The latest developments in radar technology for automotive

The latest developments in radar technology for automotive


[Music] I have the dubious honor of being the last talk for the conference so I want to take you all for hanging in there for the last for last talk be mine yeah so I look after radar at nxp it’s really the radar signal processing products nxb enjoys the leadership market position right now in in radar signal processing in automotive so we’re number one with something over 50% market share and that this is built really over the course of ten years or more of innovating with radar signal processing particularly paying attention to integration and performance for power because of the position that we have and it gives a special insight into how the market is evolving and the different ways that autonomous functions are rolling out over time compared to other kinds of safety functions and it also gives us sort of a unique view across all the different OEMs and Tier II ones that we work with day to day about the architectural impacts and the architectural evolution that’s taking place to accommodate the kinds of sensing and processing that’s required for these kinds of autonomous functions so let’s begin oh I need my so we’ll begin looking first at some major trends that drive vehicle sensing and processing architectures and there’s there’s – there’s new cars that’s when program functions end cap functions as well as different kinds of autonomous features we’ll talk more about that talk a bit about the role of radar in these trends and the sort of attach rates and the attach rate trend that that we’re seeing with radar the trends that we’re seeing in terms of size and power and cost across these different kinds of functions and talk a bit about how regards evolved and where it’s going next and one key area of course is resolution and talk about the challenges to increasing the role of radar we’re seeing a large increase in the attach rate of radar and the use of radar over time especially in a kind of autonomous functions that we’re talking about and talk about the future of radar Kimmy you’ve probably seen the picture like this quite a bit over the last couple days raise a hand if you’ve seen this picture before or looked at this kind of thing over the last couple days do right so it’s a it’s a pretty popular sort of a definition under different kinds of autonomous modes and it’s separated in two levels and you’ve got driver assistance with level one which is sort of the basic things that we’ve seen before leg lane departure warning and lane lane keeping it’s on level two partial automation we’re seeing a little more autonomous Punk functionality there with traffic jam assistance and more kind of intelligent cruise control autonomous emergency braking and then we get into level 3 in level 4 level threes that word conditional is important because the conditional word implies that the autonomous driving load is able to interrupt the driver and bring them back into the control loop so driver monitoring driver awareness is a very important part of deploying these kinds of level 3 functions level 4 is where the drivers sort of out of the control group and then the car is essentially driving itself in that autonomous driving mode and of course level 5 it’s you may not even have a driver interface the car is really all autonomous all the time it’s tempting for us in looking at a chart like this to think of this as a linear progression in fact it kind of looks like a road map right and on the x-axis you got time enough y-axis there is performance it just looks like this is rolling over time and some sort of neat orderly fashion well we’re humans it doesn’t happen that way actually this is all happening at once it’s all happening in time happening in a mishmash of ways and it’s happening so for example we grow rfqs request for quotes from car makers for starter production sop vehicles in 2021 that will have level 3 level 4 autonomous applications this features sort of comfort features or advanced safety features as part of the car and for example highway pilot or highway chauffeur or however the OEM brands that particular feature this is an autonomous driving mode where you’re on a highway and it’s kind of like cruise control on steroids right where it’ll do all the Delaine changing for you it’ll manage your speed and will take you to a particular point in a highway and say ok we’re here now here’s the off-ramp you can go that’s that’s a comfort feature that’s like a noonim that’s a level 3 autonomous function a level 4 a feature there’s something like a parking valet so you drive up to shopping mall or whatever it is you hop out of the car and the car will go and find the parking garage and park itself usually this means very constrained parking violence but in next-generation parking valet applications it will be unconstrained in terms of the kinds of parking garage that I can go into so there are level 3 level 4 Tongass features rolling out and not in premium vehicles these are new sort of premium vehicle functions that will be appearing in your neighborhood BMW Daimler Volkswagen dealer in 2021 but at the same time there are level 2 functions that are being driven vehicles so all of these great endcap function it’s like autonomous emergency braking for a collision warning that’s being driven into low-end model so the challenge for the OEM especially OEM is like for example Volkswagen or Toyota who have a wide range of