That's what I want to clarify the diffusion capacitance is part of the Cbe. But even bootstraped, say divided by 100(beta), the capacitance load by the base of the power transistor are still high.
Like in the book said, when collector current is 1A, the diffusion capacitance is over 2uF!!! Divided by 100 is still 20nF at the input to ground.
Back to my pre driver, if I divided 1600pF by 100, it's still 16pF. So changing from the KSC3503 to 2SC4793 increases from 19pF(2.6+16) to 42pF. so it is really a little over double, not exactly 10 time worst if you look at the Cob alone.
Make sense, not what I want to hear, but make sense for the pre-drivers. I think it's more like Vbe varying 100mV and the collector voltage varies 40V. But it's a valid point.
In output transistor case, the input capacitance can be as high as 2uF at 1A current as in Mr. Cordell's book, divide by 1000 is still 2000pF. Normal Cob is about 500pF for the big transistors. In this case, it's more questionable. What do you think?
Last edited:
I am more worry about the max slew I can get, compensation is not hard at all.
I ask this question because of my experience of the 3EF diamond http://www.diyaudio.com/forums/solid-state/278494-issue-3ef-diamond.html
I have the output waveform collapsed on me as I crank the frequency up. The major problem is the folded pre driver cannot drive much as it's folded, unlike the regular 3EF. I tried without dummy load and it does not improve at all. BUT if I cut 2 out of 5 pairs of big output transistors, it helped. I even simulate using RC loading the drivers and I can make it collapse. The problem is the input capacitance. I am contemplating in buying 2SC5242 and 2SA1962 that has lower Cob, but then I thought about this diffusion capacitance Mr. Cordell talked about. It seems like there is no point to change output transistors. The capacitance load from the output transistors reflect back to the drivers and to the pre-drivers. Only way so far is to increase the current of the driver and pre-drivers and they get hot.
For the pre-drivers, you have a point, the Cob do make a difference as I only run 35mA through the pre-drivers, the diffusion capacitance is not dominating the input capacitance. If I go back to KSC3503/KSA1381, it has Vbe of 0.66 rather than the 2SC4973 that is 0.58V. I loss 200mV headroom for the bias spreader.....that is already only 1.2V due to the folded pre-driver. I need to find a power transistor that has higher Vbe than the MJL3281/1302 to gain back the headroom!!!
Last edited:
Have you tried a large capacitor across the diamond predriver emitters? This would allow that stage to work in class AB at high frequencies. Also the 10uF cap across the output bases might be too large, causing the drivers to rectify into it and eventually raising the bias so they turn off.
Yes, the pre-drivers, the drivers. The only way is to increase current to insane amount. I don't have the 10uF across the emitters of the drivers, it is not helping. I settled with 33ohm on R2, R33 and R17 to get the best compromise that the waveform collapse at 1.3MHz and recovered when I back down to 900KHz.
The supply current ran up to over 2.5A when I crank up the frequency above 1MHz also in order to charge the input capacitance of the big output transistors.
I think the collapsing is a problem only for the folded pre-drivers as the driving current is limited to the CCS. For regular 3EF with all normal EF drivers, you should not have the problem. BUT I believe the supply current is still going to ramp up as you increase the frequency. I just hate to deal with the true 3EF that I need to compensate for 6Vbe.
You are right, I am going overboard. We better not take up more space on this thread. I hope you can put in your comments on my other thread about 3EF Diamond. It's really a dance between frequency response, bias spreader voltage and all different issues. I have a suspicion very few people even attempted the 3EF Diamond. I got it running stable already, just giving a final touch to make sure the pre-drivers is not loading the VAS.
Thanks
Last edited:
. Can you explain?Hi Alan,
As pointed out earlier, the Cbe is bootstrapped by the gain of the emitter follower. If Cbe were 10,000pF and the small-signal gain of the EF were 0.95, then effective capacitance would only be 500pF. In the case of a predriver, the EF gain may be closer to 0.99.
But there is a bigger question here. Why do you need to change out the predriver transistor to get lower Vbe? This does not make much sense to me. Can you explain? Maybe there is a misunderstanding at the root of this. The 3503/1381 are outstanding predrivers.
Cheers,
Bob
Hi Mr. Cordell
Thanks for your reply. It's a long story. I designed a 3EF Diamond OPS pcb, I found I need very low bias spreader voltage. Attached is my OPS schematic. I have to use the CFP spreader because it works really good in adjusting tempco by balancing TPOT2 and the TPOT1 to get the right compensation. The problem is the CFP needs more headroom to prevent the transistors from saturating. The spreader voltage is down to 1.1V or less. I had problem of the spreader saturated and could not be adjusted. I need transistors with Vbe to gain back the headroom.
