PMA said:
Sorry, I didn't make myself clear. The noise from the CRT monitor can easily creep into almost any high-gain DUT.
The oscillation problem I saw *one time* only occurred with with specific product. I'm still not sure of the exact mechanism -- it wasn't worth my time to figure it out. I solved the problem by re-routing the speaker (dummy load) wires, or grounding the heatsink with the load resistors, or something like that. It was years ago and I don't recall the exact solution.
The point is simply that the oscillation that John Atkinson saw in the Parasound amplifier was not necessarily an instability in the amplifier, but instead could have been an interaction with the System One that would not happen in a real stereo system.
Or it could have been an instability in the amp -- I don't know. In my experience, there are usually a couple of different values of load capacitance that create the worst problems. Usually somewhere around 0.01 uF is the toughest, but the local maximum tends to be quite sharp and it is easy to miss a potential instability if you are just testing with decade values. It is better to at least use a 1 - 2 - 5 - 10 scale to make sure you don't miss something.
By the time you get up to 1 uF or so, the amp often becomes more stable than with lower values of capacitance. A 10 uF cap driven by a 10 kHz square wave will create large peak currents and stress the amplifier that way, but (in my designs, at least) the amp exhibits less overshoot than with smaller values of capacitance.
h_a said:
I hope as many people as possible are going to read, read and then read again your message!
There's an exchange of messages above regarding this topic. The short answer is: poor device models. The second short answer: Spice (as any other simulation tool) is for the already illuminated. Otherwise, it is as useful (and has the same type of limitations) as a set of screwdrivers. From this perspective, Scott and any other IC designer in the industry is living in the best of the worlds.
The thing you have to understand about Scott's situation is that he's not using MicroCap or any of the usual simulation programs. He's using a custom in-house program available only to those at Analog Devices. Furthermore, they fabricate their own devices to their own specifications...not an option available to the average DIY member. As such, his success with modeling is unattainable to the man on the street. To my way of thinking, his success with simulation is as remote from our circumstances as if he were an astronaut on the moon.
Grey
Grey
Thank you for your agreement!!
Grey, I know you always point at these two reasons; in fact Spice is just an equation solver and that is not such a complicated task that 20 years of development (like in the case of Micro Cap) would not be sufficient to achieve reasonable results in the majority of cases.
I really agree with syn08 on the quality of the device models; not only of the derivation of the pure numbers from the measurements, but also the models themselves. Of course Scott is far ahead in both areas, no doubt.
However I'm really astonished how little (electrical) complexity is sometimes sufficient to fool Spice.
All the best, Hannes
PS: maybe I should spice your GR25 😀
Grey, I know you always point at these two reasons; in fact Spice is just an equation solver and that is not such a complicated task that 20 years of development (like in the case of Micro Cap) would not be sufficient to achieve reasonable results in the majority of cases.
I really agree with syn08 on the quality of the device models; not only of the derivation of the pure numbers from the measurements, but also the models themselves. Of course Scott is far ahead in both areas, no doubt.
However I'm really astonished how little (electrical) complexity is sometimes sufficient to fool Spice.
All the best, Hannes
PS: maybe I should spice your GR25 😀
I'm self employed (in today's economy this is also known as involuntary retirement) so I use the free Linear Technology simulator. It is very powerful, in this case that is just another word for complicated.
Recently I spent some time trying to understand why my design had such poor rise times in simulation. It took me a while to discover that LTC spice defaults to something like a 1uS rise time, after setting rise time to 100nS my problem went away.
Among other features, the LTC simulator will do Monte Carlo analysis, a process that is prohibitively expensive with a real circuit on a breadboard.
There is a LTC users group on Yahoo LTC spice users
They are quite helpful often your question has already been answered. They also have component library extensions that includes some vacuum tubes.
The biggest problem with the simulator is that it is quite slow if you want for example 100dB signal to noise answers. I usually simulate and optimize subsections at lower resolutions before splicing it all together for a run of the complete design. One other hint, a circuit bordering on instability or clipping often simulates considerably slower than one with good phase/gain or head room margins.
Recently I spent some time trying to understand why my design had such poor rise times in simulation. It took me a while to discover that LTC spice defaults to something like a 1uS rise time, after setting rise time to 100nS my problem went away.
Among other features, the LTC simulator will do Monte Carlo analysis, a process that is prohibitively expensive with a real circuit on a breadboard.
