5G Network Architecture by Andy Sutton (IET 2018 Turing)

okay thank you for the kind introduction and my thanks to the RF and microwave technology network for the kind invitation to be with you today hello my role within BT is to look at the end-to-end 5g network architecture so that's really the presentation I'll be giving today what I'd like to do is touch on a lot of the kind of network aspects we've heard a lot today about services products some of the capabilities we could see in 5g networks but actually what does that mean for the network so let's start with a quick review of the IMT 2020 requirements the official specifications from the ITU that we have to address with with 5g technology and then let's explore the 5g architecture as standardized with in 3gpp now I say a standardized of course some of this work is still ongoing by June of the here we should have a fully standardized 5g architecture with the next generation core assuming next-generation core is concluded within release 15 we'll talk a little bit then about the functional decomposition of the round we've heard about sea run today but sea ran means many different things to many people from cloud round coordinated round it's a cooperative run and actually the functional decomposition of that round will determine the level of coordination and capability we can deliver so let's explore that a little and then let's going to develop that into a 5g network architecture and to end and then we'll look at how we optimize that architecture for products and services in particular for low latency services I'd then like to highlight a little bit of the work we're doing within BT and EE on our trials programs and then just summarize briefly I will talk about a 5g network architecture but it's probably worth finding a little bit of context first we're intending to launch what's known as an option 3x network so that will be a network which is effectively based on an evolved packet core and EPC plus I talked in last year's presentation here which is on i-80 TV about the various architecture options so 3x is our short-term target will then evolve that network to a 7 X network so it'll look like the architecture I described today but it will involve release 15 LTE into working with the next-generation core network and of course the full 5g architecture an option to network it's effectively simply an option 7x after reforming of spectrum from Elte to new radio so that 7x network is going to work will be for quite some time so you can find an overview of this in my presentation from last year so we've seen this particular slide a few times today but it's worth to spend in a few moments on it because we are trying to address a very broad range of capabilities from enhanced mobile broadband to ultra reliable and low latency communication and massive machine-type communication now most of our focus to date is on enhanced mobile broadband that's a bit of go-to-market proposition we're working a lot with verticals and with our colleagues in research on ultra reliable and low latency communication and we're also working on machine-type communication the Internet of Things but primarily a lot of that has been addressed today through the evolution of 4G with the introduction of narrowband IOT and EMT C technologies the MMT C aspect the massive machine type communication aspect will be standardized in release 16 and evolve as a second phase effectively of 5g now interestingly in terms of data rates at the top it talks about gigabytes in a second so gigabytes plural 8 bits in a byte so in excess of 16 gigabits per second and we'll see what that means for the ITU spec in a moment lots of heavy data applications around 3d video Ultra High Definition augmented reality virtual reality and we've just heard a lot about some of those examples so these kind of services will will drive huge data rates onto networks and huge levels of concurrency on networks as well so in terms of a few selected parameters from the spec from the IDE to you initially we're talking about targeting data rates of 20 gigabits per second in the downlink and 10 gigabits per second in the uplink now those were the initial specifications obviously we're not going to see those on networks from day one certainly not with a three and a half gigahertz spectrum we'll see those when we introduce the higher frequency bands with much broader channels but this is really the kind of target so we think about a similar target in the 4G world we set ourselves a target of one gigabit per second we standardized 4G back in 2008 we're just at the point now what we're starting to see networks around the world including the EU network in the UK actually demonstrating and launching that one gigabit per second peak capability ten gigabits per second in the uplink is quite a challenge as well and when we think about uplink we need to think about different types of use of equipment now it's not all necessarily going to be a mobile phone or even a smart device that's wireless that could be devices connected into the network that have a wireless last mile but actually plugged into the mains and are quite high-power devices so if you look at the radio specifications you'll see that we can support the same high efficient high efficiency radio interface in both the downlink and uplink as well as some fallback options for the uplink as well in the release 15 specs so we are looking at a kind of fixed applications of this wireless technology as well as mobile applications some targets around spectral efficiency here both of which significant uplift on what we saw with 4G 30 bits per second per Hertz you know we really are doing pushing the boundaries here and 15 in the uplink as well so the 30mm spec we had for LTE and we're more than doubling the uplink requirements here from LTE as well what's particularly interesting is now we have this requirement for a typical