This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

IWR1443: Radar mapping of a stationary surface

Part Number: IWR1443
Other Parts Discussed in Thread: IWRL1432, IWR6843

Tool/software:

From the following post, it appears a stationary radar such as the IWR1443 is capable of mapping the surface of a large stationary target such as a pile of grain or the ground. 
e2e.ti.com/.../iwr1443-surface-profile-with-mmwave-radar-sensor

Can you provide more information about this application?  Is there an application note you can point me to or, better yet, could we arrange a call with an application engineer that is knowledgeable about this application.

The most common application for radar appears to be detecting moving targets, such as object detection for autonomous vehicles, or a single distance to a stationary target, such as level in a tank.  I am interested in using a stationary radar to map distances to a large, potentially uneven, surface with enough precision to estimate volume of the target.  The desired output would be a constellation of point representing the surface of a large stationary target.  Is that possible with the IWR1443 using existing radar processing algorithms that would be available from TI?  Is there a better radar chip for this purpose?

  • Hi 

    Thanks for your query. Please allow us a couple of days to respond

    Regards

  • Hello,

    This application is called Level Sensing and also more widely know as "Tank Level Probing Radar". We have open source example code for 1443, but we suggest our newest 77GHz low power radar sensor IWRL1432 for the level sensing application.

    https://dev.ti.com/tirex/explore/node?a=1AslXXD__2.20.00.05&node=A__AGnBmdgJtg7AQ5g-DG0mZw__radar_toolbox__1AslXXD__2.20.00.05&placeholder=true

    https://www.ti.com/video/6338474763112

    For the idea of "mapping" the surface of a static object this can be done at shorter ranges. We have an example of surface classification between grass and not grass with the help of a machine learning model, all ran on the radar chip itself.

    https://dev.ti.com/tirex/explore/node?a=1AslXXD__2.20.00.05&node=A__AD4hNRpdJvmncumsh1bO9Q__radar_toolbox__1AslXXD__2.20.00.05&placeholder=true

    https://www.ti.com/video/6342338470112

    Best Regards,

    Pedrhom

  • This does not resolve my issue.

    I would like to know if it is possible to duplicate the results of a LIDAR which is constellation of points that are the 3D coordinates of the top surface of the pile.  The response from my E2E post does not give information about this application.  Tank level detection produces a single level which works for liquid but not a solid pile.  The surface classification is interesting and may be useful but not the application I am interested in.  Perhaps another radar chip would be better for the this application.

  • Hello,

    Apologies for the slow response time, I have been out of office for the last week. Tank level detection gives a range profile where you can look at a singular peak sure, but the data around the peak can give context. It is for that reason radar can and is used for level sensing with uneven surfaces such as a grain silo. The surface classification example is done with only the range profile. I assumed this is what you meant with "estimate volume of the target" with the target being "a large stationary target such as a pile of grain or the ground" but maybe I have misunderstood? Nevertheless if you want a pointcloud you can as well, although this approach may see some losses in range accuracy in exchange for an X Y and Doppler reading that a pointcloud gives.

    Best Regards,

    Pedrhom

  • Thanks for the reply.
    It sounds like the algorithm is the same as it is for objects.  Except the Doppler FFT is not needed.  And the angle estimation is used to find locations for other points on the surface that aren't necessarily peaks on the range FFT.  Does that sound correct? 
    Do I need a module with multiple antennas in the Y dimension as well as X to get an X-Y constellation?  Or perhaps physically rotating the radar at stepped angles?
    Are there any app notes or tutorials on this other than:
    https://dev.ti.com/tirex/explore/node?node=A__AFll7zqKFZCLfpXJLkw18g__RADAR-ACADEMY__GwxShWe__LATEST
    https://dev.ti.com/tirex/explore/node?node=A__AKOmExskMEWAq1-czyBa0w__RADAR-ACADEMY__GwxShWe__LATEST
    which are written for objects.

    Do you have a recommended radar module that would work best for this application?  Would IWR6843 work better than IWR1443?

    Thanks,
    Tim

  • Hello Tim,

    You are correct on all accounts. The range profile does not need more than 1RX and 1TX, but you will need multiple antennas in order to get XY, as well as XYZ Doppler. With 6843's 4RX and 3TX antennas, you will be able to do all of this. You can enable/disable antennas as needed for the application at hand.

    Please clarify for me, are you interested in actually knowing the shape of the uneven surface is, or were you inquiring about for your end goal? Essentially do you just want to know the height of the pile of grain/dirt/uneven surface at different points? What level of resolution and accuracy is of interest here? This will help me guide our discussion better.

    Best Regards,

    Pedrhom

  • Pedrhorn,

    We need a solution that can yield a relatively high degree of accuracy over a 30 m high by 30 m diameter pile.  Perhaps a constellation point resolution of 30 cm or less in both X and Y (with Z being the distance to the surface), without major movement of the radar, would be sufficient.  In many cases the avalanche characteristics of the material in the pile will not be consistently predictable.  And the shape may be changing regularly with material being removed or added with various means. 

    I know solutions like the one described in this paper exist where the radar traverses the pile:
    https://www.mdpi.com/1424-8220/21/3/757

    But I was hoping for a solution that could use the beam-forming capability of the radar to measure distances to the surface instead of large movement of the radar over the surface to gather distances.

    We might still be interested if a lesser range or a lower resolution is possible.  We would like to know what you think might be possible.

    Thanks,
    Tim

  • Hello Tim,

    Thanks for the additional information. What is shown in the paper is doable with 6843, although to what exact level of accuracy at your specified ranges I cannot confirm, but 30cm is plausible. Note that the paper is using three 1443 TI mmWave devices, if it is doable with 1443 then it will definitely be doable with 6843. In this case I would recommend the Long Range People Detection demo as your starting point. Within TI, we use the terminology of beam forming being on the receiver, and beam steering on the transmitter. We use beamforming on every demo, but only one demo at the moment uses both beam forming and beam steering and that is the Long Range People Detection. Now I know the demo name is very different than what you are trying to accomplish, but what you can do is make the radar very sensitive and disable clutter removal so that static points appear.

    Is the idea to use multiple radars as here as well? My concern is memory based if too few sensors are trying to chart/map too large of an area. That and potentially wanting too large of a radar cube.

    Best Regards,

    Pedrhom