Green two‐tiered wireless multimedia sensor systems: an energy, bandwidth, and quality optimisation framework
2016; Institution of Engineering and Technology; Volume: 10; Issue: 18 Linguagem: Inglês
10.1049/iet-com.2016.0406
ISSN1751-8636
AutoresNguyen‐Son Vo, Dac‐Binh Ha, Berk Canberk, Junqing Zhang,
Tópico(s)Advanced MIMO Systems Optimization
ResumoIET CommunicationsVolume 10, Issue 18 p. 2543-2550 Special Issue: Green Computing and Telecommunications SystemsFree Access Green two-tiered wireless multimedia sensor systems: an energy, bandwidth, and quality optimisation framework Nguyen-Son Vo, Corresponding Author Nguyen-Son Vo vonguyenson@dtu.edu.vn Faculty of Electrical and Electronics Engineering, Duy Tan University, Da Nang, VietnamSearch for more papers by this authorDac-Binh Ha, Dac-Binh Ha Faculty of Electrical and Electronics Engineering, Duy Tan University, Da Nang, VietnamSearch for more papers by this authorBerk Canberk, Berk Canberk Department of Computer Engineering, Istanbul Technical University, Istanbul, TurkeySearch for more papers by this authorJunqing Zhang, Junqing Zhang School of Electronics, Electrical Engineering and Computer Science, Queen's University Belfast, Belfast, UKSearch for more papers by this author Nguyen-Son Vo, Corresponding Author Nguyen-Son Vo vonguyenson@dtu.edu.vn Faculty of Electrical and Electronics Engineering, Duy Tan University, Da Nang, VietnamSearch for more papers by this authorDac-Binh Ha, Dac-Binh Ha Faculty of Electrical and Electronics Engineering, Duy Tan University, Da Nang, VietnamSearch for more papers by this authorBerk Canberk, Berk Canberk Department of Computer Engineering, Istanbul Technical University, Istanbul, TurkeySearch for more papers by this authorJunqing Zhang, Junqing Zhang School of Electronics, Electrical Engineering and Computer Science, Queen's University Belfast, Belfast, UKSearch for more papers by this author First published: 01 December 2016 https://doi.org/10.1049/iet-com.2016.0406Citations: 7AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinkedInRedditWechat Abstract In wireless multimedia sensor systems (WMSSs), the devices are equipped with multiple energy-constrained camera sensors (CSs) distributed over bandwidth-constrained and lossy wireless channels, in catastrophe-prone areas. Meanwhile, multimedia applications, e.g. video streaming, require considerable energy and bandwidth resources to gain long lifetime and high streaming quality. This study proposes an energy, bandwidth, and quality (EBQ) optimisation framework for green two-tiered WMSSs. The first tier contains the CSs and the second tier includes cluster heads (CHs) selected from the CSs with higher available energy and processing capacity. In the EBQ optimisation framework, a rate allocation optimisation problem is formulated under given constraints of available backhaul bandwidth of the CHs and quality of received videos at base stations (BSs). This problem is solved for optimal encoding rates to packetise each video captured from different environments into multiple descriptions for transmission. Consequently, the average energy consumption per CS is minimised for long lifetime while conserving the bandwidth of the CHs and guaranteeing high quality of received videos for the purpose of monitoring at the BSs. Simulations demonstrate that the proposed EBQ optimisation framework can efficiently enhance the performance of green two-tiered WMSSs in terms of minimum energy consumption, bandwidth efficiency, and high quality. 1 Introduction Natural disasters (e.g. volcanoes, tsunamis, typhoons, floods, and cyclones) and rapid urbanisation threats (e.g. water pollution, carbon emission, surface temperature, and urban noise) have been disrupting and destroying people's lives around the world, especially in many countries in South-East Asia. Extensive rescue solutions, for instance, improving infrastructures for dyke systems, storm shelters, flood control works, and/or government policies, have not been efficient due to the lack of information and communications in remote catastrophe-prone areas [1]. In the most effective ways, a risk management system based on information and communication networks (ICNs) for natural disasters and rapid urbanisation threats must be built to provide capabilities that can help the government officials, first responders, emergency agents, and citizens to grasp the real varying danger in remote catastrophe-prone areas, so as to quickly and exactly make decisions on mitigation, preparedness, response, and recovery. To ensure flexibility and low cost, ICNs can be established based on partial or full integration of conventional wireless sensor networks (WSNs), wireless mesh networks, mobile ad hoc networks, and mobile cellular networks (MCNs). In this paper, we focus on the integration of WSNs and MCNs to propose a green