video server and a client instead of just the shortest path include among others:
  • reduced correlation among packet losses

  • increased channel resources that can support the application’s demands in QoS

  • the power consumption is more evenly spread in the network nodes preventing node failures

  • ability to adjust to arbitrary congestion occurrences in different parts of the network

The problem of minimizing delay among a video server and a client through optimum selection of multiple paths is addressed in [5]. In [6] an R–D optimization problem is solved using a Markov Decision Process (MDP) framework. The authors studied the case of multiple servers containing data from the same requested video stream and the distortion optimization occurs at the receiver. A single path optimal packet scheduling mechanism for multiple description coded video sequences is presented in [7]. Unlike the work that has been done so far, this paper presents a scheduling scheme over multiple paths, based on video distortion estimation prior to transmission that ensures minimum power consumption and QoS degradation. The scheme takes advantage of the increased aggregate bandwidth of multiple paths and the hierarchical clustered rooting that reduces the power consumption of the sensor nodes.

The multipath selection algorithm that is investigated in this paper is a combination of max-flow and shortest path algorithms [1]. The aim of the algorithm is to select among all the available paths between the source and the destination, these that satisfy the transmission rate requirements of the video stream. If the maximum aggregate bandwidth between the source node and the destination node is greater or equal to the video bandwidth required, then there is at least one path among the source and the destination that can support the end-to-end bandwidth requirement. Therefore the following algorithm can find a set of paths that support the highest overall end-to-end transmission bandwidth.

Initially, the max-flow in the network is computed over all paths in the network. Then the shortest path is selected and the lowest link bandwidth in the shortest path is subtracted from the available bandwidth of each link of the shortest path. This process is repeated until the total bandwidth offered by the multipath set is sufficient for the video application. It can be proved that the algorithm’s iterations converge at the B, where B is the bandwidth constrain (i.e., the video bandwidth required).

This multipath selection algorithm ensures that the resulting set of multiple paths among the source and the destination will satisfy the transmission rate requirements of the video stream. In this study the above algorithm is combined with a low-energy hierarchical routing protocol in order to satisfy the power consumption constrains as well.
  1. A.

    Modified LEACH protocol

     
The LEACH protocol is a hierarchical clustering routing protocol for wireless sensor networks [2] that achieves better resource allocation and power control over the network. Its main characteristics include:
  • stochastic autonomous and flexible clustering

  • local control for data transmission

  • low energy access to the wireless medium

  • data manipulation based on the application characteristics (i.e., data multiplexing at cluster heads)

In brief, according to LEACH every wireless sensor node in the network is organized into clusters according to their distance (indicated by the SNR) from the cluster head nodes. The sensor nodes are transmitting data directly to the cluster head, where the collected data are processed and then transmitted to the corresponding Base Station. The result is low power consumption in the sensor nodes and high consumption at the cluster heads. In order to avoid a failure at the cluster head, which will cause the entire cluster to disconnect, the LEACH protocol suggests a periodic random selection of cluster heads. The final result is a very power efficient routing scheme that minimizes the power requirements in the intra-cluster transmissions (sensor node-to-cluster head).

For the purposes of this study the LEACH protocol has been modified in order to support the multi-path routing capability. Particularly, the direct communication link among the cluster heads and the receiving host (e.g., Base Station) has been modified to allow link establishment among the cluster heads. Hence, a video sensor node can select a number of available paths through other cluster heads in order to transmit its data to the receiving node. This modification to the LEACH protocol that allows cluster heads to connect to each other wirelessly decreases the transmission power, since the distance between the cluster heads is usually smaller than between a cluster head and the receiving end (e.g., Base Station). Moreover, the multipath routing among cluster heads upgrades the bandwidth efficiency of the original LEACH protocol, since instead of just one path between a cluster head and the receiving node, there are multiple paths with higher aggregate capacity limit, which is very important for video streaming applications. Figure 1 illustrates the modified LEACH hierarchical clustering with the multiple links among the cluster heads.
https://static-content.springer.com/image/art%3A10.1007%2Fs11036-008-0061-5/MediaObjects/11036_2008_61_Fig1_HTML.gif
Figure 1

