Hi Bob,
Thanks for the reply.
I described how to adjust in the last paragraph of post #6713. I did not invent this, I read from here someone said you can adjust the tempco. I just experiment with it. It really works.
I adjust TPOT2 to 0 first to take out the PNP, so I have only single transistor bias spreader. My experience with a few transistors is it always ran away. So I increase the TPOT2 to turn on the PNP, this will lower the bias voltage. Then I adjust TPOT1 to increase back the voltage. I tested it again for drift, readjust if necessary. It really works for me beautifully.
The problem with the output transistor is their Vbe is only 0.56 to 0.58V at idle of 100mA or so. You really don't have 1.4V. When I used the KSC/KSA as pre-drivers( Vbe=0.66V), my spreader voltage was down to barely 1V!!! And it went down as the circuit heated up. You don't have multiply factor to get -4mV/deg C. Not even close.
I think it's very important to use a spreader transistor with low Vbe, and I found one...KSD1691 that is same package and pin out as KSC3503. Vbe is 0.58V vs KSC of 0.66V. You get better multiply ratio with a transistor with low Vbe. theoretically, you get -4mV/deg when the spreader voltage is 2 X 0.58=1.16V. But it still needed help with the CFP configuration. I use ZTXP2012 for the PNP as it's Vbe is 0.58V also.
As is right now, with the 2SC4793/2SA1930 that has Vbe=0.58V as pre drivers, my spreader voltage is about 1.215V for 140mA per stage ( I use 0.22ohm emitter resistor and get 30mV across). If I change back to KSC/KSA, I will loss 160mV and my spreader has to go down to 1.055mV. Using the two new spreader transistors, the two Vbe=1.16V. This mean the collector of the NPN spreader transistor is going to be below the base already. When the circuit heats up, the transistor will be close to saturation.
Now, as your suggestion, I can gain back at least 100mV. I made a mistake of laying out the pcb too narrow, I don't have much room to put in a bigger heatsink for the driver transistors. it's getting hot right now. I have to consider making a custom heatsink to be able to run 100mA. I am going to do a second layout once I test the whole amp for sure.
I really think it's worth while to do the folded 3EF, thermal stability is very good, you gain 1.4V swing on each side. It makes layout a lot easier. I don't like the normal 3EF. In order to have better thermal compensation, you have to have the drivers and pre-drivers and spreader in the middle of the board to have more even heat. You are forced to have power transistor on both sides of of the more sensitive circuits. Unless you use 4 layer board, it is almost impossible to layout so you have NPN PNP NPN PNP alternating power transistor to keep the ground current local. If you have NPN on one side, PNP on the other side, you are going to have large current traveling back and fore the middle of the board where the sensitive circuits are. That's the first and foremost reason I gone to 3EF Diamond. I managed to put the drivers and predrivers on one end of the board and alternate NPN PNP power transistor.
What do you think about my theory of the oscillation of the 3EF diamond? I don't think it's worth while to go the diamond, it's too dangerous and even if it does not oscillate, it rings with square wave test. But it's worth while to go folder pre drivers. I did measurements, the tempco of the drivers and the pre-drivers really cancelled out as I bolted them together on a separate heat sink.
Again, thanks for your time.
Last edited:
Hi Mr. Cordell
I was thinking about this, I am not worry about current hogging. I am more worry about increase distortion as the current increase as the voltage swing farther away from 0V. There is a compression effect when the base current of the power transistors increase. The 1ohm will equivalent to 5ohm increase on each of the 5 power transistor. This is symmetrical on both polarity and create odd harmonics.
Thanks
Alan,
You are worrying about beta droop at high current in the output transistors. Absent beta droop, there would be no compression effect that you describe. You are using 5 pairs of output transistors. They are loafing even at fairly high power, and the ones you are using have fairly small beta droop at reasonable operating currents. I don't think that is a problem you should be worrying about.
Cheers,
Bob
Hi Alan,
I don't think it is a good idea to use a circuit that someone else posted that you don't understand the workings of. For you to use the CFP bias spreader with confidence at low voltage spread, and being able to adjust its tempco, you should understand how it works and be able to articulate it to us. In other words, you need to do some investigation on your own, both with theory, simulation, and bench measurement. If you then can explain it to us, we will either agree or see problems with either your understanding or with the circuit. This way we all win, and profit from each other's work.
