The current deepest radio surveys detect hundreds of sources per square degree below 0.1 mJy. While the exact proportion is debated in the literature it seems that there is a growing consensus that a large fraction of these sources are dominated by star formation. However, the faintness of these galaxies at most other wavelengths makes determining the nature of individual sources difficult. If future, deeper surveys performed with the next generation of radio instrumentation are to reap high scientific reward we need to develop reliable methods of discrimination between radio emission powered

by AGN and that powered by star formation. In particular, we believe that such discriminations should be based on purely radio, or relative to radio, diagnostics. These discriminators include radio morphology, radio spectral index, radio polarisation, variability, radio luminosity and flux density ratios compared to other wavelengths e.g. different parts of the IR regime.

We will present work detailing the advantages and limitations of these various diagnostics methods on current surveys. However, this work can already provide several insights into the star formation at high redshift. As well as reproducing the well known rise in the comoving SFR density seen at higher redshifts, we also see evidence for the continued dominance of LIRGs and ULIRGs to the total star forming budget across redshifts 1-3. Additionally, while we see that the IR-radio relation for SFGs does hold to high redshifts ($z\geq 1$) there is a mild deviation depending on the mid-IR waveband used and the range of IR SEDs found. We will discuss the possible reasons behind this change.

We shall also emphasise the importance of more accurately establish the radio luminosity/star formation rate relation in the local Universe over a wider range of luminosities and rest-frame frequencies.