Monte Carlo
Hi Dimitri,
Regarding the trannies which are supposed to be matched (LPT and CM), varying the so called 'value' by 10% had so significant effect (~3%) on THD figures.
But I know of instances where the effect is much larger. Happily, not in this case.
edit: As for a Monte Carlo sim, you can step several component values, simultaneously or nested.
Cheers,
Edmond.
Hi Dimitri,
Regarding the trannies which are supposed to be matched (LPT and CM), varying the so called 'value' by 10% had so significant effect (~3%) on THD figures.
But I know of instances where the effect is much larger. Happily, not in this case.
edit: As for a Monte Carlo sim, you can step several component values, simultaneously or nested.
Cheers,
Edmond.
Last edited:
[snip]
If you are to simulate at 20KHz you will have to set the timebase so incredibly small that you get computational rounding errors. Besides, simulating with an extremely small time scale becomes extremely slow on most machines.
@goatee below: Yes, that's what I'm saying.. Waiting for the sim to reach a steadystate on a very small timescale takes way too long, even on my overclocked machine
Hi MagicBox,
I'm not sure what kind of simulator you are using, but the slowness you described I have never encountered. The THD20 of the blameless amp takes less than one second (with a core duo CPU running at 3.16GHz)
Cheers,
E.
I use MultiSim 2009, but I most of the time used the start from zero condition rather than the DC calculation. Getting to a stabilized DC point from zero takes a looooooong time. The simulator generally doesn't like to calculate the DC bias on more complex designs. However, I figured out a way for it to calculate the DC point now. For most simming it doesn't matter though - the only big caps in a circuit are the NFB cap and the input cap which have no charge at all.
I think there is seldom much to be learned from including very low time constants in the sim - although it is helpful if the DC operating point sim is close for DC servos, sub audio bypass
low frequency audio amp distortion is pooly simmed by Spice due to the lack of "Live" device temperature modeling with coupled thermal models
feedback DC blocking C shouldn't give major changes in operating points, distortion for signals larger than the output Vos and could be shorted for sim
low frequency audio amp distortion is pooly simmed by Spice due to the lack of "Live" device temperature modeling with coupled thermal models
feedback DC blocking C shouldn't give major changes in operating points, distortion for signals larger than the output Vos and could be shorted for sim
Hi Edmond,
I agree with you. Like you and many others here, I have done a Bzillion THD20 simulations on many designs, often on my laptop with LTspice. The accuracy is good and the simulation time is reasonable. I have also done 19+20kHz CCIF IM simulations, which can take even longer than THD20.
It is certainly true that performing the simulations without long time constants in the circuit is very helpful. I run my power amplifier simulations without an input coupling capacitor and without a feedback network LF capacitor. In simulation, it is easy to null out any DC offset for purposes of simulation without capacitors that introduce long time constants.
Cheers,
Bob
Hmm... Multisim. I've tried Multisim about 10 years ago. I didn't like it, but admitted, that was a much older version of course.
Probably, LTspice is faster. Just give it a try (and delete or short circuit the input and GFB caps, just as Bob does).
Cheers,
E.
Alternate architectures
A question,
It has been customary to stick with the cannonical architecture LTP-VAS-OUT based on discrete parts, with but exceptional mention to alternatives (I recall something from Pavel Macura years back) hybrid approaches like high performance audio Op Amps and low voltage gain power output stages enclosed in a global feedback loop.
I assume this particular forum is not conditioned by the kind of purist prejudices against commercial IC's but rather by rational performance based judgement, so if there are no popular hybrid approaches, then probably there are good reasons for this. Or not?
I wonder because, lacking better information, I am led to think modern high performance audio Op Amps like the OPA1641 and derivatives or similars, probably designed by skilled people backed by important corporate R&D resources and benefitting from features like thermal tracking and reduced parasitics etc. due to integration, should command some form of advantage.
