Colpitts crystal oscillator in LTSPICE

[bitrex]

I'm trying to simulate a low frequency Colpitts oscillator in LTSpice
using an NPN darlington and a crystal, and I'm having a lot of trouble
getting it to start up. Does anyone have a working one of these? I'm
just using a capacitor to simulate the crystal, the parameters I am
working with are something like:

motional capacitance: 5.3ff
ESR = 11k
ESI ~ 4500 henries
EPR = 100 meg
EPC = 1.5pf

Does this seem plausible?

[Phil Hobbs]

The crystal has to be in its inductive region for a Colpitts to work.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal Consultant
ElectroOptical Innovations LLC
Optics, Electro-optics, Photonics, Analog Electronics

160 North State Road #203
Briarcliff Manor NY 10510

hobbs at electrooptical dot net
http://electrooptical.net

[Phil Hobbs]

On 7/25/2015 10:59 AM, Phil Hobbs wrote:
> On 7/25/2015 10:56 AM, bitrex wrote:
>> I'm trying to simulate a low frequency Colpitts oscillator in LTSpice
>> using an NPN darlington and a crystal, and I'm having a lot of trouble
>> getting it to start up. Does anyone have a working one of these? I'm
>> just using a capacitor to simulate the crystal, the parameters I am
>> working with are something like:
>>
>> motional capacitance: 5.3ff
>> ESR = 11k
>> ESI ~ 4500 henries
>> EPR = 100 meg
>> EPC = 1.5pf
>>
>> Does this seem plausible?
>
> The crystal has to be in its inductive region for a Colpitts to work.
>

Oh, and you can make simulated oscillators start faster by putting a
current source in parallel with the inductor. Make it 1 uA or
something, and drop it to zero near the start of the simulation. That's
the SPICE equivalent of plucking a guitar string.

Try it with 1 nA when you're done, to check for startup problems.

[DecadentLinuxUserNumeroUno]

On Sat, 25 Jul 2015 10:56:05 -0400, bitrex
<bit...@de.lete.earthlink.net> Gave us:

Oscillators (of this type) can have what are known (referred to) as
"hard start" issues.

Usually a little inductance and or capacitance on the transistor leg
provides enough lead or lag to allow the oscillations to get going.

[Jim Thompson]

On Sat, 25 Jul 2015 10:56:05 -0400, bitrex

Here's a programmable crystal Spice model...

******************************************************************
.SUBCKT MyXTAL 1 2 PARAMS: Fo=10Meg Cp=5pF Rs=100 Q=25K Series=0
.PARAM P1={6.283185*Fo}
.PARAM P2={Q*Rs}
R1 1 3 {Rs}
C1 1 2 {Cp}
L1 4 2 {P2/P1}
C2 3 4 {1/(P1*P2-(1-Series)/Cp)}
.ENDS MyXTAL
******************************************************************

For Fo the parallel resonance, set Series=0, for series resonance, set
Series=1

(You can often get all these parameters from the crystal datasheet.)

As Phil pointed out most feedback-based oscillators (which queue up in
a linear operating region) need a kick start.

Start with Q low, otherwise you will have timestep issues, and won't
see any oscillation. Then tighten up Q as you make max timestep
smaller.

My rule-of-thumb for oscillators is max timestep = 1/Fo/32

...Jim Thompson
--
| James E.Thompson | mens |
| Analog Innovations | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| San Tan Valley, AZ 85142 Skype: skypeanalog | |
| Voice:(480)460-2350 Fax: Available upon request | Brass Rat |
| E-mail Icon at http://www.analog-innovations.com | 1962 |

I love to cook with wine. Sometimes I even put it in the food.

[bitrex]

Thanks, I'll give that a try.

[bitrex]

Thank you. Usually most oscillators I design get going simply by
setting the option "Start external supply voltages at 0 volts" or
whatever - that gives them enough kick. It doesn't seem to be happening
in this case.

Does the crystal stabilized Colpitts need any other modifications from
the standard version? One can just swap the inductor for a crystal and
it should work the same, right?

[John Larkin]

On Sat, 25 Jul 2015 10:56:05 -0400, bitrex
<bit...@de.lete.earthlink.net> wrote:

Select "skip initial operating point" in the transient analysis, or it
will never start. Or goose it as Phil suggests.

XOs are terrible to sim in time domain. The sims take forever and
there's no reasonable way to measure the frequency to PPM resolution.
Once I get one running, I cut over to AC loop analysis to fine-tune
things.

Post your netlist so people can play with it.


--

John Larkin Highland Technology, Inc
lunatic fringe electronics

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

[Jim Thompson]

That's another way of accomplishing a kick start.

If it doesn't start, the gain/phase criteria isn't met.

>
>Does the crystal stabilized Colpitts need any other modifications from
>the standard version? One can just swap the inductor for a crystal and
>it should work the same, right?

I don't speak buzzword (Colpitts, Pierce, etc) oscillators because
they're rarely used with microchips... most of my clocks are simply a
crystal around an 'HCU04 or equivalent, and my PLL VCO's are AGC'd
differential configuration.... starting with my MC1648 from the mid
'60's.

[Phil Hobbs]

Nope. With an inductor there's only one place it can oscillate. With a
crystal and any external inductance, there are two, one below the series
resonance, and one above the parallel resonance.

Also of course the rate of change of reactance with frequency is doubled
for a crystal, because the L & C are fighting each other.

The reason your usual procedure isn't working here is that the circuit
constants are too extreme--you aren't getting any initial amplitude on
the series LC. Try the current source.

[Phil Hobbs]

On 7/25/2015 11:37 AM, Jim Thompson wrote:

Filed away. Thanks!

[Jim Thompson]

I'm inherently lazy... so I will spend days developing Spice gimmicks
to save a few aggravating minutes time doing tweak, tweak, tweak...
and Spice models... that's how I got into the Spice behavioral
modeling game ;-)

Here's one of several "tanks" I have built into my symbol libraries...

