Re: [time-nuts] High-end GPSDO's
On Sun, 19 Aug 2018 15:27:03 -0700 Ralph Devoe wrote: > Thanks for the notes on the FS740. I missed it back last November. It > looks so user-friendly that I might have to get one. Most SRS stuff is > really well done. Yes, indeed. All SRS schematics I have read looked really like someone who knew what he is doing did the design. The newer designs also include comments in the schematics about the design decisions, which is really nice IMHO. > The question I have is whether it is practical to get to the 10(-13) level > with a good Rb and a good GPSDO. The Rb's aging rate is low enough, but, as > Corby has pointed out, they are sensitive to temperature / pressure so > that they can move faster than the GPSDO can correct for. I'm not clear how > far this can be pushed. If you have a HP5065 and use Corby's modification, then getting to 1e-13 should be reasonably easy. If you have a PRS10 or LPRO, you will have to do temperature and probably also pressure compensation to get to the stability needed. But I cannot say whether you can actually get down to those numbers. I am currently building a setup to measure the temperature and pressure dependence on an LPRO, but it will take some time until everything is ready, and then it will take a couple of months until I have reliable data. BTW: I recommend using some digital scheme for offset frequency generation and not touching the C-field adjust of the Rb. Either a simple DDS with enough bits should do, or a divider scheme like Rick's synthesizer[1]. Attila Kinali [1] "A narrow band high-resolution synthesizer using a direct digital synthesizer pollowed by repeated dividing and mixing," by Rick Karlquist, 1995 http://www.karlquist.com/FCS95.pdf -- The bad part of Zurich is where the degenerates throw DARK chocolate at you. ___ 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] High-end GPSDO's
Thanks for the notes on the FS740. I missed it back last November. It looks so user-friendly that I might have to get one. Most SRS stuff is really well done. The question I have is whether it is practical to get to the 10(-13) level with a good Rb and a good GPSDO. The Rb's aging rate is low enough, but, as Corby has pointed out, they are sensitive to temperature / pressure so that they can move faster than the GPSDO can correct for. I'm not clear how far this can be pushed. Ralph ___ 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] Choosing a GPS IC for carrier phase measurements
Hi > On Aug 19, 2018, at 5:26 PM, Nicolas Braud-Santoni > wrote: > > On Sat, Aug 18, 2018 at 08:25:11PM +0200, Attila Kinali wrote: >> On Thu, 16 Aug 2018 17:45:30 +0200 >> Nicolas Braud-Santoni wrote: >> >>> The main issue I'm running into is that most timing GPS modules will >>> happily give you carrier phase measurements... for their internal >>> oscillator, and the few ICs I can find that would possibly do the job, like >>> ublox's UBX-M8030-KT-FT, do not have publicly-available datasheets & docs. >>> (I tried contacting ublox to ask, and never got an answer...) >> >> Yeah. u-blox isn't as nice as they used to be to small customers :-( > > Ah, that's a pity. :( > > FWIW, I'm going to try going through a ublox reseller that says they have > that timing GNSS IC available, ask whether I can purchase in small quantities > and whether I could have the datasheet. > > >> 1) use the timing of the PPS to deduce what the phase relation between >> your clock and the internal oscillator of the LEA is. >> In principle, this is possible, but I have not worked out the math, >> so I cannot say for sure. > > I've considered that, and it ends up being mostly equivalent to what I'm > currently doing. Part of the issue is that I don't want to wait ~1 month > for a PLL lock, but I also need/want an integration time about that long, > as that's about where the GPS becomes more stable than my local XO. If you *assume* 2 ns on the GPS ( could be better … might be worse) then at 100,000 seconds ( = about a day) you are at 2x10^-14. That’s a pretty good LO. A month gets you to 6.7x10^-16. A LO that is in that range is also in the “very expensive” range. Going to an L1 /L2 approach will drop the GPS errors by an order of magnitude or more. Given that you already are in the “very expensive” range, the cost of the receiver should be trivial. Bob > > I was able to work around the problem in part, by dynamically adjusting the > constants of my IIR (and so the integration time), and it works pretty OK > despite being highly non-linear, but there is only so much one can do when > fixing hardware deficiencies in software. :( > > >> 2) replace the internal oscillator with one phase locked to your OCXO. >> The internal clock of the LEA is derived from a single TCXO. You can >> easily unsolder it and feed your own signal in. > > That seems pretty much equivalent to using a “naked” GPS IC, as the part > I care about is clocking it with my XO and getting phase measurements > (wrt. the time-code and the carrier) out. > > OTOH, it might