Spatial smoothing for 2D array

I am trying to test spatial smoothing for Music algorithm for direction finding with a 2D array. spsmooth function worked very well for 1D array, however, for 2D array the function does not work. The change in azimuth/elevation or change in both dimensions translates into one dimension (for one Tx signal). All changes looks like happen in either x/y dimension. It does not also serve the actual purpose (eg. decorrelating incoming signal, it becomes hard to locate the angle for both signals). I am testing the function with real measured value with antenna.

Is there any such a function like spsmooth for 2D array?

[Brahim Saadi]

I don't think so. You would have to implement 2-D spatial smoothing yourself.

[Santhana Raj]

You can perform the spsmooth on each row of your co-variance matrix and then also perform on each column, to get spatial smoothing on 2D.

Answer 1

[Honglei Chen]

The spatial smoothing implementation has a lot to do with how the subarrays are divided. For a linear array, although not a must, a lot of people does what spsmooth does. However, for a 2D aperture, there are a lot more freedom on designing those subarrays. May I ask if you are using a rectangular array, or some other arbitrary geometries? And could you share how you partition the subarrays? All these factors impact the implementation.

[Sanjoy Basak]

Thanks for answering. I started with rectangular array, however, after getting this issue of spatial smoothing, I switched to a simple one, L shaped array with 6 elements. I attached the geometry here. There are 4 horizontal elements and 3 vertical (one shared by both axis). With 2D music I started having this issue of spatial smoothing. Here it was possible to compare the results obtained from 1D search. The Music 1D search can easily obtain the azimuth and elevation perfectly, but then the issue comes about pair matching.

Could you tell me how to implement spatial smoothing algorithm here for 2D Music?

[Honglei Chen]

Sorry I didn't follow up sooner. It looks like the spatial smoothing requires some symmetry in the geometry so you can do the averaging between two subarrays. It is hard to explore that in the L shaped array so the solution is not trivial to implement. You may want to use some other DOA estimator that is more immune to the coherent signals?