two-tiered wireless multimedia sensor system (WMSS) for the purpose of grasping the varying danger in catastrophe-prone areas. In WMSS [2], the sensor nodes, which are connected in high density, are equipped with camera sensors (CSs) to capture their surrounding phenomena. They then communicate with each other to send the captured multimedia data, e.g. video contents, to their cluster heads (CHs). The CHs are selected from the CSs with most powerful resource to process complicated tasks and communicate with each other and with the base stations (BSs) in MCNs. In this way, the two-tiered WMSS can collect video contents of ambient phenomena and send them to the BS to monitor, analyse, measure, and make decision, for governance purposes toward sustainable development. Although WMSSs have been fostered by the development of affordable and low-power hardware device technologies to enable delivery of multimedia contents, the quality of service (QoS) of most potential applications in WMSSs is limited to some predetermined levels. This is because the inherent characteristics of devices and channels in WMSSs are energy- and bandwidth-constrained and error-prone. To achieve high QoS otherwise, multimedia applications consume considerable amount of energy and bandwidth resources of WMSSs. More new added studies related to different aspects of techniques, protocols, and standardisations in the literature have been proposed so that WMSSs can gain high efficiency of energy and bandwidth resources and high QoS, to name a few, e.g. coverage-enhanced algorithms [3, 4], routing [5, 6], relay selection [7], clustering [8], scheduling [9], and source encoding schemes [10-12]. Optimising the monitored coverage area and the number of active CSs is the major challenge in WMSSs. The problem is that many CSs may totally or partially capture a particular area causing wasted energy for processing and transmitting too much redundant/replicated video contents. To cope with this problem, coverage-enhanced algorithms have been introduced to reduce the overlap area by adjusting sensing directions of the CSs [3] and handling multiple events [4]. In this way, unnecessary CSs are made idle to save energy for long network lifetime while ensuring large monitored coverage area. Prolonging the network lifetime of WMSSs has been also achieved by routing protocols [5, 6]. Particularly, it has been shown in [5] that the proposed content relevance opportunistic routing protocol can increase the network lifetime of up to 20% while providing lowest end-to-end delay compared with traditional one. In the same objective but different solution, the authors in [6] have observed that the selection of forwarding candidate set in opportunistic routing protocol impacts on the network lifetime and the delay. To this end, an angle-based QoS and energy-aware dynamic routing protocol has been proposed to appropriately select the forwarding candidate set, and thus prolong the network lifetime while guaranteeing the delay constraint. Similarly, Yao et al. [7] presented transmission power control and relay node selection strategies, which are jointly optimised to improve multimedia transmission quality based on the harvestable energy input of each CS. Cross-cluster handover mechanism and path redirection scheme [8] and utility-based scheduling algorithm [9] have been studied as efficient ways to gain both high energy efficiency and low delivery delay. However, all the aforementioned studies have not analysed the characteristics of video contents and the conditions of wireless channels to optimally encode the video contents for transmission with minimum energy consumption while guaranteeing high quality and satisfying limited bandwidth resource in WMSSs. The most efficient solutions for high efficient video streaming over wireless channels are source encoding schemes [10-12] to optimise the tradeoff between energy/lifetime and quality. In [10], the authors modelled a power-rate-distortion (RD) framework for video encoding. This framework enables us to encode source videos to minimise the distortion while satisfying a given available energy supply of devices. By further taking into account wireless transmission, Wang et al. [11] proposed a joint optimisation of video quality, content protection, and communication energy efficiency over error-prone wireless channels in WSNs. In particular, channel-aware selective encryption at application layer and unequal error protection-based network resource allocation at lower layers were proposed to minimise the extra encryption dependency overhead and gain high communication efficiency, respectively. As a result, the video transmission quality is significantly improved while guaranteeing content protection and energy efficiency. Interestingly in [12], by extending the power-RD framework to include wireless transmission