Embodiments of the present invention provide training sequence generation method which includes generating a pseudo random sequence; obtaining a chirp coefficient of a modulator using a negated chirp coefficient to modulate the pseudo random sequence; constructing a training symbol segment that includes L subcarriers in a frequency domain, transforming the training symbol segment from the frequency domain to a time domain to obtain a training symbol segment in the time domain, and generating a training sequence based on the training symbol segment in the time domain and outputting the training sequence.