Abstract: Compressed sensing is a kind of information sampling method on Sub-Nyuqist rate. The symbol detection methods on CS are usually reconstructing Nyquist signal from Sub-Nyquist samples, and then detect receiver symbol according to the traditional symbol detection theory. As to the high sample rates problem of reconstruction based compressed sensing detection methods, this paper studies the relationship between information sampling and generalized likelihood ratio detection, and proposes a compressed detection method without reconstruction signals at Nyquist rate. This method separates the reference part and information symbol part by dual channel delay structure, and detects the sub-Nyquist symbol on the sparse correlation characteristic between the two parts. The experiment results show the method is more stable to multipath and ISI.