开云体育

I can't remember when the UniversalOpamps first came into being, but there's a good chance there's been a bug in them from the start.

Although I realised there was a problem long ago when I modified my copy of the old combined library file of UniversalOpamp2 to fix it, I had since forgotten about it. I failed to flag it up.

However, when dealing with a query on another channel, I came across a significant shortfall in expected performance when using the latest stand-alone UniversalOpamp models. I was attempting to emulate the TI OPA2211 opamp with one. The datasheet for the OPA2211 states that the differential input resistance is typically 20kΩ, and the common mode input impedance is 2MΩ. Usually, the CM impedance makes little difference to the performance of most circuits because it is typically very high, so I was surprised when putting Rin=20k that the amplifier lost about 5dB of expected closed loop gain when the source impedance was significant, by comparison.

I set up an impedance testjig for the UniversalOpamp3a that measured the input impedance in both modes.
TL;DR: neither the differential or common mode input impedance was 20kΩ!

The differential input resistance was 40kΩ, while the common mode input resistance was 10kΩ. WRONG ANSWERS! I've uploaded the test schematic: UniversalOpamp_bug1.zip

The reason this happens is obvious when you look at the .subckt for UniversalOpamp3a:?

.subckt level3a 1 2 3 4 5
S1 5 3 N002 5 Q
S2 4 5 5 N002 Q
A1 2 1 0 0 0 0 N004 0 OTA G={Avol/Rout} ref={Vos} Iout={slew*Cout} Cout={Cout} en={en} enk={enk} in={in} ink={ink} Vhigh=1e308 Vlow=-1e308
C3 5 4 1p
C4 3 5 1p
R3 3 2 {2*Rin} noiseless
G1 0 N002 N004 0 {1/Rout}
R2 N002 0 {Rout} noiseless
C1 N002 0 {X*Cout/Avol}
R4 3 1 {2*Rin} noiseless
R5 1 4 {2*Rin} noiseless
R6 2 4 {2*Rin} noiseless
G2 0 N004 4 N004 table(0 0 10 {2*slew*Cout})
G3 N004 0 N004 3 table(0 0 10 {2*slew*Cout})
R9 3 N004 {2*Rout} noiseless
R10 N004 4 {2*Rout} noiseless
.param Rout=100Meg
.param Cout={Avol/GBW/2/pi/Rout}
.model Q SW(Ron=10 Roff=10Meg Vt=0 Vh=-.1 Vser={Rail} ilimit={Ilimit} noiseless)
.param X table(phimargin,29.4,3.5,32.1,2.9,33.8,2.6,35.8,2.3,38.1,2,40.9,1.7,43.2,1.5,45.9,1.3,49.2,1.1,53.2,0.9,58.2,0.7,64.7,0.5,73,0.3,86.1,0.05)
.param Avol=1Meg GBW=10Meg Slew=10Meg rail=0 Vos=0 ilimit=25m
.param en=0 enk=0 in=0 ink=0 phimargin=45 Rin=500Meg
.ends level3a

Resistors R3-R6 implement the supposed input resistance. Essentially, both opamp inputs have resistors of 2*Rin connected to both supply rails, and by implication (with ideal voltage sources) - to ground. Therefore each input has a resistance of Rin to ground. But - and this is a big but:

Rincm = (2*Rin//2*Rin)//(2*Rin//2*Rin) = 10kΩ
Rindm = (2*Rin//2*Rin)+(2*Rin//2*Rin) = 40kΩ

Since the default for the Rin parameter is 500kΩ, the issues this typically causes are small compared to other parameters with typical feedback network impedances. But with the Rin parameter set to match the desire differential input impedance commonly found on datasheets of (bipolar) low noise opamps, things can go awry with those same feedback components.

This should be fixed. For practical components, the common mode I/P impedance is always higher than the differential I/P impedance. So, I'd say this was a bug.