I did post here to ask you what parameters in data sheet will give me the insight which has lower Vbe. After experimenting, I conclude that those that has higher current capability gives lower Vbe. I since found all the transistors that I needed and it worked beautifully. I got the spreader voltage up to 1.2V and I tested running at 140mA each stage and the voltage across the emitter resistor of the power transistor stay at 30mV within 1mV from stone cold to not touchable hot. I even drive the circuit with 1.5MHz that increase the supply current to over 2.5A for 20seconds and stop to measure. The voltage of the emitter resistor increase to 35mV but back down to 30mV again within 20 seconds. Thermal circuit really works.
To answer your question, in order to get the headroom of the spreader, I had to change the 3503/1381 to 2SC4793/2SA1930. the Vbe of 3503/1381 are 0.66V, being a folded EF, this really eat up the headroom of the bias spreader. The Vbe of the larger 2Sxxxxx transistor is 0.58V. Just by changing these, I gain 160mV head room!!! This together with using transistors with low Vbe in the spreader circuit led to my success in thermal compensation of this OPS.
This bring back to my post. The Cob of the bigger 2Sxxxxx is 26pF vs 2.6pF of the KSC/KSA. This increased the loading of the VAS and hurts slew rate. So before I try to use the KSC/KSA again, I would like to make sure that this extra capacitance does matter. Sadly, being a folded EF, the max current is set by the CCS to about 35mA. So gm is about 35/26=1.35. Diffusion capacitance=1.35/(6.28 X 100EE6)=2.15nF. Using 0.99 as you suggested, the input capacitance is 2150pF/100=21.5pF actually loading the VAS. So comparing KSC/KSA to the 2Sxxxxx, the capacitance increased from (2.6+21.5)pF to (26+21.5)pF, which is a little more than double. So, now, I want to see whether I can squeeze out a little more headroom from other part and bring back the KSC/KSA. I am ordering the NJW0281/0302 and 2SC5242/2SA1962 to test the Vbe. NJW looks promising as the data sheet shows the Vbe is higher the MJL3281/1302 that I am using. I can pick up about 80mV( I hope) of headroom.
On the side note, maybe this is of your interest. I have a thread about the 3EF Diamond that I ran into problem with oscillation. I post my theory and solution, you can double check my theory if it is of interest to you: http://www.diyaudio.com/forums/solid-state/278494-issue-3ef-diamond.html
My conclusion is the oscillation is caused by the feedback from the collector of the pre-driver back to the input. Problem showed up when I change from KSC/KSA that have Cob=2.6pF to a transistor that has higher current capability. Cob increase with the increase size of the die, it is my conclusion that Cob is the feedback path from the collector to the base that cause the oscillation. Everything works as soon as I move the collectors of the pre-drivers to the rail to break the feedback path. Now I get only 10% overshoot in square wave test with ringing die down in one cycle. I feel it's rock solid. So I ended up with a "Folded 3EF" instead of 3EF Diamond.
Apparently, very few if anyone tried the 3EF Diamond as I have to thread up for quite a few days already and nobody even responded. I really like the folded configuration as I can screw the pre-driver and driver together and based on my extensive testing and data taking, they really cancel out the tempco. So I only have the tempco of the power transistors to worry about. I want to make it work and I think I already make it work already.
Also you can see in the other thread I have wave form collapse beyond 1MHz. I think this is when the pre-driver and driver loss control as you described in p197 in your book. The folded EF has more problem as the current is limited by the CCS.
I hope you can talk a little more in your next book on situations where you need low bias spreader voltage and maybe using low Vbe transistors to help the situation. It really works. The CFP spreader is very good in adjusting tempco. I am going to put two trim pots, one for R37 and the other stays. With that, I can balance the two to get the right tempco. I first set the TPOT2 to minimum and adjust the TPOT1 to get the right bias. then test the drift. If it drift high, I increase R37 which will lower the spreader voltage. Then I adjust the TPOT1 to get back the voltage. This will increase the tempco as I increase the multiplying factor. Test and adjust until I get no drift.
Thanks for your time.
Attachments
It is more a layout problem that sway me to this. Layout is everything. If you do normal 3EF, you are pretty much forced to put the drivers and predrivers in the middle of the pcb with big power transistors on both sides. You cannot put the drivers and predrivers at the end as the temperature will be very different than the big transistors.
Also, because of the high current, it is not practical to have alternate NPN PNP power transistors to keep the current local. You are forced to have NPN on one side and PNP on the other side. This will cause severe current running across the pcb with the more sensitive circuit in the middle catching all the cross fire. The only other way is to have the spreader sense both the power transistors and the drivers separately to compensate. That really complicate things.
With Diamond, I can isolate the drivers and predrivers and cancel their tempco by screwing them together on a separate heat sink. I only need to compensate the power transistor alone. I managed to put the drivers and predrivers on one side of the pcb and alternate NPN PNP power transistors.
I got it working, thermal compensation is very good when I fix everything. Bias is rock solid from stone cold to not touchable hot. Normal 3EF need to compensate all the transistors, here all drivers and predrivers are self compensated. Because the drivers are bolted together and the spreader is bolted on top of the power transistors, the thermal response is very good. The voltage across the emitter stay to within 1mV from stone cold to not touchable hot. I simulated short high power passage by running 8Vpp 1.7MHz input to ramp the supply current to over 2.5A for 20 seconds before removing the signal, then immediately monitor the voltage across the emitter resistor. It went from 30mV to 35mV only and back down in 20 second after I remove the signal. I think this is very good. I don't think you can get this that easy with the normal 3EF.