The illustration of the modified LEACH with ten clusters and the added capability of multipath routing among cluster heads

3 Video distortion model

  1. A.

    Previous Loss Models

     

There is a large number of research works that has been reported and regards modeling the impact of packet loss on video distortion. Such models can be fall in two categories. In the first category, the models consider that distortion is proportional to the number of losses within a video sequences [8], [9]. These studies also suggest that the average distortion of multiple losses can be derived as a superposition of the uncorrelated error signals. However, these models are accurate for low residual error rates and when such errors are sufficiently apart of each other and there are no burst errors. The models in the second category consider the correlation between error signals, giving rise to more complex loss patterns, burst of losses and losses separated with small lags, than just isolated losses. Evidently, burst losses lead to larger distortion than individual single losses. In this case, the burst length affects the video quality in a distinct way and it has been determined analytically for different packet losses including burst errors and errors with lag [10].

However, all the models mentioned above have not considered the inherent feature of H.264/AVC encoder that can select between a number of previously encoded frames highly correlated with the current frame, as reference for motion-compensated prediction of each inter-macroblock or macroblock partition. The use of multiple reference pictures allows H.264/AVC to achieve significantly better compression than any previous standard and on the same time, it affects the error propagation in the case of an error frame. Accurate distortion models are very important especially when decisions like rate-distortion optimization and packet scheduling are based on these models.
  1. B.

    Proposed Model

     

As opposed to the additive model of [8], which assumes that the distortion in the decoded sequence is a superposition of the individual isolated distortions due to error frames, the proposed distortion prediction model is a recursive formula that takes into account two important parameters. Firstly, the correlation among video frames during the intra-frame period and secondly, the impact that the use of reference frames has on the distortion propagation phenomena in the case of a frame loss. Furthermore, the proposed distortion model incorporates the random behavior of losses in the wireless medium [11]. Specifically, the random nature of the wireless channel among video sensor nodes in a WMSN, will result into losses of the transmitted video data. It has been reported in [10] that these losses can be characterized as single or isolated losses, burst of losses and losses separated with a small lag. Apparently, the effect that each of these three types of losses has on the received video sequence is very different because of the inter-frame and intra-frame dependences in the encoded video. Therefore, the proposed model includes all the pre-mentioned parameters, thus can accurately predict the resulted video distortion due to any error pattern that may include one or more of isolated, burst or errors with lag.

In order to analytically express the distortion model, a list of previously encoded reference frames with size MREF that is used during the encoding and decoding processes for motion-compensated prediction, is defined. This parameter accounts for the impact of the number of reference frames on the distortion propagation. Moreover, each frame is coded into a number of video packets according to each size. Finally, a simple error concealment mechanism, which replaces a lost frame with its previous at the decoder, is applied. The proposed model includes analytical models for a single frame loss, a burst of losses with variable burst length B (where B≥2) and frame losses separated by a lag.
$$D_n = D_{n - 1} - \sum\limits_{\mathop {k = F_{n - 1} }\limits_{i = 0} }^{\mathop {i = N - 1 + k}\limits^{k = F_{n - 1} ^{ + i} } } {\Lambda _{\left( i \right)} \sigma ^2 \left( {F_{n - 1} } \right) + \sigma ^2 \left( {F_{n - 1} } \right) + } \left\{ {\begin{array}{*{20}c}{\sum\limits_{\mathop {k = F_n }\limits_{i = 0} }^{\mathop {i = N - 1 + k}\limits^{k = F_n + i} } {\Lambda _{\left( i \right)} \sigma _s^2 \left( {F_n } \right){\text{, isolated errors}}} } \\{\sum\limits_{\mathop {k = F_n }\limits_{i = 0} }^{\mathop {i = N - 1 + k}\limits^{k = F_n + i} } {\Lambda _{\left( i \right)} \sigma ^2 \left( {F_n } \right){\text{, burst errors}}} } \\{\sum\limits_{\mathop {k = F_{n - 1} }\limits_{i = 0} }^{\mathop {i = F_n - 1 - F_{n - 1} }\limits^{k = F_n - 1} } {\Lambda _{\left( i \right)} \sigma ^2 \left( {F_n } \right) + \sum\limits_{\mathop {k = F_n }\limits_{i = 0} }^{\mathop {i = N - 1 + k}\limits^{k = F_n + i} } {\Lambda _{\left( i \right)} \sigma ^2 \left( {F_n } \right){\text{, lag errors}}} } } \\\end{array} } \right.$$
(1)

