Bob Cordell's Power amplifier book

Jump to Latest
#7,821
Bob Cordell said:
I would like some comments from anyone on the approach of determining VAF by the use of small-signal 1kHz ac Zout measurements at a nominal typical operating point.
Click to expand...
I thought about building a jig which did that. Grounded emitter amplifier loaded by triple cascode current source whose Zout >>> amplifier transistor's ro. Measure the gain, divide by gm, presto you get ro. Then solve for Early Voltage VA from the Ohm's Law equation
  • (VA + Vce) = ro * Ice
The problem is biasing: how do you set Vce accurately? (Ice is easy, it's just the triple cascode's output which you can measure with a DVM). In particular, how do you set Vce accurately without having the bias mechanism load down the ultra high impedance collector node?

If Ice = 1mA and Vce = 5V and VA=1000V then gain= 38400 and ro= 1 Megohm. To measure VA within 1% you need the bias circuit's input impedance to be > 100 Megohms. Yuck! At 1 kHz, the impedance of a 1.6 picofarad capacitor is 100 Megohms.

So I stuck with (low duty cycle pulsed) DC measurements. I'll let someone else be the hero who measures VA using AC techniques at 1 kHz.

(Oh by the way, measure the Early Voltage "VA" of the NPN transistor MJE340. You'll find its VA exceeds 1000 volts. Yes really. VA=1000 is not an unrealistic exaggeration.)
 
#7,822
Bob Cordell said:
Secondly, the Early effect in reality can be nonlinear and a function of the operating current of the transistor.
Click to expand...
Bob, I always thought Early effect was sorta 'linear' ... unlike the truly EVIL variation of Cob, often the biggest THD culprit for THD20k.

Is there a good accurate explanation of Early non-linearity on the web?

Gotta be unnerstanabel to sum 1 hu neber wen 2 skul 🙂
 
#7,823
Hawksford certainly showed the dynamic distortion importance: https://web.archive.org/web/2013012..._lab/malcolmspubdocs/J10 Enhanced cascode.pdf

even if the "base current canceling" circuits are much older

I used the name "super-pair" from a '60 circuits reference book and attributed it to Baxandall - turns out the refernces I found weren't complete - the circuit goes back even further

and we've discussed it, related issues several times here at diyAudio

http://www.diyaudio.com/forums/solid-state/25172-baxandall-super-pair.html
 
Last edited:
#7,825
Mark Johnson said:
I thought about building a jig which did that. Grounded emitter amplifier loaded by triple cascode current source whose Zout >>> amplifier transistor's ro. Measure the gain, divide by gm, presto you get ro. Then solve for Early Voltage VA from the Ohm's Law equation
  • (VA + Vce) = ro * Ice
The problem is biasing: how do you set Vce accurately? (Ice is easy, it's just the triple cascode's output which you can measure with a DVM). In particular, how do you set Vce accurately without having the bias mechanism load down the ultra high impedance collector node?

If Ice = 1mA and Vce = 5V and VA=1000V then gain= 38400 and ro= 1 Megohm. To measure VA within 1% you need the bias circuit's input impedance to be > 100 Megohms. Yuck! At 1 kHz, the impedance of a 1.6 picofarad capacitor is 100 Megohms.

So I stuck with (low duty cycle pulsed) DC measurements. I'll let someone else be the hero who measures VA using AC techniques at 1 kHz.

(Oh by the way, measure the Early Voltage "VA" of the NPN transistor MJE340. You'll find its VA exceeds 1000 volts. Yes really. VA=1000 is not an unrealistic exaggeration.)
Click to expand...

These are all good points. Observing the hi-Z node is the trickiest part, as you observe. Let's assume for the moment that we observe it with a JFET op amp connected as a unity-gain buffer. We'll assume that the Vce voltage is held within the available input range of the op amp. We then use the output of the op amp to feed a DC servo circuit that biases the base of the DUT. A control voltage applied to the servo integrator sets Vce. The servo must have very high attenuation at the test frequency so as not to cause any shunt feedback impedance-reducing effect.

The ac test signal is then applied to the hi-z node through a 1 Meg resistor. Ro at the hi-z node is then inferred from the amount of signal attenuation through the 1 Meg resistor. Attenuation of 0.5 would imply Ro = 1 Meg. In this approach the transistor is not connected as an amplifier whose gain is measured. It is really a current source doing a balancing act with the current source load.

A variant of the approach would be to hold the hi-z node at zero volts and set Vce by adjusting the value of the negative rail. In this arrangement, the negative rail must be very well bypassed to ground so that its impedance to ground at the test frequency is << re' of the DUT.

I'm guessing 10% accuracy is enough for the purposes here.

Cheers,
Bob
 
#7,826
kgrlee said:
Bob, I always thought Early effect was sorta 'linear' ... unlike the truly EVIL variation of Cob, often the biggest THD culprit for THD20k.

Is there a good accurate explanation of Early non-linearity on the web?

Gotta be unnerstanabel to sum 1 hu neber wen 2 skul 🙂
Click to expand...

