Sensor and Server
CSN Sensor Package Detail
The CSN sensor package consists of a 3-axis class-C MEM accelerometer (currently a Phidget 1043) and Linux micro-computer (Raspberry-Pi 3b). The sensor has a native sampling rate of 250 samples per second (sps), which decimated in the microcomputer to to 50 sps to reduce the data volume and to ensure the device does not function in the audible range.
The package maintains time synchronization through the Network Time Protocol (NTP), that is further refined by resampling of the signal between fixed time points. The use of NTP to maintain time rather than GPS-based mechanisms allows the sensor to be deployed without a sky view (i.e. indoors). Most sensor deployments have a small UPS device included to allow operations to continue during power outages of several hours. Sensors are stuck to the floor with two-sided sticky pads.
Custom software running in the microcomputer (client software) manages the sampling of accelerometer, decimating and buffering the data, time synchronization, and communicating with the cloud. It also performs higher level processing of the signal such as picking and ground motion parmeter estimation. It archive the data on a local SD card in case of an Internet failure. The software is written in Python and is remotely updateable.
The client software performs a earthquake detection analysis (picking) on the incoming data and will trigger whenever the acceleration exceeds a station-configurable factor, typically 0.005g. A trigger initiates a process to measure pick attributes such as the peak acceleration and its onset time in the following 3 seconds. It is planned that the process will also determine the Earthquake Early Warning (EEW) measurements. The pick information is then sent immediately to the active-Q process in the Amazon Cloud Services (AWS). The latency of the pick transmission is very small (see figure). The waveform data is sent at regular intervals to the AWS (typically every 10 mins). This setup allows the client to respond quickly to an earthquake, but otherwise use low-overhead communication.
Sensor Accuracy and Sensivity
The CSN package currently uses the Phidget 1043 sensor. A table showing the comparison between this class-C sensor and a class-A sensor (Episensor) and a cell-phone accelerometer is shown below.
Sensitivity vs Distance/Magnitude
The frequency magnitude performance of the sensor is shown in the following plot. For example, the Phidget is sensitive to magnitude 3.5 and above at a distance of 10 km, but the Android phone requires magnitude 5 for the same distance.
Comparison of CSN with Episensor
This example below shows a side-by-side comparison of the CSN sensor and an Episensor accelerometer colocated in the Pasadena Seismic Vault. The event is the 2014/03/29 La Habra Earthquake (magnitude 5.1), whose epicenter is 25 km away.
Cloud Data Processing and Analytics
After the waveform data have been initially sent to the cloud, they are downloaded to a server facility at Caltech for analysis, archival and distribution purposes. The data are stored in 10- minute chunks in SAC (Seismic Analysis Code) format. A database of the individual chunks, and of the waveform collecting stations is kept. A server application delivers these data to users by cutting and joining the segments into user-requested time windows. The request results data can be streamed over the internet to the user's computer.
The 'pick' data from events detected in realtime (by each of the stations so configured) are sent to the cloud and stored there in a cache. Each pick consists of a short message that includes information about the size of the event, the location of the station detecting it, and an accurate timestamp of when the event occurred. The cache is regularly emptied by a server facility at Caltech. This process incorporates two functions, the first of which is to store all the picks in a database, for analysis, archival and distribution purposes, and the second of which is to relay the realtime picks to remote clients. These clients currently comprise the Earthquake Early Warning system's FinDer [Ref] application, and the ShakeMap [Ref] system that produces maps of regional ground motion.
For each 10-minute chunk of waveform data we compute the average, the mean, the median, and the root-mean-square values, and store these in a database. These give a measure of the average performance of the station, and its noise environment. For select stations we Compute the average H/V (horizontal/vertical) ratio as a function of frequency in order to determine the shallow site characteristics.
Real Time Monitoring
The software executing in each of the stations includes features that allow it to be interrogated and controlled over the network. Interrogations include demands for the latest sensor data, the latest pick events, the elapsed time since the software started execution, demands for information about the station's configuration (its 'metadata'), and so forth. In some cases the responses to interrogations are in the form of raw ASCII text, in some cases they are JSON encoded data, and in other cases they are Web HTML data that will render charts, for example, in the requester's browser. One use case for the real time monitoring data is a so-called 'Building Display', an application that requests pseudo real time data from all the stations in a specific building, and renders the data graphically and dynamically to show, for example, the motion of the each of the building's floors and the inter-story drift in real time.
Data Quality Control
A suite of software tools are used to regularly monitor the quality of the data in the system. These tools include scripts that produce graphical schematics showing the completeness of data produced, sent to the cloud, downloaded, and archived into storage. The tools include scripts that tabulate the status of all stations in the system, showing whether they have produced waveform data and pick data, and when the most recent communication was received from each station. The schematics and reports are sent to the system's operations managers on a regular basis. In addition to the regular monitoring of the data by the aforementioned tools, data quality control may also be monitored in some cases by observation of the real time displays described in the previous section.