A special current mirror

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
As a part of an amp design (in CFA topology), I've done quite a lot of SPICE tinkering to find the 'perfect' current mirror.

Requirements List:

a) Will operate at Iq(in) = 10mA

a) I(out) = 5 * I(in)
(or a factor nearby, which should have good reproduceability and low temperatur drift)

b) R(out) > 1 MOhm

c) sane transient behaviour

d) low THD

e) doesn't eat too much voltage from the rail

My current favorite is:
http://www.linearaudio.de/scratch/new-cm-darl-casc-2.pdf

This works well for all points a)-d), e.g. the output impedance is 6 MOhm.

But it consumes 2.5V of the available voltage range, which is a pity. I can work around this, but a genuine solution would be most welcome.

Of course the transistor count is quite high, but this is only an esthetic problem, the total silicon costs will be less than 1 Euro (or USD).

Suggestions?

Regards,
Peter Jacobi
 
Hi Peter,

I have an alternative as suggestion. I didn't simulate yours, so I don't know how these two compare.

This alternative has an accurate 5x ratio between input and output, with a gain error of order 1/hFE^2. All base currents are compensated for.
The circuit is simulated with an input current of 50-150mA sine wave. The output has a 250-750mA sine wave current, which translates into 2.5-7.5V at the output due to 10 Ohm loading. The THD is very low, all harmonics lower than -132dB. The transistors in this simulation have a hFE of 100.
I did not use emitter resistors as the transistors are completely equal in the simulation. In practice it will be a good idea to add small resistors for the bottom line transistors to reduce the effect of Vbe mismatch.
Minimum voltage at the output is approx. 2V above the ground.

Steven
 

Attachments

  • mirror5x.gif
    mirror5x.gif
    15.7 KB · Views: 1,592
Hi Steven,

Thank you for looking into this.

I simulated your topology using
the same BJT choices (BD139/BC547c) using the same Fairchild models.

This made the results somewhat mixed:

Not having identical Qs puts the balance off, so I needed 49.963mA
to balance the 10mA input (that's
49.991mA for the other circuit).

You indirectly pointed out, that emitter resistors were uncessarily large and making them just large enough to balance the devices would save 500mV voltage swing.

This worked fine for my original circuit too, with some some reduction in output impedance (from 6 to 1MOhm, which is still fine) and a
welcome reduction in K3.

Your circuit has only half the output impedance, about 0.5MOhm.

Doing the distortion analysis driving
a high impdecance load (I should have clarified, that this is the intention), your circuit has visible K2 (which may be no problem) and some reduction in higher harmonics:

Your circuit, driving 8Vpp into 1MOhm:

Harmonic Frequency Normalized
Number [Hz] Component
1 1.000e+03 1.000e+00
2 2.000e+03 2.280e-02
3 3.000e+03 2.486e-04
4 4.000e+03 1.098e-05
5 5.000e+03 6.564e-06
Total Harmonic Distortion: 2.280050%

The original circuit, driving 8Vpp into 1MOhm:
Harmonic Frequency Normalized
Number [Hz] Component
1 1.000e+03 1.000e+00
2 2.000e+03 4.730e-03
3 3.000e+03 3.106e-04
4 4.000e+03 4.381e-05
5 5.000e+03 4.051e-05
Total Harmonic Distortion: 0.474181%

(These are the numbers with 50Ohm emitter resistors, they remain consistetnt but reduce somewhat for smaller Rs).

Thanks again for your suggestions!

