The Sorensen Audio Experiment, design description version 3.0
4.2. Error Amplifier /
Integrator
4.4. Power output stage
(including MOSFET driver)
5. Implementation Considerations
Copyright (c) 2001-2004 Johan Sörensen.
Permission is granted to copy, distribute and/or modify this document
under the terms of the GNU Free
Documentation License, Version 1.1
or any later version published by the
Free Software Foundation;
with the Invariant Sections being
section 1. (this copyright notice), with no
Front-Cover Texts, and with no Back-Cover
Texts.
A copy of the license is included in the
file entitled "License.htm".
The aim
with this design is to realize a High Fidelity, high power audio amplifier, in
a simple and relatively cheap way. That is, at least cheap compared to
commercial High End audio equipment. This design is the third generation in the
Sorensen Audio Experiment series, therefore designated as version 3.0.
Note! The
contents of this document describe the current state of the project, which is
considered to be “finished”. There could still be a need for some improvements
to this documentation, specifically in terms of how-to-build instructions.
(Other suggestions are also welcome!)
Note
also that I am not advocating
that others should build this! You are free to do so, using the information
you find here, but then you do it on your own risk!
This is not a project for a beginner!
The basic design
principle is class D amplifier operation, where the output transistors
alternate between only two states; completely off and completely on. This
alternation is done many times per second, creating a pulse-width modulated
output signal. This is then filtered, to closely reproduce the desired music
signal in the audible range.
There are
different ways of creating a pulse-width modulated signal. A common one is to
compare an analog input signal with a high-frequency triangle wave, and drive
the output to high when the input is higher than the momentary triangle wave
voltage, and drive the output to low otherwise. This creates a pulse-width
modulated output signal, with a switch frequency equal to the frequency of the
triangle wave.
This basis
for design is slightly different. The picture below shows the generalized
elements of the circuit.
This
circuit can be described as a comparator driving an output stage (to either
“high” or “low” state), with a feedback loop around the whole thing, and a
noise-shaping loop filter providing the input to the comparator. A theoretical
analysis of the noise-shaping filter can be found at http://listen.to/audioexperiment.
There is
nothing setting a fixed switch frequency in this picture. If all components
were ideal, the switch frequency would be virtually infinite. But in reality,
components are far from ideal (comparators have lag-time, transistors take time
to switch on and off, and so on…). This means that all the components in the
feedback loop in this design together sets the switch frequency. This is
further explained below.
·
Sound
quality near “High End” state of the art. (This is subjective, but usually, low
THD and high damping factor is a good thing J.)
·
Power
delivery capability (per channel) of at least 100W continuously in 8ohm
with < 1% THD.
·
The design
of the output stage, and the voltage swing increase circuit driving that,
should be independent of the supply voltage. (The theoretical minimum supply
voltage needed to deliver 100W in 8ohm is ± 40 Volt for a single ended output). In practice, with a conventional
rectified transformer supply, a 38 V AC transformer yielding ~54 V DC after
rectification is needed.
·
A
well-constructed embodiment of the design should be possible to be EEC/FCC type
approved.
The
descriptions in this chapter refer to the schematics
diagram, version 3.0E.
The present
design described in this document creates the pulse-width modulated output
signal through self-oscillation. Oscillation is something that is normally
avoided at all cost in amplifiers in general, but this case is very different.
In this
design, the comparator, the voltage swing increase circuit, and the power
output stage, together make a very high gain amplifier, which is (only)
optimized to swing to the positive or negative supply voltage alternatively. A
feedback loop is then applied around this composite amplifier, and an
integrating error amplifier controls the input to the aforementioned composite
amplifier. This in effect creates a very fast on-off regulation, which
oscillates at nearly 1MHz in the present design. This pulse-width modulation
frequency is significantly higher than the highest audible frequency.
