• State TI Thinks Resolved
  • Locked Locked
  • Replies 5 replies
  • Answers 2 answers
  • Subscribers 31 subscribers
  • Views 231 views
  • Users 0 members are here
Support feedback
Options
Options
Related

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.

IWR6843AOPEVM: Intuition for accuracy and precision of range/speed

Morten Nissov
Intellectual 380 points
Part Number: IWR6843AOPEVM

I am curious regarding the statistics of a measurement from the radar, in particular the range and speed.

Taking range as the example, I know one can calibrate the range bias and calculate the range resolution given the chirp parameters. If we suppose a range measurement can be given by

measured_range = range + bias + noise

Calibration should ensure that the bias is relatively small. The question then is how we can describe the uncertainty (i.e. noise or precision). The range resolution tells us the minimum separation between two objects for them to be seen as independent peaks in the FFT.

What if there's a single object, can we apply a similar train of thought? Perhaps we can suppose that the uncertainty about this single peak is +- half the range resolution? Is this true?

Another thought I had was that the variance could be range-dependent, as the further an object, the greater the time offset between Tx and Rv chirp, resulting in a lower overlap and larger portion of the chirp to be thrown away. This seems unlikely: looking at the low range default from https://dev.ti.com/gallery/view/mmwave/mmWaveSensingEstimator/ver/2.0.0/ we have the following givens

  • max range 10m
  • ramp end time 49 us

the one way travel time for 10m is 0.03us. So even at the maximum there is hardly an effect.


Another parameter specified here is the number of range bins (256). Given that these bins are representing 0->max, one could perhaps say that the accuracy is locked to the bin size, in this case 10m over 256 bins -> 3.9cm which is somehow not the same as the "range interbin resolution" (4.883 cm)

In conclusion, I have some doubt on this. Likely in part due to incomplete understanding of the overall processing pipeline. Not to mention how velocity plays into this.

  • 0
    TI__Mastermind 37780 points

    Hi,

    I don't believe we have a model for the noise associated with our range measurements. In many of the experiments we run, we have moving people and objects, whose movement has much more uncertainty than the radar measurement. The most stable scene we measure is likely in our level sensing applications, which see a variance in the millimeter range measuring the same scene over multiple attempts.

    Morten Nissov said:
    Another parameter specified here is the number of range bins (256). Given that these bins are representing 0->max, one could perhaps say that the accuracy is locked to the bin size, in this case 10m over 256 bins -> 3.9cm which is somehow not the same as the "range interbin resolution" (4.883 cm)

    We get more precise measurement than the bin-size by interpolating between the bins. 

    Best,

    Nate

  • 0 in reply to Nathan Block
    Intellectual 380 points

    Could we say that the measurement error is bounded by the max distance / number of bins then?

    Regarding your first statement, perhaps the radar is very consistent in its measurements, but I can't see how the error (defined as difference between measured and true value) can be less than, in this example, 3.9cm. As finer deviations would fall in one or the other bin no?

    The interpolation is between a peak and the second largest peak on either side? If I'm understanding this correctly then this would mean that the range measurement is split somewhere in the middle between these two bins? Is a similar procedure done for velocity?

  • 0 in reply to Morten Nissov
    TI__Mastermind 19270 points

    Hello.

    Nate is out of office and will be able to provide a response next week.

    Sincerely,

    Santosh

  • 0 in reply to Santosh Krishnan
    Intellectual 380 points

    Thanks for getting back to me, no problem at all.

  • 0 in reply to Morten Nissov
    TI__Mastermind 37780 points

    Hi Morten,

    Morten Nissov said:
    Could we say that the measurement error is bounded by the max distance / number of bins then?

    Ideally yes. The error in the distance measurement should not be greater than the range resolution.

    Morten Nissov said:
    The interpolation is between a peak and the second largest peak on either side? If I'm understanding this correctly then this would mean that the range measurement is split somewhere in the middle between these two bins? Is a similar procedure done for velocity?

    This is correct for at least the people counting toolchain. When dev.ti.com is no longer down, please look at the PDF documents in the People Counting section for more details.

    Best,

    Nate