Crystal Oscillator without Load Caps

[MrAl]

Hello there,


Has anyone here worked with a crystal oscillator without the additional two load caps that are normally used for parallel resonant oscillation?
What i wanted to investigate was the frequency STABILITY, not the ACCURACY relative to the frequency stamped onto the crystal package.
In other words, did you notice that say a 50PPM crystal worked at or better than 50PPM with the caps, or without the caps?
Note again that i am not talking about the accuracy of the crystal. It's ok if the frequency is quite a bit off i just want to investigate the stability over temperature with and without the caps. As many of you already know, the caps themselves pull the frequency down a little, and that means that the oscillation temperature stability depends partly on the cap values themselves as to how they change over temperature.

I think this topic was talked about before somewhere but not sure where.

 

[RCinFLA]

Unless the oscillator circuit is almost intentionally badly designed the temperature tracking should not deviate more then about 20% of crystal's TC characteristics. This of course gets tougher for graded AT crystals with less then 10 ppm TC characteristics.

The PPM pulling depends on quartz cut, frequency, and mode of operation. All overtone modes are run in series resonant mode. Fundamental oscillation mode can be designed for series or parallel mode, most often parallel mode since no coils are required.

Series mode operation does not imply the crystal is operating at its series resonant freq. Almost always it is operated below crystal series resonance to avoid crystal spurs that exist above crystal's series resonant frequency.

For parallel mode intended operation, the stamped frequency represent the frequency (with manufacturer's tolerance) with crystal spec'd load capacitance. This specified load capacitance can be anything from 7 pF to 32 pF. Most parallel AT cuts are now specified for 20 pF load. AT cut parallel resonance pulling relative to crystals series resonant mode depends on specified load capacitance. 20 pF load generally run about 20 to 50 ppm below series resonance. Preferred high frequency limit for parallel AT cut is about 40 MHz. It is almost hard to manufacture an AT cut crystal with worse then a 60 ppm TC these days. Most run in the 20-30 ppm TC range from -10C to +50C.

For CMOS oscillator design the more the crystal load capacitance, the more the power drain by oscillator. 32.768 kHz BT cut clock crystals are typically spec'd for less then 10 pF load. Many cheap clocks do not account for PCB layout capacitance which can be a few pF. If you see two 20 pF feedback caps in circuit chances are it is a 10 pF load crystal and the PCB & I.C. capacitance has not been taken into account so oscillator runs low in frequency.

Eliminating loading caps is not going to make oscillator more stable over temp. It can even make it worse since oscillator amp input and output loading variation over temp will have more deviating effect. Swamping out the amps variation with a good external cap improves stability over temp. Also, the oscillator needs a reliable feedback ratio'g network which the external caps provide. Without a stable feedback ratio the loop gain of the oscillator can range widely over temp resulting in more freq variation, startup issues, and perhaps even cause oscillator to quit over temp. The quartz can even fracture if there is too much drive. 32 kHz watch crystals are very vunerable to overdriving.

For poorly manufactured crystals, aging can be more of a problem then TC. When it comes to I.C. and crystal manufacturing cleaniness is name of game. Quartz 'dust' particles and poor electrode spot depositions can flake off, mounts can relax, over time changing crystal frequency, usually higher in freq.

 

[Diver300]

All overtone modes are run in series resonant mode. Fundamental oscillation mode can be designed for series or parallel mode, most often parallel mode since no coils are required.

That's not quite right. Any adjustable oscillator will vary the load, so even with a series resonant crystal, the adjustment will take the load slightly to the inductive or capacitive side of the series resonant point.

Also, the company I ran made several thousand overtone oscillators that ran the crystals at about 10 - 12 pF load.

The oscillators were Colpitts, with an inductor and capacitor in the emitter circuit to prevent fundamental oscillation. Colpitts oscillators need a capacitive load in the emitter circuit, and the inductor and capacitor were tuned to somewhere between fundamental and third overtone. That meant that the tuned circuit appeared inductive at fundamental frequencies, so the oscillator wouldn't run like that. At third overtone, the tuned circuit appeared capacitive, allowing the circuit to run.