Note that unavoidably this very same parts (high performance Op Amps) are present somewhere in every audio chain, so "Op Amp sound" is not an issue.
In fact my amplifiers are just designed that way but will not claim insofar the "the amplifiers I build are the best sounding ones I've ever heard" syndrome takes precedence.
So the question. Is there a solid set of reasons why a discrete design will be better than an hybrid one?
Said otherwise, given the same level of design skill and resources, will the discrete amplifier allways outperform the hybrid one?
I wonder
Rodolfo
A question,
It has been customary to stick with the cannonical architecture LTP-VAS-OUT based on discrete parts, with but exceptional mention to alternatives (I recall something from Pavel Macura years back) hybrid approaches like high performance audio Op Amps and low voltage gain power output stages enclosed in a global feedback loop.
I assume this particular forum is not conditioned by the kind of purist prejudices against commercial IC's but rather by rational performance based judgement, so if there are no popular hybrid approaches, then probably there are good reasons for this. Or not?
I wonder because, lacking better information, I am led to think modern high performance audio Op Amps like the OPA1641 and derivatives or similars, probably designed by skilled people backed by important corporate R&D resources and benefitting from features like thermal tracking and reduced parasitics etc. due to integration, should command some form of advantage.
Note that unavoidably this very same parts (high performance Op Amps) are present somewhere in every audio chain, so "Op Amp sound" is not an issue.
In fact my amplifiers are just designed that way but will not claim insofar the "the amplifiers I build are the best sounding ones I've ever heard" syndrome takes precedence.
So the question. Is there a solid set of reasons why a discrete design will be better than an hybrid one?
Said otherwise, given the same level of design skill and resources, will the discrete amplifier allways outperform the hybrid one?
I wonder
Rodolfo
"the same" design skill, time resource would be a major limitations - you're likely looking at more effort, deeper understanding required to better the overwhelmingly preferred in practice, highly developed standard circuit that you can get pretty good knowledge of by reading Self, Bob maybe a few dozen papers in JAES on the conventional LIN/RCA topology amplifier design - far cheaper in the time resource than developing similar depth of understanding on your own
its not like reading Halcro's patents are really going to reveal a fraction of what you would need to know even to implement his approach
would just tweaking the output bias to a few hundred mA and using better op amps make the QSC CE output topology work to competitive levels with "enhanced" LIN/RCA designed by the book(s)? is it just a mental block rejecting floating supplies?
I suspect there an issue in the protection from a whole zoo of overload/clipping phenomena "topologically built in" when you use the standard audio amp unity gain output stage - even there patching up the VAS clipping behavior may still be needed
could be a fun intellectual challenge but unlikely to "win" any converts even if wildly successful in making yet another amp with excellent measurements
its not like reading Halcro's patents are really going to reveal a fraction of what you would need to know even to implement his approach
would just tweaking the output bias to a few hundred mA and using better op amps make the QSC CE output topology work to competitive levels with "enhanced" LIN/RCA designed by the book(s)? is it just a mental block rejecting floating supplies?
I suspect there an issue in the protection from a whole zoo of overload/clipping phenomena "topologically built in" when you use the standard audio amp unity gain output stage - even there patching up the VAS clipping behavior may still be needed
could be a fun intellectual challenge but unlikely to "win" any converts even if wildly successful in making yet another amp with excellent measurements
The biggest reason why opamps generally aren't used in discrete amp circuits in my opinion is their limit in supply voltage. Most go up to +/-15V, some a bit higher. When you get into higher power rails along with higher power output the border between opamp and chip amp gets really thin.
Hi-speed opamps are used abundantly in high performance linear PSUs, some of which can source/sink. In essence these circuits are just a huge power amp that handle transients well above audio range.
It's relatively easy to use highperformance opamps as a difference driver for a power amp, but the power stage and the powerstage driver supply need a floating topology while you want a low voltage lineair supply for the opamp. Or severely limit the output stage.
