Linear predictive digital filtering of narrowband processes in additive broadband noise

Linear prediction filters (LPF) have recently found a large number of applications in such areas as speech analysis and power spectral estimation. An important application of these filters is to aid the cancellation or additive broadband noise from narrowband signal components. Provided the signal bandwidth is significantly less than the noise bandwidth, the LPF may be used to suppress the additive noise without requiring an external reference noise input (as is required in many noise cancelling applications). The purpose of this report is to provide an analytical basis for bounding the performance of a digital LPF when applied to the problem of cancelling broadband additive noise from narrowband signals. Experimental results obtained with a hardware implementation of an adaptive Wiener filter are shown to verify the analytical results.