The C55 R43 R40 form an low-pass with time constant t1 - that creates the floating "reference level".
The parts at the inverting input form a "shaper".
The opamp will compare the average ref level at the noninverting input with the actual modulation level at the inverting input, thus it recovers the digital signal off the input independent of the signal's amplitude (it compares the "fast modulation changes" against the "average reference level").
The "longest period td" of the digital modulation signal (1/0 changes) should be much smaller than the t1.
PS: this circuit was pretty popular in 70ties as we were building the DIY radio-control for models. The signal coming from the receiver (with varying amplitude carrying the digital modulation signal) passed that opamp, and the output TTL/CMOS level was then fed into a shift register. The outputs of the shift register were wired to the servo's PWM inputs..