On Sat, 31 Mar 2007 20:41:21 +0100, Eeyore
wrote:



John Larkin wrote:

Eeyore wrote:
John Larkin wrote:
Eeyore wrote:
John Larkin wrote:

I like to protect transistors by digitizing their voltages and
currents, and the heatsink temperature, running a realtime thermal
simulation, and limiting *junction temperature*.

Digitising ? That's not going to be cheap.

Digitizing is cheap, especially when it lets you safely get, say,
twice the usable power from a given mass of transistors and heatsinks.

You don't. Audio amps already push output devices pretty much as far as they're
happy with.

But they don't do it intelligently. A proper protection scheme can
deliver a lot more usable power for the same amount of "push."

You're missing the point. The 'push' is determined by the audio signal. Mess with that and
you have distortion.

You're missing my point. If you can get twice the undistorted output
by using a smarter protection circuit, it's still twice the output.

How can you increase the allowable dissipation with a protection circuit ? Dissipation is
determined by the audio signal and load impedance.

The point is to *know* the actual dissipation, not make a crude guess
about it. Using a simple current limit, foldback limit, or the trick
from the patent, if you want to make your amp reliable in actual use,
you have to set the limits low, because your junction temp
"calculation" sucks and can err wildly depending on input levels and
duration, real/reactive loads, line voltage, and ambient temp.


The amplifier must function unimpeded for any valid combination of input level and output
load.

What does "unimpeded" mean for a huge input and a shorted load?

A short isn't a valid load.

So it's OK for the amp to explode when shorted? A current limit that
protects against a shorted load will severely limit peak power output
into a normal load.

How about this?

17 KW peak power out. Power supply rails programmable from +-75 to
+-175 to match various loads. Intelligent digital
simulated-junction-temperature shutdowns. The P and N fets are
interleaved for best cooling, and the shiny bars are nickel-plated
copper heat spreaders. VF display shows everything: RMS output, power
supplies, temperatures, junction temps, all sorts of stuff.

Protecting power devices is a computational problem, and the most
accurate way to that is digitally. You *could* do the junction temp
simulation with some diffamps and multipliers and such, but since I
had a CPU here anyhow, it was a lot easier to do in code.

John