In the above formula, Dn is the total video distortion due to an erroneous pattern of n-frames, where n ≥ 1 and Fn is the frame number of the nth erroneous frame. Moreover, Λ(i) represents the distortion propagation effect until the end of the intra period N due to an error occurred at frame k. Additionally, the error power introduced in a single frame Fn is denoted by \(\sigma _s^2 \left( k \right)\) and the total video distortion due to error frame k and its error power propagation to the following frames is denoted by Ds(k). Correspondingly, σ2(k) and D are the MSE and the sum of the MSE values over all frames in the intra frame period, of more general loss patterns, respectively.

The proposed model presented in this section, has been evaluated against real measurements. Extensive simulations have been performed for different combinations of error patterns. Additionally, the proposed model was compared against the additive distortion model introduced in [8]. The simulation results that prove the accuracy of the proposed model for different video sequences are presented in details in [11].

4 System model

In a wireless multimedia video sensor network where the sensor nodes are wirelessly connected to each other, the QoS requirements of a video service can not be met due to the error prone behavior of the wireless channel and the limitations of the transceiver hardware of the nodes [12]. Therefore, a multipath approach can be adopted, where the source (i.e., a video sensor node) delivers the data to its destinations via multiple paths, thereby supporting an aggregated transfer rate higher than what is possible with any one path. In order to enhance the perceived video QoS and at the same time extend the power efficiency of all the sensor nodes in the system, two packet scheduling algorithms are proposed in this Section, following a power consumption model that is based on the LEACH protocol and was proposed in [2].
  1. A.

    Power consumption model

     
In order to model the power consumption of a video sensor node a simple radio communication model is considered. According to this model, a node may consume power during the reception or the transmission of data. It is important to underline that during this study, only the server (i.e., the video sender) and the receiving nodes have the ability to process the video data (encode and decode) and the energy loss required for these processes are not considered in the analysis. In particular, the energy required by the sensor node to transmit video packet of n bits is proportional to the square of the distance r2 between the transmitting and the receiving nodes. This relationship stands in the case of a compact wireless sensor network setup where the nodes are not located far apart from each other, which holds true for a security surveillance video sensor network. Hence, if the energy consumed in the node’s electronic apparatus is Eelec and the energy requirements for the signal amplification during transmission is Eamp then the energy consumption in a video sensor node EVSN_T for the transmission of an n-byte long video packet is given by:
$$E_{{\text{VSN\_T}}} = n \cdot E_{{\text{elec}}} + n \cdot E_{{\text{amp}}} \cdot r^2 $$
(2)
While the energy requirements for receiving an n-byte long video packet EVSN_R are represented by the following relationship:
$$E_{{\text{VSN\_R}}} = n \cdot E_{{\text{elec}}} $$
(3)
Following the energy analysis in [2] for minimum energy multihop routing, each sensor node is considered to have equal energy resources and the power consumption in every cluster is the same. According to the LEACH protocol, during the first phase of cluster head selection each cluster head is selected randomly. The cluster head advertises its status to the neighboring nodes. The sensor nodes that receive multiple advertisements from cluster heads, decide upon which cluster will join based on the channel conditions and the received signal power. The decision phase is followed by the notification phase during which each node notifies the selected cluster head about its decision to join the cluster. In every cluster the cluster head chooses a number of nodes to join in. Therefore, considering uniformly distributed wireless video sensors, each cluster includes l / k nodes including the cluster heads, where l is the total number of sensor nodes and k is the optimum number of nodes in a cluster. Thus, the energy that is consumed by each cluster head during this cluster head decision process is calculated by:
$$E_{{\text{ch\_elec}}} = n\left( {\frac{l}{k}} \right)E_{{\text{elec}}} + nr_{{\text{ch}}}^2 E_{{\text{amp}}} $$
(4)
Where, n is the size of the message, rch is the mean distance between the cluster head and the neighboring nodes. Similarly, the energy consumed by the non-cluster heads nodes during the cluster head selection procedure is:
$$E_{{\text{non\_ch\_elec}}} = \left( {1 + k} \right)lE_{{\text{elec}}} + lE_{{\text{amp}}} r_{{\text{ch}}}^2 $$
(5)

Eqs. (2), (3), (4) and (5) calculate the power consumed in each wireless sensor node during the clustering process and the video packet transmission and reception.

