Abstract: Cyclostationary feature detection is a blind detection technique which could accurately find the signal in low signal-to-noise ratio, but with lange computation. A parallel asymptotic optimal chi-square detection algorithm was proposed. The optimal cycle frequency and delay was determined through identifying the cycle spectrum characteristic and traversal cycle autocorrelation; the test statistic was constructed based on the cycle autocorrelation and the asymptotic covariance matrix; the real and imaginary parts of two-dimensional vector and spectrum components were merged，and parallel processing was used in the iterative process; the detection probability and the detection time were obtained through simulation. Meanwhile, according to the biased estimation and asymptotic distribution characteristics of maximum likelihood ratio statistic, the expression of detection probability was derived, and the theoretical results of optimum delay are ingood agreement with that of the simulation; the multi-antenna technology makes full use of the spatial diversity; much higher detection probability could be obtained even under the strong interference, and the parallel processing could reduce detection time effectively.