Re: [time-nuts] Training period for a Rb clock using GPS
Hi Hal, Thanks so much for the detailed post. I have a follow up question: What is the relationship between the training time and the appropriate value of the time constant (currently set at 18 hours)? The time constant isn't the size of a moving average window is it? Thanks again for your help. We are a bit clueless here but trying to learn... =Abhay On Wed, Jun 2, 2010 at 2:02 AM, Hal Murray hmur...@megapathdsl.net wrote: par...@berkeley.edu said: I am a newbie at this, but have been playing around with 2 prs10s. For our application we need to run the clocks without gps, but we do get to sync it to gps *initially* for as long as we want. However, what we've noticed is that when we train it for short periods of time ( 1 hour a day) the clock drifts for a few microseconds a day once we've disconnected gps, but when we train it for say 12 hours, its drift seems to be much less (sub sub microsecond/day). We were wondering why this should be so! Look at it the other way. How long should it take to train it? Let's use rough numbers. There are 1E5 seconds per day. Your few microseconds is 1E-6 seconds. That's an accuracy of 1 part in 1E11. Your sub-sub is 1/10 microsecond or 1E-7 seconds. So that's an accuracy of 1 part in 1E12. The data sheet says: Aging (after 30 days) 5E-11 (monthly) 5E-11 is 50E-12, so that's 2E-12 per day which is what you saw. The data sheet also says: The PRS10 can time-tag an external 1 pps input with 1 ns resolution. These values may be reported back via RS-232, or used to phase-lock the unit to an external reference (such as GPS) with time constants of several hours. There are 4E3 seconds in an hour and 1E9 nanoseconds per second. So in an hour, you can get close to 1 part in 1E12. But that's assuming that the input PPS signal is right-on. There are two types of GPS receivers. Most use a free running clock and generate the PPS pulse with the closest clock edge. They typically have noise on the order of 15-50 ns. Fancy ones will tell you how far off they think it is. The really fancy ones will have a VCXO so they can slew the clock to the right offset. One magic word is hanging bridges. It comes up in discussions occasionally. For lots of info on that area: http://www.gpstime.com/files/PTTI/PTTI_2006.pdf 31 pages, lots of good stuff, aka time sink. More here: http://www.leapsecond.com/pages/vp/heater.htm 2 or 3 screens, good stuff, a quick read. So with only an hour, it's not unreasonable that you are off by a factor of 10, but you might have to get unlucky for a hanging bridge to get you. But there is another factor to consider. What sort of filter is the software using between the PPS input and the knob that adjusts the frequency? More from the data sheet: When tracking an external input, the time constant can be set from 5 minutes to 18 hours. I think the manual says the default is 65K seconds. That's 18 hours. Unless you changed it, that explains why 1 hour wasn't enough. It might get better if you give it more time and/or tweak the time constant if you can only get 12 hours. -- These are my opinions, not necessarily my employer's. I hate spam. ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there. ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.
Re: [time-nuts] Training period for a Rb clock using GPS
Hi If you have an 18 hour time constant you would need a training period of 5 to 10 X 18 hours to get the system to settle. For a one hour training period the time constant should be in the 5 to 10 minute range. Bob -Original Message- From: time-nuts-boun...@febo.com [mailto:time-nuts-boun...@febo.com] On Behalf Of Abhay Parekh Sent: Thursday, June 03, 2010 12:02 PM To: Discussion of precise time and frequency measurement Subject: Re: [time-nuts] Training period for a Rb clock using GPS Hi Hal, Thanks so much for the detailed post. I have a follow up question: What is the relationship between the training time and the appropriate value of the time constant (currently set at 18 hours)? The time constant isn't the size of a moving average window is it? Thanks again for your help. We are a bit clueless here but trying to learn... =Abhay On Wed, Jun 2, 2010 at 2:02 AM, Hal Murray hmur...@megapathdsl.net wrote: par...@berkeley.edu said: I am a newbie at this, but have been playing around with 2 prs10s. For our application we need to run the clocks without gps, but we do get to sync it to gps *initially* for as long as we want. However, what we've noticed is that when we train it for short periods of time ( 1 hour a day) the clock drifts for a few microseconds a day once we've disconnected gps, but when we train it for say 12 hours, its drift seems to be much less (sub sub microsecond/day). We were wondering why this should be so! Look at it the other way. How long should it take to train it? Let's use rough numbers. There are 1E5 seconds per day. Your few microseconds is 1E-6 seconds. That's an accuracy of 1 part in 1E11. Your sub-sub is 1/10 microsecond or 1E-7 seconds. So that's an accuracy of 1 part in 1E12. The data sheet says: Aging (after 30 days) 5E-11 (monthly) 5E-11 is 50E-12, so that's 2E-12 per day which is what you saw. The data sheet also says: The PRS10 can time-tag an external 1 pps input with 1 ns resolution. These values may be reported back via RS-232, or used to phase-lock the unit to an external reference (such as GPS) with time constants of several hours. There are 4E3 seconds in an hour and 1E9 nanoseconds per second. So in an hour, you can get close to 1 part in 1E12. But that's assuming that the input PPS signal is right-on. There are two types of GPS receivers. Most use a free running clock and generate the PPS pulse with the closest clock edge. They typically have noise on the order of 15-50 ns. Fancy ones will tell you how far off they think it is. The really fancy ones will have a VCXO so they can slew the clock to the right offset. One magic word is hanging bridges. It comes up in discussions occasionally. For lots of info on that area: http://www.gpstime.com/files/PTTI/PTTI_2006.pdf 31 pages, lots of good stuff, aka time sink. More here: http://www.leapsecond.com/pages/vp/heater.htm 2 or 3 screens, good stuff, a quick read. So with only an hour, it's not unreasonable that you are off by a factor of 10, but you might have to get unlucky for a hanging bridge to get you. But there is another factor to consider. What sort of filter is the software using between the PPS input and the knob that adjusts the frequency? More from the data sheet: When tracking an external input, the time constant can be set from 5 minutes to 18 hours. I think the manual says the default is 65K seconds. That's 18 hours. Unless you changed it, that explains why 1 hour wasn't enough. It might get better if you give it more time and/or tweak the time constant if you can only get 12 hours. -- These are my opinions, not necessarily my employer's. I hate spam. ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there. ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there. ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.
Re: [time-nuts] Training period for a Rb clock using GPS
Hi, the simple stupid answer is: 24/365 Ernie. -Original Message- From: Bob Camp li...@rtty.us To: 'Discussion of precise time and frequency measurement' time-nuts@febo.com Sent: Thu, Jun 3, 2010 6:07 pm Subject: Re: [time-nuts] Training period for a Rb clock using GPS Hi If you have an 18 hour time constant you would need a training period of 5 o 10 X 18 hours to get the system to settle. For a one hour training period the time constant should be in the 5 to 10 inute range. Bob -Original Message- rom: time-nuts-boun...@febo.com [mailto:time-nuts-boun...@febo.com] On ehalf Of Abhay Parekh ent: Thursday, June 03, 2010 12:02 PM o: Discussion of precise time and frequency measurement ubject: Re: [time-nuts] Training period for a Rb clock using GPS Hi Hal, hanks so much for the detailed post. I have a follow up question: What is he relationship between he training time and the appropriate value of the time constant (currently et at 18 hours)? The time constant isn't the size of moving average window is it? hanks again for your help. We are a bit clueless here but trying to earn... Abhay n Wed, Jun 2, 2010 at 2:02 AM, Hal Murray hmur...@megapathdsl.net wrote: par...@berkeley.edu said: I am a newbie at this, but have been playing around with 2 prs10s. For our application we need to run the clocks without gps, but we do get to sync it to gps *initially* for as long as we want. However, what we've noticed s that when we train it for short periods of time ( 1 hour a day) the clock drifts for a few microseconds a day once we've disconnected gps, but hen we train it for say 12 hours, its drift seems to be much less (sub sub microsecond/day). We were wondering why this should be so! Look at it the other way. How long should it take to train it? Let's use rough numbers. There are 1E5 seconds per day. Your few microseconds is 1E-6 seconds. That's an accuracy of 1 part in 1E11. Your sub-sub is 1/10 microsecond or 1E-7 seconds. So that's an accuracy of 1 part in 1E12. The data sheet says: Aging (after 30 days) 5E-11 (monthly) 5E-11 is 50E-12, so that's 2E-12 per day which is what you saw. The data sheet also says: The PRS10 can time-tag an external 1 pps input with 1 ns resolution. These values may be reported back via RS-232, or used to phase-lock the unit to an external reference (such as GPS) with time constants of several hours. There are 4E3 seconds in an hour and 1E9 nanoseconds per second. So in an hour, you can get close to 1 part in 1E12. But that's assuming that the input PPS signal is right-on. There are two types of GPS receivers. Most use a free running clock and generate the PPS pulse with the closest clock edge. They typically have noise on the order of 15-50 ns. Fancy ones will tell you how far off they think it is. The really fancy ones will have a VCXO so they can slew the clock to the right offset. One magic word is hanging bridges. It comes up in discussions occasionally. For lots of info on that area: http://www.gpstime.com/files/PTTI/PTTI_2006.pdf 31 pages, lots of good stuff, aka time sink. More here: http://www.leapsecond.com/pages/vp/heater.htm 2 or 3 screens, good stuff, a quick read. So with only an hour, it's not unreasonable that you are off by a factor f 10, but you might have to get unlucky for a hanging bridge to get you. But there is another factor to consider. What sort of filter is the software using between the PPS input and the knob that adjusts the frequency? More from the data sheet: When tracking an external input, the time constant can be set from 5 minutes to 18 hours. I think the manual says the default is 65K seconds. That's 18 hours. Unless you changed it, that explains why 1 hour wasn't enough. It might get better if you give it more time and/or tweak the time constant if you can only et 12 hours. -- These are my opinions, not necessarily my employer's. I hate spam. ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there. __ ime-nuts mailing list -- time-nuts@febo.com o unsubscribe, go to ttps://www.febo.com/cgi-bin/mailman/listinfo/time-nuts nd follow the instructions there. __ ime-nuts mailing list -- time-nuts@febo.com o unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts nd follow the instructions there. ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.
Re: [time-nuts] Training period for a Rb clock using GPS
Hi That will give you the best answer with a simple loop. The problem is that best may not be good enough to actually get your Rb on time / on frequency. Something more sophisticated than a simple loop may be needed. Bob -Original Message- From: time-nuts-boun...@febo.com [mailto:time-nuts-boun...@febo.com] On Behalf Of Abhay Parekh Sent: Thursday, June 03, 2010 12:28 PM To: Discussion of precise time and frequency measurement Subject: Re: [time-nuts] Training period for a Rb clock using GPS Ok, great. So if we can train for h hours we should set the time constant somewhere between h/10 and h/5. It would be safer to pick something closer to h/10 since when the clock powers up it might start in the wrong place so a smaller value helps the clock move quickly into the right area, but h/5 will act as a better buffer against hanging bridges. Is my reasoning correct? Thanks =Abhay On Thu, Jun 3, 2010 at 9:07 AM, Bob Camp li...@rtty.us wrote: Hi If you have an 18 hour time constant you would need a training period of 5 to 10 X 18 hours to get the system to settle. For a one hour training period the time constant should be in the 5 to 10 minute range. Bob -Original Message- From: time-nuts-boun...@febo.com [mailto:time-nuts-boun...@febo.com] On Behalf Of Abhay Parekh Sent: Thursday, June 03, 2010 12:02 PM To: Discussion of precise time and frequency measurement Subject: Re: [time-nuts] Training period for a Rb clock using GPS Hi Hal, Thanks so much for the detailed post. I have a follow up question: What is the relationship between the training time and the appropriate value of the time constant (currently set at 18 hours)? The time constant isn't the size of a moving average window is it? Thanks again for your help. We are a bit clueless here but trying to learn... =Abhay On Wed, Jun 2, 2010 at 2:02 AM, Hal Murray hmur...@megapathdsl.net wrote: par...@berkeley.edu said: I am a newbie at this, but have been playing around with 2 prs10s. For our application we need to run the clocks without gps, but we do get to sync it to gps *initially* for as long as we want. However, what we've noticed is that when we train it for short periods of time ( 1 hour a day) the clock drifts for a few microseconds a day once we've disconnected gps, but when we train it for say 12 hours, its drift seems to be much less (sub sub microsecond/day). We were wondering why this should be so! Look at it the other way. How long should it take to train it? Let's use rough numbers. There are 1E5 seconds per day. Your few microseconds is 1E-6 seconds. That's an accuracy of 1 part in 1E11. Your sub-sub is 1/10 microsecond or 1E-7 seconds. So that's an accuracy of 1 part in 1E12. The data sheet says: Aging (after 30 days) 5E-11 (monthly) 5E-11 is 50E-12, so that's 2E-12 per day which is what you saw. The data sheet also says: The PRS10 can time-tag an external 1 pps input with 1 ns resolution. These values may be reported back via RS-232, or used to phase-lock the unit to an external reference (such as GPS) with time constants of several hours. There are 4E3 seconds in an hour and 1E9 nanoseconds per second. So in an hour, you can get close to 1 part in 1E12. But that's assuming that the input PPS signal is right-on. There are two types of GPS receivers. Most use a free running clock and generate the PPS pulse with the closest clock edge. They typically have noise on the order of 15-50 ns. Fancy ones will tell you how far off they think it is. The really fancy ones will have a VCXO so they can slew the clock to the right offset. One magic word is hanging bridges. It comes up in discussions occasionally. For lots of info on that area: http://www.gpstime.com/files/PTTI/PTTI_2006.pdf 31 pages, lots of good stuff, aka time sink. More here: http://www.leapsecond.com/pages/vp/heater.htm 2 or 3 screens, good stuff, a quick read. So with only an hour, it's not unreasonable that you are off by a factor of 10, but you might have to get unlucky for a hanging bridge to get you. But there is another factor to consider. What sort of filter is the software using between the PPS input and the knob that adjusts the frequency? More from the data sheet: When tracking an external input, the time constant can be set from 5 minutes to 18 hours. I think the manual says the default is 65K seconds. That's 18 hours. Unless you changed it, that explains why 1 hour wasn't enough. It might get better if you give it more time and/or tweak the time constant if you can only get 12 hours. -- These are my opinions, not necessarily my employer's. I hate spam. ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo
Re: [time-nuts] Training period for a Rb clock using GPS
Yes, that makes sense. I think that we can arrange things so that we train for 10-12 hours. Do you not think that that is a long enough time for a single loop to be effective? Thanks again! =Abhay On Thu, Jun 3, 2010 at 9:37 AM, Bob Camp li...@rtty.us wrote: Hi That will give you the best answer with a simple loop. The problem is that best may not be good enough to actually get your Rb on time / on frequency. Something more sophisticated than a simple loop may be needed. Bob -Original Message- From: time-nuts-boun...@febo.com [mailto:time-nuts-boun...@febo.com] On Behalf Of Abhay Parekh Sent: Thursday, June 03, 2010 12:28 PM To: Discussion of precise time and frequency measurement Subject: Re: [time-nuts] Training period for a Rb clock using GPS Ok, great. So if we can train for h hours we should set the time constant somewhere between h/10 and h/5. It would be safer to pick something closer to h/10 since when the clock powers up it might start in the wrong place so a smaller value helps the clock move quickly into the right area, but h/5 will act as a better buffer against hanging bridges. Is my reasoning correct? Thanks =Abhay On Thu, Jun 3, 2010 at 9:07 AM, Bob Camp li...@rtty.us wrote: Hi If you have an 18 hour time constant you would need a training period of 5 to 10 X 18 hours to get the system to settle. For a one hour training period the time constant should be in the 5 to 10 minute range. Bob -Original Message- From: time-nuts-boun...@febo.com [mailto:time-nuts-boun...@febo.com] On Behalf Of Abhay Parekh Sent: Thursday, June 03, 2010 12:02 PM To: Discussion of precise time and frequency measurement Subject: Re: [time-nuts] Training period for a Rb clock using GPS Hi Hal, Thanks so much for the detailed post. I have a follow up question: What is the relationship between the training time and the appropriate value of the time constant (currently set at 18 hours)? The time constant isn't the size of a moving average window is it? Thanks again for your help. We are a bit clueless here but trying to learn... =Abhay On Wed, Jun 2, 2010 at 2:02 AM, Hal Murray hmur...@megapathdsl.net wrote: par...@berkeley.edu said: I am a newbie at this, but have been playing around with 2 prs10s. For our application we need to run the clocks without gps, but we do get to sync it to gps *initially* for as long as we want. However, what we've noticed is that when we train it for short periods of time ( 1 hour a day) the clock drifts for a few microseconds a day once we've disconnected gps, but when we train it for say 12 hours, its drift seems to be much less (sub sub microsecond/day). We were wondering why this should be so! Look at it the other way. How long should it take to train it? Let's use rough numbers. There are 1E5 seconds per day. Your few microseconds is 1E-6 seconds. That's an accuracy of 1 part in 1E11. Your sub-sub is 1/10 microsecond or 1E-7 seconds. So that's an accuracy of 1 part in 1E12. The data sheet says: Aging (after 30 days) 5E-11 (monthly) 5E-11 is 50E-12, so that's 2E-12 per day which is what you saw. The data sheet also says: The PRS10 can time-tag an external 1 pps input with 1 ns resolution. These values may be reported back via RS-232, or used to phase-lock the unit to an external reference (such as GPS) with time constants of several hours. There are 4E3 seconds in an hour and 1E9 nanoseconds per second. So in an hour, you can get close to 1 part in 1E12. But that's assuming that the input PPS signal is right-on. There are two types of GPS receivers. Most use a free running clock and generate the PPS pulse with the closest clock edge. They typically have noise on the order of 15-50 ns. Fancy ones will tell you how far off they think it is. The really fancy ones will have a VCXO so they can slew the clock to the right offset. One magic word is hanging bridges. It comes up in discussions occasionally. For lots of info on that area: http://www.gpstime.com/files/PTTI/PTTI_2006.pdf 31 pages, lots of good stuff, aka time sink. More here: http://www.leapsecond.com/pages/vp/heater.htm 2 or 3 screens, good stuff, a quick read. So with only an hour, it's not unreasonable that you are off by a factor of 10, but you might have to get unlucky for a hanging bridge to get you. But there is another factor to consider. What sort of filter is the software using between the PPS input and the knob that adjusts the frequency? More from the data sheet: When tracking an external input, the time constant can be set from 5 minutes to 18 hours. I think the manual says the default is 65K seconds. That's 18 hours. Unless you changed
Re: [time-nuts] Training period for a Rb clock using GPS
Hi Your loop for 10 hours would be around 1 or 2 hours. That's 60 X 60 X (1 or 2) seconds = 3,600 to 7,200 seconds. If GPS is good to +/- 20 ns out of your receiver in your location then you would get 20 x 10^-9 / (3600 or 7200) = 2.7 to 5.5 X 10^-12 inside the loop. The Rb should be below that level over the same time period. Simple answer - yes it should be good enough. Bob -Original Message- From: time-nuts-boun...@febo.com [mailto:time-nuts-boun...@febo.com] On Behalf Of Abhay Parekh Sent: Thursday, June 03, 2010 12:46 PM To: Discussion of precise time and frequency measurement Subject: Re: [time-nuts] Training period for a Rb clock using GPS Yes, that makes sense. I think that we can arrange things so that we train for 10-12 hours. Do you not think that that is a long enough time for a single loop to be effective? Thanks again! =Abhay On Thu, Jun 3, 2010 at 9:37 AM, Bob Camp li...@rtty.us wrote: Hi That will give you the best answer with a simple loop. The problem is that best may not be good enough to actually get your Rb on time / on frequency. Something more sophisticated than a simple loop may be needed. Bob -Original Message- From: time-nuts-boun...@febo.com [mailto:time-nuts-boun...@febo.com] On Behalf Of Abhay Parekh Sent: Thursday, June 03, 2010 12:28 PM To: Discussion of precise time and frequency measurement Subject: Re: [time-nuts] Training period for a Rb clock using GPS Ok, great. So if we can train for h hours we should set the time constant somewhere between h/10 and h/5. It would be safer to pick something closer to h/10 since when the clock powers up it might start in the wrong place so a smaller value helps the clock move quickly into the right area, but h/5 will act as a better buffer against hanging bridges. Is my reasoning correct? Thanks =Abhay On Thu, Jun 3, 2010 at 9:07 AM, Bob Camp li...@rtty.us wrote: Hi If you have an 18 hour time constant you would need a training period of 5 to 10 X 18 hours to get the system to settle. For a one hour training period the time constant should be in the 5 to 10 minute range. Bob -Original Message- From: time-nuts-boun...@febo.com [mailto:time-nuts-boun...@febo.com] On Behalf Of Abhay Parekh Sent: Thursday, June 03, 2010 12:02 PM To: Discussion of precise time and frequency measurement Subject: Re: [time-nuts] Training period for a Rb clock using GPS Hi Hal, Thanks so much for the detailed post. I have a follow up question: What is the relationship between the training time and the appropriate value of the time constant (currently set at 18 hours)? The time constant isn't the size of a moving average window is it? Thanks again for your help. We are a bit clueless here but trying to learn... =Abhay On Wed, Jun 2, 2010 at 2:02 AM, Hal Murray hmur...@megapathdsl.net wrote: par...@berkeley.edu said: I am a newbie at this, but have been playing around with 2 prs10s. For our application we need to run the clocks without gps, but we do get to sync it to gps *initially* for as long as we want. However, what we've noticed is that when we train it for short periods of time ( 1 hour a day) the clock drifts for a few microseconds a day once we've disconnected gps, but when we train it for say 12 hours, its drift seems to be much less (sub sub microsecond/day). We were wondering why this should be so! Look at it the other way. How long should it take to train it? Let's use rough numbers. There are 1E5 seconds per day. Your few microseconds is 1E-6 seconds. That's an accuracy of 1 part in 1E11. Your sub-sub is 1/10 microsecond or 1E-7 seconds. So that's an accuracy of 1 part in 1E12. The data sheet says: Aging (after 30 days) 5E-11 (monthly) 5E-11 is 50E-12, so that's 2E-12 per day which is what you saw. The data sheet also says: The PRS10 can time-tag an external 1 pps input with 1 ns resolution. These values may be reported back via RS-232, or used to phase-lock the unit to an external reference (such as GPS) with time constants of several hours. There are 4E3 seconds in an hour and 1E9 nanoseconds per second. So in an hour, you can get close to 1 part in 1E12. But that's assuming that the input PPS signal is right-on. There are two types of GPS receivers. Most use a free running clock and generate the PPS pulse with the closest clock edge. They typically have noise on the order of 15-50 ns. Fancy ones will tell you how far off they think it is. The really fancy ones will have a VCXO so they can slew the clock to the right offset. One magic word is hanging bridges. It comes up in discussions occasionally. For lots of info on that area: http://www.gpstime.com/files/PTTI/PTTI_2006.pdf 31 pages, lots of good stuff, aka time sink
Re: [time-nuts] Training period for a Rb clock using GPS
We have noticed two things with the PRS10s and other Rbs. They need about 24hrs to settle before they really start to perform well and if there are any significant temperature swings, expect them to react to it. We have a PRS10 in a Meinberg M900 that takes about 1 day to recover from 2-3C temp swing in a datacenter. It will be off (as compared to the other 1-PPS that are not in that room) by about 50-100ns. Normally it's within +-10ns when compared to the other GPSDOs. We noticed this happened when the air-handlers would alternate on about a 2-week cycle. Scott On 06/03/2010 12:09 PM, Abhay Parekh wrote: Awesome. Thanks so much! =Abhay On Thu, Jun 3, 2010 at 10:07 AM, Bob Campli...@rtty.us wrote: Hi Your loop for 10 hours would be around 1 or 2 hours. That's 60 X 60 X (1 or 2) seconds = 3,600 to 7,200 seconds. If GPS is good to +/- 20 ns out of your receiver in your location then you would get 20 x 10^-9 / (3600 or 7200) = 2.7 to 5.5 X 10^-12 inside the loop. The Rb should be below that level over the same time period. Simple answer - yes it should be good enough. Bob -Original Message- From: time-nuts-boun...@febo.com [mailto:time-nuts-boun...@febo.com] On Behalf Of Abhay Parekh Sent: Thursday, June 03, 2010 12:46 PM To: Discussion of precise time and frequency measurement Subject: Re: [time-nuts] Training period for a Rb clock using GPS Yes, that makes sense. I think that we can arrange things so that we train for 10-12 hours. Do you not think that that is a long enough time for a single loop to be effective? Thanks again! =Abhay On Thu, Jun 3, 2010 at 9:37 AM, Bob Campli...@rtty.us wrote: Hi That will give you the best answer with a simple loop. The problem is that best may not be good enough to actually get your Rb on time / on frequency. Something more sophisticated than a simple loop may be needed. Bob -Original Message- From: time-nuts-boun...@febo.com [mailto:time-nuts-boun...@febo.com] On Behalf Of Abhay Parekh Sent: Thursday, June 03, 2010 12:28 PM To: Discussion of precise time and frequency measurement Subject: Re: [time-nuts] Training period for a Rb clock using GPS Ok, great. So if we can train for h hours we should set the time constant somewhere between h/10 and h/5. It would be safer to pick something closer to h/10 since when the clock powers up it might start in the wrong place so a smaller value helps the clock move quickly into the right area, but h/5 will act as a better buffer against hanging bridges. Is my reasoning correct? Thanks =Abhay On Thu, Jun 3, 2010 at 9:07 AM, Bob Campli...@rtty.us wrote: Hi If you have an 18 hour time constant you would need a training period of 5 to 10 X 18 hours to get the system to settle. For a one hour training period the time constant should be in the 5 to 10 minute range. Bob -Original Message- From: time-nuts-boun...@febo.com [mailto:time-nuts-boun...@febo.com] On Behalf Of Abhay Parekh Sent: Thursday, June 03, 2010 12:02 PM To: Discussion of precise time and frequency measurement Subject: Re: [time-nuts] Training period for a Rb clock using GPS Hi Hal, Thanks so much for the detailed post. I have a follow up question: What is the relationship between the training time and the appropriate value of the time constant (currently set at 18 hours)? The time constant isn't the size of a moving average window is it? Thanks again for your help. We are a bit clueless here but trying to learn... =Abhay On Wed, Jun 2, 2010 at 2:02 AM, Hal Murrayhmur...@megapathdsl.net wrote: par...