Ultra low frequency VCO
in [Design]
|
Prev: AofE 3rd Edition
Next: Labelling Prototypes
From: JosephKK on 12 Jul 2010 22:57 On Mon, 12 Jul 2010 09:37:07 -0400, Phil Hobbs <pcdhSpamMeSenseless(a)electrooptical.net> wrote: >JosephKK wrote: >> On Fri, 9 Jul 2010 10:22:34 -0700 (PDT), j <jdc1789(a)gmail.com> wrote: >> >>> Resolution of noise vs frequency, (as in bw), is the issue in phase >>> noise measurements. The OP never stated the offset from the carrier >>> nor bandwidth. Or maybe I just missed it. >>> >>> Itâs not clear to me why JosephKK thinks this would be either a time >>> consuming or difficult measurement to make. Assuming the appropriate >>> measurement system is in hand 100 dBc numbers are easily achievable. >>> Whether itâs 60 Hz or several GHzâs the global issues are the same in >>> making a phase noise measurement. >>> >>> But having said the above, without the OP responding I guess it really >>> doesnât matter. But Iâd like to know more about the application and >>> derive solutions from there. >> >> >> OK. For a carrier of 60 MHz. Pick an instrument or test setup of your >> choice, state the model[s]. Clearly explain just what is going on in the >> measurement and the time it takes to accumulate sufficient data to make >> the measurement. Explain why it takes that much time to reach a reliable >> measurement of -100 dBc phase noise at that carrier frequency. >> >> Now see how well it scales to one million times lower fundamental >> frequency without a similar scaling in measurement time. > >It's the modulation frequency that's relevant, not the carrier >frequency. Measurements get slower when you reduce the bandwidth. > >(You can see why this doesn't work if you imagine running it >backwards--mixing or multiplying up to some very high frequency doesn't >allow you to make a measurement with 1 Hz bandwidth any faster. > > >Cheers > >Phil Hobbs Now what is the equivalent bandwidth of -100 dBc for a 60 Hz carrier? Since you said 20 log() basis 60 * 10^-5 is 600 microHz. That would have to take some minutes, and if you wanted a proper 10 to 1 measurement buffer, it takes ten times longer. Call it 10,000 seconds? A few hours. And the reference stability etc., i remarked on is coming into play.
From: JosephKK on 12 Jul 2010 23:02 On Mon, 12 Jul 2010 09:39:54 -0400, Phil Hobbs <pcdhSpamMeSenseless(a)electrooptical.net> wrote: >JosephKK wrote: >> On Fri, 09 Jul 2010 11:56:28 -0400, Phil Hobbs >> <pcdhSpamMeSenseless(a)electrooptical.net> wrote: >> >>> On 7/9/2010 8:59 AM, JosephKK wrote: >>>> On Thu, 08 Jul 2010 15:37:28 -0400, Phil Hobbs >>>> <pcdhSpamMeSenseless(a)electrooptical.net> wrote: >>>> >>>>> Phil Hobbs wrote: >>>>> >>>>>> I don't know that -100 dBc/Hz is that hard at 60 Hz. I bet you could do >>>>>> that by running a bog standard multivibrator at 1024*1024*60 Hz and >>>>>> dividing down. You'd need a sine shaper, but the phase noise goes down >>>>>> by N**2, so you'd get 100 dB improvement just from that. Alternatively, >>>>>> you could make an LC VCO and divide that down. >>>>> 120 dB. Can't count today. >>>>> >>>>> Cheers >>>>> >>>>> Phil Hobbs >>>> Sure, you can mathematically "predict" it, but how do you measure it? >>>> Or do you switch to another metric which can be both predicted and >>>> measured? >>> Let's keep the math bashing to the other thread, okay? >>> >>> Although it isn't highly relevant to the OP's problem, it wouldn't be >>> very difficult to measure the residual FM--use MOSFET buffers to drive >>> two divider strings running from independent power supplies, and >>> cross-correlate their outputs, exchanging them periodically to get rid >>> of the drift in the correlator. For the correlator design, see Hanbury >>> Brown and Twiss, circa 1963--and they did it with discrete bipolars. >>> >>> There are hard measurements, but this isn't one of them. >>> >>> Cheers >>> >>> Phil Hobbs >> >> My issue was not so much the direct difficulty of the measurement, there >> are several fairly straight forward setups. But with the _time_ it would >> take to make the measurement using many of those setups. The elapsed >> time seriously aggravates other measurement issues, notably including >> calibration. > >Modulation frequency isn't affected by heterodyning or frequency >multiplication and division. If you take a 60 MHz sine wave and FM it >at 1 Hz modulation frequency and 1 MHz peak frequency deviation (M=1E6), > then divide it by a million, you get a 60-Hz sine wave modulated at 1 >Hz with a 1-Hz peak frequency division (M=1). > >Cheers > >Phil Hobbs I am sorry. I think i am misreading your post, are you saying you can get a 1 MHz deviation on a 60 Hz carrier? Naw, you must be trying to say something else and i misunderstood.
