Semantic communications look great in simulations and fall apart in the real world — researchers now have a clearer idea why.
A new paper examines semantic communication systems built on MIMO and OFDM — two established wireless technologies — and finds that the gap between theoretical and real-world performance comes down to a handful of physical-layer problems. Power amplifiers distort signals in nonlinear ways, peak-to-average power ratios fluctuate unpredictably, and actual wireless channels are frequency-selective in ways that tidy simulation environments ignore. The researchers argue that the channel's frequency selectivity is the single biggest culprit, and that targeted fixes can bring real deployments closer to the numbers researchers publish.
Semantic communications take a different approach from conventional wireless: instead of sending raw bits, the system tries to transmit meaning — compressing source data, channel coding, and modulation into a jointly optimized pipeline. That efficiency promise is exactly why wireless researchers have poured effort into it. But a technique that only works in a lab has no commercial path, and this analysis is a useful reminder that the physics of real antennas and amplifiers don't care about benchmark scores.
The field has seen this movie before — deep learning-based channel estimation hit similar walls when lab prototypes met actual interference. Closing the sim-to-real gap is unglamorous work, but it is the work that actually matters.