21

(Meta: I do not know an appropriate place for this on Stack Exchange. There does not appear to be any groups related to autonomous driving technology and computer 3D vision / perception systems.)

For self driving vehicles using 3D depth perception LIDAR on a highway with hundreds of other vehicles also using various other LIDAR sweep beam or spot-field (kinect style) emission scanners, how is it able to distinguish its own signal returns, from the scanning being done by the other systems?

For an extremely large multilane highway, or complex multi-way intersections such emissions can be seen in all directions, covering all surfaces, and there is no way to avoid detecting the beam emissions from other scanners.

This seems to be the main technical hurdle for implementing LIDAR for autonomous driving vehicles. It does not matter if it works perfectly if it’s the only vehicle on the road using LIDAR.

The real question is how it deals with being inundated with spurious signals from similar systems in a future scenario where LIDAR is present on every vehicle, potentially with multiple scanners per vehicle and scanning in all directions around each vehicle.

Is it capable of functioning normally, can it somehow distinguish its own scanning and reject others, or in the worst case can it fail completely and just report garbage data that is useless, and it doesn’t know that it’s reporting garbage data?

This at least seems to be a strong case for having passive 3D computer vision that’s just based on natural light and stereo camera depth integration, as is done in the human brain.

Dale Mahalko
  • 1,570
  • 2
  • 13
  • 15
  • 1
    Or you know...the eye hazard. As for distinguishing your own return signal from others, you could do some modulation or auto-correlation. I'm not sure how compatible that is with time-of-flight schemes but it would increase processing in something that already needs to differentiate extremely small time differences. – DKNguyen Aug 05 '19 at 19:16
  • How do cellphones all talk on the same frequency? – Voltage Spike Aug 05 '19 at 19:24
  • @VoltageSpike This is going to rapidly turn into a extended discussion, but until we have optical antennas, both lasers and photodiodes much more limited in terms of wavelength flexibility, phase information, unlike a cell phone antenna so there are a lot less tricks you can play with the signal. I think you might only be able to mess around with amplitude modulation and not much else. No frequency or phase modulation of any sort (I think). – DKNguyen Aug 05 '19 at 19:32
  • 3
    @DKNguyen, you can modulate your laser with an RF subcarrier, and do any kind of RF or phase modulation you like on that subcarrier. – The Photon Aug 05 '19 at 20:27
  • 3
    @VoltageSpike Well, TDMA (Time Division Multiple Access) is one way, but CDMA (Code Division Multiple Access) is far more sophisticated. Alas, I studied those things more than 20 years ago and I hadn't a refresher since then. I remember when I studied them I thought "TDMA or FDMA? Meh! Easy concepts.", whereas CDMA was mind-boggling! Essentially you mix each stream (from each user) with a different numeric orthogonal code sequence. Extraction of the right stream is done by math operations involving the specific code sequence. The rest of the streams appears just as background noise. – LorenzoDonati4Ukraine-OnStrike Aug 05 '19 at 21:25
  • 1
    @ThePhoton When you say "phase modulation" are you actually referring to modulating the phase of the photon/electromagnetic wave as would normally be meant if we were talking about radio? Or are you referring to modulating the phase of the amplitude variations of (brightness) the laser? Because my understanding is that as long as lasers are not being produced by optical antenna, we have little control over the actual phase of individual photons. – DKNguyen Aug 05 '19 at 21:26
  • @DKNguyen, maybe if the LIDAR signal is made by coded pulse streams, some sort of CDMA detection is possible. Just random thoughts. – LorenzoDonati4Ukraine-OnStrike Aug 05 '19 at 21:27
  • @LorenzoDonati Yes pulse streams I would understand since that is just variations in brightness/amplitude. It's when people start talking about all the applying techniques from radio that require controlling the phase or frequency of light that I become wary. Since modulating the phase would imply that we have control over the coherency of the light, and modulating the frequency would imply we can have the laser change colour to a sufficiently large and predictable degree to be useful. As far as I know, we are still quite far from that due to the lack of optical antennas. – DKNguyen Aug 05 '19 at 21:29
  • 2
    @DKNguyen, I'm talking about modulating the subcarrier, AM, PM, FM, whatever. – The Photon Aug 05 '19 at 21:34
  • 1
    @DKNguyen Take a look at https://en.wikipedia.org/wiki/Electro-optic_modulator, the units I used were not near that fancy, but they would do AM and PM with a bandwidth in the 10's of Khz. – GB - AE7OO Aug 05 '19 at 22:25
  • @GB-AE7OO Interesting. Lots of weird optical devices out there. – DKNguyen Aug 05 '19 at 22:27
  • @DKNguyen As far the concept of a optical antenna goes, that might apply if we were talking about omni's, but we are not. A LIDAR system has the transmitter/receiver pair at the focus point of a reflector. Think about early RADAR systems. – GB - AE7OO Aug 05 '19 at 22:36
  • Cellphones generally do not talk on the same frequency if it can be avoided. There are a range of frequencies that can be used. https://en.wikipedia.org/wiki/GSM_frequency_bands Is there a way for a laser to similarly modulate its emission color across a spectrum range? – Dale Mahalko Aug 06 '19 at 01:55
  • When testing a long range low power radar, I had occasion to use a long fiber-optic delay line. We modulated the laser amplitude at GHz and demodulated it at the end of the line to simulate transmit delays up to a mile or so. So, it is possible to modulate laser light with many GHz of bandwidth. It may be much harder to do it in Lidar than it is in a delay line. But the basic idea is the same. – user57037 Aug 06 '19 at 03:09
  • @DaleMahalko "There does not appear to be any groups related to autonomous driving technology and computer 3D vision / perception systems." Just fyi there is of course [Robotics SE](https://robotics.stackexchange.com) with [56 questions tagged `lidar`](https://robotics.stackexchange.com/questions/tagged/lidar) including [one of mine](https://robotics.stackexchange.com/q/18014/17872) – uhoh Aug 06 '19 at 23:53
  • and [this answer](https://stats.meta.stackexchange.com/a/2795/163067) mentions [Signal Processing SE](https://dsp.stackexchange.com/) and there are [256 questions tagged `computer-vision`](https://cs.stackexchange.com/questions/tagged/computer-vision) in Computer Science SE, and also see [Where to ask questions about computer vision?](https://meta.stackexchange.com/q/284827/303080), but obviously this site works too. – uhoh Aug 06 '19 at 23:54

