cute70.jpg
CuteStudio Ltd.

Amplifiers

Compared to the damage done by mastering amplifiers may be a minor worry but there are things to look for when choosing one. The most important aspect of an amplifier is it's design, that can then be well implemented with decent quality components.

  1. Design of an amplifier needs to some first
  2. Quality of implementation comes second

Like much HiFi however often a great deal of money is spent on implementing poor designs, when it would cost exactly the same to build a truly great amplifier. Some component choices do influence design however, such as the choice between tubes/valves and solid state.

Basic design issues

There are some good hybrid designs - for instance tube front ends with solid state followers that have pros and cons but are not looked at here.

  1. Tube or solid state?
  2. Class A, AB, B or D?
  3. Feedback schemes..

Tubes/valves vs transistors

  1. Tubes are old fashioned by the most linear device is still a triode, in particular directly heated cathode ones, but even the simple dual triodes like the 12AX7 are still very good. In addition to their linearity they run at high voltage so the actual section of the transfer curve used can be rather small, adding to the linearity.
  2. BJTs have a rather non linear response but make very good followers.
  3. MOSFETs are more linear than BJTs and are available with great power.
  4. FETs are usually too small for outputs but are closest to tubes in linearity and behavior.

As the most linear devices are tubes we should therefore choose those, unless we are using class D which is a switching design.

Conclusion

Use tubes for amplification.

Amplifier class

  1. Class A. This is the best sounding class, despite being slightly asymmetrical in concept. The reason is that it uses a constant power so the power supply does not intrude on the sound.
  2. Class B. Class B is popular because it's cheap to make. To work properly it has to switch between the +ve and -ve transistors at exactly 0mA which is a problem because:
    • That's the most common place for an amplifier so it affects the sound the most
    • Devices tend to be the most non-linear when they switch off.

  3. Class AB is class B with a little bias current run through the transistors to make them a bit more linear at the critical crossing region. Unforunately this results in a situation where both devices are switched on around the central point but one switches off as the voltage leaves that region. This causes the impedance to half every time that happens, which means in anything but a small signal the impedance is continually changing with the waveform.
  4. Class C is a partial conduction not suitable for audio.
  5. Class D is best done by very fast chips and as such is not so DIYable, although TPA3116 modules are very cheap and can be upgraded to excellent sound.

Conclusion

Use class A, but class D has some interesting properties.

Feedback

Negative Feedback NFB is a mechanism of feeding back an inverted signal from the output back to the input. This improves frequency response, impedance and the gain is controllable but as with everything it has a cost.

The cost is that NFB only really works as advertised in a linear system, any non linearity gets multiplied which has the effect than non linearities merely spread upwards in frequency and spread themselves over the bandwidth. It's a type of Conservation of Non-linear Distortion effect.
This effect still helps the THD (Total Harmonic Distortion) figures look good but the correlation of sound to THD is not good.
So NFB's pros and cons are:
  • Pros: Better drive (lower output impedance), Less noise, Less gain but to a preset, controllable level.
  • Cons: Multiplication of harmonic distortion, instability (i.e slow, phase shifting amplifiers can become oscillators).
    The most insidous and often unnotoiced con is that the obvious distortion is replaced by a subtler one. For this reason NFB is really only best applied to amplifiers which are already linear, which is a problem for BJT based amps.


Feedback in tube amplifiers

Global NFB (GNFB) is usually used in tube amplifiers in moderate amounts, despite the long known fact that feedback works best if there is lots of it. Often a mild amount is worse than none at all - just enough to multiply but not enough to do any good. Individual stage feedback is wlways present anyway, a triode has an instrinsic local feedback, all followers do and anything with a (un bypassed) cathode resistor does to. Local feedback works around each device itself and so it very 'fast', and therefore always the best feedback. Additionally the harmonics it multiplies are simpler, not the sum of the preceeding 4 stages but just of the device itself.

Tube amplifiers tend to 'get away' with the bad 'Williamson' practice of GNFB though because they are already pretty linear, so the multiplication of the non linearities hasn't got much to work on.

  • Global feedback is bad, especially in phase compromised amplifiers
  • Local feedback is good and unavoidable anyway


This leads us to consider the common but poor practive of including the OPT (output transistor) in the GNFB loop. In 99% of amplifiers this is done as it's easy - especially in push-pull amplifiers, but it's still wrong. The reasons it's a very bad idea are:
  1. The OPT is a band limited phase lagging component that has no business in a feedback loop
  2. The phase lag causes the 'phase margin' to be tight, meaning that instability is a spectre which can cause overshoot, ringing and in extreme cases (of adding more feedback - which remember is good) turning your amplifier into an oscillator. I.e. this phase lag stops the proper application of decent amounts of feedback.
  3. The band limiting nature causes the feedback to be reduced at the bass and treble. Higher quality OPTs can alleviate this but the effect still occurs so even if we manage a decent amount of feedback (before noticeable instability) the feedback where we need it - at the frequency extremes - will be inadequate.


These problems (and the solution) is something Thorsten Loesch has pointed out but for reasons I'm unable to fathom is is still controversial, possibly because the 'Williamson' GNFB over everything has been used for so long. However rebuilding a cheap tube amp from the 'GNFB including OPT' method with a circuit that avoids the OPT being in the GNFB exhibited such a leap in sound quality those fears do not seem to have any practical basis.

The solution of course is to run any GNFB just around the tubes themselves (a fast, wideband fairly linear block) and use the low impedance generated to drive the primary of the OPT, allowing the OPT to get on with it's own job without any laggy 'help'. In tests it turns out that the jump in sound quality is real and significant.

Conclusion

We need to use local feedback and avoid feedback over output transformers - even with the more linear tubes.
The best amplifier will therefore be:

  1. Tube based
  2. Class A (Single ended)
  3. Not use GNFB over the transformer.


Using these principles an extremely well sounding amplifier can be built for the cost of a pretty cheap mediocre one. The sound will be light, sweet, realistic with good bass and sweet treble and the music will come alive, freed of the constraints of a tired feedback loop trying to keep fundamentally nonlinear components together. It's just physics/electronics/theory but put together it does make a difference.


Copyright © 2007-2017, CuteStudio Ltd.
Page generated in 0.044s, Powered by Silk V1.2-0 from Cutestudio