車載以太網(wǎng)在汽車電子電氣架構設計中應用研究

李陽春 翟慶

（華晨汽車工程研究院，沈陽110141）

主題詞：以太網(wǎng) 電子電氣架構 CAN總線 智能網(wǎng)聯(lián)汽車

NOMENCLATURE

AVB Audio Video Bridge

CAN Controller Area Network

ECU Electric Controller Unit

EMC Electro Magnetic Compatibility

LIN Local Interconnect Network

PHY Physical Layer

OBD On Board Diagnosis

EE Electrical and Electronic

MOST Media Oriented System Transport

TDMA Time Division Multiple Access

1 Introduction

The demands of vehicular markets and customers around global have been increasing,that leads to new technology progress in next decade,especially the technology trends of connected and intelligent vehicles are evolving fast,the challenges,in this context,of much demands on fast and large amount of data transmission are posing in front of automotive industry around world.

Traditional bus technologies are showing incapability in meeting ever increasing demands automotive industry.Ethernet has been brought into automotive industry after technology migrant from IT,while its protocolalready exists and is mature,just modification is needed for physicallayerto fulfillEMC requirements.Besides excellent bandwidth, the harnesses with Ethernet technologies are lighter and cheaper than traditional ones.These characteristics are the fundamental reasons for OEMsto apply Ethernettechnologies forvehicular development.Until now,however,it is not possible to replace other bus like CAN/LIN with Ethernet from engineering point of view.During EE architecture design,based on function implementation,several kind of bus shall be applied.However,what is strategy for Ethernet application in EE architecture design,to fulfillthe requirement of function implementation or data processing from technology point of view,it is the key issue which confuses OEMs a long time.A strategy is provided in this paper.

2 EE Architecture In Perspective

2.1 Network Technologies

Usually,CAN is main stream bus in EE architecture.Definitely,there are other kinds of bus,Table 1 shows comparison of in-vehicle networking technologies[1].

Table 1 Comparison of in-vehicle networking technologies

These buses are important communication technologies prevailing in vehicle today.The first important thing is that each technology was developed and/or is used with a specific application field in mind:CAN for robust ECU communication,LIN for low cost,MOST for high-end audio,FlexRay for X-by-wire,pixel links for unprocessed video,and consumer links for consumer device integration,the OEM actively drove some of the standardization work behind these buses.Table 1 shows that this has led to very different technologies,not only in respect to the data rates supported,but also in respect to the communication mechanisms and robustness methods used for these buses.Each new use case thus leads to new requirement, standardization efforts, communication principle and qualification processes.While powerful invehicle networking is the fundamental requirement for functional innovation, the number of networking technology needs to be as small as possible.Additionally,OEMs face the challenges of ever faster changing customer expectation and product diversification in the form of new models,new derivatives and potentially shorter life cycles.Modular EE architecture is prominent way to handle these cases and challenges.In-vehicle network for overall demands for future automotive EE architecture has to be flexible and supportive to respond these challenges.

Fig.1 summarizes that,when automotive industry started,cars were purely mechanical devices.Over time electronics were added and are increasingly replacing traditionally mechanical functions as car move toward being high-end computers that drive autonomously.At the beginning,all the technologies in car were propriety.Then automotive started to standardize especially no differentiating functions.With all electronics,cars are supported by new network.

Fig.1 Long-term trends in vehicle networking

2.2 New Perspective

From a consumer industry perspective,it looks very complex to introduce a new in-vehicle networking technology,the reason why Ethernet can be immigrated into Automotive is that,current networking do not fulfill essential automotive requirements.

（1）Long lifetime-the customers want a reliable and robust cars with minimal maintenance.In consequence,OEMs have to consider the aging effects of all components inside the cars,active as well as passive parts,like capacitors.

（2）Upgradeability-it is desirable to be able to update the car,the possibility of adding function later increase the resale value of car.The implications are quite different for different in-vehicle network.In case of automotive Ethernet,one existing ECU in the network would need to have a switch and a currently used PHY in order to allow a new ECU to be integrated into the network.Smart new concept for ensuring the upgradeability thus has to be developed for Ethernet.

（3）Long pauses between uses-Even if car has been parked(Ignition=off)for a long time,car can be started when customers want to use it again.With the increasing amount of ECUs inside the car,this is a challenge as electronics consume power and thus drain the battery of car.The ECUs with standard Ethernet component will consume too much quiescent current.Respective ECUs thus cannot stay connected to battery when car is not in use or other intelligent means for just keeping the parts connected to the power supply that are needed in the off status have to be provided,this is one of the many criteria that need to be considered in EE architecture.