There is a LTC users group on Yahoo LTC spice users
They are quite helpful often your question has already been answered. They also have component library extensions that includes some vacuum tubes.
The biggest problem with the simulator is that it is quite slow if you want for example 100dB signal to noise answers. I usually simulate and optimize subsections at lower resolutions before splicing it all together for a run of the complete design. One other hint, a circuit bordering on instability or clipping often simulates considerably slower than one with good phase/gain or head room margins.
Grey,
in case I am allowed to ask you, what is your real experience with simulation software, namely evaluation and professional versions of MicroCap or PSpice?
Regards,
Pavel
in case I am allowed to ask you, what is your real experience with simulation software, namely evaluation and professional versions of MicroCap or PSpice?
Regards,
Pavel
janneman said:
Again, I don't remember exactly what I changed. The more I think about it, the more it seems like this is what I did:
Our dummy load comprises 4 each 8 ohm 250 watt 1% Dale non-inductive power resistors. They are mounted on a 1/8" thick aluminum plate. The plate was sitting on top of a cheap, old B&K (not Bruel & Kjaer) dual-channel analog scope that we use to monitor the waveforms. If I recall correctly, the oscillation went away when we put an insulator (wood or cardboard or a magazine) between the dummy-load plate and the oscilloscope. Presumably there was some sort of (capacitive?) loop created. Remember that the PC, the AP, and the 'scope all had AC safety grounds, so there is plenty of potential for mischief.
The "bench" at our AP test station is an old surplus metal desk. It is extremely rugged and heavy, made from steel, but with a formica cover on the top. Sometimes I have seen different results with high gain DUTs (eg, phono stages) depending on how (or if) I ground the metal desk.
The point is that when you hook up a DUT to a measurement system, you are not necessarily measuring *just* the DUT. You are measuring the entire system, including the measurement system. And more than once I have seen "problems" that were due to the measurement system and not the DUT. It is possible that this is what happened with the Parasound amp that Stereophile tested. I think one of the posts said that JA was unable to duplicate the problem, so I would guess that in this case the oscillation was an artifact of the test setup and not an instability of the amplifier. But that is only a guess.
janneman said:
The manual for the System One is probably the worst manual for any product I have ever seen. But I didn't even know there was an internal fuse for the ground connection until we had owned the unit for at least five years. We were getting a funny result on one particular unit and I was not seeing what I expected to see when I changed from "Floating " to "Grounded". After an hour or two of fiddling around I finally discovered the internal fuse and found that it was blown. I have no idea how long it had been that way. It could have been blown for years for all I know. It would be nice if there were an external indicator that let you know the status of that fuse. It's not all that hard to blow it.
john curl said:
John,
Once again, you are mis-characterizing what I said. Read the exchange. It was a friendly technical exchange. Look at the language. I had no intention of demeaning your amplifier or you. But I did want to understand the limitations of not using a coil.
I have said many times that the JC-1 is a fine amplifier, and I still say that. I must emphasize that there is nothing wrong with an amplifier that uses no coil and cannot tolerate a 2 uF ceramic capacitive load at its terminals. I was just trying to understand if yours was able to do that and whether you would illuminate us on how.
I will state quite openly that I do not think that I could design an amplifier without a coil that would pass such a test, so I was really curious to know if you were able to accomplish that.
I also note that I was not referring here to the oscillation that happened in the review. I do, of course, thank Charles for illuminating us on the possible interaction with the AP system.
My question about whether you had SPICE'd the JC-1 into a capacitive load was just that - a simple question, not a challenge, and certainly not a demand.
If you don't like something I ask or say, that's fine; I just ask that you portray it accurately and not grossly exaggerate our differences.
Peace,
Bob
Bob Cordell said:
john curl said:
Bob Cordell said:
QUOTE]Originally posted by Charles Hansen
Funny how that works. Get rid of the feedback in the Wilson current mirror and it lowers the high-order distortion.... Now where have I heard that before....hmmm....let me think....[/QUOTE]
It’s likely to increase noise in return?
I thought that John uses vertical mosfets in the blowtorch? John?
How much breaking time for such circuits?
Thank you for all this hints 🙂
Funny how that works. Get rid of the feedback in the Wilson current mirror and it lowers the high-order distortion.... Now where have I heard that before....hmmm....let me think....[/QUOTE]
It’s likely to increase noise in return?
