Blind and Semiblind Channel Estimation - Nonlinear Signal and Image Processing: Theory, Methods and Applications - page 280

Among the different blind equalization techniques, those based on HOS sat-
isfy the two above conditions.
A comprehensive study of the reliability of HOS-based blind equalizers to
GSM is given in Reference 49. They compare the performance of two blind
channel estimators (EVI 53 and WS 54 ) with that of two nonblind techniques
(least-squares and cross-correlation based). For the blind techniques a data
burst contains 142 data bits only, without training sequence, whereas for the
nonblind techniques a standard GSM burst with a mid-amble of 26 training
bits is transmitted. The channels are considered to be time invariant over
a data burst duration, and they were generated according to Reference 16.
The experimental results show that the blind channel estimator EVI entails a
mean SNR loss of 1.2 to 1.3 dB in comparison to the least-squares solution.
Furthermore, its performance remains unaffected by the co-channel interfer-
ences at signal-to-interference ratios (SIR) larger than 10 dB. The HOS-based
blind equalizers converge relatively slowly. The best need an entire GSM data
frame to converge.
An improved blind channel estimation method for GSM using EVI has been
reported. 55 A technique similar to turbo coding is used to iteratively improve
the channel estimate obtained by the blind method. It also solves the (complex)
scalar ambiguity inherent in all blind channel estimation approaches.
The GSM standard has been enhanced to allow higher data rates and to
provide quality services by allocating a larger bandwidth (e.g., PCS-1900) or
by using higher-order modulations (8-PSK in EDGE). Then the equalization
task becomes more difficult because the Doppler spread increases, making
the channels time variant even within the data burst. Higher-output SNR
and high-quality channel estimates have to be ensured in order to deal with
higher-order constellations such as 8-PSK. At this moment it is difficult to
believe that a completely blind solution can satisfy all these conditions. The
semiblind approach may be a more feasible solution for practical systems.
It saves a significant part of the transmission spectrum and provides better
overall performance. However, a good blind method is the core of a semiblind
method. Hence, further development of blind methods is well justified.
8.6 Blind Receivers for CDMA
Radio access in the third generation of mobile communications is based on
CDMA. Therefore, there has been a great deal of research work on CDMA
in recent years. The basic properties of CDMA will not be considered here
because they are presented in many text books, e.g., Reference 56. In the
following we describe the characteristics of the downlink and uplink channels
in the context of channel estimation. Then blind receiver structures for CDMA