Another way
of understanding the operation of this circuit is that the error amplifier
compares the integrated momentary value of the output stage with the desired
value defined by the input signal, and then drives the output to the positive
or negative supply respectively depending on the outcome of the comparison. Of
course, driving the output to the positive supply for a while invariably and
quickly leads to a integrated output value that is slightly “too high”, thus
forcing the output to the negative supply instead, and so on… The point is that
this process is so quick and accurate compared to the fluctuations in the input
signal, that the latter is followed quite closely by the filtered output
signal.
The
remaining sections in this chapter contain some notes, explanations and/or
ideas concerning the different sub-circuits of the complete design.
The LF356
(U2) serves as an integrating error amplifier. (The integration is a critical
part, in that it does noise shaping.)
The value
of C8 determines the integration constant. The suggested value of C8 was
determined empirically – you may try to tweak this up or down, and observe the
effect on the overall amplifier operation (in particular, the audible result in
terms of distortion).
* Note: evaluate
other operational amplifiers than LF356 for this. (A component with very low
voltage offset and low bias currents on the inputs is preferable here!)
The LM311
comparator unfortunately only has an open collector output, which means that
it’s output can only pull to the negative supply, and there has to be a pull-up
resistor to the positive supply. To alleviate the bad positive drive
capability, a BF245A JFET transistor is applied directly after the LM311
output. Since the JFET is configured as a voltage follower, it in turn has a
not-so-good negative pull capability (due to R51). This is finally alleviated
by D33-R47, such that the final output of this stage (the Source on J4), has a
reasonably good positive as well as negative drive capability.
In this
design, two complementary P- & N-channel MOSFET transistors are used at the
output. These are driven by one MAX626, via a capacitive level shift and clamping
circuit at the FET transistor gates.
The first
74HCT14 inverter, between the comparator and the two other delay matching
circuits is there only as an impedance buffer.
The
RC-filters, bypassed in different directions with switch diodes, preceding the two
other 74HCT14 inverters, are there to introduce suitable turn-on delays for the
MOSFETs. This is to avoid cross-conduction, i.e. a small dead time is allowed
between the moment where one MOSFET is turned off, and the other one is turned
on. The values for C19 and C20 depend on what kind of MOSFETs that are used;
bigger and slower FETs will need more turn-on delay. Also, there is a notable
discrepancy between the values needed for a circuit on a breadboard, and one on
a PCB. Once the design has passed the beta-testing stage, I will document all
component selection cross-dependencies found.
Making an
adequate output filter is as much about getting good components as anything
else – in particular inductors that withstand high currents without big losses
and/or distortion. I prefer the concept of core-less inductors, since there
cannot be any magnetic saturation effects causing distortion (as the may be
particularly in toroid inductors, depending on the core material).
Designing a
2nd order Butterworth low-pass-filter with a transition frequency of
50 kHz, yields filter component values according to the table below:
RLoad |
L (uH) |
C (nF) |
8 |
36 |
280 |
4 |
18 |
560 |
I have
wound my own inductor coils using 1mm thick wire (coated with a thin insulation
- the kind of wire used in transformers…). 22 turns with a diameter of 4cm
should yield an inductance of ~20uH.
Should you use another wire thickness, or other dimensions, the inductance can
be calculated with
L = (0.08d2n2) / (3d +9l)
where
L is given
in uH
d = diameter in cm
n = number
of turns
l = length
of the coil
There are
issues relating to the PCB design implementation and output filtering, which
are important for the overall performance in a real embodiment of the design.
Be careful how you draw the power supply and earth strips, considering where
high frequency currents will flow. The abstract schematics design does not
fully reflect the capacitor bypassing needed in various places for successful
operation!!!
Create one
ground plane for the Low Voltage parts (the ones supplied with ± 5V), and one ground plane for the High Voltage parts.
Then pull wires from these two grounds to a common ground point at the power
supply.
In the
output filter, the ground side of C14 must be connected to the high voltage
ground; this is the same ground point you use for bypassing the high voltage supplies with capacitors of at
least a few tens of µF near the output transistors. Also, bypass the supplies
with 330nF capacitors, with good high frequency characteristics. (As close to
the output transistors as possible.)
