Discrete Opamp Open Design

bobaruni, please don't be discouraged by Guru Wurcer's comments. He's really much nicer than his avatar. :)

I for one would be very interested in an LTSpice file of your upside down SK99b as it touches upon some Jurassic experience of mine on low feedback amps.

I'm not hurt by his comments as I did not design the SK99B or the Jensen Op Amp that it's based on, if there is something better (as I'm sure there is), I definitely want to learn how it works and build it :)

Anyhow, here is an attached LTspice ASC file using only the default models.

The THD is extremely low, not sure if this sim would truly reflect what is possible in real life but will find out next week when the parts arrive. Then there's the problem of how to test it!

1Khz at 1V, gain = 1, inverting mode.

Harmonic Frequency Fourier Normalized Phase Normalized
Number [Hz] Component Component [degree] Phase [deg]
1 1.000e+03 1.000e+00 1.000e+00 180.00° 0.00°
2 2.000e+03 1.569e-08 1.570e-08 88.88° -91.12°
3 3.000e+03 2.316e-08 2.316e-08 97.31° -82.69°
4 4.000e+03 2.150e-08 2.150e-08 100.50° -79.50°
5 5.000e+03 2.205e-08 2.205e-08 103.50° -76.50°
6 6.000e+03 2.596e-08 2.596e-08 105.22° -74.78°
7 7.000e+03 2.472e-08 2.472e-08 110.66° -69.34°
8 8.000e+03 2.414e-08 2.414e-08 108.93° -71.07°
9 9.000e+03 2.657e-08 2.657e-08 110.71° -69.29°
Total Harmonic Distortion: 0.000007%

Bob.
 

Attachments

  • fft.png
    fft.png
    36.4 KB · Views: 458
  • SK99B compliment.ASC
    6.9 KB · Views: 67
  • schem.png
    schem.png
    87.7 KB · Views: 452
I was planning on setting the gain to 100x or more then using a 24Bit 192k Soundcard to sample and perform an FFT, but I'm more worried about generating such a pure sine wave to inject into the thing.
Also, as I'm new to the world of LTspice, I'm not sure if the way I'm obtaining THD information is correct?
 
Also, as I'm new to the world of LTspice, I'm not sure if the way I'm obtaining THD information is correct?
If you're getting numbers out of LTspice, and measuring the right node then there shouldn't be a problem. The main thing to think about is if the distortion figures seem to be too poor, which typically means that compression has been applied to the measurement results. The latter is set in the Control Panel or via the plotwinsize directive, in case you're not aware of this.

The reason for compression is that LTspice can create massively sized results files on the hard drive very easily, and this tames that capability if it makes no sense to record a high level of accuracy.

Frank
 
Frank, we're in luck. Guru Wurcer likes us Ozzies. You can be Russell C and I'll be Mel G.
As in, 30 Odd Feet of Grunts ... :D

Frank, how did you get on with your improved BF862 model?

Presently, I'm moving my *.asc to a970/2240, BF862 & J305 but its slow for this LTspice newbie.
Sorry, I'm a bit behind the 8 ball at the moment, will try and get that happening ASAP ... :)

Frank
 
If you're getting numbers out of LTspice, and measuring the right node then there shouldn't be a problem. The main thing to think about is if the distortion figures seem to be too poor, which typically means that compression has been applied to the measurement results. The latter is set in the Control Panel or via the plotwinsize directive, in case you're not aware of this.

Frank

Ok, I have turned off compression, and now all the high frequency crud in the FFT has gone (I was worried this was some sort of oscillation) and the THD in the log is even lower. Thanks for your help Frank :)
 
Disabled Account
Joined 2012
general comment -- due to mismatched components in the input stage (in practice - including the driving source Z), a small amount of CM signal usually feeds thru to the output. Thus, my suggestion of >-100dB rejection.

So what else contributes to input stage cmr? In traditional (?) diff input, increasing Zo of the diff pair's current source helps (often via cascoded current sources and other such Zo increase methods). Here...? Thx-RNM

Scott -- what would you suggest could be done to get some more cmr out of the SW-OPA ? Shooting for 10-20 db more. Could you make a cmr test with cascoded current sources or what ever you suggest. Thx-RNM -somwhere in Bangkok.
 
Twitchy SW-OPA?

I'm at the point where I'd like to compare my poor efforts at a better FET-990 with Guru Wurcer's SW-OPA. To do this, I'm trying to use a common set of SPICE models .. to whit BF862, J111 & J305 from Scott and also 2sc2240/2sa970 or 4401/3 from Scott, Cordell, fas42 or Staccatiss.

But I've come across a feature that fas42 encountered earlier. SW-OPA seems inordinately sensitive to the transistor models used and also their Iq.

The offset voltage changes up to more than 4V with different BJTs, FETs and small changes in Iq for input & output stages.

I'm using R3 to adjust the DC offset but this is not only overly sensitive but requires huge changes in value to accomodate different BJTs. R3 goes from 37R with fas42's 4401/3s to 20R with Scott's 2240/970s.

I gather, one of the hallmarks of this design is that the 125R resistors should have exactly 2x the current of the 250Rs but this doesn't seem to be the case .. some 5-8% imbalance from this ratio which sorta correlates with the imbalance in the input device currents ...

