DOA esitmation based on support vector machine — Large scale multiclass classification problem

Direction-of-arrival estimation is one of the most important issues in the signal processing field. Support vector machines have gained more and more attention in the DOA estimation in the recent ten years. Most researchers deal with the DOA estimation as a multiclass classification problem, and some approach have been proposed to solve the multiclass classification problem. Although the support vector machine has appeared some superior in its independence from array characteristics, the existing multiclass classification techniques encounter troubles when the class number is very large for DOA estimation. A multi-level technique is proposed to solve the large scale multiclass classification problem. Theoretical analysis shows that the multi-level support vector machine can reduce the computation load significantly and simulation experiments are carried out to confirm the conclusion above.