For a single sensor receiver, the analog channel output signal of PAM or QAM is a cyclostationary process. Thus a fundamental question arises as to whether or not it is possible to identify the unknown linear channel based solely on the second order (cyclic) statistics of the cyclostationary channel output. Gardner in [9] investigated the use of the second order cyclic spectra in channel identification and proposed a scheme requiring the use of a training period during which unknown training sequence is transmitted at a slow rate. His work inspired many other investigations of channel identification from cyclic spectra. Unlike the parametric time-domain approaches described in Chapter 5, here we describe some of these frequency domain methods that rely on cyclic spectra of the channel output explicitly.