A subspace method for model order estimation in CDMA

In the code-division multiple access (CDMA), the arising data model is linear. The parametric form of the data is known via the knowledge of the chip sequences but the number of parameters (number of delays) is unknown, because total number of paths is unknown. A number of techniques for model order estimation have been proposed in the literature. Information-theoretic techniques are computationally costly or have modest performance. Methods based on singular value analysis often yield sub-optimal result and sometimes require selection of predetermined thresholds. We introduce a new approach for model order and parameter estimation. In the new method, different signal subspaces are compared to the test subspace using the MUSIC (MUltiple SIgnal Classification) estimator. The proposed method is simpler and more efficient than the popular information-theoretic minimum description length (MDL) criterion in the CDMA application