|
Re: Quick and dirty GPSDO build by F2DC and me.
A graph showing how the set frequency is corrected over time to keep the output at 10MHz
The measurement was started 30 seconds after power on. After roughly 30 minutes the TCXO seems to become stable but all variations after 7 minutes are already below 1e-8 The delta count measured for 1e+9 pulses after 30 minutes is maximum 2 so it may be good to do a test with a order higher count -- HBTE Files section:?/g/HBTE/files Erik, PD0EK |
|
Re: Quick and dirty GPSDO build by F2DC and me.
To understand the impact of temperature fluctuations the simulation is extended to have? a perioding temperature change causing a 2 Hz temperature fluctuation over a 1.6 hour period. The feedback loop was made adaptive, e.g. the measurement time is reduced during a lot of drift and increased when no drift to improve reaction speed and measurement accuracy
The fluctuations are reduced with a factor of 10, to +/- 0.1Hz using a reference frequency of 45MHz so 2e-9 accuracy. Adding feed forward to compensate during the (more or less) stable drift periods the fluctuations are somewhat reduced but the lack of data (each measurement period ranging from 1 to 128 seconds observes less than 10 correction pulses) makes an effective drift prediction and correction difficult ?Notice the time scale has increased as the reduction of error (e.g. a prediction of the drift) has enabled in average longer measurement periods To reach 0.01Hz stability (well below 1e-9 error) the fluctuating drift has to be less than 0.1Hz and the period of the drift preferably above 10 hours otherwise a 45MHz clock and below 400seconds measurement time will not be able to measure sufficient data to counter the drift. -- HBTE Files section:?/g/HBTE/files Erik, PD0EK |
|
Re: Quick and dirty GPSDO build by F2DC and me.
To better understand I did some simulations.
Once the measurement time is above 10 seconds the 100ns PPS jitter is no longer relevant given the measurement time and measurement frequency. Drift is a problem as the current feedback loop has no feed forward to compensate for a constant drift so the drift leads to a frequency error In the simulation I have been able to add the feed forward and the drift can be compensated. Its clear that temperature stability will be the main factor determining the frequency accuracy as the feed forward to compensate for the drift requires some measurements to measure the drift and if the drift changes a frequency error will persist In below chart the simulation tries to estimate the drift and correct for it but is unable to calculate the exact drift as the counting of the frequency and the correction of the frequency has limited accuracy. Delta is in Hz, Delta*100 is in 0.01Hz.? -- HBTE Files section:?/g/HBTE/files Erik, PD0EK |
|
Re: Quick and dirty GPSDO build by F2DC and me.
¿ªÔÆÌåÓýThe jitter in the 1PPS signal depends on the GPS receiver, an older unit from 10 years back may have around 100nS jitter, a recent unit around 20-30nS.? Receivers are of two kinds position finding ¨C optimised for fast location? and? timing which are optimised for timing accuracy after they have surveyed a fixed position. ?A timing grade GNNS receiver can offer 1-3nS precision after processing. ? Nearly all cheap modules are position finding, they have higher jitter but are usable for home use. ? Any GPSDO sits between two limits, the 1PPS has very high 24 hour accuracy ?( parts in 10E-13 )but has some second to second jitter. A crystal oscillator? may have very good stability over 1 second but wonder many parts per million over a day or much less. So a GPSDO has to ?filter the intrinsic drift of the crystal oscillator? by using the precise but jittery GPS 1PPS. ?Note that the jitter is not Gaussian noise but ?mainly a periodic fluctuation due to the receiver processing . Ionospheric noise introduces around 1ns noise. ? For any crystal oscillator there is? an optimum? ¡°cross over¡± frequency? for the control loop to minimise jitter in the GPSDO output. A good commercial GPSDO can use a time constant of 400-800S with a high grade ovenised oscillator. ? An ambient temperature crystal is the worst case.? Shielding the crystal from air currents helps as does increasing the thermal mass. A TCXO ( temperature compensated ) is a bad choice in some cases as some modern units use stepped compensation. The best choice is an ovened crystal where the crystal is heated to a constant temperature which can be adjusted to an optimum temperature for that crystal.?? ? An ovenised crystal will use 1-10 W of energy all of the time? but it should have very good stability allowing heavy filtering of GPS jitter. Digital division and counting with a frequency locked loop? is the worst possible method as it cannot track and correct continuously, only allow the error to accumulate then make a step change after a long period.? Although it is akin to a huff and puff the H&P corrects rapidly so the action appears continuous whereas frequency locking with very long periods leads to discrete jumps in frequency. ? An optimum method counts oscillator cycles during 1 sec periods for rough correction during initialisation then integrates the phase difference every second. It is not necessarily complex, a popular design uses two cheap IC¡¯s and an Arduino nano .? Lars GPSDO can be found on Github. ? 73 Alan G8LCO ? ? ? Sent from for Windows 10 ? Groups.io Links: You receive all messages sent to this group. View/Reply Online (#1664) | Reply To Group | Reply To Sender | Mute This Topic | New Topic THIS GROUP: <groups.io/g/HBTE> RELATED GROUP: <groups.io.g.BITX20> Your Subscription | Contact Group Owner | Unsubscribe [g8lco1@...] _._,_._,_ ? |