Power Amplifier

Before I started with the design of my power amplifier I read the Audio Power Amplifier Design Handbook from Douglas Self.
I highly recommend everyone who wants to design his own power amp to read this book to avoid many of the pitfalls you otherwise can and will step into.

I decided to stick with the proven 3-stage design concept of differential input stage, voltage amplification/phase splitter stage and power stage but didn't want to build a fully discrete power amplifier.
National Semiconductors, now part of Texas Instruments have two power amp predrivers that incorporate the differential input stage and the voltage gain/phase splitter stage, the LME49810 for Bipolar output stages and the LME49830 for FET output stages.

The difference in power dissipation capability between BJT's end FET's is roughly a factor of two.
Given the size of the PCB I had in mind and the power levels involved I would need three to four FETs and they don't come cheap so I decided for Bipolar Junction Transistors (BJT).

The bass speaker I selected, a Seas Design L26ROY has a nominal impedance of 4 Ohms.
Before starting the design I made a plot of the dissipation for a +/-55V powerstage with Ohmic and reactive loads to see how many power BJT's I would need to parallel up to cope with the power dissipation in the power stage.

The BJT's I selected are Sanken 200 Watt complementary pairs, the 2SC3264 and 2SA1295.
In Excel I plotted the 200 Watt (dark blue) and 400 Watt (red curve) SOA curves, without taking de secondary breakdowen of the BJT's into consideration since that kicks in above 100V with these transistors. 
The green curve is the 8 Ohms load line, that falls well within the 200 Watt SOA curve.
The light blue line is the 4 Ohms load curve that falls just inside the 200 Watt SOA curve.
Since speakers aren't pure Ohmic I also plotted the reactive load curves, asuming that the worst case voltage over the BJT would be 90 Volts since a speaker is never a purely inductive or capacitive load.

The 8 Ohm reactive curve (purple) falls neatly inside the 400 Watt SOA curve so that is OK but the 4 Ohm reactive curve doesn't.
So I could add a third BJT in parallel but the PCB area restriction for my power amps didn't allow for this.
In stead I compromised and added a current limit protection to the design.
The current limit is set to 20Amps which increases the minimum impedance the amplifier can drive into clipping to 4.5 Ohm reactive load.
The current limit is more a power limit since by adding two resistors (R113 and R114) it becomes a Vce dependent current limit so the higher the voltage over the BJT, the lower the current limit kicks in.
Up to now I havent noticed any audible distortion at high power levels, near clipping.
The current limit protection is primarely intended to protect the amplifier at high power levels into impedances lower than 4 Ohms with the added benefit of a shortcircuit protection.

The curves also indicates that the dissipation in the power stage is highest between a Vce of 25 and 65 Volts.
I also made two simulation plots that show the power dissipation waveform for a 100Hz sinewave into an inductive and a capacitive load.

Top trace is the Collector current (IC) through the top side power-BJT, the second trace the power in the BJT (Ic * Vce).
With an inductive load the current is lagging so the second power peak is highest. With a capacitive load the current is leading and the
first peak is higher. 

The power amplifier design itself.

The design is straight forward, LME49810 with complementary pre-drivers (MJE15034, MJE15035) and complementary power stage (2x2SC3264, 2xSA1295), the bias transistor and power limit circuits. The PCB allows to mount either a single or a 2-parallel output stage.
The LME49810 has a Baker Clamp circuit so that when the amplifier clips it doesn't saturate and recovers much faster.
It also has a clip indication output, that I used to signal an overload condition to a front side LED.
T100 regulates the bias current through the power stage and is mounted directly on top of one of the power transistors with a nylon screw for isolation to minimise the thermal resistance and bias current variations when the amplifier heats up. P100 sets the bias current in the power stage.
25mA - 30mA is the value where the amplifier shows a moderate -85dB distortion at the second harmonic. 
Capacitor C109 determines the dominant pole of the amplifier in the LME49810 voltage gain stage and is set to limit the bandwidth to 100KHz.

The datasheet of the LME49810 indicates a large input capacitor to keep the DC balance of the amplifier.
There is a more elegant way to do this with a DC servo aplifier. I used an OPA140 (IC101) which resulted in an DC output offset of <3mV and a -3dB point of 1.6Hz.  That seems quite low but the real high pass filtering will be done in the digital domain on the DSP board.
The servo circuit also compensates for DC offsets of the driving pre-amp up to +/-1V.

In order to protect the loudspeakers, especially the tweeter, against startup and shutdown plops of the amplifier a special Amplimo audio relay was added in the output of the amplifiers.
A simple circuit detects any DC voltage on the amplifier output, de-activates the audio relais and disconnects the output of the amplifier to the loudspeaker.
The relais is activated by the Inrush Current Controller board.
It activates the audio relais after all powersuppplies have reached their nominal values and the output of the amplifier is stable.
The audio relais are also de-activated if any of the amplifiers gets over heated.
That functionality is described in more detail in the Fan Amp Controller board.

Care must be taken with ground connections in the PCB layout. I used three ground star points on the PCB. One starpoint for the secundary powersupply caps, one for the RC load (R127, C112) that provides amplifier stability at higher frequencies and the third starpoint for the small signal returns.
All three starpoints feed separately to the "master" star point of the 10mF powersupply buffer caps.
The poweramp buffer caps are therefore real ground star point for the amplifiers.
All connections to the output of the amplifier are also layed out in a start point. 
(see PCB photos in the picture section of the amplifier)
There are in total three powersupply starpoints in the system, one for the low amp, one for the mid and high amp and one for the auxilliary supplies.
They are again summed at the ground plane of the AD-DA board.