be much easier than getting a datasheet out of u-blox, so > I will keep that in mind, in case I cannot do it the way I wanted. > Thanks a lot for the suggestion. > >> Unfortunately, I was explicitly asked not to share this information :-( > > :'( > > >>> So, what are people around here using for that purpose? Bonus points if the >>> chip can work with a differential-signal clock. :) >> >> The alternative is to build your own GPS receiver. If you only want >> GPS L1 C/A, then you can use the design of The Witch Navigator[1] >> with one of the VHDL/Verilog projects out there (e.g. cu-hw-gps [2]) >> >> If you want to go for L2C, L5 or Galileo, you have to do your own coding :-) > > Thanks a lot for the pointers :) > > I indeed low-key considered rolling my own GNSS receiver, as there are now > some RFSoCs that would make it not too bad, but I decided against it as: > > - Trying to make a good GPSDO is hard enough as-is ;) > - I would need anyway to be able to validate that the PLL works correctly > and gives the expected accuracy, with a known-good GNSS receiver. > > > Best, > > nicoo > ___ > 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] Choosing a GPS IC for carrier phase measurements
On Sat, Aug 18, 2018 at 10:13:49PM +0200, Magnus Danielson wrote: > >> 2) replace the internal oscillator with one phase locked to your OCXO. > >> [...] > > > > If you go this route, the close in phase noise of your signal needs to be > > in the > > “pretty good TCXO” range. There are indeed some sources out there that > > are way to noisy …. > > Indeed. You want both pretty good wideband noise and close-in. Yeap. The TCXO I currently use is meant for GPS applications, but I'm been switching to a Si549 [0], which boasts jitter as low as 95fs RMS. :O Part of the reason for the switch is that, in my current setup, I get enough noise from my MCU's DAC (for the voltage control of the XO) that it sometimes creates additional, measurable phase noise. :( Obviously, going for entirely-digital control removes the problem: the only remaining timing-sensitive, analog parts being the RF input for the GNSS chip, and the clock distribution from the XO to the GNSS IC and to the frequency synthesis chip (I currently have an on-board Si5334 so I can output any frequency I need, but I'm planning on using a Si5341 for the new PCB) [0] https://www.silabs.com/documents/public/data-sheets/si549-datasheet.pdf Best, nicoo signature.asc Description: PGP signature ___ 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] Choosing a GPS IC for carrier phase measurements
On Sat, Aug 18, 2018 at 08:25:11PM +0200, Attila Kinali wrote: > On Thu, 16 Aug 2018 17:45:30 +0200 > Nicolas Braud-Santoni wrote: > > > The main issue I'm running into is that most timing GPS modules will > > happily give you carrier phase measurements... for their internal > > oscillator, and the few ICs I can find that would possibly do the job, like > > ublox's UBX-M8030-KT-FT, do not have publicly-available datasheets & docs. > > (I tried contacting ublox to ask, and never got an answer...) > > Yeah. u-blox isn't as nice as they used to be to small customers :-( Ah, that's a pity. :( FWIW, I'm going to try going through a ublox reseller that says they have that timing GNSS IC available, ask whether I can purchase in small quantities and whether I could have the datasheet. > 1) use the timing of the PPS to deduce what the phase relation between > your clock and the internal oscillator of the LEA is. > In principle, this is possible, but I have not worked out the math, > so I cannot say for sure. I've considered that, and it ends up being mostly equivalent to what I'm currently doing. Part of the issue is that I don't want to wait ~1 month for a PLL lock, but I also need/want an integration time about that long, as that's about where the GPS becomes more stable than my local XO. I was able to work around the problem in part, by dynamically adjusting the constants of my IIR (and so the integration time), and it works pretty OK despite being highly non-linear, but there is only so much one can do when fixing hardware deficiencies in software. :( > 2) replace the internal oscillator with one phase locked to your OCXO. > The internal clock of the LEA is derived from a single TCXO. You can > easily unsolder it and feed your own signal in. That seems pretty much equivalent to using a “naked” GPS IC, as the part I care about is clocking it with my XO and getting phase measurements (wrt. the time-code and the carrier) out. OTOH, it might be much easier than getting a datasheet out of u-blox, so I will keep that in mind, in case I cannot do it the way I wanted. Thanks a lot for the suggestion. > Unfortunately, I was explicitly asked not to share this information :-( :'( > > So, what are people around here using for that purpose? Bonus points if the > > chip can work with a differential-signal clock. :) > > The alternative is to build your own GPS receiver. If you only want > GPS L1 C/A, then you can use the design of The Witch Navigator[1] > with one of the VHDL/Verilog projects out there (e.g. cu-hw-gps [2]) > > If you want to go for L2C, L5 or Galileo, you have to do your own coding :-) Thanks a lot for the pointers :) I indeed low-key considered rolling my own GNSS receiver, as there are now some RFSoCs that would make it not too bad, but I decided against it as: - Trying to make a good GPSDO is hard enough as-is ;) - I would need anyway to be able to validate that the PLL works correctly and gives the expected accuracy, with a known-good GNSS receiver. Best, nicoo signature.asc Description: PGP signature ___ 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] History Channel "In Search of" Time Travel