Peter Jacobi
 
Hi Peter,

Your observations are right; I was too quick posting the results befor properly analyzing.
I used default NPN transistors in the simulation and liked the results. I should have been more suspicious. Warned by your results I looked to the models and it appeared that the default NPN transistor in the simulator does not have the Early effect modelled, just as some other non-linearities. So I changed that into more real transistors, the 2N4401 (I only have a limited library). For the output I used the 2N1893, that has a factor 4 lower hFE. Now it became clear that the current ratio was not exact 5 anymore due to the Early effect of the lower rail transistors. This has been improved by adding Q9 and Q10. Now the Vce of all current determining transistors is almost equal, and the current ratio is quite accurately 5x. The fact that the 2N1893 has a lower hFE does not have much effect on that, since its base current is added to the output again via Q8. The circuit is quite insensitive to hFE changes of the output transistor. And because all transistors carry a reasonable current, the high frequency behaviour is quite good. Actually all transistors carry the same current, except for the output transistor. Above 1MHz distortion increases. My simulation showed that the output resistance is approx 450kOhm, not extremely high, but workable. It lacks a cascode. Adding emitter resistors to the lower rail transistors makes no significant difference to the output resistance.
I found THD (1kHz) to be around 0.04%, mainly second harmonic, higher order very low. At 1MHz THD is 1%.


Steven
 

Attachments

  • current5xa.gif
    current5xa.gif
    4.6 KB · Views: 1,302
further possible improvement is available if the super-pair works as well in the real world as it does in sim, the voltage burden is about 3 diode drops like steven's circuit but D1 isn't strictly necessary so the circuit should still behave down to 1.5 V with D1 shorted

> 10 MOhm Zout to ~30 KHz, 2nd -96 dB @ 10 KHz 50 Vpp, 3rd ~-120dB

the mje is from power amp VAS sims in the symmetrical amp thread, rework the compensation network for different pass Q by watching step response from I1 to Vout

(oops, the AC gain is 6x, i just balanced the 10 mA dc bias of Q2 ( from R12, V2) against an extra current mirror cell - just delete one for 5x ac gain, if R12 or other current source biases both top and bottom symetrical current mirrors its current is 1st order canceled out)
 

Attachments

  • 5x_mirror.png
    5x_mirror.png
    40.5 KB · Views: 1,376
a bit of topic but still Current Sources

Nice Current source designs (Y) !

My question relates to matching an NPN and PNP based current source !
I need an diff-stage (PNP's) with in the top a PNP based current source and in the bottom an NPN based current mirror.
Since i need both legs of the diff-stage , putting one BJT as diode is no option.
So i need a 3 or 4 transistor current mirror , were iref is copied twice and iref is matched to the current of the (upper) current source !

any one ?

grtz

Simon
 
SPICE models used

FYI, I used the BC547C model from Philips, included in:
http://www.semiconductors.philips.com/models/spicespar/data/zip/SST.zip

The BD139 model is from Fairchild, available at:
http://www.fairchildsemi.com/models/PSPICE/Discrete/Bipolar_Transistor.html

I consider them to be fairly realistic within the limits of the BJT model used.

To jcx: if you are using the default 2N3904 model of LTSpice, you should be aware that it is idealized and doesn't look like the result of full parameter extraction. So you may get overly optimistic results from simulation.
(You can get a better 2N3904 model from Fairchild)

Regards,
Peter Jacobi
 
wide-swing cascode current mirror

Steven + jcx:
Thanx for your (updated) suggestions. I'll model them with matching devices and try to make a comparison table.

To try a different approach and get a significant improvement in voltage swing, I made a BJT version of the wide-swing cascode current mirror, popular in CMOS VLSI.

I tried this before and failed, but now I have a sort-of working version:

http://www.linearaudio.de/scratch/new-cm-wideswing-2.pdf

This topology gets the output voltage within 1.5 times Vbe of the rails, i.e. only about 1V is lost!

Output current matching is fine impedance adequate (300k), but problems remain:

- There is a pole near 10MHz

- Noticeable K2 (3%)

- K3 and up somewhat higher than in the other designs considered.

- not forgetting: component count

Regards,
Peter Jacobi
 
For convenience and easy reference, I changed Peter's pdf in a gif, shown below.

Interesting stuff, these current mirrors. A very basic function but a lot of opportunities to do things right or wrong. In the end the influence of the mirror on the total performance of a circuit may be marginal, still it is fun trying to make a better one.