This analytical model of power consumption in wireless multimedia sensor networks is combined with a packet scheduling algorithm named “Power aware” packet scheduling, and forms the decision bases for dropping packets prior to transmission in order to avoid loosing important packets due to power failures of cluster head nodes. The performance of the proposed “Power aware” scheduler in terms of power consumption and perceived video quality is compared to a “Baseline” packet scheduler that does not consider the possibility a node failing due to energy drought. The detailed description of the two algorithms is provided in the following sections.
  1. B.

    Baseline packet scheduling

     

In a wireless multimedia sensor network that has been organized into clusters based on the proposed modified LEACH protocol the cluster heads are connected to each other via multiple paths. Under these conditions the “Baseline” packet scheduling algorithm schedules the transmission of video packets via multi paths by dropping the excess video traffic in order to prevent network congestion. This algorithm does not take into account the energy levels of the cluster heads involved in the transmission of the video packets prior to the transmission. The name “Baseline” is selected since this algorithm will be the baseline for the simulations in the rest of this paper.

In more details, channel resources in a WMSN are scarce and there are cases when the transmission requirements exceed the available aggregate transfer rate of the multiple paths. If the required rate for error free transmission (RTR) is higher than the current available aggregate transmission rate (ATR) then the sender decides which video packets will be optimally dropped in order to adapt its current rate to the allocated one. The packets to be dropped are selected according to their impact to the overall video distortion. A combination of one or more video packets may be omitted prior to the video transmission by the video source. Dropping a packet imposes a distortion that affects not only the current video frame but all the correlated video frames. The intelligence of the packet scheduling algorithm is that utilizes the distortion prediction model (1), which considers the correlation among the reference frames, thus it selects the optimum pattern of packets to drop in each transmission window.

This process is neither time nor power consuming, as the transmission window is generally small and the mathematical calculations are not of high complexity. The transmitted packets are distributed among the available routes according to their impact in the video distortion; hence packets of high importance are transmitted through the higher capacity routes. When the remaining power of a cluster head node drops bellow a threshold value (5% of the original power levels in the beginning of the simulation) where it can no longer receive or transmit a packet, this node ceases from being a cluster head and according to the modified LEACH protocol a new node with the highest energy resources among the rest in the cluster is selected as the new cluster head. Therefore, before the new transfer window opens new routes are formed between the video source and the destination node. Then based on the new network setup the “Baseline” scheduler decides if the aggregate channel capacity of the multipaths is sufficient for supporting the transmission of the newly arrived video packets. Moreover, the algorithm selects the optimum distribution between the available routes of the video packets to be transmitted. The transmitted packets that were temporarily stored in the buffer of the old cluster head and could not be transmitted in time before the node’s power reached the threshold are considered lost by the receiver node.

This algorithm, which is modeled as a flow chart in Fig. 2, is called the “Baseline” since it only considers the bandwidth limitations of the wireless video sensor network in every transmission window. This algorithm will be compared against the second scheduling approach which, apart from the channel resources it reflects upon the power consumption of the sensor nodes, as well.
  1. C.

    Power aware packet scheduling

     
https://static-content.springer.com/image/art%3A10.1007%2Fs11036-008-0061-5/MediaObjects/11036_2008_61_Fig2_HTML.gif
Figure 2

Process flow of the “Baseline” packet scheduling algorithm

Similarly to the “Baseline” scheme this algorithm utilizes the distortion model in (1) and assigns the importance of each packet according to its impact in the received video distortion. In addition, the decision on which and how many packets will be dropped prior to transmission, whenever the aggregate multipath transmission rate cannot meet the QoS requirements of the transmitted video, is based on each packet’s importance. Evidently, the two algorithms have a similar logic as far as the packet dropping for adapting the current transmission rate to the channel conditions, is concerned. Their difference lies on the fact that this scheduling algorithm implements a sensor power level prediction routine, which allows the scheduling of the packet transmission or dropping to be decided on the bases of both the bandwidth limitations of the channel and the energy efficiency of the cluster heads nodes prior to transmission.