vehicle models the challenge is to come up with scalable architectures where the low-end they have fairly straightforward end cap functions that they can deploy but they could scale the architecture up to these sorts of little level three level four autonomous comfort / advanced safety features but not so very clear market demarcation here so when you look at it for level 1 level 2 that very strong and cap drawing so if you’re familiar with n cap it’s do car assessment program it’s a lot of innovation the road map for those safety features as larger Europe is largely driven by euro end cap and there’s a specific road map that euro end cap drives in terms of these kinds of safety features it’s a voluntary program both OMS while volunteer they’re the car models to go through tests at and Capon if they need all of the various requirements that the points add up to a certain value then they get a 5 star rating 5 star rating an end cap is huge impact on car sales it’s a major market lever this has been driving horrendous volumes and ramps and radar modules because radar is the primary and most important sensor that’s being used for these kinds of safety functions often it’s camera very rarely it’s lighter across the board it’s radar so radars are primary sensor that’s been applied to these various cognitive active safety features especially for end cap and going to these level 3 level 4 radar is a foundational sensor level finds a little bit different so we’re seeing level 4 level 4 level 3 level 4 functions rolling out now in 2021 in sort of mainstream vehicles that you’ll be able to buy in a car lot 5 is sort of an entirely different kind of mobility model usually that aligns with new fleet Fleet Services or uber uber taxis or trains for example there’s one now so various kinds of different different mobility different fleet different vehicle sort of market approaches and that’s that’s level five but on a day to day basis you’ll be seeing level three level four it’s happening very fast so you have these two market trends you’ve got MCATs worse than that it that it no that’s not it what do they do [Music] you’ve got the internet loud band cap trend and you see that over the years we’ve gone from a bad bruise patrol lane departure warning lane departure warning with that was not that particularly successful as an end cap safety feature in terms of reducing fatalities or or or act various kinds of accidents autonomous emergency braking has had a huge impact and in and autonomous the emergency braking aiibi is one of those end cap features that makes a lot of use of radar that’s being driven into more and more car models and emergency steer assist is on the roadmap and there’s also endcap advanced feature so euro end cap has an end cap and category where the OEMs can get extra points by adding safety features beyond what the end cap roadmap calls for but there’s there’s specific sensor and processing architecture so for sort of the entry-level end cap you’ll have a forward-facing radar for a EB or ACC and you’ll have two rear-facing radar for various kinds of lateral assist functions like blind spot detection and things like that with Junction Assist which is coming out in 2020 that’s caused the deployment of forward-facing radar on the front corner so there’s an example where the end cap roadmap is really driving a proliferation of radar more and more around the vehicle and for endcap advanced features we’re seeing more rhythm more deployment of radar level 3 level 4 have been sort of been in prototype stage but like I said 2021 you’ll see level 3 level 4 comfort features safety features appearing in mainstream vehicles and we have this we have this contention though between the the architectural requirements for end cap versus the architectural requirements for level 3 level 4 tires [Music] the real so but let’s have a look at the top oh hi high-resolution radar for a minute but let’s look at the different kinds of sensor technologies that we have can you actually hear me still free your hand even hear me yeah okay cool guy even the guys in the back Cory even the ones closer to the Train can hear me so that’s a good sign so it’s easy also to kind of get a little bit competitive about these different kinds of sensor technology it’s not like you know like I was going to take over the world or radars going to take over the world or cameras going to take over the world because we improved way cameras where facts matter is is you see all these sensing technologies for a long time they’re all improving in their own special way they’re all improving and either coming down in costs coming up a resolution decreasing in size they’re all improving in different ways but Canada there’s none of these sensor technologies that are going to disappear any time soon there are some that are more prevalent but then others radar certainly is the most prevalent of these right now because of the very unique capabilities that radar has in terms of range and Doppler and being able to see around objects and so on but what are the areas that radar is really in need of some improvement and we’re getting a lot of pressure from the OEMs and the Tier one is with resolution and this really has to do with the level three level for autonomous driving features they want to be able to do very detailed long-range environmental modeling which is mapping of the surroundings of the car they want to be able to like you saw in the last video detects objects on the road and