use of experience and indication as to what it should be previously we've always talked about peak rates but now we're saying actually you know typical user experience is a target value for enhancement and use case should be a hundred megabits per second in the downlink and 50 megabits per second in the uplink now we've seen a number of Talk's today that have shown what we can do with 100 megabits per second and actually to deliver hundred megabits per second to a large number of users in a dense urban area means we need to be targeting some of those higher peak rates to lift the average area capacity density up so as a network architect whilst a peak rates are something to chase and something that we we aspire to actually the benefit of those peak rates is actually earlier capacity density for a larger number of users and then we've got some one-way user playing latency over the radio so we've got a target to four milliseconds for enhanced mobile broadband and the ITU specified one millisecond one way for ultra reliable and low latency communication however 3gpp have specified 0.5 so say the ITU specs are a minimum requirement and in a number of cases 3gpp will exceed this 3gpp is already talking about simulations that show data rates higher than these minimum requirements as well of course control plane latency is important so how long does it take to take a device from an idle state to actually transmitting packets in the network the base spec asks for 20 milliseconds although 3gpp is going to meet the the stricter requirement of actually targeting 10 milliseconds now that sounds quite generous when we talk about some of the other latency targets but of course this is a state transition it's not a single transmission in one direction it's a message interchange so there's going to be a number of messages backwards and forwards who need to optimize a number of messages that have to be interchanged for that state change to take place and obviously the location of the control plane function as well and then as we move forward to massive machine type communication you know 1 million devices per square kilometer so massive density of devices in the network and it be those kind of density drivers that move us from the NBI of T type technologies we have today into the requirements for 5g mm TC and the last one I've picked out and it is just a sample really is this minimum requirement for EMV being you are LLC mobility interruption time of 0 milliseconds which suggests a make before break mobility procedure now we had this in WCDMA of course for inter frequency we had to go hard under her intra frequency but then we actually moved to a different situation in 4G of course where we now do a break and make and it's typically up to 30 millisecond mobility interruption time on a 4G network so actually to do this it's going to really call upon the separation of control and user plane and actually it may well be your control plane is connected on a lower frequency band and you opportunistically connect the data plane to cells that you see off from the capacity so we think about networks in different way we heard from the hawawa presentation before about this kind of cellulous concept so straight into the network architecture now we've seen diagrams like this for quite some time mobile really from gsm or certainly from GPRS we started to see this kind of reference point representation of mobile networks what we've got along the bottom I think michiga talked about this earlier as well is the actual data plane itself so I use a plane along the bottom from the user equipment over the new radio interface into the radio access network itself and that's actually not as simple as it looks because that functional decomposition which we'll explore shortly out to the user plane function so effectively we've collapsed the SMP gateway from LTE PC into a single node and actually the practical realization of the EPC in most networks is a single node for SMP gateway albeit there's some interfaces to manage with s5 etc and then we go out on the n6 so the SGI as we know it today that's where the kind of smarts tend to be the GI LAN environment and then out to a data network and that data network could be the public internet it could be a private intranet it could of course be service platforms on the operators network so it could be a CDN for example or some other capability above that we have the control plane the next generation control plane broken down to subscriber management authentication functions and unified data management as well as a core access and mobility function a session management function now we've got the addition of the network slice selection function which actually wasn't in the original draft architecture so things have moved on quite a lot in the last year since I spoke here policy control and then an application function sitting outside of that now that application function could well be an IMS for example an IP Multimedia Subsystem to support a voice service but actually whilst this is a standard representation I don't think it's the way we view mobile networks going forward and the next diagram we've definitely seen a couple of times so far is this kind of service based architecture we're actually what we're saying is that we've got a whole bunch of software capability that we need to put on some kind of common compute platform now that common compute platform could be local I've drawn this here with a kind of a data bus but of course it could be fully distributed as well and we may want to put functions in different parts of the network to manage different capabilities we have the ability to do that we introduced two new nodes to make this work we've got the NEF the network exposure function and that network exposure function allows us to access shared data layer for data analytics for example we've also got a network repository function the