PSN-L Email List Message

Subject: Re: Computer Timing Problems / Solutions
From: Larry Cochrane lcochrane@..............
Date: Sat, 16 Apr 2005 17:21:07 -0700


Ian wrote:
> Regarding the standard a/d board and without knowing its details, I note 
> that there is a crystal onboard and that the spec of the a/d is 100 
> samples/second.  What would be the accuracy obtained for the time 
> between the P and S wave by simply counting samples between them and 
> multiplying that by the sample period?  Would this suffice?  The flip 
> side of this question is: if that crystal isn't very good, then is all 
> this for naught!?

I assume you are taking about my A/D board... I check the frequency of the 4 Mhz 
oscillator used on my A/D board before I solder it to the board. If it's over/under 
+-75 Hz I do not use it. With a good timing reference, the A/D board can correct for 
the oscillator being off frequency by adding or subtracting time to the time 
accumulator at some time interval. This process compensates for the oscillator 
running a little higher or lower in frequency. For all of this to work correctly the 
time reference must be stable. GPS is the best since the 1PPS signal is accurate to a 
few microseconds. With WWV/WWVB or using the PC's time it's harder for the board to 
do this correctly since the time reference can jump around a few milliseconds or more.

> Thinking more about this board, is the sample data buffered (other than 
> that provided by the uart)?  Are samples lost/overwritten if the pc is 
> temporarily busy?

The board buffers the data for one second and then sends it out to WinSDR. The packet 
is also saved in memory so WinSDR can request a retransmission if the data never gets 
to the PC. The A/D board can save about 30 seconds worth of data that can be resent 
to the host.

> Also, after last week's discussion, was the conclusion that the accuracy 
> which can be achieved by a good NTP service sufficient for absolute 
> timing and that the emphasis should be on the relative time between the 
> 2 waves.  Again, what is a reasonable spec?

Here's my .02 cents on this. If you are only going to record teleseismic events, 
anything under 100 milliseconds is probably an overkill. If you are going to record 
local events the time should be accurate to within one or two sample periods. If you 
are running at 100 samples per second this is 10 to 20 ms.

Larry Cochrane
Redwood City, PSN


Public Seismic Network Mailing List (PSN-L)

[ Top ] [ Back ] [ Home Page ]