Error feedforward as used in the Quad 405, actually has one more signifficant repercussion, in a classic 3-stage amp - it (theoretically) reduces it to a 2 stage amp, so stability is much easyer to achieve. I've ctually built a few amps using this principle, and they normally don't evenneed a miller dominant pole cap.
@Wavebourn
I believe there were several topics on this forum where you compared your 'linear approximation' output stage with taht wof the Quad 405 - and in all of them it was mentioned that this is simply not the same and not directly comparable. Quad uses a bridge to do what janneman explains in his post above, that is, by means of error feedforward attemot to COMPLETELY eliminate (or rather null out) any output stage nonlinearity. Yours does not.
This is the main difference between error correction/cancellation and feedback - EC tries to subtract an EXACTLY oposing error, so in theory, get zero error. Feedback only reduces error by the ratio of open and closed loop gain. The former is subtraction, while the latter is division. The former can at least in theory be trimmed for ideal cancelation, while the later can only produce ideal cancelation if OLG is infinite. This is a rather fundamental difference, so it strikes me as odd that you need to be repeatedly reminded of it.
@Wavebourn
I believe there were several topics on this forum where you compared your 'linear approximation' output stage with taht wof the Quad 405 - and in all of them it was mentioned that this is simply not the same and not directly comparable. Quad uses a bridge to do what janneman explains in his post above, that is, by means of error feedforward attemot to COMPLETELY eliminate (or rather null out) any output stage nonlinearity. Yours does not.
This is the main difference between error correction/cancellation and feedback - EC tries to subtract an EXACTLY oposing error, so in theory, get zero error. Feedback only reduces error by the ratio of open and closed loop gain. The former is subtraction, while the latter is division. The former can at least in theory be trimmed for ideal cancelation, while the later can only produce ideal cancelation if OLG is infinite. This is a rather fundamental difference, so it strikes me as odd that you need to be repeatedly reminded of it.
ilimzn;
Approximation means that parts of function smoothly continue each other. All parts of the amp working on load have the same gain because of different dynamic output resistances and deepth of feedbacks. Otherwise it would be approximation of non-linear transfer function.
You may, however, view the amp under the angle of "bridge" that consists of output resistances and resistances in feedback, if it is convenient for you. But when I asked you the last time to show where is the bridge in Quad you pointed on the frequency compensation network. 😉
You may find the bridge in my amp, if you want. But it is 3-dimensional, because I use 3-step approximation. 😎
Approximation means that parts of function smoothly continue each other. All parts of the amp working on load have the same gain because of different dynamic output resistances and deepth of feedbacks. Otherwise it would be approximation of non-linear transfer function.
You may, however, view the amp under the angle of "bridge" that consists of output resistances and resistances in feedback, if it is convenient for you. But when I asked you the last time to show where is the bridge in Quad you pointed on the frequency compensation network. 😉
You may find the bridge in my amp, if you want. But it is 3-dimensional, because I use 3-step approximation. 😎
And you know, why 3-step approximation?
Just a coincedence. I've found that modern power BJTs designed for audio work well on currents up to 3 A only. Also, their b-e junctions break-up on voltages a bit more than needed to fully bias some HEXFETs for 30A current. As the result, I made a power amp of very high sonic quality using very few active elements, thanks to coincedence.
Just a coincedence. I've found that modern power BJTs designed for audio work well on currents up to 3 A only. Also, their b-e junctions break-up on voltages a bit more than needed to fully bias some HEXFETs for 30A current. As the result, I made a power amp of very high sonic quality using very few active elements, thanks to coincedence.
Sliding Plateau Bias in modern class A amps also offers some power savings without crossover distortion while using the best, most proven circuit topologies.
The latest Krell Class A amps use a sliding plateau bias which allows the amp to idle at a very low bias, and bumps up the Class A bias upon demand until that demand gets reduced for several seconds.
I use a simple physical hi/lo Class-A bias front panel switch on my amps.
The latest Krell Class A amps use a sliding plateau bias which allows the amp to idle at a very low bias, and bumps up the Class A bias upon demand until that demand gets reduced for several seconds.
I use a simple physical hi/lo Class-A bias front panel switch on my amps.
Nice discussion here about the 405, I made a replica in 1985. But the the bottom line is: does it sound? NO!