Though this behaviour is with a gain of 100x as shown in the attached GIF, the FET990 circuit is far more tolerant and output offset changes much less with different transistors & Iq.

Am I missing something or does this mean we have to match complements?

I enclose my *.asc & also the models used which you can see are taken from various gurus.

I used to work for J. Edgar Hoover's nephew, I'll let your imagination run with that one. For you Ozzies, J. E. H. violated rule number 3.
Scott, your evil avatar is taking over :eek:
 

Attachments

  • #1568.zip
    9.9 KB · Views: 72
  • SW-OPA#1568.gif
    SW-OPA#1568.gif
    49.5 KB · Views: 399
Disabled Account
Joined 2004
I'm at the point where I'd like to compare my poor efforts at a better FET-990 with Guru Wurcer's SW-OPA. To do this, I'm trying to use a common set of SPICE models .. to whit BF862, J111 & J305 from Scott and also 2sc2240/2sa970 or 4401/3 from Scott, Cordell, fas42 or Staccatiss.

Scott, your evil avatar is taking over :eek:

I'll help you out this weekend, and get you going (I have not had these problems). I found an old laptop to put LTSPICE on so I won't irritate our lawyers. The OS resistors depend on absolute Vbe a little I'll show you how to compute a close value (you only need the diode function on your handheld VOM).

BTW took two strips of 100 4401/4403 from ebay and find most matched better than 1-2mV Vbe. :eek:

Hopefully Lord Shiva will look favrably on our efforts.
 
Last edited:
I'll help you out this weekend, and get you going (I have not had these problems). I found an old laptop to put LTSPICE on so I won't irritate our lawyers. The OS resistors depend on absolute Vbe a little I'll show you how to compute a close value (you only need the diode function on your handheld VOM).

BTW took two strips of 100 4401/4403 from ebay and find most matched better than 1-2mV Vbe. :eek:

Hopefully Lord Shiva will look favrably on our efforts.
We grovel at your feet Oh Lord Shiva. Our brother Wurcer is restored to the Light :)

Was that between 4401s & 4403s or within a strip of 4401s, within a strip of 4403s? ie did the 4401s match the 4403s?

"OS resistors" Was is das? asks dis bum hus unnerstening of 3 or 2LA is wonky :eek:
 
I'll help you out this weekend, and get you going (I have not had these problems). I found an old laptop to put LTSPICE on so I won't irritate our lawyers. The OS resistors depend on absolute Vbe a little I'll show you how to compute a close value (you only need the diode function on your handheld VOM).
I uploaded ricardo.txt cos the transistor models should give the same results in other SPICES unless LTspice is doing something other SPICEs aren't.

But what you say implies that using R3, one of the BF862 ltp source degenerating resistors, to adjust offset is suboptimal and there is a better tweak point.

On the FET990, I use the input current mirror resistors which is similar to how its done on many IC OPAs.
 
Last edited:
Disabled Account
Joined 2004
Could you please point out that particular ebay seller, thx!

Unfortunately, I bought them 2 or 3 yr ago and bagged them up. I meant the matching of 4401 to 4401 etc. Obviously buying strips probably gets devices off of the same wafer near eachother.


EDIT - I don't see any on eBay right now from the same manu, but there are some pretty cheap and of course the pictures don't match the items. One guy picturing strips of 2N2907's. :(
 
Last edited:
Disabled Account
Joined 2004
We grovel at your feet Oh Lord Shiva. Our brother Wurcer is restored to the Light :)

Was that between 4401s & 4403s or within a strip of 4401s, within a strip of 4403s? ie did the 4401s match the 4403s?

"OS resistors" Was is das? asks dis bum hus unnerstening of 3 or 2LA is wonky :eek:

The 20 Ohm ones, they simply depend on the vbe differences, assuming reasonable thermal connection. This is not hard to do. I have lots of those little TO-92 heatsinks left from the group buy, but just putting the right devices "face to face" and wraping them with a piece of copper tape should be more than enough.
 
Unfortunately, I bought them 2 or 3 yr ago and bagged them up. I meant the matching of 4401 to 4401 etc. Obviously buying strips probably gets devices off of the same wafer near eachother.


EDIT - I don't see any on eBay right now from the same manu, but there are some pretty cheap and of course the pictures don't match the items. One guy picturing strips of 2N2907's. :(

Could you name a manu of yours devices?
 
The 20 Ohm ones, they simply depend on the vbe differences, assuming reasonable thermal connection.
BINGO! The 20Rs on the Output Stage biasing. :)

Do we tweak the 20Rs so that with 'no' load (incl. F/B bits so no Iq is lost), there is no voltage between the bases of the 'bias' devices and the output?

But that means another tweak pot for i/p stage offset. :mad: Got a recommendation for which one to use?

[edit]Just realised the Output Iq isn't just what's going through the 4R7s but a substantial amount goes through the 'bias' transistors & the 20Rs too. When you say "Adj for 8mA", is that just through the 4R7s or the total through the outputs? [/edit]

I have lots of those little TO-92 heatsinks left from the group buy, but just putting the right devices "face to face" and wraping them with a piece of copper tape should be more than enough.
.. from another project, I recommend a drop of cyano-acrylate in addition to the Cu strap. Substantial improvement in thermal matching.
 
Last edited: