Re: [time-nuts] inexpensive fiber optic distribution
I wondered when the conversion factor noise was going to show up in the conversation. I had asked time-nuts about this a long time ago. I was experimenting with optical isolation since lightning seems to be my friend. Ended up with minicircuit transformers and cable. Later discovered an HP article something like 1987 that showed how to do the conversion with the optical units of the time. I did tinker with SFPs and it was painful due to the connectors. So seeing the nice boards that have been made was great and just maybe enough to get me to make a set of boards to experiment with again. Great thread. Regards Paul WB8TSL On Fri, Mar 22, 2019 at 6:01 AM Hal Murray wrote: > > bow...@gmail.com said: > > The infrastructure is easily enough interfaced to, even a simple > > microprocessor can do it, using switches allows for distribution with > known > > characteristics. > > It would be interesting to measure the delays through a switch. > > The switches are retiming the data stream. It has to buffer up enough > data > before it starts sending to make sure that it won't run out before the end > of > packet if the switch clock is fast and the source clock is slow. There is > also bit/byte synchronization, and crossing a clock boundary before things > get > started. > > > > -- > These are my opinions. I hate spam. > > > > > ___ > time-nuts mailing list -- time-nuts@lists.febo.com > To unsubscribe, go to > http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com > and follow the instructions there. > ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] inexpensive fiber optic distribution
I'd be curious in these numbers as well. When I used to deal with this type of an issue for a living I typically didn't worry much about consistent delays of well under a millisecond. I'd be curious to know what the typical numbers are. Mark Spencer m...@alignedsolutions.com 604 762 4099 > On Mar 22, 2019, at 10:22 AM, Hal Murray wrote: > > > bow...@gmail.com said: >> The infrastructure is easily enough interfaced to, even a simple >> microprocessor can do it, using switches allows for distribution with known >> characteristics. > > It would be interesting to measure the delays through a switch. > > The switches are retiming the data stream. It has to buffer up enough data > before it starts sending to make sure that it won't run out before the end of > packet if the switch clock is fast and the source clock is slow. There is > also bit/byte synchronization, and crossing a clock boundary before things > get > started. > > > > -- > These are my opinions. I hate spam. > > > > > ___ > time-nuts mailing list -- time-nuts@lists.febo.com > To unsubscribe, go to > http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com > and follow the instructions there. > ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] inexpensive fiber optic distribution
bow...@gmail.com said: > The infrastructure is easily enough interfaced to, even a simple > microprocessor can do it, using switches allows for distribution with known > characteristics. It would be interesting to measure the delays through a switch. The switches are retiming the data stream. It has to buffer up enough data before it starts sending to make sure that it won't run out before the end of packet if the switch clock is fast and the source clock is slow. There is also bit/byte synchronization, and crossing a clock boundary before things get started. -- These are my opinions. I hate spam. ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] inexpensive fiber optic distribution
I'd recommend that fiber-optic transmission should be used where it has definite advantages, such as long distance, for galvanic isolation, or in serious EMC situations. It's fun to experiment with for all kinds of things, but it has no advantage over traditional wiring/cable for signal routing in a lab with "normal" operating and distance requirements. I don't see much point (other than the fun of fooling around with it) in the added complexity, versus the usual, simple, cabling that gets the job done. Also remember, you pay a noise penalty in the conversion of electrical signals to light and back. Ed ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] inexpensive fiber optic distribution
On 3/21/19 1:19 AM, Gerhard Hoffmann wrote: BTW our modules for Cisco had special contents in some registers to make sure that nobody could use alien modules. There must have happened some social engineering if I read the post above :-) I suspect more that someone sued under the Magnuson Moss Act - You can't deny warranty service for using "third party plug compatible" - and there's some things about "not concealing details of the interface" which would make creating such plug compatible devices. This dates way back to the IBM mainframe days. Note that this is probably (by court precedents that are available) only for "computing and communication equipment" - little crypto id chips in printer cartridges may not fall in that bucket. ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] inexpensive fiber optic distribution