2 Answers2

28

Commercial LIDAR units modulate the light with a very long pseudo-random sequence.

The modulation is primarily to (1) have a modulation for determining distance and (2) to avoid interference with ambient sources of DC and AC light.

The long sequence makes it unlikely that any other source, even a modulated one like another LIDAR, will line up and interfere.

Bob Jacobsen
  • 1,860
  • 1
  • 10
  • 10
  • Exactly. The interference issues exist at least as significantly from ambient noise, so the "competing" noise is relatively small, by comparison. – paul garrett Aug 06 '19 at 21:08
  • Is there a reason you don't refer to it as FHSS? Is it not actually FHSS? – forest Aug 07 '19 at 15:48
  • The “FH” part doesn’t feel quite right to me. The base color of the laser doesn’t change, so the hopping is in time more than frequency. Yes, the modulation sequence means the laser is SS, but the side bands are really tiny. – Bob Jacobsen Aug 07 '19 at 15:59
10

Depends on the LIDAR. My experience with it is out of date (over 10 years), but I cannot imagine that the basics have changed that much.

Most will be using a form of lock in to discriminate their signals. They treat other LIDARs as noise just as they do anything that is not locked to their signal. While you don't have the same frequency agility that a radar does, you do have the ability to modulate your carrier using the many forms of modulation. They can definitely change modulation schema as required to find the least noise.

A modern DSP version of a Lock in Amplifier or the equivalent would be used at minimum.

After more "looking" around, it also looks like some research versions have the ability to combine transmitters into a single "beam"(wrong term I know, but easy to understand) and this would make it much harder to confuse and increase imaging definitional all at one go. Way beyond what I worked on.

GB - AE7OO
  • 1,043
  • 7
  • 19