NRAO Home  >  Green Bank  |  Wiki Topic:    GB > Main > TWikiRegistration? > ToneyMinter > 7A10

Users | Groups | Offices | Changes | Index | Contents | Notify | Search | Jump to Topic:

Achieving Integration Times Less Than 1 Second With The Spectrometer

Background

GBT proposal 7A10 wishes to monitor a source to search for radio bursts that are narrow-band and quickly change frequency with time. They require 0.35 second time resolution or better but yet they still need high spectral resolution. This precludes the use of the spigot.

Configuration of Spectrometer

The proposal calls for 200 MHz bandwidth and the lowest resolution with 8192 channels. This configuration uses only a single bank of the spectrometer. Using frequency switching with frequency offests of 0,0 MHz you get four phases (spectra) during each integration. Using this setup the minimum integration time is 0.4 seconds with the frequency switching giving an effective resolution of 0.1 seconds.

NFS Latency

Taking a 60 second scan with the spectrometer setup normally results in an extra 4 seconds being needed for the spectrometer to write the data to disk. By writing the data to a local disk (see instructions below) the latency can be reduced to 4 seconds for a 30 minute scan.

Setting Up The Spectrometer To Write To A Local Disk

• You should have the operator stop sdfits on datacapture using TaskMaster.
• The observer should run "turtle" instead of "astrid" to avoid gfm fussing at him/her.
• Stop the spectrometer manager, change the environment to write to local disk and start a temporary spectrometer manager:

• Now take the data (data will be in /export/home/earth/scratch/gbtdata//Spectrometer)
• When done taking data return the spectrometer to its normal state using:

• Have the operator restart sdfits
• At next opportunity when spectrometer is not being used move data to /home/gbtdata/
• Now you can run sdfits to fill the data as if it was taking normally

Maximum Scan Length / File Size Limit

It turns out that the maximum file size that can be written by the M&C system is 2.0 GB. This is likely a limit within cfitsio. For 200 MHz bandwidth, 8192 channels, 2 polarizations using only Bank A of the Spectrometer, this limit is reached in about 54 minutes. Scans should not be longer than about 50 minutes in this mode.

Test Observations

Test observations were made during a maintenance day on January 9, 2007. The data are located in /home/gbtdata/TGBT07A_010_01. First I tested what the minimum legal integration time would be and found it to be 0.4 seconds. I then did several 5 minute scans (Tracks) and observed less than 2 seconds of latency. I then did a 30 minute scan and observed 4 seconds of latency. I then tried to do a 60 minute scan but had it abort just shy of 55 minutes into the scan because the spectrometer had a fault in writing the data to disk. This turned out to be because the file size become larger than 2.0 GB.

Time To Copy Data

It took about 8 minutes to copy about 3.0 GB of data from the local disk on earth to /home/gbtdata.

Overhead Time

I would suggest leaving about 15 minutes of time at the start of the observations for setting up the spectrometer to write to the local disk. I would allow another 15 minutes at the end of the observing to change the spectrometer back to its normal setup.

Disk Space Available / Total Observing Time To Fill Disk

The local disk has about 90 GB of free space. For the tested observing mode this allows for about 40 hours of observing data to be written to the disk.

-- ToneyMinter - 10 Jan 2007