Federated learning enables intelligent reflecting surface in fog-cloud enabled cellular network

Introduction

The usage of healthcare sensors has been growing in network capacity and ubiquitous wireless networking recognition to numerous leading enabling technologies such as the ultra-dense network (Chen et al., 2021). The quarry of 100 billion healthcare sensors devices by the upcoming fifth-generation (5G) wireless network has been widely met (UDN), Huge multi-input, multi-output system (MIMO) and communication using millimetre waves (mmWave) (Kousiouris et al., 2019). For the healthcare system, the distributed fog cloud implemented at different communication networks. However, the high complexity and hardware costs and the increased energy demand are still open questions. Densely installing base stations (BSs) or access points (APs) in a UDN, for example, not only increases hardware and maintenance costs but also exacerbates the network interference issue. Extending large MIMO from sub-6 GHz to mm Wave frequency bands is also typically expensive (Zhao et al., 2020). As a result, advanced, spectral, and energy-efficient but cost-effective solutions for future/beyond-5G wireless networks still needed for the healthcare sensors network (Qu et al., 2020).

The FL-IRSTS system can be intellectuals controlled by a cleverly reflecting surface (IRS) to extend flag concentrated obtained at the goal. This is often in stark differentiate to past procedures that made strides in remote communications by tweaking the sender or recipient. An IRS made up of a few IRS units, each of which can speak to the occurrence flag at a diverse point. The remote flag voyages from the source to the IRS optimized the goal in such IRS-assisted communications. Such a communication approach is precious when there’s a weak remote channel between the origin and destination due to boundaries or destitute natural conditions, or when they don’t have a clear line of location (Zhou et al., 2020). The IRS anticipated playing a noteworthy part in 6G systems due to their capacity to design remote situations, agreeing to numerous remote communications specialists. NTT DoCoMo and MetaWave, a startup in Japan, illustrated the utilize of IRS-like innovation for helping remote communications within the 28 GHz band. The expansive MIMO innovation utilized in 5G communications compared to IRS reflects remote signals. Hence, employments less control, whereas tremendous MIMO transmits signals and needs a parcel more control (Saha, Misra & Deb, 2020).

The computational offloading of IoT healthcare sensors, combined with IRS-assisted communications, enables a fog-cloud network in this paper. In the context of IRS-assisted wireless communications, we address the issue of Internet of Things healthcare sensors offloading computation tasks to a base station fitted with fog nodes. IoT healthcare sensors in the study plan to send computation tasks to the base station through wireless channels, with an IRS assisting communication between them. As compared to the art of studies, the study has the following contributions.

• The study devises a new IRS that enables the energy-efficient and cost-efficient distributed system for healthcare applications in distributed fog cloud. The proposed system is different than (Saha, Misra & Deb, 2020; Hosseinalipour et al., 2020) studies where these studies proposed IRS aware energy-efficient computational offloading for healthcare applications. However, this study considered a more complex communication system as compared to existing studies.

• The study first creates an algorithm framework to detect feasibility when given an arbitrary set of information rate constraints to solve the energy and cost-efficient computational offloading problem. When we determine that the problem is feasible, the algorithm will also have a possible solution. Then we devise a solution that maximizes the earning potential of fog cloud nodes.

• The importance of IRS devices to gain significant numerical performance. According to the performance assessment and results from the discussion, IRS will substantially improve the feasibility of distributed healthcare fog-cloud networks for information rate constraints. Furthermore, numerical results show that our algorithm converges quickly and can significantly boost fog-cloud nodes to achieve energy and cost results.

The rest of the paper designs as follows. “Related Work” expresses the related work of studies to the task scheduling problem. “Problem Description” illustrates the problem under research works establishes it. The proposed algorithm framework and its components discussed in “Proposed FL-IRSTS Strategy Framework”. The performance evaluation of algorithms defined in “Performance Evaluation and Results”. Conclusion and future work of the paper enclosed in “Conclusion and Future Work”.

