DYNAMICALLY INVARIANT AB LINEAR OPERATION (DIABLO) 6/15/95

James C. Strickland

BACKGROUND

For more than 30 years, the front-ends of most audio power amplifiers have followed a
similar pattern consisting of a differential amplifier driving a transconductance stage, as
shown in Figure 1. This arrangement is a Class A stage having the limited efficiency and
2:1 peak- to-quiescent current output capability typical of such operation. As a result, it
has become necessary to interpose one or more Class B current amplifying buffer stages
after the transconductance stage to properly drive the output devices.

Over the years of the audio power amplifier design’s evolution, it has been learned that
minimizing the number of stages in the signal path usually results in significant
improvements in closed-loop stability and sonic quality. An exception to this is the
addition of common-base or cascode stages as shown in Figure 2, which can result in
greater stability due to increased frequency response resulting from radical reduction of
Miller-effect capacitances.

A significant gain in short-loop design art has been possible due to the use of power
MOSFET output stages, which can be driven directly from an appropriate
transconductance stage. To date, these bufferless designs have been affected by increasing
the power level of the Class A transconductance stage--a process which quickly becomes
unwieldy for numerous technical reasons, not the least of which is heat and related
effects.

Designers have been naturally reluctant to forego the linearity of Class A

transconductance systems in favor of more efficient, but far less linear, Class AB systems
of conventional design. Clearly, if possible, a new transconductance-stage format having
the linearity of a Class A system and the power characteristics of a Class AB system
could significantly advance the art of power amplifier design. It has recently been
discovered that such a combination is possible and quite feasible to implement.

PRINCIPLE OF OPERATION

In conventional transconductance systems, the two paths from the differential amplifier
to the transconductance devices are developed and processed independently and linearly in
each path. In the new system, these two paths are made interactive and nonlinear, in a
manner which always combines to the highly linear output characteristic of a Class A
system yet having the significantly expanded peak current output and efficiency of a

1
Class AB system. The necessity for this unusual approach is shown in the following
sections.

IMPLEMENTATION

As shown in the improvement of Figure 3, the basic elements of Figure 2 have been
augmented by the addition of a voltage regulator network consisting of diodes D1 and D2
with trimming resistor Rt in series, an additional common-base or “cascode” stage Q4 and
emitter resistors R3 and R4. This important change in connectivity, combined with
appropriate adjustments of operating currents, results in dramatic and unexpected
improvements in the operation of the system.

The high-impedance outputs of Q2 and the current mirror cooperate to form a single-
ended, push-pull current-output stage. The differential amplifier is typically sourced by a
conventional current source of 10 mA, yielding 5 mA quiescent current at each differential
collector. The current mirror thus presents an output of 5 mA to balance out the current
of Q2. This 5 mA is utilized to bias the voltage regulator network and to provide
quiescent and signal currents to Q3 and Q4.

If a system is to ultimately transcend the limitations of Class A operation, the peak-to-

quiescent current-output ratio must be significantly increased beyond the 2:1 value of
Class A connectivity. The function of the voltage regulator network is to bypass a high
percentage--typically 80 percent--of the 5 mA quiescent “string” current developed in Q2
and the mirror around Q3 and Q4, leaving them with a quiescent current of typically 1
mA. Since peak current capability remains unchanged, reducing the quiescent term
propagation into Q3 and Q4 by 5:1 improves their peak-to-quiescent current capability by
5:1.

With typical semiconductor components, the drop on each regulator diode is 50-150 mV
greater than the base-to-emitter voltage (Vbe) required for 1 mA in Q3 and Q4. Thus a
stable 1 mA current can be formed with emitter resistors in the neighborhood of 50-150
ohms per device, cooperative with the two junction drops of the diodes temperature
compensating the two junction drops of Q3 and Q4. For flexibility, the small bias
trimming resistor Rt is included, allowing the emitter resistor values to be raised if desired.
Dynamic effects of the emitter resistors are discussed later.

The arrangements so far make Class AB operation possible, having established elevated
peak-to-quiescent capability. It will now be shown how invariant transconductance is
obtained in such operation.

Signal-Current Processing

2
For this analysis, signal-current is defined as the difference in the currents flowing in the
collector of Q2 and the output of the current mirror. This difference current has only two
possible paths it can take--through Q3 and Q4. Thus:

The difference of output currents of Q3 and Q4 will always exactly equal the
signal-current--independently of the fraction carried by each device.

This important result is equivalent to stating that the signal-current fractions in Q3 and
Q4 have a constant sum, regardless of the proportions of the two fractions. This design
element is the key to achieving invariant transconductance. Q5 or Q6 can present full
transconductance contribution down to nearly zero current. Once this point is reached,
the other path must provide all of the transconductance supplied equally by both devices
at the quiescent point. If either of Q5 or Q6 were driven linearly (at high slope) to
effectively zero current, an abrupt slope decrease would occur which would require an
equally abrupt compensating slope increase in the other path. Such mathematically -
discontinuous behavior--where the first derivative fails to exist at a point--is very difficult
to synchronize properly in two paths, leading to high-order harmonic generation.

To avoid this problem, a relatively smooth and continuous transfer function is needed for
transferring the transconductance function from one path to the other. Mathematically,
the differential equation di/dV = ki, having exponential solution, is ideal because it
produces an asymptotic approach to zero current. It will now be shown that this
function is also physically ideal, because it is highly feasible to implement in the
important region approaching zero current.

