"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.
Heh, when getting a GPS Disciplined Oscillator set up I was feeding the 10 MHz oscillator into an SDR to use as a reference clock and was calibrating against a local cell tower. 7 Hz beat/offset at 1700 MHz!
>> 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.
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?
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.
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.
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.
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.
Interference is a real problem with FMCW radars, either maliciously in the case of electronic warfare, or accidentally in the case you mentioned, with many radars in the same space using the same frequency band. Wifi and cell phones use time division or frequency division multiplexing techniques, but radars (at least current-gen) generally do not.
There are mitigation techniques like randomization of chirp frequencies, choosing different idle times between frames, and signal processing techniques to try to detect interference and filter it out. In the general case, FMCW techniques will always have interference problems.
This is one reason amongst many others that military radars do not use FMCW but instead coded pulse compression techniques.
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.
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.
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?
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
How do I know that each of the pulses I sent out are the same ones I receive in reflection, in terms of frequency (remember, we have moving targets and therefore Doppler shift here), and amplitude? The reflections of my own pulses are pretty quiet compared to the pulses of, say, an oncoming vehicle's own radar transmissions.
How do I discern those individually-received pulses from the very similar pulses that are being received from other transmitting vehicles? (Sure, their timings should be be different, and the pulses may be very brief. But they may also overlap rather significantly, and errors can be very bad here. We need some rather good certainty here -- however that certainty is accomplished.)
Claude 3.5 Sonnet suggests the following for your questions : Distinguishing Radar Pulse Reflections
To answer these questions, we need to consider several key aspects of radar technology and signal processing. Let's break it down:
Identifying your own pulses:
The main ways to distinguish your own reflected pulses from other sources are:
a) Frequency: While there is Doppler shift due to moving targets, the shift is typically small compared to the carrier frequency. Modern radars use sophisticated signal processing to account for and measure this shift.
b) Pulse coding: Radars often use unique pulse patterns or modulation schemes. This allows the receiver to correlate incoming signals with the known transmitted pattern, effectively filtering out other sources.
c) Timing: The time delay between transmission and reception is used to calculate distance. Signals arriving at unexpected times can be filtered out.
d) Direction: Using directional antennas or phased arrays, the radar system knows where it's transmitting and expects returns from that direction.
Dealing with weak reflections and interference:
a) High sensitivity receivers: Modern radar systems use very sensitive receivers to detect weak return signals.
b) Signal processing: Advanced digital signal processing techniques can extract weak signals from noise and interfering sources.
c) Frequency diversity: Some systems use multiple frequencies to improve detection and reduce interference.
Discerning from other vehicle transmissions:
a) Frequency allocation: Automotive radars typically operate in specific frequency bands. Within these bands, individual systems may use slightly different frequencies.
b) Time diversity: The probability of pulse overlap decreases with shorter pulse durations. Many modern systems use very short pulses or continuous wave techniques.
c) Spatial filtering: Directional antennas help focus on specific areas, reducing interference from other directions.
d) Signal characteristics: Different radar systems may use distinct modulation schemes, pulse repetition frequencies, or other signal characteristics that can be used to differentiate them.
You know what, you're right. Next time let's just have a college-style lecture complete with latex formatting for the math equations in a casual comments section.
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.
This also assumes the radars are using as much bandwidth as Wi-Fi, and that the noise floor is the same, and that Wi-Fi-like modulation works fine for radar, etc.
Keep in mind this was an OOM argument. How much info does one really need to collect from the radar? 1-10kb/s?
Even with 10 cars trying to scan the same region, that's about 5 OOM of headroom compared to wifi.
Also consider what modern military radiation radarss can do. The F-35 can actively track 50 targets, all in one direction. And it can do that while potentially 100s of aircraft all are sending radar beams into the same space, with some even activelly trying to jam the F-35 radar.
Obviously, really old and cheap radars can have interference issues. But any such limitation is not due to the Physics or even engineering, but rather on the cost of a radar sophisticated enough to handle its environment.
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.
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.
> 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.
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.
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.
Chaff wouldn't be a fixed object: released from a moving car in response to a tailgater, the chaff would be moving at the same speed as that car, though it would presumably slow down rapidly in the air. Depending on how sensitive the tailgater's radar is, it might appear as if the followed car is braking rapidly.