NRF and data offers a whole range of services including a kind of discovery function where by software entities serving these control paint purposes you can identify other entities in the computer environment and can connect directly when they need to so we can be very slick in terms of how we do that communication I've included two slides here one which explains the terms and one which explains the interfaces as well so that's just really kind of back up I just wanted to highlight the fact they're in the pack so looking at the functional decomposition of the radio access network we've got our data plane now as it is we've got the ran broken down into two functional blocks a distributed unit and a central unit with a new interface that's been designated f1 connecting between the two and the f1 specifications are available in the December release of release 15 so what does that f1 interface look like well what's been standardized in 3gpp is a higher layer split so effectively if we look at the protocol architecture P DCP the packet data convergence protocol sits in the central unit whilst RLC radio link control along with mac and phy sit down in the distributed unit and it is worth to spend in a few moments to understand these terms that they are going to be the basis of all the 3gpp specs and if you've worked on cloud round before D you often means a digital unit which is actually the baseband unit which is now the central unit so the terminology is important just to clarify the context of the conversation we've got additional protocols sitting in the sea you as well we've got the new S dot protocol which has been added to manage quality of service and how we manage cross flows in 5g which is in addition to the protocol stack we had in 4G but going for this particular functional split as a higher layer split it means we don't need to run anything like Cyprian ISA pre over the wide area network there's still an option to do that we'll come on to that in a moment but we can run this on an IP transport network layer over a Carrier Ethernet interface so it looks an awful lot like mobile backhaul as we know it today and from a dimension in perspective and a performance perspective it is very very much the same because we've got non real-time functions back here in the sea you then we don't have particularly exacting timing requirements on that F 1 interface connection so what will actually determine the latency you allow on the F 1 interface will not be the requirements of the radio rather the requirements of the services that you overlay on top of the infrastructure I'm showing the F 2 interface here what I'm calling the F 2 interface and I'm suggesting in many cases certainly in European deployments I would anticipate that would be an on-site interface in the same way Cypriots today in most cases from a baseband unit in the cabinet on site up to the remote Radiohead's or active antennas in this particular case from the D you on site which could be externally mounted as well up to the active antennas or it could actually be passives we talk a lot about active antennas and massive MIMO in 5g and it brings huge benefits absolutely but not every deployment will necessarily need active antennas a massive MIMO certainly when we get down to 700 megahertz you know we're not necessarily gonna be deployed massive mind more rage and those kind of frequency bands so we've got a number of options that's how we build that the f2 interface term is not standardized in 3gpp it has been used by a number of industry fora and it has been used by a number of vendors and it kind of makes sense as a way of describing that interface we'll see if the terms adopted or not in time so the practical realization of this network then will be a little different of course we're going to have to build that IP or that Ethernet connectivity between the distributed units on the cell site and the central unit sitting deep in our network and we'll step through that architecture as to where those central units could sit over the next few slides in the middle of that you're going to have some kind of fiber or wireless backhaul you're going to have a number of layer 2 switches layer 3 reuters etc and I've inserted IPSec though now if we're building a simple a mbb Network with P DCP sitting on the Cu you don't necessarily need to use ESP you could use a null cipher you may choose to use ESP depending how you overlay in other services but of course the packet data convergence protocol manages the cipher in which is going to sit back here which may well be deep in your network in a trusted and secure environment it's very likely though because it's an Ethernet interface you're going to want to run authentication so you don't want to protect that interface anyway so therefore an IPSec functional entity B that hardware be it a virtualized appliance or actually just be it software running on a virtualized compute platform would provide a level of protection and security there so in terms of expectations on latency we've got a number of requirements from industry fora and standards and we've also got some use cases now I've talked to lots of people about these including automotive gaming health as well and this is a kind of general consensus of a virtual reality and augmented reality we're talking between 7 and 12 milliseconds yeah tactile internet certainly less than 10 milliseconds vehicle-to-vehicle yes less than 10 seconds and the point is that mission critical systems on vehicles won't rely on cellular connectivity I spoke to a chief scientist from a major automotive manufacturer a little while ago and he said I do not need 5g to build autonomous vehicles he said however I can make the user experience a lot better if I can have connected an autonomous vehicle because that can pre ant the need to do emergency braking and brake more gradually before an incident for example I could take other