L^@REFDES %pin1 %pin2 {@Zo/(6.283185*@Fo*@Q)}
C^@REFDES %pin1 %pin2 {@Q/(6.283185*@Fo*@Zo)}
R^@REFDES %pin1 %pin2 @Zo

(PSpice can do math inside a symbol without needing a subcircuit call)

In subcircuit jargon this would be:

.SUBCKT TANK pin1 pin2 PARAMS: Fo=1Meg Zo=10K Q=10
** Zo is impedance at resonance
L pin1 pin2 {Zo/(6.283185*Fo*Q)}
C pin1 pin2 {Q/(6.283185*Fo*Zo)}
R pin1 pin2 {Zo}
.ENDS TANK

I also have tanks with inductive and capacitive taps... probably
should do a tuned transformer with everything selectable... to emulate
IF strips ;-)

[Tim Wescott]

Have you checked the thing in open loop to make sure it has a chance of
starting?

Is the transistor biased correctly?

Why a Darlington? That seems to not just be asking for trouble, but
sending a footman with an engraved invitation.

You say you're just using a cap to simulate the crystal -- don't you mean
a cap and an inductor, or (better) two caps and an inductor?

Like John said -- post your circuit, maybe there's something obvious we
can help you with.

--
www.wescottdesign.com

[piglet]

Those values would seem to indicate a 32.768kHz time-keeping kind of
crystal. It might be even the darlington input Z and gain are too low?

Perhaps try to get it to sim-osc with an op-amp first and then go descrete?

piglet

[Tim Wescott]

For that matter, why not choose a rail-rail op-amp in a SOT-23 or smaller
that has known-good large-signal behavior?

Just sayin' -- you can always do better with discretes, but it may take
lots of work to get there.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

[krw]

On Sat, 25 Jul 2015 11:02:56 -0400, Phil Hobbs
<pcdhSpamM...@electrooptical.net> wrote:

>On 7/25/2015 10:59 AM, Phil Hobbs wrote:
>> On 7/25/2015 10:56 AM, bitrex wrote:
>>> I'm trying to simulate a low frequency Colpitts oscillator in LTSpice
>>> using an NPN darlington and a crystal, and I'm having a lot of trouble
>>> getting it to start up. Does anyone have a working one of these? I'm
>>> just using a capacitor to simulate the crystal, the parameters I am
>>> working with are something like:
>>>
>>> motional capacitance: 5.3ff
>>> ESR = 11k
>>> ESI ~ 4500 henries
>>> EPR = 100 meg
>>> EPC = 1.5pf
>>>
>>> Does this seem plausible?
>>
>> The crystal has to be in its inductive region for a Colpitts to work.
>>
>
>Oh, and you can make simulated oscillators start faster by putting a
>current source in parallel with the inductor. Make it 1 uA or
>something, and drop it to zero near the start of the simulation. That's
>the SPICE equivalent of plucking a guitar string.

I used to ramp the supplies to get oscillators to start (and the
simulator converge).

>Try it with 1 nA when you're done, to check for startup problems.
>

Does that really tell much, given component tolerance?

[bitrex]

On 7/25/2015 12:12 PM, John Larkin wrote:
> On Sat, 25 Jul 2015 10:56:05 -0400, bitrex
> <bit...@de.lete.earthlink.net> wrote:
>
>> I'm trying to simulate a low frequency Colpitts oscillator in LTSpice
>> using an NPN darlington and a crystal, and I'm having a lot of trouble
>> getting it to start up. Does anyone have a working one of these? I'm
>> just using a capacitor to simulate the crystal, the parameters I am
>> working with are something like:
>>
>> motional capacitance: 5.3ff
>> ESR = 11k
>> ESI ~ 4500 henries
>> EPR = 100 meg
>> EPC = 1.5pf
>>
>> Does this seem plausible?
>
> Select "skip initial operating point" in the transient analysis, or it
> will never start. Or goose it as Phil suggests.
>
> XOs are terrible to sim in time domain. The sims take forever and
> there's no reasonable way to measure the frequency to PPM resolution.
> Once I get one running, I cut over to AC loop analysis to fine-tune
> things.
>
> Post your netlist so people can play with it.
>
>

Here's a Dropbox link to the files:

https://www.dropbox.com/sh/txmcmhn6p7fhuh2/AACKLnAh-41bI2jO7UT4wjCza?dl=0

I'm actually attempting to use the Darlington buffer inside a LM13700
section to build a Colpitts crystal oscillator, which will then be 2
quadrant multiplied by an external signal within the transconductance
amp itself...

[John Larkin]

I'd guess that C2 and C3 are way too big.

[bitrex]

Lowering C2 and C3 to 470p, making the emitter resistor 4.7k, keeping
the feedback cap the same and setting the crystal resonant frequency to
500kHz makes it start up.

The waveform is nothing to write home about, however.

[John Larkin]

On Sat, 25 Jul 2015 21:17:32 -0400, bitrex

Don't quit now; keep going.

And maybe make the upper one a bit smaller than the lower one. But I
never understood Colpitts oscillators.

I did recently design one using a coaxial ceramic resonator, at 600
MHz. It works fine even if I don't understand it.

[bitrex]

Wes Hayward's book "Radio Frequency Design" has a pretty good section on
analysis of the various LC oscillator topologies on pages 265-290, and
how to plot the gain and phase angle as a function of emitter current.

The book is very math-heavy in general, it's like the anti-AoE.

[bitrex]

There's also apparently a variation on the Colpitts where the crystal is
inserted in the feedback path instead of parallel to the capacitors, to
resonate at the series resonant frequency. I may try that one as well.

[John Larkin]

On Sat, 25 Jul 2015 23:53:34 -0400, bitrex

Spice has made all sorts of math unnecessary. Math tends to sputter
out when things get nonlinear anyhow.

