Hi Bob/John/Nelson
I don't know whether you are taking aim at someone who only uses simulation ...
a good guide is "if the computer says something won't work, it is probably right. If the computer says it will work, you check it"
... but from my point of view you need the simulations to confirm that low distortion is possible. If the simulation shows high distortion, it's likely to be right.
cheers
John
I don't know whether you are taking aim at someone who only uses simulation ...
a good guide is "if the computer says something won't work, it is probably right. If the computer says it will work, you check it"
... but from my point of view you need the simulations to confirm that low distortion is possible. If the simulation shows high distortion, it's likely to be right.
cheers
John
Ever drive a 924 or a 944? That was essentially what I was referring to. Why quibble? Learn and grow! For the record, I owned and drove a Renault Dauphine approximately 95,000 miles in 5 years and I know its characteristics well. I have also owned a Porsche 924 and currently own a 944. I know their driving characterisics just as well.
Nelson Pass said:[snip]You may have been there at AES when Lipshitz and Van der Kooy
presented their (masterful) analysis on the Quad current dumping
amplifier. When asked by a member of the audience how it
sounded, one of them said (and I paraphrase here), "I don't
know, we didn't listen to it."
Nelson,
I share your irritation. We all know that listening tests have nothing to do with a technical analysis. Yet there is always a misguided fellow human being that asks the most irrelevant question for the context...
Kudo's for whoever gave the answer for not letting himself being drawn out and giving the reasonable answer.
Jan Didden
I am trying to make a comparison between indifferent phono design and excellent phono design by using autos as a comparison. The name Porsche is written on both the 924 and 944 in prominent letters and they were at least 'set up' by Porsche at the factory to be able to go into a controlled drift with a competent driver at the wheel.
The Renault would actually roll over, if you got into a tight spot, rather than drift. That is a serious design difference. When I switched over to Pirelli tires on the Renault, it too could be put into a controlled drift as they would break away into a drift, before the Renault started to think about rolling over.
The best phono design, in my opinion, requires that you accept what is actually being developed by the phono cartridge in reasonable worst-case conditions and design the phono stage to not clip, slew rate limit, or even significantly distort during cartridge mistracking.
Bob appears to ignore mistrackiing as a design reference. That is the difference in our positions.
The Renault would actually roll over, if you got into a tight spot, rather than drift. That is a serious design difference. When I switched over to Pirelli tires on the Renault, it too could be put into a controlled drift as they would break away into a drift, before the Renault started to think about rolling over.
The best phono design, in my opinion, requires that you accept what is actually being developed by the phono cartridge in reasonable worst-case conditions and design the phono stage to not clip, slew rate limit, or even significantly distort during cartridge mistracking.
Bob appears to ignore mistrackiing as a design reference. That is the difference in our positions.
S-K Filter Simulation and Monte Carlo
Nelson,
I have seen the same behavior in simulations and believe it has to do with the phenomena occurring at the onset of oscillation. In order for oscillation to start the simulator may need to provide a "kick" to push the system under consideration off a local minimum. One simple workaround is to intentionally introduce a small amount of random noise or a noise pulse.
We have also seen the converse of this problem, where an amplifier will apear stable under quiescent conditions but will oscillate when presented with certain input signal conditions.
I am glad to see that someone has finally mentioned MC techniques. It is probably the only feasible way to guarantee proper circuit behavior over the range of voltage, temperature, and manufacturing variations. this becomes even more the case when large numbers of input conditions must be comprehended
One extension to MC that we have found particularly useful is Design of Experiment (DOE). This technique constructs a mult-dimensional curve that fits data points taken on each simulation run and then uses a curve fitting algorithm to construct a multi-dimensional response surface. Once this surface has been defined it is possible to estimate output results for any set of inputs that lie within the bounds of the response surface. The DOE tool also generates parametric sensitivity curves, so one can determine sensitivity of each input parameter. We use the JMP(R) tool where I work, but do not know what the cost of the tool is or whether lower cost alternatives are available.
Nelson,
I have seen the same behavior in simulations and believe it has to do with the phenomena occurring at the onset of oscillation. In order for oscillation to start the simulator may need to provide a "kick" to push the system under consideration off a local minimum. One simple workaround is to intentionally introduce a small amount of random noise or a noise pulse.
We have also seen the converse of this problem, where an amplifier will apear stable under quiescent conditions but will oscillate when presented with certain input signal conditions.
I am glad to see that someone has finally mentioned MC techniques. It is probably the only feasible way to guarantee proper circuit behavior over the range of voltage, temperature, and manufacturing variations. this becomes even more the case when large numbers of input conditions must be comprehended
One extension to MC that we have found particularly useful is Design of Experiment (DOE). This technique constructs a mult-dimensional curve that fits data points taken on each simulation run and then uses a curve fitting algorithm to construct a multi-dimensional response surface. Once this surface has been defined it is possible to estimate output results for any set of inputs that lie within the bounds of the response surface. The DOE tool also generates parametric sensitivity curves, so one can determine sensitivity of each input parameter. We use the JMP(R) tool where I work, but do not know what the cost of the tool is or whether lower cost alternatives are available.
The simulation varied all parameters simultaneously through fifty different iteration, I didn't isolate individual parameters. It was basically a sim of fifty unmatched randomly selected transistor pairs.
Yes, I did keep Ic matched, I forgot to mention I also used a extremely low frequency dc servo made from a voltage controlled current source and a ridiculously large capacitor.
I only very recently grasped it (and probably not completely, just intuitively), I'll see how well I can describe its operation later.
darkfenriz said:
To be honest, the only real way to know for sure would be to build two identical output stages/amplifiers - one using the Sanken parts and one using the Onsemi parts.
I'm taking a punt with the Onsemi parts though. At the end of the day, I really don't think there would be all that much in it, either way.
But don't be fooled by the 200W NPN Sanken 60MHz fT spec.
Cheers,
Glen