actions to avoid the incident completely so the term connected an autonomous vehicle CA V is very much a dialogue that we should be using when we talk about the future of cars and other vehicles then manufacturing robotics and safety systems could be as low as one millisecond but it could be out as far as 10 milliseconds so not many things on here are screaming at me saying Andy is a network architect you need one millisecond latency on your network everywhere so let's see what we can deliver on the network and how we can then scale that solution so a representative network just to talk you through the concepts of deploying the architecture and we model this kind of thing as you can imagine in in considerable detail so I'm showing an access network there with small cells and macro cells I'm showing access aggregation and core as three tiers of the network so in our case in BT that access site is a Tier one exchange the aggregation is what we call a metro sites and then the core is one of our regional hub sites and they connect out to the IP pier in point and out to external networks as appropriate so taking that kind of model and looking at latency figures now this is illustrative this is based on analysis of 8,000 sites on our network we're doing more and more work to increase the number of samples we have here which may change some of these figures we've increased it to 16,000 at the moment but just to give you a feel about 12,000 sites about 100 and about 10 at the different tiers of the network the one-way transport latency 0.6 1.2 and 4.2 milliseconds and again we're doing various things to optimize that and change that so this was illustrative in as much as it allowed us to start placing components in the network in the first phase but it's very much a work in progress what I then done is taken those figures that we've we've calculated for our network and added the 3gpp figures for the run for EMB B so for milliseconds each way and for URL LC 0.5 milliseconds each way to say actually if I want to serve about 10 milliseconds as a round time trip for enhance mobile broadband then I could do that an aggregation site that assumes of course the content you're serving is in the aggregation site if you're going to the internet for that content then you may as well put your core network back here so you really didn't have a content distributed and we'll explore that in a moment with a look at multi access edge compute capability if I want to serve a low latency use case but actually this is looking pretty good because in my mind low latency is sub 5 milliseconds seems a reasonable figure ultra-low latency could be sub 1 millisecond or 2 milliseconds we don't really have agreed industry definitions but effectively we can serve network based low latency services and that fundamentally changes the economics everybody by having common infrastructure in the network and not having to deploy compute into every single end device it fundamentally changes the way we build networks cause it puts some pressure on the network from a reliability perspective but we're building lots more reliability internet works both through the standards and through the way we're architect in our network now to support the emergency services network so let's start to overlay some platforms on to that sample conceptual network that we're going to build access aggregation core the functions in red our 5g specific the IP set kind of comes with it but it's not 5g specific so I've not highlighted in red though the user unified data management sitting inverse and evolution the HSS the subscriber management services that'll sit fairly deep in your network the control plane function could sit back there initially as well the user plane function for enhanced mobile broadband we'll probably sit there because in many cases the subscribers going to want to go out to the internet anyway to the IPPR in point so we could sit deep in the network for that first use case however were increasingly serving content from on net caches and over fifty percent are the fixed broadband content we serve on the bt fixed network comes from on net caches come from that CDN capability so we're working with a lot of major content partners to bring content on net you can see the 21c the heart of our network our PE Reuters and P Reuters running through the core of that synchronization at the bottom so this is the addition of PR TC s so primary reference timing clocks effectively to deliver phase synchronization not just frequency because we're talking very much about TDD radio interface so the need for facing kin there and you've got the IPSec at the top so this would give you an enhanced mobile broadband network capability but let's build that out we could of course push user playing functions and mech out to the aggregation so mechas multi-access edge compute was previously mobile edge compute but in etsy a little while ago we change the term for multi access so it's available for fixed and mobile it's an important part of our convergent story Michigan mentioned earlier some of the FMC work we're doing with kings and other partners as well so by doing this now with MEC here we could have content we could have computers a service we could be hosting net apps for example and having the UPF that means we've got direct connectivity into that multi-access edge compute that could be built on common NF VI to common compute platforms have software instances of course we could then push that further forward in the network to offer even lower latency to handle greater scalability certainly if we listen to the kind of numbers that we heard this morning about number of small cells that could be required in networks then we may want to distribute more and more functionality by putting IPSec at the edge you remove the IPSec overheads from the the core of the network of course by serving from the edge you remove from