1. PMCC: MAIN FEATURES .......................................................................................................... 2 1.1 1.2 1.3 1.4 1.5 1.6 1.7 2 INTRODUCTION ......................................................................................................................... 2 DETECTION BY CORRELATION .................................................................................................. 3 TIME CORRECTION FOR A NON PLANAR ARRAY ........................................................................ 3 THRESHOLD OF DETECTION: THE CONSISTENCY ....................................................................... 4 THE PROGRESSIVITY ................................................................................................................. 4 CHECKING DATA QUALITY ........................................................................................................ 5 TIME FREQUENCY ANALYSIS AND POST-PROCESSING ............................................................... 5 DESCRIPTION AND TUNING OF THE MAIN PARAMETERS .......................................... 7 2.1 2.2 2.3 2.4 ARRAY PARAMETERS ................................................................................................................ 7 DETECTION PARAMETERS ......................................................................................................... 7 FREQUENCY PARAMETERS ........................................................................................................ 9 FAMILY PARAMETERS ............................................................................................................. 10 3 REFERENCES ............................................................................................................................. 11 4 APPENDIX ................................................................................................................................... 12 CEA/DASE PMCC Documentation Main Features and tuning parameters 1. 1.1 PMCC: MAIN FEATURES INTRODUCTION In contrast to a set of isolated sensors, a dense array, whose aperture is of the order of the wavelengths of the signals of interest, allows similarity measurements of the recordings to avoid uncertainties encountered with individual arrival-time picking. Similarly to seismology, most of the infrasonic waves can be represented at a local scale by a set of planar waves using the well-known relation f (r , t ) e i (t kr ) where: 2f k is the wave vector associated to frequency f and phase velocity c c 2f 2 is the angular frequency. T The frequency content of a recorded wave can easily be determined using a single station. At the opposite, a set of sensors is needed to calculate the propagation parameter k . When the aperture of this set of sensors is equivalent to some wavelengths of signal, this set is named an array. At the opposite, when the aperture is much larger than the wavelength, it is named a network. In the case of a network, the signal is often very different from one sensor to another and the measure of the propagation parameters is derived from the set of arrival-times by inversion, as described by Husebye (1969) in seismology. On the opposite, in the case of an array, we use the similarity of the signals to compute arrival time differences using classical techniques of signal processing theory. This set of arrival time differences is used to compute the propagation parameters with a Husebye’s derived method. The most classical method for estimating these wave parameters in the case of an array is a systematic search in a specific domain of wave vector using the signals recorded on the sensors. For example the 2f disc defined in the wavenumber plane by the relation k corresponds to all the waves with a V min frequency f, with any azimuth and with a velocity V Vmin . For each discrete wave vector of this regularly discretized domain, the time delay at each sensor is calculated and the delayed signals are summed. When the signals are mainly composed of random background noise, the energy variation of the sum is small over the entire wave vector field. In contrast, if the signals are associated with a specified vector k 0 , the energy obtained for k 0 will be much larger than for the other vectors. A lot of methods have been proposed by different authors to find the wave vector which produces the maximum energy [Capon, 1969]. This is