The .four command uses a rectangular FFT window, and as a result any DC drift in the circuit will cause a high, slanted noise floor. A way of fixing this is to replace all bias/bootstrap capacitors with 1F or even 1GF capacitors, so that small operating point calculation errors don't lead to significant drift. Setting LTSpice to ignore the first sine cycle of simulation helps too, because the abrupt start of the sine wave causes some drift.
Guys, there really has been a lot of explaining across the forum about why simulation has problems and how to get it to work right. This is a good thread:
http://www.diyaudio.com/forums/soft...simulation-settings-inconsistent-results.html
The link in my signature is somewhat obsolete, since after I made that several LTSpice guides and an LTSpice WIKI have been created, but it does help.
Guys, there really has been a lot of explaining across the forum about why simulation has problems and how to get it to work right. This is a good thread:
http://www.diyaudio.com/forums/soft...simulation-settings-inconsistent-results.html
The link in my signature is somewhat obsolete, since after I made that several LTSpice guides and an LTSpice WIKI have been created, but it does help.
The amp should mostly see the reservoir caps at audio. I usually use a 500nH inductor and a .5R resistor in series with the voltage source. However the rail resistance is very nonlinear because of the delivery method, which is to charge the reservoirs every 8.33mS. I don't see it necessary however to model the rectifier to model the RF behavior of the amp itself, since the effect of them should be pretty well localized to the area of the reservoirs.
Wires generally have 25nH or so per inch. Since lytics have such high capacitance, current can take the path of least inductance through them, so ESL tends to be the length of the leads plus the distance between the leads neglecting the cap height (this applies for radial caps, not axial). Most good lytics have ESR>50mR, and I've seen few modern lytics with more than 70mR.
Now that I think about it, if you locate rectifiers and reservoirs next to the amp, this is no longer true. Since you have highly capacitive switches turning on and off, that changes the whole rail damping situation from one cycle to the next, and so you may get ringing and/or oscillation bursts every 8.33mS. So, how to deal with this situation? Locating the rectifiers close to the amp would make the rail parasitics a moving, unpredictable target, unless they were expertly snubbed from start to finish, decoupling to trafo.
Perhaps splitting the reservoir into two sections, one at the rectifier and the other at the amp board would be a better idea?
Last edited:
There's a lot to consider regarding PSU arrangements I see. I plan in the near future to measure some characteristics of the transformer I am going to use. Then try and add it to LTSpice using some information gootee has posted on this forum.
This thread holds much info.
http://www.diyaudio.com/forums/power-supplies/216409-power-supply-resevoir-size-180.html#post3410867
You have just confirmed how valuable this process could be.
This thread holds much info.
http://www.diyaudio.com/forums/power-supplies/216409-power-supply-resevoir-size-180.html#post3410867
You have just confirmed how valuable this process could be.
I fully agree. Here is an enlightening exercise.
Probe the output current of the LTP and do an FFT. Note the absolute level of the harmonics. IE a 2nd harmonic at -60db (ignoring the fundamental) would be 1uA H2. Now measure the absolute harmonic levels of all the separate components loading the LTP; all of these added together should result in the first number. Which source is dominant? Simply by following the harmonic signatures, you can track down any distortion mechanism in a simulation. That is one thing simulators allow that would be much more difficult in real life. One could sort of automate this process using .four commands.
Really, if you know all the mechanisms and understand how they all interact, the other thing necessary is to know which effects are more dominant and how they contribute to the whole. I find the latter is addressed mainly by investigation, and that is how I use simulation. Theory comes in when you need to know how the mechanisms will shift when you change operating points or components, for better or worse. For example, before I ever investigated current mirror, I knew about all the mechanisms that would interact, but I didn't know which ones mattered and when. That was my main source of error.
All good stuff. I'll add that Eugene Dvoskin’s Total Harmonic Analyzer from the LTspice Yahoo Group does this all for you with some caveats as in my earlier post.
Also Self's book shows the THD & harmonic signatures of many different distortion mechanisms. I do this in my tpc-vs-tmc-vs-pure-cherry thread.
It's particularly useful with simple circuits cos at each stage, usually 1 mechanism is dominant so you can track these down until you have 1ppm THD20k.
A 'good' amp should have THD mostly 3rd when optimised. You can usually do something about 2nd but Beta fall-off at high current, plain non-linearity at high input voltages ultimately lead to distortion which should show itself as 3rd. At 1ppm THD20k, I'm happy to let some 2nd dominate cos eliminating this might be very expensive.
