IWR6843ISK-ODS: Ghosts targets generated by wall reflection

Hi, there:

May I know which demo you are trying here?  Multipath reflection is a challenge.   We do not have a general solution to eliminate multi-path.  We have notice that multi-path reflection changes faster and is in general weaker.  Some method can be considered to help reduce ghost reflection.  For example, Capon based people counting demo chain detect less ghost than range-Doppler chain, you can also play with the CFAR threshold.  Or build in more intelligence on the higher level module, for example, If you know the room size, you can also filter some ghost.  

Best,

Zigang

Thank you for your reply,

In this demo our purpose is easy, we just extract point clouds information from IWR6843 by UART and DATA COM and display the point clouds by Python. Our code is improved from people counting visualizer. After we got the point clouds, we cluster clouds into different groups. Then we can classify the Doppler values belong which group.

If multi-path reflection is in general weaker, is there any relation between multi-path reflection and SNR value?

We indeed know the room size, can you provide specific method about the room size filtering ghost points? It would help a lot, thank you

Best,
Charles

HI, Charles:

I see you developed your own visualizer based on people counting visualizer.   But which demo binary are you using?   Can you try people counting binary to see whether performance will be better?

I would say reflection loss is proportional to RCS/(R^4), where R is the target range.  In the case of multi-path, is will be proportional to RCS1*RCS2/(R1^2 * R2*2 * R3^2), and etc. Because there will be multiple RCS and multiple distance involved in multi-path reflection.    You should search in the literature to confirm though.  

Regarding room size, I only know any detection that is outside the room size should not be considered.  Multi-path through wall often results in a detection on the other side of the wall.   But this idea will not help the ghost detection in your video.   

Best,

Zigang

Hello Zigang:

In our task, we already used 3D_people_count_68xx_demo.bin . Multi-path reflection seems a big challenge, I will have a look about
SNR definition in IWR6843.

We will keep trying to exclude the ghost targets by other method, probably Machine Learning algorithm.

Oh, one more question, Do you have any suggestion in experiment scenes setting? For example, the scene should be wider space but smaller boundary box or less stastic object?

All best,

Charles Yen