conditions, a joint performance optimisation of maximum network lifetime and minimum video reconstructed distortion for energy-constrained WMSSs was addressed. The authors proposed a framework for joint source/channel rate adaptation, which is found by solving three sub-optimisation problems of rate control, distortion control, and energy conservation. In our work, we aim to address similar optimisation problem of minimising energy consumption for long network lifetime and guaranteeing high video streaming quality studied in [12]. However, we further apply layered multiple description coding with embedded forward error correction (LMDC-FEC) [13-16] to scalable extensions of high efficiency video coding (HSVC) [17]. LMDC-FEC-based HSVC allows us to packetise the video contents into multiple descriptions for transmission in order to achieve highest data protection against error and delay multi-hop channels as well as to adapt to diverse limited bandwidth in WMSSs, i.e. without channel rate adaptation considered in [12]. Therefore, instead of finding many optimisation variables which may make Lagrangian-based solution for global optimal results prohibitively high and impractical in large-scale networks [12], we only find optimal encoding rates allocated to each video content by using genetic algorithms (GAs) within reasonable time frame [16, 18]. In particular, we propose an energy, bandwidth, and quality (EBQ) optimisation framework to efficiently deliver video contents over the green two-tiered WMSS. In this framework, a rate allocation optimisation problem (RAP) is formulated, where RD and priority/importance characteristics of video contents, video encoding scheme (i.e. LMDC-FEC-based HSVC), and conditions of WMSS (i.e. energy-constrained CSs, bandwidth-constrained CHs, and lossy features of wireless channels) are jointly taken into account. We then solve the RAP for optimal rates allocated to each video content so that the video contents are packetised into multiple descriptions for transmission from the CSs to the BSs. Consequently, the proposed green two-tiered WMSS can collect the video contents of ambient phenomena for the purpose of monitoring, analysing, measuring, and making decision toward sustainable development in the manner of minimum average energy consumption per CS while guaranteeing high received video quality at the BSs and satisfying given limited backhaul bandwidth at the CHs. The rest of this paper is organised as follows. We introduce system models and problem formulations of the two-tiered WMSS in Section 2. On the basis of these models and formulations, Section 3 presents the RAP and solution using GAs. In Section 4, simulation results are shown to evaluate the performance of the proposed EBQ. Finally, we conclude the paper in Section 5. 2 System models and problem formulations 2.1 System models In our system models, we present a typical architecture of two-tiered WMSS. The EBQ optimisation framework (hereinafter referred to as EBQ) is then proposed to show how video contents are captured by the CSs and sent to the BS via the CHs optimally. In addition, we introduce a source video packetisation scheme to packetise the video contents into multiple descriptions for transmission from the CSs to the BS. Some important notations used in our system models are specified in Table 1. Table 1. Notations Symbols Descriptions C number of CHs Si number of camera sensors (CSs) of the CHi, i = 1, 2,…, C CSi, j jth CS of the CHi, j = 1, 2,…, Si Vi, j jth video content captured by the CSi, j M number of layers/descriptions of each video Ri, j upper bound of encoding rate, i.e. full resolution, of Vi, j Li, j size of Vi, j measured in bits at full resolution optimal encoding rates of Vi, j, m = 1, 2,…, M reconstructed distortion of Vi, j encoded at rate size of Vi, j measured in bits after encoding with embedded FEC average energy consumption per CS to transmit the Vi, j from the CSi, j to the CHi average bandwidth served by the CHi number of hops to transmit Vi, j from the CSi, j to the CHi in the first tier number of hops to forward Vi, j from the CHi to the BS in the second tier description error rate of the hth hop on the route from the CSi, j to the CHi in the first tier description error rate of the kth hop on the route from the CHi to the BS in the first tier probability that there are m out of M description of Vi, j correctly received 2.1.1 Two-tiered WMSS We consider a typical architecture of two-tiered WMSS [2] as shown in Fig. 1. The two-tiered WMSS consists of a number of CSs and one BS in which the video contents are received from the CSs via the CHs for the purpose of monitoring. The CSs are partitioned into C clusters; each cluster is in charge of collecting video contents from its catastrophe-prone area. The ith cluster has Si CSs (namely CSi, j), i = 1, 2,…, C and j = 1, 2,…, Si, handled by the ith CH (namely CHi). In this way, the WMSS is arranged in two tiers. The first tier is comprised of the static or mobile CSs with lower available energy and processing capacity. The CSs carry out simple tasks of collecting video contents from the environment, analysing the RD characteristics, and encoding the video contents into descriptions for sending to the BS via the second tier. The second tier includes the CHs, which are in charge of performing complicated tasks, e.g. controlling the CSs in the first tier, solving the RAP in the EBQ, and forward the descriptions to the BS. In this paper, we assume that the CHs with higher available energy and processing capacity are elected by using lightweight and dependable trust system for clustered WSNs given in [19]. Interconnections between the CHs and between the CSs in the form of multi-hop communications are provided to guarantee high connectivity of WMSS. Fig. 1Open in figure viewerPowerPoint Architecture of two-tiered WMSS 2.1.2 EBQ for two-tiered WMSS The EBQ designed for the two-tiered WMSS is illustrated in Fig. 2. The main objective of EBQ is to formulate and solve the RAP to minimise average energy consumption per CS for long lifetime of WMSS while conserving backhaul bandwidth resource of the CHs and guaranteeing high quality of received video streams at the BS. Whenever there is a request for video contents sent by the BS, the EBQ performs three steps as follows: Step 1: The CHs communicate with each other to vote a super node (SN), which is the most powerful among the others. The CSi, j captures the video content Vi,j at full resolution Ri,j depending on its integrated camera hardware configuration, in a predefined duration. All captured video contents are analysed to obtain their RD models by the CSs. Then, the CSi,j send the information of RD model of Vi,j to the SN via the CHi. Step 2: The SN collects the energy conditions of the CSs, the backhaul bandwidth guarantees of the CHs, and the channel state information, e.g. lossy features of wireless channels, from the whole WMSS. Together with the RD models, the SN formulates the RAP and solves it for optimal encoding rates , allocated to the Vi,j, here m = 1, 2,…, M and . The optimal rate results are sent back to the corresponding CSi,j for packetising. Step 3: On the basis of the optimal rate results, the CSi,j packetises the Vi,j into descriptions for transmission to the BS via the CHi. The source video packetisation scheme is presented in the sequel. Fig. 2Open in figure viewerPowerPoint EBQ optimisation framework 2.1.3 Source video packetisation scheme In our EBQ, the RAP is formulated based on the RD models of video contents sent from the CSs. We apply the RD model introduced in [18] to scalability extension of high efficiency video coding (SHVC) [17]. In this RD model, the source reconstructed distortion of a video content is a decaying exponential function of encoding rate (Kbps), given by the following equation (1)where γi,j and βi,j are the dependent parameters of Vi,j. In step 1 of EBQ, the captured video contents are analysed to obtain their RD models featured by γi,j and βi,j [18]. Recently, SHVC has been developed by the joint collaborative team on video coding to exploit the benefits of scalable H.264/advanced video coding (AVC) [17]. We therefore utilise SHVC to cope with the diverse resolution of different CSs and condition of wireless channels in the two-tiered WMSS, as AVC does. To further overcome the problems of lossy and delay multi-hop, we apply LMDC-FEC [13-16] to SHVC [17] to packetising the video contents into descriptions for transmission over the two-tiered WMSS. Although LMDC-FEC-based SHVC yields higher coding complexity than pure scalable AVC, it meets the current road map of processing technology, which has been developed faster than other technologies. The LMDC-FEC-based SHVC packetisation scheme is presented in detail as below. Let us consider the Vi, j and its RD model as shown in Fig. 3. After obtaining the optimal rates from step 2 of EBQ such that , the Vi, j is divided into M layers [16]. The mth layer is denoted by two rate points and . In this way so far, the reconstructed distortion of Vi, j depends on not only the number but also the priority order of correctly received layers. If the mth layer is not correctly received, all layers from m + 1 to M become useless even they are correctly received. Fig. 3Open in figure viewerPowerPoint LMDC-FEC-based SHVC packetisation scheme To remove the priority order for higher flexibility and performance of reconstruction, the mth layer is then partitioned into m blocks, each has the same size of (2)where Li,j is the original size of Vi,j measured in bits and the floor operator is used to round A to the nearest integer less than or equal to A. This operator results in , but it does