[upsid...@downunder.com]

On Sat, 25 Jul 2015 11:02:56 -0400, Phil Hobbs
<pcdhSpamM...@electrooptical.net> wrote:

>On 7/25/2015 10:59 AM, Phil Hobbs wrote:
>> On 7/25/2015 10:56 AM, bitrex wrote:
>>> I'm trying to simulate a low frequency Colpitts oscillator in LTSpice
>>> using an NPN darlington and a crystal, and I'm having a lot of trouble
>>> getting it to start up. Does anyone have a working one of these? I'm
>>> just using a capacitor to simulate the crystal, the parameters I am
>>> working with are something like:
>>>
>>> motional capacitance: 5.3ff
>>> ESR = 11k
>>> ESI ~ 4500 henries
>>> EPR = 100 meg
>>> EPC = 1.5pf
>>>
>>> Does this seem plausible?
>>
>> The crystal has to be in its inductive region for a Colpitts to work.
>>
>
>Oh, and you can make simulated oscillators start faster by putting a
>current source in parallel with the inductor. Make it 1 uA or
>something, and drop it to zero near the start of the simulation. That's
>the SPICE equivalent of plucking a guitar string.

To satisfy the Barkhausen requirements, shouldn't you model the
startup condition as a linear amplifier and a frequency selective
feedback network ?

Inject some low level white noise (thermal noise) into the amplifier
input and for each cycle around the feedback loop, the amplitude at
the peak frequency will grow, while the relative noise bandwidth
decreases drastically and soon or later you will have a single
spectral line. Finally the amplifier runs in a saturated state with
constant amplitude output.

The amplifier should initially be biased into Class-A so that it will
amplify the weak thermal noise, but of course, when the maximum
amplitude has been reached during startup, it can/will run in Class-C.

If the oscillator is initially biased into Class-C, it is not capable
of amplifying the weak thermal noise and you need some external
kick-start, e.g. rapidly applying the power supply voltage. This also
explains why some oscillators starts nicely, when battery power is
applied, but fails to start when connected to a mains power supply,
which takes a while to start when the storage capacitors are slowly
charging, after plugging it into mains socket.

[Phil Hobbs]

>To satisfy the Barkhausen requirements, shouldn't you model the
>startup condition as a linear amplifier and a frequency selective
>feedback network ?

Sure. That's why I suggested using a much smaller current step omce the sim looks OK, to check for startup problems.

If the initial transient dies away rather than continuing, the gain is too low.

Thing is, being an iterative ODE solver, SPICE uses all sorts of heuristics and fairly coarse convergence checks that make it flaky and unreliable for very small signals. You have to supply an initial amplitude big enough to not get ignored if you want reliable (simulated) startup.

When not using ALC, it's a good idea to arrange the bias so that amplitude limiting happens due to cutoff, which is clean and fast, rather than saturation, which is neither.

Cheers

Phil Hobbs

[upsid...@downunder.com]

On Sun, 26 Jul 2015 01:48:27 -0700 (PDT), Phil Hobbs
<pcdh...@gmail.com> wrote:

>
>>To satisfy the Barkhausen requirements, shouldn't you model the
>>startup condition as a linear amplifier and a frequency selective
>>feedback network ?
>
>Sure. That's why I suggested using a much smaller current step
>omce the sim looks OK, to check for startup problems.

Trying to start a (simulated) oscillator with external transient
either relies on:

a.) the transient puts the amplifier momentarily from Class-C to
Class-A and amplifies everything during startup

b.) the startup transient contains so steep waveforms containing
frequencies in the feedback filter bandwidth that then can be
amplified

> If the initial transient dies away rather than continuing, the gain is too low.

In case b.) the amplifier gain would have to be as high as 100 dB, if
you have a crystal in the feedback path :-)

>Thing is, being an iterative ODE solver, SPICE uses all sorts of heuristics
>and fairly coarse convergence checks that make it flaky and unreliable for
> very small signals.
>You have to supply an initial amplitude big enough to not get ignored if
>you want reliable (simulated) startup.

Have you tried to "unwind" the startup with a number of cascaded
stages ?

You would require a noise source (say a resistor at room temperature
generating -174 dBm/Hz noise temperature) followed by your
amplification stage (transistor with fT limitations) followed by your
oscillator feedback frequency response. Instead of using feedback,
connect the output of the filter to the next identical
amplifier/filter stages.

With 20 dB stages, you would need 10-30 stages, until the amplifier
stages are overdriven.

>When not using ALC, it's a good idea to arrange the bias so that amplitude limiting happens due to cutoff, which is clean and fast, rather than saturation, which is neither.

That is an other kettle of worms how to handle the steady state
situation (phase noise etc.)

>
>Cheers
>
>Phil Hobbs

[upsid...@downunder.com]

Of course, this simulation doesn't solve the other Barkhausen
criterion regarding the phase shift in the feedback loop. If for some
reason (a resonator on the collector side in frequency multipliers)
causes a negative feedback phase on the resonator frequency, the
oscillator doesn't oscillate.

As the old practical wisdom goes, amplifiers oscillate, oscillators
don't oscillate :-)

[Phil Hobbs]

Sure. The amplifier has to be in normal class-A bias with adequate gain
to begin with. It's all the ABSTOL and RELTOL and VOLTTOL and
convergence hacks and stuff, related to SPICE being an iterative solver,
that makes starting up from roundoff noise flaky.

A real oscillator just needs a loop gain larger than 1.0, because it
builds up exponentially from noise. A SPICE oscillator does too, but it
takes awhile and doesn't always succeed unless you get it out of the mud
to start with.

There's nothing wrong with using a (sufficiently small) transient to get
it going. All you care about for startup purposes is the gain in the
small signal regime, not necessarily at 0.000000000000000000... volts.