In the UK, www.history.com forcibly redirects to www.history.co.uk On Sun, Aug 19, 2018 at 1:44 PM, Didier Juges wrote: > That would be very nice since I missed it. > Thanks in advance, > Didier KO4BB > > On Sat, Aug 18, 2018, 2:24 PM Dan Rae wrote: > > > If anyone wants to see this I could upload a .mkv copy to Wetransfer for > > download. > > Dan > > > > ___ > > 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. > ___ 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] TNS-BUF update
Without the benefit of agressive low pass filtering the noise of an LT3042 at 10Hz is at least an order of magnitude worse than that of an unfiltered LED for the same output voltage. The noise of Nitride passivated LEDs degrades significantly with aging and there is a strong correlation between LED noise and its quality / aging degradation. http://przyrbwn.icm.edu.pl/APP/PDF/119/a119z4p10.pdf Bruce > On 20 August 2018 at 03:42 Gerhard Hoffmann wrote: > > > > > Am 19.08.2018 um 14:43 schrieb Bruce Griffiths: > > Yes, I looked at the plots and I see exactly what I described. > > 2nV/rtHz at frequencies somewhat above the reference current source output > > RC filter, rising to a high value at frequencies below the current source > > output filter pole. > > This is an inherent property of the architecture. > > I really don't know how you come to that conclusion. Even the HLMP-6000 > needs more than > 100 Hz to be better than the 3042 in its standard data sheet circuit, > without extra filtering > provided, and the LED delivers only half the DC voltage, another 6 dB. > > 'Ordinary' LEDs from Toshiba or Osram(ex Siemens) are 20 to 30 db worse, > and they are > quality parts, not unknown junk box parts. > Their 1/f corner is - OMG!!!11! - the plot goes only to 1 MHz! > > Postulating superior LED performance at 1Hz from that is quite venturesome. > LEDs are good in comparison to el cheapo bandgaps and high voltage "Zeners", > but they cannot do magic. > > > And please don't say, it's the light. The measurements are made in a box > in a box in a box > to handle the pV noise densities. My preamps are now at 70 pV/rtHz and > there is a > chopper in statu nascendi at 100 pV sub- 0.1 Hz. Yes, I see the noise > of the 0.6 Ohm switches. > > But I need a better FFT analyzer now. A working SPI driver @ > 1Mtransfers/sec for the > BeagleBoneBlack would do. > > cheers, Gerhard > > > ps. I'm also interested in that time travel movie. > > > ___ > 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] TNS-BUF update
There are some NIR sensitive photodiodes that take advantage of the NIR transparency of some black epoxies. Bruce > On 20 August 2018 at 02:48 Dana Whitlow wrote: > > > Be watchful about the black epoxy. It may just be a dyed (otherwise > clear) epoxy, > and some of the black dyes in common use pass near IR quite readily. This > would > probably be most troublesome if the ambient light source were of the > incandescent > persuasion. > > Also, small diodes in clear packages can also make photodetectors. I once, > in > desperation at a customer's site, successfully kluged a slightly > sub-nanosecond > detector out of a 1N914 (or one of those types like that) and a 9V battery. > > Dana > > > On Sun, Aug 19, 2018 at 8:15 AM, Didier Juges wrote: > > > That's a very good point. I have used LEDs as photocells many times, the > > first time was in 1974 when it was easier for this university student to > > get LEDs than photocells. > > > > I have also made a bi-directional optically isolated data link using a > > single fiber optic cable and two fiber optic transmitters (LEDs) in the > > HFBR-500 series. > > > > A couple coats of black paint, or a dab of black epoxy covering the LED > > should work. Include the back side of the PWB for good measure, this is > > time-nuts :) > > > > Didier KO4BB > > > > > > On Sun, Aug 19, 2018, 3:38 AM Poul-Henning Kamp > > wrote: > > > > > > > > In message <1818735266.3837388.1534666949...@webmail.xtra.co.nz>, Bruce > > > Griffit > > > hs writes: > > > > > > >It exploits the fact that for a RED LED at least the difference > > > >between the LED forward voltage and the transistor Vbe is ~ 1V and > > > >has a fairly low tempco and has low noise (at least for RED LEDs). > > > > > > ... in darkness. > > > > > > When using LEDs this sort of way, they should always be totally > > > shielded from all external light. > > > > > > -- > > > 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. > > > ___ > 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] High-end GPSDO's
