Maintaining a Vbe Multiplier's bias value
in [Design]
|
Prev: Cushman CE-24A
Next: TIA Photodiode Bootstrap at 10MHz
From: Jon Kirwan on 9 Feb 2010 05:39 On Mon, 8 Feb 2010 19:16:24 -0800 (PST), George Herold <ggherold(a)gmail.com> wrote: >><snip> >"I'm wondering about additional topology changes to improve >the performance still more." > >Hi Jon, I've been 'sorta' following your thread on s.e.basics. I >wonder if you abandoned class A operation too early? Why not keep >things linear evreywhere and avoid the �dead band�? So what if you >need a bigger heat sink. It�s certainly a lot simpler. > >George H. Well, George... No, I've not abandoned it. Actually, it's my hope to wind up building the amplifier and then operating it (by hopefully choosing a design where that is possible) in different modes for the learning experience of it. I hope that is in the cards. I really do. But to make a sharp point on it, although it's probably just an extreme case, I remember reading about a 10W amplifier, single channel, dissipating 120W! Creeps me out. So I definitely _want_ to consider other classes of operation. And cripes, I want to learn, anyway. So why not keep my options open? Jon
From: Jon Kirwan on 9 Feb 2010 05:44 On Mon, 8 Feb 2010 18:23:16 -0800 (PST), "miso(a)sushi.com" <miso(a)sushi.com> wrote: ><snip> >Less words and real schematics would get you more readers. [The only >thing worse than ascii equations are ascii schematics.] ASCII is what I'll post. It's the only way to get them archived or properly posted to a text newsgroup. I no longer have access to the binary for schematics, sadly. If I lose some people because they cannot manage fixed-spaced fonts, I guess I lose them. I could place links up on my domain, I suppose. But in this case, the schematics are really very basic and not overly burdensome in ASCII. Besides, Win Hill posted some really nice examples here, before. Folks seemed to live with that. Not sure why you are picking on me, here. >In any event, just google improved vbe multiplier. I've seen all sorts >of circuits published to get lower impedance at the nodes. Okay. I'll do that if folks here aren't interested at all in talking about it. Jon
From: Jon Kirwan on 9 Feb 2010 06:17 On Tue, 9 Feb 2010 00:28:27 -0600, "Tim Williams" <tmoranwms(a)charter.net> wrote: >"Jon Kirwan" <jonk(a)infinitefactors.org> wrote in message >news:okr1n55h5dvjjklg760dllkqq50v7s38ib(a)4ax.com... >> Part of the function of the Vbe multiplier is to also track >> the Vbe requirements for the output stage as it heats up and >> cools down. > >The general idea is to put the Vbe transistor on the same heatsink as the >outputs, if not glued to a transistor directly. > >Unfortunately, for widely mismatched current densities, this doesn't work. >http://webpages.charter.net/dawill/Images/Ampere.gif >In this boringly typical circuit, the 2N3904 Vbe mult. doesn't have enough >tempco to compensate the far beefier (= lower current density??) output >darlingtons. Which makes sense to me. I think I already understood this, generally, if not in intimate detail. One of the reasons I included in the opening salvo, talking about Eg matching. >I was thinking of adding another CCS so a constant voltage drop appears on >the Vbe's base divider resistor. Algebraically subtracting a fairly stable >voltage results in the effective tempco (percentwise) increasing. The base >divider ratio has to be changed to compensate. I need to think about this, more. As you write above, several alternatives appear in my mind and if you wouldn't mind including a short example, I'd appreciate it. >> In this case, I want it to track the output stage so I'm >> going to have to couple it thermally in some useful way. What >> I'm considering, right now, is how to make it immune to >> unregulated supply variations and VAS output voltage swings. > >Don't worry about stability -- as John said, bypass and forget about it. I can't agree, yet. Slow changes, without a crazy-sized cap there, will have the same effective R_ac I'd mentioned before. The cap's Z just won't change it. And I'm not ready to chalk everything up and pile it all onto the global NFB, either -- not because I disagree with you or John, because I can't... I just don't know enough either way. But because this whole post is about _learning_ something. In particular, I was very specific about what I'd