BJT for the lowest phase noise XTAL oscillator

[stromer]

Re: xtal topology bjt

FEBO oscillator

I believe that the frequency part of FEBO is down to John Ackermann, and he mainly uses uses disciplined oscillators - so you would probably have a Rubidium (maybe even Ceasium?) standard as well as a quartz oscillator.

 

[stromer]

Re: low noise butler oscillator

Hi Ulrich

I found this confusing.
. The title is "Butler" oscillator, but the posted ALC version looks to be from the a Colpitts family?
. Am I interpreting the crystal correctly - 155.5MHz, C1=8.4fF, R1=0.85-Ohms? If so there would be as much loss in the 10k resistor as there is in the crystal.
. Even with the above loss, I would have expected an operating current closer to 1-mA than 4.3-mA. This seems to point to losses in other components - can you clarify?
. Did you compare the performance of this device with that of a current-starved oscillator running under similar conditions and with similar values for the Colpitts capacitors? I would not have expected a significant difference (other than in start-up time).
Thanks
Stromer

---------- Post added at 17:21 ---------- Previous post was at 17:15 ----------

Well, nice simulation which also violates the laws of physic.
Ulrich

Except that a quick look at the site suggests that this is likely to be the output from a "disciplined" oscillator.
If so, the noise will fall was frequency reduces until we either reach the loop gain reaches its maximum or the noise becomes dominated by the reference used to discipline the oscillator.
Stromer

---------- Post added at 17:36 ---------- Previous post was at 17:21 ----------

The best is to run the transistor at 15% of Icmax for best phase noise (flicker noise).
Be warned - these rules of thumb have limited validity, particularly if transistor flicker noise is significant...
For Colpitts designs, there are good reasons for the usual practice to set the base-emitter capacitors substantially larger than the emitter-collector capacitor (emitter-to-ground in Ulrich's ALC circuit).
If you take the output current from the collector, the signal swing at the collector should be smaller than at the base. If not, Miller effects will increase the effective feedback which is (often) equivalent to additional circuit loss. In this case it is advantageous to use a cascode for the output current.
I'm believe that JKAE would agree with all of this.
Follow-on crystal filters with Q substantially reduced by circuit loading are useful for achieving the lowest noise floors - though of course they have no effect on frequencies inside the resonator bandwidth.

---------- Post added at 17:53 ---------- Previous post was at 17:36 ----------

---------- Post added at 18:01 ---------- Previous post was at 17:53 ----------

"However, I need help on the amplitude noise, which is higher than the phase noise at some offsets. Especially the "hump" at 1M-10MHz is troublesome."
There is nothing apparent in your circuit to cause a 1-MHz to 10-MHz peak, though it is possible it is a resonance with a decoupling capacitor. But it is equally likely due to an integrated Voltage regulator somewhwere in the system; this could powering your oscillator or somewhwere in the measurement system. You could try an RC filter in the power line to see if it helps.
Stromer

 

[Dr.Drew]

So the goal of the feedback is to reduce the the DC fluctuation by a factor 10 000 (80dB), the flicker and related noise of the oscillator transistor under large signal condition and therefore reduce the modulation and up conversion within the bandwidth that makes the oscillator "noisy" .
This is standard feedback theory , Ulrich

In this case an substitution of osc transistor noise by DC feedback noise occures. Therefore flicker corner of active device moves from few kHz down to few tens-hundrets Hz. Which value of flicker corner is achievable using modern components? Which transistors are more suitable for DC feedback utilizing an current mirror?

 

[vfone]

It is proven that the type of the transistor in the feedback noise have almost no contribution to the overall flicker noise of the oscillator.
But it may affect somehow the wideband noise of the oscillator.

 

[stromer]

"In this case an substitution of osc transistor noise by DC feedback noise occurs."
Thinking "out loud" to give more detail (not necessarily always entirely in agreement)...
N.B. the "mental reference circuit" for the non-ALC version is a common-emitter Colpitts design

The feedback is of the demodulated signal - so in a well-designed ALC the residual flicker noise should be dominated by the contribution from the amplitude detector.
Why, then is the demodulator less noisy than the oscillator?

