Hmm.. isn't he using a 230Vac transformer? This will produce about 325Vdc which will appear across the transistor when powering up into a capacitive load, so he certainly does need a 400V transistor.
The optimum current for the 1N5388 at 200V is around 5mA. This value can be found in the datasheets as it is the actual test current the other values are based on. Goldenbeer is right about R2 being too large, but 1mA doesn't cut it.
Another improvement could be the use of a constant current source (set at 5mA) instead of a resistor to feed the zener.
Edit: Maybe this has been answered already, but it has become a rather long thread 😉. Why the 200V output when using a transformer with 325 secondaries?? IME, it's easier to go along with what you got instead of fighting it. A 125V drop is pretty substantial. Even at low currents.
Another improvement could be the use of a constant current source (set at 5mA) instead of a resistor to feed the zener.
Edit: Maybe this has been answered already, but it has become a rather long thread 😉. Why the 200V output when using a transformer with 325 secondaries?? IME, it's easier to go along with what you got instead of fighting it. A 125V drop is pretty substantial. Even at low currents.
Last edited:
+1 Always use an over-spec'd pass device with a voltage rating comfortably above the full primary voltage. I've had fireworks in my amps because I used components that were only suited for 'stable' conditions. You could consider MOSFETs too. Easier to find high voltage, high current types. And current models have very low Rson and Cgs.
Sigh… Can you cite the gm for any bipolar transistor then, in terms that somehow manage to avoid using hFE, or its hypercritical dependence on Vthresh?
MerlinB, you ol' sorcerer, get out the wand and wiggle out of that. I've been working at this for 40 years, and I'm still waiting for this to be competently answered without resorting to Inductive Magic. 🙄
And IC … doesn't depend on hFE, which is IC/IB … where IB is right back to the e((VBE - Vthresh)/kT) equation, again?
Your original point was that hFE is near-useless 'cuz it varies so much with junction temperature and so on. Yet compared to the even less predictable gM, …
I'll let it go: you can call voltage amplification universal if you like, but it is a stretch of the imagination for bipolar transistors. They are both in fact and mathematically current-multiplier devices, which like E=IR and I=E/R can be converted to a voltage expression. Yet, when designing, if one actually sits down with a pad of paper and a calculator, the capricious gm of a bipolar is completely tamed into something we non-iterative mortals can compute with hFE. And yes … device to device, there used to be a fairly wide range of hFE. Just like there is a substantial variability of gM for a random handful of ECC–83 or KT–88's. We match 'em for a reason. And we design based on their 'spec mean values'. And designs are supposed to accommodate nominal device-to-device variability. And temperature coefficients (bipolar). And different manufacturer's sloppiness.
GoatGuy
Your original point was that hFE is near-useless 'cuz it varies so much with junction temperature and so on. Yet compared to the even less predictable gM, …
I'll let it go: you can call voltage amplification universal if you like, but it is a stretch of the imagination for bipolar transistors. They are both in fact and mathematically current-multiplier devices, which like E=IR and I=E/R can be converted to a voltage expression. Yet, when designing, if one actually sits down with a pad of paper and a calculator, the capricious gm of a bipolar is completely tamed into something we non-iterative mortals can compute with hFE. And yes … device to device, there used to be a fairly wide range of hFE. Just like there is a substantial variability of gM for a random handful of ECC–83 or KT–88's. We match 'em for a reason. And we design based on their 'spec mean values'. And designs are supposed to accommodate nominal device-to-device variability. And temperature coefficients (bipolar). And different manufacturer's sloppiness.
GoatGuy
I don't follow. You know what Ic is, because you design it. You can therefore calculate gm with a high degree of accuracy.
If you happen to know the hFE (good luck) then you can also calculate the base current. Plug in a different transistor and you will still have the same gm at the same Ic, but you probably won't have the same hFE and base current! That's why designers try to make circuits that are immune to wide variations in hFE.
@Merlinb, funk1980
You are right, forget about the circuit in post #40. It doesn't cover all fault cases, or power up&down into capacitive loads or interruptions.
I messed around with some possible error protections, but they will always require either some high powered resistors or zeners, or both, or 400V power Darlingtons, which are expensive and have junk performance.
I figured it's not worth to persue this path, when all can be had with a TL783C, better, and cheaper.
You are right, forget about the circuit in post #40. It doesn't cover all fault cases, or power up&down into capacitive loads or interruptions.
