Richard,
Why not fully balanced ?
And how about caps parallel to R26, R27 ?
Best regards from Oberkochen. Hope all is well.
Patrick
Why not fully balanced ?
And how about caps parallel to R26, R27 ?
Best regards from Oberkochen. Hope all is well.
Patrick
Hi Richard, interesting cascoded current mirror. I used something like it in making a very high linearity, oscillator buffer for Sound Technology, back in the early 80's. It is complex, but very linear. I still prefer resistive loading on the input stage, but I could be wrong at this time, because Charles has done quite well with whatever he is doing. The biggest problem can be noise injection from the active devices. This where an accurate Spice noise simulation can be useful.
Pavel, i think Q2+4, Q6+9 are just there to cascode the mirror transistors Q3+8, they are not really mirrors, even if they look this way. Q3+8 see always the same voltage as Q1+7, this might be the idea for this arrangement.
I think Charles calls these mirrors "cascoded mirrors".
Not sure if Q5+10 are still needed, Richard? You could use beefier transistors for Q2, 4, 6, 9, and discard the cascode.
The Wilson mirrors or those which are shown in the simplified AD844 data sheet have inherent feedback I think....
Tino
I think Charles calls these mirrors "cascoded mirrors".
Not sure if Q5+10 are still needed, Richard? You could use beefier transistors for Q2, 4, 6, 9, and discard the cascode.
The Wilson mirrors or those which are shown in the simplified AD844 data sheet have inherent feedback I think....
Tino
Justcallmedad said:
Interesting circuits. A few points:
- The 2SJ74 has been officially discontinued. This is a very sad situation, as there is nothing left to replace it with. Linear Systems is working on a replacement, but so far they have not been able to make one.
- You did not specify the MOSFETs used for the folded cascodes. But remember that there is no current gain in the cascodes. So why use a MOSFET? The only good reason is that the input stage swings only half the voltage of the cascode stage, and it is easier to find high voltage MOSFETs than JFETs. But otherwise I hate MOSFETs. The vertical ones have incredibly non-linear interelectrode capacitances. The lateral one *could* be good if only some company could be bothered to make them. In this instance an appropriate device could be the J77/K213 pair, except that the P-channel Hitachi parts are very "triode-like" and not very good complements. The Exicon (Semelab) parts are better, but they don't make the small MOSFETs, only the big ones useful for output transistors. And if you throw these things on a curve tracer, you will see that the JFETs have nearly the same transconductance as a big MOSFET for a given current.
As I look back over what I've written, it becomes clear that perhaps the biggest challenge of audio design is finding good parts! You look for the best ones you can find, but even those have significant problems. So you spend a lot of time designing around their limitations. And then - BAM - they discontinue the danged things...
zinsula said:
Don't give me credit for that! This circuit has been around for decades, although largely overlooked in favor of the Wilson mirror.
One of the best sources of information on current mirrors comes from Dennis Feucht. Unfortunately his book is out of print and *very* difficult to find. I spent about 6 years on a waiting list before getting a used copy.
He has a website and sells an "updated" version, but it is more like a "Cliff Notes" version. Still it is probably worth having. Feucht also has the best information on the inherent instability of the emitter follower.
zinsula said:
Yep.
Edmond Stuart said:
I'm one of the guys which used to program such things and yes, we heard about "singular value decomposition" and way more other linear algebra notions and tools.
My skepticism regarding the "Spice" kind of tools, in particular for discrete devices circuits, comes precisely from understanding how these tools are working and what are their limitations.
Scott just hit the nail: these tools are nothing more than a) a fancy user interface b) a linear system solver (usually with sparse matrix) and c) a linearizing algorithm for non-linear equations. Errors like the one you mentioned are ultimately unavoidable; there's a limit of the arithmetic precision of a certain CPU and rounding errors that are accumulating. And certainly avoiding things like ideal sources, extremely large values, etc.. helps. Mathematically, there are a few things that can be optimized to avoid such errors and the way and quality of the implementation is strongly dependent on the Spice brand.
The latest PSpice version that I recently received from my friends at Cadence (for which I used to work some time ago) has done some amazing progress in improving the convergence: and Autoconverge tool, nicely integrated in the user interface, that auto-adjust the convergence parameters in an attempt to obtain convergence. It adds time to the simulation, but most of the annoying issues are now history.
It is absolutely naive to blindly rely on simulations, at least for discrete designs where the models are, at best, questionable, and I can't imagine a sane engineer that would think otherwise. The situation is quite different for Scott's designers; they use either extremely well engineered device models, extracted (including tolerances) and specific to a certain manufacturing process. Or even better, fully characterized macromodels for the building blocks. Such an approach is the minimum requirement to rely your designs on a dumb matrix solver.
There is another aspect on how much knowledge a simulator adds, beyond delivering a set of numbers, but that's another topic.
Anyway, Pavel just delivered a magnificent example of what a simulator can do properly (simulating the ratios between various harmonic components) and (at the same time) and example of what a simulator can not do properly (lacking any model for the JFET short channel effects, mobility degradation, etc... the absolute values of these harmonic components will be, most likely, quite different in practice). Is this a Spice issue? Certainly not. The tool properly linearized and solved the linear equations sets.
john curl said:[snip]He was the boss, and he was skeptical of the scheme of using the new and powerful ECAP program to validate our circuits. [snip]
Hey! I did a product review on ECAP for The Audio Amateur, at the time. Forgot the year. Does that make me a geezer 🙁 ?
Edit: should I change my sig ???
Sorry for /OT...
Jan Didden
Jan, I am confused. ECAP was a Fortran based program of many hundreds of pages, that cost big money, and ran on mid size IBM computers. What are you referring to? Anything useful?
john curl said:Charles, don't tell Bob that OLB = CLB when the amp is OLA.
OLA Object Layout Adaption
OLA Office of Legislative Affairs
OLA Office of Legislative Analysis
OLA Office of the Lead Agent
OLA Official Languages Act (Canada)
OLA Official Languages Agency (Hong Kong)
OLA Ohio Library Association
OLA Ohio Lumbermen's Association
OLA Oklahoma Library Association
OLA Oligonucleotide Ligation Assay (method of testing for a specified mutation)
OLA On-Line Academy
OLA On-Line Archive
OLA One Letter Acronym
OLA Online Advertising, Inc.
OLA Online Authorization (implementation of EFT for POS systems)
OLA Ontario Library Association
OLA Ontario Lung Association (Canada)
OLA Open Learning Agency
OLA Open Learning Australia
OLA Open Library Architecture (OpenEDA)
OLA Operating Level Agreement (information technology services)
OLA Operational Level Agreement
OLA Optical Laboratories Association
OLA Optical Laboratory Association
OLA Optical Line Amplifiers (Ciena, Pirelli, NEC, Lucent)
OLA Optical Link Amplifiers (Optivision)
OLA Ordnance Loading Apron
OLA Oregon Library Association
OLA Orlando Latin American
OLA Our Lady Assumption (school)
OLA Output Level Adjust
OLA Overlap and Add
- Status
- Not open for further replies.