Final Report Summary - IRMA (Information Role Models with Applications)
The design of estimators and detectors is well covered by the Bayesian probabilistic framework. This framework is used extensively in the design of communication devices, where it is applied to every component in a receiver, i.e. channel estimation, detection, decoding, synchronisation, etc. In theory, designing an optimal detector/decoder for a quantity based on a given observation boils down to computing an a-posteriori probability distribution. In reality, this distribution is often either unknown, or too complex for practical use. It is common practice to devise detours and shortcuts, mostly based on heuristic reasoning, to provide low-complexity practical alternatives to the direct application of Bayesian theory.
We have studied one such alternative, which we call the 'role model' approach. In this approach, a detector is optimised by mimicking the output of a superior 'role model' detector with access to better observations. In preliminary work, we have demonstrated that the role-model approach is optimal, i.e. converges to the Bayesian solution, under certain conditions for the observations available and for those of the role model detector. We have also found that, unlike the direct Bayesian solution, the role-model approach yields a convex optimisation problem that can be solved using well- established numerical techniques, giving it a considerable practical advantage.
During the course of this fellowship, the project has achieved the following objectives:
- to extend the theoretical foundations for the role-model approach;
- to refine the application of the role-model approach to problems in receiver design;
- to establish connections between the role-model approach and other known problems of information theory and digital communications;
- to investigate applications of the role-model approach outside the field of communications, in particular to incomplete data problems.
The project has achieved significant results in the characterisation of the performance of iteratively decoded codes with suboptimal decoders. These results are significant, since they allow an accurate performance characterisation (in terms of extrinsic information transfer (EXIT) charts) of the decoding trajectories that the iterative decoder follows. This is in contrast to previous work in the literature where the performance of suboptimal receivers was only approximated with some degree of success. The fellow has recently discovered the application to characterise the performance of bit-interleaved coded modulation (BICM) using the role model approach. BICM is the de facto standard technique for combining coding and modulation in current and upcoming telecommunications systems.
We have studied one such alternative, which we call the 'role model' approach. In this approach, a detector is optimised by mimicking the output of a superior 'role model' detector with access to better observations. In preliminary work, we have demonstrated that the role-model approach is optimal, i.e. converges to the Bayesian solution, under certain conditions for the observations available and for those of the role model detector. We have also found that, unlike the direct Bayesian solution, the role-model approach yields a convex optimisation problem that can be solved using well- established numerical techniques, giving it a considerable practical advantage.
During the course of this fellowship, the project has achieved the following objectives:
- to extend the theoretical foundations for the role-model approach;
- to refine the application of the role-model approach to problems in receiver design;
- to establish connections between the role-model approach and other known problems of information theory and digital communications;
- to investigate applications of the role-model approach outside the field of communications, in particular to incomplete data problems.
The project has achieved significant results in the characterisation of the performance of iteratively decoded codes with suboptimal decoders. These results are significant, since they allow an accurate performance characterisation (in terms of extrinsic information transfer (EXIT) charts) of the decoding trajectories that the iterative decoder follows. This is in contrast to previous work in the literature where the performance of suboptimal receivers was only approximated with some degree of success. The fellow has recently discovered the application to characterise the performance of bit-interleaved coded modulation (BICM) using the role model approach. BICM is the de facto standard technique for combining coding and modulation in current and upcoming telecommunications systems.