Question about Communication Toolbox's 802.11n MIMO channel model example. Where is oversampling occuring?

This is a question directed to anyone familiar with the Communications System Toolbox, or communications theory in general.

Mathworks has published an example of using comm.MIMOChannel to model a channel according to the 802.11n (Wifi) specification here:

http://www.mathworks.com/help/comm/examples/ieee-802-11n-channel-models.html

My confusion is regarding initial block of code:

M = 2;                      % Modulation order
Rsym = 10e3;                % Input symbol rate
Rbit = Rsym*log2(M);        % Input bit rate
Nos = 4;                    % Oversampling factor
Rs = Rbit*Nos;              % Input sample rate

The oversampling factor (Nos) is clearly stated as 4. And the documentation says "The input to the channel simulator is oversampled by a factor of four." However, later in the code, samples are generated at the symbol rate using the pskmod() function and fed directly into the channel step function. I see no palce where the input is oversampled (unless this is happening inside the channel step function). So where is the oversampling occuring?