. The situation is relatively clear if, in the absence of ALC-the limiting mechanism is collector-base conduction - small changes in amplitude cause large changes in circuit conditions (transistors with short reverse-transit times are less susceptible..).
. If the circuit is operated in current starvation the situation is less clear. Flicker noise will change the current through the bias resistor, and also the collector current, so:
. . If we use resistive feedback, the base current will change the DC collector Voltage. Clearly, this may also be avoided by using a damped inductor for DC feedback (would that this were practical in an IC - we have to use other techniques...).
. . Which potentially leaves the changes in collector current - which in current starvation is directly reflected as a proportionate change in output amplitude. ALC can reduce this variation, as only a small proportion of the signal has to pass through the detector transistors. However, there is another method available - buffer the base current so that your control current drives mainly the collector (you probably want the level shift anyway...).

All this ignores temporal changes in the crystal - but Q changes are at minimum accompanied by frequency changes equal to the change in 1/Q, and amplitude variation on this scale should not significantly impact frequency in a suitably-designed oscillator.

Provisional conclusion:
ALC can reduce both DC shift and amplitude variation, with direct impact on frequency , but
appropriate circuit techniques can apparently provide the same suppression of bipolar transistor flicker noise (and may be less subject to gain ringing - or even squegging)

 

[biff44]

Why, then is the demodulator less noisy than the oscillator?

In simple terms, a transistor in an oscillator has energy going through it over and over and over. Think of how an oscillation builds up--white noise gets amplified in the oscillator transistor, get maybe 10 db higher, then goes thru it again, gets 10 db higher still, ...and so on, until the white noise power has increased to maybe 50 milliwatt level. Then it keeps doing that until the phase noise is at steady state. In all, that microwave power goes thru the device hundreds of times! This happens so fast, with respect to an audio frequency, that the 1/f noise impacts the microwave signal the same way every time thru the oscillator transistor. So, whatever 1/f noise there is, it gets multiplied X times in the oscillator transistor.

If there are other components that have their own 1/f noise, the signal only goes thru them once (no memory effect), so even though the feedback path has some gain, it is not as bad as going thru the oscillator active device.

Rich

 

[stromer]

Why, then is the demodulator less noisy than the oscillator?
"In simple terms, a transistor in an oscillator has energy going through it over and over and over."

??? Rich: That verbal argument can be taken two ways - the other being that each time the signal interacts with the maintaining amplifier the amplitude is corrected towards the available drive; this would suggest that the power dissipated in the resonator (crystal) should correspond precisely to the available drive. This is a in practice a reasonable assumption - the flicker-noise amplitude problem lies with the variation of the available drive, which is one of the two items I address in my post.
There is also an issue with flicker noise modulating base-emitter diffusion capacitance, but that is another story (and one of many reasons I use faster transistors than convention dictates, and suppress amplitude flicker noise in other ways).

 

[biff44]

Every time the signal goes thru the active device, the resonator phase shift interacts with the active device's phase shift, and where they cancel out conjugately--that is the frequency that you get the oscillation to occur at. Unfortunately, since the "gain" of the oscillator circuit is pretty flat (even for a narrowband resonator), and the phase shift is also fairly flat (when you think of a +/-1 Khz sort of range), that feedback mechanism is pretty weak!

The 1/f can disrupt the instantaneous phase of the active device slightly, so the resonator conspires to actually force the frequency momentarily off center--giving phase noise.

That is why feedback correction in an external circuit (such as a bias circcuit with 70 dB of active correctikng gain, or an external resonator or delay line discriminator feedback are so effective--you get far more stabilizing effect than you get from a passive resonator alone.)

 

[Dr.Drew]

As i have understood, stromer, ALC allow to supress an AM flicker-noise at amplifier output. This noise modulates an resonant frequency of quartz resonator which causes an flicker phase noise in output spectrum (AM-to-PM conversion?).

Another effect of flicker noise is an modulation of C-B capacitanse which tends to jitter of amplifier phase characteristic.

Supression of active device flicker noise allow to move flicker corner down. For this an current mirror can be used (schematic has been presented by Ulrich Rohde). But transistor in current mirror has it's own flicker noise. What flicker corner in case of real transistors in current mirror will be? Equal to flicker corner of these transistors?

 

[stromer]

Every time the signal goes thru the active device, the resonator phase shift interacts with the active device's phase shift, and where they cancel out conjugately--that is the frequency that you get the oscillation to occur at. Unfortunately, since the "gain" of the oscillator circuit is pretty flat (even for a narrowband resonator), and the phase shift is also fairly flat (when you think of a +/-1 Khz sort of range), that feedback mechanism is pretty weak!

The 1/f can disrupt the instantaneous phase of the active device slightly, so the resonator conspires to actually force the frequency momentarily off center--giving phase noise.

