Operational and functional capabilities of the SNDA as a system for SASA data analysis are illustrated below by its application to the problem of detection and parameter estimation of a so called "hidden explosion" seismic signal. The possible scenario of avoiding a Comprehensive Test Ban Treaty is to perform secret nuclear test by triggering a nuclear device with a seismic signal from a rather strong earthquake. In this case the explosion wave phases are obscured by coda waves of the earthquake. Latter typically have an intensity significantly stronger than seismic noise at observational sites of the monitoring network. Nevertheless, registering of a noise field by a SASA and processing SASA recordings with the help of special statistically optimal algorithms provides a chance of reliable CTBT monitoring even in the case of implementing of the avoiding scenario described.

Figures (shown below) illustrates results of application of the SNDA advanced processing methods to the "hidden explosion" problem. In this study we used multichannel seismograms from underground nuclear test at Novaya Zemlya site (24 Oct. 1990) and earthquake in Hindu Kush (25 Oct. 1990) registered by NORESS SASA. Simulation of NORESS recordings containing a "hidden explosion" signal obscured by an earthquake coda and seismogram processing for explosion signal detection and parameter estimation were made with the help of a special SNDA script, comprising a variety of the SNDA stack commands and SA-procedures.

Figure below shows P- wave seismogram from Novaya Zemlya explosion (NZE) (trace (1)) and P-wave with coda wave seismogram from Hindu Kush earthquake (HKE) recorded by the central NORESS sensor. The seismograms were filtered in the frequency band (0.5-5) Hz, resampled, shifted in time and scaled by SNDA stack commands.

The simulated "hidden explosion's" 25-channel NORESS data are displayed in figure below. The created by the SNDA tools mixture of the real NZE and HKE NORESS seismograms contains the NZE P-wave obscured by the HKE coda with the RMS SNR=0.5 and the onset time at 23 sec later the HKE P-wave arrival. This data is the raw material for the succeeded analysis. Note that the explosion signal is not recognizable on this seismogram mixture du to similarity of amplitude and frequency contents of the NZE P-waves and HKE coda waves.

The following figure demonstrates the results of detecting of the NZE P-wave on the HKE coda background with the help of the adaptive statistically optimal detector (ASOP) [1-3]. The detection procedure is applied to the output of the beamforming procedure with the beam steered to NZ site (trace 3). The output of the same beamforming procedure applied to "pure" HKE seismograms is shown at the trace (4) for comparison with the trace (3). One can see that conventional beamforming does not provide suppression of HKE coda waves sufficiently for reliable detecting by the standard STA/LTA detector. At the same time trace (2) containing the ASOP statistic time series demonstrates presence of the peak from NZE P-wave which significantly exceeds the ASOP statistic fluctuations due to HKE coda wave oscillations. Comparison of the trace (2) with the automatically chosen threshold (equal to the doubled root mean square value of this trace) allows to detect reliably NZE P-wave and assign the appropriate time interval containing suspected "hidden explosion" signal as the object for succeeded thorough analysis (trace 1).

The following figure illustrates capability of a the adaptive group filtering (AOGF) algorithm [2-4] for extracting of waveforms of weak seismic phases obscured by a coherent noise using a SASA data. The HKE coda is a strongly coherent one that yields in the insufficient suppression capabilities of the conventional beamforming being applied to its NORESS recordings (trace 3). At the same time just the strong HKE coda coherence allows the AOGF procedure to gain the effective coda wave suppression that is seen in the AOGF output trace 3. At the trace (1) the output of the AOGF applied to the "pure" NZE seismograms is depicted for comparison with the trace 3. One can see that waveform of the NZE P-wave is reproduced by the AOGF rather accurately. Note, that adaptation of AOGF algorithm was made using NZE+HKE NORESS seismograms mixture at the time intervals (0-22) sec and (35-60) sec., i.e. before and after the interval containing the NZE P-wave. These intervals were automatically chosen as a result of detection procedure preceding AOGF.

At this figure one can see the output traces of the AOGF applied to the "pure" NZE and NZE+HKE mixed NORESS recordings.

Superposition of these traces by the SNDA graphic means (see picture below) allows to assess distortions of the NZE P-waveform which are stipulated by the signal extraction from HKE coda waves in this rather difficult case with initial SNR=0.5.

The picture below illustrates estimation of the NZE P-wave onset time with the help of maximum likelihood algorithm [1-3] applied to AOGF output (trace 3 at the picture). The algorithm is realized in the SNDA in two forms: as the stack command (for scripts) and interactive graphic command.

At the picture below the power spectra of beamforming and AOGF outputs are shown. They are calculated for the time interval where the NZE P-wave exists. The figure demonstrates the small distinction of the spectrum of extracted NZE P-wave (after AOGF) from the spectrum of the initial NZE P-waveform. At the same time the spectrum of the conventional beamforming output does not allow to estimate the NZE P-wave spectrum in given case with SNR=0.5, that makes it impossible to implement the standard source identification procedures.

The following pictures are devoted to results of arrival direction estimation for a suspicious wave detected in the HKE coda. The estimation was performed with the help of variety of spatial spectral analysis algorithms. Let us emphasis that evaluation of seismic ray arrival direction together with estimation of P and S phase onset times provide the information needed for location of event epicenter based on a data from single SASA.

It displays the spatial spectrum calculated using NZE+HKE seismogram mixture at the time interval containing NZE P-wave with the help of conventional broad band F-K analysis algorithm [3]. One can see that the spectrum maximum which is the estimate of a wave arrival direction only slightly differ from arrival direction of HKE P-wave (According with the geography and the standard travel time table it has the azimuth = 102.4 degrees and apparent velocity 14.8 km/sec.).

Application of the high resolution spectral analysis (HRSA), namely the modified Capon algorithm with estimation of an array data matrix power spectrum by the multidimensional autoregressive-moving average modeling [3], allows to detect that two waves are present in the HKE+NZE seismogram mixture at the analysis time interval (see picture below). The global maximum testify to presence of HKE P-wave.

Measuring of the second maximum location made with the help of the graphic interactive SNDA tool gives the arrival direction estimation for the second wave equal to: azimuth = 26,5 degrees, apparent velocity = 7.4 km/sec (see picture bbelow).

These values are rather far from real NZE P-wave arrival direction equal to: the azimuth = 32.9 degrees, the apparent velocity = 10.4 km/sec.

At last, the advanced algorithm was implemented for accurate estimation of the NZE P-wave arrival direction based on the NZE+HKE seismogram mixture. This is the adaptive maximum likelihood algorithm for direction estimation of a signal plane wave arriving to a SASA site together with coherent interfering waves; it is supposed that latter can be observed independently at a time interval which does not contain the signal wave, and algorithm adaptation can be made using this observations [4]. The algorithm resulted in the spatial spectrum estimate (see the picture below).