Spice simulation

1audio said:


John:
I'll give you a crash course (which probably will be all I know) to get you up the next time I'm nearby. Assuming you have the computer.

Getting models and making them work is the bigger pain. We all need sources for models for Jfets, MosFets and who knows what else that aren't in the default LTspice libraries. And a quick writeup on how to include them (I have to search the web every time I do that).

I personally don't think Spice is the way to -130 dB distortion products. Its a start but the devices themselves vary too much for the models to have that much accuracy. For the incredibly low distortion numbers to be real it needs to be built and verified. Its great for Monte Carlo analysis to see if you can make more than one of something. Its also great to test an off the wall idea to see if it will work.
-Demian

Hello Demian

What simulator do you prefer to use.

Regards
Arthur
 
Member
Joined 2004
Paid Member
PHEONIX said:


Hello Demian

What simulator do you prefer to use.

Regards
Arthur

I learned how to use them with Pspice pre-graphic front end. Being lazy and cheap I use LTspice myself. I have figured out how to tweak settings to get results.

Why -130 dB distortion? The passive component non-linearities start surfacing there and technique/layout need to be dealt with before the real product will match the performance.
Demian
 
1audio said:
[snip]
Why -130 dB distortion? The passive component non-linearities start surfacing there and technique/layout need to be dealt with before the real product will match the performance.
Demian

The effect of nonlinear resistors can be avoided. Take the NFB divider network, for example. Just use (a lot of!) all the same resistors from the same batch. Perhaps overkill, but (theoretical) no distortion at all.

BTW, our PGP amp, using ordinary metal film resistors and only two for the NFB divider network, measures AND simulates a THD-20k below -120dB. Admittedly, not -130dB, but that will be our next target.:D
 
Member
Joined 2004
Paid Member
Your making an assumption about resistors- that they will all have identical distortion mechanisms to proportional levels. It may be true but the following could get in the way:
The construction and materials change for different values and a 10:1 ratio will have different thicknesses or different compounds (alloys) to create the values. And the heating effects will be different.

The stray effects on the PC board from the board itself and the nearby components could be greater than the resistors. PPM levels and below are very difficult to maintain and do so consistently, and they are not part of the models in Spice.

If you look at he construction of a precision Kelvin-Varley divider with sub PPM resolution they are more about technique than materials.

The simulations are a good start but not a substitute for the real world. As you have noticed getting from -120 db to -130 db isn't easy.
 
1audio said:
Your making an assumption about resistors- that they will all have identical distortion mechanisms to proportional levels. It may be true but the following could get in the way:
The construction and materials change for different values and a 10:1 ratio will have different thicknesses or different compounds (alloys) to create the values. And the heating effects will be different.
[snip]

NO! :sigh: I said: "using the same resistors from the same batch". That also means using resistors of the same value. So for a 10:1 divider you will need ten (10) resistors, I repeat, of the same value, and of course with the same nonlinear (if any) behavior. Then nil distortion from the resistors.
 
Mooly said:
Downloaded LTSpice/swd cadIII last weekend. Thinking if I don't do this now I never will.
BY THE CRINGE--- there has to be an easier way than this. Taken me a week to get an AC voltage over two series resistors.


Mooly,

Look in Program Files\LTC\SwCADIII\examples\Educational . There is a lot of useful information.

Also, the Files section in the Yahoo LTSpice group has even more useful information for learning how to do stuff. Be sure to look for the frequently-updated All_Files.htm in Files > Tables of Contents and download it as a guide for looking for what you need.


Jeff
 
Member
Joined 2004
Paid Member
Edmond Stuart said:


NO! :sigh: I said: "using the same resistors from the same batch". That also means using resistors of the same value. So for a 10:1 divider you will need ten (10) resistors, I repeat, of the same value, and of course with the same nonlinear (if any) behavior. Then nil distortion from the resistors.

I follow and understand, however it won't work for just any resistor, it would depend on the distortion mechanism and how random it is. With good selected resistors it should work I think, but I'm not the mathematician here.
 
Edmond Stuart said:


NO! :sigh: I said: "using the same resistors from the same batch". That also means using resistors of the same value. So for a 10:1 divider you will need ten (10) resistors, I repeat, of the same value, and of course with the same nonlinear (if any) behavior. Then nil distortion from the resistors.

It would even out the thermal effects, but not voltage dependency, as far as I can see (assuming that voltage dependency is non-linear, which I frankly don't know if it is).

A clever way to tackle thermal distorsion, though, instead of just using a higher wattage.
 
