Shot-gather data in deep seismic reflection profiling have complicated characteristics of non-stationary, weak energy, a wide variety of strong noise, and so on. How to increase signal to noise ratio of the shot gather has been one big hard nut to crack. We use the soft threshold filter based on S transform to process the obsolete data and show its chart flow is valid. Stockwell (1996) proposed S transform, which is a time-frequency transformation method to analyze non-stationary signal. Compared with other methods for analyzing time-varying signal, the basic wavelet of S transform does not meet an admissibility condition of zero mean in the time domain. The broadness of time window in S transform has an inverse ratio to frequency, which is wide at low frequency and narrow at high frequency. So, the time-frequency resolution in S transform is related to the signal frequency, which has high frequency resolution at low frequency and high time resolution at high frequency. A simple operation of sum along the time axis in the S transform domain can get the Fourier spectrum. Therefore, it is simple to invert S transform. The hard threshold and the soft threshold are the most common methods in de-noising. The soft threshold is better than the hard threshold generally, though it may cause over-smoothing effects. The threshold is calculated from the noise before the first break in the S transform domain. The data are collected in the junction of the southwest Tian Shan and Tarim Basin, northwestern China. This shot set is exploded with 40 kg charge placed in a 24 m-deep borehole and gathered on a 2 ms sampling rate with a 50 m group interval. The raw shot set has a variety of strong noise, and the signal after 5 s is difficult to recognize. After time-invariant bandpass filtering (5 similar to 25 Hz) and AGC, the signal to noise ratio is obviously improved. However, mixing interference in 5 similar to 25 Hz hampers the identification of thin layers at depth. So, we filter the data further using the soft threshold filter based on S transform to highlight the weak reflection and the Moho reflector. We use the soft threshold filter based on S transform to process the deep seismic reflection data. Our results show that the processing flow does improve signal to noise ratio in deep seismic reflection effectively, which suppresses the frequency mixing interference and enhances weak reflection signal, favorable to track wave groups and further to recognize thin layers. The method increases stratigraphic resolution of the Moho discontinuity especially, which provides the base for subsequent data processing and interpretation of deep seismic reflection profiling. In addition, S transform can be calculated in the frequency domain with the benefits of high calculation efficiency. So, the soft threshold filter based on S transform can also be used in de-noising of other seismic data.
