Blind and Semiblind Channel Estimation - Nonlinear Signal and Image Processing: Theory, Methods and Applications - page 257

8

Blind and Semiblind Channel Estimation

Visa Koivunen, Mihai Enescu, and Marius Sirbu

CONTENTS
8.1 Introduction ......................................................... 257

8.2 Time-Varying Channel Model ....................................... 259

8.3 Signal Properties Used in Blind Equalization ....................... 264

8.4 Receivers for MIMO Systems ....................................... 270

8.5 Blind Receivers for GSM/EDGE .................................... 276

..........................................

8.6 Blind Receivers for CDMA 280

8.7 OFDM ............................................................... 285

8.8 Semiblind Techniques ............................................... 288

8.9 Conclusion ........................................................... 290

References .................................................................. 290

8.1 Introduction

In digital wireless communications, the impairment caused by time-varying
multipath fading must be compensated for. The transmitted signal propagat-
ing via multiple paths experiences various delays due to the differing lengths
of the paths. The resulting intersymbol interference (ISI) may distort the re-
ceived signal so severely that the transmitted symbol sequence cannot be
recovered. The effects of the ISI can be mitigated in the receiver with either
direct equalization or channel identification together with, for example, max-
imum likelihood sequence estimation (MLSE). Typically, no prior knowledge
of the channel impulse response is available and the channel is time-varying.
Hence, it needs to be estimated and the estimates updated on a regular basis.
The estimation is commonly performed by transmitting a known sequence of
training symbols in a periodic fashion or by using a separate pilot signal. The
receiver generates a local copy of the training data and the channel coefficients
or equalizer are estimated. In the presence of Gaussian noise, the estimators
derived are typically linear. This type of periodic transmission of training data
or pilot signals significantly reduces the effective data rates. For example, in