To pull this back to time nuts territory, this brings up the idea of simply using traditional ethernet infrastructure, be it 100M/1G/10G/whatever at the PHY layer to achieve distribution of 1pps. The infrastructure is easily enough interfaced to, even a simple microprocessor can do it, using switches allows for distribution with known characteristics. Not sure that you'd need to climb up even into Layer 2, but if you do, that would allow for identification of the sources by MAC address. You's still have to worry about collisions I suppose. There are other networking technologies that could also be used with less impact and better characteristics, with infiniband (lossless, low latency) coming immediately to mind. On Thu, Mar 21, 2019 at 11:03 AM Julien Goodwin wrote: > > The lock in is almost entirely vendor profiteering, optical monitoring & > wavelength switching are standards (where implemented), but plenty of > vendors use (almost always trivially bypassable) lock-in to sell SFPs > they rebrand from the same OEMs at up to 100x markup. > > The folk in the big telcos (or in my case, big content providers) who > run the global backbones have plenty of stories. > As one of the guys who has to make these decisions for a big company, I have to pipe up in our defense. It has very little to do with profiteering, though it is easily seen as such. In reality, there are a number of vendors of transceivers out there (or other similar parts) ranging from excellent to really horrible. The 'lock-in' is to restrict the pool to those that have been qualified and tested. This vastly reduces support and service hassles on both ends of the phone call. The incremental revenue on a transceiver of, in one case I can document, of about $600, even when multiplied by the number required for a full system, isn't even in the decimal points of the total cost of the system, which can easily exceed $3-4M. Bob ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] inexpensive fiber optic distribution
On 3/20/19 11:42 PM, Warren Kumari wrote: On Thu, Mar 21, 2019 at 7:05 AM Anders Wallin wrote: We've tinkered with SFP-to-SMA adapter boards like this: https://github.com/aewallin/SFP2SMA_2018.03 http://www.anderswallin.net/2018/04/500-mhz-sfp-board-v4/ And if you want a "ready made" solution instead of making the board yourself, etc there is http://shop.sysmocom.de/t/development-boards/sfp I have some on order to play with, but am traveling, and so cannot play with them yet. W This is very interesting stuff... There have been times I was thinking "what I really need here is a optical link for the 10 or 100 MHz".. But "hundreds of dollars" was a bit steep. This is starting to get more reasonable (still, about $200 for a link, by the time you buy/assemble adapter boards, the $7 modules, and some fiber). As far as the reprogramming goes (another comment on the list) - I understand that is to make the unit have some particular flavor SFP-wise. If you were plugging them into a Cisco switch and all you had was ones configured for Juniper switches, then you'd want to do that. At the multikilobuck level to do this, you'd need a LOT of them to make it worthwhile. The folks building radio telescope arrays with hundreds of cables probably fit in that use case. A devoted time-nut using it to distribute their reference around the house or lab, probably not. although designed for 1 Gbit (1.25 GHz, or 800 ps 'bits') they work down to 10 MHz (and 5 MHz IIRC). With a decent interface-board and a short bit of fiber the ADEV noise floor is very close to what is measurable with a 3120A (aka TimePod). IIRC a rough rule-of-thumb is that the fiber can add around 1e-16/meter. So add 1km of fibre and you can expect fluctuations at 1e-13 level - roughly... AW ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] inexpensive fiber optic distribution
Am 21.03.19 um 07:27 schrieb Hal Murray: jim...@earthlink.net said: It seems like you could probably figure out how to interface to these things and use them to distribute timing signals. The document is the "SFP MSA". I still have the XFP version somewhere here. I don't know how well they will work for something like a PPS. Somebody should try it. The signal pattern is mostly trying to be friendly to the PLL that does clock recovery and/or the AGC that sets up the switching level. If I wanted to send a PPS, I'd use something simple like Manchester encoding. That assumes you can line your transmit clock up with the PPS. It won't work if you just want a link-extender for a PPS. Manchester should work. Simply trying to transmit a differential CML level 1pps won't do. The signals are ac-coupled typically 100nF/50 Ohm right at the