Awesome, I was just getting ready to look through your lectures for one on oscillators. I was using an OpAmp as a buffer, but it started self-oscillating.
6:08 regions for most oscillators. Since they are inherent, let us utilize their asymptotic nature at those corner frequencies. 14:02 virtual damping -- the phase timing jitter explained on a decaying LC sinusoid with impulse response. Very nice talk. 24:51 an oscillator has no way of knowing its previous phase -but what if they could? 27:03 Impulse sensitivity function ISF captures the sensitivity of an oscillator to the input impulse. 1:06:17 is like a good record you have to listen a few times, including 1:07:52 1:08:07 - minimum ISF sensitivity at the peaks 1:11:01 - time-shifted ISFs - (a tensor, superposition) in motion 1:11:30 - creation of superposition of tensor superposition state 1:20:30 - resonance with nearly moving tensor, especially quadrature. 1:30:55 - invite to read two papers
I watched with satisfaction the firsts 184 videos and this is the first that I found not so clear. Perhaps it is too fast and the slides parts are not strictly correlated (highlighted) with speech
Because these oscillators don't operate in a constantly on state and in steady-state the transistor only turns on to draw enough current through the tank when the emitter/source is at the lower voltage.
Time "01:15:00" rms-Jitter vs. time for free-running osc., you said equal rise/fall-time will reduce flicker noise portion (independent part) bcs DC value of ISF becomes zero. This is right only if NMOS/PMOS flicker noise currents are correlated. But they're NOT. So, each device has its own ISF, and the noise of each will be unconverted separately. Am I correct?
It is correct that the noise of the NMOS and PMOS are uncorrected. There is a more detailed analysis that calculated the effective ISF of each of the NMOS and PMOS devices and takes the aggregate into account. I usually show that slide in the longer presentations, but not in this video. It shows that there is still a reduction, although it is not possible to zero it out.
@@AliHajimiriChannel Thanks for the reply. I've read your paper on ring osc. Yes, equal tr/tf decreases the corner frequency "fc" and the PN at far-out offsets is improved. I mean, "fc" is around tens of megahertz in ROs right? -especially in short channel devices. But in far-in say 100KHz, PN improvement may not be appreciable. Am I right? By the way, your results on your paper are for current-starved RO type where you mainly play with the starved transistors to make tr/tf equal. But in a simple RO, your charging and discharging paths may experience different load cap (say for fanout=2, "CL" is twice in charging mode since PMOS is twice the NMOS). So, PN improvement may not be so touchable even in far-out. Do you agree?
Hello, Thank you very much for this video, it is very informative, and I learned a lot. I do have a question though (basic one). In the beginning of the video you state that the jitter variance widen over time and I don't quite get it. I understand that the phase "error" you add at each clock cycle stays there for ever, therefore over time you keep on adding errors with a given distribution. Which means over time you have more and more chance to be really off the "ideal" edge position (vs. time). But I don't get why the the distribution itself widens (ie. variance increases over time, t^2 in your graph)... Can you help me understand that? Regards.
I really like the lectures of Ali Hajimiri. Nevertheless, I got some questions: In the according paper equation (21) and that shown in th-cam.com/video/wByzymJ0Ppc/w-d-xo.html are not equal (different by a factor of 2). Is it because of single- and double-sided spectra? Moreover, I get a little confused by the usage of S_phi. Sometimes it is used as the spectrum of Phi and sometimes as the PSD of Phi.
Is the capital gamma function of (wo *t) the reflection coefficient of the impedance matching? If so, that woul explain that cancelling the reactive component of the negative impedance optimize to the minimum the phase noise magnitude.
Ali Hajimiri is the best with no questions !!!!
Thank you so much!!
Who can explain it better than the author himself :-)
He explains it so well even in the paper. It's the one paper that I could understand with minimum head scratching.
Awesome, I was just getting ready to look through your lectures for one on oscillators. I was using an OpAmp as a buffer, but it started self-oscillating.
6:08 regions for most oscillators. Since they are inherent, let us utilize their asymptotic nature at those corner frequencies. 14:02 virtual damping -- the phase timing jitter explained on a decaying LC sinusoid with impulse response. Very nice talk. 24:51 an oscillator has no way of knowing its previous phase -but what if they could?