@berkeley.edu said: I am a newbie at this, but have been playing around with 2 prs10s. For our application we need to run the clocks without gps, but we do get to sync it to gps *initially* for as long as we want. However, what we've noticed is that when we train it for short periods of time ( 1 hour a day) the clock drifts for a few microseconds a day once we've disconnected gps, but when we train it for say 12 hours, its drift seems to be much less (sub sub microsecond/day). We were wondering why this should be so! Look at it the other way. How long should it take to train it? Let's use rough numbers. There are 1E5 seconds per day. Your few microseconds is 1E-6 seconds. That's an accuracy of 1 part in 1E11. Your sub-sub is 1/10 microsecond or 1E-7 seconds. So that's an accuracy of 1 part in 1E12. The data sheet says: Aging (after 30 days)5E-11 (monthly) 5E-11 is 50E-12, so that's 2E-12 per day which is what you saw. The data sheet also says: The PRS10 can time-tag an external 1 pps input with 1 ns resolution. These values may be reported back via RS-232, or used to phase-lock the unit to an external reference (such as GPS) with time constants of several hours. There are 4E3 seconds in an hour and 1E9 nanoseconds per second. So in an hour, you can get close to 1 part in 1E12. But that's assuming that the input PPS signal
Re: [time-nuts] Training period for a Rb clock using GPS
I should have said, we give it a 24hr time period to settle regardless of the time constant. Scott On 06/03/2010 12:23 PM, Scott Mace wrote: We have noticed two things with the PRS10s and other Rbs. They need about 24hrs to settle before they really start to perform well and if there are any significant temperature swings, expect them to react to it. We have a PRS10 in a Meinberg M900 that takes about 1 day to recover from 2-3C temp swing in a datacenter. It will be off (as compared to the other 1-PPS that are not in that room) by about 50-100ns. Normally it's within +-10ns when compared to the other GPSDOs. We noticed this happened when the air-handlers would alternate on about a 2-week cycle. Scott On 06/03/2010 12:09 PM, Abhay Parekh wrote: Awesome. Thanks so much! =Abhay On Thu, Jun 3, 2010 at 10:07 AM, Bob Campli...@rtty.us wrote: Hi Your loop for 10 hours would be around 1 or 2 hours. That's 60 X 60 X (1 or 2) seconds = 3,600 to 7,200 seconds. If GPS is good to +/- 20 ns out of your receiver in your location then you would get 20 x 10^-9 / (3600 or 7200) = 2.7 to 5.5 X 10^-12 inside the loop. The Rb should be below that level over the same time period. Simple answer - yes it should be good enough. Bob -Original Message- From: time-nuts-boun...@febo.com [mailto:time-nuts-boun...@febo.com] On Behalf Of Abhay Parekh Sent: Thursday, June 03, 2010 12:46 PM To: Discussion of precise time and frequency measurement Subject: Re: [time-nuts] Training period for a Rb clock using GPS Yes, that makes sense. I think that we can arrange things so that we train for 10-12 hours. Do you not think that that is a long enough time for a single loop to be effective? Thanks again! =Abhay On Thu, Jun 3, 2010 at 9:37 AM, Bob Campli...@rtty.us wrote: Hi That will give you the best answer with a simple loop. The problem is that best may not be good enough to actually get your Rb on time / on frequency. Something more sophisticated than a simple loop may be needed. Bob -Original Message- From: time-nuts-boun...@febo.com [mailto:time-nuts-boun...@febo.com] On Behalf Of Abhay Parekh Sent: Thursday, June 03, 2010 12:28 PM To: Discussion of precise time and frequency measurement Subject: Re: [time-nuts] Training period for a Rb clock using GPS Ok, great. So if we can train for h hours we should set the time constant somewhere between h/10 and h/5. It would be safer to pick something closer to h/10 since when the clock powers up it might start in the wrong place so a smaller value helps the clock move quickly into the right area, but h/5 will act as a better buffer against hanging bridges. Is my reasoning correct? Thanks =Abhay On Thu, Jun 3, 2010 at 9:07 AM, Bob Campli...@rtty.us wrote: Hi If you have an 18 hour time constant you would need a training period of 5 to 10 X 18 hours to get the system to settle. For a one hour training period the time constant should be in the 5 to 10 minute range. Bob -Original Message- From: time-nuts-boun...@febo.com [mailto:time-nuts-boun...@febo.com] On Behalf Of Abhay Parekh Sent: Thursday, June 03, 2010 12:02 PM To: Discussion of precise time and frequency measurement Subject: Re: [time-nuts] Training period for a Rb clock using GPS Hi Hal, Thanks so much for the detailed post. I have a follow up question: What is the relationship between the training time and the appropriate value of the time constant (currently set at 18 hours)? The time constant isn't the size of a moving average window is it? Thanks again for your help. We are a bit clueless here but trying to learn... =Abhay On Wed, Jun 2, 2010 at 2:02 AM, Hal Murrayhmur...@megapathdsl.net wrote: par...@berkeley.edu said: I am a newbie at this, but have been playing around with 2 prs10s. For our application we need to run the clocks without gps, but we do get to sync it to gps *initially* for as long as we want. However, what we've noticed is that when we train it for short periods of time ( 1 hour a day) the clock drifts for a few microseconds a day once we've disconnected gps, but when we train it for say 12 hours, its drift seems to be much less (sub sub microsecond/day). We were wondering why this should be so! Look at it the other way. How long should it take to train it? Let's use rough numbers. There are 1E5 seconds per day. Your few microseconds is 1E-6 seconds. That's an accuracy of 1 part in 1E11. Your sub-sub is 1/10 microsecond or 1E-7 seconds. So that's an accuracy of 1 part in 1E12. The data sheet says: Aging (after 30 days)5E-11 (monthly) 5E-11 is 50E-12, so that's 2E-12 per day which is what you saw. The data sheet also says: The PRS10 can time-tag an external 1 pps input with 1 ns resolution. These values may be reported back via RS-232, or used to phase-lock the unit to an external reference (such as GPS) with time constants of several hours. There are 4E3 seconds in an hour and 1E9 nanoseconds per second. So
Re: [time-nuts] Training period for a Rb clock using GPS