module border. I was one of the ~5 people who did the electronics of the 10 GBPS XFP modules at Infineon Fiber Optics (RIP, sold) and someone decreased the coupling capacitors to 10n for a very minor advantage. It's 10 GBPS, pure RF after all. No. That gave us 1 bit error per day (complete show stopper) and a month of delay. I'm puzzled that it works so far off-frequency with these SFPs. In our 10 GBPS XFPs it would definitely not work. The eye opener at the input to the transmitter would go nuts, and the receiver clock recovery would not work. BTW our modules for Cisco had special contents in some registers to make sure that nobody could use alien modules. There must have happened some social engineering if I read the post above :-) regards, Gerhard ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] inexpensive fiber optic distribution
On Thu, Mar 21, 2019 at 8:00 AM Hal Murray wrote: > > jim...@earthlink.net said: > > It seems like you could probably figure out how to interface to these > things > > and use them to distribute timing signals. > > There are two big advantages of fibers: > They work well for long distances. > No ground loops. > > The target market for those contraptions is networking. The signal is > digital, and they get to work with a "nice" signal, probably 8B/10B. > That's 8 > bits of data in 10 bits on the wire. Roughly, they pick 256 out of 1024 > patterns that have minimal long strings of 0s and 1s and use a few more > patterns for control - things like idle, and begin/end of packet. It > works > fine if AC coupled. > > I don't know how well they will work for something like a PPS. Somebody > should try it. The signal pattern is mostly trying to be friendly to the > PLL > that does clock recovery and/or the AGC that sets up the switching level. > > If I wanted to send a PPS, I'd use something simple like Manchester > encoding. > That assumes you can line your transmit clock up with the PPS. It won't > work > if you just want a link-extender for a PPS. > > If you are connecting to a FPGA, use whatever they support for high-speed > serial interface. Again, that assumes you can line your signal up with > the > transmit clock. > > I'm pretty sure the signaling is differential, either PECL or one of the > newer > standards like LVDS. > > > > They take a standard singlemode duplex fiber and have a pinout suitable > for > > plugging into a Cisco, etc. switch. Apparently, there's some ... > > I'm pretty sure none of the programming is on the high speed path. Maybe > changing the signal levels. I don't know how much of it is total BS, aka > vendor lock in, as compared to actually doing something useful. I'd be > surprised (but not much) if they didn't work fine with no programming. Yup, the majority of the programming is assigning a "personality" (vendor lock). There is sometimes also a small bit of config (some let you assign minimum receive before it declares the link up, or (very unusual) the frequency for tunable optics (usually this is tuned in the device, but some let you assign in the SFP). AFAIK, all the ones you can easily purchase (eg FS or flexoptix) come with at least some personality. W > > > > -- > These are my opinions. I hate spam. > > > > > ___ > time-nuts mailing list -- time-nuts@lists.febo.com > To unsubscribe, go to > http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com > and follow the instructions there. > -- I don't think the execution is relevant when it was obviously a bad idea in the first place. This is like putting rabid weasels in your pants, and later expressing regret at having chosen those particular rabid weasels and that pair of pants. ---maf ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] inexpensive fiber optic distribution
In message <20190321062751.4c65b406...@ip-64-139-1-69.sjc.megapath.net>, Hal Murray writes: >I don't know how well they will work for something like a PPS. Somebody >should try it. The signal pattern is mostly trying to be friendly to the PLL >that does clock recovery and/or the AGC that sets up the switching level. If we are only inside a lab, look at TOSLINK which can be made to do interesting stuff if you play a bit with it. -- Poul-Henning Kamp | UNIX since Zilog Zeus 3.20 p...@freebsd.org | TCP/IP since RFC 956 FreeBSD committer | BSD since 4.3-tahoe Never attribute to malice what can adequately be explained by incompetence. ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] inexpensive fiber optic distribution
