The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Design >> Mixed-Signal Design >> question about oscillators
https://designers-guide.org/forum/YaBB.pl?num=1123060319

Message started by svensl on Aug 3^rd, 2005, 2:11am

Title: question about oscillators
Post by svensl on Aug 3^rd, 2005, 2:11am

Hello all,

I have a simple question about VCO's. We all know that with a rising level of input magnitude the VCO output frequency will go up. What happens if the input signal is a sine wave centred on zero, ie, amplitude levels of +-0.5V? Is the output frequency still highest at 0.5 and lowest at -0.5, meaning, can the VCO distinguish between positive and negative input levels?
With the mathematical model I could offset the input, but then the output PSD would have a dc signal in it which I don’t want.

Thanks,

Title: Re: question about oscillators
Post by Andrew Beckett on Aug 3^rd, 2005, 4:51am

Surely this will depend entirely on how the VCO is designed? I don't think a generic answer is possible. I would expect in most cases that provided the entire swing of the VCO input is within the input range of the VCO, the output frequency would usually (for most VCOs) have the lowest frequency at the minimum value of the sine wave, and the highest frequency at the maximum.

This is a bit of an odd question... perhaps there's something behind what you're asking? It doesn't make much sense to me (or rather the answer seems too obvious unless there's some detail you've omitted).

Andrew.

Title: Re: question about oscillators
Post by Visjnoe on Aug 5^th, 2005, 1:59pm

Hi,

First, let's assume that VCTRL = 0V is a valid input DC level for the VCO design (VCTRL being the node/port which controls the varactor capacitance).

So, @VCTRL = 0V, the VCO is oscillating at a certain frequency wo. Now, what happens if you apply a sine (frequency f1) with amplitude +/-A centered around 0V?

VCTRL becomes: VCTRL = 0 +Asin(2*pi*f1*t).

What happens can be most easily seen in the frequency domain: besides the fundamental tone @wo, two sideband-tones @wo-f1 and @wo+f1 will appear. Their magnitude relative to the fundamental tone will be determined by the KVCO (frequency gain/Volt) of your VCO.

You can verify this formally by putting the expression for VCTRL in the general VCO equation. I refer to the book by Razavi for more details.

Kind regards,

Peter

Title: Re: question about oscillators
Post by Visjnoe on Aug 5^th, 2005, 2:01pm

Hi,

there's a slight mistake in my previous post:

the tones appear @wo -2*pi*f1 and @wo+2*pi*f1
(rad/s versus Hz... :))

Kind Regards,

Peter

Title: Re: question about oscillators
Post by svensl on Aug 13^th, 2005, 12:48am

Hello,

Yes, I know what you are talking about since I have read Razavi's pages on VCO. However, this won't solve my offset problem. Please consider a VCO in the forward path of a system with negative feedback. If I apply a +/- signal I need to offset the free running freq. to keep a positive frequency. (Andrew, sorry for the misunderstanding).
This will introduce a dc signal in the output which can be seen in the PSD. To avoid it I could place an 1/s integrator before the VCO which will suppress the dc signal. Then, everything will be fine. However, is there is there a different way of suppressing the dc signal? I hope I am making more sense this time.

Cheers.

Title: Re: question about oscillators
Post by Andrew Beckett on Aug 14^th, 2005, 12:07am

Are you suggesting that the VCO frequency will be DC with a 0 input voltage? I'd say such a VCO would be rather hard to design! (especially if you want high frequencies out of it
as well) - except of course if the VCO simply stops working with 0 volt input...

VCOs normally have some sort of centre frequency and gain, and a finite tuning range.

Perhaps I've misunderstood your question (still)?

Regards,

Andrew.

Title: Re: question about oscillators
Post by svensl on Aug 14^th, 2005, 1:02am

Well, yes and no. Thinking of it it will indeed be very hard to have the freq. to be dc at 0 volt input. My models so far were doing excatly that. But I can easlily change that. At the moment I am still using ideal models as the one from the AMS section. Thanks for you feedback though.

Speaking of modelling. I was wondering about the power consumption of VCOs, i.e ring oscillators. Considering the acquisition range only we would expect the input the the VCO to change and thus the output frequency. Is it correct to assume that the power during acquisition bill be dynamic and can be expressed as P=Vdd²C_Lf_{out_average}?

Here, C_L would be the load of each cell, or rather the parasitic capacitance associated with a simple inverter cell and f_{out_average} be the average output frequency of the VCO. I would also think that the above equation would have to multiplied by the number of stages in the ring oscillator.

Is it correct to use above equation for the power conumption of a VCO during acquisition?

Title: Re: question about oscillators
Post by Andrew Beckett on Aug 15^th, 2005, 12:32pm

Most "ideal" VCO models would still be based around a
centre frequency. Even ideal models need to be vaguely model reality!

