hitsware said:>Ha, Ha, Ha!!! Very funny!.
I fail to see the humor.
Lateral mosfets do everything
talked about here and with 3 pins.
Not as much transductance (perhaps)
as Vfets, or current capability,
BUT!
What's so funny?
It just struck me funny. What can I say!
Laterals certainly have their place, and do enjoy the advantage of the zero TC point lying at the typical idle bias point, making them very temperature stable. Whether this outweighs the disadvantages of lower transconductance, higher Rdson, lower maximum current, and slower speed is up to the judgment of the designer. Its just a matter of which dragons you wish to slay. Good amplifiers can definitely be made with laterals. It is my impression that they are substantially more expensive, however.
Cheers,
Bob
You know: My name is John and I am an audio designer. Is there a 12 step program for this?
>It is my impression that they are substantially more expensive, >however.
Availibility too. The price is not too bad
if you buy while a major distributor has stock.
I got my last ones @ Avnet for ~ $3 or $4 ea.
(IIRC)
But they don't have them in stock all the time.
FWIW
http://home.comcast.net/~mnjmiller/mosamp.gif
Availibility too. The price is not too bad
if you buy while a major distributor has stock.
I got my last ones @ Avnet for ~ $3 or $4 ea.
(IIRC)
But they don't have them in stock all the time.
FWIW
http://home.comcast.net/~mnjmiller/mosamp.gif
I moved some posts, as requested, to the Biasing/temperature compensation of Thermal Traks thread as requested.
Heres the Thread:
See you there!
Heres the Thread:
See you there!
jcx said:
International Rectifier used to make some of these also. We used them in our first amp, the all-MOSFET V-3. The problem was that the current sensing part was specified to 1% accuracy when the part was running at 10 A or so, but we used them more like 500 mA. At that current the sensing part would only track within about 20%. So we had to individually calibrate each device. IR doesn't make those parts anymore.
Bob Cordell said:
Having designed commercially-successful amps with verticals, laterals, and bipolars, here is my summary:
a) I would never use verticals again. Their Cgs becomes extremely non-linear as the Vds goes to zero (in other words, as a source-follower output stage approaches clipping). This makes for an amp that sounds distinctly worse than when using laterals. Please note that this was in the context of a zero-feedback design. Using feedback (eg, "error correction") will tend to mitigate this problem, but then you still have the sonic penalties that feedback imposes.
b) All of the point Mr. Cordell makes above with regards to laterals are correct. Each individual lateral device costs about $5, so this isn't much more than a vertical device. But to overcome their disadvantages, you need to parallel about 8x as many parts. Then the darn things work great except that now the cost of the output stage is significantly more than a vertical stage.
c) I have found bipolars to be the best solution. They sound the best, don't require the careful matching, and only cost around $2 each in quantity. One of their bad habits, secondary breakdown, is easily avoided by using a bridged output stage. This requires only half the rail voltage and makes secondary breakdown a moot point.
Charles Hansen said:
Hi Charles,
It's great to see you posting here. As someone who has done commercial amps with all three technologies, your perspective and experience is really valuable in this thread.
As previously on this thread we have discussed many of the pros and cons of BJT and MOSFETs (without a lot of resolution, I might add), your specific point about the input capacitance of vertical MOSFETs makes for a good focused discussion that we can zero in on.
Although I won't directly address the issue right now, I do have a question of clarification for you. You specifically mentioned Cgs becoming very nonlinear as Vds goes to zero, which will happen near clipping in a source follower. Did you instead mean to say Cgd? It is my understanding that Cgs changes rather little with Vds, but that it is Cgd that increases fairly steeply as Vgs (actually Vgd) goes to zero.
A related question is what kind of driving impedance were you presenting the vertical MOSFEts with in those designs? Were the MOSFETs fed from low-impedance emitter followers (for example), or were they fed from a higher impedance source (perhaps even directly from a VAS).
Thanks,
Bob
Bob Cordell said:
Yes, of course you are right, I meant Cgd. Sorry for the mixup, I had a bad accident last fall and the pain meds make my brain a little fuzzy. (The Cgs is negligible in a source-follower application, as long as you aren't overloading the output stage.)
The design I was using had a relatively high output impedance drive, around 2.5 kohms. This resulted in excessive high frequency distortion at high powers. For example at 10 kHz, the 10 watt distortion was around 0.25%, but this would climb to over 10x that figure at 100 watts.
We built a prototype that added some source followers as buffers between the gain stage and the output stage. It measured much better, but sounded significantly worse.
