A low-complexity semi-blind channel estimation method for OFDM systems

Abstract

Some blind channel estimation methods for OFDM systems exploring the cyclostationarity embedded in the cyclic prefix have been reported in several papers. Some of these methods can be only applied to short OFDM symbols, but fail for long symbols. Other methods suffer from large amount of computation of cyclic autocorrelation values and are difficult to be used in practical situation. This paper presents a new method of semi-blind channel estimation in which the computation is largely reduced. Simulation result demonstrates that the pro- posed method performs well for long OFDM symbols and is superior to those reported earlier for several channel model. It is more likely to be used in practice. I. INTRODUCTION Since pilots or training signals reduce the transmission capability, blind and semi-blind channel estimation always attracts considerable research attention. It is shown that the second-order cyclostationarity embedded in the signals can be used for blind channel estimation provided that some necessary and sufficient conditions are met(1)-(4). In OFDM system, the cyclic prefix happens to induce cyclostationar- ity and helps identifying the channel blindly. Literature (5) and (6) independently developed subspace channel estimation method for OFDM system based on cyclostationarity and both achieved fairly good performance. However, their simulation results are both acquired under the assumption that the number of the subcarriers is not very large. For practical situation when hundreds of subcarriers are used, performance of both two methods above are degraded. Literature (7) proposed another method. It exploits the zeros of the cyclic autocorrelation and extracts the zeros of the channel. However, because of the symmetry of the cyclic autocorrelation of the OFDM signal, many zeros are on the unit circle and must first be eliminated. For deep fading channel, there are some channel zeros near the unit circle, which are submerged by those of cyclic autocorrelation of OFDM signals and are difficult to be extracted. For all above methods, computation of many cyclic auto- correlation values are a time consuming task. The number of autocorrelation values that are needed for estimation is twice as many as one OFDM symbol length. Thus the methods are This work is supported by funds for National Natural Science under Grant 60241003. too complex to be used in practical situation. In this paper, we make use of the fact, which is ignored by all above methods, that the cyclic autocorrelation of OFDM signal are made up of several impulses and most of the values along the τ-axis are zero. The delay spread of the channel will not make the impulses interfere with each other if only the delay spread does not exceed the whole OFDM symbol. We propose a simpler semi-blind channel estimation method based on this characterization. It need only a few cyclic autocorrelation values and saves a lot of computation. Also it will not suffer from the large number of zeros on the unit circle as in (7). finally, we point out that all blind methods estimation mentioned above can only obtain a vector linear related to the true channel, so one pilot is needed in practice for the final estimation. The loss of bandwidth efficiency caused by the pilot is so little that it can be neglected. Simulation result shows that the new method is applicable to fairly long OFDM symbols with superior performance over those reported earlier