Indoor Experimental Trial on AI-Based Pilotless Transmission

This paper presents a pilotless transmission scheme that combines a learned constellation with a neural receiver, and validates the approach through over-the-air indoor experiments. Motivated by the need for evidence beyond simulation when applying artificial intelligence/machine learning at the air interface, we first use a channel emulator to establish basic characteristics and to evaluate a wide range of parameters including channel-model family, delay spread, signal-to-noise ratio, and velocity. The channel emulator-based experiments show that the proposed pilotless scheme achieves a block error rate (BLER) generally comparable to a 5G New Radio baseline. As the BLERs converge, eliminating pilots (demodulation-reference signals (DMRSs)) overhead provides the expected throughput gain. Moreover, under strong time- and frequency-selective conditions, the conventional pipeline typically requires DMRS pattern optimization, whereas the proposed neural receiver maintains its performance level without such tuning. We then validate the approach in real indoor channels. Over-the-air measurements in an obstructed room, an open lobby, and a corridor enclosed by walls, floor, and ceiling with partial non-line-of-sight yield throughput gains of 6.5% to 18.2%.