But high order stuff is evil and should be tracked down & eliminated without mercy.
I'm loving all the discussion and the work you put in Bob. I've really noticed a shift on DIYAudio recent years when it comes to negative feedback, or feedback in general. I'm glad to see it's treated more like process technology, where all kinds of feedbacks (local feedbacks for devices/stages and ofcourse the GNFB) are treated separately while maintaining bandwidth. I remember was being scoffed at stating that feedback needs speed and a lot of it.
Well, a simple way to look at it. A feedback amp is like an LCD panel. If you have few pixels, you have an inaccurate image. You need a lot of pixels (resolution) for an accurate image. The more bandwidth, the better and more accurate it can reproduce audio waves. To produce 20KHz accurately irregardless of transient state, you'll need at least 10 times the bandwidth. More is better. This can push into the MHz range and to test stability and accuracy, I tend to use THD100K - THD200K to optimize as best as possible in conjunction with squares.
It's nice to see that speed is chased in the form of trying to optimize indiviual poles, attempting to maintain speed in every stage. It's also why I think CFB amps do so well.
I've recently been developing a new type of VAS and together with the inputstage it still does 1MHz recognizable squares with a good 300V slewrate
It sits on a breadboard and with all the wires and parasitics it still does amazingly well. I'll publish it for discussion soon.
Well, a simple way to look at it. A feedback amp is like an LCD panel. If you have few pixels, you have an inaccurate image. You need a lot of pixels (resolution) for an accurate image. The more bandwidth, the better and more accurate it can reproduce audio waves. To produce 20KHz accurately irregardless of transient state, you'll need at least 10 times the bandwidth. More is better. This can push into the MHz range and to test stability and accuracy, I tend to use THD100K - THD200K to optimize as best as possible in conjunction with squares.
It's nice to see that speed is chased in the form of trying to optimize indiviual poles, attempting to maintain speed in every stage. It's also why I think CFB amps do so well.
I've recently been developing a new type of VAS and together with the inputstage it still does 1MHz recognizable squares with a good 300V slewrate
It sits on a breadboard and with all the wires and parasitics it still does amazingly well. I'll publish it for discussion soon.
Last edited:
I'm loving all the discussion and the work you put in Bob. I've really noticed a shift on DIYAudio recent years when it comes to negative feedback, or feedback in general. I'm glad to see it's treated more like process technology, where all kinds of feedbacks (local feedbacks for devices/stages and ofcourse the GNFB) are treated separately while maintaining bandwidth. I remember was being scoffed at stating that feedback needs speed and a lot of it.
Well, a simple way to look at it. A feedback amp is like an LCD panel. If you have few pixels, you have an inaccurate image. You need a lot of pixels (resolution) for an accurate image. The more bandwidth, the better and more accurate it can reproduce audio waves. To produce 20KHz accurately irregardless of transient state, you'll need at least 10 times the bandwidth. More is better. This can push into the MHz range and to test stability and accuracy, I tend to use THD100K - THD200K to optimize as best as possible in conjunction with squares.
It's nice to see that speed is chased in the form of trying to optimize indiviual poles, attempting to maintain speed in every stage. It's also why I think CFB amps do so well.
I've recently been developing a new type of VAS and together with the inputstage it still does 1MHz recognizable squares with a good 300V slewrate
Hi MagicBox,
You're welcome. There is a lot to negative feedback and in the past it has been maligned by some who did not understand it. That's what got me much more involved with the deatails of it back in the late 1970's (before we had SPICE as we know it today). When properly used, NFB produces excellent results. As you correctly point out, high speed in circuits can greatly increase the effectiveness of NFB.
Cheers,
Bob
I think that term speed and consequently delay are not good term to describe NFB amp.
Those terms are main explanation to be used by anti NFB people to argue aganst use of NFB.
BR Damir
Damir you make a good point. The term speed can have ambiguous meanings. For example, bandwidth vs slew rate.
Wide bandwidth and small amounts of phase lag are better descriptions of what helps make feedback work well. Circuit characteristics which make a high ULGF possible while retaining very good stability help make NFB better.
It is also true that good feedback amplifiers can have very low open-loop bandwidth, and this can be termed a lack of speed. This can wrongly be used by the wide-open-loop-bandwidth advocates to justify their claim that feedback is not good unless it has wide open-loop bandwidth.
A high gain-bandwidth product with little unnecessary phase lag is probably a better way to describe what we want.
