Wireless traffic usage forecasting methods can help to facilitate proactive resource allocation solutions in cloud managed wireless networks. In this paper, we present temporal and spatial analysis of network traffic using real traffic data of an enterprise network comprising 470 access points (APs). We classify and separate APs into different groups according to their traffic usage patterns. We study various statistical properties of traffic data, such as auto-correlations and cross-correlations within and across different groups of APs. Our analysis shows that the group of APs with high traffic utilization have strong seasonality patterns. However, there are also APs with no such seasonal patterns. We also study the relation between number of connected users and traffic generated, and show that more connected users do not always mean more traffic data, and vice versa. We use Holt-Winters, seasonal autoregressive integrated moving average (SARIMA), long short-term memory (LSTM), gated recurrent unit (GRU) and convolutional neural network (CNN) methods for forecasting traffic usage. Our results show that there is no single universal best method that can forecast traffic usage of every AP in an enterprise wireless network. The combined models such as CNN-LSTM and CNN-GRU are also used for spatio-temporal forecasting of a single AP traffic usage. The results show that considering spatial dependencies of neighboring APs can improve the forecasting performance of a single AP if it has significant spatial correlations.

Spectrum sharing in radar bands with interference forecasting for enhanced radar protection can help design proactive resource allocation solutions which can achieve high data rates for wireless communication networks on one hand and help protect the incumbent radar systems. We consider radar spectrum sharing in 5.6GHz where a weather radar operates as a primary system and the dominant secondary system is an enterprise network consisting of access points (APs) in a university campus. Our work models transmit the power of the APs as a time series with multinomial distribution based on real collected data. The aggregated interference due to the transmissions from the APs at the radar is forecasted using a long short-term memory (LSTM) based neural network. Monte Carlo dropout is utilized to generate prediction intervals that capture the uncertainties in the interference from the APs. Finally, by using both average and upper limits of predicted interference time series a cloud-assisted efficient sharing and radar protection algorithm is proposed. Tracking the rotating radar is not required in the proposed system. The results show that the proposed efficient sharing and radar protection system ensures better radar protection and increased throughput for wireless communication users.

Prediction of wireless network parameters, such as traffic (TU) and channel utilization (CU) data, can help in proactive resource allocation to handle the increasing amount of devices in an enterprise network. In this work, we examined the medium-to-long-scale forecasting of TU and CU data collected from an enterprise network using classical methods, such as Holt-Winters, Seasonal ARIMA (SARIMA), and machine learning methods, such as long short-term memory (LSTM) and gated recurrent unit (GRU). We also improved the performance of conventional LSTM and GRU for time series forecasting by proposing features-like grid training data structure which uses older historical data as features. The wireless network time series pre-processing methods and the verification methods are presented as time series analysis steps. The model hyper-parameters selections process and the comparison of different forecasting models are also provided. This work has proven that physical layer data has more predictive power in time series forecasting aspect with all forecasting models.

Interference prediction with neural networks (NNs) can help in proactive resource management for spectrum sharing in 5.6 GHz radar bands. This can achieve high data rates for secondary users (SUs) of the shared spectrum and enhance the protection of the incumbent radar systems. The recently introduced efficient sharing and radar protection (ESRP) system with interference prediction used NN-based long short-term memory (LSTM) and Monte Carlo (MC) dropout to utilize the uncertainties in the interference from the access point (APs). Due to the random nature of radio propagation, the permissible probability of harmful interference at the radar (ε p ) for the ESRP system varies depending on the MC dropout and prediction intervals (PIs) which represents the amount of uncertainty captured in the system. In this work, we use a gated recurrent unit (GRU) which is simpler and faster than LSTM for interference prediction. We also investigate how the different MC dropout values can vary the parameter ε p and improve the radar protection performance of the ESRP system. The results show that radar protection performance can increase by using GRU and the values of MC dropout play an important role in the ESRP system ensuring better radar protection with a small trade-off for throughput of the SUs which are the APs.