An adaptive spectral subtraction algorithm to remove persistent cultural noise

Individual seismic stations and seismic arrays suffer from unwanted seismic noise as a result of inevitable population growth and development. This encroachment of noise degrades performance. To respond this this, seismic stations can be relocated to a quieter area. However, moving seismic arrays is often not possible due to logistical reasons. In this study, we examine the feasibility and merit of an adaptive denoising algorithm to reduce the impact of persistent anthropogenic noise. We build our algorithm on spectral subtraction techniques that have been commonly applied to speech and audio traces and develop a noise-suppression technique that is tailored for seismic data. Using the continuous wavelet transform, we subtract estimates of the noise in the frequency domain. We first evaluate this algorithm on synthetic data, consisting of a set of carefully selected events on a low-noise array environment in Alaska. Then we apply this technique to a seismic array in Turkey known to suffer from persistent anthropogenic noise. Our results on individual seismic traces, both on the synthetic and real-world data, demonstrate that our noise-suppression technique is quite successful at improving the signal-to-noise ratio of key seismic phases. The strengths of this approach include its intuitiveness, its ability to adapt to changes in the background noise, and the ability to reduce noise while preserving the phase of the original signal---a prerequisite for use in array analysis. When the denoised traces are used for array analysis we do not find the noise suppression to be as effective as it is on individual traces. We explore a number of reasons why this performance is less desired. Despite the results of our particular implementation, these data demonstrate that the larger family of spectral subtraction techniques offer considerable adaptability and are worthy of further exploration.