Hello Amit:
I do not know what "CJ" band is. If it is C-band, ~4 GHz, you can find several solutions:
1. Original Dicke modulator utilized an attenuating disk rotated by a motor so it entered a waveguide section to attenuate by ~20 dB, or was pulled out to get 0 dB loss. I used a similar design but instead of a round disk I used a "propeller" with two symmetrical wings. This offers no vibration and twice the frequency, two per revolution. The position of the propeller is sensed with a photosensor and drives the sync detector. The photosensor should be mounted so the phase of the "reference" pulses can be adjusted for the best output.
2. There are MMIC switches now available, by Analog Devices, Linear Technology and RFMD, with a low loss and a good isolation, to be used as switches. You can also buy or design a PIN-diode switch. All such switches have >1 dB loss which makes them worse than the above having 0 dB loss.
3. A nice solution was used by Paglione, also at 4 GHz: to avoid switch loss, he used two LNAs before his mixer. He switched LNA DC bias to alternate, one of the LNAs was connected to antenna, the other, to reference noise source.
You can see that I chose a similar approach in my Dicke radiometer at 11 GHz. The LNB I used can switch the linear polarization of input signal, also by having two alternating LNAs before mixer.
Then I inserted a Mylar attenuating foil in the LNB input pipe so its emission only gets into one LNA while the other input is exactly perpendicular. The result is excellent. Maybe you could also find a C-band LNB with this feature, or, you can design it.