On Strong Consistency of A 2-Dimensional Frequency Estimation Algorithm

BQ Miao,Y Wu,LC Zhao

IF: 1.4

1998-01-01

Statistica Sinica

Abstract:In this paper, we prove that an algorithm proposed in Rao, Zhao and Zhou (1993) gives a strongly consistent estimation of 2-dimensional frequencies. where y(m, n) consists of p distinct noise contaminated 2-D sinusoids in the (µ, ν)-plane and w(m, n)'s are independent and identically distributed random variables with zero mean and variance σ 2 . This signal model can be found in synthetic aperture radar imaging, frequency and wave number estimation in array signal processing and nuclear magnetic resonance imaging and so on. To recover the p 2-D sinusoids, the techniques often used are periodogram, autoregressive spectral estimation and maximum entropy spectral estimation etc. (See, say Dudgeon and Merseresau (1984).) But all these methods require an exhaustive search in a 2-D space to obtain the estimates of the 2-D angular frequencies and need plenty of data. In the paper by Kung, Arun and Rao (1993) and Shaw and Kumaresan (1986), two sets of 1-D frequencies {µk} and {νk} are estimated separately instead of one set of 2-D frequencies {(µk ,ν k)}. Hua (1992) proposed a new algorithm by a matrix enhancement and matrix pencil (MEMP) approach. However, in the set {µk} (or {νk}), two or more components may be equal, but they are treated as different parameters in the estimation of the set {µk} (or {νk}). This concludes that the number of parameters estimated by MEMP may be larger than is necessary. In pairing the components in the sets {µk} and {νk}, the computation cost by MEMP in an optimum pairing process is proportional to p!. In order to reduce the cost, Hua (1992) suggested a pairing process with the computation cost proportional to p 2 but its performance is not clear. As