Experience from my process of
building and testing the circuit on PCBs suggests that this kind of design
[i.e. switching power amp.] is quite sensitive to PCB layout (since high
powers and switching currents are involved), so it may be a lengthy and arduous
process test out different PCB variations to find out what is working and what
is not, if you don’t know exactly what you’re doing the first time…
So far,
only a few rudimentary tests and measurements have been made, using an
oscilloscope and a normal multimeter instrument, and a PC as a signal source.
(As of June 8, 2002)
The tests
were made on a PCB 3.0D, with a power supply giving +/- 54V with no load (drops
to 45-50V fully loaded). This was aided with a +/-30V laboratory power supply,
which is further regulated down to +12V and +/-5V for the various parts of the
amplifier circuit. A large 8ohm resistor served as a test load.
The output
transistors were mounted on a small heat sink (approx 6 by 3 cm).
Some notes:
And of
course, there are no THD measurements etc. yet; what I’ve done so far serves to
demonstrate the quantitative (if not qualitative) abilities of the design J.
Just for
the record, here is a summary of milestones in the development of the 3.0
design and PCBs:
Date |
Event |
Feb. 2001 |
Created
the Audio Experiment Website. Started collecting ideas & background
material on the making of a MOSFET switching output stage, capable of
sustaining 100-200W output power into 8ohms. |
Dec. 28,
2001 |
Started
building circuit of yet-to-be 3.0A on a prototype board. |
Jan. 2002 |
Much
experimenting and discussions on the Audio Experiment forum (and offline) led
to 3.0A design with MAX626 driver and capacitive coupling to gates of the
output MOSFETs. |
Feb. 1, 2002 |
Had a
prototype for 3.0A on the breadboard working quite well with the limited
laboratory power supply. (+/- 30V) |
Feb. 11, 2002 |
Had a
first PCB (160x100mm) with two channels ready for the 3.0A design; however,
this was unstable and unreliable, and sounded bad when both channels were
operated concurrently. |
Feb. 24, 2002 |
After
some testing and tweaking of the 3.0A PCB, it was found to have low
efficiency (~70%) and, was generally unreliable. |
Mar. 2002 |
Started
design on a new PCB, and prepared it to use either LT1016 or LM311 as
comparator. (LT1016 is much faster, but therefore also more prone to pick up
noise.) |
Apr. 7, 2002 |
Second
PCB (3.0B) ready and working to some degree with the LM311 comparator. |
Apr. 16, 2002 |
The great
moment when the new design started amplifying music signals in my living room
“for real”. This time quite reliably, and with (subjectively) acceptable
efficiency. (The amplifier casing was only very slightly elevated in
temperature even after hours of playing at high levels.) |
Apr.23, 2002 |
However,
when trying to measure efficiency, I realized that the lack of DC stability
on the outputs wasn’t without problems. Thus the card had to be augmented
with this feature. (I had had some foresight when designing it, so it was
possible to retrofit a couple of extra transistors onto the card J.) |
April 2002 |
Although
I was quite happy with the operation of the 3.0B PCB, I started designing a
new PCB (3.0C) where each channel was totally independent (on 80x100mm). |
May 2, 2002 |
The
tweaked second PCB was put into operation again, successfully. |
May 25, 2002 |
The third
PCB set had proven impossible to get to work, apparently due to PCB layout issues.
(I hadn’t been as generous with ground planes as on the 3.0B PBC L.) So, a fourth design was started. |
June 5, 2002 |
Two 3.0D
PCBs were initially tested – sounding great J. |
June 8, 2002 |
Tested
operating limits and efficiency of 3.0D PCBs; see section 6.1. |
Aug. 4, 2004 |
Clean-up
of document. Removal of input buffer OP that was present in the 3.0D design. |
Thanks to:
Svante
Gellerstam for listening to my stuff and encouraging the project right from the
start (1996!).
Daniel
Ståhl for providing various electronics expertise and measurement help.
My wife for
tolerating the audible activities.
My
father-in-law for supporting the project,
Joseph
Meisenhelder for sharing his experience on MOSFET driving techniques.