A New Approach to Lossy Network Compression of a Tuple of Correlated Multivariate Gaussian RVs

The classical Gray and Wyner source coding for a simple network for sources that generate a tuple of multivariate, correlated Gaussian random variables $(Y_1,Y_2)$ is re-examined using the geometric approach of Gaussian random variables, and the weak stochastic realization of correlated Gaussian random variables. New results are: (1) The formulation, methods and algorithms to parametrize all random variables $W : \Omega \rightarrow {\mathbb R}^n $ which make the two components of the tuple $(Y_1,Y_2)$ conditionally independent, according to the weak stochastic realization of $(Y_1, Y_2)$. (2) The transformation of random variables $(Y_1,Y_2)$ via non-singular transformations $(S_1,S_2)$, into their canonical variable form. (3) A formula for Wyner's lossy common information for joint decoding with mean-square error distortions. (4) The methods are shown to be of fundamental importance to the parametrization of the lossy rate region of the Gray and Wyner source coding problem, and the calculation of the smallest common message rate $R_0$ on the Gray and Wyner source problem, when the sum rate $R_0+R_1+R_2$ is arbitrary close to the joint rate distortion function $R_{Y_1, Y_2}(\Delta_1, \Delta_2)$ of joint decoding. The methods and algorithms may be applicable to other problems of multi-user communication, such as, the multiple access channel, etc. The discussion is largely self-contained and proceeds from first principles; basic concepts of weak stochastic realization theory of multivariate correlated Gaussian random variables are reviewed, while certain results are developed to meet the requirement of results (1)-(4).