Plenary Speaker at ITW 2022 Mumbai

Careful joint constructions of codes and decoders generally presuppose, often implicitly, isotropic IID models of noise, sublimated into algebraic notions, such as minimum distance. The difficulty of the design of codes and decoders is such that it has to a large extent dominated the construction of the communications architecture stack, from modulation to detection. In this talk, we envisage a different philosophy. We consider taking as a starting point that the core parameter in operating communications system is the effect of the channel characteristics, such as the noise, which are beyond the engineer’s control and often highly time-varying, and which should be considered in terms of their realizations, not their average behavior. The construction of the code can be in effect relegated merely to that of a good hash for verification of the validity of a codeword. In that case, the decoder goes from being code-centric to noise centric. We provide some examples of that philosophy, from guessing random additive noise decoding (GRAND) to noise recycling, that use the real-time realizations of the noise and any statistical knowledge of it, in a code-agnostic way. We argue that the stack can becomes simplified, more modular and far more efficient. All work is joint with Ken Duffy.