1audio said:


I personally don't think Spice is the way to -130 dB distortion products. Its a start but the devices themselves vary too much for the models to have that much accuracy. For the incredibly low distortion numbers to be real it needs to be built and verified. Its great for Monte Carlo analysis to see if you can make more than one of something. Its also great to test an off the wall idea to see if it will work.
-Demian

Demian,

WRT -130dB, very true.

However if you are resourceful you can, for example, change device
models of known parameters to observe which elements that part of
the circuit is sensitive to.

Changing BC547 to 547C, for example can isolate beta and it's
importance in a specific location noted.

You start to get an intuitive sense about where in circuits different
things are necessary, high beta, flat beta, high ft, low cob, low cob
modulation with voltage etc.

Using this technique, I spiced a 0 GFB I-V for Sabre DAC chip, which
went under Ltspice zero line at -166dB /20kHz / 0dB FS.

Of course -166dB is in dreamland WRT anything real and achievable.
However the major benefit was that I could then start simplifying
and trimming the circuit back whilst observing which components
and parameters had the least effect. Same goes for harmonic
structure.

In the end, I settled on simpler design, at lower current, with
fewer esoteric devices that still spices to around -140dB.

As such real world chances are bettered and if I achieve say 20dB
worse again, it's still a great result.

cheers,

Terry
 
Member
Joined 2004
Paid Member
Terry:
Sounds like a rational and reasoned use of Spice with a proper respect for its limitations. I have done similar efforts in the past since its much faster to tweak in the simulator. However the results are only as good as the models, and if the models are really not good it can still show where the sensitivity in the circuit lies. And you must understand layout well or you would not have come close to those results. I hope the finished circuit works as well as the simulation suggests.
-Demian
 
Edmond, forget what I said. I realized that with your approach you probably have all resistors in series, which splits the voltage equally. (In contrast to using a buch of higher valued resistors in parallel to get higher wattage to minimize thermal effects). On the other hand, what about the increased inductance in the feedback loop, if doing it your way?
 
Edmond Stuart said:


NO! :sigh: I said: "using the same resistors from the same batch". That also means using resistors of the same value. So for a 10:1 divider you will need ten (10) resistors, I repeat, of the same value, and of course with the same nonlinear (if any) behavior. Then nil distortion from the resistors.


Hello Edmond

I once saw the inside of a Halcro DM68( I think ) and they used a combination of series and parallel resistors to construct the feedback resistor , from output to inverting input of amplifier. What struck me about it was that they used quite a few resistors rodenstein MK8 metal film types . I thought the network was large to make up a low reistance value to meet the equivalent input noise figure of the amp of 5nV/sqrt(Hz).

The M70 Yamaha (0.0005% @ 20Khz) is covered in carbon resistors which surprised me, I think this shows that leaded reistors are pretty low in THD. On the other hand thick film smd resistors have measurable distortion and should be avoided for ultra low THD products.

Regards
Arthur
 
Christer said:
Edmond, forget what I said. I realized that with your approach you probably have all resistors in series, which splits the voltage equally. (In contrast to using a buch of higher valued resistors in parallel to get higher wattage to minimize thermal effects). On the other hand, what about the increased inductance in the feedback loop, if doing it your way?

Hi Christer,

Just as the voltages are equally divided over the R's, so the inductances are equally divided. No problem, it doesn't affect the division ratio.
But agreed, the total inductance is larger. However, in a practical design, you have to bridge the rather long distance between the output and the inverting input anyhow. Whether using a long copper trace or a lot of series resistor, in both cases we have to deal with some inductance.

Cheers,
Edmond.
 
Edmond Stuart said:


But agreed, the total inductance is larger. However, in a practical design, you have to bridge the rather long distance between the output and the inverting input anyhow. Whether using a long copper trace or a lot of series resistor, in both cases we have to deal with some inductance.

Yes, that was my point. I also realized what you say, but depending on layout, you might get a considerably longer path through the resistors than otherwise. On the other hand, I guess it should rather decrease the bandwidth and improve stability, than the opposite.
 
PHEONIX said:

The M70 Yamaha (0.0005% @ 20Khz) is covered in carbon resistors which surprised me, I think this shows that leaded reistors are pretty low in THD. On the other hand thick film smd resistors have measurable distortion and should be avoided for ultra low THD products.

Arthur,

My recollection is that carbon film resistors are noisier than metal film. Is this your experience? How does thin film compare to thick film for distortion?

Regards,

Jeff