Realistic traffic modeling plays a key role in efficient Dynamic Spectrum Access (DSA) which is considered as enabler for the employment of wireless technologies in critical industrial automation applications (IAA). The majority of models of spectrum usage are not suitable for this specific use case as they are based on measurement campaigns conducted in urban or controlled laboratory environments. In this work we present a time-domain traffic model for industrial communication in the 2.4 GHz industrial, scientific, medical (ISM) band based on measurements in an industrial automotive production site. As DSA is usually implemented on Software Defined Radios (SDR), our measurement campaign is based on SDR platforms rather than sophisticated spectrum analyzers. We show through the estimation of the Hurst parameter that industrial wireless traffic possesses inherent self-similarity that could be exploited for efficient DSA. We also show that wireless traffic could be modeled as a semi-Markov model with channel on and off durations Log-normally and Pareto distributed, respectively. We finally estimate the parameters of the derived models using Maximum Likelihood estimation.
Cognitive radio (CR) is a key enabler of wireless in industrial applications especially for those with strict quality-of-service (QoS) requirements. The cornerstone of CR is spectrum occupancy prediction that enables agile and proactive spectrum access and efficient utilization of spectral resources. Hidden Markov Models (HMM) provide powerful and flexible tools for statistical spectrum prediction. In this paper we introduce a HMM-based spectrum prediction algorithm for industrial applications that accurately predicts multiple slots in the future. Traditional HMM prediction approaches use two hidden states enabling the prediction of only one step ahead in the future. This one step is most often not enough due to internal hardware delays that render it outdated. We show in this work that extending the number of hidden states and formulating the prediction problem as a maximum likelihood (ML) classification approach enables a prediction span of multiple slots in the future even with fine spectrum sensing resolution. We verify the suitability of our approach to industrial wireless through extensive simulations that utilize a realistic measurement-based traffic model specifically tailored for industrial automotive settings.
The cornerstone of cognitive systems is environment awareness which enables agile and adaptive use of channel resources. Whitespace prediction based on learning the statistics of the wireless traffic has proven to be a powerful tool to achieve such awareness. In this paper, we propose a novel Hidden Markov Model (HMM) based spectrum learning and prediction approach which accurately estimates the exact length of the whitespace in WiFi channels within the shared industrial scientific medical ISM) bands. We show that extending the number of hidden states and formulating the prediction problem as a maximum likelihood (ML) classification leads to a substantial increase in the prediction horizon compared to classical approaches that predict the immediate (short-term) future. We verify the proposed algorithm through simulations which utilize a model for WiFi traffic based on extensive measurement campaigns.