HP paper
With reference to John Curl:
http://www.diyaudio.com/forums/showthread.php?postid=244369#post244369
I just digged up the paper describing the 13-26mV voltage drop across the emitter resistors.
It is an article in Hewlett Packard Journal, february 1971, Volume 22, Number 6, page 11-16, written by Barney Oliver (actually Dr. Bernard M. Oliver), at that moment HP's vice president for research and development. In this article he describes the two principle types of distortion in class B amplifiers, difference in current gain of the two transistors on one hand and crossover distortion on the other. It is derived that the output resistance of the amplifier is most constant in the crossover region, when g.R =1 with R is the emitter resistor and g is the transconductance at the operating point (in case of voltage drive). This is easy to see. If only one of the transistors is conducting (class B region) the output resistance would be R because 1/g is very small (large current), either from the NPN or from the PNP. With no drive (class A region), the output resistance would be (1/g+R)//(1/g+R)=R for 1/g=R, or g.R=1. The output resistance of NPN and PNP are equal and in parallel. But the output resistance is not constant, there is a bump in between these regions. This bump dissapears for a smaller bias current; when g.R=1/2 the output resistance falls motonically from 2R to R, if driven harder. So, the optimum R is between 1/2g and 1/g. This can be translated to a voltage across the emitter resistor in the range of kT/q ... kT/2q, which is 26mV ... 13mV at room temperature.
Mr. Oliver points out that this is a very small value compared to the Vbe change of 250mV if the temperature changes from 0 to 100 degr C, making thermal tracking of the biasing diodes a potential problem. At that moment he suggests to make R >> 1/g for thermal stability and rely on negative feedback to reduce the resulting distortion. I think this is a pity because of the negative aspects of negative feedback related to the increasing amount of higher order harmonics in the distortion compared to the lower order harmonics. Anyway, to avoid wasting too much power in the (large) emitter resitors he suggests to bypass these with diodes. He did the same for his "The Barney Oliver Audio Amplifier", where the emitter resistors are 8.2 Ohm (!), bypassed by a diode.
Around the same time others have made similar calculations or simulations, like Blomley in Wireless World (also february 1971!
) and later Self.
Steven
With reference to John Curl:
http://www.diyaudio.com/forums/showthread.php?postid=244369#post244369
I just digged up the paper describing the 13-26mV voltage drop across the emitter resistors.
It is an article in Hewlett Packard Journal, february 1971, Volume 22, Number 6, page 11-16, written by Barney Oliver (actually Dr. Bernard M. Oliver), at that moment HP's vice president for research and development. In this article he describes the two principle types of distortion in class B amplifiers, difference in current gain of the two transistors on one hand and crossover distortion on the other. It is derived that the output resistance of the amplifier is most constant in the crossover region, when g.R =1 with R is the emitter resistor and g is the transconductance at the operating point (in case of voltage drive). This is easy to see. If only one of the transistors is conducting (class B region) the output resistance would be R because 1/g is very small (large current), either from the NPN or from the PNP. With no drive (class A region), the output resistance would be (1/g+R)//(1/g+R)=R for 1/g=R, or g.R=1. The output resistance of NPN and PNP are equal and in parallel. But the output resistance is not constant, there is a bump in between these regions. This bump dissapears for a smaller bias current; when g.R=1/2 the output resistance falls motonically from 2R to R, if driven harder. So, the optimum R is between 1/2g and 1/g. This can be translated to a voltage across the emitter resistor in the range of kT/q ... kT/2q, which is 26mV ... 13mV at room temperature. Mr. Oliver points out that this is a very small value compared to the Vbe change of 250mV if the temperature changes from 0 to 100 degr C, making thermal tracking of the biasing diodes a potential problem. At that moment he suggests to make R >> 1/g for thermal stability and rely on negative feedback to reduce the resulting distortion. I think this is a pity because of the negative aspects of negative feedback related to the increasing amount of higher order harmonics in the distortion compared to the lower order harmonics. Anyway, to avoid wasting too much power in the (large) emitter resitors he suggests to bypass these with diodes. He did the same for his "The Barney Oliver Audio Amplifier", where the emitter resistors are 8.2 Ohm (!), bypassed by a diode.
Around the same time others have made similar calculations or simulations, like Blomley in Wireless World (also february 1971!
