Realtime Wideband Modem in the Cloud - Technology and Economics

Processing of wideband communications waveforms can be accomplished entirely in software using standard servers. The software-centric implementation dramatically reduces development time and maintenance costs. Modern servers can sustain real-time processing of 100s and even 1000s of mega-samples per second, which is sufficient for almost all fielded systems. This paper discusses the technical challenges and economic considerations for implementing the real-time processing on a commercial cloud (AWS). The signal processing aspects have already been covered in author's previous papers [1]–[3]. The cloud implementation brings up multiple data distribution and networking tradeoffs. The multicast-based architecture presented in earlier papers turns out to be unsuitable for the cloud. Having shown the feasibility of the architecture, we turn our attention to the economics of it. The paper presents benchmarking results that quantify the cost per hour per megahertz for different instance types. We also discuss the costs associated with creating and maintaining a high-rate link to move raw digitized RF samples from the antenna to the cloud. These costs have a direct influence of the feasibility of enhanced resilience by geographic redundancy.