• State Resolved
  • Locked Locked
  • Replies 3 replies
  • Subscribers 31 subscribers
  • Views 1107 views
  • Users 0 members are here
Support feedback
Options
Options
Similar topics

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: The calculation method of noise measurement covariance matrix R

Cary Wang
Intellectual 580 points
Part Number: IWR1443
Other Parts Discussed in Thread: IWR1642

I would like to consult the calculation method of noise measurement covariance matrix R (including distance variance, velocity variance, angle variance) .

Generally, observation noise covariance matrix is determined by the measurement accuracy of the sensor. But when I see in the cases of SRR, distance variance, velocity variance, angle variance are determined according to CRLB (Cramer Rao lower bound criterion , in function convertSNRdBToVar). That is Var (f) >=12/ ((2*pi) ^2*SNR*N (N^2-1)), while also considering the range resolution and other information, but this value is the minimum value of variance.

While in the IWR1642 routine, the distance variance is determined according to the range^2*SNR (in the function : RADARDEMO_clusteringDBscan_calcInfoFixed).

 I can understand that the signal-to-noise ratio can reflect the measurement accuracy of the sensor, but the algorithm I do not particularly understand.

I would like to know, for the two calculation of the noise covariance matrix method, which is better? Are there any better algorithms to determine the covariance matrix of the observation noise? Are there any related materials for reference? Thanks!

  • 0
    TI__Mastermind 23570 points

    Hi Cary,

    Wanted to check if you were asking this questions in reference to specific implementation example that we have provided.

    It would help us answer in the right context if you can point out which code example or processing chain you are referring to.

    For example in context to SRR demo : short-range radar reference design

    There is a similar thread that discusses this :  Compiler/AWR1642: questions about "convertSNRdBToVar" algorithm (function) in AWR1642 SRR released C code

    Thank you,

    Vaibav

  • 0
    TI__Expert 4720 points
    Hi Cary,

    The question of which is better is determined by what your application is.

    The SRR demo is used for targets at 80-120m, and the range resolution is fixed to 30cm. This leads to the case where for a car (at 40+m), there is only one detected object (for the entire car). Hence, the only source of variance estimation is the SNR of the target.

    The IWR1642 demo (Traffic monitoring) is assumes that there are many detected objects per target. For e.g. there are at least 3 objects per target. Hence, we can estimate the variance using the variance of the point cloud corresponding to one target, and the SNR of the target.

    So if you can guarantee that there is a point cloud around your target, then you can use the Traffic monitoring approach, otherwise, you can use the SRR approach.

    As far as references, you can search for papers/textbooks on 'Kalman filters', 'particle filters', 'Data association for multi-object tracking', etc. There are many approaches to creating a good tracker. For, e.g the 'People Counting' demo (dev.ti.com/.../ uses another approach for its tracker based on the following resource - Multiple-Target Tracking with Radar Applications - Blackman.

    Regards
    Anil
  • 0 in reply to Anil Mani
    Intellectual 580 points
    Dear Anil and Vaibav,
    An intensive analysis, Thanks for your help!