like to study right now. Vbe multipliers and various incarnations that may help to deal with current ripple (from 20Hz to 20kHz, I suppose.) I'm wanting to make sure my analysis so far isn't grossly wrongly made, accepting corrections as they arrive, and I'd like to consider interesting ideas, too. Hopefully, my question here on this narrow subject won't be taken as "Well, what does he know about the field of audio amplifier design?" If that's the question to be asked, the answer is easy. "Not much." Cripes, I just started looking at the whole idea about two weeks ago. If I knew enough to ask all the right questions on this topic in barely more than 10 days, I'd likely be headed into being the next Bobby Fischer of audio amps! I'm just a hobbyist, for gosh sake. I found a few circuits on the web that included a collector resistor in the Vbe multiplier and, at first, had no idea why it was there. I asked in .basics and no one else seemed firm about knowing, either. I grew more curious about it and sat down and lo, and behold, the scales fell from my eyes and I could _see_! I could actually see why it was there. Not only why, but how to estimate quantitative values for it and what to expect as a result. It's that sense of discovery that sometimes pushes one further. So I want to understand a little better how one might do even more about compensating vs current variations? In this focus, I don't even need to care about amplifier design, at all. It is purely about the Vbe multiplier right now and nothing else. Sure, audio amplifier design questions caused me to look more closely at this structure. That was my inspiration to set down this short path, right now. But is it wrong to want to explore this area a little more before moving on? >Most of the dynamic VAS/CCS current flows into the output stage, since >that's what it's there for anyway. I think I see that, though I'll see it a lot better later on. Hopefully where I'll be able to put quantities to it. >The capacitor helps turn on the N side / >turn off the P side for rising edges and vice versa. I think I gather that much. It's got very low impedance when the dV/dt is there. >As for PSRR, the CCS's and gobs of feedback keep that in check. Yes, and yes. John L. mentioned this, too, a week ago and more. No question I've got the point, there. The CCS's aren't perfect and where they don't do so well, it gets all nicely lumped into the global NFB and left for it to deal with. I don't mind, though, investigating things just a little more. And I will very soon start taking on the CCS's themselves. I know a few and I know there are a lot more than I'm not even remotely aware of, too. So that is going to be fun. But I'm not one to just borrow and run. I need to _understand_ the mathematics and try my hand at deriving certain features in quantitative ways, not just qualitative ones. I expect to analyze at least four or five different CCS structures before I move on, in what quantitative detail I can manage at the time. >Of course, >in principle you need something to start the CCS's. ICs do this with a JFET >(i.e. current regulating diode) or bandgap reference (e.g., TL431), or >sometimes both, to set a master current, from which everything else is >mirrored. I've seen that done time and time again in ICs. I can remember tracing my fingers from one to another to another as I spent time understanding them better. Last July, in fact, here in this very group, I posted this about the LM334: >: By the way, I just looked at the general schematic for the LM334 on >: National's datasheet and with a quick sweep of my arms came up with a >: design Iset/Ibias of 8, not 16 as they show on page 5. I'm off by a >: factor of two. >: >: My logic went like this. 1/2 of the I from V+ flows via Q6 to the R >: rail. 1/4 via Q4 and 1/4 via Q5. Q5's 1/4*I flows via Q1 to the R >: rail, too. So now up to 3/4*I into the R rail. Q4's 1/4*I passes >: through two paths. The Ic(Q2)=Ic(Q1)/2... but Ic(Q1)=1/4*I, so that >: is 1/8*I, leaving the other 1/8*I for Q3's Vbe conduction, which also >: flows to the R rail. So the R rail gets 7/8*I and the V- picks up >: 1/8*I. Multiplying through by 8 to get rid of the divisor, I see a >: factor of 8 for Iset/Ibias... not 16. >: >: Can someone do a quick description about how to arrive at something >: more like 16? I'm missing a clue (or two.) No one here _did_ fully answer my question, Tim. There it is, and I did try to follow it. I'm aware of the frequent practice, at least. >Most discrete circuits just use a resistor, which is "0%" PSRR, >but it's not all that bad because the currents are balanced (*on average*, >which means you'll see IMD products when it's moving). > >Tim Intermodulation distortion?? I never saw the term IMD before, but that seems as though it must be what you just said. Fits, anyway. Which brings me back to the MC1495, again. And yes, I think I see why you bring it up. Jon
