The spacing between antennas has to be at least $\lambda/2$ where $\lambda$ is the wavelength.The key to using a diversity scheme is that the fading channels between the transmitter's antenna and the receiver antennas should be as independent as possible so the greater the distance between the receive antennas the better.Recieve diversity does not improve received signal strength, it just improves the receiver's sensitivity and in thus doing improve link performance.
Given that you have 1 transmitter and 2 receivers (i.e a single-input single output SIMO system), there are two options you can use to demodulate at the receiver:
1) You can independently demodulate the two symbols, average them and choose the most likely transmitted symbol.This option will double the processing time at the receiver or will use twice as many resources at the recievers so it isn't the most ideal option.
2) What is commonly done is that there is channel estimation done at the receiver (this commonly done by using pilot symbols and the LMS algorithm) and those the channel estimates are used to multiplex the symbols at the receiver to create the decision variable, $Z$.The decision variable, Z, can then be used for demodulation.i.e
$$
\mathbf{y = H}x + \mathbf{n}
$$
$$
Z = \mathbf{H^Hy} = \left( \sum_{n = 1}^N |h_n|^2 \right)x + \mathbf{H^Hn} = \left( \sum_{n = 1}^N |h_n|^2 \right)x + \mathbf{\hat{n}}
$$
where $\mathbf{y} = [y_0 ,y_1]$ is the received symbol vector for the received symbols at receiver 0 and receiver 1, $\mathbf{H} = [h_0 , h_1]^T$ is the channel estimates for the different fading channels to the reciever, $x$ is the transmitted symbol and $\mathbf{n} = [n_0 , n_1]$ is the noise vector.The operation $\mathbf{H^H}$ is the Hermitian transpose of the vector $\mathbf{H}$ and is equivalent to performing a transpose and then taking the complex conjugate (i.e $\mathbf{H^H = (H^*)^T} $). As the equation shows, a divesity gain of order $N$ can be achieved using this method.
General diagram of a SIMO scheme:
