Estimation of media noise parameters from experimental measurements of nonstationary correlation matrices

Abstract

c i Logk -Colorado 305 Interlocken Pkwy. Broomfield. CO 80021 (303) 464-6628 (303) 464-6776 FAX d e p t @colorpdo.cirrus.comIn this paper, we experimentally measure the nonstationary (non-Toeplitz) correlation matrix associated with media noise and fit it, in a least squares sense, to a simple, closed form, theoretical conelation matrix associated with a parametric media noise model [I] which is experimentally justified by [2,3]. We obtain estimates of thc media noise p a r a t e r s by finding the best correla-

Pattern-dependent noise prediction in signal-dependent noise

“…Note that the noise correlation and, thus, the predictor taps are still conditioned on . It follows that the corresponding predictor error variance is (13) This gives the same result as the low-order AR approximation based on the correlation-matrix decomposition discussed in [4]. It is feasible (and perhaps advisable) to vary the number of predictor taps from one branch to the next, as certain branches are more prone to the noise and, thus, suffer from more severely correlated noise than are others.…”

“…The noise model of our interest is based on a series expansion of the noisy transition response with respect to the position-jitter and width-variation parameters [11], [12]. This model, which we will call the position-width (PW) noise model, allows a relatively simple block-diagram level representation of the channel while proving a good match with experimental data [11], [13]. In this section, we shall focus on the discrete-time second-order PW (SPW) model shown in Fig.…”