not a trivial problem because data are discrete in the space domain (i.e. only few sensors are used). This implies that for each frequency, false results can be obtained due to correlated signals over one or more periods (ambiguity effect). To study these effects, Capon suggested to compute the array beam forming function B( k ) e ikr j . It represents the array j response for a wave which arrives vertically under it, i.e. with a horizontal planar wave reaching all the sensors at the same time. The phase velocity of the wave measured is then infinite, which leads to a maximum amplitude for k 0 . The main assumption linked to this type of methods is the search of a signal modeled by a planar wave 2/16 CEA/DASE PMCC Documentation Main Features and tuning parameters recorded on all the sensors of the array. In practice, some of these assumptions are confirmed to various extents as diverse sources cause interference with these signals. That is why a more flexible method, less constraining with respect to the model, is proposed. It is based on conventional signal processing techniques to detect a stable signal on two or more records, partly by relaxing the planar wave model rigidity. The PMCC method (Progressive Multi-Channel Correlation), originally designed for seismic arrays, proved also to be efficient for analyzing low-amplitude infrasonic coherent waves within non-coherent noise. 1.2 DETECTION BY CORRELATION A temporal signal s(t ) can be represented by its Fourier transform S ( f ) A( f )e i ( f ) where A( f ) represents the spectral amplitude and ( f ) is the phase. The background noise is characterized by a rapid variation of both A( f ) and ( f ) from one sensor to another, even if they are closer than one wavelength of signal. On the opposite, in case of signal propagating between the sensors, the following relations are available: A2 ( f ) A1 ( f ) 2 ( f ) 1 ( f ) (r2 r1 ) These relations indicate that no deformation exists between the two signals, and that the only difference is a delay depending on the relative positions of the sensors (i.e.: (r2 r1 ) k (r2 r1 ) in the case of a planar wave). Based on these two observations, a signal-processing tool can be used to detect a signal present on the records si (t ) and s j (t ) . The correlation function is used to measure the time delay t ij between the two records. In case of a wave propagating without distortion (assuming a planar wave), this delay is the same for all frequencies of the signals: tij 1 ( j ( f ) i ( f )) 2f This measurement is made in the time domain with values ranging from -1 to 1. Taking into account all frequencies, it measures in a given time window the similarity of the signals when shifted in time. The maximum of the correlation function gives the time delay between the signals. This method enables a decision to be made on whether there is a signal in a set of simultaneous records, independently of any information on previous records. 1.3 TIME CORRECTION FOR A NON PLANAR ARRAY In some cases, for a non planar array, the travel time differences due to elevation differences between sensors become not negligible. For such arrays, the elevation of each array element should be taken into account. The calculation of the time delays between two sensors is to be improved. It can be shown that the time delays then also depends on the local sound speed and the elevation angle. In a first order, this correction is applied: j x sin( ) y cos( ) z cos(i) VT c 3/16 CEA/DASE PMCC Documentation Main Features and tuning parameters where: is the backazimuth (often called azimuth), angle of wave front approach, measured clockwise between the north and the direction towards the source, c is the local sound speed d is the horizontal distance between sensor j and the barycenter j is the time-delay between sensor j of coordinates (x,y,z) and the center of the array (0,0,0) i is the incidence angle between the direction of propagation of the wave front and the vertical VT sin( i ) 1.4 d , is the horizontal trace velocity across the array. c THRESHOLD OF DETECTION: THE CONSISTENCY To avoid ambiguity problems when correlating the records from sensors too far apart, the analysis is initialized on the smallest groups of three sensors. The correlation function is used to calculate the propagation time tij of the wave between sensors i and j. For each subnetwork (i,j,k), the closure relation tij +tjj+tki = 0 should be obtained. In the presence of background noise the phase is unstable. Therefore, the delays measured in this case are the result of random phase combinations. These delays, independent of the amplitude of each elementary wave, become random, and the closure relation given above is no longer valid. The consistency of the set of delays obtained using all the sensors of R n is then defined as the mean quadratic residual of the closure relations: rijk tij t jk tki 6 2 i, j , k Rn cn r ijk n(n 1)( n 2) i j k If this consistency is below a given threshold cThreshold , a detection is observed on R n . 