Let's Talk about Current Sources

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

analag

Well-known member
Joined
Apr 23, 2005
Messages
1,942
Location
Mars
ism-fig6a.gif
Gentlemen tell me which is better.
 
Nice topic !

A lot of respects in which one or the other could be better than the other.

[edited to better indicate lame joke]Left one can be better (as in smaller) since a diode smaller than a TO92.[/edited to better indicate lame joke]

More seriously (and assuming discretes, uncorrelated devices and assuming high hFE), the right one might be
fluctuating less because current is VbeQ3 / R3 where for the left one the relation (tracking) between D2 & VbeQ1 enters as well.
 
The right one - unless you want to realize multiple sources with one reference element only (but then you'd use a blue led instead of two diodes).
 
The CMOS(ex IC designer) guy here at work swears that diode junctions are quieter than BJT junctions but not by any amount that you would see with packaged parts. 

I think Analag was specifically talking about the function of the circuit, probably not the parts count or anything like that.  the difference in size between a couple diodes and a BJT in size is negligible.
 
In terms of current control one is significantly better than the other...the imposition of the so called infinite resistance is better realized.
 
As I recall, in Walter Jung's recent article on current sources, he found the two-diode source better than the BJT source, but other configurations (including LEDs, zeners, cascoded sources, etc.) were a good deal better than either.

Peace,
Paul
 
I'd pick the one on the left myself.   Maybe I'd also use a current source instead of the 10K that's feeding the diodes - set for about 4 - 5 mA?  An LED might be better yet than diodes?   A cascode on Q1 to raise the impedance even more?   I'm getting more complicated and losing the plot...
 
I hope it is fine that I took the freedom to edit the title for improved spelling. It's such an important topic it might be worth to be found by the search engine.

In terms of current control one is significantly better than the other... The imposition of the so called infinite resistance is better realized.

As has been mentioned output impedance of the right one is better. There are three basic limitations regarding output impedance for the left one:

* Early effect modulates the hFE of Q1, causing variable base current. As the emitter current remains fixed the collector current must change, resulting in finite output resistance. This can be more intitively seen as an equivalent resistor from collector to base.

* Collector-base capacity causes a similar loss, with increasing importance at high frequencies. There is also considerable nonlinearity in this mechanism (much more than for the base current loss), possibly resulting in increased HF distortion at critical nodes.

* Early effect modulates the Vbe of Q1. As its base potential is ~fixed by D1 and D2 the emitter and hence collector current must vary. Again a contribution to finite output resistance.

The right implementation can reduce the last mentioned effect with feedback from Q3. This does not however have any influence on the first two mechanisms, and hence the resulting output impedance is still relatively low. Emitter referenced cascodes are way better (if I find time I'll post a few implementations).

As I recall, in Walter Jung's recent article on current sources, he found the two-diode source better than the BJT source.

Note that Walt mainly investigated PSRR, which is related to but not the same as output impedance.

Samuel
 
Svart said:
I think Analag was specifically talking about the function of the circuit, probably not the parts count or anything like that.  the difference in size between a couple diodes and a BJT in size is negligible.

I've edited my post above in an attempt to make the silly non-contribution even more clear.
Note that the words 'More seriously' were originally already there.
I'm not sure you're pulling my leg here or not  ;)

Bye,

  Peter
 
Samuel Groner said:
There are three basic limitations regarding output impedance for the left one:

* Early effect modulates the hFE of Q1, causing variable base current.

Hi Samuel,

Got any illustrating figures at hand ? The caused changes in Zout for instance ?
Mechanisms are what gives insight, but in addition it's interesting to get the feeling one should spend that additional BJT or not.

Like: hFE for instance may be pretty high, so a variation in base current is still a variation in base current, but not big enough to influence things noticably.

It's late here, the coin didn't drop here anymore why Early effect modulates the hFE of Q1, but not so for Q2.

Bye,

  Peter
 
It's late here, the coin didn't drop here anymore why Early effect modulates the hFE of Q1, but not so for Q2.

I don't quite follow why you think hFE Early effect should not apply to Q2? The local feedback does just address the Vbe, but not the hFE modulation.

Mechanisms are what gives insight, but in addition it's interesting to get the feeling one should spend that additional BJT or not.

