Power allocation in NOMA systems for example on the basis Proportional Fairness calculations depends on knowledge of user throughput on a specified sub-band, which implies that users have already been allocated to particular sub-bands. Meanwhile, maximum throughput can generally be achieved where there is the greatest possible difference in transmission power for the users on a given sub-band, so that optimal allocation of users to sub-bands requires knowledge of the power available for each user. A mechanism is proposed based on iteratively applying a waterfilling algorithm to distribute power across a progressive subset of sub-bands to provisionally distribute the power budget across that subset of sub-bands, where at each iteration the water filling algorithm is carried out for each possible combination of users assignable to the newly considered sub-band using a floor for that sub band proportional to the reciprocal of the square of the highest channel gain value of any user in that combination, and calculating a throughput for that combination with the corresponding power attribution, whereby the combination retained for the next iteration (with an additional sub-band) is whichever gives the highest throughput. This process is thus repeated until users are assigned to all sub-bands, whereupon power allocations from the last iteration are definitive.