1.5 THE PROGRESSIVITY To minimize errors in the calculation of the wave parameters, distant sensors are progressively added using a criterion based on a comparison between their distance to the subnetwork and the computed wavelength. This progressive use of distant sensors has two main effects: the removal of false detections which could be due to correlated noise at the scale of the starting subarrays, and a better estimation of the wave parameters by increasing the array aperture. After being initialized with a small subnetwork of three sensors, in order to avoid ambiguity problems inherent in the correlation of signals from distant sensors, the wave parameters calculated on the initial subnetwork R n are used when adding other sensors. For that, a propagation of a planar wavefront is assumed. The new measured time delay is given by the maximum of the correlation function which is the closest to the one that has been estimated. Each elementary detection is therefore defined by several parameters such as the consistency value, the number of sensors participating to the detection, the frequency, the horizontal trace velocity and the backazimuth. Such a detector is independent of the signal amplitude and uses only the intrinsic information of the recordings. As long as the closure relation is valid, the use of sensors increasingly further apart gives more precise wave parameters since the aperture of the network increases with each new sensor. The final solution is given by the largest subnetwork. 4/16 CEA/DASE PMCC Documentation Main Features and tuning parameters 1.6 CHECKING DATA QUALITY In the PMCC configuration file (see in Appendix), a list of starting subnetworks of three sensors is defined. As explained previously, the detection is initiated through each of these subnetworks. Then, as long as there are available sensors and the consistency value remains lower than its threshold, new sensors are aggregated. To avoid wrong results due to the lack of data in the recordings, an automatic procedure checks the data quality. If the initial subnetworks contain sensors with consecutive zeros in the recordings, this procedure looks for other set of three sensors belonging to the array. Among all possible combinations calculated from the remaining sensors, the best subnetworks are selected. The principle is to sort them according to symmetry and size criteria. Equilateral triangle of small aperture is the best configuration. The maximum number of new eligible subnetworks corresponds to the number of subnetworks defined in the configuration file. 1.7 TIME FREQUENCY ANALYSIS AND POST-PROCESSING The processing is performed consecutively in several frequency bands fi, and in adjacent time windows tj covering the whole period of analysis. To avoid unrealistic wave detection, a further condition is introduced. A set of several elementary detections in the time-frequency domain is considered to represent one detected wave (corresponding for example to different frequency bands or adjacent windows). Conversely, several waves with different parameters may coexist in the same time window but in different frequency bands. Each wave must be identified separately. To do this, a nearest-neighbor search of elementary detections in the (ti,fi,VTi,i) domain is used [Cansi, 1995; Cansi, 1997]. The final detection is thus an aggregate of close-enough points in this domain. Different distances can be used to connect close-enough points, but the more usual is a weighted Euclidian distance defined by: The ’s are used as weighting factors to allow the comparison of quantities of different meanings. These weights can be tuned independently. As opposed to the other weights, V is the only dimensionless parameter. The velocity weight is expressed in percent. Figure 1 presents an example of PMCC results before and after this processing. 