The problem of blind and semiblind channel estimation and symbol detection is considered for long-code wideband code division multiple access (CDMA) systems, including systems with multirate and multicode transmissions. A decorrelating matched filter, implemented efficiently in state-space, eliminates multiaccess interference and produces a bank of vector processes. Each vector process spans a one-dimensional (1-D) subspace from which channel parameters and data symbols of one tiger are estimated jointly by least squares. A new identiflability condition is established,. which suggests that channels unidentifiable in short-code CDMA systems are almost surely identifiable when aperiodic spreading codes are used. The decorrelating Matched filter is implemented efficiently based on time-varying state-space realizations that exploit them structure of sparsity of the code matrix. The mean square error of the estimated channel is compared to the Cramer-Rao bound, and a bit error rate (BER) expression for the proposed algorithm is presented.