Am i using the correct network type? Network accuracy not improving.

So i am currently training a neural network which will eventually take the place of a series of functions in a very large matlab script.

The current goal of the network is to take 2 inputs, and from these gain 2 outputs (MIMO). I get very good accuracy if i use a narxnet, however i cannot implement this in the properly in the script it has to slot into.

Since the feed forward net can be implemented in the main script, I am currently using this. However, as it stands the network is massively simplifying, with the same value for a huge range of inputs, and regardless of how accurrate i attempt to set the target to.

I think i am implementing correctly, so am i using the wrong type of network for my problem?

Input and Target datasets are two 2x79999 matrices.

X = tonndata(Input,true,false);
T = tonndata(Target,true,false);
net = timedelaynet(1:2,75);
view(net)
[Xs,Xi,Ai,Ts] = preparets(net,X,T);
net.performFcn='msereg'; 
net.performParam.ratio=0.1; 
[net,tr] = train(net,Xs,Ts,Xi,Ai);