I messed around with some possible error protections, but they will always require either some high powered resistors or zeners, or both, or 400V power Darlingtons, which are expensive and have junk performance.
I figured it's not worth to persue this path, when all can be had with a TL783C, better, and cheaper.
Mmm… not true, especially on the first cycle: the 2.2 μF capacitor across the Z1 zener will not charge up fast enough thru the 270 kΩ trickle resistor to prevent Q1 from seeing a sizeable fraction of VREC.
For safety, the transistor should be full-rated. No downside.
GoatGuy
I wouldn't use a TL783. Yes it can be floated, but the max in-out differential is 125. That's the amount the OP wants to drop. Bad idea to use a regulator at its max ratings. Next the thermal ratings. Let take an example of 50mA it has to supply. That's 6,25W dissipation. Thermal resistance of the though-hole 783 is around 3C/W. A reasonable sized heatsink (10C/W) + thermalpad/grease (0.5C/W). Added all up makes for around 84 degrees celsius. That's above ambient. Inside a tube amp case, 35 degrees is common. So 35 + 84 = 119 degrees. Waaaaay too hot.
I'm growing amused here… So MerlinB you've got an NPN, you've designed it to run at 10 mA = IC, and your gM is 0.010/0.025 = 0.4 S or 400 mS (ma/V). What's your next step? How would you propose biasing said transistor? How does this relate to the highly variable IC case of an emitter follower circuit? How does any variation at all of IC not alter gM to the point of being a near-useless characteristic for bipolar transistors?
Really, dude… you may not like hFE, but it is the gain-abstraction that a designer inevitably comes to count on, within a range of values, from a range of manufacturers, for realizing real-world and not theoretical circuits.
I'm patient. I'm hopeful!
GoatGuy
Last edited:
Voltage rating of transistor - I think I need 400V because I can potentially see 353Vce when the final capacitor is charging, there is no load on the output, and line voltage is 125VAC. I see goldenbeers' design has only a very small capacitor (100nF) between ground and the emitter. Does keeping the charge time super-short like that mitigate the need to use the higher-rated part? I'm not conversant with silicon failure modes, so I just try to engineer everything with ample headroom. With a 100nF value, I would probably choose either ceramic or film type, so I don't need to worry about future-failure, but I'm curious also what would happen if I used an electrolytic in this position and it started leaking 20 years hence. (I work on a lot of 50-75 year old amps, so these questions cross my mind frequently).
Funk1980 - 200V was selected as the target voltage because that's what the manufacturer suggested for that amplifier. It is well-known that these amps will run at much higher voltages, but there are no benefits to that (except easier supply design?) and I suspect that higher voltages will encourage the tubes and electrolytic capacitors in the amplifier to expire more quickly. The transformer was selected based largely on cost and availability at Digikey [flat rate shipping to Canada FTW]. There seems to be a dearth of transformers in the 80V or 160V range. My earliest design was based on an 80VAC transformer feeding a voltage doubler, but then I noticed the transformer cost $38...The 230VAC transformer at $18 (CAD) is much more palatable.
Goldenbeers suggested that R1 (1K) in my design is too big. I must admit, I basically chose this value at random, thinking it would be nice to get ripple SNR down to 30dB before rectification. The C1 capacitor value was chosen largely based on the intersection of the size:cost curve. Now I wonder, though, if C1 should not be smaller. I suppose if I were smart, I would look at R1 and C1 and think about minimizing inrush load that the transformer sees. I wonder why the datasheet doesn't say anything about over-current amount and time? A slow-start for this amplifier is not necessary IMO but it couldn't hurt to take a half-second or more to charge that first capacitor.. I need to dig up the equations and play with some numbers, I guess.
I'm also not clear on the reason for the suggestion of four zeners instead of one. Cost-wise, it's exactly the same, down to the penny. Is it to try and keep the junction temperature more stable? Voltage accuracy will probably be better, assuming a truly random distribution throughout the tolerance allowance. I'm not *really* worried about this. If I decide that I need voltage accuracy, I think I would try replacing the zener with an LR8.
The big cap at the output, I added this for luck, and because it always seems you can never have too much capacitance in an amplifier power supply. But this is nothing like any other amplifier I have studied before....usually we are worried about being able to reproduce bass pulses with an under-sized transformer (so that peak amplifier power can briefly exceed the transformer's capability), but that's not the case at all with this amplifier. One thing that this supply should be able to handle is an instant change in load by 10mA or so without sagging the B+. I guess, given how much ripple it can eliminate, the transistor stage is able to do this already?
