Simulating LT1166-based Current Source in LTSpice

The '1166 model is essentially transistor-level and it has trouble converging in some instances (the sim run itself is not especially taxing on the CPU). It may help to have the alternate solver invoked, as it uses an extended-precision math function. In tough cases, you can skip the dc solver in the control setup and use a pulse source signal to help kick-start the sim. Changing the R5 value in the LT1166 entry within LTC.lib can help, but it pushes the bias point a bit higher than the nominal 20mV.

The instability you notice though is due to the loop gain being a function of the load resistance (driven where you show a ground symbol?), so at low load resistances, the current will swing wildly with small voltage commands at the '1166. If there is voltage headroom to spare, a coupling resistance could be added that would ensure a minimum load resistance and the loop could be compensated properly for that worst-case (through selection of the integration cap at the error opamp). You could analyze the opamp loop more easily by replacing the '1166 section with a gain=1 e block driven by a 1us RC.

Thanks for the very helpful response.

I used the AD8226 with an external gain resistor as a quick alternative to an AD8228 configured for a gain of 100.  A colleague did the job "properly" and imported the AD8228 model into the circuit.  That cured the convergence problem (although LTSpice still worked hard to find the operating point).

I've been using the alternate solver by default.

I'll try both your suggestions: replacing the '1166 block, and experimenting with a coupling resistance (although there isn't much voltage headroom to spare).  The load resistance is where the ground symbol is shown.

Sometimes this circuit is used with a current transducer, so the load resistance is very close to zero ohms.

Thanks for the very helpful response.

I used the AD8226 with an external gain resistor as a quick alternative to an AD8228 configured for a gain of 100.  A colleague did the job "properly" and imported the AD8228 model into the circuit.  That cured the convergence problem (although LTSpice still worked hard to find the operating point).

I've been using the alternate solver by default.

I'll try both your suggestions: replacing the '1166 block, and experimenting with a coupling resistance (although there isn't much voltage headroom to spare).  The load resistance is where the ground symbol is shown.

Sometimes this circuit is used with a current transducer, so the load resistance is very close to zero ohms.