Every implementation is different, but as shown I'd expect an order of magnitude improvement for the right one. At ~100 kHz the output impedance will be roughly the same due to capacitive effects.

Samuel
 
clintrubber said:
Samuel Groner said:
There are three basic limitations regarding output impedance for the left one:

* Early effect modulates the hFE of Q1, causing variable base current.

Samuel Groner said:
It's late here, the coin didn't drop here anymore why Early effect modulates the hFE of Q1, but not so for Q2.

I don't quite follow why you think hFE Early effect should not apply to Q2? The local feedback does just address the Vbe, but not the hFE modulation.

Hi Samuel,

I interpreted your preference for the right circuit & mentioning the issue with Early/hFE/Q1 as that it didn't seem to pop up for Q2, hence my non-understanding.
OK, everything seems to fit again  ;)

  Bye,

  Peter

 
pstamler said:
As I recall, in Walter Jung's recent article on current sources, he found the two-diode source better than the BJT source, but other configurations (including LEDs, zeners, cascoded sources, etc.) were a good deal better than either.

Peace,
Paul
I second this. It's a very interesting thing to read. Here are the links to the article:
Sources (part 1)
Sources (part 2)

My current favorite (pun intended) is using an LM336Z as the voltage reference.
 
I like the simplicity of using one of the FE current source diodes, since they come in many handy values. Here is an example of a follower I have used many times:
 

Attachments

  • TMP46.JPG
    TMP46.JPG
    155.2 KB · Views: 284
Hijack: that's an interesting buffer -maybe it could be presented as a "ruby" buffer? (short for "rubirio" 8) )

.....but IMHO it would be even better if:
-Q3 Q4 collectors would be connected to Q6 emitter
-Q1 Q2 collectros would be connected to Q8 emitter
-Q2 Q6 Q4 Q8 would have B-E resistors. (now.. what exact values??..)

yay or nay?
 
Randy Slone (whom I contacted at some point) touches on this subject briefly but neatly in his book
"High Power Audio Amplifier Construction Manual" pages 65 to 68. He demonstrates that the one on the right has six times better current regulation than the left one with the forward biased diode as the regulating element.
 
tv said:
Hijack: that's an interesting buffer -maybe it could be presented as a "ruby" buffer? (short for "rubirio" 8) )

.....but IMHO it would be even better if:
-Q3 Q4 collectors would be connected to Q6 emitter
-Q1 Q2 collectros would be connected to Q8 emitter
-Q2 Q6 Q4 Q8 would have B-E resistors. (now.. what exact values??..)

yay or nay?

Might help a little by increasing Vce's.  This bootstrapped collector approach is often unstable btw, although if oscillations can be controlled has very low input capacitance midband.

A resistor from base to base of the output devices wouldn't hurt.  Size it to provide some turnoff current without raising Q5/Q7 dissipation unduly---maybe 750 ohms?

One thing that bothers me a bit is the very low and ill-determined currents in Q1 and Q3.  If they were much smaller geometry devices this would be o.k. but as it is, with no more than the base currents of Q2 and Q4 they are going to be operating way down on the gain-bandwidth curve.  Use of integrated darlingtons like the MPSA13/MPSA62 etc. would work, since the internal equivalents of Q1/Q3 are tiny.

The description in the caption is also a bit misleading.  I'd venture the tempco of the current regulator diodes is a relatively minor influence on bias stability.  The matching of the junction potentials is key.

Another note:  surely "0.5K" is a misprint, particularly if the circuit is going to drive the advertized 5 ohm load.

I remember the author's name from a number of circuits published in Ideas for Design around this time.  They were generally good stuff.
 
A resistor from base to base of the output devices wouldn't hurt.  Size it to provide some turnoff current without raising Q5/Q7 dissipation unduly---maybe 750 ohms?
what about adding a parallel "D.Self 1uF cap" for "even better better"?

One thing that bothers me a bit is the very low and ill-determined currents in Q1 and Q3.  If they were much smaller geometry devices this would be o.k. but as it is, with no more than the base currents of Q2 and Q4 they are going to be operating way down on the gain-bandwidth curve.  Use of integrated darlingtons like the MPSA13/MPSA62 etc. would work, since the internal equivalents of Q1/Q3 are tiny.
adding Q2 Q4 B-E resistors would do that, no? Providing sort-of-fake CCS's for Q1/Q3 loading...
 
Back
Top