From: Phil Hobbs on 12 Jul 2010 23:23 JosephKK wrote: > On Mon, 12 Jul 2010 09:37:07 -0400, Phil Hobbs > <pcdhSpamMeSenseless(a)electrooptical.net> wrote: > >> JosephKK wrote: >>> On Fri, 9 Jul 2010 10:22:34 -0700 (PDT), j <jdc1789(a)gmail.com> wrote: >>> >>>> Resolution of noise vs frequency, (as in bw), is the issue in phase >>>> noise measurements. The OP never stated the offset from the carrier >>>> nor bandwidth. Or maybe I just missed it. >>>> >>>> It's not clear to me why JosephKK thinks this would be either a time >>>> consuming or difficult measurement to make. Assuming the appropriate >>>> measurement system is in hand 100 dBc numbers are easily achievable. >>>> Whether it's 60 Hz or several GHz's the global issues are the same in >>>> making a phase noise measurement. >>>> >>>> But having said the above, without the OP responding I guess it really >>>> doesn't matter. But I'd like to know more about the application and >>>> derive solutions from there. >>> >>> OK. For a carrier of 60 MHz. Pick an instrument or test setup of your >>> choice, state the model[s]. Clearly explain just what is going on in the >>> measurement and the time it takes to accumulate sufficient data to make >>> the measurement. Explain why it takes that much time to reach a reliable >>> measurement of -100 dBc phase noise at that carrier frequency. >>> >>> Now see how well it scales to one million times lower fundamental >>> frequency without a similar scaling in measurement time. >> It's the modulation frequency that's relevant, not the carrier >> frequency. Measurements get slower when you reduce the bandwidth. >> >> (You can see why this doesn't work if you imagine running it >> backwards--mixing or multiplying up to some very high frequency doesn't >> allow you to make a measurement with 1 Hz bandwidth any faster. >> >> >> Cheers >> >> Phil Hobbs > > Now what is the equivalent bandwidth of -100 dBc for a 60 Hz carrier? > Since you said 20 log() basis 60 * 10^-5 is 600 microHz. That would have > to take some minutes, and if you wanted a proper 10 to 1 measurement > buffer, it takes ten times longer. Call it 10,000 seconds? A few hours. > And the reference stability etc., i remarked on is coming into play. You're confused, I'm afraid. -100 dBc phase noise in a given bandwidth (say 1 Hz, but it doesn't matter) is 7 microradians RMS. Using a 5V swing and a CMOS analogue gate as a phase detector, that's dV = 7e-6 rad RMS * 5V/(pi rad) = 11 microvolts RMS, which is trivial to measure in a 1 Hz bandwidth in a few seconds--it's 80 dB above the noise of a good op amp, so you just have to wait for the filter to settle. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal ElectroOptical Innovations 55 Orchard Rd Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
From: Phil Hobbs on 12 Jul 2010 23:29 JosephKK wrote: > On Mon, 12 Jul 2010 09:39:54 -0400, Phil Hobbs > <pcdhSpamMeSenseless(a)electrooptical.net> wrote: > >> JosephKK wrote: >>> On Fri, 09 Jul 2010 11:56:28 -0400, Phil Hobbs >>> <pcdhSpamMeSenseless(a)electrooptical.net> wrote: >>> >>>> On 7/9/2010 8:59 AM, JosephKK wrote: >>>>> On Thu, 08 Jul 2010 15:37:28 -0400, Phil Hobbs >>>>> <pcdhSpamMeSenseless(a)electrooptical.net> wrote: >>>>> >>>>>> Phil Hobbs wrote: >>>>>> >>>>>>> I don't know that -100 dBc/Hz is that hard at 60 Hz. I bet you could do >>>>>>> that by running a bog standard multivibrator at 1024*1024*60 Hz and >>>>>>> dividing down. You'd need a sine shaper, but the phase noise goes down >>>>>>> by N**2, so you'd get 100 dB improvement just from that. Alternatively, >>>>>>> you could make an LC VCO and divide that down. >>>>>> 120 dB. Can't count today. >>>>>> >>>>>> Cheers >>>>>> >>>>>> Phil Hobbs >>>>> Sure, you can mathematically "predict" it, but how do you measure it? >>>>> Or do you switch to another metric