In details, the “Power aware” packet scheduling algorithm showed in Fig. 2, estimates the power that will be consumed by every cluster head in all the multiple paths in the network if the “Baseline” scheduling criteria were to be applied. Thus, it can predict whether a sensor node will be able to receive and transmit all the packets that will go through it in the next transmission window without consuming all its power. In case where the scheduling algorithm predicts that one or more cluster head nodes will not be able to receive and transmit the data that pass through them before their power supply is ended, then it calculates how many packets of the next transmission window these nodes will not be able to process and based on the distortion model of (1) the algorithm selects to drop those packets prior to transmission.

Therefore, this scheme can control the life span of the video sensor network by estimating the remaining power efficiency of each cluster head node in all the selected routes. Moreover, this approach reduces the resulted distortion of the decoded video sequence since it decides upon which and how many packets will be dropped due to transmission rate limitations and power failure of the nodes prior to the transmission. The simulation results presented in this study indicate the efficiency of the proposed packet scheduling algorithm (Fig. 3).
https://static-content.springer.com/image/art%3A10.1007%2Fs11036-008-0061-5/MediaObjects/11036_2008_61_Fig3_HTML.gif
Figure 3

Process flow of the “Power aware” packet scheduling algorithm

5 Simulation results and discussions

  1. A.

    Simulation setup

     
The network topology used during the simulations is illustrated in Fig. 4. Initially the network is comprised of 120 wireless video sensor nodes distributed randomly over an L × L area with L = 100 m organized into eight clusters. These sensor nodes are assumed to be capable of capturing, encoding and broadcasting live video sequences to a receiving point. According to the modified LEACH protocol the elected cluster heads are connected with each other, creating multiple paths towards the receiving point. Table 1 includes the parameters and their values that were initially selected for the simulation. Each path is subjected to variable background traffic with constant bit rate (CBR) over UDP in order to increase the virtual collisions. The video sequence is encoded according to H.264/AVC standard with a reference frame list of size five frames for compensated prediction [13]. The video testing sequence Foreman is used at QCIF resolution with 300 video frames at a frame rate of 30 fps with a constant quantization step that results at an average PSNR of 36dB. The inter-frame period is 36 frames and is set to be equal with the transmission window. The video frames are encapsulated into RTP packets of size 1,024 bytes [14]. The generated video packets are delivered through the 802.11 system at the form of UDP/IP protocol stack in a unicast transmission mode (one video source—one sink) through multiple paths.
  1. B.

    Simulation results

     
https://static-content.springer.com/image/art%3A10.1007%2Fs11036-008-0061-5/MediaObjects/11036_2008_61_Fig4_HTML.gif
Figure 4

WMSN topology setup with multipath routing between cluster heads according to modified LEACH. Eight different clusters are shown where each sensor node (SN) is wirelessly connected directly to its local cluster-head. The multiple paths assigned among cluster-heads are shown in red

Table 1

Values of parameters used during the simulation

Description

Parameter

Value

Power dissipated at the amplifier for uplink transmission

Eamp

10 pW/bit/m2

Power dissipated at the electronic circuit of the sensor node (ignoring video codec)

Eelec

0.0013 pW/bit/m4

Number of wireless multimedia sensor nodes

N

120 nodes

Network covered area

L

100 m × 100 m

Initial power available at each node

Estart

0.1 W

RTP packet size

k

1024 bytes

Video frame rate

r

30 fps

Number of reference frames

MREF

5 frames

The proposed “Baseline” and “Power aware” scheduling algorithms have been compared and evaluated in terms of perceived video quality and power consumption. The simulation was performed by transmitting the testing video sequence Foreman once in every simulation cycle and scheduling the packet transmission according to “Baseline” and “Power aware” algorithms.

