Luke, Seems that you might indeed become a time nut, though most of those on the list are focused on atomic clock topics and the time precision that atomic clocks provide. That said, sounds like you have a good start. While I have not used the GPS module you mention, GPS timing sources typically show timing accuracies of 20 to 100 ns. The short term noise arises from the signal-to-noise of the "as received" signals, so the 99% accuracy of 60 ns likely means that you can assume that all of the pulses are within +/-100 ns. The time noise is "white" for GPS signals. Note also that this is time accuracy, not frequency accuracy. Frequency accuracy is maintained by GPS Ground Control Segment with reference to the US Naval Observatory Time Scale. At the risk of unraveling your existing measurement architecture, you might consider measuring the clock phase (pendulum passing) with respect to the GPS pulse. This would be a time interval measurement with an accurate (GPS) reference pulse and a time interval based upon your TCXO. This reduces the burden of calibrating the TCXO and will eliminate concerns about the frequency accuracy and stability of the TCXO. For example, if your TCXO is off frequency by 1 ppm, you would get a non-cumulative error of 1 ppm in the time interval measurement rather than a 1 ppm error in the clock rate. The 1 ppm TCXO error drops out of the solution for rate of the clock; only changes in the TCXO frequency enter and they are typically much smaller.
Hope this is useful; others on the list will likely have other input. Mike -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Luke Mester Sent: Sunday, December 01, 2013 2:54 AM To: [email protected] Subject: [time-nuts] Using a UBlox NEO-6 GPS module for calibrating a PIC microprocessor based timer. One of my hobbies is collecting and repairing mechanical clocks. I was looking at buying one of the specialized electronic timers used to measure the performance of mechanical clocks. I really couldn't justify the cost just for hobby use. Since I have electronics and programming skills I decided to build my own timer using a PIC chip. This became a much bigger project than I expected! I have my clock timer running and have most of the software features that I need working. I then realized that I need some way to calibrate it and verify it's accuracy. I didn't have any source of accurate time available. After searching the internet and finding this mailing list I decided to try a GPS module. I bought a $20 module from DX.com. It has a built in antenna, voltage regulator, serial interface and most important, a 1 PPS output.The GPS is a UBlox NEO-6M. After reading the specs on this module I see that they claim a 99% accuracy of <60ns for the time pulse signal. What does this mean? What about the other 1%? How much variation can the time pulse have? If it's really 60ns it's much better than I need. I'm hoping that some of the time experts on this mailing list can give me some idea what to expect from this GPS module. Also, if there are any settings that I should change to get better timing performance. There are a huge number of settings available when I run it's configuration program. I have no idea what most of them do. I'm using one of the hardware timers on the PIC chip to measure the time interval. The PIC is running with a 100ns (10MHz) instruction cycle. The timer will provide 100ns resolution. I'm getting occasional variations of about a microsecond. Because I'm using interrupt driven code to capture the timer value there will be some unavoidable jitter in the timing. I was expecting about 4 or 5 cycles (400ns - 500ns) but I'm getting more than twice that. Is it safe to assume that these are due to problems in my hardware or software? Is this from variations in the GPS PPS output? Maybe I'm just not interpreting the data correctly. Below are links to some data plots: Four plots are shown. The first two are the Rate and Beat error that my timer reports while monitoring the GPS PPS signal. Rate is normally the average of two beats ( two time interval samples). If a clock is not in beat (the tick and tock take different amounts of time) the displayed rate would jump back and forth. Averaging two beats eliminates this jump. I have disabled this average in my code so that the rate is now showing each beat and not the average of two. I turned it off because I expect that this averaging could hide possible problems with my timer. Beat error is the difference between the two beats. This shows the rate change for each pair of beats. This is needed so that you can get the clock pendulum or balance wheel adjusted properly. Raw Data <http://mesterhome.com/clock/data/RawData.png> Average Data <http://mesterhome.com/clock/data/AveData.png> Average data has been filtered with a 100 sample running average. The plot looks really good. The average is just hiding the instability. I also noticed variations that appeared to be due to temperature changes. I borrowed a temperature data logger from work and did some testing. The temperature and rate graphs track perfectly. I can see my furnace cycling and my programmable thermostat moving the temperature setting up and down! That got me interested in trying a TCXO instead of the standard crystal that I was using. A $3.00 TCXO from EBay made a huge difference! Both of these plots have the running average applied. You can't see the temperature changes with the raw data. Crystal <http://mesterhome.com/clock/data/RTvTP.png> TCXO <http://mesterhome.com/clock/data/RTvTPTC.png> In case anyone is interested, here is a link to a data file captured from the clock timer. It's in CSV format. The first column labeled "Rate" is the time for each beat of the clock. "Rate Avg" is a running average rate and "Beat E" the beat error. Data file <http://mesterhome.com/clock/data/tcxo.csv> Finally, I think I might be turning into a time nut! For the clocks that I work with this timer is already far better than I need. Millisecond accuracy Is good enough to test most mechanical clocks. Microsecond is great! I know that a microprocessor based timer is capable of better performance. I then had the problem of what I could use to measure the performance of my timer. I Needed a better clock than my timer. Now I'm wondering if this cheap Chinese GPS is good enough. I'm having fun tweaking the hardware and software to see just how good I can get it to perform! _______________________________________________ time-nuts mailing list -- [email protected] To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there. _______________________________________________ time-nuts mailing list -- [email protected] To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.
