09:15 〜 09:30
[J04-1-03] Reducing Digitiser Latency for Earthquake Early Warning: New Strategies for Seismic Hardware
The latency between earthquake origin and issuing an alert is one metric used to assess the performance of Earthquake Early Warning Systems. Total latency accumulates from many sources (e.g. P-wave propagation, digitization, transmission, receiver processing, triggering, event declaration). Data delivery latency is the time taken for digitization, packetisation, transmission and receipt of data. The two main contributors to data delivery delays are data packetization (Brown et al., 2011) and digitisation.
Data packetisation: SEEDlink protocol is commonly used, however, packet size is fixed to 512-byte miniSEED records; each with a 64-byte header. SEEDlink can cause typical large data transmission latencies of >0.7 s at typical sampling rates. We propose a new ultra-low-latency protocol, GDI, an efficient and flexible method of exchanging data between seismic stations and data centers in large networks. Instead of fixed-length packets, GDI dispatches data sample-by-sample as acquired by the data-logger. A self-adaptive scheme assesses available bandwidth and judges the smallest packet size needed for the fastest transmission. A light 4-byte header maximizes efficiency for small packet sizes.
Digitisation: Most current seismic data-loggers implement acausal (linear phase FIR) filters; but these cannot be computed in true real-time and give spurious precursors to seismic wave onsets (e.g. Scherbaum & Bouin, 1997). Acausal filtering typically contributes >0.3 s to data delivery delays. Causal (minimum phase) FIR filters are typically non-standard in seismic data-loggers, yet, the improved latency and sharper onset determination mean that causal filters offer significant advantages for EEWS, reducing digitization delays to 20–60 ms (at sample rates of 100–250 sps).
Using a twofold strategy of causal filtering combined with a rapid data transmission protocol, such as GDI, we show significantly reduced data delivery latencies to 40–60 ms (at typical sample rates).
Data packetisation: SEEDlink protocol is commonly used, however, packet size is fixed to 512-byte miniSEED records; each with a 64-byte header. SEEDlink can cause typical large data transmission latencies of >0.7 s at typical sampling rates. We propose a new ultra-low-latency protocol, GDI, an efficient and flexible method of exchanging data between seismic stations and data centers in large networks. Instead of fixed-length packets, GDI dispatches data sample-by-sample as acquired by the data-logger. A self-adaptive scheme assesses available bandwidth and judges the smallest packet size needed for the fastest transmission. A light 4-byte header maximizes efficiency for small packet sizes.
Digitisation: Most current seismic data-loggers implement acausal (linear phase FIR) filters; but these cannot be computed in true real-time and give spurious precursors to seismic wave onsets (e.g. Scherbaum & Bouin, 1997). Acausal filtering typically contributes >0.3 s to data delivery delays. Causal (minimum phase) FIR filters are typically non-standard in seismic data-loggers, yet, the improved latency and sharper onset determination mean that causal filters offer significant advantages for EEWS, reducing digitization delays to 20–60 ms (at sample rates of 100–250 sps).
Using a twofold strategy of causal filtering combined with a rapid data transmission protocol, such as GDI, we show significantly reduced data delivery latencies to 40–60 ms (at typical sample rates).