hello...(this is -obiously- my first post here; I am quite new to psychoacoustic and audio coding, but I guess that only mean there is lots of room for improvement )

I made an implementation of the basic AAC psychoacoustical model in MATLAB. It takes a audio signal (.wav) and give back (what is supposed to be) the minumum audible noise spectral power density (equivalent to the SMR?). It seems to work (the curves does look the way they should), but my problem is how to test it more accurately...

I thought of generating a "just audible colored noise" using the power spectrum I got, but I am not sure of how to do that. What I tried is:
I generate a white noise, calculate it's SDP (constant, as it's a white noise), multiply it by the SDP I want (the one from my psychoacoustical model), re-generate a complex signal using that multiplyed SDP square root as norm and the phase of my white noise as angle (our ear is suppose to be insensible to signal phase right?). So I now have a signal in frequency known to have the wanted energy spectrum; I take it's inverse fourrier transform, which gives a complex signal in time... now what do I do with that complex signal? If I take it's real part, it seems to have the right spectrum, but...
Is that correct? do you know a standard way of generating a colored noise having a given energy spectrum?

But even if I got my just audible noise, to test it I got to play both signal (my original .wav and that noise, which I am expected not to hear) at the same time. how do I do that with MATLAB? It'snot equivalent to adding them, right?

Is there some other way of testing my psychoacoustical model? Some artifact/effect that would be caracteristic?

Thanks a lot!

You are trying way too hard

Treat your SPD as the frequency domain of an FIR filter. Convert it to an impulse train (time domain) and convolve it against your white noise source. Adjust to scale.