I received an amplifier for repair. There was a lot of DC on the output and the servo could not take care of this. The problem was easy to fix as the diff amps their cascodes and current mirrors did not use matched devices for both the positive and negative complementary front end. I put in matched devices and the the servo took care of the then 10mV offset. Now it is <3mV. It is loosely based on Bob Cordell's output stage with a conventional fron end (Diff amp well degenerated and the VAS runnig at about 20mA. The bias spreader is made from a pair of NPN and PNP TO220 devices which are clamped to one of their respective sex output devices.
The amplifier runs off a +/-90v rail and uses a triple output from page 99 (figure 5.1b) of Bob's first edition book.
The bias spreader uses a 25 turn pot so very fine adjustment is available. The spreader also monitors the emitter voltage of the first pair of drivers which are alos TO-220.
The output devices are 2SC5200 and 2SA1943 all matched. The drivers are the same as the outputs. 0.33R emitter resistors are used with base stoppers of 3R9 on each output device, 4R75 on each driver and 100r on each base of the first triple drivers. there is no oscillation that I can see up to 100MHz.
There are 12 up and 12 down output devices on a massive heatsink! Weighs abour 12Kg alone. The power supply is a switchmode with PFC front end so it is well regulated. Runs at about 70KHz and he uses 35,000mfd on each rail. The PFC supplies the great regulation.
The re of the 12 // devices is small indeed
Now trying to set the bias to allow the amplifier to run in class A up to maybe 50 watts is my client's request. So first I want to set bias so that crossover THD is very small.
Referring to the attched PDF, the crosscoupled resistors on the pre-pre driver is 158R and that of the drivers is 15R (To allow all base charge to be swept away quickly).
Again referring to page 101-103 of Bob's book, he talks about this 26mV across each emitter resistor and having an optimum bais current PER DEVICE of 79mA. With 12 devices that is almost 1amp of idle current. No problem as the power supply and heatsink are up to the task. I checked the bias spreader that it was reducing it's voltage by about 2.2mV/deg C.
This 26mA through each emitter resistor does not ring true to me. What if your resistors are 0.1 ohm then 26mV across each represents 0.26A per device of idle current!
Back to the triple. There are 6Vbe drops so the bias spreader needs to develop at least 3.6 volts + the drops across the 0.33R emitter resistors.
I have a meter from top emitter to bottom emitter which gives me the idle current of one pair of devices. This voltage/0.66 = idle current for one pair.
I have a second meter across the bias spreader.
I have my Audio Precision reading THD and I am also looking at the THD residual on a second scope.
When I try to adjust idle current to be 100mA per device (1.2 amps total) I have to push the bias spreader voltage close to 4 volts. This does not sit well with me.
The THD is quite low at midband (<0.006%) driving an 8 ohm load. At 20KHz it is <0.07% when sweeping power from 1 watt to 350 watts.
However as I said above my client wants about 50 watts of pure class A meaning about 3.5 amps peak of idle current = about 600 watts of idle heat.😎
Look forward to some insight.
The amplifier runs off a +/-90v rail and uses a triple output from page 99 (figure 5.1b) of Bob's first edition book.
The bias spreader uses a 25 turn pot so very fine adjustment is available. The spreader also monitors the emitter voltage of the first pair of drivers which are alos TO-220.
The output devices are 2SC5200 and 2SA1943 all matched. The drivers are the same as the outputs. 0.33R emitter resistors are used with base stoppers of 3R9 on each output device, 4R75 on each driver and 100r on each base of the first triple drivers. there is no oscillation that I can see up to 100MHz.
There are 12 up and 12 down output devices on a massive heatsink! Weighs abour 12Kg alone. The power supply is a switchmode with PFC front end so it is well regulated. Runs at about 70KHz and he uses 35,000mfd on each rail. The PFC supplies the great regulation.
The re of the 12 // devices is small indeed
Now trying to set the bias to allow the amplifier to run in class A up to maybe 50 watts is my client's request. So first I want to set bias so that crossover THD is very small.
Referring to the attched PDF, the crosscoupled resistors on the pre-pre driver is 158R and that of the drivers is 15R (To allow all base charge to be swept away quickly).
Again referring to page 101-103 of Bob's book, he talks about this 26mV across each emitter resistor and having an optimum bais current PER DEVICE of 79mA. With 12 devices that is almost 1amp of idle current. No problem as the power supply and heatsink are up to the task. I checked the bias spreader that it was reducing it's voltage by about 2.2mV/deg C.
This 26mA through each emitter resistor does not ring true to me. What if your resistors are 0.1 ohm then 26mV across each represents 0.26A per device of idle current!
Back to the triple. There are 6Vbe drops so the bias spreader needs to develop at least 3.6 volts + the drops across the 0.33R emitter resistors.
