[CDO14] Properties, performance and practical interest of the Widely Linear MMSE beamformer for nonrectilinear signals
Revue Internationale avec comité de lecture :
Journal Signal Processing,
vol. 97,
pp. 269-281,
2014
Mots clés: Noncircular, Widely linear, (non)Rectilinear, MMSE, Interferences, SAIC, Beamformers, Improper, Rectilinearity detection, Blind source separation
Résumé:
Widely Linear (WL) Minimum Mean Square Error (MMSE) estimation has received a great
interest these last 20 years for second order (SO) noncircular signals. In the context of
radio communications networks, it has been shown in particular that WL MMSE receivers
allow to implement Single Antenna Interference Cancellation (SAIC) of one rectilinear
interference, such as Binary Phase Shift Keying (BPSK) or Amplitude Shift Keying (ASK)
interference, or of quasi-rectilinear interference, such as Minimum Shift keying (MSK),
Gaussian MSK (GMSK) or Offset Quadrature Amplitude Modulation (OQAM) interference,
hence their great interest for Global System for Mobile Communications (GSM) cellular
networks in particular. However, one may wonder whether WL MMSE receivers remain
attractive for SO noncircular nonrectilinear interferences, not so scarce in practice. The
purpose of this paper is mainly to answer to this important question by giving, in a selfcontained
and unified way, some new insights into the behavior, properties and
performance of the WL MMSE beamformer in the presence of arbitrary noncircular
signals and interference which are not necessarily rectilinear. It is shown in particular
that, surprisingly, WL MMSE receivers lose their practical interest for strong interferences
which are not rectilinear. This breakthrough thus generates a new open problem for the
choice between linear and WL MMSE receiver corresponding to the detection of
rectilinearity (and/or quasi-rectilinearity), instead of noncircularity, in a given noisy
observation. Although this question is out of the scope of this paper, we finally propose
preliminary tools based on blind source separation methods to solve this problem