Related Work

In this session, the manuscript discussed the existing efforts in intelligent reflecting surface (IRS) for applications in terms of computational offloading in terms of computation and communication technologies. The main problem to be solved based on machine learning and federated learning. Furthermore, joint technologies such as fog-cloud computing and communication technology by intelligent reflecting surface (IRS) can give many advantages in practice to the sensor devices and healthcare applications. The related work part studies the efforts of both fog-cloud computational offloading and communication technology in IRS. Whereas, energy-efficient fog-cloud aware computational offloading aware framework suggested in Chen et al. (2021), Kousiouris et al. (2019), Zhao et al. (2020), Qu et al. (2020). In many studies Zhou et al. (2020), Saha, Misra & Deb (2020), fog-cloud paradigm aware system suggested to minimize end to end latency of healthcare applications. The works (Hosseinalipour et al., 2020; Sharma, Park & Cho, 2020) analyze a network with a large number of IoT sensors users and a single base station with a single fog node. By comparing the obtained utilities under the two conditions, mobile devices choose computation offloading to edge cloud or local computing on the devices. Both network communications use the same wireless channel and conflict with one another. The dynamics of offloading IoT healthcare sensor data offloading decision-making modelled as a game, then analyzed using future game theory. Instead of just one radio, discuss a wireless obstacle setting with several channels in Lakhan et al. (2021b).

Table 1 shows existing objectives, methods, centralized machine schemes, federated learning techniques and different offloading and scheduling suggestions in the literature work. These studies trained the mobility model of application in adaptive environment based on machine learning unsupervised learning. The single parameters enabled (Single-para) static model based on supervised learning widely implemented in existing IRS system. The goal was minimized the overall load balancing issues of the network. Whereas, multiple parameters enabled training model based on unsupervised without labeling for IRS system widely investigated in the defined studies in the table. The data uploading rate of a single healthcare sensor uses computation offloading compared to the data uploading rate of other sensors on the same wireless channel of the fog-enable have many base-stations aware system devised in Lakhan et al. (2021c). The goal is to minimize the energy efficiency of the sensors in the network. The work (Lakhan & Li, 2020) addresses the computation offloading of a power system to a fog node, where the addition of energy-efficient communicating nodes and complicates the computation offloading policy design. The latency and energy-efficient aware computation offloading for edge computing in MIMO the multicell scheme studied recently. The goal is to transmit precoding matrices of mobile devices, as well as computational resources assigned by the edge cloud to the devices, which are included in the optimization variables to minimize mobile devices’ energy consumption (Mujeeb-ur Rehman et al., 2021). IRS-aided communications have recently received a lot of attention in the literature. IRS support transmissions from IoT sensors to base stations because our work also focuses on uplinks since IRSs can manage to represent incident wireless signals in the desired way. In Lakhan et al. (2021a), Chen, Liu & Peng (2019), Lakhan & Xiaoping (2018) investigate the effect of channel estimation errors on uplink data speeds as defined above. In addition to the uplink studies mentioned above, downlinks examined in Lakhan & Li (2019) for IRS assisting communications between mobile devices and base stations. The fact that there seem to be more downlink studies than uplink studies may explain that the base station will centralize transmit beamforming optimization in the former case. The direct communications between healthcare sensors and IRS widely improve the communication capacity during offloading and downloading.

To our knowledge, no research has been done on the energy and delay-effectiveness of IoT healthcare sensors in distributed fog-cloud networks with IRS. Existing studies used the IRS in the fog computing system to solve the latency minimization problem. This paper proposes a novel IRS for a fog-cloud-assisted healthcare sensor system, in which the cost and energy efficiency of nodes are optimized.

Problem Description

The study devises novel federated learning awareness, and IRS enable base-station assisted of fog nodes at the user the network as shown in Fig. 1. The system consists of different components which are accountable for doing the entire process in the network. The user applications (app) are healthcare and connected to the proximity fog-enable base stations. All the base stations configured by the IRS controller in the cellular network with the one-kilometre range. The fog cloud agent (FCA) manages all the fog-nodes controller in the system.

Energy consumption

The system shown in Fig. 1 is ubiquitous and distributed with different nodes for healthcare applications. The energy model in IRS aware distributed base station is a bit complex. Therefore, the system measured the energy in various models in the system.

Energy consumption between sensors and base-station

To comprehend the issues with cellular B power consumption, one must first investigate the design of these systems and the power consumption of their components. Access links connect mobile stations to a core network through B. As shown in Fig. 2A, these stations cover a cell divided into several sectors, with each B surrounded by a sector antenna. According to their coverage area, cellular B divided into macro-, micro Femto (indoor), and pico B, with each cell having its scale, output capacity, and data rate (Lakhan et al., 2021c). Because of their limited coverage range and low radiation power demand, small BSs use less power (Lakhan & Li, 2020; Mujeeb-ur Rehman et al., 2021; Lakhan et al., 2021a). As shown in Fig. 2B, a macro-BS site usually consists of many pieces of power-consuming equipment. The operational power of the macro-BS can model as follows.

(1) $$\begin{array}{c}{B}_{Energy-1}=(E_{Sect}\times E_{TX}){\displaystyle \frac{P_{PA}+P_{BB}+P_{RF}}{(1-\lambda ES)(1-\lambda DC)(1-\lambda Cool)}}\\ +P_{wm}+P_{au}.\end{array}$$

In Eq. (1), PA, BB, RF implies the energy of the power amplifier (PA), the optical signal processing or baseband unit (BB), and the transceiver (RF). The outcome of PA is a linear function of B base station transmission energy E_TX, and defined as ${\displaystyle \frac{E_{TX}}{PA}}$ and it represents PA power Efficiency. Where, 1 − γES, 1 − γDC, 1 − γCool represent the losses sustained by the rectifier, regulator, and active cooling, which are scaled linearly with the power consumption of the other components, respectively. P_mw stands for auxiliary equipment, P_au is a backhaul such as lighting and closed-circuit healthcare nodes and denotes the microwave backhaul connection. Since a BS has several sectors and antennas, the power consumption of these components must be multiplied by the number of base stations E_Sect, and the B power consumption must be multiplied by the number of transmitting antennas E_TX for each sector.

Energy consumption between base-station to base-station

The model determines the energy-consumption of between base stations in the following way.

(2) $$B_{Energy-1,2}={\displaystyle \frac{W_i}{B_w}}$$

In Eq. (2), B_w is the available bandwidth between base stations to transfer the requests in the system.

(3) $$Mobility=BDensity\lambda {\displaystyle \frac{B^2}{k{m}^2}.}$$

Eq. (3) determines the Energy efficiency (Mbit/Joule) between base stations during mobility of the workloads.

Sum of energy consumption

The proposed system consists of different computing nodes such as fog nodes and cloud node, where all nodes are heterogenous in their properties. Therefore, the power consumption directly proportional to the speed of the node during execution in the system. The high power node consumed much more power as compared lower power node. The study determined the power consumption of the node in the following way.

(4) $$\begin{array}{c}E=\sum _{i=1}^I\sum _{k=1}^K\sum _{s=1}^S{x}_{isk}{\displaystyle \frac{W_i}{{\zeta }_k}\times T_i^e\times }\hfill \\ P_k+BT.P_W+TM.P_W+B_{Energy-1}+B_{Energy-1,2}.\hfill \end{array}$$

The Eq. (4) determined the power consumption of resource during processing of applications in the system.

Proposed FL-IRSTS Strategy Framework

The study devises Federated Learning Intelligent Reconfigurable Surface Task Scheduling (FL-IRSTS) algorithm framework consists of different dynamic heuristics to solve the problem into sub-steps as shown in Algorithm 1. The algorithm consists of other heuristics to improve training, latency efficiency, and energy efficiency to solve the problem as shown in Algorithm 1.

Algorithm 1 is the framework which consists of different strategies as discussed in problem description and solve the problem in different steps. Each strategy to be explained in different sub-sections.

Performance Evaluation and Results

This evaluation part conducted the experiments of healthcare workloads in the system. The existing approaches: Approach1 (Zhao et al. (2020); Saha, Misra & Deb, 2020; Hosseinalipour et al., 2020; Sharma, Park & Cho, 2020) and Approach2 (Kousiouris et al., 2019; Mujeeb-ur Rehman et al., 2021; Zhou et al., 2020) implemented that are nearly same structure as proposed scheme IRSTS framework in the system.