Andrew.

Title: Re: question about oscillators
Post by ywguo on Aug 17^th, 2005, 6:22pm

Svensl,

Quote:

I was wondering about the power consumption of VCOs, i.e ring oscillators. Considering the acquisition range only we would expect the input the the VCO to change and thus the output frequency. Is it correct to assume that the power during acquisition bill be dynamic and can be expressed as P=Vdd2CLfout_average?

First, why do you need model the power consumption of VCOs?

Second, the power consumption of VCOs depends on its design but not acquistion or steady state. Furthermore,
P=Vdd2CLfout_average
is not a general equation for VCO.

Best regards,
Yawei

Title: Re: question about oscillators
Post by svensl on Aug 18^th, 2005, 1:38am

Yawei,

Thanks for your feedback.
I know that the implementation of the VCO dictates the power consumption. But considering only a ring oscillator made up of N single-ended inverter cells, then would it not be right to assume that P=Vd²C_LF_{vco_ave} is approx. valid?

You said that " The power consumption of VCOs depends on its design but not acquistion or steady state."
But the time required for the PLL to lock to a new frequency can be expressed as t_aqu = t_capture + t_lock .
For the estimation of the power consumption during the capture process, I assumed that the VCO frequency follows the variation on the control voltage. Thus, P=Vd²C_LF_{vco_ave} where F_{vco_ave} is the difference between the start frequency, and the end frequency.
The locked condition can be considered a stable state. There are no drastic changes and most of the variables like K_vco and f_vco are constant. To estimate the power during the locked range I consider that for a non-differential implementation, at any given time, just one cell is switching. Thus the current would be something like I=V_ddC_L/t_d where t_d=1/2/N/f_vco.

Out of curiosity, how would you model the power consumption of a ring-osc. during aquisition range?

Cheers,

Title: Re: question about oscillators
Post by ywguo on Aug 18^th, 2005, 2:27am

Svensl,

I have never modeled the power consumption of VCOs in acquistition mode. Normally, I only measure the power consumption in steady state.

For a ring oscillator made up of N single-ended inverter cells, I think that P=2N*Vd²C_LF_vco is approx. valid, where C_L is the load capacitance of each stage.

Best regards
Yawei

Title: Re: question about oscillators
Post by svensl on Aug 18^th, 2005, 2:35am

Thanks. This is exciting.

I was also wondering how one would express the dynamic range of a ring-oscillator? For example, in SC circuits where the main noise is kT/C we have something like DR=V_in²C_sample/(8kT).
Is there a similar equivalent for VCO's? If so, are there any references I could look up.

Thanks

Title: Re: question about oscillators
Post by ywguo on Aug 18^th, 2005, 4:37am

Svensl,

According to my personal opinion, in the past 10 years, the most important contributors to the theory about the noise in VCOs are Alper Demir, Ali Hajimiri, and John G. McNeill.

Please refer to their papers and thesises.

Best regards,
Yawei

Title: Re: question about oscillators
Post by Jitter Man on Aug 18^th, 2005, 10:32am

Demir's work is difficult to penetrate, Hajimiri's work is seriously flawed, and MacNeill's work is largely on jitter. You might be better served taking a look at the material on this site. For example, the cyclostationary noise paper found at http://www.designers-guide.org/Theory/. You can also try http://www.designers-guide.org/Analysis/PLLnoise+jitter.pdf or http://www.designers-guide.org/Analysis/rf-sim.pdf.

For more design insights, try http://www.designers-guide.org/Books/dg-osc/index.html.

Title: Re: question about oscillators
Post by Paul on Aug 19^th, 2005, 4:41am

Jitter Man,

I agree that Demir's work is very mathematical and that McNeill uses more the time-domain approach, but I am somewhat surprised by the strength of your statement on Hajimiri's work. Can you give some arguments why you consider it as "seriously flawed", even though many people use it as a starting point for their designs? I believe this could be very useful for many of us.

What do you think of T.Weigandt's work?
http://kabuki.eecs.berkeley.edu/~weigandt/phd.pdf

In advance many thanks for your comments.

Paul

Title: Re: question about oscillators
Post by svensl on Aug 19^th, 2005, 9:00am

Thanks for the various references.
I agree that Demir's work is difficult to understand. I have had a look at the papers posted here and at a quick look at the book "Designer's Guide to High Purity Oscillators", but the process of learning is slow.

Please let me state a simple question. In switched-capacitor integrators the dynamic range is mainly determined by the kT/C noise. For example we have, DR=V_sw²*OSR*C_sampling/8/k/T.
I understand that in oscillators the dominant noise source is phase and jitter noise. Then, could we write the same expression for DR for ring-oscillators? For example the noise power for a 3-stage ring oscillator is given as |V_{n_tot}|²=8kT*2/(8gm)*(w_o/dw)². This is from Razavi's "A Study of Phase Noise in CMOS Oscillators". So could we say the DR is then again the ratio if the input singal power to |V_{n_tot}|²? I am just trying to make more sense out of this by relating it to something I know, namely SC circuits.