Cheers,
Bob
You're right, it was a bit of an ambigious term
What I'm referring to with speed, is the propagation delay / switchingspeed of each active device. This ultimately is what defines the overall maximum GBW product one can achieve.
The chase for a high GBW product probably describes it best - it's this feature that defines the 'resolution' of the amplifier. It's the main ingrediënt for NFB to work as more begin to understand
What I'm referring to with speed, is the propagation delay / switchingspeed of each active device. This ultimately is what defines the overall maximum GBW product one can achieve.
The chase for a high GBW product probably describes it best - it's this feature that defines the 'resolution' of the amplifier. It's the main ingrediënt for NFB to work as more begin to understand
You're right, it was a bit of an ambigious term
What I'm referring to with speed, is the propagation delay / switchingspeed of each active device. This ultimately is what defines the overall maximum GBW product one can achieve.
The chase for a high GBW product probably describes it best - it's this feature that defines the 'resolution' of the amplifier. It's the main ingrediënt for NFB to work as more begin to understand
Although I use it loosely, I like to use the term "excess phase". This is phase lag that is added by extra poles in the circuit that are well above ULGF that create phase lag without much gain loss. There can be many such poles in an amplifier. Putting it another way, if you have an ideal Miller compensated amplifier it will have 90 degree phase lag at ULGF and thus a 90 degree phase margin. If in practice the amplifier has 120 degree phase lag and 60 degree phase margin at ULGF, then I would tend to say it has about 30 degrees of excess phase.
Cheers,
Bob
Ok, and how would the lesser of excess phase
??? I don't think there is any connection between the resulting phase margin and NFB "working well".
Compensated audio amplifiers are always minimum phase systems (unless somebody would fancy an all pass filter in the signal path) so the gain and phase are not independent. Therefore, given a circuit compensation order (1st order Miller, 2nd order TPC or TMC, 3rd order Cherry's NDFL, and that's about it in practice), if one would want to increase the amount of NFB (aka loop gain), some phase margin has to be sacrificed. That's why
is an oxymoron. To take your example, all Miller compensated amps with the same ULGF do have the same (minimum) phase.
If you are talking about the far residual poles effect, then the concept of GBW no longer makes sense (and the minimum phase theory also no longer necessary applies).
Last edited:
Waly, when you talk about "circuit order" in your post above, you are strictly speaking talking about the number of "dominant" poles, or the order of the compensation network(s). The circuit will in fact have many more poles at higher frequency, beyond the ULGF.
What Bob is pointing out is that for a given phase margin and order of compensation, reduction of "excess phase" (e.g. using faster transistors in the output stage) will allow the ULGF to be increased. In turn, this will increase the loop gain in the audio band and reduce distortion (assuming that in the process of reducing the "excess phase", the open-loop distortion has not increased by a larger factor than the loop gain).
What Bob is pointing out is that for a given phase margin and order of compensation, reduction of "excess phase" (e.g. using faster transistors in the output stage) will allow the ULGF to be increased. In turn, this will increase the loop gain in the audio band and reduce distortion (assuming that in the process of reducing the "excess phase", the open-loop distortion has not increased by a larger factor than the loop gain).
Last edited:
Hi Waly,
Less excess phase in an amplifier's open-loop response allows a higher ULGF, which in turn allows more loop gain and distortion reduction at 20kHz.
You are largely right: all compensated amplifiers have a minimum phase open-loop gain function, and amplitude roll-off and phase lag are most certainly related. That is why I described my "loose" use of the term excess phase. However, those excess poles at higher frequencies cause the system to be of a much higher order than first. They tend to add what I call "excess phase" because their phase detracts from phase margin more than would be made up by a reduction in ULGF caused by their roll-off contribution. Sorry for the awkward wording.
BTW, although it lies at a fairly high frequency, many Miller-compensated VAS do in principle go to an all-pass characteristic unless a small resistor is inserted in series with Cdom.
Cheers,
Bob
There is no such causality. As you say, faster output transistors would allow a larger ULGF (or GBW) after compensation, and that's about it. "Excess phase" has no role in this process, it's a byproduct not a root cause.
The "excess phase" concept has it's role mostly in non-minimum phase systems, where the gain and phase can be adjusted independently.
Really?Would you care to elaborate?
The only parameter I know that doesn't follow the minimum phase path is Beta around and beyond Ft (the bipolar transistor transition frequency). A theoretical case, with no impact in real life audio. I've seen people caring about that in very high frequency (GHz) oscillators.