What is the relationship between the training time and the appropriate value of the time constant (currently set at 18 hours)? The time constant isn't the size of a moving average window is it? My assumption is that the filter is a something like a simple exponential decay. The error will decay (roughly) like 1/e per time-constant. There could easily be scale factors due to quirks in documentation or implementation. For a given time constant, how long you need to wait depends on both how far off you start and how good you need to get. Ok, great. So if we can train for h hours we should set the time constant somewhere between h/10 and h/5. It would be safer to pick something closer to h/10 since when the clock powers up it might start in the wrong place so a smaller value helps the clock move quickly into the right area, but h/5 will act as a better buffer against hanging bridges. Is my reasoning correct? I think you are close, but you also have to consider if the unit will be warmed up and stable quickly enough. If the warm up time is close to the training time, it's not likely to work very well. I think this type of problem is best solved by a few experiments. Do you know how accurate you need it to be? Can you get close enough? Another consideration is temperature shift. Are you training the unit in the same environment as you will be using it? -- These are my opinions, not necessarily my employer's. I hate spam. ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.
Re: [time-nuts] Training period for a Rb clock using GPS
Robert, Thanks that is really helpful. =Abhay On Tue, Jun 1, 2010 at 7:48 PM, Robert Benward rbenw...@verizon.net wrote: Abhay, The microprocessor learns the drift of the oscillator while it is connected, so that when GPS is lost, it can make those corrections in anticipation of drift. The longer it's locked to GPS, the better it learns the drift (more history). Don't confuse the word drift with locked, even when locked, the oscillator is drifting and continuously being corrected. These corrections (the voltage to the EFC or electronic frequency correction) are an indication of the oscillator drift, even though the output is exactly 10MHz. Others can provide links to useful articles. Bob - Original Message - From: Abhay Parekh par...@berkeley.edu To: Discussion of precise time and frequency measurement time-nuts@febo.com Sent: Tuesday, June 01, 2010 10:36 PM Subject: [time-nuts] Training period for a Rb clock using GPS Hello, I am a newbie at this, but have been playing around with 2 prs10s. For our application we need to run the clocks without gps, but we do get to sync it to gps *initially* for as long as we want. However, what we've noticed is that when we train it for short periods of time ( 1 hour a day) the clock drifts for a few microseconds a day once we've disconnected gps, but when we train it for say 12 hours, its drift seems to be much less (sub sub microsecond/day). We were wondering why this should be so! I apologize if this is obvious to most of you but I would greatly appreciate your help in shedding some light for us. Thanks! =Abhay ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there. ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there. ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.
Re: [time-nuts] Training period for a Rb clock using GPS
par...@berkeley.edu said: I am a newbie at this, but have been playing around with 2 prs10s. For our application we need to run the clocks without gps, but we do get to sync it to gps *initially* for as long as we want. However, what we've noticed is that when we train it for short periods of time ( 1 hour a day) the clock drifts for a few microseconds a day once we've disconnected gps, but when we train it for say 12 hours, its drift seems to be much less (sub sub microsecond/day). We were wondering why this should be so! Look at it the other way. How long should it take to train it? Let's use rough numbers. There are 1E5 seconds per day. Your few microseconds is 1E-6 seconds. That's an accuracy of 1 part in 1E11. Your sub-sub is 1/10 microsecond or 1E-7 seconds. So that's an accuracy of 1 part in 1E12. The data sheet says: Aging (after 30 days) 5E-11 (monthly) 5E-11 is 50E-12, so that's 2E-12 per day which is what you saw. The data sheet also says: The PRS10 can time-tag an external 1 pps input with 1 ns resolution. These values may be reported back via RS-232, or used to phase-lock the unit to an external reference (such as GPS) with time constants of several hours. There are 4E3 seconds in an hour and 1E9 nanoseconds per second. So in an hour, you can get close to 1 part in 1E12. But that's assuming that the input PPS signal is right-on. There are two types of GPS receivers. Most use a free running clock and generate the PPS pulse with the closest clock edge. They typically have noise on the order of 15-50 ns. Fancy ones will tell you how far off they think it is. The really fancy ones will have a VCXO so they can slew the clock to the right offset. One magic word is hanging bridges. It comes up in discussions occasionally. For lots of info on that area: http://www.gpstime.com/files/PTTI/PTTI_2006.pdf 31 pages, lots of good stuff, aka time sink. More here: http://www.leapsecond.com/pages/vp/heater.htm 2 or 3 screens, good stuff, a quick read. So with only an hour, it's not unreasonable that you are off by a factor of 10, but you might have to get unlucky for a hanging bridge to get you. But there is another factor to consider. What sort of filter is the software using between the PPS input and the knob that adjusts the frequency? More from the data sheet: When tracking an external input, the time constant can be set from 5 minutes to 18 hours. I think the manual says the default is 65K seconds. That's 18 hours. Unless you changed it, that explains why 1 hour wasn't enough. It might get better if you give it more time and/or tweak the time constant if you can only get 12 hours. -- These are my opinions, not necessarily my employer's. I hate spam. ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.