5/16 CEA/DASE PMCC Documentation Main Features and tuning parameters Figure 1 - Results of PMCC calculation on signals generated by a gaz explosion in Belgium recorded at the Flers experimental infrasound station set up in Normandy (France). The PMCC results (horizontal trace velocity and azimuth) are presented in time / frequency diagrams. Values are given according to the color scales. The results are presented from 0.1 to 4 Hz in 10 equally spaced frequency bands. Azimuths are given clockwise from North. Top: elementary detection. Bottom: final results after the post-processing, 6/16 CEA/DASE PMCC Documentation Main Features and tuning parameters 2 DESCRIPTION AND TUNING OF THE MAIN PARAMETERS Depending on type of signals to be processed (eg waveforms and frequency content) and the background noise level, the PMCC parameters have to be set in order to ensure an optimum detection providing a compromise between a low detection threshold and low numbers of false detections. Too large threshold values will allow a low level of detection, but a large number of false detections are expected. On the opposite, too small values are not appropriate to detect coherent waves with poor signal-to-noise ratio, but a very few number of false detections will be obtained. PMCC requires input of two configuration files: pmcc.ini and filters.ini. These contain, respectively, the computation parameters and the filter coefficients. Examples of these two files as well as the format of the result files are provided in the Appendix. The main PMCC parameters are related to the array, the detection, the filtering and the post processing. They are described below along with some guidelines underlying the choice of suitable values. 2.1 ARRAY PARAMETERS 1. [X,Y] are the sensor coordinates (relative to the barycenter, in km). For a relevant calculation of the wave parameters, they should be accurate enough regarding the sampling rate, the array aperture and the wavelength of the signals. 2. NbSensors: number of sensors of the arrray. 3. NbSubNet is number of subnetworks to be processed. In order to avoid ambiguity problems inherent in the correlation of signals from distant sensors, the calculation is generally initiated with small subnetworks of three sensors. Then, for each subnetwork, distant sensors are progressively added using a criterion based on a comparison between their distance to the subnetwork and the computed wavelength. This progressive use of distant sensors has two main effects: the removal of false detections which could be due to correlated noise at the scale of the starting subarrays, and a better estimation of the wave parameters by increasing the array aperture. Since all sensors of the array will be added as long as the consistency remains a threshold value, it is not necessary to define all available subnetworks. Furthermore, the computation time grows linearly with NbSubNet. For an 8-element array, a maximum reasonnable number of NbSubNet is 5. If the aperture of the full-array is significantly smaller than the wavelength, in order to make the process less sensitive to highly correlated undesirable signals, the largest subnetworks can be defined. 4. GeoSubNet describes each selected subnetwork, with numbers referring to the sensors. Their aperture should be in the same order of wavelength of the signals and their geometry should be as symetric as possible. The equilateral triangle is the optimum configuration. 5. DelayCorrection (s) is the time correction to be applied to each channel (null is the default value). 2.2 DETECTION PARAMETERS 1. WindowLength (s) is the length of time window to compute the correlation. It can not be lower than the maximum propagation time of the wave between two sensors. This value is generally increased to ensure a smoothing of the time base, which significantly reduces the number of false detections and improves statistics on the measurements of the wave parameters. A 7/16 CEA/DASE PMCC Documentation Main Features and tuning parameters reduction in the number of families (final detection of PMCC aggregating close-enough pixels in the time / frequency / speed / azimuth domain) is then expected, providing a synthesis of information. In the case of long duration phenomena such as microbaroms or MountainAssociated Waves, WindowLength can contain several periods (up to 10). In case of transient signals, WindowLength can also be greater than the maximum propagation time between sensors. The disadvantage of a broad time window is the anticipation of the detection and the increase of its duration. In order to take into account this effect, a time picking procedure has to be used for an accurate arrival time measurement. 