Authors:

Takao Inoue and Robert W. Heath Jr.

Reference:

Submitted to IEEE Trans. on Signal Processing., August 2009

Abstract:

Predictive vector quantization is a high resolution vector quantization technique used in applications such as speech, image, and video processing where the correlation of data over time or space can be exploited. Unfortunately, conventional formulation does not readily extend to the case where the data lies on the Grassmann manifold due to its non-Euclidean geometric structure. In this paper, we propose a Grassmannian predictive coding algorithm where the geometric difference and prediction are used to formulate an extension of predictive vector quantizer to the Grassmann manifold. We analyze the quantization error and derive bounds on the distortion attained by the proposed algorithm. Application of the proposed algorithm to single-user and multiuser limited feedback multiple-input multiple-output wireless communication systems operating under temporally correlated channels shows that significant symbol error rate and achievable rate improvements can be obtained.