How do I convert Noise (A-weighted) -102.5 dBu to THD %rating ??
You don't. For a range reasons.
THD-N is total harmonic distortion plus noise. THD is just the harmonic distortion, and by definition does not include noise. (In addition, using a weighting means the noise has had a frequency dependant function applied to as part of creating a single figure of merit. This is a one way operation, which means you can't reverse out any frequency dependencies.)
Harmonic distortion and noise are often lumped together as
noise, but the names are deceptive, and don't fully convey what is occurring.
Noise is in its broadest sense anything we don't want in the signal. But we can divide than down into stuff that is dependant on the signal itself, and stuff that is independent of the signal. We can then call the stuff that is dependant on the signal "distortion" and the stuff that is independent of the signal "noise". Harmonic distortion is the most popular and generally useful way of measuring disortion (although it isn't the only way and it doesn't capture all possible mechanisms), and if you want a single figure of merit you add the noise figure onto THD, and get THD-N. aka
SIgnal to
Noise
And
Distortion. SINAD. SINAD implicitly admits other distortion than just harmonic, although it is usually a synonym for THD-N.
Noise can be divided into correlated noise - that is noise that has some internal structure - such as mains hum, and non-correlated noise, which is at its base AIWN, additive independent white noise. Such noise is stochastic in nature, and in electronic systems is usually a mix of Johnson-Nyquist noise, shot noise, and 1/f noise (aka flicker noise). These noises are not necessarily individually white, (1/f noise is pink) but they have the property that you cannot predict what is to come next based on what has come before, so they are not correlated, and can be considered as filtered white noise). Generally they come under the term "hiss". Our ear has different frequency sensitivities, so noise has more or less efffect in different frequency bands, so applying a frequency dependant function to the noise spectrum when the power in the noise is summed is a useful thing to do. Hence the common A-weighted noise figure.