Transient behavior of the LMS algorithm: a deterministic approach

The dynamic behavior of the LMS (least mean square) algorithm is analyzed for an important class of nonstationary desired and input signals. Differential equation analysis is used to model the LMS algorithm and predicts in closed form the performance for signal sets that are members of the exponential family, such as damped sinusoids and phasors. The analysis provides new insight into the use of adaptive processing in applications where signals are primarily sinusoidal in nature, such as radar, sonar, and other communication areas.<
>