Symbol-rate estimation based on filter bank

In non-cooperative communication applications, such as automatic modulation recognition, it is necessary to estimate blindly some parameters of the received signal. We address the symbol-rate estimation of single-tone (i.e., single-carrier) digitally modulated signals. In the proposed algorithm, we first filter the received signal by using a lowpass filter bank, and then pass each of the filtered outputs through a fourth-order nonlinear unit. By proper design, several nonlinear outputs have spectral peaks at the frequency of the symbol-rate. The proposed algorithm extracts several symbol-rate candidates from each nonlinearity output and assigns a confidence metric to each of these candidates. Based on the symbol-rate candidates and their associated metrics, a majority vote scheme is then employed to determine the final estimate of the symbol-rate. The validity of the proposed algorithm has been verified through simulations.