Pii: S0267-7261(98)00005-0

A novel blind deconvolution methodology for identification of the local site characteristics based on two seismograms recorded on the free surface of a sediment site is presented. The approach does not require recordings at depth nor at a nearby rock outcrop, and eliminates the need for any prior parameterization of source and site characteristics. It considers that the surface recordings are the result of the convolution of the ‘input motion at depth’ with transfer functions (channels) representing the characteristics of the transmission path of the waves from the input location to each recording station. The input motion at depth is considered to be the common component in the seismograms (same input in a statistical sense). The channel characteristics are considered to be the part in the seismograms that is non-common, since the travel path of the waves from the input motion location at depth to each recording station is different, due to spatially variable site effects. By means of blind deconvolution, the algorithm eliminates what is common in the seismograms, namely the input motion at depth, and retains what is different, namely the transfer functions of the site from the input location to each recording station. It estimates the site response in both frequency and time domains, and identifies the duration of the site’s transfer functions. The methodology is applied herein to synthetic data at realistic sites for performance validation. The blindly estimated results are in almost perfect agreement with the actual site characteristics, indicating that the approach is a promising new tool for seismic site-response identification from recorded data. q 1998 Elsevier Science Ltd. All rights reserved.