quote:

---

Product Philosophy and Design Theory

When I started designing amplifiers 25 years ago, solid state amplifiers had just achieved a
firm grasp on the market. Power and harmonic distortion numbers were king, and the largest
audio magazine said that amplifiers with the same specs sounded the same.

We have heard Triodes, Pentodes, Bipolar, VFET, Mosfet, TFET valves, IGBT, Hybrids,
THD distortion, IM distortion, TIM distortion, phase distortion, quantization, feedback,
nested feedback, no feedback, feed forward, Stasis, harmonic time alignment, high slew,
Class AB, Class A, Pure Class A, Class AA, Class A/AB, Class D, Class H, Constant bias,
dynamic bias, optical bias, Real Life Bias, Sustained Plateau Bias, big supplies, smart
supplies, regulated supplies, separate supplies, switching supplies, dynamic headroom,
high current, balanced inputs and balanced outputs.

Apart from digitally recorded source material, things have not changed very much in twenty
five years. Solid state amplifiers still dominate the market, the largest audio magazine still
doesn't hear the difference, and many audiophiles are still hanging on to their tubes.
Leaving aside the examples of marketing hype, we have a large number of attempts to
improve the sound of amplifiers, each attempting to address a hypothesized flaw in the
performance. Audiophiles have voted on the various designs with their pocketbooks, and
products go down in history as classics or are forgotten. The used market speaks
eloquently: Marantz 9's command a high price, while Dyna 120's are largely unwanted.

There has been a failure in the attempt to use specifications to characterize the subtleties of
sonic performance. Amplifiers with similar measurements are not equal, and products with
higher power, wider bandwidth, and lower distortion do not necessarily sound better.
Historically, that amplifier offering the most power, or the lowest IM distortion, or the lowest
THD, or the highest slew rate, or the lowest noise, has not become a classic or even been
more than a modest success.

For a long time there has been faith in the technical community that eventually some
objective analysis would reconcile critical listener's subjective experience with laboratory
measurement. Perhaps this will occur, but in the meantime, audiophiles largely reject bench
specifications as an indicator of audio quality. This is appropriate. Appreciation of audio is
a completely subjective human experience. We should no more let numbers define audio
quality than we would let chemical analysis be the arbiter of fine wines. Measurements can
provide a measure of insight, but are no substitute for human judgment.

As in art, classic audio components are the results of individual efforts and reflect a coherent
underlying philosophy. They make a subjective and an objective statement of quality which
is meant to be appreciated. It is essential that the circuitry of an audio component reflects a
philosophy which address the subjective nature of its performance first and foremost.

Lacking an ability to completely characterize performance in an objective manner, we should
take a step back from the resulting waveform and take into account the process by which it
has been achieved. The history of what has been done to the music is important and must
be considered a part of the result. Everything that has been done to the signal is embedded
in it, however subtly.

Experience correlating what sounds good to knowledge of component design yields some
general guidelines as to what will sound good and what will not:

1) Simplicity and a minimum number of components is a key element, and is well reflected in
the quality of tube designs. The fewer pieces in series with the signal path, the better. This
often true even if adding just one more gain stage will improve the measured specs.

2) The characteristic of gain devices and their specific use is important. Individual
variations in performance between like devices is important, as are differences in topological
usage. All signal bearing devices contribute to the degradation, but there are some different
characteristics are worth attention. Low order nonlinearities are largely additive in quality,
bringing false warmth and coloration, while abrupt high order nonlinearities are additive and
subtractive, adding harshness while losing information.

3) Maximum intrinsic linearity is desired. This is the performance of the gain stages before
feedback is applied. Experience suggests that feedback is a subtractive process; it removes
information from the signal. In many older designs, poor intrinsic linearity has been
corrected out by large application of feedback, resulting in loss of warmth, space, and detail.

High idle current, or bias, is very desirable as a means of maximizing linearity, and gives an
effect which is not only easily measured, but easily demonstrated: Take a Class A or other
high bias amplifier and compare the sound with full bias and with bias reduced. (Bias
adjustment is easily accomplished, as virtually every amplifier has a bias adjustment pot, but
it should be done very carefully). As an experiment it has the virtue of only changing the
bias and the expectations of the experimenter.

As the bias is reduced the perception of stage depth and ambiance will generally decrease.

This perception of depth is influenced by the raw quantity of bias current.

If you continue to increase the bias current far beyond the operating point, it appears that
improvements are made with bias currents which are much greater than the signal level.

Typically the levels involved in most critical listening are only a few watts, but an amplifier
biased for ten times that amount will generally sound better than one biased for the few
watts.

For this reason, designs which operate in what has been referred to as "pure" Class A are
preferred because their bias currents are much larger than the signal most of the time.

As mentioned, preamp gain stages and the front ends of power amplifiers are routinely
single ended "pure" Class A, and because the signal levels are at small fractions of a watt,
the efficiency of the circuit is not important.
Problems with push-pull amplifier designs associated with crossover distortion have been
discussed elsewhere at length, and one of the primary results is non-monotonicity. Class B
and many AB designs have distortion products that dramatically increase with decreasing
signal. This is reduced greatly by Class A mode, but crossover distortion remains as a
lower order discontinuity in the transfer curve.

A very important consideration in attempting to create an amplifier with a natural
characteristic is the selection of the gain devices. A single ended Class A topology is
appropriate, and we want a characteristic where the positive amplitude is very, very slightly
greater than the negative. For a current gain device, that would mean gain that smoothly
increases with current, and for a tube or field effect device a transconductance that smoothly
increases with current.

Triodes and Mosfets share a useful characteristic: their transconductance tends to increase
with current. Bipolar power devices have a slight gain increase until they hit about an amp
or so, and then they decline at higher currents. In general the use of bipolar in a single
ended Class A circuit is a poor fit.

Another performance advantage shared by Tubes and Fets is the high performance they
deliver in simple Class A circuits. Bipolar designs on the market have between four and
seven gain stages associated with the signal path, but with tubes and Mosfets good objective
specifications are achievable with as few as one gain device in the signal path.

Regardless of the type of gain device, in systems where the utmost in natural reproduction is
the goal, simple single ended Class A circuits are the topologies of choice.

---

quote:

---

1) Simplicity and a minimum number of components is a key element, and is well reflected in
the quality of tube designs. The fewer pieces in series with the signal path, the better. This
often true even if adding just one more gain stage will improve the measured specs.

---

quote:

---

3) Maximum intrinsic linearity is desired. This is the performance of the gain stages before
feedback is applied. Experience suggests that feedback is a subtractive process; it removes
information from the signal. In many older designs, poor intrinsic linearity has been
corrected out by large application of feedback, resulting in loss of warmth, space, and detail.

---