Per probe normalisation is a way of adjusting a set of quantitations to emphasise the differences between different data stores, whilst removing any overall differences in average quantitation level.
The quantitation adjustment works by adjusting the medians of each probe to be the same across the set of stores which are being normalised.
You can choose how this normalisation is applied. You can either subtract the median value from each individual value, or you can divide by the median and log2 transform the result. You should choose subtract if you data was already quantitated on a log scale, and divide if it wasn't. In each case your probes will have a median of 0. Alternatively you can scale your data so that each probe has a value between 0 and 1 where 0 is the lowest value for that probe across all datasets, and 1 is the highest value.