I thought that John uses vertical mosfets in the blowtorch? John?
How much breaking time for such circuits?
Thank you for all this hints 🙂
h_a said:
Obviously your mileage varied. First off MODELS, MODELS, MODELS discrete devices usually have woefully inadequate models. It takes a huge investment in people (device physicists, characterization engineers) and millions of dollars worth of equipment to build a model library. We now work in SOI (no parasitic devices) this makes a huge difference. Our simulator also has temperature as an independent variable so device temperatures can vary independently during transient simulation. You don’t breadboard 1.2 GHz op-amps, the wavelength of the signals certainly assures this. And yes people still can’t believe this, you can design a ~100 or so transistor diff A/D driver with -80dB of distortion at >100MHz< and have the silicon results lay right on the simulations.
Once again, there IS NO “SPICE”. By which I mean there is nothing going on but automation. Funny buying an HP35 to replace your slide rule is OK. BTW I could never afford one. The one I got my first week on the job cost the company about 2 weeks pay, about 4 times what a top laptop costs now. I also hated the programming features of the HP calculators and never used them. OTOH you would have to pry my 11C from my cold dead hands.
Take a big sheet of white paper and tape it to the wall. Now take the JC-2 schematic and label the nodes and branches using the simplest hybrid-pi models for the 6 devices. Now write by hand the matrix equations to solve for the V’s and I’s. OK now using whatever you learned in college algebra and circuit theory classes solve for the V’s and I’s. If you want an AC analysis you’ll have to use complex math. Be warned you’ll need a lot of scrap paper to keep track of all the intermediate terms of those determinants. Very soon you might start realizing that there is a lot of tedious repetition involved and even one little +- inversion will ruin the whole thing. You are also not thinking creatively while doing it and not really learning anything about this or any other circuit. That’s all SPICE (or any other similar tool) is doing. Nothing more than a calculator.
In 1969 at MIT we used APL
to fill in and solve simple matrixes as part of Jim Roberge’s class, same thing same answers. Also, please go lightly with that "We believe in SPICE results" We are not all named Dorothy though many of us are munchkins.
Best of luck Hannes
Bob Cordell said:
First of all, it is important to note that a BJT emitter follower is inherently unstable into a capacitive load. (This has nothing whatsoever to do with the presence or absence of any feedback loops or "error correction".) It requires some sort of changes to the circuit to make it stable.
The standard approach for an audio power amp is to isolate the capacitive load with an inductor. If you look at the app notes for many op-amps, they simply suggest to put a series resistor of 10 to 100 ohms to isolate the load. Obviously this will not work well with an audio power amp.
The idea of isolating the speaker from the amplifier did not sit well with me. So when I designed my first BJT output stage I decided to make it be stable without the use of an inductor. It took over a month of full-time experimentation to find a system that works. I don't want to divulge the details, as I consider it proprietary. But for those who read the ARRL handbook (as I did as a teenager), look up the section on "neutralization". (I don't know if this information is still in the current editions. You may have to find an old one that still has information on tubes.)
Like everything else, if it were easy, everyone would be doing it. Instead, everyone does the easy thing and uses an output inductor. But doing things the same old way doesn't lead to any advances....
No Scott, you can just BUILD a prototype, and measure it. I have the first JC-2 phono prototype at my office, built into the equivalent of an aluminum baking pan. I also have the original GD modules that became the line amp. To do a matrix would be ridiculous.
I now use the HP32 calculator. The more like a 35, the better. I never liked programming the HP65, either, but I did have some pre-made programs for the 65 that were very useful.
At least your position is made clear, Scott. I hope that mine has been as well.
I now use the HP32 calculator. The more like a 35, the better. I never liked programming the HP65, either, but I did have some pre-made programs for the 65 that were very useful.
At least your position is made clear, Scott. I hope that mine has been as well.
Bob Cordell said:
In my experience, this is where SPICE is at its least usefulness.
When you are looking at oscillations in the MHz range, then you need to have ultra-accurate device models. Even the board layout and parasitics of the passive components becomes very important.
And if you study the inherent instability of an emitter follower into a capacitive load, you will quickly find that it is *not* well understood. (If it were, there would be widely known and used methods to fix the problem. And adding a band-aid inductor to isolate the load does not *fix* the problem, it only hides it.) So if the problem is not well understood, I would hardly expect the models to be accurate enough to allow SPICE to be useful here.
