We have investigated the role that different square wave signals play in the dynamics of a time delay network of McCulloch-Pitts elements by computer simulations. The different cycle (positive and negative) square wave signals induce the following network features: signals, whose cycles exactly divide the delay time, give rise to system memory; on the other hand, signals, whose cycles do not have the numerical relationship mentioned above with the delay time or are larger than it, give rise to disordered integral activity of the network, or patterns which cannot exhibit the step of the inputs after the inputs are eliminated. Furthermore, we have studied whether this network can discriminate the difference in cycle between two square wave signals applied sequentially to it. The results show that the network can only distinguish the difference in cycle between signals such that the first one can be remembered.