What is the benefit of using two transistors and a diode in place of a Darlington pair?
I understand the sizing considerations for R2 now, BTW. Thanks for explaining that.
How were sizes derived for C1, C2 in Goldenbeers' supply? Can somebody point me at the math to figure out how big the voltage ripple will be after R1? I have been using a calculator that gives me figures for RC and LC filters, but when I try to treat the CRC filter as two RC filters with one R zero ohms, it tells me there is no ripple attenuation, which I don't believe. Was C1 in this design chosen to reduce the load on the transformer while C2 charges? Also, why so much capacitance around the zeners? Surely we don't need 22uF to stabilize the mV fluctuations?
Wes
Funk1980 - 200V was selected as the target voltage because that's what the manufacturer suggested for that amplifier. It is well-known that these amps will run at much higher voltages, but there are no benefits to that (except easier supply design?) and I suspect that higher voltages will encourage the tubes and electrolytic capacitors in the amplifier to expire more quickly. The transformer was selected based largely on cost and availability at Digikey [flat rate shipping to Canada FTW]. There seems to be a dearth of transformers in the 80V or 160V range. My earliest design was based on an 80VAC transformer feeding a voltage doubler, but then I noticed the transformer cost $38...The 230VAC transformer at $18 (CAD) is much more palatable.
Goldenbeers suggested that R1 (1K) in my design is too big. I must admit, I basically chose this value at random, thinking it would be nice to get ripple SNR down to 30dB before rectification. The C1 capacitor value was chosen largely based on the intersection of the size:cost curve. Now I wonder, though, if C1 should not be smaller. I suppose if I were smart, I would look at R1 and C1 and think about minimizing inrush load that the transformer sees. I wonder why the datasheet doesn't say anything about over-current amount and time? A slow-start for this amplifier is not necessary IMO but it couldn't hurt to take a half-second or more to charge that first capacitor.. I need to dig up the equations and play with some numbers, I guess.
I'm also not clear on the reason for the suggestion of four zeners instead of one. Cost-wise, it's exactly the same, down to the penny. Is it to try and keep the junction temperature more stable? Voltage accuracy will probably be better, assuming a truly random distribution throughout the tolerance allowance. I'm not *really* worried about this. If I decide that I need voltage accuracy, I think I would try replacing the zener with an LR8.
The big cap at the output, I added this for luck, and because it always seems you can never have too much capacitance in an amplifier power supply. But this is nothing like any other amplifier I have studied before....usually we are worried about being able to reproduce bass pulses with an under-sized transformer (so that peak amplifier power can briefly exceed the transformer's capability), but that's not the case at all with this amplifier. One thing that this supply should be able to handle is an instant change in load by 10mA or so without sagging the B+. I guess, given how much ripple it can eliminate, the transistor stage is able to do this already?
What is the benefit of using two transistors and a diode in place of a Darlington pair?
I understand the sizing considerations for R2 now, BTW. Thanks for explaining that.
How were sizes derived for C1, C2 in Goldenbeers' supply? Can somebody point me at the math to figure out how big the voltage ripple will be after R1? I have been using a calculator that gives me figures for RC and LC filters, but when I try to treat the CRC filter as two RC filters with one R zero ohms, it tells me there is no ripple attenuation, which I don't believe. Was C1 in this design chosen to reduce the load on the transformer while C2 charges? Also, why so much capacitance around the zeners? Surely we don't need 22uF to stabilize the mV fluctuations?
Wes
@Wes, please forget the design from post #40. It is not stable.
I would edit the post to remove it, but no longer can do so.
@funk1980
Zener D5 (with R1,2,3) protects the TL783c against overvoltage. Although you should replace D3 with a 100V power zener as well if you're starting up into big capacitive loads.
I would edit the post to remove it, but no longer can do so.
@funk1980
Zener D5 (with R1,2,3) protects the TL783c against overvoltage. Although you should replace D3 with a 100V power zener as well if you're starting up into big capacitive loads.
I don't see what is funny. Merlin is right. Collector current in any competently-designed emitter follower circuit is almost independent of beta. The tiny variation which is left will vary gm, of course, but all that does is produce a tiny variation in gain and a tiny variation in output impedance. Beta affects input impedance, but in most cases this just needs to be 'high enough'.GoatGuy said:
You need to know which ballpark beta is in when designing the bias network.