) and later Self.Steven
author = "B. M. Oliver",
title = "Distortion in complementary-pair class-{B}
amplifiers",
journal = j-HEWLETT-PACKARD-J,
volume = "22",
number = "6",
pages = "11--16",
month = feb,
year = "1971",
CODEN = "HPJOAX",
ISSN = "0018-1153",
bibdate = "Tue Mar 25 14:12:15 MST 1997",
acknowledgement = ack-nhfb,
classcodes = "B1220 (Amplifiers)",
keywords = "amplifiers; complementary pair class B amplifiers;
crossover distortion; difference distortion; electric
distortion; feedback distortion suppression; negative;
transistor beta",
treatment = "P Practical; T Theoretical or Mathematical",
from http://www.math.utah.edu/pub/tex/bib/hpj.html#Oliver:1971:DCC
mlloyd1
title = "Distortion in complementary-pair class-{B}
amplifiers",
journal = j-HEWLETT-PACKARD-J,
volume = "22",
number = "6",
pages = "11--16",
month = feb,
year = "1971",
CODEN = "HPJOAX",
ISSN = "0018-1153",
bibdate = "Tue Mar 25 14:12:15 MST 1997",
acknowledgement = ack-nhfb,
classcodes = "B1220 (Amplifiers)",
keywords = "amplifiers; complementary pair class B amplifiers;
crossover distortion; difference distortion; electric
distortion; feedback distortion suppression; negative;
transistor beta",
treatment = "P Practical; T Theoretical or Mathematical",
from http://www.math.utah.edu/pub/tex/bib/hpj.html#Oliver:1971:DCC
mlloyd1
Im planning a Leach with 12 transistors using +/- 80V rails.
I have massive sinks so I would like to bias into Class A, I can dissipate around 300W per channel.
Im a little confused because I have read that you dont want to go over 50mV across the emmitter resistor, and that you dont want to go as low as 0.2R in high bias for stability reasons, but I cant see how it could possible work.
Heres a couple of options, which is the best way to proceed?
v across resistor resistor bias x 6 total idle watts
0.1 0.33 .818181818 290.9090909
0.065 0.20 1.909090909 312
the spacing disapears for some reason, if I use:
0.20 ohm resistors I get 312 watts with a Vre of 0.65V
0.33 ohm resistors I get 290 watts with a vre 0f 0.1V
I have massive sinks so I would like to bias into Class A, I can dissipate around 300W per channel.
Im a little confused because I have read that you dont want to go over 50mV across the emmitter resistor, and that you dont want to go as low as 0.2R in high bias for stability reasons, but I cant see how it could possible work.
Heres a couple of options, which is the best way to proceed?
v across resistor resistor bias x 6 total idle watts
0.1 0.33 .818181818 290.9090909
0.065 0.20 1.909090909 312
the spacing disapears for some reason, if I use:
0.20 ohm resistors I get 312 watts with a Vre of 0.65V
0.33 ohm resistors I get 290 watts with a vre 0f 0.1V
Last edited:
Dear,
This golden "25mV rule" Do you mean 25mV over each emitter resistor? or 25mV total over both NPN and PNP emitter resistors? (means 12,5mV over each emitter resistor duh
)
In this case using 0.1R emitter resistors is in most situations a no go in a class B amplifier, since you need a lot of idle current to reach the 25mV value.
If memory serves me well, the Parasound JC1 has 0.1R emitter resistors.
With kind regards,
Bas
This golden "25mV rule" Do you mean 25mV over each emitter resistor? or 25mV total over both NPN and PNP emitter resistors? (means 12,5mV over each emitter resistor duh
)In this case using 0.1R emitter resistors is in most situations a no go in a class B amplifier, since you need a lot of idle current to reach the 25mV value.
If memory serves me well, the Parasound JC1 has 0.1R emitter resistors.
With kind regards,
Bas
The 25mV rule applies to an optimally biased ClassAB complementary EF output stage.
It does not apply to CFP. It does not apply to Quasi. It does not apply to ClassA, It does not apply to single ended.
The 25mV rule is actually 26mV less a correction for the effective internal impedance of the output device.
As Re goes down so does the Vre. eg. Re=0r47 Vre~24 to 26mV, Re=0r1 Vre~18 to 20mV.
You cannot bias a +-80Vsupply amplifier into ClassA for conventional speakers.
Let's look at the numbers for an 8ohm speaker.
Vpk ~70V.
Ipk~ 70/8= 8.75A.
total Iq >= 8.75/2 ~4.4A.
Total power dissipated in the output devices is 80* 2 * 4.4 = 704W.
Dissipation per device is ~59W.
What size of heatsink would this need?
What device could stay within it's temperature de-rated SOAR while Pq is 59W?
What amp like this could survive a 4ohm speaker load?
Could any amp like this survive a 1r3 test load?