the core transmission costs as well so there's a whole bunch of less obvious benefits to distributing more capability in your network so we've now got a more distributed network so I could comfortably do a few milliseconds round time trip from this point which would meet many many low latency requirements and I could do that from a network that is robust that is resilient and it's highly available but what if you do want to do one millisecond or though there abouts well actually we could look at distribution even further in the network we could distribute control plane functions as necessary we can access locally or we push right the way out to the very edge of the network so here I'm showing a five GDU with its Cu component its user plane function and it's multi access edge compute all on the cell site these distributed units themselves are effectively compute platforms it's really software so actually we can put more and more capability onto those platforms now that could be a standard macro cell or alternatively it could be a bespoke solution that we build in a factory in a hospital in a manufacturing plant for example so there's lots of ways we can actually customize that as a special projects application to meet specific demand and we're engaging with a wide range of verticals at the moment to understand those opportunities and how we bring propositions to market we talk to small cells very briefly certainly miss Gibbs presentation earlier mentioned the need for huge numbers of small cells to really lift area capacity density we agree that small cells an integral part of 5g we're doing a couple of small cell trials at the moment so that 4G focused but five-year be big 5g will put an even bigger drive and demand on that fire earth small cell program so in terms of connecting these things in we're looking at a basic connectivity solution which is fiber based I'm showing an example here of a fiber ring and I've cut the fiber Spurs the dashed line though suggesting some kind of wireless backhaul as well we're particularly interested in what's happening in the higher millimeter wave bands as well particularly the working at C is G on millimeter where transmission looking at Deeb and so 140 to 170 gigahertz some very very high capacity short-range links though we've simulated 50 gigabits per second over a few hundred meters so that could be plenty for the gun thing we're trying to achieve in terms of wireless and fiber distribution for the next phase where we're going to have a hybrid of that fiber where we have it wireless you know where we need to effectively but then what happens when we really do go for ultra dense networks with 26 gigahertz well 26 gigahertz we've heard a number of references today to self backhaul in I am a believer in that I think it's gonna be a very efficient way of dropping a sighted there's a few ways it could happen there's three point two five gigahertz of spectrum in that twenty six gig band as we call it so your operators may have enough spectrum to allocate some for back or some for radio access or actually with the spatial multiplexing benefits of massive MIMO potentially there's a reused probability as well for radio access possibly not over the whole 360 degrees of that cell or whatever angle it's covering dependent on the antenna array but potentially some areas whilst directing backhaul in another area as well the backhaul will be line of sight you can probably use a higher-order modulation scheme with less coding and you would do for the radio access environment as well this is the number of things to consider and effectively that means we then deliver a very very dense network with simple connectivity now of course what that does mean is anything you're doing on wireless or fiber based backhaul will have to scale to support that massive area capacity density so whilst today I'm busy rolling out 10 gigabit per second circuits for back hall into the evolved for G network in the future whilst 10 gig may be sufficient for a 5g macro with the option to split that we saw before P DCP RL C between the CEO and D you these small cells are going to need for 25 gig backhaul which means the fiber connection points are probably gonna need a hundred gig so a 100 gig which is currently a core network technology needs to very quickly become a metro and then an access technology of course the volumetrics with that will drive down price but we desperately need to see innovation in that kind of fixed environment as well to bring the fiber cost down and there's lots of work went on a number of universities I think we might hear a little bit more about that this afternoon as well so in terms of what we've been doing just to start to sum up we've run a number of trials we're working with many partners we're working with huar we were working with Nokia we're working with Ericsson we're working with a wide number of universities from Surrey to King's to Cambridge to Bristol to the University of Salford we're doing work on microwave and millimeter wave transmission and what I'd like to do is to show a video in a moment that shows the actual 2.8 gig end-to-end activity we finished shortly we've also been doing some massive mimal testing our labs at astral park our research teams are really starting to understand the detail of spatial multiplexing and massive MIMO what that means for us and also we recently did a demonstration of okay uplink and downlink decoupling to show how we can support the wider coverage from the 3.5 gig downlink on the uplink as well so some interesting developments in that space so I'll just find this video for a moment it's only about one minute 40 seconds but it gives you a feel for some of the experimentation and development activities happening in our company hi my name's Tom Lennox I'm the director for technology services here at either what we're delighted to demonstrate here today is an end-to-end 