Sampath, B.Li, YeLiu, K.J. RayA subspace-based blind channel identification algorithm using only the fact that the received signal can be oversampled is proposed. No direct use is made of the statistics of the input sequence or even of the fact that the symbols are from a finite set and therefore this algorithm can be used to identify even channels in which arbitrary symbols are sent. A modification of this algorithm which uses the extra information in the more common case when the symbols are from a finite set is also presented. This LS-Subspace algorithm operates directly on the data domain and therefore avoids the problems associated with other algorithms which use the statistical information contained in the received signal. In the noiseless case, it is possible for the proposed Basic Subspace algorithm to identify the channel exactly using the least number of symbols that can possibly be used. Thus, if the length of the impulse response of a channel is JT, T being the symbol interval, then it is possible to use this algorithm to identify the channel using an observation interval of just (J + 3)T. In the noisy case, simulations have shown that almost exact identification can be obtained by using a few more symbols than the theoretical minimum. This is orders of magnitude better than the other blind algorithms. Moreover, this algorithm is computationally very efficient and has no convergence problems.en-USdigital communicationsestimationsignal processingblind identificationequalizationSystems Integration MethodologyTechnical Report