I have a meter from top emitter to bottom emitter which gives me the idle current of one pair of devices. This voltage/0.66 = idle current for one pair.
I have a second meter across the bias spreader.
I have my Audio Precision reading THD and I am also looking at the THD residual on a second scope.
When I try to adjust idle current to be 100mA per device (1.2 amps total) I have to push the bias spreader voltage close to 4 volts. This does not sit well with me.
The THD is quite low at midband (<0.006%) driving an 8 ohm load. At 20KHz it is <0.07% when sweeping power from 1 watt to 350 watts.
However as I said above my client wants about 50 watts of pure class A meaning about 3.5 amps peak of idle current = about 600 watts of idle heat.😎
Look forward to some insight.
Hi MOER,
It would be good to have an entire schematic for viewing. I have Bob's 2nd edition, but it's about 1000 miles from here. 🙁 Can you paste an excerpt referring to the 26mV discussion? I suspect this may refer to a device model in which the transistor's internal emitter resistance is 26mV/emitter current, eg. Re = 26 Ohm when emitter current is 1mA. The .33 ohm resistors will dominate by far.
I don't dispute your needing about 4V spreader to develop 100mA per device. And with 3.5A peak current and 35V rails, you're close to pure Class A over the entire available voltage.
You can download Cordell's 1st edition ( I did not refer to his 2nd edition)from the web and it is chapters 5 and 10
https://pearl-hifi.com/06_Lit_Archi...dell_Bob/Designing_Audio_Pwr_Amps_Cordell.pdf
The front end and VAS have no effect on the bias issue. I have no desire to back engineer and trace out the schematic.
This issue of gm doubling has always been troublesome for me as an amplifier designer. A pure class A amplifier where the output device is driven from a constant current source can have no gm doubling since the device(s) are always fully conducting. This is of course a signle ended design.
A "tradtional" class B amplifier(push pull) which is biased into full class A (to whatever degree) will of course have gm doubling since we have two active devices - one on top and one on the bottom pulling current at the same time.
I also have an issue of pure class A. We are taight in small signal design that a class A stage driving another should be designed that the second stage only pulls less than 10% of the standing current from the first.
So if we have a 10w at 8 ohm class A amplifier we must deliver 9v RMS to the load. This means that the peak current is 1.6A. My logic implies that this 1.6A must be 10% and no more of the standing current in the output stage, Thus the standing current must be ast least 9A RMS.
Is this extreme, yes it is. So I built two amplifiers, one with 1.5A RMS and one with 9A RMS standing currents. The 9A version sounded better.
Now I do confess to build a 100w 8 ohm amplifier based on this would be kind of crazy, but some food for thought.
https://pearl-hifi.com/06_Lit_Archi...dell_Bob/Designing_Audio_Pwr_Amps_Cordell.pdf
The front end and VAS have no effect on the bias issue. I have no desire to back engineer and trace out the schematic.
This issue of gm doubling has always been troublesome for me as an amplifier designer. A pure class A amplifier where the output device is driven from a constant current source can have no gm doubling since the device(s) are always fully conducting. This is of course a signle ended design.
A "tradtional" class B amplifier(push pull) which is biased into full class A (to whatever degree) will of course have gm doubling since we have two active devices - one on top and one on the bottom pulling current at the same time.
I also have an issue of pure class A. We are taight in small signal design that a class A stage driving another should be designed that the second stage only pulls less than 10% of the standing current from the first.
So if we have a 10w at 8 ohm class A amplifier we must deliver 9v RMS to the load. This means that the peak current is 1.6A. My logic implies that this 1.6A must be 10% and no more of the standing current in the output stage, Thus the standing current must be ast least 9A RMS.
Is this extreme, yes it is. So I built two amplifiers, one with 1.5A RMS and one with 9A RMS standing currents. The 9A version sounded better.
Now I do confess to build a 100w 8 ohm amplifier based on this would be kind of crazy, but some food for thought.
Indeed. The 26mV is the optimum for minimal xover distortion in a class AB amp.
For class A, obviously you up the bias current (voltage from bias spreader) until you have the wanted bias current for class A.
Jan
For class A, obviously you up the bias current (voltage from bias spreader) until you have the wanted bias current for class A.
Jan
The 26mV is according to Bob Cordell, who got it from an obscure HP paper, which doesn't appear to me to actualy state exactly this. But there are far too many amplifiers out there with optimal or specified bias different from 26mV for this claim to be valid.
This sounds to me pie in the sky. Barney Oliver from HP made a logical case for the 26mV which you can read online.
Also Bob C discusses it in his book with arguments why it is optimal.
You on the other hand just gives some personal opinion with no arguments.
I know where my money goes.
Jan
Also Bob C discusses it in his book with arguments why it is optimal.
You on the other hand just gives some personal opinion with no arguments.
I know where my money goes.