GRG
GRG
LineSource said:
I don't like tools that decide instead of me what do I want, that's why my class A amps always stay well biased and have a load impadance switch on the rear panel. The advantage of class A amps is low distortions on low power when they are most audible. That's why A+C approximation amps sound so natural, they switch faster than amps with adaptive bias.
Traditional amps with abused symmetrical emitter followers always generate crossover distortions due to rectification of the signal on base-emitter junctions of output emitter followers, especially working on real complex loads. High current biasing and deep feedbacks only push this distortions to be faster, on shorter period of time, on lower powers, but sensitivity of human perception depends on sound power, so such approach is contradictory to our perceptions. Asymmetrical amplifiers sound better exactly because they distort more on higher power when our perception is less sensitive to distortions.
Also, if you analyze "natural" distortions you may hear thast the more distortions we hear, the wider is their spector. Take brass pipes, strings, human voice, everything complies with this natural law! The more volume, the more distortions.
Tahe the signal with constant main frequency level and slowly increase it's power. Subjective volume changes a bit. Take the same signal and gradually add harmonics, and subjective volume changes a lot. Remove even harmonics, and you get very loud, abused sound. Clarinet sounds shrilly. It has no odd harmonics, its spector is wide. But it is still sounds natural. Now, leave only clarinet type distortions, but of very high order, and your perception can't understand, is the sound loud or quiet (audible high order harmonics, but no corresponding low order distortions), is the source of the sound abused (symmetrical distortions) or not (no low order harmonics).
Symmetrical amplifier with more distortions on high end of spectrum and with more distortions on lower power is the biggest mistake of modern engineers that is repeated and repeated and repeated....
...A hungry monkey sees a banana tree. The monkey shakes a tree, bananas do not fall. The internal voice says: " Sit down, think ". She took a stick, brought down bananas, had a meal.
The engineer shakes a tree. An internal voice: " Sit down, think ". The engineer: " It is necessary to shake more strongly! "
...the more strongly are abused symmetrical emitter followers, the more we are hungry for a pleasant sound. It is simple...
Wavebourn said:
Maybe you haven't realised yet that the C and L in the dumper are not only the bridge but at the same time the compensation components. You really should read up on the original article. It can only increase your respect for the designers that came up with several innovations in one fell swoop.
Jan Didden
LineSource said:
Sliding bias means that the amp changes operation conditions with the signal. I cannot understand why this is good except for lower dissipation. But that can be done with optimal class AB, with better sonics IMHO.
Jan Didden
Wavebourn said:
I certainly did not, you should go back and check it again. or, better still, use the search on this forum to locate a link to Pete Walkers original papers and some followups, where it is clearly shown.
Do not mistake an inductor and a capacitor for frequency compensation networks (although, due to the way the bridge was implemented, this function is somewhat inherent) - in the Quad 405 embodyment of Walkers invention, two legs of thebridge use reactive components, in order to prevent power loss due to resistive drops in the bridge. There is a trade-off with this approach, however - and IIRC I also mentionedit, which is, the amp output impedance gets a potentiallly signifficant inductive component. Based on some simulation and experiments i did just recently, i think this might be responsible for most of the negative that was said about how the 405 sounds. The particular inductor used in the whole Quad current dumping series of amps limits output slew rate the more, the lower the load impedance, so much so I would not feel confortable driving 4 ohm loads. Either way, it is entirely possible to implement the bridge using resistive components, and in fact, one of the papers demonstrates this, and I myself used it on a small prototype too.
Hi Wavebourn,
I've found the gate charge issues to be complicated to eliminate without resorting to a power BJT stage to eliminate with mosfets. In this regard, BJTs can be easier to successfully design with.
-Chris
I am somewhat lost on this. I don't follow what you are saying at all. For instance, what do you mean by this?
Are you referring to SOA here?
If so, what are the power limitations of the mosfet under the same conditions?
I've found the gate charge issues to be complicated to eliminate without resorting to a power BJT stage to eliminate with mosfets. In this regard, BJTs can be easier to successfully design with.
-Chris
janneman said:
I highly respect designers who have ideas and share them with me. 🙂
All components are parts of all imaginable bridges. Dependence of transfer function of output transistors on currents, voltages, frequencies, are as well. 😉
janneman said:
Also, it is done with my notebook computer. 😉
Anatech, I am not resorting to BJTs. I just use them in a peaceful coexistance.
