Current source conundrum

[Mr Evil]

I was playing around trying to remove the need for separate current sources for an input stage when I realized that base current compensated current mirrors are actually the combination of a current mirror and a two-transistor current source. This lead to the attached schematic.

The conundrum is: Why are R16/R17 needed? It simulates fine without them, but when built on a breadboard, the circuit will not turn on (no current through Q7/Q9). With the extra resistors added, it starts working the same as the simulation.

[Mr Evil]

Here's an FFT showing 1kHz, 5Vrms into 8R. Ignore the 15kHz peak - that's just background noise. It's approximately 0.002% THD, which is ok.

[kaos]

I don’t see why the circuit should start in the first place. The simulator must make certain assumptions when it starts running (or it models a very fast ramp up speed from the power supplies, kick starting the circuit through junction capacitance, or it assumes certain leakages in the devices that may or may not be present in the real devices at room temp). The presence of R16 & R17 in the breadboard may introduce enough spurious noise in the real circuit to initiate start up (I’ve seen that before in high impedance current source prototypes). Don't really know the answer for sure though.

[Mr Evil]

Thanks for your thoughts.

I considered that perhaps the resistors in conjunction with the Early effect could give some positive feedback (reduced current through Q7/9 decreases the voltage drop across R16/17, increasing Vcb, which will tend to increase the current slightly because of the Early effect). To test this, I substituted R16/17 with Zener diodes, expecting that the circuit would fail to start with them, which is what happened.

I'm not sure if that really explains why it starts in the first place in reality but not in the sim. I tried simulating a slow ramp of the supply voltages but that didn't change anything. What is the simulator not taking into account?

[Mr Evil]

Err, that should read negative feedback above.

[kaos]

I wouldn’t expect it to start with zeners in place either, as they would represent an additional potential to overcome for whatever marginal start-up mechanism is occurring.

[Mr Evil]

With a Zener in series with the resistor it still starts, so that can't be stopping it, or at least not as much as whatever starting effect the resistor is having.

[kaos]

weird ... what kind of resistor and zener are they (not that that's going to enlighten me any)? Does it still start with no input signal present?

[Mr Evil]

An ordinary metal film resistor and BZX55C 5V6 Zener. The tail current is about 5mA, so the voltage across the resistor is a bit more than the Zener - I'll do some tests later to see exactly what value of resistor is needed (I know 100R doesn't work, but 1K2 does).

I tested it with the input open, grounded, and with a signal (square wave) present, but that doesn't seem to make any difference.

Also, With the resistors removed, if I turn up the output stage bias by increasing RV1, I can get it to oscillate at about 0.5Hz, presumably it's repeatedly almost starting and then turning off again.

[Lumba Ogir]

Mr Evil,
why do you refuse to supply suitable constant current sources for the LTPs?