Forums

Heck of a way to run an oscillator?

Started by skeptical engineer May 13, 2014
<haiticare2011@gmail.com> wrote in message 
news:5a9edf0e-46e7-46b8-9dd3-42a6ceda9ada@googlegroups.com...
> I haven't read it yet, but it's got me wondering if a counter on the > oscillator could provide a feedback loop to tune the saw via a heater. > Frequency counters are easy to make. I just wonder whether a temperature > sensor > could adjust the oscillaor, or perhaps more realistically, provide a > lookup > table to compensate for the SAW temperature variation.
How does that give better drift than the crystal in the frequency counter? Tim -- Seven Transistor Labs Electrical Engineering Consultation Website: http://seventransistorlabs.com
On Fri, 16 May 2014 14:31:31 +0100, "Kevin Aylward"
<ExtractkevinRemove@kevinaylward.co.uk> wrote:

>"Jim Thompson" wrote in message >news:vbj9n9p3gebit51f22nagtmlfo3uaebe7p@4ax.com... > >On Thu, 15 May 2014 09:29:42 +1000, Clifford Heath ><no.spam@please.net> wrote: > >>>On 15/05/14 04:43, haiticare2011@gmail.com wrote: >>>> On Wednesday, May 14, 2014 10:46:54 AM UTC-4, Joerg wrote: >>>>> Cheapskates use logic parts for that because you get a six-pack for=
a
>>>> OK thanks. - What logic part would produce a sine wave output? >> >>>One that has no extra gain than what is required to start, so it's not >>>constantly slamming against the amplitude limit. One that limits its >>>amplitude into a resistive load. Or one fitted with an AGC circuit. It >>>helps if the clipping is somewhat soft. Look at the 74HCU04 inverters >>>for a start. > >>"Clipping" is for digital guys. AGC is necessary for low distortion. > >Usually, however, clipping/clamping/limiting pretty much invariably =
produces=20
>the lowest phase noise. Simple principle, is the more devices, the more=20 >noise. > >I have actually spent some time running sims on just about every =
reasonable=20
>oscillator topology you can imagine for phase noise. One transistor =
wins.
> > >Kevin Aylward B.Sc. >www.kevinaylward.co.uk >www.anasoft.co.uk - SuperSpice=20
Thank you. Though somehow it did not really surprise me. ?-) =20
On Fri, 16 May 2014 14:31:31 +0100, "Kevin Aylward"
<ExtractkevinRemove@kevinaylward.co.uk> wrote:

>"Jim Thompson" wrote in message >news:vbj9n9p3gebit51f22nagtmlfo3uaebe7p@4ax.com... > >On Thu, 15 May 2014 09:29:42 +1000, Clifford Heath ><no.spam@please.net> wrote: > >>>On 15/05/14 04:43, haiticare2011@gmail.com wrote: >>>> On Wednesday, May 14, 2014 10:46:54 AM UTC-4, Joerg wrote: >>>>> Cheapskates use logic parts for that because you get a six-pack for a >>>> OK thanks. - What logic part would produce a sine wave output? >> >>>One that has no extra gain than what is required to start, so it's not >>>constantly slamming against the amplitude limit. One that limits its >>>amplitude into a resistive load. Or one fitted with an AGC circuit. It >>>helps if the clipping is somewhat soft. Look at the 74HCU04 inverters >>>for a start. > >>"Clipping" is for digital guys. AGC is necessary for low distortion. > >Usually, however, clipping/clamping/limiting pretty much invariably produces >the lowest phase noise. Simple principle, is the more devices, the more >noise. > >I have actually spent some time running sims on just about every reasonable >oscillator topology you can imagine for phase noise. One transistor wins.
IIRC, Wes Hayward's book "Introduction to Radio Frequency Design" contained a design around a long-tail pair that with suitable component selection, the transistors wouldn't saturate. Unfortunately I have misplaced my copy, so I can't check it.
"josephkk"  wrote in message 
news:76rdn9h2vrkk65sc49lh9psitndj3s5v1c@4ax.com...