On Sat, 18 Aug 2018 20:53:19 +0200 Attila Kinali wrote: > Just to avoid confusion: I don't want to bash the FS740, not at all. > It's a very well designed device with lots of ingenious solutions for > small details (see also my quick review of it at [2]), but it's still > just a GPSDO and has to live with the limitations of the GPS/GNSS > system. Addendum: I just found out that Wikipedia has a page on the error contributions of GPS/GNSS [1]. In particular, I would like to draw your attention to the table of User Equivalent Range Errors in the second section. Each m of error is the equivalent of ~3ns. Attila Kinali [1] https://en.wikipedia.org/wiki/Error_analysis_for_the_Global_Positioning_System -- It is upon moral qualities that a society is ultimately founded. All the prosperity and technological sophistication in the world is of no use without that foundation. -- Miss Matheson, The Diamond Age, Neal Stephenson ___ 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] TNS-BUF update
Also check your red LEDs. I have some that have a Vf of over 3V. They could be a blue LED/phosphor (not likely, blue -> red conversion is rather inefficient) or two red LEDs in series. ___ 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] TNS-BUF update
Am 19.08.2018 um 14:43 schrieb Bruce Griffiths: Yes, I looked at the plots and I see exactly what I described. 2nV/rtHz at frequencies somewhat above the reference current source output RC filter, rising to a high value at frequencies below the current source output filter pole. This is an inherent property of the architecture. I really don't know how you come to that conclusion. Even the HLMP-6000 needs more than 100 Hz to be better than the 3042 in its standard data sheet circuit, without extra filtering provided, and the LED delivers only half the DC voltage, another 6 dB. 'Ordinary' LEDs from Toshiba or Osram(ex Siemens) are 20 to 30 db worse, and they are quality parts, not unknown junk box parts. Their 1/f corner is - OMG!!!11! - the plot goes only to 1 MHz! Postulating superior LED performance at 1Hz from that is quite venturesome. LEDs are good in comparison to el cheapo bandgaps and high voltage "Zeners", but they cannot do magic. And please don't say, it's the light. The measurements are made in a box in a box in a box to handle the pV noise densities. My preamps are now at 70 pV/rtHz and there is a chopper in statu nascendi at 100 pV sub- 0.1 Hz. Yes, I see the noise of the 0.6 Ohm switches. But I need a better FFT analyzer now. A working SPI driver @ 1Mtransfers/sec for the BeagleBoneBlack would do. cheers, Gerhard ps. I'm also interested in that time travel movie. ___ 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] TNS-BUF update
Be watchful about the black epoxy. It may just be a dyed (otherwise clear) epoxy, and some of the black dyes in common use pass near IR quite readily. This would probably be most troublesome if the ambient light source were of the incandescent persuasion. Also, small diodes in clear packages can also make photodetectors. I once, in desperation at a customer's site, successfully kluged a slightly sub-nanosecond detector out of a 1N914 (or one of those types like that) and a 9V battery. Dana On Sun, Aug 19, 2018 at 8:15 AM, Didier Juges wrote: > That's a very good point. I have used LEDs as photocells many times, the > first time was in 1974 when it was easier for this university student to > get LEDs than photocells. > > I have also made a bi-directional optically isolated data link using a > single fiber optic cable and two fiber optic transmitters (LEDs) in the > HFBR-500 series. > > A couple coats of black paint, or a dab of black epoxy covering the LED > should work. Include the back side of the PWB for good measure, this is > time-nuts :) > > Didier KO4BB > > > On Sun, Aug 19, 2018, 3:38 AM Poul-Henning Kamp > wrote: > > > > > In message <1818735266.3837388.1534666949...@webmail.xtra.co.nz>, Bruce > > Griffit > > hs writes: > > > > >It exploits the fact that for a RED LED at least the difference > > >between the LED forward voltage and the transistor Vbe is ~ 1V and > > >has a fairly low tempco and has low noise (at least for RED LEDs). > > > > ... in darkness. > > > > When using LEDs this sort of way, they should always be totally > > shielded from all external light. > > > > -- > > 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. > ___ 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] TNS-BUF update
