Crystal capacitors to GND or VCC

Log in or register to post comments
Go To Last Post
31 posts / 0 new
Author
Message
Brutte
Brutte's picture
Level: Raving Lunatic
Joined: Thu. Oct 5, 2006
Posts: 3231 View posts
Location: Poland
#1
Posted by Brutte: Sat. Jul 31, 2010 - 04:06 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

What is the difference in connecting two crystal capacitors in three situations below (AT90USB162 and other AVRs):

    A: both to GND (this is the solution found in AN and datasheets) B: both to VCC
    C: one to GND and one to VCC?

Since GND and VCC lines conduct same currents and are symmetrical in terms of voltage fluctuations, it seems to me A and B are equivalent and it does not really matter where the crystal quartz capacitors are connected to.

My question is: Why do all ATMEL's (related) application notes and datasheets state it must be GND?

Second question (if nobody could answer the first one other way than "Because.") is: what about method C? Is there any difference?

I will traditionally answer the questions of "Why would you need that?". Simple answer: my PCB layout simplifies greatly if these capacitors could be connected the way I like (and not necessarily to GND)

No RSTDISBL, no fun!

Tags:

Learning and Information, General Electronics, AT90USB162
Top
david.prentice
david.prentice's picture
Level: 10k+ Postman
Joined: Sat. Feb 12, 2005
Posts: 29255 View posts
Location: Wormshill, England
Posted by david.prentice: Sat. Jul 31, 2010 - 04:29 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Both Vcc and GND are signal GND as far as the oscillator is concerned.

But you already knew that, or else you woul not have raised the subject.

I am not sure whether the startup time will be affected at power-up. You can either analyse it yourself or just measure it.

So if it suits you, go for it.

Regarding layout and schematic conventions, you normally have a large GND plane. GND is normally -ve. This suits most current semiconductors. So humans are going to expect and assume regular electronic practice.

In the days of Germanium PNP transistors, electronic circuits would often use a +ve GND. There was a period of limbo when you could never assume a GND polarity. Fortunately the world has migrated to a -ve GND convention.

Your service engineers will love you for your approach.
You can equally drive on the left side of the road. (I do). I am not so sure that your fellow Poles would appreciate it.

David.

Top
Brutte
Brutte's picture
Level: Raving Lunatic
Joined: Thu. Oct 5, 2006
Posts: 3231 View posts
Location: Poland
Posted by Brutte: Sat. Jul 31, 2010 - 05:11 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Quote:
So if it suits you, go for it.

So, in your opinion, all three solutions are equivalent and do not violate Atmel datasheets? And it is also obvious that when all Atmel's ANs and datasheets state (regarding C) "connect them to GND" it means "connect them to any of power planes"? And same relations apply to all capacitors like on Aref pin and all inductors like Avcc line inductor?

I can understand it is a custom - a common practice to connect crystal capacitors to GND, but this is not what I found in a datasheet and what is elementary to you is not so obvious to mee. Atmel did not state "A common practice is to connect them to GND" but "connect them to GND". I can see the difference, and this is what I am asking about because of mentioned reasons.

No RSTDISBL, no fun!

Top
ka7ehk
ka7ehk's picture
Level: 10k+ Postman
Joined: Fri. Nov 22, 2002
Posts: 20376 View posts
Location: Tangent, OR, USA
Posted by ka7ehk: Sat. Jul 31, 2010 - 05:20 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

One of the consequences of connecting the oscillator caps to Vcc is noise. Even though it is well bypassed (and often not so well), there WILL be noise at the clock frequency. This can lead to timing jitter. It can also lead to unreliable operation in the presence of external electromagnetic events, such as motor start/stop, relay activation, and PWM control. This is especially true of the low-power oscillator.

The real desire, as far as I can tell, is that the ground pin next to the crystal pins is also the ground of the inverter for the oscillator. Thus, you will often see references to connecting the two caps to the ground PIN, rather than the ground plane, and making the connection to the ground plane AT the pin, rather than the caps.