However, I'm not sure it'd be effective: the whole problem with tailgaters is that they follow too closely, so there might not be enough space for the chaff to decelerate and trigger the tailgater's radar. It would be an interesting test, though. But I think an active radar emitter that detects the tailgater's radar and generates a matching signal that appears like a rapidly-braking car would be better, though obviously more technically difficult.
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.
Non-directional doppler radars are cheaper than the kind of directed radars that would be needed to map the distance to all static objects.
Also a dopler radar can scan the whole sector with every emission, and simply measure the distance to the closest moving objects (and their speed) with each scan.
A radar that creates a 2d image of the sector usually needs to send a targeted beam in one direction at a time. Such a radar is not only much more expensive, it also adds latency.
Visible light is better suited for 2D imaging, anyway.
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.
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.
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.
For automotive radar, I'm not sure the requirement is to determine the true speed of the detected object. The requirement is to determine if the radar vehicle is likely to collide with the detected object.
If the detected object is moving tangentially with the radar, relative velocity would be zero, because it's staying a fixed distance from the vehicle while both are in motion, and would seem to not be in the path of a collision.
"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.
Same like how you get to work with single-digit-hertz beats when tuning one string to match another.
Heh, when getting a GPS Disciplined Oscillator set up I was feeding the 10 MHz oscillator into an SDR to use as a reference clock and was calibrating against a local cell tower. 7 Hz beat/offset at 1700 MHz!
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.
Author here: Thanks for pointing this out. You are right, I will edit the article to fix.
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?
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.
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.
How is it continuous wave if it has a non-unity duty cycle?
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.
> 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.
Interference is a real problem with FMCW radars, either maliciously in the case of electronic warfare, or accidentally in the case you mentioned, with many radars in the same space using the same frequency band. Wifi and cell phones use time division or frequency division multiplexing techniques, but radars (at least current-gen) generally do not.
There are mitigation techniques like randomization of chirp frequencies, choosing different idle times between frames, and signal processing techniques to try to detect interference and filter it out. In the general case, FMCW techniques will always have interference problems.
This is one reason amongst many others that military radars do not use FMCW but instead coded pulse compression techniques.
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.
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.
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?
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
How do I know that each of the pulses I sent out are the same ones I receive in reflection, in terms of frequency (remember, we have moving targets and therefore Doppler shift here), and amplitude? The reflections of my own pulses are pretty quiet compared to the pulses of, say, an oncoming vehicle's own radar transmissions.
How do I discern those individually-received pulses from the very similar pulses that are being received from other transmitting vehicles? (Sure, their timings should be be different, and the pulses may be very brief. But they may also overlap rather significantly, and errors can be very bad here. We need some rather good certainty here -- however that certainty is accomplished.)
Claude 3.5 Sonnet suggests the following for your questions : Distinguishing Radar Pulse Reflections
To answer these questions, we need to consider several key aspects of radar technology and signal processing. Let's break it down:
Identifying your own pulses: The main ways to distinguish your own reflected pulses from other sources are:
a) Frequency: While there is Doppler shift due to moving targets, the shift is typically small compared to the carrier frequency. Modern radars use sophisticated signal processing to account for and measure this shift.
b) Pulse coding: Radars often use unique pulse patterns or modulation schemes. This allows the receiver to correlate incoming signals with the known transmitted pattern, effectively filtering out other sources.
c) Timing: The time delay between transmission and reception is used to calculate distance. Signals arriving at unexpected times can be filtered out.
d) Direction: Using directional antennas or phased arrays, the radar system knows where it's transmitting and expects returns from that direction.
Dealing with weak reflections and interference: a) High sensitivity receivers: Modern radar systems use very sensitive receivers to detect weak return signals.
b) Signal processing: Advanced digital signal processing techniques can extract weak signals from noise and interfering sources.
c) Frequency diversity: Some systems use multiple frequencies to improve detection and reduce interference.
Discerning from other vehicle transmissions: a) Frequency allocation: Automotive radars typically operate in specific frequency bands. Within these bands, individual systems may use slightly different frequencies.
b) Time diversity: The probability of pulse overlap decreases with shorter pulse durations. Many modern systems use very short pulses or continuous wave techniques.
c) Spatial filtering: Directional antennas help focus on specific areas, reducing interference from other directions.
d) Signal characteristics: Different radar systems may use distinct modulation schemes, pulse repetition frequencies, or other signal characteristics that can be used to differentiate them.
Speaking about signal-to-noise ratio, this is an example of a low one.
