Rapidly developing Compressed Sensing theory looks promising for many practical applications, since it allows us to reconstruct K-sparce signals and to reduce some hardware requirements. In this work, we consider the problem of changing noise properties after recovering and its influence on the radar false alarm rate. Due to nonlinearity of the recovering algorithm there is no analytical solution allowing finding a noise distribution after the reconstruction. Therefore, by an empirical approach we come to a solution, where the well-known cell averaging constant false alarm rate detector can be used for a compressed sensing radar. We analyze its performance by simulation and test it with real radar data.