Our Technology

Seismic Warning Systems started by studying applications of earthquake warning to understand how best to design an effective earthquake warning system. By starting with a clean slate, we avoided performance compromises.
Sensor Station
QuakeGuard Network sensors
Dual Sensors

Each sensor station we deploy has two identical sensors placed 20 to 100 meters (66 to 330 feet) apart. This is very different from stations built for earthquake research. It may seem like an extravagance, but the two sensors provide significant benefits for an earthquake warning system.

Tell me more: Technology »Science »Partners »Build »
  • 2 force balance broadband tri-axial accelerometers
  • 124dB dynamic range
  • 3g clip threshold
  • 1000 samples-per-second
  • Synchronous sampling of all 6 channels
  • Installed in 2 meter (6 feet) deep post holes
    • Provides a stable diurnal temperature
    • Reduces possibility of vandalism
Sensor Siting
  • 6-8 km apart in urban areas; 1-3 km from known faults
  • Sensors are placed on a fault parallel axis (mostly)
  • Soil conditions should be uniform between the two sensors
  • About 500 stations will provide coverage for California
  • Redundant operation tolerates sensor failure
  • Homogeneous sensor network simplifies analysis
  • Homogeneous network reduces deployment and support costs
  • A sensor can be repaired or calibrated without taking the station offline
False Alarm Prevention
  • Earthquakes hit both sensors the same way at the same time
  • Local sources affect each sensor differently
  • Comparison of two streams of data rejects electrical and cultural noise
  • An earthquake can be confirmed with one station instead of 3-4
Seismic vs local source waveforms
CC-Attribution-NonCommercial-ShareAlike 4.0/SWS
Epicenter Location
  • Arrival time lag determines azimuth
  • First motion analysis resolves ambiguity
  • Can locate earthquake with one station instead of 3
Back azimuth diagram
CC-Attribution-NonCommercial-ShareAlike 4.0/SWS
Station Hosting Criteria

The need to separate the sensors by at least 20 meters (65 feet) (100 meters is preferred) limits the locations suitable for placing sensors. The requirements for sensor station locations are:

  • Close proximity to the fault
  • Located close (1-3 km) to a needed station site
  • Suitable locations for sensors:
    • Far from magnetic fields (e.g. transformers)
    • Far from major sources of vibrations
    • At least 1.5 structure heights from tall structures
  • Open ground between sensors (no structures, concrete)
  • Good Internet access
  • Reliable power
Sensor station construction
Sensor installation diagram - annotated
CC-Attribution-NonCommercial-ShareAlike 4.0/SWS
Sensor Station
GPS antenna

In order to estimate the effects of the very largest earthquakes, we need to measure how much the ground shifts along the fault as the earthquake is happening. For this purpose, QuakeGuard stations all have a GPS receiver measuring the position of the ground. These data are combined with data from the accelerometers to provide measurements that are better than either is able to provide on its own.

Tell me more: Technology »Science »
Sensor Fusion
A multi-rate Kalman filter fuses GPS and accelerometer data to produce high sample rate measurement of displacement. The filter is computed in real time. Displacement parametrics are provided as part of earthquake data packets.
  • GPS rate: 20 sps
  • Accelerometer rate: 1000 sps
High-Fidelity Displacement Data
  • Combine low-rate GPS with high-rate acceleration data
  • Uses multi-rate Kalman filtering
  • Output is high-rate displacement for seismic analysis
High-rate displacement
Real-time Rupture Tracking
  • Measure relative displacement in real time
  • Identify near-fault stations
  • Determine direction and extent of rupture
South Napa GPS map
Sensor Station
CPU board
CC-Attribution-NonCommercial-ShareAlike 4.0/SWS
Local Processing

All processing of sensor data is done at the sensor site. This eliminates any delays associated with sending the data elsewhere for processing. It also means that considerably more computational power is available for earthquake analysis.

Tell me more: Technology »
Multiple CPUs

Each station has a 4 core processor which handles communications, system monitoring, diagnostics, network management, operating system, encryption, and some scientific algorithms.

Distributed Computing
  • Eliminates the single point failure inherent in a data center
  • Greatly increases available computing resources: 2.5 TMIPS and 20 TFLOPS for the system
Sensor Station
DSP board
CC-Attribution-NonCommercial-ShareAlike 4.0/SWS
Digital Signal Processing

Specialized computers exist optimized for analyzing data from sensors are called Digital Signal Processors (DSP). Our stations have high performance DSPs capable of performing all necessary analyses in a small fraction of a second.

