Adaptive approximate Kernel orthogonalization for antenna array processing Articles uri icon

publication date

  • August 2013

start page

  • 4091

end page

  • 4100

issue

  • 8

volume

  • 61

International Standard Serial Number (ISSN)

  • 0018-926X

Electronic International Standard Serial Number (EISSN)

  • 1558-2221

abstract

  • In a previous paper we have presented a kernel method for antenna array processing using gaussian kernels as basis functions (Approximated Kernel Orthogonalization, AKO) and shown that it presents near optimal performance with an affordable computational burden and small size of the resulting model. However, AKO was only able to operate in batch mode, i.e., it had to process chunks of data (between 10 and 30 frames), which implies a delay in the system response and inability to rapidly respond to changes in the environment. We present here an adaptive extension, the so called Adaptive Approximate Kernel Orthogonalization (AAKO) Method, that is able to on-line process the received frames without loosing performance with respect to the batch version and without any delay in the output, which opens up the possibility of real time implementations.

keywords

  • adaptivity; antenna array processing; kernel method; mercer's kernels; non-stationary ; support vector machines