Figure 5 illustrates the PSNR per frame for both scheduling schemes during the last two simulation cycles (cycles 15 and 16) represented in the figure by frames 0–300 and 301–600 respectively. In simulation cycle 16 the remaining “live” sensor nodes are scarce and far between as it is shown in Fig. 6, thus the available paths among the video source and the destination are limited. Under these network conditions the packet losses are severe and the resulting PSNR under the “Baseline” scheduling is very low. However as it can be seen in Fig. 5 the resulting PSNR under the “Power aware” algorithm (under the same conditions in the 16th simulation cycle) is significantly better and what is more important it is almost the same as in the 15th simulation cycle where transmission losses were fewer than in the 16th cycle. The “Power aware” scheduling algorithm achieves better perceived quality due to limited PSNR variations. In particularly, it improves the perceived QoS of the video sequence.
https://static-content.springer.com/image/art%3A10.1007%2Fs11036-008-0061-5/MediaObjects/11036_2008_61_Fig5_HTML.gif
Figure 5

PSNR per frame for “Baseline” and “Power aware” scheduling algorithms, the Frame index is extended to cover for the two consecutive video streams where most losses occurred

https://static-content.springer.com/image/art%3A10.1007%2Fs11036-008-0061-5/MediaObjects/11036_2008_61_Fig6_HTML.gif
Figure 6

Number of active sensor nodes during each simulation cycle

In addition to the above, the average video distortion under different transmission rates illustrated in Fig. 7, indicates that under the proposed “Power aware” scheduler the average PSNR can be improved by up to 8 dB at 40 kB/s. The average PSNR is constantly higher for the “Power aware” scheme than the “Baseline” algorithm because it selectively drops packets that will be lost during the transmission when a node’s power fails, based on the distortion prediction model. The available transmission rate is calculated in every transmission window; hence the peaks and lows in the plot indicate transmission windows that include a higher number of important video packets. The “Baseline” scheme is more susceptible to dropping video packets of higher importance.
https://static-content.springer.com/image/art%3A10.1007%2Fs11036-008-0061-5/MediaObjects/11036_2008_61_Fig7_HTML.gif
Figure 7

Average PSNR versus aggregate channel available transmission rate. Comparison between Baseline and Power aware scheduling algorithms

As it was expected the “Power aware” scheduling algorithm outperforms the “Baseline” scheme in terms of energy efficiency. Figure 8 indicates that under the “Power aware” scheme the aggregate power levels are significantly higher than the aggregate power level resulting from the “Baseline” algorithm. In fact it can be seen that the “Power aware” scheduling is almost six times more efficient than the “Baseline” after 16 simulation cycles. This is due to the fact that “Power aware” scheduling estimates how many packets will potentially be lost during the transmission because of a node failure and selectively drops the low importance packets prior to transmission. As it has been shown this does not impose larger distortion to the received video sequence because the decision of which packets will be dropped is based on the distortion model (3).
https://static-content.springer.com/image/art%3A10.1007%2Fs11036-008-0061-5/MediaObjects/11036_2008_61_Fig8_HTML.gif
Figure 8

Power consumption over simulation time. Comparison between Baseline and Power aware packet scheduling algorithms

6 Conclusion

In this paper two scheduling algorithms “Baseline” and “Power aware” were proposed that optimize the video transmission over a wireless multimedia sensor network. A modified version of the LEACH protocol has been proposed that allows the establishment of multiple routes among the elected cluster heads. Therefore, multiple paths connect the video sender and the destination point, increasing the performance of the WMSN in terms of perceived video quality and power consumption. In addition to selecting multiple paths in order to increase the available transmission rate, the two scheduling algorithms selectively drop video packets prior to transmission in order to adapt the rate of the video sender to the limitations of the wireless channel. Moreover, the “Power aware” algorithm estimates when a potential power failure of a node may occur, thus predicts how many video packets will be lost. Based on this information, the scheduling scheme can decide to drop a pattern of packets that will not increase the distortion of the received video. Simulation results proved the efficiency of the proposed scheduling algorithm in terms of perceived video quality (PSNR) and the power consumption.

Acknowledgements

This work is supported by the project PENEDNo.03636, which is funded in 75% by the European Social Fund and in 25% by the Greek State—General Secretariat for Research and Technology.

Copyright information

© Springer Science+Business Media, LLC 2008