Fuck that. Fuck everything about that.
Bad human.
If I wanted a hyper-confident response that is unfettered by such constructs as context and introspection from a bot I already know where to find one.
Sincerely,
Not a fucking bot.
You're welcome. I read over it before I posted and it's basically everything I would have written.
It is largely hand-waving nonsense. Many words, and much of them are bullshit.
If you had written it, I'd be willing to discuss exactly why I think that about this prose.
But you did not write it, and I have zero interest in conducting a third-party discussion with a bot.
You know what, you're right. Next time let's just have a college-style lecture complete with latex formatting for the math equations in a casual comments section.
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.
Did I just drop into a time warp and end up in a future where LLMs bicker tirelessly in support of bad analogies?
This also assumes the radars are using as much bandwidth as Wi-Fi, and that the noise floor is the same, and that Wi-Fi-like modulation works fine for radar, etc.
Keep in mind this was an OOM argument. How much info does one really need to collect from the radar? 1-10kb/s?
Even with 10 cars trying to scan the same region, that's about 5 OOM of headroom compared to wifi.
Also consider what modern military radiation radarss can do. The F-35 can actively track 50 targets, all in one direction. And it can do that while potentially 100s of aircraft all are sending radar beams into the same space, with some even activelly trying to jam the F-35 radar.
Obviously, really old and cheap radars can have interference issues. But any such limitation is not due to the Physics or even engineering, but rather on the cost of a radar sophisticated enough to handle its environment.
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.
Cars use a lot of space and don't pack very tightly. Most of the time.
Any idea what the range of influence on the noise floor is/will likely be?
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.
When I said "we don't worry about it," I meant the problem has been acceptably mitigated.
What that really means is that several talented people dedicated their entire careers to worrying about it!
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.
I really like the figures in this and the previous entry:
https://www.viksnewsletter.com/p/how-automotive-radar-uses-c...
If the author is here I would be curious to know your process and tools to generate the graphs and figures?
Author here: I primarily use Excalidraw. I wrote an article detailing how I go about writing my newsletter here.
https://www.viksnewsletter.com/p/how-i-write-an-engineering-...
Great walk through, thanks for sharing!
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.
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.
I have a Toyota and the owner's manual says the radar does not detect fixed objects, like parked cars.
Chaff wouldn't be a fixed object: released from a moving car in response to a tailgater, the chaff would be moving at the same speed as that car, though it would presumably slow down rapidly in the air. Depending on how sensitive the tailgater's radar is, it might appear as if the followed car is braking rapidly.
However, I'm not sure it'd be effective: the whole problem with tailgaters is that they follow too closely, so there might not be enough space for the chaff to decelerate and trigger the tailgater's radar. It would be an interesting test, though. But I think an active radar emitter that detects the tailgater's radar and generates a matching signal that appears like a rapidly-braking car would be better, though obviously more technically difficult.
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.
Non-directional doppler radars are cheaper than the kind of directed radars that would be needed to map the distance to all static objects.
Also a dopler radar can scan the whole sector with every emission, and simply measure the distance to the closest moving objects (and their speed) with each scan.
A radar that creates a 2d image of the sector usually needs to send a targeted beam in one direction at a time. Such a radar is not only much more expensive, it also adds latency.
Visible light is better suited for 2D imaging, anyway.
I'm pretty sure that's the case because it won't detect things going slower than 5 mph.
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.
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.
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.
Fascinating.
I guess it my be better to stick with old-fashioned caltrops instead of chaff.
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/...
Fortunately, we have excellent devices for tricking visual systems, like the one rendering this text.
It might be useful to solve the problem of improperly dimmed headlights of other cars interfering with proximity detection.
cf. https://news.ycombinator.com/item?id=39244889
The radar cross section of an automobile within 100 meters is probably too high unless it's made of celery or has blackbody paint.
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.
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
Sure but there's no doppler effect if the target moves tangentially to your radar.
Author here: You can still use angle of arrival estimates from the MIMO antenna system to track the object. https://www.viksnewsletter.com/p/how-radar-measures-angle?r=...
For automotive radar, I'm not sure the requirement is to determine the true speed of the detected object. The requirement is to determine if the radar vehicle is likely to collide with the detected object.
If the detected object is moving tangentially with the radar, relative velocity would be zero, because it's staying a fixed distance from the vehicle while both are in motion, and would seem to not be in the path of a collision.