Because we have to know the phase of elements in the channel matrix first before we do the EGC,however,the channel should be randomly generated every second,so I wonder that if we can do EGC in the real world or not?
Yeah, so you continuously have to estimate the channel. You need to do that anyway, no matter whether you use any combining method or not, because phase is typically important to the receiver!
So, think of EGC as a kind of channel knowledge-based beamforming.
Therefore: Yes, EGC is used. The difference to MRC is that there's no need to estimate the receive path's SNR – and depending on the signal and receiver model, that SNR estimate might be non-trivial (e.g. when you don't know the different noise figures of your receivers, or when you can technically easily know the phase of your receivers, but not the strength of the signal, e.g. because it's very weak, before you used combining).