About circuit design optimization _ a general question

Member
Joined 2009
Paid Member
Good morning to Everyone
I immediately apologize for a reasoning that probably will sound inconclusive
I recently discovered the LTSpice program but I couldn't get decent THD vsHz graphs
With the very kind and precious help of Mr Mooly who provided me with the most appropriate settings I began to see some THD graphs and I fell in love with them
I started playing with voltages and values and types of transistors in the most banal circuits in order to reach a lower THD and I saw that the result changes considerably from total non-functioning to rather interesting results.
Now the question that's on my mind.
Looking at the most popular discussions here in the forum I think I'm quite isolated in my curiosity to see the limits of simpler circuits (i.e. low parts count)
I guess that simpler topologies must have intrinsic limits and this makes them not interesting for who wants to reach the very best measurements.
But what are their real limits?
I mean there are a lot of variables already in a simple circuit from the type of parts and values and supply voltages etc.
If optimizing a simple circuit already appears so challenging and demanding what would be optimizing a very complex one (i.e. with many many parts)?
I have seen a discrete opamp with 80 passive and active parts o_O
The number of variables increases dramatically at a level that is pratically impossible to manage.
I have the feeling that people are more challenged about doing something complex than getting the most out of something simple.
like those chefs who don't care about the quality of the raw materials but add tons of different sauces to their recipes that only confuse the sense of taste
Maybe it's my impression
On another hand I was also struck by a review of a preamp some time ago
The reviewer, clearly an electronics expert, believed the circuit quite a bit. In fact, he saw some even trivial flaws in it
Yet that circuit tested on the bench still gave truly remarkable measurements in terms of THD+noise
His explanation was that the results were achieved with an optimization of a circuit that on the paper looks quite simple.
What is your approach on the subject ?
And moreover how do you optimize your designs ?
 
Last edited:
  • Thank You
Reactions: 1 user
Complexity is in the eye of the beholder.
One needs to understand the critical knobs to turn to optimise. Many knobs have low sensitivity to optimization, while others are very sensitive.
Often components are added to make a circuit easier to analyze with fewer knobs to optimise.
Take for instance a long tail pair. Resistor vs a current source in the tail. The current source dramatically improves CMRR, so other optimization parameters regarding CMRR are less critical.
In many ways more components can be simpler to analyze and optimise.
 
  • Thank You
Reactions: 1 user
Audio does not have a single attribute to optimize. Rather, there are a dozen independent attributes. The goal is to stay within reasonable bounds on all of them. One should not try to minimize or maximize a single attribute if doing so will come at the expense of others.

Circuits with many components contain building blocks that can be analyzed more-or-less independently. One designs the building blocks and then designs their interactions. This avoids dealing with a large number of combinations. Often, one uses more components to get closer-to-ideal performance.
Ed
 
  • Thank You
  • Like
Reactions: 1 users
I assume it's about audio circuits.
Another experience plays an important role: the audible modulation of the signal due to material resonances and the current quality of the parts: "Noise". So: the more parts, the more the modulation: the higher the audible "noise". So: the fewer parts, the lower the audible "noise"-)

By the way: an OP-amp with 80 components was certainly not developed for audio purposes. Amplifier technology is developed for all "forest and meadow applications". Usability for audio is not self-evident.
Only about every 10th electrical part is suitable for audio. 90 % make a lot of "noise" or even do not get the tones together;-)
... I have also heard parts with an "echo", a "shadow";-)
 
  • Like
  • Thank You
Reactions: 1 users
Member
Joined 2009
Paid Member
Hi i thank you very much All of you for the very kind and valuable advice
I have to apologize I count better and the discrete opamps i mentioned as "only" 52 parts either active or passive
From what i see it is one of the most complex in terms of parts count at least Also a very high +/-50VDC supply
It is uses as buffer with gains 2 and 10 in a preamp from Goldmund that i listened many years ago and that seemed to me like the concept of the wire with gain materialized I was fooled into thinking I was hearing real instruments A very impressive experience
I am quite amazed that products so complex can be so transparent Also opamps have many parts I am reading from the web that
The standard 741 Op-amp circuit contains 20 transistors and 11 resistors

I would have just another general question How much space sim SW and lab measurements have in establish the quality of a design
I mean are the steps like this

choice of topology > circuit drawing > simulation > construction of the prototype > lab measurements > fine tuning (?)

i would love to hear all the steps to start from the right one
 
A lot of design can be done with hand DC and AC circuit analysis. Simulation can handle non-linear circuits and complex interactions. This is where the design space can be explored and 99% of bugs can be eliminated.

When these steps are done well, the prototype should perform as expected. Any problems found late require high effort to fix.
Ed
 
  • Thank You
Reactions: 1 user
Even the development of audio devices does not correspond to the idea of an ideal development process.
And if you want to be involved in the "world participation chain" (not "value chain") with lots of numbers ("money"), you can't offer just a few numbers (quantitative measurable...) so it comes down to lots kilograms, lots parts, lots volume, lots of numbers in terms of watts and distortion
- not sound: that is not quantitative measurable...-)
 