As Pavel has shown recently on this thread, SPICE can be helpful in understanding certain fundamental relationships of the device performance. But I would never rely on it to accurately predict whether or not a design was stable into a capacitive load. Especially when it is so easy to test in the real world....
Re: neutralization
Yep,
ARRL, 1941 edition, pgs. 78pp., 212pp
Another way to make an emitter follower output resistive is to find the right R//C at the base, which does the trick per impendance transform. The RC must match the -3dB-corner of transistor beta and the beta roll-off is assumed as 1st order, which are the weak points of this.
- Klaus
Yep,
ARRL, 1941 edition, pgs. 78pp., 212pp
Another way to make an emitter follower output resistive is to find the right R//C at the base, which does the trick per impendance transform. The RC must match the -3dB-corner of transistor beta and the beta roll-off is assumed as 1st order, which are the weak points of this.
- Klaus
Hp32e
- Klaus
After the slide rule, a HP32E (is this the same as HP32?) was my first calculator. In fact I had several of them, the earlier series had long-term contact problems as the IC's weren't soldered, only pressed down onto a foil-on-foam PCB with a plastic frame.... I wished there still were RPN-based calculators around (which are when you have a PC, install a FORTH-system)john curl said:
- Klaus
Charles Hansen said:
Agreed. It's pretty difficult to simulate a harmonic oscillator (even if it's not by design 🙂 in Spice, and to correctly predict the frequency and the amplitude of the oscillations.
The only thing that Spice can always do for you (within the current discrete devices models accuracy) is to analyze a stability criteria for the feedback loops (and yes, even a degeneration is a feedback loop and can be analyzed as such). If the open loop circuit doesn't meet the stability criteria by a consistent margin, then that's something I wouldn't put on a PCB. If models are not validated, this can be used more as a qualitative analysis (or go/no go criteria) rather than expecting precise results.
h_a said:
By all means, if if makes you happy to simulate the GR-25, then do so, although it would probably be more appropriate to post the results in the GR-25 thread than here.
The purpose, in the broadest sense, of electronics as a hobby is to give pleasure. If simulation is your stimulation, then do so with my blessing. Speaking only for myself (a member of an ever-shrinking minority, it seems), I prefer to actually build the circuits so as to have something I can listen to. I have yet to hear of a simulation program that's sophisticated enough that it will simulate the playback of music through the circuit that's being modeled (while compensating for the limitations of PC playback in real time, naturally). Given my time limitations, I don't feel that it's worth my time to build the same circuit twice--once in virtuality, then again in reality. It's much more efficient to build once, then listen during the time that I would have spent building the virtual version of the circuit.
I have no quarrel with people who simulate for pleasure. My problem is with those who mistake it for reality and cop an attitude towards those who decline to join their religion. Or, heaven forbid, point out the flaws in their belief system.
PMA said:
Formerly Pspice. I forget what version I had; this was six or eight years ago. I've got MicroCap, but haven't done much more than poke at it.
This is the point in the conversation where you inevitably see a succession of posts saying something along the lines of:
Poster A--Pspice! Pspice is a dog! I use Auto-Simu 4.7.1.2a and I get perfect simulations every time.
Poster B--Auto-Simu! Sweet Mother of Gawd! How anyone can stand that piece of crud is beyond me. I use Mega-Simi because they have the best models.
Poster C--You're both full of it! Dune-spice rocks!
(Yes, that was a shameless attempt to make anyone who reads science fiction smile...)
And so on. It's not enough to use a simulation program, you have to choose sides. Then you've got to argue over models. Then you've got to debate over which 'features' (aka bugs) in the program caused your obviously erroneous results. And if you have any energy left--or in my case, time--you get to build a circuit...then makes excuses as to why reality steadfastly refuses to map to your "perfect" simulated circuit.
If that's your idea of fun, then have at it.
It's not entertaining to me on any level.
Grey
Simulators are just tools. They do some things well and others not so well. A chisel can be used as a screwdriver, it's not a good idea. On the other hand those who say "My dad used a trained groundhog to nibble on wood, what's good enough for my dad is good enough for me", exhibit an age based myopia. The trick is to take the best of he old and the new, that's where the progress is.
- Status
- Not open for further replies.