Jim

Jim Wagner Oregon Research Electronics, Consulting Div. Tangent, OR, USA http://www.orelectronics.net

Top
david.prentice
david.prentice's picture
Level: 10k+ Postman
Joined: Sat. Feb 12, 2005
Posts: 29255 View posts
Location: Wormshill, England
Posted by david.prentice: Sat. Jul 31, 2010 - 05:47 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Quote:
So, in your opinion, all three solutions are equivalent and do not violate Atmel datasheets?

No. My opinion is that all three solutions are equivalent.

The non-GND solutions do violate Atmel datasheets. Data sheets advise how to use the product. They will also make certain guarantees if you follow their advice. The will normally disclaim any other practice.

Vcc is only a signal ground if it is well decoupled. Jim has illustrated some real-life problems.

You can drive your car on the left of the road. The car will work. If you do not crash, you will get from A to B. If you get arrested, you can always argue that David drives on the left. It works ok for him.

David.

Top
Brutte
Brutte's picture
Level: Raving Lunatic
Joined: Thu. Oct 5, 2006
Posts: 3231 View posts
Location: Poland
Posted by Brutte: Sat. Jul 31, 2010 - 06:57 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Quote:
ground pin next to the crystal pins is also the ground of the inverter for the oscillator

AVRs do not have special GND pin (all are equivalent) and inverter's GND is not any different from inverter's VCC, isn't it?
Quote:
One of the consequences of connecting the oscillator caps to Vcc is noise
noise on VCC with respect to GND is exactly the same as noise on GND with respect to VCC (with minus), isn't it?

I can understand that if you design power plane with current sinks/sources and these are not tied to star topology (but bus or loop), we can have voltage drops and additional inductance involved, but I hope we are not discussing elementary errors in here.

Is it possible to have noisy VCC with respect to GND and un-noisy GND with respect to VCC? I am not an expert, so please explain the difference and how to explain the superiority of GND over VCC with good or bad decoupling.

It seems I have a problem with formulating my question precisely:
If I connect caps with "B" method instead of A, this violates Atmel's recommendations, but is electrically equivalent. On the other hand caps to GND "A" method is a custom only, but you cannot infer this is equivalent "B" reading datasheet - you need to know that. For any engineer it is obvious VCC is only relative to GND and GND is relative to VCC. All voltages/currents as functions of time are exactly anti/symmetrical in any circuit (this a consequence of Thevenin's and Norton's theorem).

No RSTDISBL, no fun!

Top
KKP
KKP's picture
Level: Resident
Joined: Tue. Jun 26, 2001
Posts: 848 View posts
Location: Silkeborg, Denmark
Posted by KKP: Sat. Jul 31, 2010 - 07:55 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

For most models and configurations, 2 to GND is the electrically correct way to do it.

For an ATMega16 with CKOPT=1 (full swing mode), the amplifier's input is referenced to vcc/2*. Thus, if you use two capacitors and have noise on VCC, this noise is seen with a gain of 0.5 by the oscillator amplifier (since the input reference moves by vcc/2, and the capacitor voltage does not). In this particular configuration, a 4 capacitor setup has better VCC rejection.

For an AT90CAN128, ATMega168, ATMega169P (which does not have full swing mode), the amplifier's input is referenced to GND+0.7V**. Here a 2-capacitor-to-GND setup has the better rejection.

The electrical characteristics of the oscillator amplifiers in the AVR( 8 ) series is pretty much (completely) undocumented.

/Kasper
*really (VCC+GND)/2
**changing VCC from 2.7V to 5.0V does not change the bias point

Top
tpappano
tpappano's picture
Level: Raving Lunatic
Joined: Mon. Dec 1, 2003
Posts: 2943 View posts
Posted by tpappano: Sat. Jul 31, 2010 - 08:07 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Quote:
All voltages/currents as functions of time are exactly anti/symmetrical in any circuit (this a consequence of Thevenin's and Norton's theorem).

I'm not sure I see how Norton or Thevenin fit here. Perhaps you mean Kirchoff?

