Averaged probability of the error in calculating wavelet coefficients for the random sample size (Q2314461)

While analyzing data obtained during the processing of signals, in certain situations, it is desirable to reduce the noise in signals. This is achieved by using a wavelet transform which takes into account the intra correlations in adjacent samples of a digital signal. The small wavelet coefficients are usually ignored and replaced by zero, without affecting the signal. This assumption compresses the data in a digital channel and transfers it faster and economically. A further commonly used technique is to ignore the coefficients whose absolute values do not exceed a given threshold. The models studied in the literature assume that there are a fixed number of empirical wavelet coefficients. Asymptotic optimal thresholds are calculated and orders of different loss functions based on ``probability of error in estimation'' criterion. In this paper, the author considers a situation where the model is such that the number of wavelet coefficients of the signal function is random with a white noise distributed as Gaussian.