There is a slight advantage of gaining 1.4V headroom for the rail also.
Last edited:
Hi Alan,
You've touched on many things here, and there are many issues. First of all, your suggestion for me to discuss further in my second edition the use of low-voltage bias spreaders (e.g., about 2Vbe) is a good one. The spread of only 2Vbe does limit some options in terms of spreader design and headroom options.
That having been said, selecting transistors in the pre-driver, driver and output stage for Vbe is just not the way to go, in my opinion.
Your use of the CFP bias spreader in this case seems to be at the heart of your problems, as it needs more headroom than the simple one-transistor Vbe multiplier. You may have to forego the CFP bias spreader. Its main advantages are its lower impedance, reducing or eliminating the need for the compensation resistor in the collector that is often used, and its higher current gain that reduces the base current drawn from the Vbe multiplier resistive divider. I have not thought carefully enough about your claim that the CFP gives better temperature compensation, but it may be some sort of second-order effect. It is not clear why the second pot helps. Beware of any approach that depends on transistor current gain for its behavior. For now, I think you may have to stick with the one-transistor Vbe multiplier with maybe the compensation resistor in the collector. The drop across this compensation resistor does not reduce headroom of the Vbe multiplier transistor.
Bear in mind that there are other ways in the folded Triple to increase the needed spread voltage, such as putting small resistors directly in series with the driver emitters. If you are running 100mA through the drivers, even 1 ohm will buy you a total of 200mV for the spreader. If you want to go even further, you can put a diode in series with each emitter; the diode dissipates very little, so its tempco will only slightly come into play as a function of the ambient temperature of the board.
The high frequency signal collapse you speak of should not be an issue of concern. Large signal operation at 1 MHz is completely unrealistic. Also, it sounds like you were discussing having not enough current to drive the bases of the drivers, as dictated by the current sources for the pre-drivers.
As far as available current to drive the output transistor bases at high frequencies, the issue is the same whether it is a standard 3EF Triple or a triple with a folded pre-driver.
Cheers,
Bob
Hi Mr. Cordell.
Thanks for your reply.
The reason I choose the CFP spreader is because I can adjust the tempco. I need two trim pot and I can adjust tempco and bias current. I don't think you can do it with one transistor. I tried with one transistor and the bias current kept going up. With two trim pot, it's easy to trim the pot to never turn on the PNP and the spreader becomes a single transistor. It does not work. All the big transistor has very low Vbe, the spreader is going to be only 1.1V or less realistically. You won't have the multiply ratio to get the right tempco. You have -4mV/deg C to compensate, you can only get -4mV/deg if the bias spreader is 1.4V.
The idea of adding a resistor at the emitter of the driver transistor is a good one. But one thing concern me is I have 5 pairs of output transistor. The base stop to each power transistor is 2.2ohm. I I use 1ohm like you suggested, that is equivalent of changing the 2.2ohm to 2.2+5=7.2ohm. This is because I have 5 pair in parallel. I think 0.5ohm will be a good one, this will be equivalent to 4.7ohm base stop. What do you think? I can up the current of the driver to 50mA, that will increase the spreader voltage by 50mV. That will help a little and I still stay with base stop equivalent to 4.7ohm.
I know choosing by Vbe is not the best way, but even after I get the extra 50mV, I still need a little more. I really want to go back to the KSC/KSA for lower input capacitance to put less loading to the VAS. I still want to order the NJW0302/0281 as the data sheet show it's going to have higher Vbe than the MJL1302/3281 that I am currently using. I have 5 pairs of output transistor, I can afford to use a lower power transistor and not worry about the SOA.....I hope!!!
Thanks
Alan
Last edited:
Can you explain in detail how you use the CFP to change the tempco without depending on one of the transistors being moved to an abnormal operating point (e.g. quasi-saturation or starved or something like that)?
Assuming that the input transistor of the CFP is mounted on the output transistor, one will expect to get about 2.2mV/C tempco slope of the temperature of that transistor, times the Vbe multiplier ratio. That ratio would normally be about 2 to achieve the needed spreader voltage. The tempco in that situation would be about 4.4mV/C
Assuming that the mV/C number and/or the Vbe multiplication number are different for various reasons, one may need to have that 4.4mV/C number be different. I assume this is the tempco adjustment you are talking about.
Placing a 1-ohm resistor in series with the emitter of the driver is not quite the same as effectively increasing the base stoppers by 5 ohms, since ALL of the output transistor bases see the same effect of the added 1-ohm resistor. This means that current hogging stability is not degraded in the same way.
With such a large amplifier with 5 output pairs you probably should indeed increase the driver bias current to 100mA - and make sure you have enough heat sink on it.
Cheers,
Bob