You said that when it is in the one-transistor mode the circuit often "ran away". By that do you mean that you would set the bias to a desired point and then it would dangerously increase? Can you explain why it was running away?
Cheers,
Bob
I think the reason it works is as you adjust TPOT2, you drive the PNP harder and make the PNP take on more and more of the current. So the NPN is passing less and less current. This means the Vbe of the NPN is decreasing as you adjust the TPOT2 up. So the voltage of the spreader goes down because the multiply ratio is fixed.
So if you adjust the TPOT1, you increase the multiply ratio. So you increase the tempco.
This my theory and I think it's brilliant!!! It's so simple it's beautiful.
When I said it ran away, I mean I set the voltage across the emitter resistor to 26mV cold, it slowly increase as it heats up. I look at the supply current, it gone up to over 1.6A after 10minutes.
Tell tale is obvious, when you adjust to Olivers cold, you can see the voltage increases after you turn on, or you can look at the supply current creeping up.
From my experience( very limited), there is no way out of this except looking for transistor with low Vbe for the spreader. Low Vbe allows you to increase the multiply ratio and thereby increase the tempco of the spreader.
We have to live with the fact the big power BJT has low Vbe even at bias current of over 100mA. You might start with 0.6V, I had the MJL gone down to 0.49V during the run away test. You'll never find or should using 1.4V as a guide line for designing. By default, you will not have enough tempco from the spreader adjusting down to 1.2V with a single transistor. I got a spreader NPN with Vbe of 0.58V and still it ran away because there is always a differential in temperature between the power transistor and the spreader no matter how good a thermal coupling you can make. I had to spreader transistor bolted right on top of the power BJT. Yes, it's on the plastic casing, but I believe it is a mile better than if you have the spreader transistor on the heatsink side by side with the power transistor.
There are only so few high power BJT. Namely MJW1302/3281 and 2Sx5200/1943 and NJW0302/0281 that has flat hfe vs current and 30MHz. You can have variant of these two in different package like MJL, 2Sx5242/1962 etc. But really they are in that family. From reading the Is of the model, the 2Sx5200/1943 should have even lower Vbe than MJL I use. I ordered all the above last night to test the Vbe. I am putting the hope on the NJW0281/0302 as ON Semi published the Vbe(on) and it shows it's higher than MJL I use at 100mA.
Last edited:
Hi Mr. Cordell
But when the power BJT conducts more current as the output move away from 0V driving a load, current increases. In turn, base current increases and cause more drop across the base drop resistor of the power BJT. So the voltage across the base stop is modulated by the output voltage. This is like compression and causes odd harmonics.
Hi Alan0354,
On Semi has announced "last buy" on those MJW transistors you want to use. That is sad from my viewpoint as I think they were the nicest BJT outputs I have ever used before.
Second issue. There is an Ebay store selling those transistors. I knew these would be fakes, but I bought one lot anyway for 5 of each transistor. Guess what? They were all defective - and fakes on top of that! About 1/2 were shorted, the rest had open connections. The price was very attractive, and I just hope you aren't buying from Ebay for these.
What marks them as fake parts? Case style is wrong, leads are wrong and the origin and date codes don't match what the data sheet indicates they should be when you compare them with real parts (which I bought through normal distribution).
-Chris
On Semi has announced "last buy" on those MJW transistors you want to use. That is sad from my viewpoint as I think they were the nicest BJT outputs I have ever used before.
Second issue. There is an Ebay store selling those transistors. I knew these would be fakes, but I bought one lot anyway for 5 of each transistor. Guess what? They were all defective - and fakes on top of that! About 1/2 were shorted, the rest had open connections. The price was very attractive, and I just hope you aren't buying from Ebay for these.
What marks them as fake parts? Case style is wrong, leads are wrong and the origin and date codes don't match what the data sheet indicates they should be when you compare them with real parts (which I bought through normal distribution).
-Chris
Thanks
Ha ha, I know better. I only buy from digikey, Newark and Allied. I don't dare to buy these on ebay at all.
