Linear Neural Network poor performance

I have been asked to build a linear neural network with a single hidden layer. All the layers (input, hidden and output) should have 3 nodes. So I have written the following code so far, with P and T a sample of the dataset:

P = [1.7500    1.8000    2.4000    1.4000; 3.3500    3.3000    3.3000    4.0000 ;4.2000    3.9500    2.5500    6.8000];
T = [0     0     0     1;  1     1     0     0; 0     0     1     0];


function testFF(P, T)

net = newff(P,T,3, {'purelin'});
net.trainParam.showCommandLine = 1;
net.trainParam.goal = 0.0001;
net.trainParam.lr = 0.1;
net.trainFcn = 'traingd';
net = train(net,P,T);

% Check network performance on training patterns.
EstimatedTrainingTargets = sim(net,P);
EstimatedTrainingTargets = round(EstimatedTrainingTargets);
Differences = abs(EstimatedTrainingTargets - T);
CorrectTrainClassificationRatio = 1 - (sum(sum(Differences)) / (2*size(P,2)))

First of all, is the above a valid implementation of a linear neural network?

Secondly, after I feed the input and output matrices, whose size is 3 X 22000, the networks prediction isnt accurate at all , for example for the first 4 inputs as given above, the output is:

 0     0     0     0     
 0     0     0     0    
 0     0     0     0    

and the CorrectTrainClassificationRatio for the whole data set is 0.4713.

I have tried different learning rates, training functions and parameters goals but the outcome doesn't change by much. How can I get better predictions?