Like all Colpitts oscillators, the circuit from the base to ground has to be inductive. In my circuit, that was the crystal, running at about a 10 - 12 pF load.

I know that a lot of overtone designs run at series, and the inductors that are used to prevent the oscillator running at fundamental frequencies also make the crystal run at the series resonant point. Also, overtone crystals have much less pulling than fundamental crystals, so getting the load capacitance wrong with an overtone crystal will often only result in a small frequency shift, so it may go unnoticed.

20 pF load generally run about 20 to 50 ppm below series resonance

. A crystal will always run at a higher frequency when it is running with a load capacitance than when it is running at series resonance. However, the crystals are adjusted for a specific load capacitance, so a crystal adjusted for 20 pF will run lower than one adjusted for the same frequency but at series resonance, when running in the same circuit.

The frequency shift can often be a lot more than 50 ppm. We used to regularly make voltage controlled crystal oscillators that moved by +/-250 ppm minimum, so they changed by over 500 ppm.

Series mode operation does not imply the crystal is operating at its series resonant freq. Almost always it is operated below crystal series resonance to avoid crystal spurs that exist above crystal's series resonant frequency.

Series mode isn't actually distinct from parallel mode. Series mode is simply the point where the crystal's current and voltage are in phase. At slightly lower frequencies the crystal appears capacitive, so needs and inductive load, and at slightly higher frequencies the crystal appears inductive and needs a capacitive load. Voltage controlled oscillators will often change the crystal load from inductive all the way though series resonant to capacitive to get the pulling range, and there is no change in oscillation mode at series resonance.

Fundamental mode and all the overtones are distinct modes and only a few specialised oscillators can get more than one overtone running at any one time.

The spurious responses that all crystals have should always be above the correct crystal frequency, but running slightly below series won't help to avoid them because the spurious responses also have their series resonant point, and their frequencies also change with load, just less than the main response. For AT cut crystals, the design of the crystal rather than the oscillator is what stops the oscillator running at a spurious response. The spurious responses are not at predictable frequencies, but they should always have less activity that the main response.

SC cut crystals have a B mode just above the main mode, and the B mode oscillates more easily, but with terrible temperature coefficient. They need carefully designed filters to stop the B mode oscillating.

 

[MrAl]

Hello again,


Thanks for the great responses there. I guess what i am after here is very specific. I would like to find out if the variation in oscillation frequency over a mild temperature variation is more stable with load capacitance or without load capacitance. As i mentioned previously, the oscillation has a sensitivity factor with respect to the load capacitance, in that the oscillation frequency changes by some factor due to the change in load capacitance. When the temperature changes over a small range like 20 deg C, the crystal doesnt change that much but since the caps change that introduces some shift in frequency too. Without caps this extra change would not occur. I also assume (for now) that the oscillator will continue to run for a long time without a problem (except for aging of course) and that phase noise will not be significant either, and the oscillator will always start up without a problem (this last one is actually probably a good assumption because they start easier with less capacitance).

To clear up some other things, this will be an AT cut and it will be used in fundamental mode and say a 1MHz to 5MHz crystal.

It is interesting, the point brought up about the CMOS itself changing threshold voltages a bit, but i think they vary in the same direction meaning some automatic canceling of the effect is inherent in the gate itself. I could study this easily by doing a couple simulations or a few simple calculations though just to see how much effect that would have, but for now lets assume that it is perfectly stable. I dont see how caps can help this situation anyway as the threshold voltages are dc quantities and the caps wont change anything about that.

 

[RCinFLA]

Hello again,


Thanks for the great responses there. I guess what i am after here is very specific. I would like to find out if the variation in oscillation frequency over a mild temperature variation is more stable with load capacitance or without load capacitance. As i mentioned previously, the oscillation has a sensitivity factor with respect to the load capacitance, in that the oscillation frequency changes by some factor due to the change in load capacitance. When the temperature changes over a small range like 20 deg C, the crystal doesnt change that much but since the caps change that introduces some shift in frequency too. Without caps this extra change would not occur. I also assume (for now) that the oscillator will continue to run for a long time without a problem (except for aging of course) and that phase noise will not be significant either, and the oscillator will always start up without a problem (this last one is actually probably a good assumption because they start easier with less capacitance).