Re: [time-nuts] Training period for a Rb clock using GPS
Hi The only thing I would add to that is: If your training starts from power up, part of your hour is spent learning something wrong. Most Rb's drift a bit as they stabilize after being turned on. Bob On Jun 2, 2010, at 5:02 AM, Hal Murray wrote: par...@berkeley.edu said: I am a newbie at this, but have been playing around with 2 prs10s. For our application we need to run the clocks without gps, but we do get to sync it to gps *initially* for as long as we want. However, what we've noticed is that when we train it for short periods of time ( 1 hour a day) the clock drifts for a few microseconds a day once we've disconnected gps, but when we train it for say 12 hours, its drift seems to be much less (sub sub microsecond/day). We were wondering why this should be so! Look at it the other way. How long should it take to train it? Let's use rough numbers. There are 1E5 seconds per day. Your few microseconds is 1E-6 seconds. That's an accuracy of 1 part in 1E11. Your sub-sub is 1/10 microsecond or 1E-7 seconds. So that's an accuracy of 1 part in 1E12. The data sheet says: Aging (after 30 days) 5E-11 (monthly) 5E-11 is 50E-12, so that's 2E-12 per day which is what you saw. The data sheet also says: The PRS10 can time-tag an external 1 pps input with 1 ns resolution. These values may be reported back via RS-232, or used to phase-lock the unit to an external reference (such as GPS) with time constants of several hours. There are 4E3 seconds in an hour and 1E9 nanoseconds per second. So in an hour, you can get close to 1 part in 1E12. But that's assuming that the input PPS signal is right-on. There are two types of GPS receivers. Most use a free running clock and generate the PPS pulse with the closest clock edge. They typically have noise on the order of 15-50 ns. Fancy ones will tell you how far off they think it is. The really fancy ones will have a VCXO so they can slew the clock to the right offset. One magic word is hanging bridges. It comes up in discussions occasionally. For lots of info on that area: http://www.gpstime.com/files/PTTI/PTTI_2006.pdf 31 pages, lots of good stuff, aka time sink. More here: http://www.leapsecond.com/pages/vp/heater.htm 2 or 3 screens, good stuff, a quick read. So with only an hour, it's not unreasonable that you are off by a factor of 10, but you might have to get unlucky for a hanging bridge to get you. But there is another factor to consider. What sort of filter is the software using between the PPS input and the knob that adjusts the frequency? More from the data sheet: When tracking an external input, the time constant can be set from 5 minutes to 18 hours. I think the manual says the default is 65K seconds. That's 18 hours. Unless you changed it, that explains why 1 hour wasn't enough. It might get better if you give it more time and/or tweak the time constant if you can only get 12 hours. -- These are my opinions, not necessarily my employer's. I hate spam. ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there. ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.
Re: [time-nuts] Training period for a Rb clock using GPS
Abhay, The microprocessor learns the drift of the oscillator while it is connected, so that when GPS is lost, it can make those corrections in anticipation of drift. The longer it's locked to GPS, the better it learns the drift (more history). Don't confuse the word drift with locked, even when locked, the oscillator is drifting and continuously being corrected. These corrections (the voltage to the EFC or electronic frequency correction) are an indication of the oscillator drift, even though the output is exactly 10MHz. Others can provide links to useful articles. Bob - Original Message - From: Abhay Parekh par...@berkeley.edu To: Discussion of precise time and frequency measurement time-nuts@febo.com Sent: Tuesday, June 01, 2010 10:36 PM Subject: [time-nuts] Training period for a Rb clock using GPS Hello, I am a newbie at this, but have been playing around with 2 prs10s. For our application we need to run the clocks without gps, but we do get to sync it to gps *initially* for as long as we want. However, what we've noticed is that when we train it for short periods of time ( 1 hour a day) the clock drifts for a few microseconds a day once we've disconnected gps, but when we train it for say 12 hours, its drift seems to be much less (sub sub microsecond/day). We were wondering why this should be so! I apologize if this is obvious to most of you but I would greatly appreciate your help in shedding some light for us. Thanks! =Abhay ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there. ___ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.