which can be both predicted and >>>>> measured? >>>> Let's keep the math bashing to the other thread, okay? >>>> >>>> Although it isn't highly relevant to the OP's problem, it wouldn't be >>>> very difficult to measure the residual FM--use MOSFET buffers to drive >>>> two divider strings running from independent power supplies, and >>>> cross-correlate their outputs, exchanging them periodically to get rid >>>> of the drift in the correlator. For the correlator design, see Hanbury >>>> Brown and Twiss, circa 1963--and they did it with discrete bipolars. >>>> >>>> There are hard measurements, but this isn't one of them. >>>> >>>> Cheers >>>> >>>> Phil Hobbs >>> My issue was not so much the direct difficulty of the measurement, there >>> are several fairly straight forward setups. But with the _time_ it would >>> take to make the measurement using many of those setups. The elapsed >>> time seriously aggravates other measurement issues, notably including >>> calibration. >> Modulation frequency isn't affected by heterodyning or frequency >> multiplication and division. If you take a 60 MHz sine wave and FM it >> at 1 Hz modulation frequency and 1 MHz peak frequency deviation (M=1E6), >> then divide it by a million, you get a 60-Hz sine wave modulated at 1 >> Hz with a 1-Hz peak frequency division (M=1). >> >> Cheers >> >> Phil Hobbs > > I am sorry. I think i am misreading your post, are you saying you can > get a 1 MHz deviation on a 60 Hz carrier? Naw, you must be trying to say > something else and i misunderstood. You can put a 1 MHz phase modulation on a 60 Hz carrier, but you sure don't wind up with anything pretty. For instance, you could put the 60 Hz on a varactor-loaded transmission line, and drive the varactors with 1.000000000 MHz. As long as the varactors were driven really differentially, you wouldn't get any 1.000000000 MHz on the line. That's way outside the quasistatic limit, of course, which is where we're all used to working. It would be a nasty splattery mess, but you'd get _something_. But that wasn't the point I was trying to make. ;) Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal ElectroOptical Innovations 55 Orchard Rd Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net
From: Phil Hobbs on 12 Jul 2010 23:43
Tim Wescott wrote: > On 07/08/2010 12:29 PM, Phil Hobbs wrote: >> Paul Keinanen wrote: >>> On Tue, 06 Jul 2010 09:52:43 -0700, Tim Wescott <tim(a)seemywebsite.com> >>> wrote: >>> >>>> On 07/06/2010 09:10 AM, Daku wrote: >>>>> On Jul 5, 8:59 pm, Tim Wescott<t...(a)seemywebsite.com> wrote: >>>>>> I'd hardly call 60Hz "ultra low frequency". But it is pretty darned >>>>>> low. >>>>>> >>>>>> All the suggestions you've gotten so far are good as far as they go >>>>>> and >>>>>> may well be perfect -- but what are you trying to do? Do you need >>>>>> sine >>>>>> wave out or square? If sine wave, how pure? Do you have any >>>>>> specifications on jitter, phase noise, or frequency accuracy? >>>>> I am trying to design a PLL for very low frequencies, e.g., power line >>>>> grid. >>>>> I am concerned with the VCO as it is a crucial sub-circuit. I am >>>>> aiming for >>>>> a phase noise of approximately -100 dBc/Hz but not very sure of the >>>>> offset >>>>> frequency. Ideally, I would like to have frequency accuracy of 1 - 5% >>>>> at most. >>>>> Also, I am aware that S-parameter methods are not appropriate at these >>>>> low >>>>> frequencies. >>> >>> If you want to track the _actual_ mains frequency, just use a mains >>> driven synchronous motor. To get the noise sidebands down, use some >>> flywheels :-). >>> >>>> I think that those specs would be difficult to achieve with an >>>> all-analog oscillator running at 60Hz. Not impossible -- I could do >>>> it, and Joerg could do it in a fraction of the time I'd take. Using >>>> some sort of direct