Tom Pappano
Tulsa, Oklahoma

Top
tpappano
tpappano's picture
Level: Raving Lunatic
Joined: Mon. Dec 1, 2003
Posts: 2943 View posts
Posted by tpappano: Sat. Jul 31, 2010 - 08:14 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Picking at some nits, however, you do have a capacitive network with the crystal caps and the decoupling cap where Thevenin and Norton can come into play. If the crystal caps common connection is tied to Vcc, you do have the decoupling cap reactance in series with the common node and ground. This is not really "equivalent" to the common node actually being tied to ground.

Tom Pappano
Tulsa, Oklahoma

Top
david.prentice
david.prentice's picture
Level: 10k+ Postman
Joined: Sat. Feb 12, 2005
Posts: 29255 View posts
Location: Wormshill, England
Posted by david.prentice: Sat. Jul 31, 2010 - 08:20 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

All three Theorems/Laws apply.

It would not be the first time that someone has laid out a pcb with a mistake. You then have to consider what is a practical solution to get you out of the sh*t.

It is obviously worth trying the 'unconventional' topology. If it works ok, you can wipe the egg of your face. If it does not work ok, neither your boss or Atmel will support you.

Next time, you take more care with the pcb layout. IMHO, there is little point arguing that the rest of the world overturning conventions just to follow your simple mistake.

Should all of Poland start driving on the left purely because you have bought a right-hand drive car?

David.

Top
Brutte
Brutte's picture
Level: Raving Lunatic
Joined: Thu. Oct 5, 2006
Posts: 3231 View posts
Location: Poland
Posted by Brutte: Sat. Jul 31, 2010 - 11:57 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Quote:
just to follow your simple mistake

This is a misunderstanding. I did not design the PCB with reversed VCC/GND capacitors. My next design could be simplified greatly if I had the option to decide where to connect XTAL caps. Connection with VCC fits the most.
I suppose my English is not precise enough and that is where misunderstandings come from. Sorry for that.
I also do not understand david.prentice comparing connecting these caps with GND/VCC to Driving Law. I am not convincing anybody to connect them to VCC or GND, I am just curious why it is made that way. Misunderstanding.
Quote:
overturning conventions

I appreciate your opinion but it seems there is a good KKP's argument opposing it:

Quote:
the amplifier's input is referenced to GND+0.7V**. Here a 2-capacitor-to-GND setup has the better rejection.

This undermines "convention theorem", because 0.7V offset violates a linearity of the circuit (none of mentioned theorems apply then). I did not simulate it, but it seems noise added to VCC will not influence VCC/2 linear circuit (if an oscillator can be treated as such). Instead a drive with 0.7V offset can introduce jitter perhaps.

Quote:
a 4 capacitor setup has better VCC rejection

You mean VCC/C1/XTAL1/C2/GND and VCC/C3/XTAL2/C4/GND? If linear, it looks like being equal to a pair of capacitors connected to VCC or GND or VCC/GND. But I suppose generator cannot be treated as such and this is why 4 caps are better than two, isn't it?
Could you KKP (Kasper) give me some reference (literature) about the subject? Unfortunately AT90USB162 has low power oscillator. I will simulate it anyway, but would like to have deeper theoretical understanding first - nonlinear circuits are cunning.

Quote:
This is not really "equivalent" to the common node actually being tied to ground

Sure it is, if capacitors are linear. And they are at this level of discussion.

No RSTDISBL, no fun!

Top
tpappano
tpappano's picture
Level: Raving Lunatic
Joined: Mon. Dec 1, 2003
Posts: 2943 View posts
Posted by tpappano: Sun. Aug 1, 2010 - 01:06 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Quote:
Sure it is, if capacitors are linear. And they are at this level of discussion.

Nope, "B" is not equivalent to "A" when you analyze the impedance "seen" by each crystal lead to *ground* and to each other. There is now an additional feedback path from osc input to output that is not present in "A". Since you are not privy to the actual internal design of the oscillator, the importance of the relationship between the crystal terminals and chip ground is unknown.

Tom Pappano
Tulsa, Oklahoma

Top
LDEVRIES
LDEVRIES's picture
Level: Raving Lunatic
Joined: Fri. May 4, 2007
Posts: 3749 View posts
Location: Geelong Australia, Home of the "Cats"
Posted by LDEVRIES: Sun. Aug 1, 2010 - 01:18 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Quote:
Vcc & Earth are at the same AC potential and therefore it makes no difference etc. etc.