（4）Low power(petrol)consumption when car is in use-the car combines a large number of functions from completely different modular.Not all functions provided inside a car are needed all the time,and so for the power budget,it is advantageous if a function that is currently not needed does not use energy.From users'side,this will become more urgent with the proliferation of electric car,in which drivers might actively switch off the infotainment or air-conditioning system in order to drive more.From OEM side,thisisalso urgentin orderto fulfill governmentalrequirements in terms ofOBD with combustion engines.

（5）Fast availability-inside a car,the customers expect all functions are available immediately,this leads to the requirements of a short start-up time.

（6）On-off change-a network expects not to face extreme temperature oreven a lotoftemperature variations.Additionally the whole network will be shut down only occasionally for maintenance reasons and equipment upgrades.In contrast,an in-vehicle network is started and shut down frequently,often several times during one day.Short start-up times are thus crucial.

（7）Quality and availability of power supply-in case of vehicle power design,the power supply is limited.Furthermore,the power needed will vary,depending on the actual use of car,more or fewer functions are connected to network and the power supply.In case of the service,even only one ECU might be powered.Without giving any further details on how these issues are addressed in EE architecture,all OEMs have different capability in this respect,they have to be taken into account when designing in-vehicle network with Ethernet.

As Dr.Varun from Robert Bosch presented that,Invehicle communication networks with newer standards such as automotive Ethernet will fulfill the requirements for higher bandwidth while providing better security and safety measures[2].These characteristics help enable highly automated driving and high definition infotainment systems in vehicles.

3 In-Vehicle Networking and Architecture Design with Ethernet

3.1 Integrated Versus Distributed EE architecture

During system design,there are two principal approaches on how to partition function block:an integrated and a distributed one.In integrated approach,many function blocks are designed into one ECU,while in the distributed approach only a few function blocks are designed into ECU.This means that to achieve the same functionality,there are fewer ECUs and a smaller communication network in the integrated than distributed approach.The customers are willing to pay for the functionalities,this generally makes the integrated approach less costly.At the same time,the integrated approach is less flexible and scalable.Everyone is offered the same function.Even if some of functionalities are activated by software,so that it is still possible to sell functionalities as options,every customer will have the same hardware provisions in the car.Table 2 provides an overview of differences between integrated and distributed architectures.

Table 2 An overview of differences between integrated and distributed architectures

Overall,the preference also depends on the market segment that OEMs want to address,either with a specific car or in general.As a rule of thumb,integration is more common for basic functionalities.Distribution is more likely iffunctionalitiesare options,new,and/orto addresses the top end of market.Naturally,there are other criteria for partitioning functionalities on ECUs than just the general preference on both architectures.

The choice impacts the need for Ethernet,the more distributed a network is,the more units need to communicate and to be connected.Thus,the more likely that Ethernet is a choice.In some cases,the availability of Ethernet might even become a prerequisite to separate functionalities,when this would result in communication with higher data rates that Ethernet supports.In other aspect, Ethernet simplifies handling a distributed architecture.It allows for scalability in the data rate by exchanging the PHY and supports various different option configurations over time and different car models in a cost optimized way.

As Torben Mielke presented,each sub-functions modeled in the logical architecture must be mapped to one ECU[3].Depending on the mapping of sub-functions to ECU,each information flow in the logical architecture mustbemapped to theappropriatecommunication protocol element.

3.2 From A Bus to A Switched Network

The traditional network technologies in-vehicle(including CAN,LIN,FlexRay and MOST)are all bus system,that leads to the available bandwidth is shared between all connected ECUs.Most often,a bus system has a line topology,with MOST beingexception,usingaring topology.It allows ECUs to communicate with each other,even communication is in one ring direction only.

In principle,a fundamental property of a bus is that all ECUs can listen to and thus receive all data that is available on the channel.There are different ways to decide whether a receiver actually does process data and the traditional in-vehicle networking technology does deploy a variety.There can be a message ID,from which a receiver decides whether data is of interest(CAN).They can follow a pre-defined schedule(LIN,FlexRay).There can be address unambiguously identifying the receiving ECUs(MOST).