On Fri, 16 May 2014 14:31:31 +0100, "Kevin Aylward"
<ExtractkevinRemove@kevinaylward.co.uk> wrote:

>"Jim Thompson" wrote in message
> >>"Clipping" is for digital guys. AGC is necessary for low distortion. > >>Usually, however, clipping/clamping/limiting pretty much invariably >>produces >>the lowest phase noise. Simple principle, is the more devices, the more >>noise. > >>I have actually spent some time running sims on just about every >>reasonable >>oscillator topology you can imagine for phase noise. One transistor wins. > >
>Thank you. Though somehow it did not really surprise me.
There is one sort of an exception, but more than likely not physically doable, and A cross coupled oscillator can, in principal, be configured to null out 1/f up conversion. Whether or not the null point stays over time is another matter. http://www.kevinaylward.co.uk/ee/phasenoise/FlickerNoiseNullification.xht x-coupled LC oscillators still won't beat single transistor xtal ones though. Kevin Aylward B.Sc. www.kevinaylward.co.uk www.anasoft.co.uk - SuperSpice
wrote in message news:f94en916gp1dmlev5mh9a03ir205hppmgu@4ax.com...

On Fri, 16 May 2014 14:31:31 +0100, "Kevin Aylward"
<ExtractkevinRemove@kevinaylward.co.uk> wrote:

>"Jim Thompson" wrote in message >news:vbj9n9p3gebit51f22nagtmlfo3uaebe7p@4ax.com... > >On Thu, 15 May 2014 09:29:42 +1000, Clifford Heath ><no.spam@please.net> wrote: > >>>On 15/05/14 04:43, haiticare2011@gmail.com wrote: >>>> On Wednesday, May 14, 2014 10:46:54 AM UTC-4, Joerg wrote: >>>>> Cheapskates use logic parts for that because you get a six-pack for a >>>> OK thanks. - What logic part would produce a sine wave output? >> >>>One that has no extra gain than what is required to start, so it's not >>>constantly slamming against the amplitude limit. One that limits its >>>amplitude into a resistive load. Or one fitted with an AGC circuit. It >>>helps if the clipping is somewhat soft. Look at the 74HCU04 inverters >>>for a start. > >>"Clipping" is for digital guys. AGC is necessary for low distortion. > >>Usually, however, clipping/clamping/limiting pretty much invariably >>produces >>the lowest phase noise. Simple principle, is the more devices, the more >>noise. > >>I have actually spent some time running sims on just about every >>reasonable >>oscillator topology you can imagine for phase noise. One transistor wins.
>IIRC, Wes Hayward's book "Introduction to Radio Frequency Design" >contained a design around a long-tail pair that with suitable >component selection, the transistors wouldn't saturate.
Sure, but something has to limit in an oscillator otherwise the output will get to trillions of volts and keep on increasing for ever. Oscillators can be designed to be current or voltage limited, but it needs one to actually work! Kevin Aylward B.Sc. www.kevinaylward.co.uk www.anasoft.co.uk - SuperSpice
On Fri, 16 May 2014 17:19:41 -0500, "Tim Williams"
<tmoranwms@charter.net> wrote:

>"Phil Hobbs" <pcdhSpamMeSenseless@electrooptical.net> wrote in message >news:7tydnQENnKRyquvOnZ2dnUVZ_rWdnZ2d@supernews.com... >>> If this were necessary, than Spice simulations would have problems as >>> they don't have noise! >> >> They do, actually, just like all other numerical simulators. It comes >> from roundoff. > >It is necessary, and a problem, from time to time. > >But this is also *because of* roundoff. > >If the derivatives are close enough to zero, they'll be approximated *to >zero*, the timestep cranks up to max and the simulation skims along, >blissfully unaware of its own instability. > >Tim
This was of course a problems when using computers initially designed in the 1950's to 1970s that did not support denormalized floating point values. With the IEEE floating point standard, anything designed after about after 1980 supports denorms. Even with IEEE floats/doubles, even adding a billion times a very small value to a huge value does not change the result a bit. Thus, it is still critical how the equations are evaluated. A mathematical statement that looks very sensible, might not work properly on a computer, so an alternative mathematical formula may have to be used to work around the computer limitations.
>"Phil Hobbs" wrote in message >news:7tydnQENnKRyquvOnZ2dnUVZ_rWdnZ2d@supernews.com...
> >> >When first switched on, the amplifier input related thermal noise is >> amplified, Part of it is going to the frequency sensitive network, >>> which attenuates the lowest and highest frequencies, leaving a broad >>> peak around the resonant frequency. > >> If this were necessary, than Spice simulations would have problems as >> they don't have noise!
>They do, actually, just like all other numerical simulators. It comes from >roundoff.
You missed my bit about "numerical noise" then ?
> >> This is not how an oscillator functions. An oscillator is not amplified > >noise.
>Oscillators build up from noise, though, and they do amplify the close-in >noise very strongly.
They do, but it is not necessary. A pulse exciting the tank will start the oscillation, independent of noise. A pendulum clock does not start because of noise.
>See e.g. >http://rfic.eecs.berkeley.edu/~niknejad/ee242/pdf/eecs242_lect22_phasenoise.pdf
Well.... Although the paper correctly points out non-linear capacitors, implying non linear time constants, as a cause of up conversion of low frequency noise, it incorrectly states/implies that "pumping" action of the oscillator also generates up conversion noise. Simple mixing does not produce phase noise, which is what usually matters, despite the creation of up converted amplitude noise. Unfortunately that paper perpetuates the myth as to the value of the Hajimiri- Lee ISF method. The HL method is wrong, and is completely useless with regard to up conversion, producing errors easily at the 50 dBc level. Its 1/f up conversion formula is wrong. Period. This is all shown here: http://www.kevinaylward.co.uk/ee/phasenoise/phasenoise.html In particular: http://www.kevinaylward.co.uk/ee/phasenoise/LTV.xht Shows why the HL ISF theory is wrong in a manner much more accessible than A.Dimre's highly mathematical paper I reference. http://www.kevinaylward.co.uk/ee/phasenoise/OrthogonalPerturbation.xht Directly demonstrates the invalidity of orthogonal perturbation (hitting the tank at peaks and zero x-ings for those more mathematically challenged) parroted in the lec 22 paper. http://www.kevinaylward.co.uk/ee/phasenoise/PhaseNoiseOscillators.xht Shows in more detail why HL is wrong, and what is the correct mechanism for up-conversion of phase noise. A. Dimir trashed the HL paper pretty much immediately it came out. Its pretty stunning why such a flawed, unusable approach is still being taught in universities. I summarise: "The HL-LTV method can be shown to be valid for spherical chickens in a vacuum"
>Oscillators will work without any noise at all. If the system is >> unstable, any initial DC condition that does not sit at a >> metastable/unstable equilibrium point will ensure that signals start >> changing in such a way as to head toward amplifier limiting. This may >> result in either steady state or chaotic oscillations.
>Not necessarily. There has to be some nonzero signal in the band where the >oscillator is unstable.
Ho...mm humm... ok to be precise, if the initial DC condition is sitting in the already mentioned unstable oscillator system at such a point that it can transverses to the designated limit cycle. Sure, the system might be unstable for various DC points, but actually sit at a point that is locally stable. "Band" implies frequency. Frequency is irrelevant. Instability has nothing to do with frequency. A chaotic oscillator has no fixed frequency. Only in the pathological sense that an impulse may be analysed as a continuous spectrum is it arguable that there is a frequency in the band where the oscillator is unstable. Even then, this is still dubious for chaotic oscillators.
> >> Usually a clean (power on) startup pulse will excite the tank, and it is >> this ringing that generates the oscillation build-up, just as it does in > >the real world.
>Sometimes. It's easily possible that the power-up transient is smooth >enough that its spectrum dies away to well below device noise in the >relevant band.
Possible, but clutching at straws me thinks.
>Good oscillators are designed so that that is the case, because otherwise >their supply rejection would stink.
Sure, precision oscillators need need very good PSRR. Mine have 100db @ 150mv drop out :-) You walked into that one... http://www.rakon.com/products/families/ocxo-ocso - Mercury Note, a one transistor oscillator has 10,000 supporting transistors...LDO, LNBG, Function Generators, Constant power circuits, limiters, LN dividers etc.... Good oscillators are also designed to start as fast as possible.
> > This bit about noise starting up the circuit is often quoted in the > books, and er... ah...somewhat mindlessly repeated, but is very dubious. > > Theoretically numerical noise could start a simulation, and sometimes > does, but it is not necessary.
>Arbitrarily sharp power-up steps kickstart the oscillation in simulation, >maybe, but building up from noise will occur very often in real life,
This missies the point being made. An unstable system is an unstable system. Noise has nothing to do with it. An oscillator in not amplified noise despite that fact that it may well amplifier noise. Kevin Aylward B.Sc. www.kevinaylward.co.uk www.anasoft.co.uk - SuperSpice
<upsidedown@downunder.com> wrote in message 
news:s65en99hfero6ircom5endqg8moamup96m@4ax.com...
> This was of course a problems when using computers initially designed > in the 1950's to 1970s that did not support denormalized floating > point values. With the IEEE floating point standard, anything designed > after about after 1980 supports denorms. > > Even with IEEE floats/doubles, even adding a billion times a very > small value to a huge value does not change the result a bit. Thus, it > is still critical how the equations are evaluated. A mathematical > statement that looks very sensible, might not work properly on a > computer, so an alternative mathematical formula may have to be used > to work around the computer limitations.
SPICE was a glimmer in some student's eye back in the 50s; it has more to do with the RELTOL parameter than the numerical stability of the underlying datatypes. By about twelve orders of magnitude. Was SPICE ever even developed on a platform that didn't support IEEE? I don't even know what they used. Something PDP? Tim -- Seven Transistor Labs Electrical Engineering Consultation Website: http://seventransistorlabs.com
On Sat, 17 May 2014 09:55:10 +0100, "Kevin Aylward"
<ExtractkevinRemove@kevinaylward.co.uk> wrote:

>>"Phil Hobbs" wrote in message >>news:7tydnQENnKRyquvOnZ2dnUVZ_rWdnZ2d@supernews.com... > >> >>> >When first switched on, the amplifier input related thermal noise is >>> amplified, Part of it is going to the frequency sensitive network, >>>> which attenuates the lowest and highest frequencies, leaving a broad >>>> peak around the resonant frequency. >> >>> If this were necessary, than Spice simulations would have problems as >>> they don't have noise! > >>They do, actually, just like all other numerical simulators. It comes from >>roundoff. > >You missed my bit about "numerical noise" then ? > >> >>> This is not how an oscillator functions. An oscillator is not amplified >> >noise. > >>Oscillators build up from noise, though, and they do amplify the close-in >>noise very strongly. > >They do, but it is not necessary. A pulse exciting the tank will start the >oscillation, independent of noise. > >A pendulum clock does not start because of noise.
If you buy a pendulum clock and the transport company very hardly slams it on your floor (power on transient) it might quite well start ticking. Use a resistor (noise source) followed by 10-100 sections of amplifiers and band pass filters to simulate an oscillator startup. Probing through these stages and you get a view how the oscillator starts.
>wrote in message news:73aen9h43rhohnjj85vcd4okdb9sktgd62@4ax.com... >> >>> This is not how an oscillator functions. An oscillator is not amplified >> >noise. > >>Oscillators build up from noise, though, and they do amplify the close-in >>noise very strongly. > >They do, but it is not necessary. A pulse exciting the tank will start the >oscillation, independent of noise. > >A pendulum clock does not start because of noise.
>If you buy a pendulum clock and the transport company very hardly >slams it on your floor (power on transient) it might quite well start >ticking.
>Use a resistor (noise source) followed by 10-100 sections of >amplifiers and band pass filters to simulate an oscillator startup. >Probing through these stages and you get a view how the oscillator >starts.
This can not be done on regular Spice. Resisters only generate noise in AC simulations. There are some Spices with Transient noise extensions, but this still wont allow you to see what might happen in the real world regarding noise. The issue is that Spice is temperamental as to when it will start on its own due to numerical noise or needs a kick start. The Colpitts oscillator in my SuperSpice examples, just starts on its own. In contrast, its often been impossible getting Cadence PSS to start an oscillator when it has specific functions in it to do so. It doesn't even allow you to have any time changing sources on the schematic in PSS runs to start it. So, there would probably be no realistic way to isolate start-up due to transient noise and start-up due to, that's what it does without noise. Noise is in the nV region. I wager in the majority of cases, moving bias conditions on power up, overwhelmingly start up a real oscillator. In fact, some oscillators won't start up at all on power on. Those ones also need a lax removal man. Even with a slow PS, a real system has so many bits and bobs like PORs and devices that there are going to be transient mV/ma floating about that will break the symmetry from a 0=0 loop response. Kevin Aylward B.Sc. www.kevinaylward.co.uk www.anasoft.co.uk - SuperSpice