5g network for Genie radio with massive MIMO on the antenna array a Phi G packet core which is virtualized all in partnership with its hawawa here at ETS out one of the great things with this end-to-end demonstration is that we're using commercial off-the-shelf hardware in our packet core so it's a while waste software on HP hardware [Music] [Music] we're consistently showing 2.8 gigabits per second end-to-end through our RAM and cord for us that's a great step forward in utilizing and showcasing the promise of 4G okay so in that particular demonstration we use one hundred megahertz of spectrum we used a seven to one downlink to uplink ratio on that TDD configuration 60/40 64 R for the base station and the ya tau e was 48 R as a configuration sitting behind what you've just seen there so to summarize 5g will address a broad number of use cases from enhanced mobile broadband to ultra reliable and low latency communication and I should also if I break it out on the bottom II I always try and make this point ultra reliable low latency is one grouping but of course some services may require ultra reliability but not low latency and others may require low latency but not reliability so there's actually three subsets of service type that you would need to address from that and of course we mentioned as well the evolution of massive machine type communication 5g requires a new architecture this is a chance to really embrace the world of network functions virtualization the world of software ization through software-defined networks and flexible network architectures I agree with mr. Kiir phi g has to be the last g because number of reasons one commercially it's not viable to keep them for lift truck upgrades every 10 years on networks of course will continue to evolve this technology but this should deliver everything we need to get people to the point where actually upgrading to the next G gives them nothing more you know we've continued to manage that service and manage that experience so 5g is fundamental in shaping the future we talked about next-generation core network and how it can be grouped into a control plane function and the user plain function some ran functions are going to be pulled in towards the core while some core functions are pulled out towards a run now that introduces a new kind of node towards the edge of your network and that new node will be in a fixed location hence the need for software ization n FBI yeah we'll need to push that capability right to the edge of the network in some places and have it in different locations for the services so that next generation control play based on service based architecture will allow a device to connect to both a locally hosted service and a more centralized hosted service so it may well be an enhanced mobile broadband internet browsing session along with some kind of critical communication activity running in parallel small cells an essential component of 5g you are LLC is an overlay on the existing infrastructure so we build the network and then we pop that on top of it and the last point there the initial machine type communication use cases will be addressed by narrowband IOT and that's been rolled out now great advantage of narrowband IOT is you get about 20 DB improvement on the link budget because the radio characteristics gives you thank you very much the network has been designed to support one millisecond of there abouts with a 0.5 millisecond specification on the URL LC it could be but those two users will be served from a very local base station and very local content you're not going to be going back out to the Internet and back out again so you can build local networks that can be Network assist of v2x for example could work with very local capability that capability could be compute capability for example or it could also be device to device and device the device ID not touched on today but that would be a vital part as well of what we build in 5g building on the work on proximity services etc in in 4G today thank you I'm Jesse from Romagna Hermitage there's very good presentation thank you thank you it all sounds very expensive so are there any thoughts on how the business models might need to change to support investment yep the key thing is we need to be focusing on new use cases so we need to be engaging with the verticals and generating revenue sources from non-traditional groups I guess for cellular I think from the conversations we've we've we've had today I've talked a few people in the room and the presentations we've seen so far those verticals are there actually so I don't think it's so much the operators trying to push the technology there was a level of pull so we need to find the opportunity to monetize those verticals but absolutely it is a significant investment we're making 60 40 60 for our antennas are incredibly expensive but actually they're not necessarily needed everywhere there will be evolutions in the technology that brings the cost down and there will be more variance between an 88 R and a 60 40 60 for our solution but also the addition of millimeter wave as well means we can actually deliver very high capacity and small form factors so the types of site we build to lift area capacity density could change as well so there's a number of things we're addressing really from both an infrastructure perspective in terms of trying to drive cost down and deliver capacity in a more efficient way and certainly in a cost per bit analysis 5g would be a lot cheaper than 4G s 4G is a lot cheaper than 3G but actually we do need to monetize the new services as well absolutely I'm around all day yes I need to dash off now for a quick interview on IET TV but after that I'm around yes [Music] [Applause]

Channel: Martín Piñeiro

Views: 96,335

Rating: 4.8400002 out of 5

Keywords: 5G

Id: aGEAQJ7U1tA

Length: 37min 56sec (2276 seconds)

Published: Fri Feb 09 2018