[Phil Hobbs]

On 7/26/2015 6:57 AM, upsid...@downunder.com wrote:
> On Sun, 26 Jul 2015 01:48:27 -0700 (PDT), Phil Hobbs
> <pcdh...@gmail.com> wrote:
>
>>
>>> To satisfy the Barkhausen requirements, shouldn't you model the
>>> startup condition as a linear amplifier and a frequency
>>> selective feedback network ?
>>
>> Sure. That's why I suggested using a much smaller current step omce
>> the sim looks OK, to check for startup problems.
>
> Trying to start a (simulated) oscillator with external transient
> either relies on:
>
> a.) the transient puts the amplifier momentarily from Class-C to
> Class-A and amplifies everything during startup

It had better be Class A at startup, or you're guaranteed a lot of
midnight phone calls. Have you ever shipped an oscillator that wasn't
Class A in quiescent conditions?

>
> b.) the startup transient contains so steep waveforms containing
> frequencies in the feedback filter bandwidth that then can be
> amplified
>
>> If the initial transient dies away rather than continuing, the gain
>> is too low.
>
> In case b.) the amplifier gain would have to be as high as 100 dB,
> if you have a crystal in the feedback path :-)

No, you misunderstand what I'm proposing. If you put a current source
in parallel with the inductor, then at t=0 the full current is going
through the crystal inductance. When you turn that off, the crystal
rings strongly at its series resonance. A microamp of crystal current
in a watch crystal is probably full amplitude, or nearly.

>
>
>> Thing is, being an iterative ODE solver, SPICE uses all sorts of
>> heuristics and fairly coarse convergence checks that make it flaky
>> and unreliable for very small signals. You have to supply an
>> initial amplitude big enough to not get ignored if you want
>> reliable (simulated) startup.
>
> Have you tried to "unwind" the startup with a number of cascaded
> stages ?

No. What would I learn from that?

>
> You would require a noise source (say a resistor at room temperature
> generating -174 dBm/Hz noise temperature) followed by your
> amplification stage (transistor with fT limitations) followed by
> your oscillator feedback frequency response. Instead of using
> feedback, connect the output of the filter to the next identical
> amplifier/filter stages.

But they wouldn't be indentical. The phases would be independent, so
there would be nothing to enforce the oscillation criterion. All you'd
get would be filtered white noise, whereas a good oscillator has a much,
much narrower line width than the crystal.

> With 20 dB stages, you would need 10-30 stages, until the amplifier
> stages are overdriven.

20 dB is too much stage gain for a good oscillator anyway--the resonator
amplitude will be too small.

>
>> When not using ALC, it's a good idea to arrange the bias so that
>> amplitude limiting happens due to cutoff, which is clean and fast,
>> rather than saturation, which is neither.
>
> That is an other kettle of worms how to handle the steady state
> situation (phase noise etc.)

Phase noise is the main reason you do it. If the amplifier is
nonlinear, all of its baseband crap and 1/f noise gets intermodulated
with the output signal.

[Phil Hobbs]

Oscillator and high frequency amplifier design are special cases, where
grinding out the math is important. Even in a classical Colpitts, the
choice of collector current, tank impedance, the ratio of the tank caps,
and the size of the emitter and base coupling caps have a lot of
influence on frequency stability and phase noise.

That's a 5-D search space, so getting it right can take awhile
numerically, and you can never really be sure you're not on a local optimum.

You only have to do it once for each topology, then you just use Mathcad
or Mathematica or something to apply it to the case at hand.

[Phil Hobbs]

I like the standard Colpitts, because you can take the output right from
the crystal, which gives you a nice wave shape for the second stage, and
less broadband noise. Also capacitors are more stable than inductors,
and you can't pull the oscillator off the crystal resonance the way you
can with a separate tank.

One good use of inductors is as overtone selectors: you resonate one of
the tank caps so that it looks like the right size capacitor at only one
of the overtones. Parzen's crystal oscillator book is a good read on
that sort of thing.

[Jim Thompson]

On Sat, 25 Jul 2015 10:56:05 -0400, bitrex
<bit...@de.lete.earthlink.net> wrote:

>I'm trying to simulate a low frequency Colpitts oscillator in LTSpice
>using an NPN darlington and a crystal, and I'm having a lot of trouble
>getting it to start up. Does anyone have a working one of these? I'm
>just using a capacitor to simulate the crystal, the parameters I am
>working with are something like:
>
>motional capacitance: 5.3ff
>ESR = 11k
>ESI ~ 4500 henries
>EPR = 100 meg
>EPC = 1.5pf
>
>Does this seem plausible?

My favorite crystal oscillator structure...

<http://www.analog-innovations.com/SED/XtalSeriesOsc.pdf>

No fiddle, no fuss, no critical matching... always works.

First used it to get to 144MHz (2m) by X3 (transistor), X3 (varactor),
from 16MHz crystal for my first (and last ;-) 2m rig... when in my mid
'20's.

[John S]

On 7/25/2015 9:56 AM, bitrex wrote:
> I'm trying to simulate a low frequency Colpitts oscillator in LTSpice
> using an NPN darlington and a crystal, and I'm having a lot of trouble
> getting it to start up. Does anyone have a working one of these? I'm
> just using a capacitor to simulate the crystal, the parameters I am
> working with are something like:
>
> motional capacitance: 5.3ff
> ESR = 11k
> ESI ~ 4500 henries
> EPR = 100 meg
> EPC = 1.5pf
>
> Does this seem plausible?