I have not seen a good explanation of it either. The Early voltage can have a dependence (sometimes mild) on Vce and/or Ic. This can be influenced by things like doping profile and emitter crowding effects and dependence of beta on Ic. The biggest cause of finite Early voltage is thinning of the effective base region by depletion region expansion as Vcb increases. I am not a device person, of course, and others I'm sure can explain the effects much better than I can.

Cheers,
Bob
 
#7,827
Bob Cordell said:
I have not seen a good explanation of it either. The Early voltage can have a dependence (sometimes mild) on Vce and/or Ic. This can be influenced by things like doping profile and emitter crowding effects and dependence of beta on Ic. The biggest cause of finite Early voltage is thinning of the effective base region by depletion region expansion as Vcb increases. I am not a device person, of course, and others I'm sure can explain the effects much better than I can.

Cheers,
Bob
Click to expand...

Supreme and Pisces are available from Stanford they are very powerful device modelers, there's probably a huge learning curve.

Stanford University TCAD Tools

Myself, I'm a fan of ultra-symmetry subtract like from like in all possible cases, which on some circuits involved tied off unused transistors just to match the non-linear capacitances.
It is finished, I just got home from my last official day as a full time employee. I don't know about other's out there but I was sort of ambivalent at the end but I did make a point of turning out the lights something I have not done for decades.
 
Last edited:
#7,830
scott wurcer said:
Supreme and Pisces are available from Stanford they are very powerful device modelers, there's probably a huge learning curve.

Stanford University TCAD Tools

Myself, I'm a fan of ultra-symmetry subtract like from like in all possible cases, which on some circuits involved tied off unused transistors just to match the non-linear capacitances.
It is finished, I just got home from my last official day as a full time employee. I don't know about other's out there but I was sort of ambivalent at the end but I did make a point of turning out the lights something I have not done for decades.
Click to expand...

CONGRATULATIONS!!!

I'm sure you will keep busy in retirement, and that helps a lot with the ambivalence. Put the old day job in the rear-view mirror!

Cheers,
Bob
 
#7,831
Bob Cordell said:
I have not seen a good explanation of it either. The Early voltage can have a dependence (sometimes mild) on Vce and/or Ic. This can be influenced by things like doping profile and emitter crowding effects and dependence of beta on Ic. The biggest cause of finite Early voltage is thinning of the effective base region by depletion region expansion as Vcb increases. I am not a device person, of course, and others I'm sure can explain the effects much better than I can.

Cheers,
Bob
Click to expand...

Very few saw it likely because there is no trivial description of the phenomena involved with the nonlinear Early effect. Wambacq and Sansen has the whole derivation and explanation (below is only a fragment):

https://books.google.ca/books?id=Fb...#v=onepage&q="early effect" nonlinear&f=false
 
#7,834
Bob Cordell said:
Observing the hi-Z node is the trickiest part, ... assume for the moment that we observe it with a JFET op amp connected as a unity-gain buffer. ...a DC servo circuit that biases the base of the DUT.

The ac test signal is then applied to the hi-z node through a 1 Meg resistor. Ro at the hi-z node is then inferred from the amount of signal attenuation through the 1 Meg resistor. Attenuation of 0.5 would imply Ro = 1 Meg. In this approach the transistor is not connected as an amplifier whose gain is measured. It is really a current source doing a balancing act with the current source load.
Click to expand...

The PNP transistors I measured, had a 21-to-1 spread of Early voltage*: the lowest was 21X smaller than the highest. If you then include a 10-to-1 range of collector currents at which you plan to measure Early voltage, that becomes a total range of collector impedance of 210-to-1. You might need more than one test jig, each jig with a different value of injection resistor, for best signal-to-noise during measurement.

*preliminary datasampling suggests that NPNs will turn out to have a wider spread (maxVA / minVA) than PNPs.
 
#7,838
Pimm did elebroate ss ccs for tubez

the Hawksford paper I linked to above does OK eplaining if not giving prior art

there you should see that Cob isn't the limit with the cascode "done right"

jcx said:
Hawksford certainly showed the dynamic distortion importance: https://web.archive.org/web/2013012..._lab/malcolmspubdocs/J10 Enhanced cascode.pdf

even if the "base current canceling" circuits are much older

I used the name "super-pair" from a '60 circuits reference book and attributed it to Baxandall - turns out the refernces I found weren't complete - the circuit goes back even further

and we've discussed it, related issues several times here at diyAudio

http://www.diyaudio.com/forums/solid-state/25172-baxandall-super-pair.html
Click to expand...
 
#7,840
Waly said:
Very few saw it likely because there is no trivial description of the phenomena involved with the nonlinear Early effect. Wambacq and Sansen has the whole derivation and explanation (below is only a fragment):

https://books.google.ca/books?id=Fb...#v=onepage&q="early effect" nonlinear&f=false
Click to expand...
Thanks for this Waly.

I've developed a headache trying to unnderstan the excerpts but took away from p187
It is seen that the Early effect is in fact "quite linear"
Click to expand...
That's sorta my understanding of the subject ...

So my question to yus gurus is whether LTspice & the Cordell et al models sim whatever 'non-linearlity' remains of this.
 
You must log in or register to reply here.
Top Bottom