VHD solutions focus on energy efficient networks (Tuysuz and Trestian 2017). Most of these solutions used IEEE 802.21 MIH (Iqbal et al. 2019) and ANDSF (Access Network Discovery and Selection Function) protocols to collect information. Moreover, existing VHD systems make it possible to save energy by minimizing the scanning/detection time needed to discover the wireless network or to select the access point most economical with energy (Xenakis et al. 2011).
1.3. Spectrum handoff in the IoT
To avoid the scarcity of frequencies introduced above, the IoT can use the concept of intelligent radio. Many researchers have studied problems linked to spectrum management and channel allocation (Wu et al. 2014b; Koushik et al. 2018; Tarek et al. 2020). In Kumar et al. (2016), the authors studied spectrum handoff schemas in a CRN context and identified three types of transfer. First of all, the reactive approach, where the spectrum handoff and reconfiguration of the radio frequency happen after the primary user has been detected. Then, there is the proactive approach, where the spectrum handoff and frequency reconfiguration take place before a primary user has occupied the channel. Finally, the hybrid approach combines the two previous approaches. Recently, intelligent radio has been used in the IoT, especially in the two vehicular ad-hoc networks. This use has made it possible to improve the network’s spectrum efficiency and the experience of itinerant users by optimizing vehicle communication capacities and the QoS of applications such as road safety and traveler entertainment (Singh et al. 2014; Kumar et al. 2017).
To ensure continuity of service, we have opted for a proactive approach using a prediction module tasked with calculating the probability of future channel availability as well as the average time it is available.
1.4. Multicriteria decision-making module for an effective spectrum handoff in the IoT
In the IoT, the multicriteria solution seems most appropriate for making decisions about VHD, and by considering a great deal of contextual information to ensure effective selection of an access network/radio channel. To overcome the complexity of implementation, we opt for a cost function solution that calculates different candidates’ scores (access network or radio channel) and selects the one with the highest score. In the approach retained, we focus both on the QoS and on energy consumption, with the weight attributed to these parameters depending on the general communication context (network, application, object and user). Contextual information from the network is collected with the help of the surveillance module (context of multiple access networks) and of the intelligent radio-detection module (CRN context).
1.4.1. General architecture
In this section, we detail the general architecture we propose (Figure 1.1). In the approach we have retained, based on costs, the decision-making mechanism makes it possible to select the access network best adapted in the case of an object with several network interfaces and, on the other hand, to adapt the transmission parameters (channel, frequency, modulation, etc.) in the case of an intelligent radio network context. This selection will be based on the information available about the application, the user, the current radio conditions and predictions about how they will evolve. Information on the radio conditions is provided by the radio channel detection module (Akyildiz et al. 2006) in the CRN context and by the network monitoring module in the context of Multiple Radio Access Networks (M-RAN). The multicriteria decision-making module and the other general architecture modules are implemented within the object, which may be a car in the case of vehicular networks.
Figure 1.1. Proposed architecture for a context aware IoT device/object. For a color version of this figure, see www.iste.co.uk/chalouf/intelligent.zip
1.4.1.1. Detection for an intelligent radio module
In the context of CRNs, the detection module provides information on the accessible radio channels, their quality and their rates of occupation. A candidate radio channel (noted in the CR channel suite) is characterized by a central (fixed) frequency, a passband width (fixed), modulations and schemas for possible codings (fixed), a probability of availability (predicted), an average availability time (predicted), an energy cost (estimated), a packet loss rate (estimated), a passband (fixed), an average delay (estimated) and an average jitter (estimated).
1.4.1.2. Prediction module
The radio environment is not stable and the radio parameters may change for many reasons such as mobility and the arrival or departure of other objects. Thus, an object may need to change access network or transmission parameter values. To avoid delays and disruptions to transmission, unusable spectrum handoffs, we use a prediction module. This module will predict future variations of some parameters to anticipate decisions on spectrum handoff and carry this out at an opportune time. In the CRN context, a period could be equal to the time interval allocated to an object to carry out spectrum detection operations, signaling and the transfer of data along a given channel. Thus, when a degradation of the QoS is perceived for the current period (t), the decision to change channel is taken for the future period (t + 1).
At each period t, the prediction module calculates and predicts some transmission parameters for the period t + 1. If the transmission conditions are satisfactory for the period t + 1, then it is not necessary to change channels. Otherwise, the object considered should modify the transmission parameters to guarantee the required QoS.
The process of prediction may require significant resources (CPU, memory and energy). Consequently, implementation of the prediction module in the object may be very restrictive, indeed impossible. Thus, we suggest migrating the prediction process to the Fog radio access network. This makes it possible to improve the object’s capacity, to economize on the battery and to provide the user with a better experience.
The parameters that will be predicted at the period t + 1 are channel availability and the average channel availability time.
1.4.1.2.1. Probability of channel availability at period t + 1
In a CRN context, according to the analytical model defined in Song and Xie (2012), the predicted probability that a candidate channel i is inactive at time t can be expressed as follows (equation [1.1]):
where
The probability that the channel i is in active at an instant t can be obtained by equation [1.2]:
where