Yes, this is almost true! It assumes that the impedances in the earth & Vcc distribution are the same at all frequencies, which in practice will never be the case and as Vcc side will have a greater impedance there will be more noise on the Vcc supply than the Ground supply.
Connecting both oscillator caps to Vcc will increase the noise on the clock signal. Connecting only one will most likely make it worse. Every time an output pin goes through a transition and switches considerable current considerable oscillator noise is generated.

Taking the david.prentices's car analogy a little further, doing this would be like attempting to deriving an engines crank angle from one of the tyres of the automobile rather than from the crankshaft direct.
Yes, I could compute a crank angle from the a tyre position, but it will have lot's of noise on it!

You won't be violate Atmel's rules. You will be violating good oscillator design practices which Atmel endorses for use in their products. Every other manufacturer of products that utilize crystals would use the same method!

Not following these recommendations, means that they you are on your own and if they product you design ever cause death or injury due to some internal latch-up, they cant be sued but you certainly will!

I cannot imagine a PCB design, where you cant get the caps to ground!

Charles Darwin, Lord Kelvin & Murphy are always lurking about!
Lee -.-
Riddle me this...How did the serpent move around before the fall?

Top
Brutte
Brutte's picture
Level: Raving Lunatic
Joined: Thu. Oct 5, 2006
Posts: 3231 View posts
Location: Poland
Posted by Brutte: Sun. Aug 1, 2010 - 01:33 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Quote:
to each other
is simply two caps connected in series (C-C) in both "A" and "B".

XTALs

Quote:
to *ground*
have:
    C1 for XTAL1 and C2 for XTAL2 with "A", C1||C for XTAL1 and C2||C for XTAL2 with "B",
    C1 for XTAL1 and C2||C for XTAL2 with "C",
    ...
where C is a capacitance of VCC/GND. The only difference is in the values of capacitors - you can set them arbitrarily to get any capacitance from any XTAL to any power line(with one equality constraint). It seems this argument does not explain superiority of GND over VCC. Or there is something I did not understand?

No RSTDISBL, no fun!

Top
LDEVRIES
LDEVRIES's picture
Level: Raving Lunatic
Joined: Fri. May 4, 2007
Posts: 3749 View posts
Location: Geelong Australia, Home of the "Cats"
Posted by LDEVRIES: Sun. Aug 1, 2010 - 01:43 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Gravitation is not fully explained either, but I wouldn't argue against. :)

Charles Darwin, Lord Kelvin & Murphy are always lurking about!
Lee -.-
Riddle me this...How did the serpent move around before the fall?

Top
Brutte
Brutte's picture
Level: Raving Lunatic
Joined: Thu. Oct 5, 2006
Posts: 3231 View posts
Location: Poland
Posted by Brutte: Sun. Aug 1, 2010 - 01:53 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Quote:
Connecting both oscillator caps to Vcc will increase the noise on the clock signal.

This is an argument of "Because." kind. I can imagine an offset argument given by KKP, although it does not explain superiority, but gives a hint - nonlinear circuits could be unsymmetrical.

Quote:
You will be violating good oscillator design practices

I do not care about practices nobody can explain and give reasonable argument why this is the way things should be made. I am not an expert in analog electronics circuitry, but as far as we are considering linear oscillators there are no reasons of telling GND, VCC or any other voltage is superior over any other. We can discuss about customs, tradition and quarrel about superiority of left or right hand side driving but it will not bring us to a conclusion.

Guys, please stop the Driving Law analogy or crankshafts because this does not bring us any closer to the answer:

Quote:
My question is: Why do all ATMEL's (related) application notes and datasheets state it must be GND?

No RSTDISBL, no fun!

Top
tpappano
tpappano's picture
Level: Raving Lunatic
Joined: Mon. Dec 1, 2003
Posts: 2943 View posts
Posted by tpappano: Sun. Aug 1, 2010 - 02:31 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Quote:
It seems this argument does not explain superiority of GND over VCC. Or there is something I did not understand?