See if this works for you

Version 4
SHEET 1 1560 740
WIRE 176 -192 -1056 -192
WIRE 416 -144 -816 -144
WIRE 784 -144 416 -144
WIRE -1056 -112 -1056 -192
WIRE 784 -112 784 -144
WIRE 416 -80 416 -144
WIRE 784 -16 784 -32
WIRE -816 16 -816 -144
WIRE -528 16 -624 16
WIRE -368 16 -448 16
WIRE -224 16 -304 16
WIRE -64 16 -144 16
WIRE 176 64 176 -192
WIRE 416 64 416 0
WIRE 416 64 176 64
WIRE -816 144 -816 96
WIRE -624 144 -624 16
WIRE -624 144 -816 144
WIRE -816 160 -816 144
WIRE 416 160 416 64
WIRE -1056 208 -1056 -48
WIRE -960 208 -1056 208
WIRE -880 208 -960 208
WIRE -624 208 -624 144
WIRE -352 208 -624 208
WIRE -64 208 -64 16
WIRE -64 208 -288 208
WIRE 176 208 -64 208
WIRE 272 208 176 208
WIRE 352 208 272 208
WIRE -880 224 -880 208
WIRE -864 224 -880 224
WIRE 352 224 352 208
WIRE 368 224 352 224
WIRE -960 240 -960 208
WIRE 272 240 272 208
WIRE -1056 320 -1056 208
WIRE 176 320 176 208
WIRE 416 320 416 256
WIRE -816 368 -816 256
WIRE 416 464 416 400
WIRE 416 464 336 464
WIRE -1056 480 -1056 400
WIRE -960 480 -960 304
WIRE -960 480 -1056 480
WIRE 176 480 176 400
WIRE 272 480 272 304
WIRE 272 480 176 480
WIRE 336 496 336 464
WIRE 416 496 416 464
WIRE 176 512 176 480
WIRE -1056 528 -1056 480
WIRE -816 544 -816 448
WIRE 336 624 336 560
WIRE 416 624 416 576
WIRE 416 624 336 624
WIRE 416 672 416 624
FLAG 416 672 0
FLAG 784 -16 0
FLAG -816 544 0
FLAG -1056 528 0
FLAG 176 512 0
SYMBOL ind -240 32 R270
WINDOW 0 32 56 VTop 2
WINDOW 3 5 56 VBottom 2
SYMATTR InstName Lm
SYMATTR Value 1701.17Hy
SYMBOL res -544 32 R270
WINDOW 0 32 56 VTop 2
WINDOW 3 0 56 VBottom 2
SYMATTR InstName Rs
SYMATTR Value 15k
SYMBOL cap -368 32 R270
WINDOW 0 32 32 VTop 2
WINDOW 3 0 32 VBottom 2
SYMATTR InstName Cm
SYMATTR Value 1.489fF
SYMBOL cap -352 224 R270
WINDOW 0 32 32 VTop 2
WINDOW 3 0 32 VBottom 2
SYMATTR InstName Cs
SYMATTR Value 1.743pF
SYMBOL voltage 784 -128 R0
WINDOW 123 24 132 Left 2
WINDOW 39 0 0 Left 2
SYMATTR InstName V1
SYMATTR Value 9
SYMBOL res 400 -96 R0
SYMATTR InstName R1
SYMATTR Value 2.2k
SYMBOL njf 368 160 R0
SYMATTR InstName J1
SYMATTR Value U309
SYMBOL res 400 480 R0
SYMATTR InstName R2
SYMATTR Value 680
SYMBOL cap 320 496 R0
SYMATTR InstName C2
SYMATTR Value .01u
SYMBOL res 160 304 R0
SYMATTR InstName R3
SYMATTR Value 10meg
SYMBOL cap -1072 -112 R0
SYMATTR InstName C4
SYMATTR Value .01u
SYMBOL diode 256 240 R0
SYMATTR InstName D1
SYMATTR Value 1N4148
SYMBOL njf -864 160 R0
SYMATTR InstName J2
SYMATTR Value U309
SYMBOL res -832 352 R0
SYMATTR InstName R4
SYMATTR Value 1.8k
SYMBOL res -1072 304 R0
SYMATTR InstName R5
SYMATTR Value 10meg
SYMBOL diode -976 240 R0
SYMATTR InstName D2
SYMATTR Value 1N4148
SYMBOL res -832 0 R0
SYMATTR InstName R6
SYMATTR Value 1.8k
SYMBOL res 400 304 R0
SYMATTR InstName R7
SYMATTR Value 470
TEXT -392 -128 Left 2 ;Crystal
TEXT 1200 200 Left 2 !.tran 0 2 1.9 startup
RECTANGLE Normal -48 320 -656 -96 2

[Phil Hobbs]

On 7/26/2015 12:02 PM, Jim Thompson wrote:
> On Sat, 25 Jul 2015 10:56:05 -0400, bitrex
> <bit...@de.lete.earthlink.net> wrote:
>
>> I'm trying to simulate a low frequency Colpitts oscillator in LTSpice
>> using an NPN darlington and a crystal, and I'm having a lot of trouble
>> getting it to start up. Does anyone have a working one of these? I'm
>> just using a capacitor to simulate the crystal, the parameters I am
>> working with are something like:
>>
>> motional capacitance: 5.3ff
>> ESR = 11k
>> ESI ~ 4500 henries
>> EPR = 100 meg
>> EPC = 1.5pf
>>
>> Does this seem plausible?
>
> My favorite crystal oscillator structure...
>
> <http://www.analog-innovations.com/SED/XtalSeriesOsc.pdf>
>
> No fiddle, no fuss, no critical matching... always works.
>
> First used it to get to 144MHz (2m) by X3 (transistor), X3 (varactor),
> from 16MHz crystal for my first (and last ;-) 2m rig... when in my mid
> '20's.
>
> ...Jim Thompson
>

I tend not to put DC on crystals, though modern ones may be less sensitive.

One thing that series resonant oscillators are especially good for is
ring-down calibrators for logarithmic amplitude detectors (DLVAs).
Exponential envelope decay -> linear ramp output (ideally).

Running at series resonance maximizes the initial amplitude of the
envelope when the oscillator switches off. One of those will calibrate
an SA615 good to at least 12 bits, usually more.