Signal-current Steering System

In the interconnected pair Q3 and Q4, the dynamic emitter resistance of each bipolar
transistor varies inversely with the instant total current in each path--following the
classical exponential transconductance relationship di/dV = ki. This resistance is typically
26 ohms at 1 mA. The signal-current develops a small AC voltage common to the entire
floating regulator network. This small AC (signal) voltage causes fractions of the signal-
current into Q3 and Q4 to be inversely proportional to the sum of the fixed emitter
resistor and the dynamic emitter resistance of the device at its instant current. Thus as +/-
signal-current is generated, one emitter-path resistance is decreased and the other
increased. As will be exampled shortly, the decrease is relatively small and the increase
much larger. As this happens, the signal-current is steered progressively into the path of
decreasing resistance and away from the path of increasing resistance--the two fractions
however always maintaining a constant sum equal to signal-current. Numerical examples
are now presented to further clarify this operation.

3
Numerical Examples

Observe now this operation with 100 ohm emitter resistors. At a signal-current where
Q3 moves from 1 mA quiescent to 2 mA, the left end of R3 will move down by 118 mV in
potential--100 mV of added drop on R3 and 18 mV of increased Vbe for Q3’s 2:1 current
increase. The regulator network also pulls down the left end of R4 by very close to 118
mV. This results in Q4’s emitter going down 40 mV (of Vbe), reducing its current to 22
percent of quiescent value, thus reducing the associated R4 drop to 22 mV. The 78 mV
reduction on R4 plus the 40 mV of decreased Vbe now accounts for the 118 mV change.

At this time, Q3’s dynamic emitter resistance is 13 ohms and Q4’s is 118 ohms.
Comparing the total (dynamic + fixed) emitter-path resistances reveals 113 ohms for Q3
and 218 ohms for Q4. Q3 is thus carrying 218/331 of the signal current and Q4 113/331.
Q3 is now carrying almost twice the signal-current fraction carried by Q4. The
asymptotic approach toward cutoff is already quite apparent. It will become much more
apparent in the next example.

An acceleration toward Class B takes place as 2 mA is exceeded on Q3. By the time 3 mA

is drawn from the emitter of Q3, it and Q5 will be handling about 99 percent of the signal-
current propagation, with only about 1 percent in the Q4-Q6 path. As signal-current
increases further, the movement toward Class B continues asymptotically with Q3 now
approaching linear operation alone--carrying virtually 100 percent of the signal-current
into Q5. Clearly, due to circuit symmetry, identical operation occurs on the opposite
phase of signal-current, when Q3 is being cut off and Q4 drives Q6 into Class B operation.

Since the differential amplifier is sourced at 10 mA, it can output peak signal-currents
into Q3 and Q4 of +/-10 mA--10 times their 1 mA quiescent level. Thus they can each
output a peak-to-quiescent ratio of 10:1. Q5 and Q6 are linearized amplifiers--they can
thus also output a 10:1 peak-to-quiescent ratio--5 times higher than that of a Class A
system. Higher ratios are feasible simply by increasing the ratio of differential amplifier
sourcing current to the quiescent current in Q3 and Q4.

Dynamic Effects of Emitter Resistor Value

It was shown earlier that the emitter resistors tend to naturally be in the 50-150 ohm
region. It can be deduced from the theory and numerical examples presented that their
value also affects the shape of the signal-current steering function--lower values resulting
in a transition into Class B approaching a classic exponential function with higher values
resulting in a more pronounced “knee” characteristic. Fortunately however, because of
the constant sum nature the system, the exact shape of the steering function is relatively
non-critical. Thus this resistor value may be chosen to optimize system bias, current

4
headroom, thermal and device-variation stability and normal “linear” cascode properties.
Over such adjustments the dynamic invariance of transconductance is fully maintained by
the principles disclosed herein.

Differential Amplifier Properties

The nonlinearity contribution of the normal “S” (saturation) curve of the differential
amplifier per se has been omitted from discussion--because of its ubiquity and its
dependence on the degree of local degenerative linearization selected for the stage.
However, because of the operation aforedescribed, the new system typically operates on
a smaller, relatively linear portion of this curve--further improving its linearity over a
Class A system. Substitution of J-FETs for the differential amplifier does not affect the
novel operation of the circuitry at all, and is typically elected for designs of the highest
fidelity.

STAND-ALONE OPERATION

Although developed primarily for the application discussed above, the new circuit can be
advantageously employed as a line amplifier such as utilized in professional and home
audio equipment. The ability to drive capacitive loads makes it well suited for driving
long shielded cables. It is also well suited for driving headphones, which require moderate
power levels.

PRESENT APPLICATION INFORMATION

Diablo is now employed in the Rockford Fosgate Punch 500m autosound

amplifier of 500 watts (mono), using 14 output MOSFETSs, the Hafler
Professional (studio-monitor amplifier) 9303 with six MOSFETs/channel and the
9505 and 7000 amplifiers with eight MOSFETs/channel. All of these amplifiers
use grounded-source (Transnova) design, as shown in US 4,467,288, which,
already having the lowest possible driver-power requirement, benefits
significantly from the improved driver system.

Conventional source-follower MOSFET designs require the same levels of gate-

charging current supplied from considerably higher operating voltages. Thus
considerably higher driver power is required in this case. The improved driver
system, with its much higher efficiency, can be quite beneficial for these high-
voltage designs, opening up for the first time the feasibility of driving the output
devices directly from a transconductance stage of reasonable dissipation level.