It does not apply to CFP. It does not apply to Quasi. It does not apply to ClassA, It does not apply to single ended.
The 25mV rule is actually 26mV less a correction for the effective internal impedance of the output device.
As Re goes down so does the Vre. eg. Re=0r47 Vre~24 to 26mV, Re=0r1 Vre~18 to 20mV.
You cannot bias a +-80Vsupply amplifier into ClassA for conventional speakers.
Let's look at the numbers for an 8ohm speaker.
Vpk ~70V.
Ipk~ 70/8= 8.75A.
total Iq >= 8.75/2 ~4.4A.
Total power dissipated in the output devices is 80* 2 * 4.4 = 704W.
Dissipation per device is ~59W.
What size of heatsink would this need?
What device could stay within it's temperature de-rated SOAR while Pq is 59W?
What amp like this could survive a 4ohm speaker load?
Could any amp like this survive a 1r3 test load?
For who may be interested, the link for downloading a PDF copy of the the Oliver paper is as follows. Thanks Steven and mlloyd1, for posting the complete journal info!
http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1971-02.pdf
Regards,
Satoru
http://www.hpl.hp.com/hpjournal/pdfs/IssuePDFs/1971-02.pdf
Regards,
Satoru
Last edited:
Dear Andrew,
I was aware it counts for class A/B EF stages. But it is still unclear for me, if it is 26mV total over both Re's or 26mV per Re?
With kind regards,
Bas
Andrew,
I understand most of this, but as I explained on the other thread, I want to bias to 300 watts. To most people this is class A, perhaps I should have been more specific, but I thought it was obvious from my bias settings what I was trying to achieve.
So if the 26mV rule does not apply to class A, does it apply to partially biased Class A?
Are you saying that if I want Class A, I can ignore 26mV rule but the amp should not enter Class AB as it may become unstable?
Long post
Luke you are completely misunderstanding what ClassA is.
A ClassA amplifier keeps all it's active devices actively controlling the output.
This is a very unusual way of stating the ClassA definition, to the extent that is is not actually the ClassA definition.
Let's re-phrase it more like what many expect.
In ClassA no devices switch off, all pass current.
But that simple definition allows wierd circuit solutions to be developed that deliberately allow active devices to pass almost constant near zero current (cf. load current) and these nearly off devices make no active contribution to the load current.
ClassAB is a push pull circuit arrangement that passes current from one supply rail to the load and returns that current through the ground.
On alternate half cycles the lower output devices sinks current from the ground through the load to end up at the negative supply rail.
ClassAB become ClassB when the two halves of the cycle meet at zero output volts and no current passes from rail to rail and each of the lower and upper halves of the circuit conduct for exactly 180degrees of the 360degrees of a waveform.
ClassB generates crossover distortion.
ClassAB biases the output stage to reduce this crossover distortion.
Increasing the ClassAB bias allows distortion to reduce to a minima and then the distortion rises with further increase in bias.
The output bias that generates the minima in the distortion is the optimum ClassAB bias.
It has been found that this is not based on optimum current. It can be modeled very closely by basing the bias point on voltage. I and some others use the term Vre as the voltage across the emitter resistor of the output device.
By using the 26mV rule and comparing it to Vre we can bias the stage pretty close to that distortion minima. A distortion analyser may/will allow this minima to be set more accurately.
Now back to the amount of bias.
Remember that further increase of the output bias leads to an increase in distortion, while the circuit continues to work in ClassAB mode.
Amplifiers can be biased into this region and can sound very good, better even than an optimally biased ClassAB amplifier. But there is a trick in there that must be achieved to allow for this better sound.
I'll come back to that.
ClassA bias is when both the upper and lower active devices conduct (and control) the output current for the whole 360degrees of waveform.
In push pull the output current is the current in one half minus the current in the other half with a tiny bit going through the Negative FeedBack Loop.
If the bias is set to 0.1A. The output current = zero, when the top half and the bottom half both pass exactly the same current. This is the same as setting the output offset to zero mVdc. Now apply control voltages to the upper and lower active devices. The control voltage reduces the current in one half while increasing the current in the other half. Let's pop in a number. 0.1A of bias reduces by 40mA to 60mA in the top half and increases by 40mA to 140mA in the lower half.
The output current is 60mA -140mA = -80mA.
The load is sourcing 80mA from ground to the negative supply rail.
Increase the top half current to 199mA. The lower half should pass ~1mA.
The load current 199mA - 1mA = 198mA sunk from the positive supply rail to ground. Note that the sign of the load current tells one whether the current is sinking or sourcing to ground and also which supply rail is meeting that demand.
