A digital over-the-air computation system based on maximum posterior probability demodulation

Abstract

The number of edge devices (EDs) has grown exponentially with the development of technologies such as the Internet of Things, distributed consensus, federated learning, and sensor networks. Physical communication between edge devices and fusion centers (FCs) has become a bottleneck affecting the efficiency of data aggregation. To overcome this bottleneck, an efficient broadband analog transmission scheme has recently been proposed named over-the-air computation (OAC), featuring data aggregation via the waveform superposition property of the wireless channel. The OAC system effectively improves the efficiency of data fusion and reduces the consumption of hardware and arithmetic resources. However, analog communications have poor communication quality and are difficult to deploy. While existing digital OAC solutions are not widely available due to the limitations of demodulation schemes. To solve this problem, we propose a digital OAC system based on maximum a posteriori probabilistic demodulation. The proposed scheme has higher generalizability. In addition, we propose a frequency offset estimation algorithm for oriented OAC uplink and a three-stage handshake compensation scheme to eliminate the effect of frequency offset on OAC systems. This frequency synchronization scheme can effectively eliminate the residual frequency offsets caused by channel imperfect reciprocity and improve the calculation accuracy of OAC. Experimentally, it is proved that the proposed scheme has higher accuracy, especially at low signal-to-noise ratios, with an average improvement rate of 56.8% compared with the existing scheme. Although the complexity of the proposed frequency synchronization scheme slightly increases, this is perfectly acceptable considering the light weight of the proposed demodulation scheme and the increase in system accuracy.