DF96, MerlinB…
I accept that you're wed to the gM being an inviolate parameter that describes bipolars as well as FETs and tubes and everything else. Yet, when I present a simple (and it is simple) example / challenge case… both of you have waved your hands about "competently designed", without turning "competent" into numbers.
Take the leap, lads.
Do so numbers.
Show your work.
Show that that which you hold as true, IS true.
I'm not trying to be a jêrk, or troll! Really! I'm just a Missouri kind of guy, who is asking to be shown how hFE isn't materially useful in competently designing a transistor amplifying stage.
Thing is, DF96 and MerlinB - I've been reading your comments for years, and I have come to really trust your opinions. They're solid, they're well considered, and you're usually spot-on. So… again, I'm not trolling.
GoatGuy
I accept that you're wed to the gM being an inviolate parameter that describes bipolars as well as FETs and tubes and everything else. Yet, when I present a simple (and it is simple) example / challenge case… both of you have waved your hands about "competently designed", without turning "competent" into numbers.
Take the leap, lads.
Do so numbers.
Show your work.
Show that that which you hold as true, IS true.
I'm not trying to be a jêrk, or troll! Really! I'm just a Missouri kind of guy, who is asking to be shown how hFE isn't materially useful in competently designing a transistor amplifying stage.
Thing is, DF96 and MerlinB - I've been reading your comments for years, and I have come to really trust your opinions. They're solid, they're well considered, and you're usually spot-on. So… again, I'm not trolling.
GoatGuy
Wow, I spent about 4 hours writing that last post (busy morning), lots of posts since I started. 😀
I'm satisfied now that I should definitely keep Q1's Vce big enough to handle the full possible 353V.
I tried following the above discussion but it's a bit over my head at this point. At least I learned that gm and transconductance are the same thing. And now I know what transconductance IS. Fewer mysteries! I also finally grokked what biasing IS while reading about transistors yesterday..this might be a simple power supply project, but I am really excited that I am learning a bunch of related things while I'm at it.
One thing I have not been able to "pull out" from the discussion is what parameters I really need to consider when selecting the voltage follower Darlington/transistor(s)/FETs/whatever.
I know I need to select a Vce with some headroom above 353V, and that I want something that can dissipate a fair amount of power. Is the power that it "throws away" what I need to look at, or the power used by the amp? What do I need to look at in a datasheet to insure that the circuit will actually work, and within reasonable safety tolerances?
For example, the Sanken 2SD2141 looks like a good choice from a casual perusal of the Digikey parametric search. http://www.semicon.sanken-ele.co.jp/sk_content/2sd2141_ds_en.pdf
But, I look at the data sheet and am left with big question marks. Years ago, when building something like this, I would simply build the circuit and rescue as many big transistors from broken PC power supplies (etc) as I could. Then I would put one in the circuit, run it with a large resistive load, and see if it worked, and if the transistor burnt up or got really hot. If I could keep my finger on the transistor for 10 seconds, I deemed it "good enough". I'm hoping to design smarter these days. 😀
If I look at the Pc-Ta graph, assume I have a large heat sink and a 35C ambient temp, it looks like my dissipation (Pc) is 10W. Do I calculate Pc by using the load of the amplifier as the current, and the voltage as Vce? So, 325V regulated to 200V gives 125Vce and a draw of 30mA on the amp gives P = (0.03 * 125)W = 3.75W -- well under 10W.
Looking at the Pc-Ta derating graph, I infer that the numbers in the safe operating area graph can be multiplied by 5 or so when using a large heatsink. Fair assumption? Next we look for the safe operating area for DC at Vce=125V, which is about 0.1A. Multiplied by 5, that gives 0.5A - lots of headroom over my max amplifier current (I am using 50mA for that, which has lots of headroom built in also).
So, I am okay for safe operating current during steady state. Do I need to take a look, then, at how long C3 will charge, keeping the current limiting effect of R1 in mind, and make sure that the single pulse current is also not exceeded? (Not quite sure how to calculate that)
The next question becomes - is there enough gain for my output current requirements? This is quite tied to Ic (collector current) which I'm not sure how to calculate? I see that Ic = hFE * Ib. But hFE is dependent on Ic!
I'm satisfied now that I should definitely keep Q1's Vce big enough to handle the full possible 353V.
I tried following the above discussion but it's a bit over my head at this point. At least I learned that gm and transconductance are the same thing. And now I know what transconductance IS. Fewer mysteries! I also finally grokked what biasing IS while reading about transistors yesterday..this might be a simple power supply project, but I am really excited that I am learning a bunch of related things while I'm at it.