From: Bitrex on 9 Feb 2010 06:49 Jon Kirwan wrote: > On Mon, 08 Feb 2010 22:17:31 -0800, John Larkin > <jjlarkin(a)highNOTlandTHIStechnologyPART.com> wrote: > >> On Mon, 08 Feb 2010 22:11:51 -0800, Jon Kirwan >> <jonk(a)infinitefactors.org> wrote: >> >>> On Mon, 08 Feb 2010 20:49:24 -0800, John Larkin >>> <jjlarkin(a)highNOTlandTHIStechnologyPART.com> wrote: >>> >>>> On Mon, 08 Feb 2010 20:43:03 -0800, Jon Kirwan >>>> <jonk(a)infinitefactors.org> wrote: >>>> >>>>> On Mon, 08 Feb 2010 17:54:13 -0800, John Larkin >>>>> <jjlarkin(a)highNOTlandTHIStechnologyPART.com> wrote: >>>>> >>>>>> Hang a big capacitor across it. >>>>> Nice try. >>>>> >>>>> Jon >>>> No, seriously, that solves a bunch of problems. >>>> >>>> John >>> Which problems does a slew-dependent, C*dV/dt bypass current >>> solve? >>> >>> Jon >> A big cap across the biasing gadget keeps the voltage drop across it >> fairly constant, of course. That nukes some of the problems you >> referred to. More peak current is available to the output bases, for >> example. > > What size cap would help with power supply ripple? Seems the > dV/dt is so small that a fair sized cap would be required to > make any difference. Similarly for low frequency amplified > signal out of the VAS. When you say "big," maybe you mean > it. > > Ban is suggesting global NFB from output back to input. > You've said as much when you say to apply "lots of NFB." I > don't doubt the sincerity of either of you and I'm certain it > will do a lot. But right now I'm interested in seeing what > can be done right on this local subcircuit and at LF as well > as higher frequencies. Unless someone wants to walk me > through the thinking towards the larger concepts. I'm good > either way, as it's the learning that takes place I'm looking > for. But without such guidance, I need to move along at the > pace I can handle while guiding myself. > > Jon Hey Jon, I found a derivation of the input impedance of the two-resistor /transistor Vbe multiplier you might be interested in looking at: http://paginas.fe.up.pt/~fff/eBook/MDA/Mult_Vbe.html For bypassing purposes the rule of thumb I've always heard is to make the impedance of the capacitor 1/10th the value of the impedance looking in to the circuit at the lowest audio frequency.
From: Jim Thompson on 9 Feb 2010 09:49
On Tue, 09 Feb 2010 02:44:31 -0800, Jon Kirwan <jonk(a)infinitefactors.org> wrote: >On Mon, 8 Feb 2010 18:23:16 -0800 (PST), "miso(a)sushi.com" ><miso(a)sushi.com> wrote: > >><snip> >>Less words and real schematics would get you more readers. [The only >>thing worse than ascii equations are ascii schematics.] > >ASCII is what I'll post. It's the only way to get them >archived or properly posted to a text newsgroup. I no longer >have access to the binary for schematics, sadly. If I lose >some people because they cannot manage fixed-spaced fonts, I >guess I lose them. I could place links up on my domain, I >suppose. But in this case, the schematics are really very >basic and not overly burdensome in ASCII. Besides, Win Hill >posted some really nice examples here, before. Folks seemed >to live with that. Not sure why you are picking on me, here. > >>In any event, just google improved vbe multiplier. I've seen all sorts >>of circuits published to get lower impedance at the nodes. > >Okay. I'll do that if folks here aren't interested at all in >talking about it. > >Jon Miso is all blabber and no content... best if ignored ;-) ...Jim Thompson -- | James E.Thompson, CTO | mens | | Analog Innovations, Inc. | et | | Analog/Mixed-Signal ASIC's and Discrete Systems | manus | | Phoenix, Arizona 85048 Skype: Contacts Only | | | Voice:(480)460-2350 Fax: Available upon request | Brass Rat | | E-mail Icon at http://www.analog-innovations.com | 1962 | I love to cook with wine. Sometimes I even put it in the food. |