You have quoted a couple of network analysis rules, but are not using an important one, Thevenin. The "b" circuit breaks down to a wye network of capacitors. The top two the small caps, the bottom one the bypass cap. The bypass cap is a voltage source of whatever power rail noise is present. It simply injects that noise into xtal1 and xtal2 via the small caps. Additionally, the three caps form a "T-pad" allowing non-xtal feedback from xtal1 to xtal2. You have had several explanations, but here is another analogy. If you were building a nice low noise audio amplifier, would you tie the bottom end of the volume control to ground, or to the raw DC output of the power supply?

Tom Pappano
Tulsa, Oklahoma

Top
ka7ehk
ka7ehk's picture
Level: 10k+ Postman
Joined: Fri. Nov 22, 2002
Posts: 20376 View posts
Location: Tangent, OR, USA
Posted by ka7ehk: Sun. Aug 1, 2010 - 03:05 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

I simply report what I see. Use a good, high-bandwidth scope. Near zero length ground lead on the probe, Connect ground lead to circuit ground.

Look at "ground" with probe. Almost no noise, even at points well separated from the connection point of the ground lead.

Look at Vcc. Noise, 5mv. Maybe 10mv. May be more. Spike every clock edge.

I do not want to connect the crystal to Vcc. Since ground is close to the caps, why even bother to try to connect them any where else?

Jim

Jim Wagner Oregon Research Electronics, Consulting Div. Tangent, OR, USA http://www.orelectronics.net

Top
nleahcim
nleahcim's picture
Level: Raving Lunatic
Joined: Wed. Feb 19, 2003
Posts: 2194 View posts
Location: Redwood City, CA
Posted by nleahcim: Sun. Aug 1, 2010 - 04:30 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Brutte: I think yours is a good question and everybody is being too quick to dismiss it. I can't stand when something is dismissed just because it is not familiar.

I would suggest that, depending on the exact design of the oscillator, that there could be times when it is in fact better to have the caps connected to VCC instead of GND.

Top
leon_heller
leon_heller's picture
Level: Raving Lunatic
Joined: Fri. Jul 27, 2001
Posts: 7680 View posts
Location: St. Leonards-on-Sea (UK)
Posted by leon_heller: Sun. Aug 1, 2010 - 04:42 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Rubbish! The circuit is usually a Pierce oscillator configuration, how is performance going to be improved by connecting the capacitors to Vcc?

Leon Heller G1HSM

Top
LDEVRIES
LDEVRIES's picture
Level: Raving Lunatic
Joined: Fri. May 4, 2007
Posts: 3749 View posts
Location: Geelong Australia, Home of the "Cats"
Posted by LDEVRIES: Sun. Aug 1, 2010 - 07:21 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

NOTE NOTE NOTE
This comment is Motor car analogy free!

Quote:
depending on the exact design of the oscillator, that there could be times when it is in fact better to have the caps connected to VCC instead of GND.

The only reference to that is the case of a making a "dodgy" oscillator oscillate.

In Freescales AN1706/D reference in the Trouble shooting section the following points are made.

Quote:
Power supply noise:
— Power supply noise can sometimes be greatly amplified by the oscillator’s amplifier. If the power supply noise is some harmonic of the crystal frequency, or vice versa, the oscillator may not begin to oscillate. A good test for this condition is to disable the board’s power supply and use a high-quality bench supply to power the board. If the oscillator starts with the bench supply but not with the board supply, look at cleaning up the +5 V rail. In most cases, if power supply noise is a problem, the voltage at XTAL will be a constant around V supply/2.

In that paragraph it also mentions the following

Quote:
In other cases, it has been seen that a noise element on the power supply actually helped the circuit begin oscillating by supplying a much needed high frequency component. Tying the stabilizing capacitors to +5 V sometimes provides a better AC reference point than ground.

I believe this would be a case of helping an inactive crystal start or overcoming some other component issue such as excessively long tracks etc.

Quote:
High-power circuit interference If the ground plane has noise due to high current pulses, try tying the stabilizing capacitors to +5V instead of ground.

