Abstract
Code Division Multiple Access (CDMA) techniques afford Local Area Networks (LANs) the support of concurrent, asynchronous communication between users without access delay. These properties are obtained by encoding users' data with high rate code sequences, so that data is spread over a much larger bandwidth than would usually be required for transmission. The necessary bandwidth is provided by using optical fibre both as the LAN medium and for incoherent optical signal processing. Conventionally, extraction of a desired user's signal is achieved by correlation using a single delay-line matched filter. Matched filters are optimal for the recovery of a known signal in the presence of additive noise. However, in a CDMA environment, their performance is limited by Multiple Access Interference (MAI), arising from the cross-correlation products of overlaid users, and degrades as the number of users increases. Adaptive Logic Networks (ALNs), a form of Artificial Neural System (ANS), are applied to the extraction of a single user's signal in a multi-user environment. In the approach taken, ALNs learn to incorporate the presence of interfering users' signals, in deciding the actual data bit received. Computer simulation is used to compare the error rates obtained by ALNs and the previously proposed correlation receivers; the performance of the latter providing a benchmark. Simulations are conducted assuming chip synchronism between users and no external sources of noise, i.e. MAI is assumed dominant. Consideration is given to systems employing both sparse optical codes and Gold-like codes as spreading sequences. In all the systems considered, ALNs are shown to enable significant reductions in error rate over the conventional correlation receivers. MAI effects, causing errors with the correlation receivers, are reduced by using additional temporal and intensity based information contained in the receiver input signal. This permits an ALN to extract details of the structure of interfering users' signals, to provide a better context for the classification of the desired user's signal. In the systems employing sparse codes, it is demonstrated that while a certain amount of MAI persists, it may be minimised by selection of the ALN input window, to provide the maximum possible information regarding the interfering users' signals. In the systems using Gold-like codes, it is shown that ALNs can be used to completely eliminate the effects of MAI. This is significant since, although this form of code sequence is suited to coherent CDMA systems, the cross-correlation products arising in incoherent optical environments are normally considered to be unacceptably high.