know if it’s if it’s an object that they need to worry about or not how high is the object on on the surface of the road and then want to use radar from these new kinds of high resolution tasks so there’s a real need for high-resolution radar and we’re very much in front of that and beginning to do a lot of innovation in that area but well one of the various needs for it well there’s classification so there’s often thinkable cameras for classification because we hear a lot about different kinds of algorithms deep learning approaches convolutional neural networks with cameras to do classification but actually we’re classifying objects with radar right right now even with radar the way it is we’re able to classify you’re able to do things with micro Doppler and various kinds of radar features in order to classify objects but that but that classification requirement is really increasing quite a bit so we want to do more and more classification and with radar we’re actually being asked to apply the same sorts of neural network and deep learning approaches with radars you see commonly use division one of the cool things you can do with radar is you actually see around objects so because of the reflective wavefront property of the radar that comes back to the receiver we can see objects like for example a car that’s behind a truck or a person that’s behind a truck we can because the radar reflects back underneath the vehicle so we can detect on and gather objects but that requires a lot more algorithmic processing higher end processing so it’s more processing requirements for the classification it’s an increase in processing ISM it’s an increase in processing requirements localized and tracking we’re doing now with radar but now we’re talking about many more objects being tracked and localized over much longer range mapping this picture is from one of our partners I’ll show you a video soon where the car enters a parking lot and gradually builds a map of the entire parking lot and those who are all the various parking spaces are I mean doing more and more of that kind of mapping and a term that you may hear a lot in the near future is imaging radar so they’re not really talking about the word imaging has a lot of baggage attached so we’re not talking about the driver being able to look at a view screen then seeing what the image that they get back from radar the stuff we’re talking about we’re talking about is a very detailed map of three three-dimensional objects in the environmental modeling in the grid mapping that’s going on and that’s imaging radar so it’s high Lucian radar being used for high-resolution modeling I mean environmental modeling and mapping around the vehicle so this is a really a video this is from one of our partners at smart Microsystems they’re based in Germany and I’m not gonna pronounce it well but pronounced strike Kosovo’s Berg and this is using our radar processor the s32 are 27 and or the race under ultras we go to college and and that XP transceiver em the mr3 they’re three so she’s one of our ultras and one of these Eagle transceivers and it’s got two px tune transmits and for RS so actually fairly fairly simplistic and what they’re doing if they’re using the eraser under ultra the r27 and the legal vm are three zero zero three for the data acquisition but they’re doing the processing offline you will see bit more as we round this corner we some cars parked along the curb and you’ll see that the radar is basically imaging it’s creating this map the map of the vehicles that are parked along the curb and where the spaces are ending it can find a drivable path the main objective of environmental modeling and all of this fusion processing is to find a drivable path that’s you’re trying to find a safe path around all of these objects so you need to know where the objects are you need to know which which direction they’re going you need to classify them because if you know what the objects are you can do some predictive algorithms as to what they’re going to do next so that’s why you need to classify local law to track them and map them it’s all about finding the drivable path that’s the key thing for the autonomous vehicle so this is using fairly low resolution radar just with more sophisticated super resolution algorithms applied to the the raw data now imagine if you could do this with something like 20 times the resolution so now rather than for receive challenge they’ve got sixteen receive channels rather than to transmit you like like twelve twelve transmit but the radars you can also use my mode and I don’t know if you have any sort of experience with the wireless infrastructure and mongol but it’s multiple in multiple out and it you basically can multiply the virtual channels you get by by combining a transfer to receive so if you have for example for receive and to transmit you have for physical receive channels with actually with mine well you’ve got eight virtual channels for beams with sixteen are x16 receiving twelve TX food through the math I don’t think I can right now in my head but it’s a channel right it’s a lot more so it you get a big increase in a virtual challenge you also get much more control in more sophisticated wireless techniques for example with being being steering where you can really control through the TX and through some kind of digital way how you manage the beam the width of the beam and in beam forming a variety of beam forming techniques so you can really narrow in on a