Do you mean NJW0281/0302 is being disconnected? I have over 50 ea of the MJL1302/3281. I am trying to replace these, so it does not border me they get discontinued.
When I find a nice transistor, I usually buy larger number of them. I have so many BC546/556, KSC3503/KSA1381, 2SC4793/2SA1930 etc.
I guess I still don't understand the operation of the CFP with respect to altering its tempco while ultimately having the same spreader voltage, even when adjusting both pots. Of course, if you do something to change the multiplier ratio, you will get a corresponding increase in BOTH the spreader voltage and the tempco, so you are back where you started.
It is a little unclear how making the NPN work harder (bear a greater fraction of the VAS current) changes the tempco much (for a given spread voltage) unless you are depending on the tempco of beta of the PNP or the tempco of Vbe of the CFP. If the former, that seems not to be a good idea. If the latter, that needs more thought, but my gut tells me it will likely not have much effect.
With respect to bias runaway, I have never had the problem. As a matter of course, when I set up the bias on a new amplifier I adjust the bias pot to the minimum end, then turn on the amplifier. I do not make a bias adjustment at that time when the amplifier is stone cold. I want a couple of minutes for the pre-driver and driver transistors to warm up, then I turn up the bias to the desired level and wait for a considerable amount of time to see where it will end up. After that, I may trim the bias a little. I then turn the amplifier off for an extended period of time until it is stone cold, then turn it on and watch bias vs. time to see that all is reasonable.
If you are getting runaway when the spreader is in normal one-transistor mode, I'm guessing it is a result of your using quite small values of RE (was it 0.15 ohm you were last at?).
Always make sure that your VAS current is reasonably stable. Also, make sure there are no parasitic oscillations, as they can cause significantly increased output stage current.
Cheers,
Bob
I would think the last statement Bob just made about oscillation is one reason not to go for those Megahertz bandwidths in an amplifier like Alan is trying to do. If you can't reasonably test for a hidden oscillation way up high you could be fighting a problem that you just don't see causing all kinds of problems. Runaway bias settings come to mind here.
Hi Mr. Cordell
Adjusting the TOPT2 balance the current between the NPN and the PNP. The more you make the PNP drives the current, the less the current goes through the NPN and the Vbe of the NPN drops. When the multiply ratio stay constant, the spreader voltage drops. BUT, the tempco of the NPN remains the same.
Then when you adjust the TPOT1 to increase the spreader voltage back, you increase the multiply ratio.......thereby increase the tempco.
Essentially, you are adjusting the Vbe of the NPN using TPOT2. Just like I choose low Vbe transistor and increase the multiply ratio to get more tempco.
Yes, at the time, I use 0.11(two 0.22 in parallel). I did not know half as much at the time. When I have time, I'll go back and redo the experiment again. But that time, when I use the big MJL as the spreader NPN to increase the tempco, and I put the MJL on the transistor in the middle position ( heat up more because it has output transistors on each side) instead of the last transistor that is cooler, I fixed the problem and it was very stable even with 0.11ohm emitter resistor.
I am using 0.22ohm right now and it works perfect with the spreader in the last position. I did see current increase when I adjusted the TPOT2 to 0 to turn off the PNP, I tried two different NPN already. So even with 0.22, the current start running up, I just did not wait how much it will increase as it increases quite fast and I don't need to let it run as hot as before.
I am starting my contract job again
so I am very busy. I am pushing the limits of the TIA circuit. The application engineer of Linear technology invites me to visit them next week because I challenge one of their opamp LTC6268. I'll redo the experiment with adjusting CFP bolt onto the last position again using 0.11ohm. My point is to find a safe way to use 0.11ohm. This is my own research. If I can do it safely, then I'll be one up in the game. I still believe it's about the design and theory. That's why I like your book so much. Getting result is not good enough, you need to know "WHY", you spend the time, you learn and you can be one up in your design. I won't stop just because people said it's stupid, just follow....Which I get a lot here.You actually make me feel guilty yesterday telling me not to just trust the CFP circuit, that I must understand why. yes, because of the contract was coming, I took the lazy way and just tried it. It works and I moved on. I did stop yesterday and think about "why" after I read your post, and I learned. I am pretty sure I am right on this. It is so important to understand and go back to the theory every time encountering a problem instead of just finding a solution and blindly following it. Fixing the problem is not good enough, you have to understand why.
