[HN Gopher] How automotive radar measures the velocity of objects
___________________________________________________________________
How automotive radar measures the velocity of objects
Author : subbdue
Score : 53 points
Date : 2024-06-23 17:17 UTC (3 days ago)
(HTM) web link (www.viksnewsletter.com)
(TXT) w3m dump (www.viksnewsletter.com)
| Animats wrote:
| _" The IF signal has a phase that is the difference between
| transmit and received signal phases."_
|
| Yes. That's a neat property of superhetrodyning - phase is
| preserved. Both the outgoing and incoming signals are down-
| converted by mixing with the local oscillator. The phase angle
| difference between out and in is the same at both the
| transmitted/received frequency and the IF frequency. But down at
| the IF frequency, you get to work at a lower frequency where it's
| easier to do A/D conversion and counting. Most software defined
| radios still have a superhetrodyne front end, so the digital
| stuff is working at the IF frequency.
|
| This is less necessary than it used to be, now that digital
| circuits can work well into gigahertz ranges.
| kazinator wrote:
| Same like how you get to work with single-digit-hertz beats
| when tuning one string to match another.
| dvh wrote:
| What will happen if every single car has radar? Wouldn't they
| interfere? You're stuck in traffic and 300 nearby cars are
| blasting radars all over the place?
| psunavy03 wrote:
| There are various ways a radar (or any RF signal) can be
| designed to recognize its own signal from all the background
| noise. We don't worry about millions of cell phones or WiFi
| routers sharing bandwidth either.
| r2_pilot wrote:
| Actually, in both cases, we do. Cell towers deliberately have
| different frequencies allocated from their neighboring
| towers, and Wi-Fi has multiple channels, several of which do
| not have any overlap.
| psunavy03 wrote:
| When I said "we don't worry about it," I meant the problem
| has been acceptably mitigated.
| mitthrowaway2 wrote:
| What that really means is that several talented people
| dedicated their entire careers to worrying about it!
| ssl-3 wrote:
| Neither of those examples answers the question.
|
| Co-channel Wifi interference is real. It really puts a damper
| on range and throughout compared to how it used to be. It is
| a largely unmitigated clusterfuck, as is the way with CSMA/CA
| once density increases enough.
|
| LTE interference isn't an important thing in practice, in
| part because because all participating devices have very
| tightly-controlled timings. It isn't a clusterfuck at all
| because of the mitigations in place, but it does require
| centralized coordination to be this way.
|
| Radars on cars don't have centralized coordination (do
| they?). What mechanism prevents their performance from
| degrading as wifi does?
| r2_pilot wrote:
| For your specific question, you can send a pulse train
| specific to the radar emitter, and you filter out anything
| that doesn't match your specific pulse train
| trashtester wrote:
| Consider how many bits you transfer over wifi before
| hitting saturation.
|
| Now consider how many bits of information is collected by
| these radars per second.
|
| That gives and indication of how much free bandwith there
| is in the radar bands if the radars are built at the level
| of sophistication we expect from wifi. (At an OOM level, if
| not accurately).
|
| Any congestion with current technology would be because the
| technology is far less optimized and standardized than
| wifi.
| ssl-3 wrote:
| Did I just drop into a time warp and end up in a future
| where LLMs bicker tirelessly in support of bad analogies?
| mitthrowaway2 wrote:
| Nonetheless, even in the very best case, every other radar
| still increases the background noise floor that each radar
| has to distinguish its own signal above. It won't ruin the
| signal completely, but it will affect how much scan time or
| output power is needed, or the detection resolution it can
| attain.
| kube-system wrote:
| > What will happen if every single car has radar?
|
| If you're on the road in a relatively affluent area where
| people drive late model cars, this is pretty close to already
| the case. Automakers have started making these systems standard
| on many/all of their models in the US for several years now.
| Toyota, for example, started rolling out these systems a decade
| ago, and have been standard on all US models since 2018.
|
| I'm not sure what these systems use in practice for
| interference mitigations, but there's a bunch of stuff that
| could be done, for instance, hopping between different
| frequencies.
| curiousfab wrote:
| Typical FMCW radars transmit very short ramps (microseconds) at
| a very long (relatively) intervals (several ten milliseconds),
| i.e. a duty cycle of less than 0.1%.
|
| In order to create interference between two radars, the ramps
| have to overlap pretty exactly, within a few nanoseconds of
| each other. This is very unlikely to happen.
|
| Modern radars employ technologies to detect and/or avoid such
| collisions.
|
| Overall it is not really an issue, even with many radars in
| crowded spaces.
| abstrakraft wrote:
| This is true for some earlier lofi radars, but as driver
| assistance and self-driving have developed, so have the
| requirements and capabilities of the radar systems. Newer
| systems generally have shorter PRIs for higher doppler
| bandwidth, and much higher duty cycles for more energy on
| target - the FCC limits power, so you've got to get energy
| from the time axis. Both of these things make the
| interference problem harder.
| aidenn0 wrote:
| How is it continuous wave if it has a non-unity duty cycle?