digital synthesis -- even if it's just a >>>> microprocessor -- running off of a crystal reference would be almost >>>> trivial in comparison and would probably take less board space and >>>> would be far more repeatable in manufacturing. >>>> >>>> If you just had to do this purely in the analog domain your best bet >>>> might be a pair of crystal oscillators, frequency steered with >>>> varactors, carefully built, and with their outputs mixed down to >>>> 60Hz. But that's a solution I would expect to see in a bit of kit >>>> from the 50's through the 80's -- anything later and I'd expect to >>>> see a DDS. >>> >>> Just a few minutes ago, the Nordel AC network (Danish isles, Finland, >>> Norway, Sweden) was running at 50.11 Hz or +2200 ppm above nominal in >>> order to allow the mains synchronized clocks to catch up. >>> A simple fundamental frequency VXCO can be pulled about +/-100 ppm >>> with the load capacitance. About 1000 ppm is the maximum with >>> adjustable serial inductance and adjustable parallel load capacitance >>> at the crystal. >>> >>> At 50/60 Hz, even a trivial processor can generate a variable >>> frequency sine wave using the NCO (Numerically Controlled Oscillator) >>> principle to generate a sine wave, which can be locked to the incoming >>> signal in some loop configuration. >>> >>> Even a trivial processor might be able to generate both sine and >>> cosine waveforms for 49.98, 50.00. 50.92 Hz etc. in parallel and >>> performing a phase comparison between all these in parallel to >>> determine the best match. >>> >> >> I don't know that -100 dBc/Hz is that hard at 60 Hz. I bet you could do >> that by running a bog standard multivibrator at 1024*1024*60 Hz and >> dividing down. You'd need a sine shaper, but the phase noise goes down >> by N**2, so you'd get 100 dB improvement just from that. Alternatively, >> you could make an LC VCO and divide that down. > > This actually kind of makes my point, which I didn't state clearly: if > you _don't_ use a divider it'll be hard. With a divider it gets easy, > as long as you ignore clock jitter in the divider (and clock jitter > probably isn't a big deal, given the output frequency). > >> You might even be able to do it with all analog--the OPA378 has 20 >> nV/sqrt(Hz) all the way down to DC. With a 5V sine wave at 60 Hz, that's >> something like 1800 V/s, so 20 nV gives you something like 10 >> picoseconds per root hertz. You probably lose a factor of sqrt(2) in >> there, but that ought to be good enough. Your ALC network would >> contribute more than that, almost for sure. > > Depending on how close to the carrier you want to get, you lose a factor > of up to infinity (if you get _really_ close to the carrier). > > The noise gain is something like 1/(s^2 + w0^2) -- it's an oscillator. > Worse, because it's an RC, the constant you're multiplying by is greater > than one -- I get Hn(s) ~ 15/(s^2 + w0^2). That's not taking the > current noise of the part into account (which, I admit, I haven't > checked on because I'm lazy). > > 1Hz away your noise gain is just about 200, for 4uV/sqrt(Hz). That's > doing OK, but at 0.1Hz away the noise gain is about 2000 -- all you have > to do is measure close enough to the carrier at a wide enough bandwidth > and your noise is too high (sure would be nice if the OP specified what > he wanted, but I think we lost him). > Sorry to be slow responding. That's an interesting point about the noise gain of the oscillator. I think we agree that the OPA378 is okay down to 1 Hz offset or thereabouts, anyway. (It's a very nice part btw--I'm using it in some millihertz things just now.) Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal ElectroOptical Innovations 55 Orchard Rd Briarcliff Manor NY 10510 845-480-2058 hobbs at electrooptical dot net http://electrooptical.net |