That's a very good point. I have used LEDs as photocells many times, the first time was in 1974 when it was easier for this university student to get LEDs than photocells. I have also made a bi-directional optically isolated data link using a single fiber optic cable and two fiber optic transmitters (LEDs) in the HFBR-500 series. A couple coats of black paint, or a dab of black epoxy covering the LED should work. Include the back side of the PWB for good measure, this is time-nuts :) Didier KO4BB On Sun, Aug 19, 2018, 3:38 AM Poul-Henning Kamp wrote: > > In message <1818735266.3837388.1534666949...@webmail.xtra.co.nz>, Bruce > Griffit > hs writes: > > >It exploits the fact that for a RED LED at least the difference > >between the LED forward voltage and the transistor Vbe is ~ 1V and > >has a fairly low tempco and has low noise (at least for RED LEDs). > > ... in darkness. > > When using LEDs this sort of way, they should always be totally > shielded from all external light. > > -- > 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] TNS-BUF update
Yes, I looked at the plots and I see exactly what I described. 2nV/rtHz at frequencies somewhat above the reference current source output RC filter, rising to a high value at frequencies below the current source output filter pole. This is an inherent property of the architecture. The LT3042 is better in this regard than TI's offerings but only because its easy to filter the reference and set the filter pole frequency much lower. Its also quieter than the Abracon low noise supply that I have. Bruce > On 20 August 2018 at 00:25 Gerhard Hoffmann wrote: > > > You did not take a look at the plots, did you? > > It is seldom to find these data taken all in an identical setup. > > > Am 19.08.2018 um 14:07 schrieb Bruce Griffiths: > > The LT3042 is still inherently very noisy at ultra low frequencies > > approaching and below the pole frequency (can be well below 1Hz) of the > > reference filter (There's a limit to the maximum capacitance > > available/feasible). Forward biased diodes including LEDs are quieter in > > this region. > > > 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. ___ 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] TNS-BUF update
Am 19.08.2018 um 10:22 schrieb Bruce Griffiths: Rick I devised the bias circuit for the TNS-BUF. It exploits the fact that for a RED LED at least the difference between the LED forward voltage and the transistor Vbe is ~ 1V and has a fairly low tempco and has low noise (at least for RED LEDs). (Most of the LTSpice LED models do not correctly predict LED forward voltage drop tempco.) Most classical schemes for biasing BJTs use a resistive voltage divider which inevitably couples power supply noise into the BJT collector current. John Miles changed the bias circuit of some classical series shunt amp buffers to one similar to this and the buffer flicker phase noise was significantly reduced. In principle an LED could be used to directly set the dc bias at the base of the amplifier transistors, however inductors may be required to shunt part of the emitter series resistance at dc to allow the desired dc collector current to be established. A pair of series connected LEDs buffered by an npn emitter follower would allow the bias voltage to be shared by all stages and allow the inductor to be replaced by a capacitor bypassing part of the emitter to ground resistance required to establish the desired collector current whilst achieving the required resistance from RF to ground for RF. Not all red LEDs are created equal. For noise, by far the best I have found is the HLMP6000 by HP / Avago / Whoever_owns_it_today. < https://www.flickr.com/photos/137684711@N07/24354944411/in/album-72157662535945536/ > 0 dB is 1nV/rtHz, +20 dB is 10 nV/rtHz and so on. LEDs are ineffective photo cells, probably because of the large band gap and their built-in color filter. I have given up to apply the blob of black laquer, I have never seen a difference. What I find impressive is the noise performance of low voltage Z-Diodes. We are always told that Zeners are noisy. No. Avalanche breakdown is noisy. Take a look at these NXP BZX84C2V7 and C3V3. Admire the low 1/f corner and note how things turn bad when we approach 5V. < https://www.flickr.com/photos/137684711@N07/24411798996/in/album-72157662535945536/ > The precious 7V LM399 reference is a complete joke in comparison. Its oven is no help here. Classical bias schemes are usually much noisier especially at low frequencies. Even regulators like the LT3042 are quite noisy at frequencies below the the pole of the reference circuit low pass filter. That does no justice to the LT3042. It features 2nV/rtHz to _very_ low frequencies. It is very easy to bypass a constant current source while it is very costly to filter a low impedance LED. The LT3042 even has a startup circuit so that it does not take forever to get to the right voltage. Most LEDs are much worse, and especially at low frequencies. The LT3042 is a piece of art. It leaves the rest of the regulators that we know in the dust, by 40 dB or better. Most of the noise < 50 Hz or so goes on my 89441A and the too-small input coupling capacitor of my preamp. (20 ADA4898 op amps in par, 220 pV/rtHz) The preamp has been fixed with a costly :-( wet slug tantalum in the meantime. 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.