Ifmore ECUs need to be included in the communication on the bus,it seems straight forward to simply attach a new ECU to the bus,as shown in Fig.2.Such a layer 1 connection requires that there is enough data rate available to support traffic for and from new ECU.Repeaters work on layer 1 and might be used to extend the range of a bus.In case of FlexRay,a star coupler is needed when the network increases such that the propagation delays of the signals become too large.Additionally,on layer 2,it needs to be ensured that new ECU is included in the channel access scheme, which might require reprogramming of all ECUs that share the same bus.In case of MOST,the order of ECU needs to be observed.

Fig.2 Differences between a bus and a switched network

Traditionally,the gateway function between two buses of same technology is performed on the network of transport layers.However,because of timing requirement and availability of star couplers,gateway between two FlexRay is not so common.For MOST,it is also unlikely.Communication from one ECU bus to another is also not that obvious,as this would require sharing message IDs.Extending an existing CAN bus would be easier.If this is not possible because of bandwidth limitation,the easiest way for a gateway to handle this is to simply pass a packet“as is”from one bus to the other.In general,gateway requires additional effort,which increases with the time criticality ofapplications.This lead to a kind of architecture,named Domain,with the idea that most communication happens within one domain.However,the data rate provided by a certain bus might not be sufficient.Additionally,it is not always possible to avoid crossdomain traffic.

With automotive Ethernet,first,each connection is Point-To-Point connection，which means that always ever only two ECUs are attached to the same link segment.The network happens on layer 2,on which switches pass data on,depending on address(in Fig.2,the switch is marked with“X”).This is key issue but completely new in architecture design.It supports all kinds of topologies and no gateways are needed from Ethernet to Ethernet,even if various differentPHY and speed grades are used.Gateways are only needed between Ethernet and other buses.

Extending the network requires adding another port to a switch or exchanging a PHY with a two port switch.Because the links are always Point-To-Point,this architecture increases instead of decreases the overall capacity inside the network.Of course,the limit of available data rate also needs to be observed when extending an Ethernet-based network.It is better for the overall data rate to attach an ECU directly or close to the main communication partner and not at the other end of network.The right architecture should always be able to provide enough bandwidth.This in return gives more flexibility for network optimization.This potentially allows prioritizing network optimization criteria differently.

In an Ethernet network,it is better to avoid the use of broadcast and multicast.It always be small in comparison with an IT network,for example,the use of multiple unicast transmissions instead of one multicast transmission,is preferable.If done with care,this is not only less strenuous for network but also scales well in case system grows.

3.3 Throughput Optimization with Automotive Ethernet

Crucial for designing communication network is that all data arrive at the receiver ECU within required timeline.In order to ensure this,it is first necessary to know all communication paths between ECUs and the amount of data to be transmitted.Normally,this is one of outputs from EE architecture and communication matrix,then decides on the in-vehicle network bus to use,how many gateways to include,and where or how to connect ECUs to network in a way that it scales over option selection,car model and functional updates.The in-vehicle network thus has to allow for future growth in data rate requirement.

For designing architecture and communication network,tools and simulation are used,especially to determine how much of available data rate is used up.Deterministic approaches do exist,but often result in the establishment of maximum possible traffic demand or occurrence of bursts.Deterministic approaches can be deployed to determine the required buffer space in an ECU.

The simulation of bus physics of an Ethernet network has to take switched architecture into account.Between switches,there are Point-To-Point links only and the computation of traffic loads on those links is comparably straightforward.The critical design element is the load inside switches.The in-vehicle network design has to ensure that a certain switch in network does not become the bottleneck because of too small buffers.Depending on the tooling preferences within OEMs,the respective tools need to be extended accordingly including AVB.

4 Conclusions

The decision on the overall functionalities to provide with a car is first of all a marketing decision and independent of capabilities of in-vehicle networking technology used and available.As soon as decisions have to be made on how to enable the functionality and configuration,EE architecture becomes important.Aspects like flexibility,scalability or how to distribute functionalities are severely impacted by the properties ofnetwork.Ethernetapplication into architecture needs a new concept and approach,which lead to architecture to have new characteristics and changes from system development point of view.However,those are necessary because automotive Ethernetis becoming a mainstream in near future.

Acknowledgement

Authors would like to take this opportunity to thank our management leadership and colleagues.In the process of composing this paper,they give us many academic and constructive advices,and help us modify the paper.Additionally,they also gave us the opportunity to do our teaching practice.