Excessive power supply noise or excessive ground plane noise is bad design and requires board layout changes. I am sure that the intent of connecting the caps to Vcc is a temporary fault finding technique and not a recommendation technique for normal low noise oscillators.

A requirement to connect the load caps to Vcc indicates a bad oscillator design! Very few crystals are bad, but there are plenty of bad oscillators!

I am looking forward to see a post from nleahcim which contains a circuit of a high performance oscillator which has an essential requirement to have load caps connected to Vcc :twisted:

Charles Darwin, Lord Kelvin & Murphy are always lurking about!
Lee -.-
Riddle me this...How did the serpent move around before the fall?

Top
KKP
KKP's picture
Level: Resident
Joined: Tue. Jun 26, 2001
Posts: 848 View posts
Location: Silkeborg, Denmark
Posted by KKP: Sun. Aug 1, 2010 - 10:30 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Brutte:
Your quoting style is bad. Taking two statements from different contexts and placing them in such a way that they appear to be in the same context is what you expect to find in, shall we say, the low end of news reporting.

First, stop modelling the amplifiers as you think the amplifier should be, and instead build a model of how they are:
For the full swing pierce oscillator, the amplifier is (with fuzz) two complementary MOSFETs with the gates connected, drains connected, sources to VCC/GND, and operating in the linear region. This amplifier has Vout=(0.5*Vcc-Vin)*2gm*Ro+0.5*Vcc. Since inputs and outputs are referred to Vcc/2, the 4 capacitor load (in which each terminal has 2*C capacitance to Vcc/2) is good (Vcc drops out of the equation). To model of this, you need a 2-port amplifier (with 4 terminals) and a gain 0.5 amplifier to give you vcc/2 for the reference terminals of the first 2-port.

The reduced swing oscillator is a single NMOS transistor with source to GND, and a constant current source in drain. Here Vout=Ro*(Isrc-gm*(Vin-Vt)). This is (enough for our purposes) linear and all the usual tricks apply. You are still going to need a 2-port for your model. Vcc is not in the equation, so it is not an input/output terminal for EMI.

If you connect the capacitors to Vcc in the reduced swing oscillator, the amplifier's transfer function, referred to Vcc, becomes
Vout'=Ro*(Isrc-gm*((Vcc+Vin')-Vt))-Vcc = -Vcc*(1+gain) -Vin*gain + biasterm. That is, Vcc now is a high gain path into, and out of the oscillator.

Any circuit analysis book should do for this. There is no magic involved. This is all nice and linear.

leon_heller:
You never built a pierce oscillator with germanium PNPs? Then if you did, you named the positive rail "GND" so I guess they were still connected to "GND".

Top
leon_heller
leon_heller's picture
Level: Raving Lunatic
Joined: Fri. Jul 27, 2001
Posts: 7680 View posts
Location: St. Leonards-on-Sea (UK)
Posted by leon_heller: Sun. Aug 1, 2010 - 10:38 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

I'm old enough to have used Ge PNPs (I remember when point-contact devices first became available), but I never built an oscillator with them.

Leon Heller G1HSM

Top
KKP
KKP's picture
Level: Resident
Joined: Tue. Jun 26, 2001
Posts: 848 View posts
Location: Silkeborg, Denmark
Posted by KKP: Sun. Aug 1, 2010 - 10:56 AM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

LDEVRIES wrote:

In that paragraph it also mentions the following
Quote:
In other cases, it has been seen that a noise element on the power supply actually helped the circuit begin oscillating by supplying a much needed high frequency component. Tying the stabilizing capacitors to +5 V sometimes provides a better AC reference point than ground.

I believe this would be a case of helping an inactive crystal start or overcoming some other component issue such as excessively long tracks etc.

A way to build such an oscillator is to have a tiny bit of hysteresis in the oscillator amplifier; In this case there is no gain for a change that is less than the hysteresis, and what you see is the circuit behaving either as an RC oscillator (slow oscillation) or delay loop (there is enough shunt capacitance across the crystal, and thus from output to input, and you get fast oscillation).
When you have sufficient noise in the system to overcome the hysteresis, it becomes sort-of a PWM amplifier for small signals, and may start.
It will also fail unpredictably.

