There are a couple of reasons for the peak gain not being 21dB:
1) The power going into the amp pushes it into saturation, so the (large signal) gain seen at the output is less than the small signal S21 gain plotted from the data file. You can plot the power out versus power in curve of the block to see this effect, or you can reduce the signal in to the small signal region.
2) In RF Blockset we interpret the unitless Simulink signal as being the source voltage, Vs, of a source with output impedance Zs, whose value you set in the Input Port block. The default is 50 ohm. This leads to 6dB less signal at the output than if we interpreted the Simulink signal as the voltage available from the source Vavs, where Vavs = Vs/2. Insert a Gain of 2 if you want to interpret the Simulink signal as Vavs.
3) The default Spectrum Scope calibration isn’t the calibration that is needed in this case, so you have to add a Gain block to calibrate it as desired (see attached model).
The reason for the shape looking different at the band edges is that the input signal is aperiodic. This would require an infinitely small frequency resolution to model it. With the finite frequency resolution in the model (frequency resolution = 1 / (length of FIR modeling filter * sample time)), the impulse response that should be at infinite delay appears erroneously at t = t_s * length of FIR. A workaround is to add an artificial delay after the amplifier that delays the model time window and prevents the “wrap-around.”
