A conventional voltage-feedback op amp places very few restrictions on designers when it comes to having reactive networks in the feedback path (Fig. 1a). However, when an integrator is needed to integrate pulses of a few tens of nanoseconds in duration with a 3-V amplitude, scant options are available. Few voltage-feedback amplifiers have the speed and input slew rate to cope with the signal input. One option is to use a passive integrator, and then employ a current-feedback amplifier (CFA) to buffer the output.
A better option is to use the CFA as an active integrator, but a CFA can't have a reactive element in its feedback path. Therefore, resistor RINT must be replaced with an inductor (Fig. 1b). This yields a transfer function similar to that of a conventional capacitor-resistor integrator. The main error source (not considered in the equations shown in the figure) comes from the inductors, which tend to have a lower Q than capacitors.
For example, if an EL2120 CFA is used, and if Rf equals 270 Ω (obtained from the manufacturer's data sheet), then a 20-µ inductor supplies a 74-ns time constant. Much longer time constants can be achieved by using CFAs that use a high-value Rf, such as the EL4393 (Rf = 1300 to 1500Ω). This allows relatively long integration periods without being forced to use inductors that, because they have so many turns, have a resistance which becomes significant.
The oscilloscope photos show 2-V, 20-ns pulses integrated by an EL2020 with a 100-µH inductor and a 680-ΩRf resistor (Fig. 2). This gives a time constant of 147 ns, and, as the calculations would indicate, the 20-ns pulses barely disturb the integrated level at the moment of the pulse (left photo). As intended, once the pulse width decreases, the "dc" level of the output increases (right photo).