That is why feedback correction in an external circuit (such as a bias circcuit with 70 dB of active correctikng gain, or an external resonator or delay line discriminator feedback are so effective--you get far more stabilizing effect than you get from a passive resonator alone.)

Rich: Yup, all this and more was already covered - except that
a) Nowhere have we seen any evidence that there is any advantage to be had in increasing the ALC gain beyond about 30-dB (this assumes a sensibly-designed starting-point, and the 30dB is above that of the standard bias circuits);
b) "Phase of active device" needs clarification. Flicker noise directly affects mainly the input impedance, and then the gm (not strictly a phase parameter) and the collector capacitance via bias and signal level variations.
c) Please note that my discussion point that you selected (out-of-context) to answer was an introduction to an alternative approach that removes the effects* of flicker noise at source, rather than trying to post-correct them.
*With the exception of changes in bipolar transistor input impedance - but this is an area that ALC does not address either.
d) A high-quality 10-MHz crystal oscillator relies on a 90-degree phase over about 10-Hz; to my way of thinking that is a great deal more than "not a lot over 1-kHz". Even cooking crystals (cheap compact ATs) deliver Q's iin the 100k range at 10MHz (90-degrees over about 100-Hz).
e) A non-substantive note: In the grounded-emitter Colpitts family of oscillators the phases are seen to sum to 180-degrees - but this is not a difference in substance, as re-assigning the ground (e.g. common collector version) restores your description.

 

[biff44]

70 db comes from either the rhode or wenzel circuit--I forget which one. Obviously, you only need enough feedback control loop gain to degernate the 1/f noise, and any extra just causes trouble.

 

[stromer]

70 db comes from either the rhode or wenzel circuit--I forget which one. Obviously, you only need enough feedback control loop gain to degernate the 1/f noise, and any extra just causes trouble.

That's about right - the DC current loop gain of a resistor-biased transistor with a beta of 100 is already 40-dB!

---------- Post added at 18:13 ---------- Previous post was at 17:47 ----------

As i have understood, stromer, ALC allow to supress an AM flicker-noise at amplifier output. This noise modulates an resonant frequency of quartz resonator which causes an flicker phase noise in output spectrum (AM-to-PM conversion?).
Another effect of flicker noise is an modulation of C-B capacitanse which tends to jitter of amplifier phase characteristic.
Supression of active device flicker noise allow to move flicker corner down. For this an current mirror can be used (schematic has been presented by Ulrich Rohde). But transistor in current mirror has it's own flicker noise. What flicker corner in case of real transistors in current mirror will be? Equal to flicker corner of these transistors?

That is understood - even covered in my text. AM-PM flicker-noise conversion is indeed a significant problem with AT crystals (and particularly with high-Q designs) when they are driven at high levels* (which is why post-filtering is often the used to achieve a low noise-floor).
What I am addressing is whether alc is the only way to reduce drive level fluctuations: - I say not. As bipolar flicker noise modulates only the base current, the requirement is that it does not affect amplitude, collector current, or base-collector Voltage. My proposal is to correct this at source, by ensuring that collector current is not affected by base current.
BTW, you are right that changes in base-emitter impedance will also affect the amplitude, but (compared with current starvation in a Colpitts-family design) ALC does little or nothing to correct this, as the ALC acts on the collector current, not on the circulating current. If the final flicker noise is dominated by changes in emitter-base impedance, the first (standard) nostrum is to increase the Colpitts capacitance in parallel with the emitter-base junction. If that is not sufficient, you need to buffer the input impedance (with an emitter follower). This of course requires particular care...
*The salient feature is elastic strain, which for a given mode equates approximately to Constant*I*Nharmonic/C1 (the approximately because the constant depends on the distribution of vibration across the electrode)

 

[Dr.Drew]

The language barrier affects...

Current mirror allow to reduce only amplitude variations at amplifier output. These variatios make much difficulties in oscillators with drive-level-sensitive resonators. If i try to use it in CRO or similar, i will not see any effect in phase noise?

 

[stromer]

but (compared with current starvation in a Colpitts-family design) ALC does little or nothing to correct this, as the ALC acts on the collector current, not on the circulating current.