To clear up some other things, this will be an AT cut and it will be used in fundamental mode and say a 1MHz to 5MHz crystal.

It is interesting, the point brought up about the CMOS itself changing threshold voltages a bit, but i think they vary in the same direction meaning some automatic canceling of the effect is inherent in the gate itself. I could study this easily by doing a couple simulations or a few simple calculations though just to see how much effect that would have, but for now lets assume that it is perfectly stable. I dont see how caps can help this situation anyway as the threshold voltages are dc quantities and the caps wont change anything about that.

Your point about greater loading reactance is true, as the more external reactance involved the more the frequency will be dependent on it. This is one reason you don't want to pull a crystal several hundred ppm.

Other things that need to be considered is reactance loading variation due to gain variation of oscillator amp causing variation in overdrive level which changes impedance loading by amp. For mass production you don't want a given lot of PCB's with different etch back tolerance and board relative dielectric constant to dominate variation in loading on the crystal.

High tolerance NPO and relatively stable negative temp coefficient capacitors are readily available to swap out above mentioned variations.

CMOS oscillators have poorer 1/f noise. If you are looking for good SBN, particularly close in noise, use bipolar oscillator.

 

[MrAl]

Hi again,


The 5 percent capacitors seem to be the easiest to obtain. Couldnt find any 1 percent'ers where you dont have to by 1000 of them or they arent 3 bucks a piece.

 

[MrAl]

Hello again,

I did a few calculations for the crystal oscillator open loop and it looks like when the load capacitors change with temperature by about 9 percent that only changes the oscillator frequency by 11PPM to 13PPM depending on the value of the two load caps. A 2pf set causes 11PPM and a 22pf set causes 13PPM. That's not that significant considering that's for almost a change of 10 percent of the capacitance of each capacitor (the actual change was from 22pf to 20pf). Since it is not going to be that much really, i guess using the load caps is still a good idea.

 

[Dragon Tamer]

I was tinkering with some crystal oscillators last weekend, and I noticed that when I didn't include the caps, the frequency for a 4MHz crystal (4.032MHz actually) was all over the place. It would swing in frequency from 17MHz to 1kHz. I was using a 4011 as an inverter for my oscillator, and the frequency changed very little with load until I started to draw more than 7mA from the output. I also had the output of the oscillator go into the input of another 4011 NAND as an inverter to act as a buffer, and the frequency didn't change at all with load.

 

[MrAl]

Hello there DragonT,

Im sorry, but i should have made it more clear that the "load" caps are the two caps that are used to 'load' a parallel operation crystal, not the output load on the oscillator which is a very different issue. Note that there are two load caps, one that connects to one lead of the crystal and one that connects to the other lead of the crystal, and the two other leads of the caps go to ground. That loads the crystal to the manufacturers specifications so that the frequency of oscillation is closer to the stamped value on the crystal package.
If you look at some crystal data sheets you'll see that the manu specifies a specific value for the load capacitance but that's a single value like 20pf. What that means is these two caps that connect directly to each lead of the crystal have to equal 20pf but the calculation for that value is with the two caps in series. That 20pf load cap spec means we would need two caps 40pf each connected directly to the crystal and the other end of each cap goes to ground. That's what is called the "load capacitance" or the "load capacitors". These caps are used when the crystal is made for parallel resonant operation which is what i am looking at here.

I guess i could draw up a circuit...
In the schematic below, CL1 and CL2 are the two load caps. The total specified load capacitance is equal to the series combination of these two caps. Rs is a series resistance used to lower the crystal drive.
Note that any capacitance connected to the very output (Fout) should not alter the oscillation frequency at all, but that's not the capacitance being considered in this thread. That would be the "output load capacitance", while the two caps CL1 and CL2 make up the "crystal load capacitance". What i was looking at in this thread is the crystal load capacitance to see how much the frequency varies when these two caps change value over temperature.