not seriously affect the performance of the packetisation scheme because the total reduction in bits is infinitesimal compared with Li,j. For simplicity, we thus do not need to dwell on bit rate adjustment algorithm to compensate for the dropped bits caused by the floor operator. A Reed–Solomon erasure code is applied to m original blocks (of size bits) of the mth layer to generate the codewords of length M blocks (of size bits) with embedded FEC. The can correct up to M–m erroneous blocks of the mth layer of Vi,j. In other words, the first m layers are reconstructed if there are m out of M blocks correctly received. Finally, to obtain M descriptions of Vi,j for transmission, we select the mth blocks of all layers of Vi,j and packetise them into the mth description. To this end, if m out of M descriptions of Vi,j are correctly received (i.e. not depending on the priority order of received descriptions), the first m layers of Vi,j are reconstructed and thus the Vi,j is played back at rate with corresponding reconstructed distortion . The total size of Vi,j with embedded FEC is given by the following equation (3) 2.2 Problem formulations 2.2.1 Energy consumption The main objective of RAP in the proposed EBQ is to minimise the energy consumption for prolonging the lifetime of WMSS. To do so, we assume that available energy levels of the CHs are much higher than that of the CSs. Therefore, we only focus on available energy levels of the CSs to compute average energy consumption per each CS, which is considered as the objective function to be minimised in the RAP. Let be the number of hops on the route from the CSi,j to the CHi and be the wireless distance measured in metres of the hth hop. Following [20], the energy consumption to transmit a bit over metres is given by the following equation (4)where θ1 = 50 nJ/bit/hop, θ2 = 0.0013 pJ/bit/m4/hop, and η = 4. To transmit bits of Vi,j over hops, the average energy consumption per CS from the CSi,j to the CHi is expressed by the following equation (5)In (5), we only consider the energy drained from transmission, not from processing. Intermediate CSs on the route from the CSi,j to the CHi are established by applying utility energy-based opportunistic routing protocol for lifetime enhancement [21]. We further assume that the lifetime of this route is much greater than the streaming session of Vi,j. The average energy consumption for transmission of all video contents per CS can be written as follows (6)where ri, j reflects the priority/importance of Vi, j captured from the catastrophe-prone area by the CSi,j. The value of ri,j follows Zipf-like distribution [22], given by the following equation (7)where αi ≥ 0 represents the skewed importance between different video contents handled by the CHi and . If αi > > 0, the Vi,j is much more important than the Vi,j+1, otherwise, if αi = 0, all video contents of the CHi have the same importance. Zipf-like distribution is used to ensure that in each CH, the video contents with higher importance are transmitted with less energy consumption while conserving bandwidth resource and guaranteeing high quality. 2.2.2 Bandwidth conserving The problem of bottleneck at the CHs in the two-tiered WMSS may occur if we do not carefully consider the maximum encoding rate of Vi,j during encoding process. The solution is that we have to derive average bandwidth coming from the CSi,j to the CHi and then limit this average value by keeping it within a given backhaul bandwidth of the CHi in the bandwidth constraint of the RAP. The average bandwidth served by the CHi can be given by the following equation (8) 2.2.3 Quality guarantee For high quality guarantee, we have to calculate average reconstructed distortion of received video streams at the BS, which is considered as the quality constraint of the RAP. Let be description error rate of the hth hop on the route from the CSi, j to the CHi in the first tier and be description error rate of the kth hop on the route from the CHi to the BS in the second tier, the description error probability over the route from the CSi,j to the BS is shown as follows (9)where is the number of hops on the route from the CHi to the BS. As analysed, if there are m out of M descriptions correctly received, the Vi,j is reconstructed and played back at rate with distortion . Let be the probability that m out of M descriptions of Vi,j are correctly received, the reconstructed distortion of Vi,j can be given by the following equation (10)where (11)Taking the importance value ri,j of Vi,j captured by the CSi,j into account, the overall average reconstructed distortion of all video streams received at the BS can be written as follows (12) 3 RAP problem and solution 3.1 RAP problem In the RAP, the average energy consumption (6) is considered as the objective function, which is minimised to prolong the lifetime