This amp used 4 single MOSFET's in a complementary-bridged output configuration. I have posted a schematic for a prototype version of this amp, which you can probably find with the search function. The next design used lateral MOSFET's. I had to parallel 8 of them in each leg for a total of 32 per channel. The measured performance was much better, and so was the sound quality (the two are not always linked, as noted above). But the problem was the cost for parts and labor. We had to hand match 64 output transistors for a stereo amp, and a $5 each (quantity pricing), this was $320 just for the output devices.
In contrast, the latest generation of On-Semi parts are built so well that they don't require matching. They are $2 each, and our 300 watt amp uses 16 devices per channel, so only $64 for the output devices of a stereo pair. Failures with these parts are exceedingly rare, and we only protect them with rail fuses. The emitter resistors are 0.1 ohms, so not much limiting action there either. Best of all is the sound.
The bitch about bipolars is achieving stability with a capacitive load. We use a emitter-follower triple as first described by Bart Locanthi in 1966. Marshall Leach concurred that this was the best output stage, and our own testing confirmed this also. With a combined beta of around 1,000,000, the triple emitter followers allow for less loading on the gain stage than the MOSFET's (if you consider the full audio band). And it is the best sounding output stage I have used.
I'm not going to give away any secrets about our stabilization scheme, but we don't use an output inductor, nor any Zobel networks. It is a do-able thing, but not trivial. I spent two or three months working out that problem.
Charles, it is good to see you contributing. Now that you have beat my socks off with your new design with my own customers, ;-) I hope that you will help to convey what we have learned over the years. This old guy (me) gets inundated with criticism when I do it alone, here. Give me a call some time, and keep Ed Oxner happy by talking to him more often as well. We are trying to get Interfet to do something real for once. Help us out, if you can.
John, you know darned well that I learned (stole?) most of my good tricks from you! If I've done anything beyond that, it is just the old "standing on the shoulders of the giants" thing.
Vishay is selling Siliconix off, and Ed is trying to broker a deal with InterFET. We will see what happens, but in my view a JFET input is mandatory for a high quality audio circuit. And better yet, complementary JFET's.
As I've learned more and more about them (largely from Ed's now out of print books), I'm convinced that the best JFET's were the old NEC J44/K163 pair. Why? Because that is the only JFET that was:
a) Designed for audio.
b) Available as true complements.
c) Able to run at a good idle current (say 5 mA or so) and be at the zero tempco operating point.
This last one is interesting. The Toshiba J103/K246 can run at zero tempco at 0.35 mA or so, but the J74/K170 can't run at a zero tempco at any idle current.
Vishay is selling Siliconix off, and Ed is trying to broker a deal with InterFET. We will see what happens, but in my view a JFET input is mandatory for a high quality audio circuit. And better yet, complementary JFET's.
As I've learned more and more about them (largely from Ed's now out of print books), I'm convinced that the best JFET's were the old NEC J44/K163 pair. Why? Because that is the only JFET that was:
a) Designed for audio.
b) Available as true complements.
c) Able to run at a good idle current (say 5 mA or so) and be at the zero tempco operating point.
This last one is interesting. The Toshiba J103/K246 can run at zero tempco at 0.35 mA or so, but the J74/K170 can't run at a zero tempco at any idle current.
Charles Hansen said:
Hi Charles,
Thanks for the clarification and additional information.
I was very sorry to hear of your accident and hope your recovery is coming well.
I'm not surprized that the MOSFET input capacitance caused degradation with a 2.5K source impedance, but am a little surprized that the introduction of the source followers did not improve the sonics as well as the number. Chalk it up to the X-factor, I guess. I have usually believed that the need for plenty of buffering by the BJTs in front of them sort of forced us to do the right thing in terms of isolating the VAS from the output stage. After all, the nonlinear input capacitance of the BJTs under the same conditions can also be substantial. The vertical MOSFET Cgd capacitance should not be a problem if they are properly buffered, but of course that buffering must not introduce its own sonic problems.
I agree that the triple emitter follower is the way to go with bipolars. This is pretty much the approach I use with my vertical MOSFET designs, but with the power stage being a MOSFET instead of a BJT.
I've been looking at the OnSemi ThermalTrak devices and they look like great parts. And the price is so reasonable.
Whether we agree or not about the effect on the sound, being able to eliminate the output Zobel and the coil, and retain unconditional stability into any reasonable capacitive or inductive load is certainly a yoeman's accomplishment. I imagine not having NFB makes it a little easier, but I'm sure still no walk in the park. I'm not sure it's a "reasonable" load, but can you actually go all the way from 1000 pF to 2 uF right at the terminals of the amplifier?
Cheers,
Bob
Bob Cordell said:
Back when I designed that amp (15 years ago), I was seriously into a "simpler is better" philosophy. The nice thing about the MOSFET output stage was that it allowed a bare minimum of two transistors. I used more transistors than that as follows:
a) Doubled the number to make fully complementary.
b) Doubled the number again to make a fully-balanced bridge.
c) Added a folded cascode stage as a level shifter to allow the elimination of a coupling capacitor.