I am surprised you don't agree about using transistor with low Vbe, it really works for me in my limited experience. Vbe is very predictable and within limits, it's repeatable as it's governed by Is. This is not bandage, it's design according to theory that low Vbe force you to use higher multiply ratio and thereby increase tempco.
Alan
Last edited:
I did not test much for oscillation the first go around as I just powered it up. There was no particular tell tale for me to suspect there was. But I since had oscillation after I changed the KSC3503/KSA1381 predrivers to a lower Vbe higher Ccb transistor. I fixed that and the board is very stable now. I worked with a lot of oscillations, it's very obvious even in 100s of MHz circuit. For these kind of low speed circuit, you can tell quite a bit by looking at the square wave pulsing test.
My problem of oscillation is not even simple transistor problems, those are easy to fix and prevent as these are slow transistors. My problem is the 3EF Diamond the creates a +ve feedback loop if you use any transistor with higher Ccb. I since fixed that.
But I will redo the test when I have time.
Alan,
I'm sure you have plenty of test and measurement experience. The fact that board layout or reaching for extreme bandwidth can create a nice RF oscillator or pickup is something that us mere mortals don't always have experience with or the test equipment with a bandwidth that can measure or show that. You can end up chasing your tail not knowing that you even have this type of problem when your O-scope can't measure that high.
I'm sure you have plenty of test and measurement experience. The fact that board layout or reaching for extreme bandwidth can create a nice RF oscillator or pickup is something that us mere mortals don't always have experience with or the test equipment with a bandwidth that can measure or show that. You can end up chasing your tail not knowing that you even have this type of problem when your O-scope can't measure that high.
If you have transistors that fT below 100MHz, a 100MHz+ scope should be able to see it.
The trick I used a lot when I was working with transistor with fT over 1GHz, you cannot see the oscillation with a 400MHz scope, but you can look at the output offset. I called it a pencil or screw driver test. Using a pencil or screw driver and touch around, you'll see the offset jump. It's like touching a raw nerve and the thing jump!!! In the 80s, nobody had 1GHz+ scope, you just adapt. You see enough, you can tell. First thing first is to poke around and look for raw nerve if you don't see oscillation with a slower scope.
If there is no oscillation problems, the circuits in text books really work and it's that easy. What make engineers earn their keeps is the ability of taming oscillations. It's a routine thing to tame oscillation in circuit design!!! I just encountered oscillation in the most unexpected place in my contract work. The voltage regulator enable input cause the regulator to oscillate even if it is a quasi logic level input. It showed up in the sensitive amplifier output and fooled me for an hour before I chased it down. It's a day to day thing!!! This just happened day before yesterday. Data sheet never flag precaution on that!!!
I learned this pencil technique when I was working with LeCroy in the 80s, opamps were slow, we had to design hybrid opamps using BFR90 or 91 RF transistors inside a DIP package. Oscillation was always a problem, I think it was even Walter LeCroy that taught me this trick.
Layout is everything, the war is won or loss on the layout. That's the reason I started with a solid power and ground planes with 0.1uF and 10uF bypass transistor on every single collector of the EF transistor. This technique might not work for GHz transistor as the loss of the capacitor comes into play. But this work every time on the slow transistors. The circuit pcb is designed to be good to over 100MHz. Base stop works.
Last edited:
It is always good to at least start out with a conservative ULGF, like 500kHz. Later, when one gets to know the behavior of the design and understand it and debug it, one can start pushing for higher ULGF. Of course, a conservative ULGF does not prevent local parasitic oscillations, so one still should have a good scope at the output of the amplifier. Minimum 20MHz bandwith, but 100MHz is desirable.
Cheers,
Bob
I have the 465 100MHz. From years of working with it, I know I can see 200MHz even though I have to put it on X10 Magnification and crank up the brightness intensity.
I will test with 0.11ohm emitter resistor again with my tamed board for run away when I have time and room on my bench.
I will test with 0.11ohm emitter resistor again with my tamed board for run away when I have time and room on my bench.
Last edited: