In other recent posts such as this one, I detailed partial response signaling common to GMSK where a known inter-symbol interference (ISI) is intentionally introduced in the time domain at the benefit of increasing the data rate for a given spectral occupancy (spectral efficiency), with the only down-side being receiver complexity: we can completely demodulate the received waveform without added distortion by correlating to every possible combination of patterns over the finite memory of the ISI (done efficiently with a Viterbi decoder). "Faster than Nyquist" signaling follows similar thoughts in allowing for intentional ISI.
This then makes me think a similar operation can be done in the frequency domain, for the same purpose- by intentionally introducing inter-carrier-interference (ICI) and then following the same logic in terms of cost and benefit as we do in partial response signaling (swapping time and frequency domains).
Does this already exist or have been pursued in any similar form? Or is there a fundamental flaw in my thought process that keeps this from being a viable solution? Certainly it may not fit well with OFDM as currently done with standard FFT and IFFT processing, so I don't want to discount it on that thought alone which is limited to the specific algorithm rather than the result.