Jan
1. Check the SOA of all output transistors (including drivers and predrivers).
2. Check the thermal conditions.
3. Check the ability of the power supply to deliver continuous power for class A.
I don't think the demand is reasonable. I never know anybody successfully use conventional convection heatsink to dissipate 650W per channel, 1300W stereo. Probably useable during winter close to the arctic, absolutely useless indoor here in the tropics, even with mighty aircon in full blast.
Only 300 W of dissipation per channel. A class idle current needs to be half of peak A class output current, so 1.75 A for 50 W/8 Ohm.
I'm not sure that load pulling only 10% of idle current was the reason for better sound. It could be fact that with increased bias, A class output stage distortion is falling and becoming only second harmonic. Odd harmonics disappear to unmeasurable levels at high bias.
Where does the 10% come from then?
If 1.6A is enough to deliver the 10W in class A, what else is necessary?
Jan
If 1.6A is enough to deliver the 10W in class A, what else is necessary?
Jan
It is easy to write a simulator for a BJT push-pull output stage.
Ic slides along an exponential function of Vbe. The push-pull pair has two such exponentials, each a mirror image of the other. The bias determines how far the current slides down the exponential. There is a local minimum for crossover distortion, but it is narrow and hard to realize.
Ed
Ic slides along an exponential function of Vbe. The push-pull pair has two such exponentials, each a mirror image of the other. The bias determines how far the current slides down the exponential. There is a local minimum for crossover distortion, but it is narrow and hard to realize.
Ed
Jan, with all respect that is completely false. My argument was explicitly that there are too many power amps out here with different biasing requirements for the claim to be credible. This is certainly an argument, and it is certainly not just 'personal opinion'. I've biased amps at everything from 5 to 50mV, following factory instructions and checking with a distortion analyser.
A couple of comments on the "26mV" issue. I have measured the perofrmance of a typical amplifier at 1 kHz and 20kHz for crossover distortion, and would point out that dynamically, a higher bias than the static value may be needed,. This is because the transistor junctions have a relatively high input capacitance, and what may be optimum at D.C. may not be at higher frequencies. In one measurement, crossover distortion at 20kHz was reduced by increasing the bias current to around 120mA - and in that experiment, the low frequency crossover distortion was slightly higher. The nominal current for 0.33 ohm emitter resistors I used would have been 78mA from the "26mV" rule, but in practice I saw little difference at 50mA, only rising significantly at 30mA or below; except to say that a Quad 303 clone circuit gave lowest crossover distortion at 30mA.
Another point is that many power transistors do not follow the ideal response on which that 26mV is based, but due to resistive and high level injection effects, tend towards a 2kT/q response. Generally this occurs at higher currents than the ones typical of the quiescent current levels normally seen, but if the response begins to change from the kT/q value that may suggest a higher bias is needed.
Another point is that many power transistors do not follow the ideal response on which that 26mV is based, but due to resistive and high level injection effects, tend towards a 2kT/q response. Generally this occurs at higher currents than the ones typical of the quiescent current levels normally seen, but if the response begins to change from the kT/q value that may suggest a higher bias is needed.
Here is a link to Dr Barney Olivers paper on optimal biasing of class B output stages.
https://hifisonix.com/cross-over-distortion/
Unfortunately the document quality is not very good, so you need to take your time to go through it.
A lot of amplifier designers underbias their output stages - one of the worst I’ve seen is in the Marantz PM7000 where just a few mA are flowing. They do this to undersize the heat sinks I think. You get good distortion results from about 15 mV through to about 30 mV across the degeneration resistor. Values outside this range seem to be problematic. YMMV.
https://hifisonix.com/cross-over-distortion/
Unfortunately the document quality is not very good, so you need to take your time to go through it.
A lot of amplifier designers underbias their output stages - one of the worst I’ve seen is in the Marantz PM7000 where just a few mA are flowing. They do this to undersize the heat sinks I think. You get good distortion results from about 15 mV through to about 30 mV across the degeneration resistor. Values outside this range seem to be problematic. YMMV.
Last edited:
Attached file of the full journal where it was published. Slighly better quality, I think. Page #11.
Another factor to consider is that the junction temperatures of output (and driver) transistors is that they are frequently higher than "room temperature", whatever that means, but something in the region of 20C, or in warmer climes, even 27C. The kT/q thermal voltage, as the T indicates, rises with temperature. So, that would suggest that the "26mV" figure might not be the correct value to use when transistors are at their operating temperature, and the current would have to be increased to compensate. As a Class AB design will have varying power levels in the output devices, tracking the temperature is important, and while many circuits have been presented to maintain a constant quiescent bias current due to the -2mV/C or therabouts change in Vbe with temperature, perhaps in light of kT/q effects on crossover distortion, a little increase in current with temperature is allowable or may be desirable.