  • Thank You
Reactions: 1 user
Member
Joined 2009
Paid Member
Hi i would like to thank you sincerely for your very valuable input as usually it happens to people without knowledge i was trying to trivializing a very complex task I should have imagined that a circuit is always the result of hours of work by expert designers
To be very honest i would like to try to sim the discrete opamp with 52 parts ... just out of curiosity
I will try to do it because as i have said i listen to the preamp using them Easily the best line preamp i have ever heard
For the 1st time i heard real instruments and not reproduced instruments i guess this is how extremely THD+noise sound o_O
The preamp in question is the Goldmung Mimesis 27 of which there is a schematic in the web but not official of course
I guess that the quality of the unit has triggered the cloning community How those clones actually sounds i have no idea honestly
that was made with two of these opamps the 1st buffering the inputs followed by a 10klog volume pot and another opamp with some gain
I am afraid it will take me all the summer to put the schematic on LTSpice ... but i am too curious to see what come out from numbers
Thank you so much to All of you
Have a nice evening :)(y)
 
  • Like
Reactions: 1 user
If I had to build an OP discretely with 52 components in order to listen to music with it, then I would say: No. Not a single clean tone comes out.
The result as an IC is different, as cross-sections, circuits, short current lengths ... provide the sonic advantage over discrete setups.
To get to the bottom of the sound of the Mimesis 27, we have to look at its practical construction. And also in which combination it sounded good.
I, for example, don't use preamps because they unnecessarily degrade the sound)-;

Let me start with an easy exercise:
It looks like in this picture that the signals from sockets and relays run via this flat ribbon including the two connectors - 3 yellow arrows. These things sound like scrap. I recommend simply connecting an input directly to the board, without relays, connectors or flat ribbon - red arrows. And then you can compare them at your leisure. I'm afraid the mimesis spell won't work for long)-;
 

Attachments

  • l.jpg
    l.jpg
    230.5 KB · Views: 53
Last edited:
  • Thank You
Reactions: 1 user
Member
Joined 2009
Paid Member
Hi yes is that one indeed I agree completely with what you say
Maybe a better way could be to place opamp buffers directly on the same board with inputs minimizing the signal path
I should have added that in the more recent products they have switched to opamps And the price is increased a lot
So considering that this preamp could sound very good the consequence could be that the new products using opamps are even better
I have already found a very interesting opamp in the Analog Device catalogue ... the ad811 It is a very powerful opamp and low THD as well
I think that design with discrete are so challenging that even high end manufacturers are abandoning them
Old school Thanks a lot again
 
Regarding transistor-level design of negative-feedback amplifiers, I've been taught (at the Delft university in the early 1990's) to design as orthogonally as possible. That is,

First choose a suitable feedback configuration without worrying about the implementation of the active part.

As long as the input stage has a reasonable gain, noise will usually be dominated by it, so optimize the input stage for noise (and possibly offset, if DC matters). Choose the input device or devices and the bias point, but don't worry about the biasing circuit yet.

Design the output stage for sufficient signal handling. A common-emitter, -source or -cathode stage is preferred to keep the signal levels in the previous stages small (unless there is a good reason to deviate from that; in general, don't be too dogmatic about the whole procedure).

Put as many stages in between as needed to get enough loop gain.

Now design the frequency compensation. Insert common-base stages (cascodes) if needed to keep the high-frequency behaviour manageable.

Design the biasing circuit, and to the extent possible, design it not to affect the signal behaviour substantially. This normally means biasing with current and voltage source implementations rather than with resistors.

The whole procedure is supposed to lead to designs that are close to optimal in performance, although there are exceptions to that. The designs usually have lots of transistors.
 
  • Like
  • Thank You
Reactions: 1 users
Member
Joined 2009
Paid Member
Hi thank you very much indeed I think i am clearing up my mind with your kind and valuable advice
I digress a little but you introduce a very fundamental concept when you say
... designs that are close to optimal in performance ...
how do you know that you have reached a circuit optimal in performance ? by measuring it ? and if so which parameters do you look at ?
 
When we go to the store and buy an amplifier, the circuits are usually the least important part of the sound. Because there is too little knowledge about materials, layout, components and more. The supposed experts do not have the right training. They also lack methods, essentially qualitative ones, because the standard engineering degree courses do not include any scientific training.
I could continue with the necessary modifications so that even a Mimesis 27 would sound much better;-)
 
Member
Joined 2009
Paid Member
If I had to build an OP discretely with 52 components in order to listen to music with it, then I would say: No. Not a single clean tone comes out...
I see but honestly i am puzzled I think i know what kind of sound i am looking for It is fundamental to have a 1st tool ... a well made recording
A recording where the details that make an instrument sound real are all captured I am sure there are and are used by professionals
I remember a series of cds where a speaker voice was captured magnifically They were used to check the intelligibility of words in a hall for conferences
The same can be done with instruments alone and ensemble The best system how someone say gets out of the way
The devices seem even disconnected It is a weird effect that i have experienced with some very good speakers They seem not connected to the power amps. If the 52 pieces preamp can play that trick in a chain i am done ... end of the game.
 
  • Like
Reactions: 1 user
"can play that trick in a chain"
Not in every chain, not even in the most. In a few only.
And it is also not based on this OP amp with 52 components)-;
Proof could be: some manufacturer will have installed this op. But you won't be as enthusiastic about their amp. Or even more so;-?
 
  • Like
Reactions: 1 user
Member
Joined 2009
Paid Member
Yes but sincerely more than the price i wonder if the decision to swith to opamps for their products has a tech basis
I guess that there must have been an evolution and the newer products are better It is a two chassis with separate power supply
Many manufacturers offer this solution for their top of the line products It is less convenient but interesting
The main offender could be the mains transformer with its vibrations and EM emission (i guess that can be even measured ?)
I like when it is separated from the circuits
 
  • Like
Reactions: 1 user