Tell me more: Technology »
State-of-the-art DSPs
  • Optimized for processing seismic signals
    • Fixed (32 and 64 bit)
    • Single and double precision floating point
  • High performance
    • Fixed point: 40 GMACsbillion multiply-accumulate operations per second
    • Single precision floating point: 20 GFLOPsbillion floating point operations per second
    • Double precision floating point: 5 GFLOPsbillion floating point operations per second
    • Single precision 2048pt FFT: 14 μs
  • Much more efficient than general purpose computers
    • Peak performance @ 2.5W
Sensor Station
Loma Prieta GGB 360 waveform
CC-Attribution-NonCommercial-ShareAlike 4.0/SWS
Advanced Algorithms

Special algorithms are used to analyze ground motions to estimate how intense the earthquake will be. This is used to ignore the small earthquakes that occur frequently. The trick is to complete the analysis quickly enough to permit alerts to be given at the epicenter.

Tell me more: Science »
Intensity not Magnitude
  • Magnitude is a single number for every earthquake
  • Intensity depends on your location
  • Moderate, nearby earthquakes feel like large, distant ones
South Napa intensity map
Public Domain/USGS
Better Performance
  • Going directly to intensity saves time
  • Going directly to intensity is more accurate
  • Using magnitude as an intermediate step adds errors
Magnitude vs, Intensity diagrams
CC-Attribution-NonCommercial-ShareAlike 4.0/SWS
Time budget
Algorithms must be fast. See The Need for Speed
Network map
Optimized Communications

Since we process all sensor data locally, we only need to send messages when there is an earthquake. We optimize our communications for small messages which can be sent very quickly, even in the presence of network congestion. This provides more warning time to the end user.

Tell me more: Technology »
Low Latency Protocols
  • All messages are single packets
  • Packet sizes are all below fragmentation thresholds
  • Simultaneous authentication and data exchange
  • Optimum dynamic timeout and retransmission windows
  • Proof against man-in-the-middle attacks
  • Immune to replay attacks
  • Perfect forward security
Centralized vs. distributed system
CC-Attribution-NonCommercial-ShareAlike 4.0/SWS
Mesh Networking

Most earthquake warning systems stream sensor data to central data centers where it is processed. This leaves the system vulnerable to failures or interruptions in power and communications. Our system is based on hardened, dedicated communcations between sensor stations using a dynamic mesh.

Data centers
A data center
CC-BY-SA-3.0/Victorgrigas, from Wikimedia Commons
Regional Data Centers

Data centers contain servers that monitor the sensor stations and distribute earthquake information to end users. Data centers are located in all urban areas close to end users to reduce communication delays. No analysis of seismic signals is performed in the data centers.

Tell me more: Technology »
Data Processing
Seismic data processing is located at the sensor stations in a fully distributed architecture
Central Data Centers
Central data centers house servers that monitor and control system operation.
Regional Servers
Earthquake warning information is distributed from many server locations to minimize communication delays and improve reliability. Regional servers can assume system monitoring and control tasks as needed.
Data Center Requirements
We lease space for our servers in qualified data centers. Each data center houses several servers to have enough capacity to handle all message traffic even if 50% of the servers fail.
  • Earthquake hardened facility
  • Uninterruptable backup power
  • Located away from other hazards, such as floods
  • Direct peering with several Internet backbones
  • Private fiber connections with our other data centers
  • Well connected to local ISPs
  • Able to prioritize bandwidth in emergencies
End user applications
NetQG control model
CC-Attribution-NonCommercial-ShareAlike 4.0/SWS
QuakeGuard Appliances

Our earthquake warning signal is received by a QuakeGuard Appliance which first verifies that the signal is valid and then generates an alert or takes an action specified by the end user. The simplest appliance model has a built in speaker and flasher for audible and visible alerting of people. Other models support a variety of interfaces to control equipment. The appliance constantly tests connections and operation to ensure its readiness in an earthquake.

Tell me more: Technology »Security »

Optimized Communications

Messages sent to the QuakeGuard Appliance use a highly optimized protocol and distribution network to provide the most warning time possible.

Control Integration

  • Direct device control with relays
  • Serial and network control protocols
  • SNMP
  • Software plugins for protocol support
  • Expandable hardware interfaces, such as CAN

Different Models

There are several QuakeGuard Appliance models supporting different control, alerting, and integration requirements.
NetQG audio model

Audio Alerting

QuakeGuard industrial model

Industrial Control

Communication Security
Each QuakeGuard Appliance uses state of the art encryption and authentication to ensure that the systems connected to the appliance are protected from malicious hacking attacks.
  • Peer to peer authentication
  • Elliptic curve cryptography
  • Authentication and authorization are separated
  • State dependent message validation