/Kasper

Top
Brutte
Brutte's picture
Level: Raving Lunatic
Joined: Thu. Oct 5, 2006
Posts: 3231 View posts
Location: Poland
Posted by Brutte: Sun. Aug 1, 2010 - 03:10 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

First KKP:

Quote:
Brutte: Your quoting style is bad. Taking two statements from different contexts

I am quoting several opinions in one post usually, but I have no idea which specific posts this remark apply (there were 4 of mine).

Quote:
For the full swing pierce oscillator(..) and operating in the linear region.

Then full swing oscillator is a linear circuit and there is no point on discussing if "A" is better than "B" - it does not matter.

Quote:
The reduced swing oscillator(..). This is (enough for our purposes) linear.
It has a constant additive term of 0.7V, so it is not linear, but affine (not y=A*x, but y=A*x+b), because of bias term. Output will not be proportional to VCC, in general.

LDEVRIES

Quote:
NOTE NOTE NOTE
This comment is Motor car analogy free!

I appreciate that. You mention startup problems(with a remedy - modifying caps) and fault finding or high performance oscillators - this still does not explain why GND is any better or more stable than VCC. As I wrote before - you can design bad GND or bad VCC plane, and also this is an interesting and important subject, but first why GND is superior over VCC, please.

Leon Heller:

Quote:
Rubbish! (..) how is performance going to be improved by connecting the capacitors to Vcc?

Who has EVER suggested in this post that "B" method is superior over "A"?? Nobody (LDEVRIES mentioned it as a joke). Neither did I. The problem is everybody wants to convince me "A" is better than "B".
nleahcim
Quote:
I would suggest that, (..) better to have the caps connected to VCC instead of GND

This is not my point! I do not want to convince anybody that VCC is superior over GND or vice versa. I just want to know why application notes and datasheets do it.

ka7ehk

Quote:
I simply report what I see(..)

First you measure VCC with respect to GND and then you compare it with GND with respect to GND.

Then take your scope and measure GND with respect to VCC and then you compare it with VCC with respect to VCC.
Take your post, find/replace VCC<->GND and post it once again.

No RSTDISBL, no fun!

Top
david.prentice
david.prentice's picture
Level: 10k+ Postman
Joined: Sat. Feb 12, 2005
Posts: 29255 View posts
Location: Wormshill, England
Posted by david.prentice: Sun. Aug 1, 2010 - 03:33 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

I think everyone has been quite clear. A, B and C should all work. You have to do a bit more network analysis for the VCC versions. You have some advantageous noise issues and some disadvantageous noise issues.

The obvious answer is for you to construct prototypes and compare their performance.

Meanwhile, I will continue to use the topology recommended by every micro manufacturer. Simply because 'if it ain't broke, why fix it?'

I look forward to reading your paper in the technical press.

David.

Top
nleahcim
nleahcim's picture
Level: Raving Lunatic
Joined: Wed. Feb 19, 2003
Posts: 2194 View posts
Location: Redwood City, CA
Posted by nleahcim: Sun. Aug 1, 2010 - 05:56 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Brutte wrote:

nleahcim
Quote:
I would suggest that, (..) better to have the caps connected to VCC instead of GND

This is not my point! I do not want to convince anybody that VCC is superior over GND or vice versa. I just want to know why application notes and datasheets do it.

You missed my point. I was saying it depends on the design of the oscillator. I do not believe it is always equal.

Top
Jepael
Jepael's picture
Level: Raving Lunatic
Joined: Mon. May 24, 2004
Posts: 6279 View posts
Location: Tampere, Finland
Posted by Jepael: Mon. Aug 2, 2010 - 12:24 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

I have seen ancient microcontroller designs where the crystal caps (or the only one) were connected to VCC.

Also one paranoid guy on the web has decided it is best to split the total capacitance equally to both VCC and GND, so he has eg. four 11pF caps instead of two 22pF caps.

I guess the caps being connected to ground is the easiest mental model and it works sufficiently. You could disconnect the caps from ground so they are only connected to each other, the crystal would still see same capacitance, but the oscillator amplifier would not. Sometimes you see a resistor on amplifier output pin, and the resistance is meant to match capacitor impedance at the resonant frequency, so it creates voltage division by half. For that to work, it needs the capacitor to be connected to somewhere like GND (or VCC) so it cannot float.

