An Accurate, efficient, and stable system to estimate the unknown input frequency of a sinusoidal signal is presented. The proposed design solves the main drawback of the existing phase-based estimator which called a derivative estimator depend on deep learning. These limitations are the inability to estimate low frequencies and the large estimation errors for the frequencies near the Nyquist rate. A Brief mathematical analysis in discrete-time of the proposed system is presented. Proposed estimator performance when the input is a single sinusoid, multiple sinusoids in the presence of additive white Gaussian noise (AWGN) are provided. The accuracy of the proposed esti-mator is the result of dividing the dynamic range of estimation to three regions (low frequencies, middle frequencies, high frequencies) and specify a different formula to calculate the estimated frequency in each region. The boundaries of each region are determined by using a Grey wolf optimizer (GWO) which training bidirectional recurrent neural networks (BRNN) to select the best weights for the estimated frequency. The simulation results ensure the accuracy and validity of the proposed estimator compared to the traditional one. The hardware implementation of enhanced estimator using field-programmable gate array (FPGA), consumed 265 mW, and worked at 375 MHz.