Steven ;)
 

Attachments

  • current5xb.gif
    current5xb.gif
    18.4 KB · Views: 1,126
Re: SPICE models used

pjacobi said:

(You can get a better 2N3904 model from Fairchild)

i agree that the Phillips' model in the current SwCad has suspiciously even values for many parameters but they are not uniformly in the "better" direction compared to the Fairchild

i have mixed and matched several model sources and the compound Baxandall "super-pair" current mirror i presented earlier seems to hold up in offering vastly superior performance - with D1 shorted i even get distortion components @~ -90dB with 20 Vpp swinging to within 1 V of the negative rail where Q1 is nearly saturated @ ~ 200mV Vce! (Q1 = 2N2219A, PN2222 or BD139)
 
Re: Re: SPICE models used

jcx said:


i agree that the Phillips' model in the current SwCad has suspiciously even values for many parameters but they are not uniformly in the "better" direction compared to the Fairchild

I don't think it matters much, with the circuits we are just looking at, but unrealistic uniformity (e.g. equal VAF for 2N3904 and 2N3906) will hide some distortion sources in more complex circuits.

jcx said:


i have mixed and matched several model sources and the compound Baxandall "super-pair" current mirror i presented earlier seems to hold up in offering vastly superior performance - with D1 shorted i even get distortion components @~ -90dB with 20 Vpp swinging to within 1 V of the negative rail where Q1 is nearly saturated @ ~ 200mV Vce! (Q1 = 2N2219A, PN2222 or BD139)

Impressive results!
Would you be so kind to post or email the SwCad .asc file?

For which load conditions are your distortion figures? My figures are usually for driving 1M or 300k.

And another question: As you have to counterbalance currents with Rs (if I understand right), wouldn't the circuit be sensisitive to temp changes?

Regards,
Peter Jacobi
 
a bit slow with the response here...

the distortion performance of the current mirror with the baxandall super pair cascode is somwhat worse with the 1M load but still distortion components are > 50 dB down within 2 V of the rail @ 10 KHz

thermal stability is not as straightforward in discrete designs due to poor thermal coupling between devices and larger device parameter spread vs monolithic designs - the flip side is that thermal coupling being lower, some types of gain modulation are reduced

in the super-pair cascode output Q1's power dissapation modulation of Hfe is effectively reduced by the current gain of Q2, and if the rest of the Qs see nearly constant and similar voltages i would expect the super-pair cascode to to offer superior "thermal distortion" performance

(rename the file with extension "asc", or just open in SwCad and save as *.asc)
 

Attachments

  • 5x_mirror.txt
    4.7 KB · Views: 129
jcx said:
further possible improvement is available if the super-pair works as well in the real world as it does in sim, the voltage burden is about 3 diode drops like steven's circuit but D1 isn't strictly necessary so the circuit should still behave down to 1.5 V with D1 shorted

> 10 MOhm Zout to ~30 KHz, 2nd -96 dB @ 10 KHz 50 Vpp, 3rd ~-120dB

the mje is from power amp VAS sims in the symmetrical amp thread, rework the compensation network for different pass Q by watching step response from I1 to Vout

(oops, the AC gain is 6x, i just balanced the 10 mA dc bias of Q2 ( from R12, V2) against an extra current mirror cell - just delete one for 5x ac gain, if R12 or other current source biases both top and bottom symetrical current mirrors its current is 1st order canceled out)


jcx :)
Would you still, today use same basic approach ?
to make one: 10 mA in ... 6x10mA out ... precise Current Mirror

=================================

Here, Schematic of The jcx 2004 mirror:
http://www.diyaudio.com/forums/attachment.php?s=&postid=302398&stamp=1074056311

=================================

Would be interesting to compare with what you would do, this year of 2008.

Mirrors that can Multiply the input current
can be very useful for me,
in my no global FB and my low global FB amplifiers.


Thanks & Audio regars
Lineup
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.