Hi Lalit,
I have done simulations in some coherent systems with sample rate exactly equal to the bit rate value, while in systems using quadrature modulator what you have mentioned about the sample rate being at least twice the bit rate should be applied. I would also like to know about the theoretical aspect. However, in OCDMA (and some other examples) I believe the reason for samples per bit being so much more than one is just to get better results. Otherwise, we still have results (with lower quality though) for smaller values.
Regards