er-Rao Lower Bound for Blind Signal Separation Henrik Sahlin and Ulf Lindgren Chalmers University of Technology, Department of Applied Electronics, S-412 96 Gothenburg, Sweden, salle@ae.chalmers.se, Telephone +46 31 772 1845 Abstract This paper considers some aspects of the source separation problem. Unmeasurable source signals are assumed to be mixed by means of a channel system resulting in measurable output signals. These output signals can be used to determine a separation structure in order to extract the sources. When solving the source separation problem the channel lter parameters have to be estimated. This paper presents a compact and computationally appealing formula for computing a lower bound for the variance of these parameters, in a general Many Inputs Many Outputs scenario. This lower bound is the asymptotic (assuming the number of data samples to be large) Cram er-Rao lower bound. The CRLB formula is developed further for the two-input two-output system and compared with the results from a Recursive Prediction Error Method.
[1]
Lang Tong,et al.
Waveform-preserving blind estimation of multiple independent sources
,
1993,
IEEE Trans. Signal Process..
[2]
Christian Jutten,et al.
Blind separation of sources, part I: An adaptive algorithm based on neuromimetic architecture
,
1991,
Signal Process..
[3]
Meir Feder,et al.
Multi-channel signal separation by decorrelation
,
1993,
IEEE Trans. Speech Audio Process..
[4]
M. H. Quenouille,et al.
The analysis of multiple time-series
,
1958
.
[5]
S. Kay.
Fundamentals of statistical signal processing: estimation theory
,
1993
.