No abstract

For t h e case of white uncorrelated inputs, most of t h e blind multichannel identification techniques are n o t very robust a n d only allow to estimate t h e channel u p to a n u m b e r of ambiguities, especially in t h e M I M O case. O n t h e o t h e r hand, all current standardized communication systems employ some form of known i n p u t s to allow channel estimation. T h e channel estimation performance in those cases c a n be optimized by a semiblind approach which exploits b o t h training a n d blind information. W h e n t h e i n p u t s a r e colored a n d have sufficiently different spectra, t h e M I M O channel may become blindly identifiable u p to o n e constant phase factor per input, a n d t h i s u n d e r looser conditions o n t h e channel. For t h e case of spatial multiplexing, possible cooperation between t h e channel inputs allows for more complex M I M O source prefiltering t h a t may allow blind M I M O channel identification u p to j u s t o n e global constant phase factor. We introduce semiblind criteria t h a t are motivated by t h e Gaussian M L a p proach. T h e y combine a training based weighted leastsquares criterion with a blind criterion based o n linear prediction. A variety of blind criteria are considered for t h e various cases of source coloring.

No abstract

No abstract

No abstract