2. TimeStep (s) is the time shift between two consecutive windows. It cannot exceed WindowLength. It defines the time width of the elementary detection in the time / frequency domain, also named pixels. A low value for TimeStep allows to monitor in greater detail the time variation of the wave parameters without interfering with the detection performance. Reducing TimeStep by a factor of 2 proportionately increases the number of pixels per family. In order to maintain a similar level of detection using TimeStep/2, ThresholdLFamMin must be multiplied by two and the standard deviations Sigma_t, Sigma_f, Sigma_v and Sigma_a can be reduced (see below). 3. ThresholdConsistency (s) is the maximum consistency threshold for detection. This parameter is the first criterion to reject a pixel, and the choice of its value has a major impact on the detection performance. ThresholdConsistency has to be higher than the sampling period, otherwise the closure relation will never be fullfilled and no detection will be obtained. In unfavorable background noise conditions, including the consequent interferences in the intercorrelation functions, ThresholdConsistency must be raised. In addition, too high a value will noticeably increase the number of false detections linked to the incoherent nature of the background noise. The tuning of this value is achieved taking into account the frequency band being studied and the mean noise level prevailing at the station 4. Q defines the threshold distance for integration of a sensor in the current subnetwork. There is integration if the distance between the sensor and the barycenter of the subnetwork is lower than Q. This threshold is useful to remove far-away sensor for which poorly-correlated signals are expected. If the maximum distance between sensors is not too large compared to the wavelength, Q can be set to a high value (typically 50). In that case, all sensors will be used for the calculation. 5. ThresholdNbSensors is the minimum number of sensors participating to a detection. It ranges between 3 and the number of elements in the array. A high level of detection is obtained with a large value of ThresholdNbSensors. 8/16 CEA/DASE PMCC Documentation Main Features and tuning parameters 2.3 FREQUENCY PARAMETERS 1. NbFilters defines the number of bands available to be filtered. For a given frequency range, NbFilters also defines the frequency width of the pixels in the time/frequency domain. Typically, for frequency range of one decade, a reasonable number of NbFilters is 10. 2. NbCoeff defines the number of AR filter coefficients available in the file filters.ini. 3. DT_Band defines the index of the filter bands to be processed (numbered from 0 to NbFilters) 4. Decim is the decimation factor. The tuning of this parameter must take account the frequency band to be processed in relation to the sampling frequency. As show by the figure below, a filter with a too narrow band produces instability in the filter frequency response, which will disturb detection performance. Typically, for a bandwidth of 0.5 Hz and a sampling frequency of 20 Hz, Decim=1. For a bandwidth of 0.05 Hz, Decim=5 provides a good compromise. [0.005-0.01] Hz Fe = 20 Hz – Decim = 1 [0.005-0.01] Hz Fe = 20 Hz – Decim = 10 0.005-0.01] Hz Fe = 20 Hz – Decim = 50 Figure 2 - Digital filter frequency response as a function of the sampling frequency (Chebyshev order: 2). In the current version of PMCC, a unique time window length can be defined. In order to cover a frequency band larger than one decade, it is preferable to split the processing in several bands and to adjust the time window length according to bands. The three bulletins are then merged for the purposes of statistical analysis and the creation of the graphs contained in the present report. For infrasound monitoring in the [0.05-5] Hz band, three bands can be defined: Band 1: The [0.5-5] Hz band dedicated to detection of transient signals whose origin may be natural or man-made, and which may propagate over distances of several hundred kilometers (Decim=1, WindowLength=30 s). Band 2: The [0.05-0.5] Hz band dedicated mainly for the detection of microbaroms and possible remote large events such as explosions, meteorites (Decim=5, WindowLength=90 s). Band 3: The [0.005-0.05] Hz band dedicated to detection of atmospheric waves such as Mountain Associated Waves or gravity waves (Decim=20, WindowLength=200 s). The fairly low order of the filters used (order 2) allows an overlap of detections among these three frequency bands; for example, microbaroms are sometimes detected in Band 1, and Mountain Associated Waves in Band 2. 