Oscillator design is not so simple as it seems, there are a lot of equations so you can solve if your crystal will work with your microcontroller. Well, if the microcontroller amplifier parameters are known anyway, I have not seen them in AVR datasheets.

For example STM32 and PIC (gasp) application notes go very deep in the crystal selection and component value selection process.

I don't know, while I question many things and like to know how things really work, I would still obey datasheets and applications notes and would not question that capacitors are best to connect to ground plane. Also what works in brand X MCUs does not mean it works in brand Y MCUs.

Top
gyoung
gyoung's picture
Level: New Member
Joined: Sun. Oct 20, 2019
Posts: 1 View posts
Posted by gyoung: Sun. Oct 20, 2019 - 03:03 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

I had a problem with an 32,768 hz oscillator on an RTC chip (I think it was Signetics) not starting more than 25 years ago. I eventually discovered my problem was that I had connected both caps to GND out of habit, but the datasheet actually showed one of the caps connected to VCC. When I made this change, the oscillator started reliably. I asked the FAE about this, he didn't know so he asked "the factory" and eventually the answer came back. The problem is that oscillators need some noise to get started. If the circuit is too quiet, the oscillator may not start. As the discussion above shows, everyone is really concerned that the power and layout of oscillators is as quiet as reasonable. The solution, also alluded to by the discussion above, is VCC, which has more noise, especially at startup, where it's really (only) needed for sake of starting the oscillator. I wish I could tell you which oscillators pin had its cap connected to VCC, I think it was the input pin, but I'm not 100% and that documentation is long gone. As for effects on the long term performance, I cannot tell you, because neither the power consumption nor accuracy of the RTC was critical. It was critical that the RTC start and work well enough, which it did - we never had any complaints attributable to RTC failures in the field.

Top
avrcandies
avrcandies's picture
Level: Raving Lunatic
Joined: Wed. Jun 30, 2004
Posts: 3487 View posts
Location: Cincinnati Oh
Posted by avrcandies: Sun. Oct 20, 2019 - 04:00 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

Ummm...that is a 9 year old post!

Wiring the cap to Vcc gives the oscillator circuit a "shove" with a "logic high", since at powerup, the cap voltage is zero (so the cap output lead to the osc rises up with the supply rising).  Over time, a voltage develops on the cap & its output moves to wherever the oscillator circuit drives it. Ideally, hooking the cap to Vcc and gnd is the same (if the source were ideal, zero impedance).   Of course, in normal practice,  sinking the pulses to gnd keeps the Vcc cleaner elsewhere for other chips to drink.  

 

Now back to 2019!

When in the dark remember-the future looks brighter than ever.   I look forward to being able to predict the future!

Top
Krupski
Krupski's picture
Level: Wannabe
Joined: Tue. Nov 17, 2015
Posts: 76 View posts
Location: NY, USA
Posted by Krupski: Tue. Oct 22, 2019 - 02:16 PM
  • 1
  • 2
  • 3
  • 4
  • 5
Total votes: 0

ka7ehk wrote:
One of the consequences of connecting the oscillator caps to Vcc is noise. Even though it is well bypassed (and often not so well), there WILL be noise at the clock frequency. This can lead to timing jitter. It can also lead to unreliable operation in the presence of external electromagnetic events, such as motor start/stop, relay activation, and PWM control. This is especially true of the low-power oscillator.

The real desire, as far as I can tell, is that the ground pin next to the crystal pins is also the ground of the inverter for the oscillator. Thus, you will often see references to connecting the two caps to the ground PIN, rather than the ground plane, and making the connection to the ground plane AT the pin, rather than the caps.

Jim

I agree there. The silly-scope always shows a little "fuzz" of electrical noise (around 100 mV p-p) on the Vdd line. I wouldn't want any of that coupled into the oscillator. On the other hand, I never use loading caps on my crystals. They work just fine without them.

Gentlemen may prefer Blondes, but Real Men prefer Redheads!

Top