Hi-speed opamps are used abundantly in high performance linear PSUs, some of which can source/sink. In essence these circuits are just a huge power amp that handle transients well above audio range.
It's relatively easy to use highperformance opamps as a difference driver for a power amp, but the power stage and the powerstage driver supply need a floating topology while you want a low voltage lineair supply for the opamp. Or severely limit the output stage.
Hi Rodolfo,
The best answer is .... it depends.
There are some good examples is my book in Chapter 27 "Integrated Circuit Power Amplifiers and Drivers."
There are at least two classes to be considered beyond just using an op amp in some way to achieve high voltage swing. First there are the fully-integrated IC power amps like the LM3886 and other Gain Clone type designs. These can develop surprizingly good performance, but don't generally aspire to the performance and power levels of all-discrete power amplifier designs.
Second, there are power amplifiers built with high-voltage IC driver chips connected to fairly conventional BJT or MOSFET output stages. An example driver is the LME49810. These designs can deliver VERY good performance, rivaling that of well-designed discrete power amplifiers and bettering many that are not designed with the highest skill or quality (yes, there are many such discrete designs out there).
Hope this helps.
BTW, I am posting from the Rocky Mountain Audio Fest. Jan Didden and I will have an exhibit table there displaying my book and his Linear Audio. Drop by and chat with us if you can.
Cheers,
Bob
These are out of the question, performance is well below in an order of magnitude with respect to state of the art for example ppm thd@20KHz. This notwithstanding, a large retinue of fans flock after the LM3886 but this is an entirely different issue.
Can you (or anybody) point to some examples including performance data?
On the other hand, perhaps the requirement of high voltage drivers pushing unity gain output stages is a tall order to fill. This is why I at least have built my amplifiers with 3x voltage gain output stages driven by audio OpAmps at normal supply levels.
Rodolfo
Not if you use them as power boosters inside of an opamp feedback loop or EC or Howland or other composite topology.
Sorry but that reminded me of application note AN-272 that's available at national.com, "Op Amp Booster Designs", where in Figure 9 they use an entire McIntosh 75 tube amplifier inside of the feedback loop of an opamp. (grin)
Last edited:
"Sorry but that reminded me of application note AN-272 that's available at national.com, "Op Amp Booster Designs", where in Figure 9 they use an entire McIntosh 75 tube amplifier inside of the feedback loop of an opamp. (grin) "
"
LoL!!
That no doubt would have cleaned that tube sound up a bit!
"
LoL!!
That no doubt would have cleaned that tube sound up a bit!
I did a test once with an LM3886 against a MOS complementary symetry output stage as power output. This test used an OPA 1641 with the power output stage enclosed in the feedback loop (and some lead compensation tweaking for stability).
The discrete MOS output stage measured less than 0.001% with and input signal simultaneously measured at less than 0.001% THD@1KHz as best as I could measure.
This is against somewhat less than 0.01% for the LM3886 which anyway was slightly better than regular specs.
Rodolfo
Hi Gootee, thank you very much for the info!!
I looked at it and noticed:
The best figures as measured for the LME49810+discrete output by Panson at best bias is 0.001% or 100 ppm if I am not wrong, and at 1 KHz.
Now, Edmond Stuart quotes THD20 figures simulated in the 10+ ppm range for discrete TMC design.
I tend to take this figures as meaningful given the large improvements in models and simulation tools (though of course should like to see actual measurements on a tangible assembly).
I did get probably lower than 100 ppm with the OPA 1641 and MOS output stage, best I could measure was 0.001% in and 0.001% out THD@1KHz. The OPA 1641 is rated at 0.00005% THD@iKHz so it has the muscle to tame a well designed output stage in the low ppm range I guess ...
Rodolfo
Last edited:
OOps sorry!!
Yet you get the idea, Stuart's results are THD@20KHz, you can expect almost an order of magnitude degradation with respect to 1KHz so still relevant I think.
Rodolfo