Title: Re: question about oscillators
Post by ywguo on Aug 22^nd, 2005, 2:03am

Paul,

T.Weigandt's work focuses in time domain, too. I studied his thesis when I prepare my master thesis.

Weigandt's work only studied relative simple factors. He ignored something like correlation.

McNeill's work is more thoroughful. It also studies the noise of VCOs in open loop or that in closed loop. I think McNeill's work is more important than Weigandt's.

Best regards,
Yawei

Title: Re: question about oscillators
Post by ywguo on Aug 22^nd, 2005, 2:18am

Svensl,

The phase noise or jitter is noise in phase domain or timing domain. The VCO's DR is not the ratio of the phase noise to input power. Because the input power is the amplitude of the signal.

Best regards,
Yawei

Title: Re: question about oscillators
Post by Frank Wiedmann on Aug 22^nd, 2005, 5:26am

Paul,

Demir explains in his paper http://home.ku.edu.tr/~aldemir/pubs/tcas001.pdf why he considers Hajimiri's method to be flawed, please refer to the next-to-last paragraph of section III. However, it might still be the case that Hajimiri's conjecture for orthogonally decomposing the perturbation into components that generate phase and amplitude deviations is a reasonably good approximation for many practical oscillator circuits. After all, the results that Hajimiri presents in his papers (e.g. http://www.chic.caltech.edu/Publications/general_full.PDF) seem to match reality pretty well. In order to really judge on this, it would probably be useful if one could get hold of Demir's reference [20] or maybe even try to contact Demir and/or Hajimiri themselves. Perhaps Jitter Man also has some additional insights on this subject.

Title: Re: question about oscillators
Post by Paul on Aug 22^nd, 2005, 6:50am

Frank,

I was not aware that this consideration was established in one of Demir's papers. Thanks for the information. Indeed it would be nice to see the development in the referenced paper, but it doesn't seem to be in the public domain.

Paul

Title: Re: question about oscillators
Post by Jitter Man on Aug 22^nd, 2005, 1:47pm

Anybody that has looked closely at Hajimiri's work from the CAD side of the world points out the flaws in the decomposition he uses. However, in most cases his decomposition, while not technically correct, gives reasonable results. However, in his paper on ring oscillators his claim that symmetry in the signals prevents flicker noise up conversion is completely false. And it is wrong for a completely different reason than problems in the decomposition.

For more information, see http://www.designers-guide.org/Forum/?board=rfsim;action=display;num=1094680012;start=1#1.

[glb]Jitter Man[/glb]

Title: Re: question about oscillators
Post by Ken Kundert on Aug 23^rd, 2005, 8:56am

You can find On the Validity of Orthogonally Decomposed Perturbations in Phase Noise Analysis by Demir and Roychowdhury here ...

http://www.designers-guide.org/doc/1998-Phase-Noise-Orthogonal-TM.pdf

-Ken

Title: Re: question about oscillators
Post by Paul on Aug 23^rd, 2005, 11:11am

Jitter Man, Ken,

thanks for the enlightening information.

Paul

Title: Re: question about oscillators
Post by Frank Wiedmann on Jan 30^th, 2007, 11:21am

I have just found out that there is a paper by Geoffrey Coram in which he presents a simple example showing that the orthogonal decomposition yields the wrong result. For me, it was much easier to understand than Demir's paper. It does not seem to be available for free download, but the exact reference can be found at http://ieeexplore.ieee.org/search/freesrchabstract.jsp?arnumber=933331&isnumber=20194&punumber=81&k2dockey=933331@ieeejrns.

Title: Re: question about oscillators
Post by loose-electron on Jan 30^th, 2007, 1:37pm

I have to sit here and chuckle. Considering I know both Ali and John.
As with a lot of academic papers, you need to look at the constraints of analysis.

Title: Re: question about oscillators
Post by Geoffrey_Coram on Jun 28^th, 2007, 7:14am

Frank Wiedmann wrote on Jan 30^th, 2007, 11:21am:

but you can PM me and I'll e-mail you a copy. :)

Title: Re: question about oscillators
Post by Paul Geraedts on Oct 25^th, 2007, 12:27pm

Hi everybody,

I think this paper will be an interesting addition to this thread. In this paper both Demir's and Hajimiri's theory is applied and the results are compared.

http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1167539

Paul

Title: Re: question about oscillators
Post by Frank Wiedmann on Oct 25^th, 2007, 11:41pm

Thanks for this reference. As Google reveals, the paper is available for free download at http://www.sigda.org/Archives/ProceedingArchives/Iccad/Iccad2002/04a_1.pdf.