9/16 CEA/DASE PMCC Documentation Main Features and tuning parameters 2.4 FAMILY PARAMETERS 1. ThresholdDistance allows the integration of one candidate pixel into one growing family selected among the existing ones. It is a dimensionless threshold defining a maximum acceptable distance in the time / frequency / speed / azimuth domain. A family is a candidate if at least one of its pixels has distance to the candidate pixel which is lower than ThresholdDistance. Since it is a normalized distance, an acceptable value is one. 2. ThresholdDate (s) defines the maximum standard deviation in time between a pixel and a family. It must be higher than TimeStep, otherwise no integration can occur. This is because the distance in time t between all candidate pixels and the family cannot be less than TimeStep, yielding to a ratio t/Sigma_t always greater than one (typical value of ThresholdDistance). 3. Sigma_t (s) defines the maximum distance in time between one pixel and a family. It must be higher than TimeStep, otherwise no integration can occur. If there is no more pixel to be integrated because of this condition, the family is closed. 4. Sigma_f (Hz) defines the maximum standard deviation in frequency. It must be greater than the frequency resolution of one pixel, eg (Fmax-Fmin)/NbFilters. If this is not the case, the distance in frequency between a candidate pixel and an existing family will always be greater than the value for ThresholdDistance. 5. Sigma_v (%) and Sigma_a (°) define the maximum standard deviation in velocity and backazimuth, respectively, for the integration of a plot into a family. These values should be larger than the expected physical and numerical resolution of the wave parameters of the detected signals. For the processing in a low frequency band (wavelength significantly greater than the array aperture), these values should be in increased. 6. ThresholdLFamMin (number of pixels) any closed family whose length is less than this threshold is removed. 7. ThresholdLFamMax (number of pixels) prevents memory troubles with families growing infinitely. It limits the extention of all families whose size extends beyond its threshold. In that case, the families are closed. ThresholdLFamMax is defined as a number of pixels and can be set to 1000. In the case of permanent detections generating by long duration phenomena (microbaroms for example lasting frequently from hours to days), the number of pixels per family quickly reaches ThresholdLFamMax. Then, contiguous families with similar propagation parameters will be separated despite the continuous nature of the signals. 8. VstoreMin and VstoreMax (km/s) are the minimum and maximum values and the average horizontal trace velocity of the family. If the calculated value is out of this range, the family is removed. These parameters allow to suppress false detections based on unrealistic values of the velocity according to know propagation models. 10/16 CEA/DASE PMCC Documentation Main Features and tuning parameters 3 REFERENCES Capon, J, High resolution frequency wavenumber spectrum analysis, Proc. IEEE, 57, 1969. Cansi, Y., An automatic seismic event processing for detection and location: the PMCC method, Geophys. Res. Lett., 22, 1021-1024, 1995. Cansi Y. and Y. Klinger, An automated data processing method for mini-arrays, CSEM/EMSC European-Mediterranean Seismological Centre, NewsLetter 11, 1021-1024, 1997. 11/16 CEA/DASE PMCC Documentation Main Features and tuning parameters 4 APPENDIX Some examples of the format of the initialiation and result files of PMCC are provided. In addition to the parameters described above, the followings have to be defined: Signal parameters: - Fe: sampling frequency (Hz) - FilePath: directory containing the signal files to be read - FileNames: signal file names to be read - TypeFiles: format of signals to be read [Ascii: 0 – Fonyx: 1 – Css: 2] - TypeCss: Css format - FoffCss: read offset for Css format File parameters: - FilePath: directory containing the signal files to be read - FileNames: names of signal files to be read - ResultFile: full path for the results file - FamiliesFile: full directory path for the family file - BulletinFile: full directory path for the bulletin file - FiltersFile: full directory path for the file containing the filter parameters - SaveIntermResults: saving intermediate results (0/1) - SaveBands: saving frequency bands - [SaveBeginDate, SaveEndDate]: saving the time window Detection parameters: - StoreRes: saving result file (0/1) - StoreFam: saving family file (0/1) - StoreBul: saving bulleting file (0/1) - ClassFile: path for use with fuzzy classification methods - StoreClass: =0 indicates that the latter methods are not implemented Fe = 20.000000 Decim = 1 WindowLength = 30.000000 WindowGap = 5.000000 StartingDate = 0,0,0,0 EndingDate = 14400,14400,14400,14400 q = 50.000000 ThresholdConsistency = 0.300000 ThresholdNbSensors = 3 NbSensors = 4 NbFilters = 10 NbCoeff = 4 FilePath = "U:\Doc\RFP\Task4\I33mg\02_08_2002_00_04\"U:\Doc\RFP\Task4\I33mg\02_08_2002_00_04\"U:\Doc\R FP\Task4\I33mg\02_08_2002_00_04\"U:\Doc\RFP\Task4\I33mg\02_08_2002_00_04\" FileNames = "ArchI33MG.1013126400.w"ArchI33MG.1013126400.w"ArchI33MG.1013126400.w"ArchI33MG.1013126400. w" TypeFiles = 2 TypeCss = s3 FoffCss = 0,864000,1728000,2592000 StoreFam = 1 12/16 CEA/DASE PMCC Documentation Main Features and tuning parameters StoreRes = 1 StoreBul = 1 StoreClass = 0 VStoreMin = 0.100000 VStoreMax = 1000.000000 FamiliesFile = "U:\Doc\RFP\Task4\I33mg\02_08_2002_00_04\hf\families.txt" BulletinFile = "U:\Doc\RFP\Task4\I33mg\02_08_2002_00_04\hf\bulletin.txt" FiltersFile = "U:\Doc\RFP\Task4\I33mg\02_08_2002_00_04\hf\filters.ini" ResultFile = "U:\Doc\RFP\Task4\I33mg\02_08_2002_00_04\hf\results.txt" ClassFile = "new_regle.fis" NbSubNet = 4 X = -0.016474, 0.63416, 0.25469, -0.87238 Y = -0.025133, 0.94778, -1.4259, 0.50321 GeoSubNet = {0,1,3}{0,2,3}{1,2,3}{0,1,2} Bands = 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 ThresholdDate = 10.000000 ThresholdLFamMin = 7.000000 ThresholdLFamMax = 300.000000 ThresholdDistance = 1.200000 sigma_t = 10.000000 sigma_f = 1.000000 sigma_v = 0.030000 sigma_a = 0.200015 OPTION = 1 DelayCorrections = 0.0, 0.0, 0.0, 0.0 SaveIntermResults = 0 SaveBands = 1, 2, 3 SaveBeginDate = 30.0 SaveEndDate = 60.0 Table 1 – Format of the pmcc.ini file. The following parameters must be defined in the configuration file filters.ini: - [Fmin, Fmax]: low and high cut-off frequencies for each filter (reduced frequencies standardized at Fe/2), - [ForwardCoeffs, ForwardCoeffs]: AR filter coefficients. Fmin = 0.005, 0.0245, 0.044, 0.0635, 0.083, 0.1025, 0.122, 0.1415, 0.161, 0.1805 Fmax = 0.0245, 0.044, 0.0635, 0.083, 0.1025, 0.122, 0.1415, 0.161, 0.1805, 0.2 ForwardCoeffs 0.0298513, 0, 0.0298513, 0, 0.0298513, 0, 0.0298513, 0, 0.0298513, 0, 0.0298513, 0, 0.0298513, 0, 0.0298513, 0, 0.0298513, 0, 0.0298513, 0, = -0.0597026, -0.0597026, -0.0597026, -0.0597026, -0.0597026, -0.0597026, -0.0597026, -0.0597026, -0.0597026, -0.0597026, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0.0298513 0.0298513 0.0298513 0.0298513 0.0298513 0.0298513 0.0298513 0.0298513 0.0298513 0.0298513 ReverseCoeffs = -3.45563, 4.49686, -2.62477, 0.58356 -3.39048, 4.38357, -2.57528, 0.58356 -3.27449, 4.18722, -2.48718, 0.58356 -3.1094, 3.91956, -2.36178, 0.58356 -2.8977, 3.59657, -2.20098, 0.58356 -2.64255, 3.23755, -2.00718, 0.58356 -2.34778, 2.86396, -1.78328, 0.58356 -2.01781, 2.4981, -1.53265, 0.58356 -1.65759, 2.16185, -1.25904, 0.58356 -1.27252, 1.87528, -0.966556, 0.58356 Table 2 – Example of format of the filters.ini file (Fe=20 Hz, Decim=1, Fmin=0.1 Hz, Fmax=4 Hz,NbFilters=10). 