A novel adaptive algorithm using an IIR narrow band filter (NBPF) is presented to detect a single sinusoid corrupted by Gaussian noise. This algorithm is not only computationally efficient but also suitable for VLSI implementation. The derivation of the adaptive algorithm for frequency begins with an autoregressive (AR) model to formulate a constrained-optimization problem. Then, a new algorithm for frequency detection is derived via the method of Lagrange multiplier. MATLAB simulation shows that our approach has good tracking ability and noise reduction. Finally, a high speed VLSI architecture is designed and implemented according to the proposed algorithm. After hardware simulation, the chip area and clock rate of the whole architecture is 1804/spl times/1804 /spl mu/m/sup 2/ and 75 MHz respectively using 0.35 /spl mu/m COMS 1P4M technology.