[Previous entry: "WWVB morning phase jump"] [Main Index] [Next entry: "Continuing 8164 tweaks"]

10/02/2004 Entry: "Tweaking the Spectracom 8164"

I'm feeding the Spectracom 8164 WWVB receiver into my main ntp time server, which also gets time information from the Z3801A GPSDO. The Z3801A provides the PPS signal, while the Spectracom uses its ASCII time code. There is an offset between the Spectracom's 1pps signal and the ASCII time mark, at least as ntpd interprets it. I'm trying to figure out what that offset is.

The distance between WWVB and my site is 1100.25 miles (1770.67km) and the calculated path delay, based on the formula in the Spectracom manual, is 5.90ms. The receiver's internal delay is nominally 17ms. So, the receiver's delay should be set to 22.9ms.

With that setting, the 1pps output from the Spectracom currently lags the 1pps output from the Z3801A by about 600us. According to my current ntp stats plot (www.febo.com/stats/ntp/), ntpd thinks the Spectracom is about 2.4ms slow.

The Spectracom adjusts its time sync in 100us increments, and unless it loses lock it stays within 1ms but seems to sometimes ramp through the +/-500us range. I think the only way to determine the true offset is to log the maximum and minimum offsets of the Spectracom 1pps against the Z3801A 1pps long enough to see the full range, and then pick the value in the middle. I'll start that experiment today.