At the time, I figured that the extra source follower sounded worse because it just added an extra stage. Now I'm not so sure about that and not really sure why it sounded worse. But as I said, the more I learned about vertical devices, the less I liked them. (Again, in the context of a zero-feedback circuit. YMMV.)
Bob Cordell said:
Back when Bongiorno was one of the solid-state amp "gurus", he derided a triple emitter follower because he claimed that each additional junction in the signal path added more distortion. (This was an article in "Audio Amateur" in the '80s.) I've come to believe that he was fooling himself. Some of his configurations were complementary feedback pairs which have lower distortion due to an additional feedback loop, but not because of "one less junction in the signal path".
I'm something of a student of audio history, and I think that Bart Locanthi got it right with what he dubbed the "T-circuit" way back in 1966 for JBL.
Bob Cordell said:
I can't say enough good stuff about them. They grew out of Motorola when they had a "six-nines" reliability goal, and they have always made a consistent product. They copied (licensed?) the Toshiba outputs with the xxx1302/xxx3281 series and these have always been good. Their prices are good, the availability is great, and now with the ThermalTrak series that has added the built-in bias diodes, I don't think there is anything else that compares. On top of that they have improved their process so that the typical beta spread is only 10%! It's hard to match them that close yourself unless you have special temperature controlled equipment. I recommend them to everyone, as they should be rewarded for making a great product for the audio market.
Bob Cordell said:
Yeah, it took me a while to do it. When I first ran into the issue (MOSFET's don't have this problem) I read everything I could on it. I asked everyone I knew about. There's just not a lot of good information out there. The best thing that was written is now out of print, it was a book written by a an ex-Tektronix guy named Dennis Feucht. But even that wasn't too helpful.
After a lot of head-scratching I remembered an old RF technique called "neutralization" that I was able to adapt, and this solved the problem. Now I can hook up any load without oscillation. The worst load tends to be around 5,000 to 10,000 pF, and there is a pretty sharp local maximum. There is still a lot of trial and error involved in getting everything stable, but it's do-able with a lot of patience.
lumanauw said:
Oh, it's just a fancy word we made up for a technique whereby we stabilize the operating point of the gain transistors. (Actually, *all* of the gain happens in the input FET. Everything else is either a level shifter or a buffer.) But the circuit sounds good, and we are able to get good measured distortion. Look at Stereophile's test of the MX-R they have posted online. At 8 ohms and 100 watts, the distortion is only about 0.015%. Not bad for no feedback!
Many of the "guru" here is approaching this. NP in his last Threshold front end, also making the 2nd stage more of a level shifter (contributing little gain). Now you said the above too
I will have to study about this "more of a level shifter 2nd stage". But a very good output stage is a mondatory for this approach, I think.
It's not "not bad", but difficult to achieve
I never hear this amp, but since it is non feedback, I'm curious, how do you describe the bass performance?lumanauw said:
It's better to read what others say than myself. Here is the link to the Stereophile review:
http://stereophile.com/solidpoweramps/407ayre/
Hi-Fi+ also reviewed it and we should have a link to that up on our website in the next few days.
Charles Hansen said:
Hi Charles,
Thanks for your answer. I agree completely with all of your points.
Bongiorno is a pretty sharp guy, so it is surprizing that he may have got it wrong with respect to the T circuit.
I have usually not subscribed to the generalization of simpler is better, but I have also seen examples out there where added complexity has been foolish and possibly sound-degrading. The key, I think, is the "right" degree of complexity for a given objective. Having had a background in linear IC design, I have usually not hesitated to add a transistor if it cleaned up a shortcoming in a another transistor in a straightforward way without introducing a gotcha. The Triple output stage is a good example of such straightforward and generally good circuit additions, in my opinion. Heck, let's face it, bypassing an electrolytic capacitor is also an example of increased complexity when you come right down to it (if it is a given that one needed the electrolytic in the first place, of course).
As I said, I think the use of the Triple concept with vertical MOSFETs completely eliminates the conern about the Cgd (at least to the analogous point that it exists for bipolars as well in the same architecture), but I am very interested in learning of your other concerns with vertical MOSFETs that you have developed over the years, since discussion of those would be extremely relevant to this thread.
Although my designs tend to use global negative feedback, I think that any concerns that you developed in regard to vertical MOSFETs in your non-feedback architectures would in most cases be just as relevant to NFB amplifiers, since the goal for NFB amplifiers (at least the one that should always be practiced) is that the open loop should be as linear as possible BEFORE closing the loop.
Cheers,
Bob