13/16 CEA/DASE PMCC Documentation Main Features and tuning parameters The results files generated are result.txt and families.txt. The result.txt file contains nine columns. Each line corresponds to an elementary detection (pixel) in the time/frequency domain. The ninth column is not read. - TPS: date (s), relative time at beginning of file - FRQ: central frequency (Hz) - AZI: backazimuth (rad) - VIT: trace velocity (km/s) - CONS: consistency (s) - NBST: number of channels contributing to the detection - RMS: amplitude (Pa/ Hz) - COR: correlation coefficient (dimensionless) - CHAN: sensors contributing to the detection 15.00 15.00 15.00 20.00 20.00 25.00 25.00 25.00 25.00 30.00 30.00 (…) 0.29500000 0.68500000 1.07500000 1.85500000 2.24500000 0.68500000 1.85500000 2.24500000 3.80500000 1.07500000 1.46500000 4.246 2.136 0.441 0.719 0.953 3.869 0.953 0.953 0.956 3.673 0.933 4.740445 0.0791 40.228505 31.045538 3.445694 0.0354 0.351246 0.0000 0.206430 0.0500 0.351246 0.0000 0.351246 0.0000 0.350808 0.0354 0.241774 0.0500 0.363258 0.0000 4 0.0500 0.0500 4 4 3 4 4 4 3 3 0.00899230 0.73 3 0.00302511 3 0.00189505 0.00031060 0.27 0.00031761 0.4 0.00064196 0.30 0.00032001 0.43 0.00033106 0.55 0.00036291 0.58 0.00039923 0.35 0.00034552 0.33 |0|1|3|2 1.18 |0|1|2 0.51 |0|1|2 |1|2|3|0 |0|1|2|3 |0|1|3 |0|1|2|3 |0|1|2|3 |0|1|2|3 |0|1|2 |0|1|3 Table 3 – Format of the result.txt file. The families.txt file contains the following seven parameters: - TPS: date (s), relative time at beginning of file - FRQ: central frequency (Hz) - AZI: backazimuth (rad) - VIT: trace velocity (km/s) - RMS: amplitude (Pa/ Hz) - COR: correlation coefficient - CONS: consistency (s) Each line in families.txt is the result of a selection of elementary detections contained in result.txt. The file families.txt lists homogeneous groups – the ‘family’ concept – within which the detections are sufficiently close neighbors in the time/frequency/speed/azimuth domain. The formation of families is dependent on the Detection Parameters defined in pmcc.ini. Each family is separated in the file by a dotted line. The file provides backazimuth and speed statistics for each family (mean value and standard deviation). -------------------------------------date: 200.00 s freq: 1.075 Hz azim: 0.93272 rads sp: 0.34248 km/s RMS: 0.00037377 Pa cor: 0.27 cons: 0.000 date: 30.00 s freq: 1.465 Hz azim: 0.93272 rads sp: 0.36326 km/s RMS: 0.00034552 Pa cor: 0.33 cons: 0.000 date: 15.00 s freq: 2.245 Hz azim: 0.93272 rads sp: 0.36326 km/s RMS: 0.00093410 Pa cor: 0.53 cons: 0.000 date: 15.00 s freq: 2.635 Hz azim: 0.95294 rads sp: 0.35125 km/s RMS: 0.00082642 Pa cor: 0.49 cons: 0.000 date: 15.00 s freq: 3.025 Hz azim: 0.95294 rads sp: 0.35125 km/s RMS: 0.00075597 Pa cor: 0.48 cons: 0.000 date: 15.00 s freq: 3.415 Hz azim: 0.93272 rads sp: 0.36326 km/s RMS: 0.00069941 Pa cor: 0.57 cons: 0.000 Azimuth mean: 0.948 Speed mean: 0.354698 Azimuth standard deviation: 0.01281 Speed standard deviation: 0.00585810 -------------------------------------(…) Table 4 – Format of the families.txt file. The detections are grouped together in a bulletin file containing the following ten parameters: 14/16 CEA/DASE PMCC Documentation Main Features and tuning parameters NAME S T S SSD A ASD fmin fmax RMS Cons Cor N Description Origin time (UT) End Time (UT) Mean propagation speed (km/s) Speed standard deviation (km/s) Mean backazimuth (deg) Backazimuth standard deviation (deg) Minimum frequency for family (Hz) Maximum frequency for family (Hz) Amplitude RMS (Pa/ Hz) Consistency mean value (s) Correlation mean coefficient Number of plots per family Table 5 – List of the result parameters. OT: Origin Time (UT) ET: End Time (UT) S (km/s): Speed (km/s) SSD (km/s): Speed Standard Deviation A (°): Azimuth (deg) ASD (°): Azimuth Standard Deviation fmax (Hz): Minimum frequency (Hz) fmin (Hz): Maximum frequency (Hz) RMS (Pa/V Hz): Root Mean Square (Pa) CONS (s): Consistency Mean Value COR: Correlation Mean Coefficient N: Number of elements of the detected family ==================================================================================== OT AND S SSD A ASD fmin fmax RMS CONS COR N 2002-01-25 2002-01-25 2002-01-25 2002-01-25 2002-01-25 2002-01-25 2002-01-25 (…) 02:42:40 02:42:45 02:43:15 02:45:00 02:44:30 03:50:25 07:13:00 2002-01-25 2002-01-25 2002-01-25 2002-01-25 2002-01-25 2002-01-25 2002-01-25 02:43:00 02:43:05 02:43:50 02:45:15 02:45:20 03:50:40 07:14:35 0.382 0.355 0.329 0.313 0.335 0.313 0.328 0.007 0.003 0.004 0.005 0.011 0.006 0.004 8.0 1.1 3.415 3.805 99.129 0.12 0.40 9 359.5 0.3 0.295 3.025 211.996 0.05 0.45 29 359.6 1.7 0.295 2.635 264.880 0.09 0.37 25 356.5 3.3 1.855 3.025 1204.654 0.06 0.35 10 0.8 0.8 0.295 1.075 795.344 0.09 0.48 20 330.1 1.5 0.295 0.685 138.614 0.12 0.57 7 310.7 2.0 0.295 3.805 229.810 0.05 0.53 119 Table 6 – Format of the bulletin.txt file. 15/16 CEA/DASE PMCC Documentation Main Features and tuning parameters Table 7 – PMCC flowchart describing the detection process. 16/16