27:03 Impulse sensitivity function ISF captures the sensitivity of an oscillator to the input impulse. 1:06:17 is like a good record you have to listen a few times, including 1:07:52
1:08:07 - minimum ISF sensitivity at the peaks
1:11:01 - time-shifted ISFs - (a tensor, superposition) in motion
1:11:30 - creation of superposition of tensor superposition state
1:20:30 - resonance with nearly moving tensor, especially quadrature.
1:30:55 - invite to read two papers
Brilliant explanation Prof Hajimiri! The explanation augments really well to the one in your paper. Thank you so much
I tried to read your paper many times, but not quite understand. Now it's clear.
I watched with satisfaction the firsts 184 videos and this is the first that I found not so clear. Perhaps it is too fast and the slides parts are not strictly correlated (highlighted) with speech
1:04:06 Why increasing the mobility can make the drain current lobe narrower?
Because these oscillators don't operate in a constantly on state and in steady-state the transistor only turns on to draw enough current through the tank when the emitter/source is at the lower voltage.
Is there a link to the two papers by Prof Hajimiri ?
Time "01:15:00" rms-Jitter vs. time for free-running osc., you said equal rise/fall-time will reduce flicker noise portion (independent part) bcs DC value of ISF becomes zero. This is right only if NMOS/PMOS flicker noise currents are correlated. But they're NOT. So, each device has its own ISF, and the noise of each will be unconverted separately. Am I correct?
It is correct that the noise of the NMOS and PMOS are uncorrected. There is a more detailed analysis that calculated the effective ISF of each of the NMOS and PMOS devices and takes the aggregate into account. I usually show that slide in the longer presentations, but not in this video. It shows that there is still a reduction, although it is not possible to zero it out.
@@AliHajimiriChannel Thanks for the reply. I've read your paper on ring osc. Yes, equal tr/tf decreases the corner frequency "fc" and the PN at far-out offsets is improved. I mean, "fc" is around tens of megahertz in ROs right? -especially in short channel devices. But in far-in say 100KHz, PN improvement may not be appreciable. Am I right?
By the way, your results on your paper are for current-starved RO type where you mainly play with the starved transistors to make tr/tf equal. But in a simple RO, your charging and discharging paths may experience different load cap (say for fanout=2, "CL" is twice in charging mode since PMOS is twice the NMOS). So, PN improvement may not be so touchable even in far-out. Do you agree?
Hello,
Thank you very much for this video, it is very informative, and I learned a lot.
I do have a question though (basic one).
In the beginning of the video you state that the jitter variance widen over time and I don't quite get it.
I understand that the phase "error" you add at each clock cycle stays there for ever, therefore over time you keep on adding errors with a given distribution.
Which means over time you have more and more chance to be really off the "ideal" edge position (vs. time).
But I don't get why the the distribution itself widens (ie. variance increases over time, t^2 in your graph)...
Can you help me understand that?
Regards.
Can I get the slides used in this lecture?
I really like the lectures of Ali Hajimiri. Nevertheless, I got some questions:
In the according paper equation (21) and that shown in th-cam.com/video/wByzymJ0Ppc/w-d-xo.html are not equal (different by a factor of 2). Is it because of single- and double-sided spectra? Moreover, I get a little confused by the usage of S_phi. Sometimes it is used as the spectrum of Phi and sometimes as the PSD of Phi.
Thank you so much Prof. Hajimiri. I think at time: 29:36, it is Gamma(w0*tau) instead of Gamma(w0*t). Is it correct or am I missing something?
You are right and it should be Gamma(w0*tau). It was correct in the original paper.
You are right. I should correct that in the slides. Thank you.
Is the capital gamma function of (wo *t) the reflection coefficient of the impedance matching? If so, that woul explain that cancelling the reactive component of the negative impedance optimize to the minimum the phase noise magnitude.
It has NOTHING to do with reflection coefficient. It is the Impulse Sensitivity Function defied earlier in the video.
I wish slides were shown for more time instead of focusing on professor.
How does the formula 2D/(del_w)^2 came?
چ😇را فکر میکنم ایرانی هستی