On 21/3/19 5:27 pm, Hal Murray wrote: > > jim...@earthlink.net said: >> It seems like you could probably figure out how to interface to these things >> and use them to distribute timing signals. > > There are two big advantages of fibers: > They work well for long distances. > No ground loops. Pure optical amplification, and multiplexing is also possible, less useful for pure time & frequency use though. However for long distance you start to get into fun issues like thermal affects etc, and running something like Sync-E might actually make more sense than a pure signal. > The target market for those contraptions is networking. The signal is > digital, and they get to work with a "nice" signal, probably 8B/10B. That's > 8 > bits of data in 10 bits on the wire. Roughly, they pick 256 out of 1024 > patterns that have minimal long strings of 0s and 1s and use a few more > patterns for control - things like idle, and begin/end of packet. It works > fine if AC coupled. 8b/10b in the older signals up to 1g, 64b/66b in 10g. These days native 25g, 50g and more are combined to make 100g / 200g / 400g Ethernet. > I don't know how well they will work for something like a PPS. Somebody > should try it. The signal pattern is mostly trying to be friendly to the PLL > that does clock recovery and/or the AGC that sets up the switching level. > > If I wanted to send a PPS, I'd use something simple like Manchester encoding. > > That assumes you can line your transmit clock up with the PPS. It won't work > if you just want a link-extender for a PPS. > > If you are connecting to a FPGA, use whatever they support for high-speed > serial interface. Again, that assumes you can line your signal up with the > transmit clock. > > I'm pretty sure the signaling is differential, either PECL or one of the > newer > standards like LVDS. Yep, all diff I/O, details are in the spec. Random sample: http://www.kt.agh.edu.pl/~lason/SFP/TRX17/SCP6Gx8c.pdf >> They take a standard singlemode duplex fiber and have a pinout suitable for >> plugging into a Cisco, etc. switch. Apparently, there's some ... > > I'm pretty sure none of the programming is on the high speed path. Maybe > changing the signal levels. I don't know how much of it is total BS, aka > vendor lock in, as compared to actually doing something useful. I'd be > surprised (but not much) if they didn't work fine with no programming. Correct, the control path is out of band, and just I2C. The lock in is almost entirely vendor profiteering, optical monitoring & wavelength switching are standards (where implemented), but plenty of vendors use (almost always trivially bypassable) lock-in to sell SFPs they rebrand from the same OEMs at up to 100x markup. The folk in the big telcos (or in my case, big content providers) who run the global backbones have plenty of stories. ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] inexpensive fiber optic distribution
jim...@earthlink.net said: > It seems like you could probably figure out how to interface to these things > and use them to distribute timing signals. There are two big advantages of fibers: They work well for long distances. No ground loops. The target market for those contraptions is networking. The signal is digital, and they get to work with a "nice" signal, probably 8B/10B. That's 8 bits of data in 10 bits on the wire. Roughly, they pick 256 out of 1024 patterns that have minimal long strings of 0s and 1s and use a few more patterns for control - things like idle, and begin/end of packet. It works fine if AC coupled. I don't know how well they will work for something like a PPS. Somebody should try it. The signal pattern is mostly trying to be friendly to the PLL that does clock recovery and/or the AGC that sets up the switching level. If I wanted to send a PPS, I'd use something simple like Manchester encoding. That assumes you can line your transmit clock up with the PPS. It won't work if you just want a link-extender for a PPS. If you are connecting to a FPGA, use whatever they support for high-speed serial interface. Again, that assumes you can line your signal up with the transmit clock. I'm pretty sure the signaling is differential, either PECL or one of the newer standards like LVDS. > They take a standard singlemode duplex fiber and have a pinout suitable for > plugging into a Cisco, etc. switch. Apparently, there's some ... I'm pretty sure none of the programming is on the high speed path. Maybe changing the signal levels. I don't know how much of it is total BS, aka vendor lock in, as compared to actually doing something useful. I'd be surprised (but not much) if they didn't work fine with no programming. -- These are my opinions. I hate spam. ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] inexpensive fiber optic distribution
