A Time Series Model for Communication Scenario Recognition with Weighted Correction

Wireless communication is in strong demand of high throughput, low latency and high reliability, while mobility brings challenges to these demands. In particular, dynamic communication scenarios bring difficulty to communication devices in adjusting transmission strategies accordingly. Communication scenario recognition provides specific information of the environ-ment to mobile devices to make more intelligent communication strategies. We propose a long short-term memory (LSTM)-based model called SERA, which exploits the temporal relationship between the samples of multivariate data collected by mobile devices. In the proposed model, we have designed weighted correction (WCR) as a real-time post-processing to smooth the recognition results. Furthermore, we implement extensive performance analyses and simulations to validate SERA. Our experiments with real-world data demonstrate a 20% improve-ment in recognition performance over existing benchmarks.