[Jim Thompson]

On Sun, 26 Jul 2015 12:26:21 -0400, Phil Hobbs
<pcdhSpamM...@electrooptical.net> wrote:

>On 7/26/2015 12:02 PM, Jim Thompson wrote:
>> On Sat, 25 Jul 2015 10:56:05 -0400, bitrex
>> <bit...@de.lete.earthlink.net> wrote:
>>
>>> I'm trying to simulate a low frequency Colpitts oscillator in LTSpice
>>> using an NPN darlington and a crystal, and I'm having a lot of trouble
>>> getting it to start up. Does anyone have a working one of these? I'm
>>> just using a capacitor to simulate the crystal, the parameters I am
>>> working with are something like:
>>>
>>> motional capacitance: 5.3ff
>>> ESR = 11k
>>> ESI ~ 4500 henries
>>> EPR = 100 meg
>>> EPC = 1.5pf
>>>
>>> Does this seem plausible?
>>
>> My favorite crystal oscillator structure...
>>
>> <http://www.analog-innovations.com/SED/XtalSeriesOsc.pdf>
>>
>> No fiddle, no fuss, no critical matching... always works.
>>
>> First used it to get to 144MHz (2m) by X3 (transistor), X3 (varactor),
>> from 16MHz crystal for my first (and last ;-) 2m rig... when in my mid
>> '20's.
>>
>> ...Jim Thompson
>>
>I tend not to put DC on crystals, though modern ones may be less sensitive.

No DC there... each end of the XTAL is 1*Vbe down from rail

>
>One thing that series resonant oscillators are especially good for is
>ring-down calibrators for logarithmic amplitude detectors (DLVAs).
>Exponential envelope decay -> linear ramp output (ideally).
>
>Running at series resonance maximizes the initial amplitude of the
>envelope when the oscillator switches off. One of those will calibrate
>an SA615 good to at least 12 bits, usually more.
>
>Cheers
>
>Phil Hobbs

[John Larkin]

I can usually find a paper or a book, where someone has done the math
for me. It's amazing how little serious math I need to do electronic
design; not much past algebra. The only really hairy stuff that we do
is digital filter design, in FPGAs, and quantization issues make
accurate analytical solutions impossible, so we wind up simulating the
final filters anyhow.

[John Larkin]

On Sun, 26 Jul 2015 09:02:04 -0700, Jim Thompson
<To-Email-Use-Th...@On-My-Web-Site.com> wrote:

>On Sat, 25 Jul 2015 10:56:05 -0400, bitrex
><bit...@de.lete.earthlink.net> wrote:
>
>>I'm trying to simulate a low frequency Colpitts oscillator in LTSpice
>>using an NPN darlington and a crystal, and I'm having a lot of trouble
>>getting it to start up. Does anyone have a working one of these? I'm
>>just using a capacitor to simulate the crystal, the parameters I am
>>working with are something like:
>>
>>motional capacitance: 5.3ff
>>ESR = 11k
>>ESI ~ 4500 henries
>>EPR = 100 meg
>>EPC = 1.5pf
>>
>>Does this seem plausible?
>
>My favorite crystal oscillator structure...
>
><http://www.analog-innovations.com/SED/XtalSeriesOsc.pdf>
>
>No fiddle, no fuss, no critical matching... always works.
>
>First used it to get to 144MHz (2m) by X3 (transistor), X3 (varactor),
>from 16MHz crystal for my first (and last ;-) 2m rig... when in my mid
>'20's.
>
> ...Jim Thompson

That is essentially the "two transistor Butler" topology. I think I've
seen it with an inductor across the crystal to null out the parallel
capacitance. For high precision, you have to keep the xtal current
low, which is a compromise.

[upsid...@downunder.com]

On Sun, 26 Jul 2015 07:55:27 -0400, Phil Hobbs
<ho...@electrooptical.net> wrote:

>On 7/26/2015 6:57 AM, upsid...@downunder.com wrote:
>> On Sun, 26 Jul 2015 01:48:27 -0700 (PDT), Phil Hobbs
>> <pcdh...@gmail.com> wrote:
>>
>>>
>>>> To satisfy the Barkhausen requirements, shouldn't you model the
>>>> startup condition as a linear amplifier and a frequency
>>>> selective feedback network ?
>>>
>>> Sure. That's why I suggested using a much smaller current step omce
>>> the sim looks OK, to check for startup problems.
>>
>> Trying to start a (simulated) oscillator with external transient
>> either relies on:
>>
>> a.) the transient puts the amplifier momentarily from Class-C to
>> Class-A and amplifies everything during startup
>
>It had better be Class A at startup, or you're guaranteed a lot of
>midnight phone calls. Have you ever shipped an oscillator that wasn't
>Class A in quiescent conditions?

I have seen oscillator designs with some kick-start pulse delivered
through a diode or some kind of power on reset type circuit that slams
on the Vcc when a sufficient voltage has been stabilized from the
mains PSU, simulating the turn on transient from battery powered
devices :-)

>
>>
>> b.) the startup transient contains so steep waveforms containing
>> frequencies in the feedback filter bandwidth that then can be
>> amplified
>>
>>> If the initial transient dies away rather than continuing, the gain
>>> is too low.
>>
>> In case b.) the amplifier gain would have to be as high as 100 dB,
>> if you have a crystal in the feedback path :-)
>
>No, you misunderstand what I'm proposing. If you put a current source
>in parallel with the inductor, then at t=0 the full current is going
>through the crystal inductance. When you turn that off, the crystal
>rings strongly at its series resonance. A microamp of crystal current
>in a watch crystal is probably full amplitude, or nearly.
>
>>
>>
>>> Thing is, being an iterative ODE solver, SPICE uses all sorts of
>>> heuristics and fairly coarse convergence checks that make it flaky
>>> and unreliable for very small signals. You have to supply an
>>> initial amplitude big enough to not get ignored if you want
>>> reliable (simulated) startup.
>>
>> Have you tried to "unwind" the startup with a number of cascaded
>> stages ?
>
>No. What would I learn from that?

