Abstract: An adaptive linear two-dimension sinusoid tracer with selectable frequency and bandwidth parameters is deduced to estimate the instantaneous value and the amplitude of a sinusoid by least square error, gradient descent method and rotation transform. A number of the tracers in parallel constitute a linear comb filter with adjustable bandwidth in every frequency that is a multi-dimension linear ordinary dynamic system. The uniformly asymptotical stability of the comb filter is validated by LaSalle invariance principle. The frequency characteristics in the form of vector for a sinusoid and matrix for signal with lots of sinusoidal components are derived by Laplace transform. All components of signal as well as their amplitudes can be estimated exactly simultaneously if all frequencies of the signal are the same as the frequency parameters of the comb filter. The frequency grid effect of the proposed algorithm as well as the effect of bandwidth parameter on precision in steady state are investigated. The algorithm's validity is verified by simulation results.