of the CSs in the WMSS by finding the optimal rates allocated to each video content. In addition, to conserve the backhaul bandwidth of the CHs and guarantee high quality of received video streams at the BS, the average bandwidth (8) served by the CHi and the average reconstructed distortion (12) at the BS, which cannot be greater than given thresholds, are taken into account in the constraints of the RAP. Mathematically, the RAP is expressed as follows (13) (14)In (14), the first two constraints are to satisfy the rate allocation characteristic discussed in Section 2.1.3. and D* in the last two constraints, which are the backhaul bandwidth of CHi and the maximum average reconstructed distortion, are used together with two adjustable coefficients δB > 0 (bandwidth consumption coefficient) and δD > 0 (quality guarantee coefficient) to limit the average bandwidth served by the CHi and guarantee high quality of received video streams at the BS, respectively. Although (13) and (14) can be converted to an unconstrained optimisation problem and then solved by applying Lagrange multipliers [23], the complexity of Lagrangian-based solution for global optimal results of RAP is prohibitively high [16, 18]. We therefore introduce GAs to solve RAP in the following. 3.2 Solution with GAs In practical systems, deploying a global and precise optimisation solution with excessive time and computational complexity is not feasible compared with the one within reasonable complexity. For exactly or approximately optimal solution within a reasonable time frame, a class of adaptive heuristic searching algorithms, i.e. GAs [24], based on the natural evolutionary principles of selection, variation, and inheritance, can be applied as a promising method. However, the problem is that the constraints in (14) are more complicated than the ones in the form of simple lower and upper bounds. This may cause GAs to result in infeasible offsprings that do not satisfy the constraints. To overcome this problem, penalty function approach is used to remove the constraints in (13) and (14) [25]. In this way, the constraints in (14) are first rewritten in sequence as follows (15)Next, the penalty function can be given by the following equation (16)where λ1, λ2, λ3, and λ4 are added to reflect the violation degree of the constraints. Then, pulling (13) and (16) together, we get the following unconstrained optimisation problem: (17)Consequently, GAs can be applied to solve (17) instead of (13) and (14). The detailed implementation of GAs for the RAP is presented in Algorithm 1. Algorithm 1.Implementation of GAsRequire: Initial parameters of GA: NP = 500: Population size, i.e. number of individuals NB = 50: Number of bits to represent an individual of solution set , we set it being equivalent to the number of layers/descriptions M as declared in Table 2 NG = 50: Number of generations PG = 0.9: Generation gap Pc = 0.1: Crossover probability Pm = 10−6: Mutation probability Ensure: 1: Generating NP random individuals of solution {Cc} 2: Calculating fitness values in (17) corresponding to each individual in {Cc} of current generation 3: while NG > 0 do 4: Putting {Cc} and in mating pool 5: Selecting NP × PG best individuals with better fitness values, i.e. lower values of , for breeding the next generation by applying stochastic universal sampling operator 6: Collecting two parent individuals to generate two offsprings by applying single point crossover operator with crossover probability Pc 7: Mutating two new offsprings with mutation probability Pm to recover the good genes that could be lost due to the operators in previous steps 8: Calculating the fitness values of all offsprings and reinserting them into the current population 9: NG ← NG − 1 10: end while 11: Finding the best fitness value and the corresponding best individual C* in the last generation Table 2. Parameter settings Notations Specifications C three CHs Si three CSs or three video contents captured in each cluster M fifty layers/descriptions , randomly distributed from 1 to 5 hops randomly distributed from 10 to 50 metres , randomly distributed from 0.0005 to 0.05 α{αi} α{1, 2, 3}, α changes from 0 to 1 to adjust αi {10,000, 100,000, 30,000} Kbps, obtained by approximately averaging the full resolutions Ri, j in each CH as shown in Table 3 D* 1000 (MSE) δB from 1 to 10 to adjust δD from 0.1 to 1 to adjust D* 4 Performance evaluation 4.1 Parameters setting We deploy the two-tiered WMSS with system and video parameters given in detail in Tables 2 and 3. The considered videos are analysed to generate their RD models by using HM reference software version 12.0 [26]. Table 3. Information of videos Vi, j Pixels|frames γi, j|βi, j Ri, j, Kbps Li, j, Kbit V1,1 foreman 352 × 288|300 9806|−0.9972 3000 L1,1 = 35600 V1,2 mobile 352 × 288|300 3652 × 105|−1.9690 9000 L1,2 =
Referência(s)