| abstrakraft wrote:
| In practice, the difference between pulsed radar and
| continuous wave radar is a continuum rather than a
| dichotomy. Historically, FMCW (frequency modulated
| continuous wave) had a high duty cycle (though not 100%,
| the ramp generators need finite time to reset (though you
| can alternate between up and downramps and get closer)).
| For some applications, though, requirements force you to
| short ramps and long PRIs, thus low duty cycles, but the
| name (FMCW) sticks.
| kazinator wrote:
| I suspect radar like this needs only a tiny time slices to do
| its work. Say for example that it's only necessary to get
| updates about the moving object 100 times per second: every 10
| ms. The radar pulse durations necessary to do the job can
| probably be measured in microseconds, though. A 10 ms
| separation between pulses measured in microseconds is a large
| amount of empty space.
| teeray wrote:
| It would be interesting to see how to attenuate or deflect radar
| emissions from cars to passively disable automatic braking. Won't
| brake if the car believes it's driving on a flat expanse of
| nothing.
| jvanderbot wrote:
| The radar cross section of an automobile within 100 meters is
| probably too high unless it's made of celery or has blackbody
| paint.
| opwieurposiu wrote:
| If you had a chaff dispenser, perhaps you could deploy a chaff
| cloud that would look like a fixed object and trigger the
| emergency braking of the cars behind you.
|
| The newest weapon in the war against tailgating.
| kayodelycaon wrote:
| I have a Toyota and the owner's manual says the radar does
| not detect fixed objects, like parked cars.
| ssl-3 wrote:
| Fascinating.
|
| I guess it my be better to stick with old-fashioned
| caltrops instead of chaff.
| kube-system wrote:
| I've seen other brands say/do the same. I suspect these
| systems would have tons of false positives otherwise,
| because cars often drive in close proximity to fixed
| barriers with high radar reflectivity.
|
| I'm sure the best way to not run into other cars on the
| road is to detect the other objects that are also moving at
| non-zero velocities. They're likely to be other cars,
| rather than road signs, poles, etc.
| kayodelycaon wrote:
| I'm pretty sure that's the case because it won't detect
| things going slower than 5 mph.
| krisoft wrote:
| The radar detects them perfectly. Higher level logic
| filters stationary objects (using the ego-vehicle's known
| speed) out to prevent nuisance detections.
|
| If they wouldn't do that filtering they would trigger a
| brake for almost all manhole covers and overhead signs
| and such.
| kube-system wrote:
| Yeah, by 'detect' above I think we've been referring to
| the end behavior (e.g. what's in the owners manual) of
| the automated systems, not the raw RF received.
| Obviously, radar reflectivity itself is not dependent on
| relative motion of the object compared to the receiver.
| abstrakraft wrote:
| It's not a problem of reflectivity, it's a problem of
| resolution. In order to detect something distinctly from
| other things (i.e. resolve that thing), you must be able
| to distinguish its reflected energy from that of other
| things by separating them along one or more dimension.
| Range is usually a good discriminator, but there are many
| things at (nearly) equal range to the radar. Azimuth is
| typically not great, because azimuth resolution requires
| a physically wide aperture, and real estate on the bumper
| is expensive. Doppler is great for moving things because
| it's easy to design a waveform with a small doppler
| resolution, and most moving things (cars, bikes, people)
| don't move at exactly the same speed as other moving
| things. However, nonmoving things have a very consistent
| velocity of precisely 0, and there are lots of them. So
| they can be very hard to resolve, and thus to detect.
| MisterTea wrote:
| My CR-V uses a combination of camera and radar. So you'd have
| to fool both somehow.
|
| https://owners.honda.com/utility/download?path=/static/pdfs/...
| sandworm101 wrote:
| Some slight errors in terminology.
|
| >> The time between two consecutive chirps is called the pulse
| repetition rate (PRT), and plays a key role in the accuracy of
| doppler velocity estimation.
|
| This is actually known as the pulse-repetition _interval_ (PRI),
| or _time_ (PRT). A "rate" is describable by a frequency. An
| interval is described with a unit of time between repetitions.
| Radar signal characteristics are a rabbit hole of such
| definitions. They really do matter once one switches from
| theoretical discussion to actual math. Confuse a rate with a
| period and your math for calculating ambiguity zones will fall
| apart.
| roger_ wrote:
| Only skimmed this but didn't see any mention of tracking, which
| is the only way (?) to get the true velocity when there's non-
| radial motion.
| enchilada wrote:
| The article says it's using doppler effect.
|
| > The velocity of the target also manifests as a frequency
| shift in the received chirp due to Doppler effect
| KeplerBoy wrote:
| Sure but there's no doppler effect if the target moves
| tangentially to your radar.
| transpute wrote:
| Wi-Fi 7 doppler radar has been used by AI/NPU laptop to detect
| nearby humans, https://www.youtube.com/watch?v=3WFx-8agAq4
|
| _> With a PC with Intel Wi-Fi sensing capabilities in sleep
| mode, the PC Wake-on-Approach is activated as it detects human
| presence. Even when a user forgets to lock the PC, a count-down
| to lock starts with no human present. False detection is
| prevented even with human presence behind and next to the PC._
___________________________________________________________________
(page generated 2024-06-26 23:01 UTC)