evaluating BJTs for a linear PSU: am I doing it right?

[exe]

Hello!

When I was building a clone of blackdog's PSU I couldn't find exact pass transistor (2sa1943-something). So, I bought a clone from ST: 2sta1943. Latter I presumably found a Toshiba part, so I bough it too. I decided to compare them. So, I created a simple test rig (see pic for schematic). I apply a small stimulus between base and emmiter, and measure signal on emitter resistor. Here is what I found: transistors behave very different! It seems to me that the ST part is much better because 90 phase shift happens at much higher frequency. I'm also surprised that TIP2955G (with Ft of just 2.5MHz) in this setup performed even better than 2sta1943 with Ft of 30 MHz. In general, I found no correlation between Ft and performance in this test :/. May be some transistors don't like low Vce, or their Ft was calculated for lower currents when they perform much better.

So, I want to choose "the best" transistor based on measurements. My question is: am I doing it correctly? I understand that due to self-heating it's not going to be very accurate, but as a figure of merit, can I use these data?

Plot description: yellow line is stimulus, blue line -- response. Please ignore offset voltage on pictures, I set it to low values manually to remove stimulus, or shunt resistor would overheat.

My bench PSU can only give 1A current, but it has two channels. So, I put two channels in parallel to get more current. The low-esr (0.05ohm as measured by transistor tester) cap of 220uF is put close to transistor because I use relatively long leads from the power supply.

Because of drop on emmitter resistor R1, Vce is just 1V, the other two volts are dropped on the shunt.

PS all parts are bought from presumably reputable sources: tme and mouser.

[David Hess]

Here is what I found: transistors behave very different! It seems to me that the ST part is much better because 90 phase shift happens at much higher frequency. I'm also surprised that TIP2955G (with Ft of just 2.5MHz) in this setup performed even better than 2sta1943 with Ft of 30 MHz. In general, I found no correlation between Ft and performance in this test :/.

Like hfe, Ft is not well controlled and can vary over a considerable range for the same part number.  If you need a tighter Ft to prevent oscillation or for performance reasons, then you may have to grade the transistors yourself.

May be some transistors don't like low Vce, or their Ft was calculated for lower currents when they perform much better.

Datasheets specify the conditions for each measured characteristic.

So, I want to choose "the best" transistor based on measurements. My question is: am I doing it correctly? I understand that due to self-heating it's not going to be very accurate, but as a figure of merit, can I use these data?

The best test is in a circuit which is representative of the application circuit.  But usually one of the three transistor configurations is close enough.