One thing I have not been able to "pull out" from the discussion is what parameters I really need to consider when selecting the voltage follower Darlington/transistor(s)/FETs/whatever.
I know I need to select a Vce with some headroom above 353V, and that I want something that can dissipate a fair amount of power. Is the power that it "throws away" what I need to look at, or the power used by the amp? What do I need to look at in a datasheet to insure that the circuit will actually work, and within reasonable safety tolerances?
For example, the Sanken 2SD2141 looks like a good choice from a casual perusal of the Digikey parametric search. http://www.semicon.sanken-ele.co.jp/sk_content/2sd2141_ds_en.pdf
But, I look at the data sheet and am left with big question marks. Years ago, when building something like this, I would simply build the circuit and rescue as many big transistors from broken PC power supplies (etc) as I could. Then I would put one in the circuit, run it with a large resistive load, and see if it worked, and if the transistor burnt up or got really hot. If I could keep my finger on the transistor for 10 seconds, I deemed it "good enough". I'm hoping to design smarter these days. 😀
If I look at the Pc-Ta graph, assume I have a large heat sink and a 35C ambient temp, it looks like my dissipation (Pc) is 10W. Do I calculate Pc by using the load of the amplifier as the current, and the voltage as Vce? So, 325V regulated to 200V gives 125Vce and a draw of 30mA on the amp gives P = (0.03 * 125)W = 3.75W -- well under 10W.
Looking at the Pc-Ta derating graph, I infer that the numbers in the safe operating area graph can be multiplied by 5 or so when using a large heatsink. Fair assumption? Next we look for the safe operating area for DC at Vce=125V, which is about 0.1A. Multiplied by 5, that gives 0.5A - lots of headroom over my max amplifier current (I am using 50mA for that, which has lots of headroom built in also).
So, I am okay for safe operating current during steady state. Do I need to take a look, then, at how long C3 will charge, keeping the current limiting effect of R1 in mind, and make sure that the single pulse current is also not exceeded? (Not quite sure how to calculate that)
The next question becomes - is there enough gain for my output current requirements? This is quite tied to Ic (collector current) which I'm not sure how to calculate? I see that Ic = hFE * Ib. But hFE is dependent on Ic!
Yes it varies, but then everything varies. The point about gm is that its variation is highly predictable, and with quite a simple formula too. Unlike hFE which varies with just about everything and isn't easy to predict. The best you can hope for is that it is 'high enough'. If you can't control something, you engineer it to be insignificant. When I buy transistors I buy them with a guarenteed minimum hFE and I design the circuit to be immune to that limiting defect; I don't go looking for an hFE of 'precisely 100 to 101' to suit my tweaky circuit.
Last edited:
Choose an emitter resistor, say 1k. Ideal voltage across it is therefore 10V. Base will be one Vbe drop above that, so apply 10.6V from a sourse which is more than capable of supplying the pesky base current. Choose a transistor with a guarenteed high-hFE so you know the base current will indeed be much smaller than the collector current. How much smaller? You don't care as long as it is less than one-hundreth of Ic, so you know it won't cause more than 1% error in Ie.
Last edited:
OK, I'll give up: you did not use gM nor its derivation gM = IC/Vthermal, which was the original premise.
I'm sorry, MerlinB, but in my pretty long experience as well as from grad-school semiconductor theory (and ridiculous amounts of lab time) … in bipolar transistors, gM is hugely variable, and over remarkable scales of transconductance, hFE is nearly a constant. So much so, that it allows accurate "napkin designs" that actually work, as designed. Maybe in this pSpice enhanced world, that's no longer much of a laudable skill. But I'm old, and I like being able to take a pad of paper, a pencil and a calculator (I've upgraded to a spreadsheet - recomps are so much easier!), and get the values spot-on straight away.
I can't do that with the gM that you recommend. I can, trivially, from the hFE parameter.
So, I propose we close this with "we agree to disagree", and remain on friendly terms. Agreed?
GoatGuy
I think you're mixing up the idea of a transistor being a voltage-controlled transconductance device on the physical level, with the practical aspects of circuit design. Yes, you need to know something about hFE to design a biasing circuit; like any component defect, you have to be aware of it and know how to deal with it. You use gm to calculate things like gain and output impedance.
- Status
- Not open for further replies.