So far the push pull ClassAB amplifier has been operating in it's ClassA mode. The output current is less than 2times the output bias current.
Further increase of the control voltage will reduce the current in one half of output stage to effective zero mA and in the other half to more than twice the bias current. Let's assume the upper half is passing 250mA. The lower half is effectively passing 0mA.
The output current is 250mA-0mA = 250mA and this is sunk to ground from the positive supply rail. The amplifier is now operating in ClassAB mode. It is no longer in ClassA mode. I will not call it ClassB mode because that requires the 180degress of waveform rule. ClassAB rarely if ever complies with that.
A ClassAB amp can continue using increasing control voltage to keep sending more current through one half of the output stage. That is until something else says STOP. This could be excessive heating, or protection circuit triggering or the PSU simply running out of puff (current).
Note, throughout this discussion we are looking at output CURRENT and load CURRENT. NOT LOAD VOLTAGE.
Most music is played with the amplifier not clipping on output voltage and not clipping on output current. For most of us it sounds too horrible to listen to a clipped signal and we choose to lower the volume control to limit the amount of clipping, maybe not quite to zero clipping but low enough to be tolerable.
The average music level will be way below the clipping level just referred to.
Music and other sound sources generally have an average to peak range of 10dB to 30dB. I'll use 20dB for the remainder of this argument.
your amplifier is on average putting out ~ 10% of it's voltage capability ref max power.
If the majority of the average output is in ClassAB mode then there is ClassAB crossover distortion. If the bias is set very low then nearly all the power coming from the amplifier is ClassAB mode power and al of this will have some crossover distortion.
If the bias is set very high then much of the average power will be delivered in the ClassA mode. Only the extreme transients will demand that the amp enters ClassAB mode. If the designer chooses to set the bais high enough that most of his clients listen to much of their sound output in ClassA mode rather than jumping from ClassA to ClassAB and back again, it is possible the high bias amplifier can sound better than a low bias amplifier.
This design choice is expensive. But some go that route.
Here's the trick I referred to earlier.
Choose an optimum bias current that is so high that almost all the sound output from that amplifier is in ClassA mode and that because it is set at optimum ClassAB bias it has also been set to stay in or very close to that distortion minima for the few infrequent signals that make the amp switch to ClassAB mode.
Increasing the number of output devices automatically increases the amount of optimum bias current.
Decreasing the emitter resistor of the output device automatcially increases the amount of bias current.
Combine these two options and you can find that the optimum ClassAB bias current is so high that any signal that is -6dB ref maximum power stays in ClassA mode. Wonder of wonders. A classAB amplifier that sounds great, does not get as hot and not as expensive as a full ClassA amplifier and retains the clean output characteristics of a ClassAB amp that is optimally biased.
I believe this is the region you want to be in.
Now, instead of working with powers start working with currents and hopefully a sensible solution will come out, after looking at all the options.
Sorry to all, that this has taken so long. But hopefully a few that got to the end can see some clarity.
Luke you are completely misunderstanding what ClassA is.
A ClassA amplifier keeps all it's active devices actively controlling the output.
This is a very unusual way of stating the ClassA definition, to the extent that is is not actually the ClassA definition.
Let's re-phrase it more like what many expect.
In ClassA no devices switch off, all pass current.
But that simple definition allows wierd circuit solutions to be developed that deliberately allow active devices to pass almost constant near zero current (cf. load current) and these nearly off devices make no active contribution to the load current.
ClassAB is a push pull circuit arrangement that passes current from one supply rail to the load and returns that current through the ground.
On alternate half cycles the lower output devices sinks current from the ground through the load to end up at the negative supply rail.
ClassAB become ClassB when the two halves of the cycle meet at zero output volts and no current passes from rail to rail and each of the lower and upper halves of the circuit conduct for exactly 180degrees of the 360degrees of a waveform.
ClassB generates crossover distortion.
ClassAB biases the output stage to reduce this crossover distortion.
Increasing the ClassAB bias allows distortion to reduce to a minima and then the distortion rises with further increase in bias.
The output bias that generates the minima in the distortion is the optimum ClassAB bias.
It has been found that this is not based on optimum current. It can be modeled very closely by basing the bias point on voltage. I and some others use the term Vre as the voltage across the emitter resistor of the output device.
By using the 26mV rule and comparing it to Vre we can bias the stage pretty close to that distortion minima. A distortion analyser may/will allow this minima to be set more accurately.
Now back to the amount of bias.
Remember that further increase of the output bias leads to an increase in distortion, while the circuit continues to work in ClassAB mode.
Amplifiers can be biased into this region and can sound very good, better even than an optimally biased ClassAB amplifier. But there is a trick in there that must be achieved to allow for this better sound.
I'll come back to that.
ClassA bias is when both the upper and lower active devices conduct (and control) the output current for the whole 360degrees of waveform.
In push pull the output current is the current in one half minus the current in the other half with a tiny bit going through the Negative FeedBack Loop.
If the bias is set to 0.1A. The output current = zero, when the top half and the bottom half both pass exactly the same current. This is the same as setting the output offset to zero mVdc. Now apply control voltages to the upper and lower active devices. The control voltage reduces the current in one half while increasing the current in the other half. Let's pop in a number. 0.1A of bias reduces by 40mA to 60mA in the top half and increases by 40mA to 140mA in the lower half.
The output current is 60mA -140mA = -80mA.
The load is sourcing 80mA from ground to the negative supply rail.
Increase the top half current to 199mA. The lower half should pass ~1mA.
The load current 199mA - 1mA = 198mA sunk from the positive supply rail to ground. Note that the sign of the load current tells one whether the current is sinking or sourcing to ground and also which supply rail is meeting that demand.
So far the push pull ClassAB amplifier has been operating in it's ClassA mode. The output current is less than 2times the output bias current.
Further increase of the control voltage will reduce the current in one half of output stage to effective zero mA and in the other half to more than twice the bias current. Let's assume the upper half is passing 250mA. The lower half is effectively passing 0mA.
The output current is 250mA-0mA = 250mA and this is sunk to ground from the positive supply rail. The amplifier is now operating in ClassAB mode. It is no longer in ClassA mode. I will not call it ClassB mode because that requires the 180degress of waveform rule. ClassAB rarely if ever complies with that.
A ClassAB amp can continue using increasing control voltage to keep sending more current through one half of the output stage. That is until something else says STOP. This could be excessive heating, or protection circuit triggering or the PSU simply running out of puff (current).
Note, throughout this discussion we are looking at output CURRENT and load CURRENT. NOT LOAD VOLTAGE.
Most music is played with the amplifier not clipping on output voltage and not clipping on output current. For most of us it sounds too horrible to listen to a clipped signal and we choose to lower the volume control to limit the amount of clipping, maybe not quite to zero clipping but low enough to be tolerable.
The average music level will be way below the clipping level just referred to.
Music and other sound sources generally have an average to peak range of 10dB to 30dB. I'll use 20dB for the remainder of this argument.
your amplifier is on average putting out ~ 10% of it's voltage capability ref max power.
If the majority of the average output is in ClassAB mode then there is ClassAB crossover distortion. If the bias is set very low then nearly all the power coming from the amplifier is ClassAB mode power and al of this will have some crossover distortion.
If the bias is set very high then much of the average power will be delivered in the ClassA mode. Only the extreme transients will demand that the amp enters ClassAB mode. If the designer chooses to set the bais high enough that most of his clients listen to much of their sound output in ClassA mode rather than jumping from ClassA to ClassAB and back again, it is possible the high bias amplifier can sound better than a low bias amplifier.
This design choice is expensive. But some go that route.
Here's the trick I referred to earlier.
Choose an optimum bias current that is so high that almost all the sound output from that amplifier is in ClassA mode and that because it is set at optimum ClassAB bias it has also been set to stay in or very close to that distortion minima for the few infrequent signals that make the amp switch to ClassAB mode.
Increasing the number of output devices automatically increases the amount of optimum bias current.
Decreasing the emitter resistor of the output device automatcially increases the amount of bias current.
Combine these two options and you can find that the optimum ClassAB bias current is so high that any signal that is -6dB ref maximum power stays in ClassA mode. Wonder of wonders. A classAB amplifier that sounds great, does not get as hot and not as expensive as a full ClassA amplifier and retains the clean output characteristics of a ClassAB amp that is optimally biased.
I believe this is the region you want to be in.
Now, instead of working with powers start working with currents and hopefully a sensible solution will come out, after looking at all the options.
Sorry to all, that this has taken so long.
But hopefully a few that got to the end can see some clarity.
Last edited:
yes, 15mVre to 25mVre, note that is across each Re. Doug Self gives a total voltage across the pair of series connected Re and for 0r22 he quotes 46.2mVre (he calls it optimal Vq).
setting to 20mVre is about right. There are only two ways that I know of to improve on this.
Use a distortion analyser to minimise the unwanted harmonics.
Use your ear to find a sound quality that you and your neighbours like.
I suspect the combination of all 3methods is more correct than using any single method.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.