At least it would educational to those that haven't thought about
oscillators as amplifiers with positive frequency selective feedback.

>> You would require a noise source (say a resistor at room temperature
>> generating -174 dBm/Hz noise temperature) followed by your
>> amplification stage (transistor with fT limitations) followed by
>> your oscillator feedback frequency response. Instead of using
>> feedback, connect the output of the filter to the next identical
>> amplifier/filter stages.
>
>But they wouldn't be indentical. The phases would be independent, so
>there would be nothing to enforce the oscillation criterion. All you'd
>get would be filtered white noise, whereas a good oscillator has a much,
>much narrower line width than the crystal.

Yes, I pointed out that in my follow-up message.

[Phil Hobbs]

Ah, right, quite so. You IC guys and your weird-ass biasing schemes. ;)

[Phil Hobbs]

The parallel inductor is very helpful if you want to run at a high overtone.

[Jim Thompson]

On Sun, 26 Jul 2015 13:00:01 -0400, Phil Hobbs
<ho...@electrooptical.net> wrote:

>On 7/26/2015 12:30 PM, Jim Thompson wrote:
>> On Sun, 26 Jul 2015 12:26:21 -0400, Phil Hobbs
>> <pcdhSpamM...@electrooptical.net> wrote:
>>

[snip]

>>>>
>>>> My favorite crystal oscillator structure...
>>>>
>>>> <http://www.analog-innovations.com/SED/XtalSeriesOsc.pdf>
>>>>
>>>> No fiddle, no fuss, no critical matching... always works.
>>>>
>>>> First used it to get to 144MHz (2m) by X3 (transistor), X3 (varactor),
>>>> from 16MHz crystal for my first (and last ;-) 2m rig... when in my mid
>>>> '20's.
>>>>
>>>> ...Jim Thompson
>>>>
>>> I tend not to put DC on crystals, though modern ones may be less sensitive.
>>
>> No DC there... each end of the XTAL is 1*Vbe down from rail
>>

[snip]

>>>
>>> Cheers
>>>
>>> Phil Hobbs
>>
>> ...Jim Thompson
>>
>Ah, right, quite so. You IC guys and your weird-ass biasing schemes. ;)
>
>Cheers
>
>Phil Hobbs

I think our _brains_ are "biased" differently >:-}

[Jim Thompson]

On Sun, 26 Jul 2015 19:59:27 +0300, upsid...@downunder.com wrote:

[snip]

>
>I have seen oscillator designs with some kick-start pulse delivered
>through a diode or some kind of power on reset type circuit that slams
>on the Vcc when a sufficient voltage has been stabilized from the
>mains PSU, simulating the turn on transient from battery powered
>devices :-)
>

[snip]

I know a low-power IC designer who relies on kick-start schemes
initiated by a uP... but no check for success. I think he's an idiot

[John Larkin]

On Sun, 26 Jul 2015 13:00:01 -0400, Phil Hobbs
<ho...@electrooptical.net> wrote:

You can do a lot of fun stuff if you don't mind using 70 transistors
to do it.

Sometimes an IC data sheet will have a transistor count. The numbers
can be amazing.

[Phil Hobbs]

On 7/26/2015 1:11 PM, Jim Thompson wrote:
> On Sun, 26 Jul 2015 19:59:27 +0300, upsid...@downunder.com wrote:
>
> [snip]
>>
>> I have seen oscillator designs with some kick-start pulse delivered
>> through a diode or some kind of power on reset type circuit that slams
>> on the Vcc when a sufficient voltage has been stabilized from the
>> mains PSU, simulating the turn on transient from battery powered
>> devices :-)
>>
> [snip]
>
> I know a low-power IC designer who relies on kick-start schemes
> initiated by a uP... but no check for success. I think he's an idiot
>> :-}

Sounds like it. There are enough non-obvious ways to fail that there's
no reason to include the obvious ones.

[Phil Hobbs]

Within an ace of saturation, like the oscillator. ;)

[Phil Hobbs]

It's really important to avoid saturating BJT oscillators. The
bias-dependent time delay alone is enough to ruin the stability and
phase noise, not counting limit cycles and so on.

[bitrex]

I'm pretty skilled with Numpy and SciPy. Do you have an interesting set
of equations I can set it to work grinding on?

[Phil Hobbs]

Not right handy--I haven't done a crystal oscillator in years.

[Jim Thompson]

On Sun, 26 Jul 2015 13:34:57 -0400, Phil Hobbs

That's why the gold-doped 2N2369's.

My chip versions use level measurement and current source control...
much better spectral purity.

[John Larkin]

It's really easier to buy oscillators. The phase noise and suchlike
specs are right there already. A pretty good OCXO can be cheaper than
buying just an SC-cut crystal.

--

John Larkin Highland Technology, Inc

picosecond timing precision measurement

[Phil Hobbs]

Yup. I mostly build oscillators in protos these days, or if I need to
do something fancy, such as tuning it faster than the RC network on the
varactor wants to let me, or getting more than an octave of tuning.
(Both from recent examples--LC not crystal, obviously.)

[John Larkin]

VCOs tend to have miserable modulation bandwidths, if specified at
all.

I'd like to make or buy a VCO that runs at 1.5 or 2 GHz maybe, with
500 MHz mod bandwidth. Looks difficult.

