The MK7 Data Logger Problem

A problem with the IDA MK7 data logger has been discovered that effects all data recorded from the broadband (STS1 and KS54000) sensors. IDA personnel have prepared a detailed report describing the nature of the problem and the impact that the problem has upon the data already collected and distributed through the IRIS DMS. A preliminary draft (size 1.7Mb) of that report is available here.

Summary

This report describes a problem in the IDA MK 7 data acquisition system (DAS) that affects all recorded and archived data originating from the broadband seismometers. The affected channels have SEED convention names of bh* , lh*, and vh*. Other channels follow a different processing path and are not affected. A series of tests conducted on a MK 7 DAS in the laboratory led to the following conclusions:

There is definitely a bug in the Digital Signal Processor (DSP) code that performs filtering and decimation of the digitized signal stream. The output of the defective code Y(t) = F(X(t)) + E(t), where F(X(t)) is the correctly filtered output of the raw digitized stream X(t), and E(t) is an error term.
The error term arises from an incorrect handling of round-off error in the convolution of the filter coefficients and the data.
The error term in the output of the first filtering/decimation, the bh* streams, is approximately equal to 1.6 x 10-3 times the lower 16 bits of the 2's complement representation of the data. Thus, the error is limited in amplitude (104 counts) and produces its largest effect when the input data values cross a N * 216 boundary where N is an integer.
The error increases with each subsequent filtering/decimation so the vh* streams are most affected..
In the spectral domain, the greatest effect of the error appears in the long-period spectrum (greater than 10 seconds).
The DSP code is simple to fix.
Most of the information can be recovered by reprocessing the broadband data (20 sps) and then applying correct filtering/decimation routines to produce the long- and very long-period data channels.

The full report contains some background information on the discovery of the problem, our diagnosis, and the characterization of the effects of the error on the data recorded and archived.

Plans

IDA is currently developing a plan of action to correct the problem. The plan will involve procedures to correct the hardware currently in the field and to reprocess data currently in the IRIS data archive. We will keep the research community apprised of our plans to implement these corrections.

Impact

IDA staff have prepared a C program to simulate the behavior of the data logger. The text of the code may be found in the back of the above draft report and by anonymous ftp The program accepts 32 bit integers as input and applies filtering and decimation routines to mimic the action of the MK7 for test purposes. A flag may be set to run the program in a mode that eliminates the problem. Also in ftp are the low order coefficients cl referenced in the Matlab code fragment within the report.

To try to give some idea of the effect the MK7 filter problem has on the recorded noise levels at IRIS/IDA stations, we performed an analysis the network data for the last half of the year 2000. For this time period, we derived lhz data from the bhz channel in the lab and compared these with the lhz recorded by the DAS. Following the procedures Luciana Astiz used in the FDSN Station book (http://www.fdsn.org/FDSNstation.htm), the seismic noise levels determined from the robust PSE analysis are shown in decibels (db) with respect to acceleration (m**2/s**4)/Hz. In these plots, the horizontal axis indicates period in seconds. The blue curve shows the spectra of the recorded lhz channel and the red curve, of the derived lhz channel.