The bit-error rate (BER) performance of a pulse position modulation (PPM) scheme for non-line-of-sight indoor optical links employing channel equalisation based on the artificial neural network (ANN) is reported. Channel equalisation is achieved by training a multilayer perceptrons ANN. A comparative study of the unequalised `soft' decision decoding and the `hard' decision decoding along with the neural equalised `soft' decision decoding is presented for different bit resolutions for optical channels with different delay spread. We show that the unequalised `hard' decision decoding performs the worst for all values of normalised delayed spread, becoming impractical beyond a normalised delayed spread of 0.6. However, `soft' decision decoding with/without equalisation displays relatively improved performance for all values of the delay spread. The study shows that for a highly diffuse channel, the signal-to-noise ratio requirement to achieve a BER of 10âˆ’5 for the ANN-based equaliser is ~10 dB lower compared with the unequalised `soft' decoding for 16-PPM at a data rate of 155 Mbps. Our results indicate that for all range of delay spread, neural network equalisation is an effective tool of mitigating the inter-symbol interference.