On Thu, Mar 21, 2019 at 7:05 AM Anders Wallin wrote: > We've tinkered with SFP-to-SMA adapter boards like this: > https://github.com/aewallin/SFP2SMA_2018.03 > http://www.anderswallin.net/2018/04/500-mhz-sfp-board-v4/ > And if you want a "ready made" solution instead of making the board yourself, etc there is http://shop.sysmocom.de/t/development-boards/sfp I have some on order to play with, but am traveling, and so cannot play with them yet. W > although designed for 1 Gbit (1.25 GHz, or 800 ps 'bits') they work down to > 10 MHz (and 5 MHz IIRC). > With a decent interface-board and a short bit of fiber the ADEV noise floor > is very close to what is measurable with a 3120A (aka TimePod). > IIRC a rough rule-of-thumb is that the fiber can add around 1e-16/meter. So > add 1km of fibre and you can expect fluctuations at 1e-13 level - > roughly... > > AW > > On Thu, Mar 21, 2019 at 6:06 AM jimlux wrote: > > > While researching this for a radio telescope, I found amazingly > > inexpensive (<$10) fiber optic transceivers for 1Gbps sorts of rates.. > > > > They take a standard singlemode duplex fiber and have a pinout suitable > > for plugging into a Cisco, etc. switch. Apparently, there's some > > internal programming which sets the hardware for the kind of switch, > > rates, etc.. fs.com has a generic ASIC inside the adapter which can be > > reprogrammed for whatever formatting, compatibility etc is needed. The > > astronomers were saying that at Berkeley, they bought a box for a few k > > which allows them to reconfigure the transceivers. > > > > > > Something like this: > > https://www.fs.com/products/12622.html > > > > Data sheet here: > https://img-en.fs.com/file/datasheet/sfp1g-lx-31-10km.pdf > > > > and a 10 meter patch cable is $5 > > https://www.fs.com/products/40203.html > > > > This is sort of "old news" in the networking world (the SFP form factor > > was defined in 2001) > > > > > > It seems like you could probably figure out how to interface to these > > things and use them to distribute timing signals. > > > > I'll be talking to the radio astronomers tomorrow again, and I'll see > > what I can find out about interfaces. I suspect they do NOT use them for > > conventional networking - they're all about running lots of bits from > > antennas into large multichannel correlators/beamformers. > > > > They're buying these things by the dozen. > > > > ___ > > time-nuts mailing list -- time-nuts@lists.febo.com > > To unsubscribe, go to > > http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com > > and follow the instructions there. > > > ___ > time-nuts mailing list -- time-nuts@lists.febo.com > To unsubscribe, go to > http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com > and follow the instructions there. > -- I don't think the execution is relevant when it was obviously a bad idea in the first place. This is like putting rabid weasels in your pants, and later expressing regret at having chosen those particular rabid weasels and that pair of pants. ---maf ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
Re: [time-nuts] inexpensive fiber optic distribution
We've tinkered with SFP-to-SMA adapter boards like this: https://github.com/aewallin/SFP2SMA_2018.03 http://www.anderswallin.net/2018/04/500-mhz-sfp-board-v4/ although designed for 1 Gbit (1.25 GHz, or 800 ps 'bits') they work down to 10 MHz (and 5 MHz IIRC). With a decent interface-board and a short bit of fiber the ADEV noise floor is very close to what is measurable with a 3120A (aka TimePod). IIRC a rough rule-of-thumb is that the fiber can add around 1e-16/meter. So add 1km of fibre and you can expect fluctuations at 1e-13 level - roughly... AW On Thu, Mar 21, 2019 at 6:06 AM jimlux wrote: > While researching this for a radio telescope, I found amazingly > inexpensive (<$10) fiber optic transceivers for 1Gbps sorts of rates.. > > They take a standard singlemode duplex fiber and have a pinout suitable > for plugging into a Cisco, etc. switch. Apparently, there's some > internal programming which sets the hardware for the kind of switch, > rates, etc.. fs.com has a generic ASIC inside the adapter which can be > reprogrammed for whatever formatting, compatibility etc is needed. The > astronomers were saying that at Berkeley, they bought a box for a few k > which allows them to reconfigure the transceivers. > > > Something like this: > https://www.fs.com/products/12622.html > > Data sheet here: https://img-en.fs.com/file/datasheet/sfp1g-lx-31-10km.pdf > > and a 10 meter patch cable is $5 > https://www.fs.com/products/40203.html > > This is sort of "old news" in the networking world (the SFP form factor > was defined in 2001) > > > It seems like you could probably figure out how to interface to these > things and use them to distribute timing signals. > > I'll be talking to the radio astronomers tomorrow again, and I'll see > what I can find out about interfaces. I suspect they do NOT use them for > conventional networking - they're all about running lots of bits from > antennas into large multichannel correlators/beamformers. > > They're buying these things by the dozen. > > ___ > time-nuts mailing list -- time-nuts@lists.febo.com > To unsubscribe, go to > http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com > and follow the instructions there. > ___ time-nuts mailing list -- time-nuts@lists.febo.com To unsubscribe, go to http://lists.febo.com/mailman/listinfo/time-nuts_lists.febo.com and follow the instructions there.