Simulation environment

The simulation environment for federated learning Intelligent Reflecting Surface is Matlab-based open-source and available on Github. We merged these techniques in Ifogsim simulator (Chen, Liu & Peng, 2019) to design the modified research environment for the applications. The study implemented the Arduino wireless-based such as sound, vibration, pressure, widely exploited in the simulation to generate data. Furthermore, MOCH healthcare sensors kit (Chen, Liu & Peng, 2019) was implemented with the Ifogsim and analyzed the performances of applications with the proposed schemes in the system. In the Ifogsim, we added the convolutional neural network-based training by using the loss function. For instance, each application is deployed at sensors level or mobile, and the loss function train all data locally on mobile devices. Then the trained model of data offloaded to the fog cloud agent (FCA). The FCA in Ifogsim is a global model where all local train models of data are shared to all local devices of applications.

Observation of federated learning-IRS

The study analyses the spike of workload requests to base stations as per B density λ (BS/km²) Per Requested Workloads in the simulation as shown in Fig. 3. The density λ offers the arrival requests of workloads to the fog-enable base stations during different intervals. At the same time, the process consumes B_{Energy 1}, B_{Energy 1,2} unit per joule decision of managing workloads with the IRS Unit. The study analyses the spike of workload requests to base stations as per B different density’s λ (BS/km²) Per Requested Workloads in the simulation as shown in Fig. 4. The density λ offers the arrival requests of workloads to the fog-enable base stations during different intervals. In contrast, the process consumes B_{Energy 1}, B_{Energy 1,2} unit per joule decision of managing workloads with the IRS Unit. It shows different density requests have required different energy level to process the recommendations in the cellular network. Figures 5 and 6 denotes the mobility and base-station to base-stations FL-IRS outperforms as compared existing state of art IRS method. Figures 5 and 6 FL-IRS unit can achieve optimal results in the existing reconfigurable cellular stations with the specific ranges with different density workloads level in the system.

Result discussion part

This part shows the performances of different schemes in the IRS that enable the unit to the designed system for the healthcare applications.

Energy consumption between sensors and base-stations

Initially, the study determines the energy-efficient performances between requested sensors workload and fog-enable base-stations as defined in Eq. (1). Figure 7 shows the federated learning aware base-station with IRS can make the optimal decision compared to the state of the art IRS method for the cellular system. The proposed system IRSTS determines the density request traffics at the runtime and learning them with efficient-energy efficiency and delay performances as shown in Figs. 7 and 8 during initial and base-station to base-station as defined in Eq. (2). Figures 9 and 10 elaborated the energy-performances of all cases as defined in Figs. 2A and 2B with the federated learning aware IRS unit with fog-enable base-stations. The existing approaches exploited IRS with the cellular base-stations without learning and with the fixed number of requests in the different intervals as did in Approach1 and Approach2. However, IRSTS schemes in the proposed distributed the teaching in other parts as FCA can decide with lower power consumption during initial energy, base-station to base-station energy and mobility energy. In this way, the efficient routing path for healthcare applications can deploy and optimize the energy objective E% of the study efficiently.

Figures 11 and 12 shown the performances of IRSTS as compared to Approach1 and Approach2 with delay objective for healthcare workloads in the system. It is cleared from results figures, the IRSTS outperformed all existing IRS enable cellular system with their approaches. The main reason is that, IRSTS divided the overall learning of the system in different layers as shown in Figs. 2A and 2B. By this way, the delay of centralized existing IRS enable fog-cloud can improved with the proposed schemes in the system.

Conclusion and Future Work

The federated learning aware Intelligent Reconfigurable Surface Task Scheduling schemes (FL-IRSTS) algorithm is proposed in this paper to achieve high-speed communication with energy and delay efficient offloading and scheduling. The trained model will decide the IRSTS configuration that best meets the goals in terms of communication rate. For each workload, multiple local models trained with the local healthcare fog-cloud network using federated learning (FL) to generate a global model. Then, each trained model shared its initial configuration with the global model for the next training round. Each application’s healthcare data is handled and processed locally during the training process. Simulation results show that the proposed algorithm’s achievable rate output can effectively approach centralized machine learning (ML) while meeting the study’s energy and delay objectives.

In the future, the study will consider the blockchain-enabled FL-IRSTS system for Industrial Internet of Things applications in the distributed fog-cloud network.