[Phil Hobbs]

On 7/27/2015 6:59 PM, John Larkin wrote:
> On Mon, 27 Jul 2015 17:30:35 -0400, Phil Hobbs
> <pcdhSpamM...@electrooptical.net> wrote:
>
>>>>
>>>
>>> It's really easier to buy oscillators. The phase noise and suchlike
>>> specs are right there already. A pretty good OCXO can be cheaper than
>>> buying just an SC-cut crystal.
>>>
>>>
>> Yup. I mostly build oscillators in protos these days, or if I need to
>> do something fancy, such as tuning it faster than the RC network on the
>> varactor wants to let me, or getting more than an octave of tuning.
>> (Both from recent examples--LC not crystal, obviously.)
>>
>> Cheers
>>
>> Phil Hobbs
>
> VCOs tend to have miserable modulation bandwidths, if specified at
> all.

A lot of them have time constants in the 10-100 microsecond range, even
for a UHF VCO. Pathetic.

>
> I'd like to make or buy a VCO that runs at 1.5 or 2 GHz maybe, with
> 500 MHz mod bandwidth. Looks difficult.

There are a bunch of parametric effects that you have to worry about in
a case like that. What do you want it for?

[John Larkin]

On Mon, 27 Jul 2015 21:22:40 -0400, Phil Hobbs

<pcdhSpamM...@electrooptical.net> wrote:

>On 7/27/2015 6:59 PM, John Larkin wrote:
>> On Mon, 27 Jul 2015 17:30:35 -0400, Phil Hobbs
>> <pcdhSpamM...@electrooptical.net> wrote:
>>
>>>>>
>>>>
>>>> It's really easier to buy oscillators. The phase noise and suchlike
>>>> specs are right there already. A pretty good OCXO can be cheaper than
>>>> buying just an SC-cut crystal.
>>>>
>>>>
>>> Yup. I mostly build oscillators in protos these days, or if I need to
>>> do something fancy, such as tuning it faster than the RC network on the
>>> varactor wants to let me, or getting more than an octave of tuning.
>>> (Both from recent examples--LC not crystal, obviously.)
>>>
>>> Cheers
>>>
>>> Phil Hobbs
>>
>> VCOs tend to have miserable modulation bandwidths, if specified at
>> all.
>
>A lot of them have time constants in the 10-100 microsecond range, even
>for a UHF VCO. Pathetic.
>
>>
>> I'd like to make or buy a VCO that runs at 1.5 or 2 GHz maybe, with
>> 500 MHz mod bandwidth. Looks difficult.
>
>There are a bunch of parametric effects that you have to worry about in
>a case like that. What do you want it for?
>
>Cheers
>
>Phil Hobbs

I want to transport some analog signals out of a high EMI environment,
over fiber. The customer wants at least 500 MHz bandwidth. Think
fiber-optic scope probe.

I got some fiber-coupled LEDs, which will modulate (baseband, pure
analog) nicely without mode jumps or interference wobbles, but they
are too slow. I was thinking PWM or FM with a laser, but the FM thing
will be difficult, as you note. Blindingly fast PWM is a possibility.
The budget doesn't include lithium niobate modulators.

[Phil Hobbs]

Near a CO2 laser?

One approach is to current-tune a laser, and demodulate it using a delay
discriminator. You can use the delay discrim for AFC as well, just to
maintain an operating point. The AFC can be really slow, so you can
filter the daylights out of it. The delay discriminator will get you a
full range signal for a small tuning input, but of course it'll be
sinusoidal, so it'll need calibrating.

That can all be done easily with connectorized fibre parts--you just
need two 50:50 couplers with a few centimetres of path difference.

Another approach would be to put AM on a subcarrier, and modulate the
diode with that.

[John Larkin]

On Tue, 28 Jul 2015 13:53:56 -0400, Phil Hobbs

No. I don't know the actual application, other than the EMI
environment is extreme. I assume that battery power should be an
option. But we get occasional requests for this sort of thing, so I'm
playing with it. Slower stuff we could just digitize.

>
>One approach is to current-tune a laser, and demodulate it using a delay
>discriminator. You can use the delay discrim for AFC as well, just to
>maintain an operating point. The AFC can be really slow, so you can
>filter the daylights out of it. The delay discriminator will get you a
>full range signal for a small tuning input, but of course it'll be
>sinusoidal, so it'll need calibrating.
>
>That can all be done easily with connectorized fibre parts--you just
>need two 50:50 couplers with a few centimetres of path difference.
>
>Another approach would be to put AM on a subcarrier, and modulate the
>diode with that.

We might AM modulate a laser directly, if we can avoid mode jumps and
interferance effects. FM or PWM would make gain independent of fiber
and connector losses, which would be nice.

[Phil Hobbs]

You can't do that over a very wide amplitude range, though, whereas with
FM you get a zero-centred output basically for free, and the tuning
range is narrow enough (some fraction of a wave number, maybe 10 GHz
p-p) that you don't have to worry about mode hops much.

Connectorized fibre couplers are pretty cheap if you get them from
China, and they'll make you the unbalanced Mach-Zehnder as a single
small assembly, complete with two photodiodes, for a reasonable price in
quantities of > 10 or so. (Check out Photop, who are the ones I've
used. Great outfit.)

[upsid...@downunder.com]

On Mon, 27 Jul 2015 15:59:25 -0700, John Larkin
<jla...@highlandtechnology.com> wrote:

>
>I'd like to make or buy a VCO that runs at 1.5 or 2 GHz maybe, with
>500 MHz mod bandwidth. Looks difficult.

Run the VCO at say 4.5-5 GHz, mix it down with a fixed 3 GHz
oscillator to get 1.5-2 GHz. It should be fairly simple to filter out
the sum (7.5-8 GHz). The 3 GHz is outside both the desired frequency
range as well as the actual VCO range.

[upsid...@downunder.com]

What did they used to send analog CATV signals over long distances
using fibre ? Apparently they just amplitude modulated the light using
the full CATV downlink frequency range.

[John Larkin]

I think they generally use lithium niobate Mach-Zender (interferance
based) modulators, which tend to be expensive.