particular beam in a particular object that you want to focus on so more channels you have the more power the more resolution you can obtain being steering beam forming my movies are really sophisticated wireless RF techniques that are old hat as far as the wireless infrastructure goes that are just really making your way into radar now into automotive radar so how do you it have you increased the resolution of radar well I’ve already talked about that a little bit envision it’s a little more straightforward you just add more pixels right you just keep going more and more pixels per image sensor and you increase the resolution there’s algorithmic stuff really cool radar it’s a little more complicated because it’s an active sir you’re you’re you’re transmitting what’s called a Turk and and you’re receiving and you have a lot of control of the waveform you have a lot of control of the modulation that you use you also have Doppler so there’s angular resolution the more receive channel so more beams that use you split up the field of view with the more angular resolution you have but there’s Doppler resolution and there’s elevation resolution which is important for finding out how tall that object is on the road so the what we call the radar data the result of all of the FFTs that we do multiplies tremendously so there’s a big increase in the amount of data and the amount of processing and the sophistication of the processing that’s going on so it’s sort of an exponential this is just looking at the FFT performance requirements and you see it’s it’s virtually an exponential curve in terms of the processing requirements to drive high-resolution imaging radar in these applications and then you’re really certain trying to achieve all these it our thoughts about pixel resolution not just about the kind of spatial resolution that you you think about with vision it’s a it’s a it gives you a range and the the depth resolution that you can achieve varies quite a bit in in terms of how in how you use the transmit and receive and the way you use the frequency that you’ve been allocated so the processing performance is escalating quite a bit even using sort of conventional frequency modulated radar and that’s a that’s another thing I don’t really spend any time on in this talk but to think about the wireless infrastructure it was originally through base stations and and then the base stations went digital unless anyone was coding and we were CD then we wanted we went with CDMA so it it was coded radio basically and we got into spread in to spread spectrum and deep and the modulation of CDMA radar over the next few years you’re going to see the same sort of transition with radar what’s going to go from really frequency modulation to more of a digitally modulated signal and that opens up a lot of resolution for radar beyond the things we talk about with frequency a modulated radar here which is what I’m talking about just radar as it is there’s a lot of legs left with radar in terms of resolution what the radar dubby it’s the way it’s modulated now but once you get into digitally modulated radar then the resolution really goes up and I think it’s very interesting the discussion between radar and lidar and things like that gets the the change is telling a little bit also you begin to look at radar as some sort of vehicle the vehicle should be used for communication at that point so let’s talk about multitasking right so there’s an interesting future for let’s call it – radar on the horizon anyway so back to the story right we’ve got this huge increase in performance requirements but we don’t get any breaks when it comes to power the RIC radar module is getting smaller and smaller the thermal requirements are getting tighter and tighter you have to do 10 times as much processing in about the same power please so it’s it’s a challenge but we’re all over that the way that nxp handles that is with a lot of innovation in terms of processing accelerators and getting the most out of the performance for power got to be flexible because to allow the OEMs to scale from basic endcap functions to more sophisticated level 3 level 4 comfort features and and inactive safety features they need to scale and use the same radar modules that same radar processing infrastructure the same software software we use across the entire platform going from low end to high end so if this is a typical injure tunity and and a roadmap is really built around this to address these problems with scalability and flexibility and and in performance / power so we talked about these two key trends we’ve got we have to drive cost competitive integrated radar solutions for endcap to allow endcap to propagate across many more so the low-end vehicle so everybody gets safe but but we also have to enable the level three level for autonomous features that are rolling out in 2021 in vehicles you can buy radar presents itself as a kind of ubiquitous kind of sensor that’s used in a wide range of safety functions and vehicles and the key focus for us and we’re really trying to lead the way is in terms of resolution and what we can get from existing radar technologies with resolution and and we’re we’re very active in in innovating in that area and working with the other leaders tier ones and carmaker isn’t helping to put that’s all thank you very much for your patience and hey for the last time [Applause] [Music]


Leave a Reply

Your email address will not be published. Required fields are marked *