The error is possibly due to amplifier input bias current, which is quite high - 3.5 µA typical, 9 µA max. The voltage drop across R4, 200 µV, suggests 5.6 µA (no place for it to go but into the amp input; presumably moderately high measurement error since voltages reported at ends of R4 are only to one post-decimal digit. The amp input offset voltage spec is not particularly good - 1 mV typ, 6 mV max, which is not bad for a fast amp, but not great if you need precision.

Since the only feedback for the amp is through the transistor, if the amp offsets due to input offset voltage and/or bias current are of the "wrong" net polarity, it will perfectly happily drive drive enough current into the transistor base to "null" the error with zero volts at the non-inverting input (NII). Though the amp output can't get all the way to zero if it has to sink any current, in driving the base it only has to source current. Clearly at half a volt output the bottom of the amp's output totem pole is off, unless the transistor has a very leaky collector-base junction. Adding a resistor base-emitter makes a path for current into R1 without the need for collector current - and makes a mess of the transfer function of the current sink (normal base current is already an error factor since it contributes to current through R1 that hasn't come through the collector, which is why very high beta or a FET is preferred).

Balancing the resistance to ground for both input of the amplifier when the signal is nominally zero (but presumably not zero ohms to ground) will help cancel error due to input bias current.

The offsets could work in the opposite way - requiring the control signal to be some millivolts above zero to get any current through the load.

 

Thanks Senacell, Danadak and ebp. Your advice has been great for sorting out and understanding this last problem. Clearly the AD8061 is not the best device in terms of input bias current. I shall look for and test something else. As for fixing the issue, initially putting a 1K resistor over the LED has extinguished the LED without affecting any of the switching waveforms. I will only use ~12mA in the 'off state anyway so I am good with that.

To all the contributors, thank you for your time and expertise. It has really helped me get to grips with this tricky problem. Looks like I have a good solution with a few bits of further tweaking to come in the future.

AnH