On generalized covariance-based velocity estimation

We look at the problem of how a moving receiver (mobile station) estimates its own velocity from the autocovariance of functions of the quadrature components of the received signal. In cellular systems, this information-bearing signal conveying voice, data, or both and transmitted from a base station is well characterized as a narrow-band Gaussian random process specified by Clarke's model. The need for such estimation arises, for example, in the design of handoff algorithms for microcellular systems. We consider here the performance of estimators based on two classes of functions: the integral-power sum of the quadrature components G(1,n)(t) = r(n)(t) + r(q)(n)(t) and integral powers of the complex envelope G(2,n)(t) = [r(2)(t) + r(q)(2)(t)](n/2), n is an element of Z(+), respectively. A particular case of this approach for n = 2 was first considered by Sampath and Holtzman [13] and subsequently by Austin and Stuber [3]. For each of the two classes of functions, we derive the optimal (in a defined sense) estimator and show that all estimators based on G(2,n)(t) have identical performance and are generally inferior to those based on G(1,n)(t), We also study the implications of Aulin's model [2] when used to evaluate the performance of estimators based on the integral-power sum of the quadrature components. Last, using a proposed analytical model, we analyze the effect of the scattering distribution on the performance of the estimators, corroborating some previous results obtained in [3] by simulation.