Initially, I used 10 pairs of BJT transistors for 30A current. Later I found that I can get the same results using only 2 BJTs and 2 FETs.
To clarify what you can't understand, please look at curves of b-e junctions of typical power transistors. They can keep well with voltages needed to bias a FET on the opposite rail. Zeners on the schematic protect b-e junctions against too much voltage, also they limit maximal output current if the amp's output is shorted. I did not try, probably B-E junctions will survive the same current as slower Zeners, but since maximum current of broken-down reverse biased transistors is not specified by Toshiba, I decided do not risk.
So, emitter resistors are selected such way so starting from 3A FETs have enough bias to start working. Slowly, gradually, without harsh high order distortions. Speaking of input capacitances, what is 10,000 picofarad against 3A current, especially if this capascitance is between output of a class A amplifier and any possible capacitance ot the load?
Initially, I used 10 pairs of BJT transistors for 30A current. Later I found that I can get the same results using only 2 BJTs and 2 FETs.
To clarify what you can't understand, please look at curves of b-e junctions of typical power transistors. They can keep well with voltages needed to bias a FET on the opposite rail. Zeners on the schematic protect b-e junctions against too much voltage, also they limit maximal output current if the amp's output is shorted. I did not try, probably B-E junctions will survive the same current as slower Zeners, but since maximum current of broken-down reverse biased transistors is not specified by Toshiba, I decided do not risk.
So, emitter resistors are selected such way so starting from 3A FETs have enough bias to start working. Slowly, gradually, without harsh high order distortions. Speaking of input capacitances, what is 10,000 picofarad against 3A current, especially if this capascitance is between output of a class A amplifier and any possible capacitance ot the load?
Wavebourn said:
Sorry but you lost me (and others apparently). I'm not joking, seriously. What is the "b-e keep-up". Are you talking about reverse Vbe, or forward Vbe? What zeners are you referring to, the ones to protect the fet gates? You don't need those on bjt's right?
Jan Didden
janneman said:
Yiu are right, I mean that BJTs see reverse Vbe equal to needed to bias FETs in opposite rails. Zeners do not protect gates, they limit output current by limiting input voltage (Yes, they do actually, but it is a side-effect). To protect driver in such conditions I used diodes that you may see on the diagram.
B-E junctions may be viewed as zeners, but currents through such Zeners are not specified by manufacturers, also exact voltages of break-down are not specified at all, however maximal reverse voltages are, and they are no less than needed to bias some HEXFETs.
I hope it helps.
Wavebourn said:
Zeners for the protection of the gates of fets are much to high in avalanche value to limit input voltages. The fets are hard on much, much earlier than any zener action starts.
Wavebourn said:
Not really...
Jan Didden
janneman said:
Now I am not with you....
Requirements to Zeners to limit G-S voltage to limit current are much strict then to limit voltage on level less that needed to damage the gate.
I have many blown up 20A fuses and damaged screwdriver, but all FETs survived tests. 😉
Hi Wavebourn,
It seems to me that your findings depend heavily on the circuit and devices used. I am not adverse to using a mosfet where it's advantages are clear. The same goes for a BJT (I'm also very familiar with the characteristic curves).
However, this thread is a discussion on Quad amplifiers and I feel we've strayed far off topic. A thread on your design or your feelings on power devices may suit a thread of their own.
-Chris
It seems to me that your findings depend heavily on the circuit and devices used. I am not adverse to using a mosfet where it's advantages are clear. The same goes for a BJT (I'm also very familiar with the characteristic curves).
However, this thread is a discussion on Quad amplifiers and I feel we've strayed far off topic. A thread on your design or your feelings on power devices may suit a thread of their own.
-Chris
anatech said:
Yes, I always tend to design circuits with the end result in mind, and to use devices properly assuming their own properties. For example, I try to use input capacitances and relatively high voltage bias for good instead of fighting against them, and so on. That's why my diagrams look so simple and work better than some much more complex designs.
Good Idea. If you don't ask questions about my design, I would not answer. In turn, I advocated the British design because I know advantages of such design by own experience that I was glad to share.
"Don't tell about taste of kivi fruits if you did not eat them". I did, experimenting extensively, and I confirm they taste really good.
- Status
- Not open for further replies.