The e4k does the filtering but has an analog output. The RTL2832 is in effect an USB ADC.
apologies if this is obvious to everyone and has already been posted (I have not checked the whole history of this ml), but indeed by connecting a couple of wires to the I or Q differential inputs of the RTL2832 you do get a fine 2 MS/s oscilloscope. This unfortunate AGC means that DC signals cannot be monitored (even after soldering the wires after the common mode cutoff capacitors), but neverless this does provide a 40 fold sampling rate improvement with respect to the use of sound cards as A/D converters (http://sequanux.org/jmfriedt/199999720input_middle.png and http://sequanux.org/jmfriedt/199999720input_top.png for the I and Q measurements when an input signal of 199.999720 kHz is generated by a synthesizer in order to synchronize with the sampled data, ie the sampling clock is off by 1.4 ppm, not too bad !).
JM