I have been working on a friends Arcam AVR360 which has been suffering some serious overheating issues on 3 of the 7 discreet amplifiers. The three bad channels are experiencing thermal runaway and the variable Vbe multiplier just doesn't track anymore. The output stage is a fairly typical class AB bjt power transistor pair (each with a 0.22R emitter resistor) driven by corresponding driver transistors. The power transistors have integral diodes for thermal matching which are used along with a Vbe multiplier, thermally coupled to a driver transistor, to bias the four emitters of the output stage and the 0.22R emitter resistors. The integral diodes provide tracking for the power stage and the Vbe multiplier compensates for the drive transistors. The design is very sensitive to bias mismatch with each +44mV adding 100mA to the bias current and at 90V total supply that's 9W per +44mV per channel, and they all share a common heatsink. The drive transistors on tiny heatsinks get especially hot.
So I have been pondering how to approach biasing the output stage to maintain thermal stability. The output power transistors look good with their integral diodes, but the driver transistors look to be the weak point, with their high dissipation and poorer thermal tracking. It would be nice to use matched transistor pairs like the BCM847 etc., using one transistor for biasing, but they lack the power handling. The simple solution is to use cascode transistors to handle the voltage and power for the driver transistor making the biasing and tracking more robust. It does add a little to the complexity but has the added advantage of increasing the bandwidth of the output stage by a factor of about five times i.e. 1.1MHz BW becomes 5MHz, that will help improve stability when feedback is applied.
a. I have included a couple of AC analyses, one with the cascode transistors and one without showing the BW change.
b. The LTSpice schematic shows two 10ma current sources in the bias chain, in the AVR360 it is these current sources that are modulated but for simplicity I have chosen to use a voltage source V5 with a 1k resistor, for simple open loop simulation of the output stage.
c. The cascode transistors Q2 and Q6 have been chosen purely for convenience of the SPICE model, something else would probably be chosen for a real design.
d. The voltage source V1 is to get the approximately 100ma bias current through the output stage.
e. The capacitor values have no great design reason but should be good enough.
f. There is no current limiting, just keeping it simple.
g. I have included the LTSpice file as well as the BJT library I used.
I don't expect the design is particularly novel, as it is a pretty obvious solution, but I haven't seen it elswhere, however I fully expect someone will show some prior references which would be appreciated. I am interested in what you might consider the pros and cons of the cascoded driver stage are. Mine are:
Pros:
1. Lower temperature rise of drivers.
2. Better matching of each driver, less chance of thermal runaway.
3. Significantly improved BW. and better phase margin as a result.
Cons:
1. More complex, extra components.
2. More base drive current needed for the drivers.
3. Reduced voltage swing.
Your input would be appreciated.