I should clarify - I'm referring here to ALC that is driven from the collector current. ALC based on crystal (or "circulating") current gives advantage for variations both of crystal R1 (ESR) and of base input conductance.
An alternative if using controlled collector current plus current starvation might be to provide non-linear series element(s) that compensate the effects of amplitude variation. This can be adjusted (using hyperabrupt varactors) based on the effect of a deliberate small change in collector current.
(BTW, unless I am misreading Ulrich's circuit, it appears to provide DC feedback only - and with no added current gain. If so, its sole advantage over a simple feedback resistor would be to bias the collector at a higher and more constant potential than resistor biasing. Please correct me if I have got this wrong.
Also on Ulrich's circuit - I note that the SPICE flicker noise parameter KF is given as zero in the CEL data sheet; this makes it an unexpected choice - unless flicker noise parameters are available for this device?)

---------- Post added at 13:26 ---------- Previous post was at 13:19 ----------

The language barrier affects...

Current mirror allow to reduce only amplitude variations at amplifier output. These variations make much difficulties in oscillators with drive-level-sensitive resonators. If I try to use it in CRO or similar, I will not see any effect in phase noise?

All resonators are in the end non-linear - for a CRO this might be signal loss -> heat -> thermal expansion.
But usually your drive level will be low enough for this to be irrelevant; in this case the main source of AM to FM conversion will be the non-linearity of junction capacitances - so ALC can also be helpful here.

 

[Dr.Drew]

Spice models for NE688xx in Design Kit from CEL have non-zero Kf.

IMHO CRO's and conventional DRO's have very low Q to observe an AM-PM conversion because of drive level variations. Main effects of flicker noise are in jitter of phase of gain. I've observed that only for leucosapphire oscillators an necessity of ALC were considered.

 

[stromer]

Spice models for NE688xx in Design Kit from CEL have non-zero Kf.

IMHO CRO's and conventional DRO's have very low Q to observe an AM-PM conversion because of drive level variations. Main effects of flicker noise are in jitter of phase of gain. I've observed that only for leucosapphire oscillators an necessity of ALC were considered.

I think we are in agreement about internal dissipation in modest-Q DROs at usual drive levels. However, the phase gain of the maintaining amplifier can be affected by flicker noise in three ways - change in (base-emitter) diffusion capacitance, change in base-emitter conductance, and change in effective collector capacitance (due to amplitude modulation). ALC will help if the last of these three causes is dominant. In all other cases I believe that suppression at source (base buffering and collector-current control) is more helpful. Theoretically at least some of the residual flicker can be compensated, but I have not seen any published work on this.

 

[TonyBo]

Hi,
The crystals I have got from the Czech Republic, seem to be doing the job. I had aging issues with some local product, but this firm seems to supply "what one asks for".

Kevin

Hi Kevin,

Do you have contact info on the Czech company which made SC cut crystals for you?

Thanks,

Tony

 

[stromer]

Do you have contact info on the Czech company which made SC cut crystals for you?

I can't answer for Kevin, but (in case you wish to start searches before he replies) I have heard good reports about Krystaly
They are in Hradec Králové
+420 495 406 687 Sales
krystaly@krystaly.cz
Krystaly, Hradec Králové, a.s.

 

[zl1ujg]

Thats the company.
I dont know if they do SC crystals, but its worth an email.
I havent had any issues, with 20 to 30 overtone crystals ordered at different times for my own use.
Good reliable product, with documentation.
Good communication. No problems
I pay by Telegraphic transfer

I am probably telling people what they already know, but...
When ordering crystals, specify what you want exactly, rather than just saying I want an crystal on XYZ frequency.
Also ask for preaged crystals, otherwise you might as well leave the oscillator running for a year (or more)
Specify the type of loading the crystal sees. Some VHF overtone circuits are used with overtone crystals in series resonance, but other circuits should be using crystals with capacitive loading
The circuit should have some amplitude limiting, outside the oscillator transistor. Otherwise the loaded Q (quality factor) of the crystal can be degraded further.
Kevin

 

[biff44]

Thats

I am probably telling people what they already know, but...
When ordering crystals, specify what you want exactly, rather than just saying I want an crystal on XYZ frequency.
Also ask for preaged crystals, otherwise you might as well leave the oscillator running for a year (or more)
Specify the type of loading the crystal sees. Some VHF overtone circuits are used with overtone crystals in series resonance, but other circuits should be using crystals with capacitive loading
The circuit should have some amplitude limiting, outside the oscillator transistor. Otherwise the loaded Q (quality factor) of the crystal can be degraded further.
Kevin

Interesting points. I would like to learn more about crystal ordering. Could you please post a "typical" specification for a crystal resonator for us to learn from. Thanks.