System Architecture for Intelligent Environment-Aware Wireless Communications

Abstract

The integration of communications and wireless sensing using the same frequency spectrum is likely to be one of the key features beyond 5G wireless systems, enabling digital transformation to its full potential. Therefore, future wireless communication networks will have to become intelligent, meaning environment-aware, adaptive and parsimonious in terms of radio resources and energy. In this chapter, we propose a system architecture that enables the introduction of intelligent environment-aware wireless communications. Special attention has been paid to the description of the wireless channel in the database with a set of radio rays, generated by the calibrated ray tracing using the ray bouncing method. The proposed approach enables straightforward calculation of the channel impulse response enriched with information about the angle of arrival and angle of departure for the band-limited multiple antenna wireless communication systems. The proposed architecture introduces the concept of local and global agents, used for retrieving associated parameters from the user requirements and the existing radio channel characteristics data.

1. Introduction

To meet the user expectations of future wireless communication networks, they have to become intelligent, meaning environment-aware, adaptive and parsimonious in terms of radio resources and energy. Radio environment awareness, which includes knowledge about the characteristics and location of radio devices, their operational characteristics and their mobility patterns, the geometry and electromagnetic characteristics of the radio environment and possible interference, is essential for introducing intelligence in wireless communications. The environment-awareness enables the prediction of radio channel properties by taking advantage of environmental information, measured channel state information (CSI) and information about the radio devices.

Today, wireless receiver estimates information about the radio environment with the help of known inserted training symbols calculating CSI. The CSI refers to channel properties and describes the combined effects of scattering, diffraction, path loss, fading and interference. The CSI is applied in the selection of the base station (BS) and radio resource management. The ratio between the training and information symbols in modern communication systems is increasing with the increase in communication signal bandwidth and the number of transmitter and receiver antennas [1], resulting in a decrease in the bandwidth efficiency of the complete wireless communication system. Today, the estimated CSI is not stored anywhere to be reused later on for the prediction of the characteristics of wireless links operating in the same area, using the same frequency band and exhibiting the same wireless channel impairments. The reuse of the stored CSI may result in a decrease in the transmitted training symbols, which can consequently result in increased bandwidth efficiency.

The idea of storing CSI in a database arose together with the initiative of cognitive radio, which was launched by Joseph Mitola III in a seminar at KTH Royal Institute of Technology in Stockholm in 1998 and published in Ref. [2]. The basic idea of cognitive radio is to efficiently use the spectrum by exploiting the spectrum holes, that is, part of unused spectrum, allowing the wireless transmission of users (secondary users) to whom the spectrum is not allocated. This basic idea led to an explosion of research and publications dealing with efficient spectrum sensing and environment awareness methods, spectrum access methods and radio resource allocation optimization, as well as machine learning methods used in predicting available spectrum holes and allocating spectrum holes to specific secondary users. One outcome of the research, which resulted also in the IEEE 802.22 standard ‘Wireless Regional Area Networks Enabling Broadband Wireless Access Using Cognitive Radio Technology and Spectrum Sharing in White Spacesʼ, was that the spectrum sensing neither centralised nor distributed is not sufficient for cognitive radio operation. The spectrum-sensing should be complemented by a database1 called the geolocation database, where the geolocated information of the wireless nodes including the cognitive BSs, custom premise equipment and primary transmitters have to be stored. The spectrum management entity uses all available information and calculates the list of the preferable channels to be used for the secondary user [3]. In cognitive radio, the CSI used is limited to the received signal strength indicator (RSSI), that is, on the level of the received signal.

An extension of the geolocation database called radio environment map (REM) was proposed by Zhao [4]. The REM is an integrated database that characterises the radio environment. It covers multi-domain environmental information such as geographical features, available services, spectral regulations, locations and activities of radios, relevant policies and past experiences. The radio environment data is presented in the form of maps, which are generated and updated by measurements collected from distributed radio devices. The REM can be also viewed as the extension of the available resource maps. The generation of radio maps from the sparse measurements in the field of radio mapping has emerged as a prominent area of research [5, 6, 7].

The idea of the REM is further developed in Ref. [8] by proposing the concept of channel knowledge map (CKM) as an enabler of environment-aware wireless communications. The CKM relies on the precise location of the transmitter and the receiver to retrieve CSI from the database. The CKM can have different formats. In its basic format is a channel gain map (CGM), which represents the channel attenuation at a particular location assuming channel bandwidth and carrier frequency. The main challenge in the construction of the CGM maps is to estimate the channel path loss at points without measurement from sparse measurement. A common approach is to use different two-dimensional interpolation techniques [5, 6, 7] or to partition the available measurements into different groups, where each group shares the same modelling parameter values that are to be determined by expectation maximisation approach [9].

The contemporary telecommunication systems are equipped with multiple antennas, at the transmitter and the receiver or at least with directional antennas; thus, the information about the radio ray angle of arrival (AoA) and angle of departure (AoD) information is beneficial. Furthermore, contemporary radio communication system operates on a wider frequency band, where multipath propagation results in selective fading, and consequently, the straightforward CGM does not provide sufficient CSI. The channel path map (CPM) and beam index map (BIM) are proposed in Ref. [10] to solve this problem. The CPM is a channel impulse response (CIR) enriched by the AoD and AoA of the radio rays, which can be easily converted to a multiple-input multiple-output broadband channel matrix. The BIM is a simplification of CPM storing information only the antenna beam indices for particular locations, which result in the least interfered radio channel. The CPM can be estimated using extensive channel measurements, which is very time-consuming, or by ray tracing simulation, which requires precise knowledge of the environment, including the electromagnetic characteristics of facets. In order to ray tracing simulation mimics the real environment, it has to be calibrated by measurements.

Up to now, there is no consensus about the system architecture and database structure that will enable intelligent environment-aware communication systems, so in this chapter, we propose the system architecture that enables experimentation with the environment-aware intelligent wireless communication in the testbed Log-A-Tec located at Jožef Stefan Institute premises [11].

The chapter is organised as follows. After this introduction, we discussed the CSI and which radio parameters can be included to describe the CSI efficiency. The next section is devoted to CIR, its formal description, efficient storage in a database and the calculation of the CIR for the band-limited communication system. The estimation of the CIR by computer simulations and/or measurements, including the information needed for computer simulations and channel model calibration, follows. After that, the frequency bands for wireless communication are discussed. The architecture of intelligent wireless networks is proposed in the next section. The description of the information exchange protocol between entities of the proposed system architecture as well as from the publicly available databases follows after that. Some publicly available databases containing information needed for environment-aware wireless communication systems are listed in the next section. The chapter concludes with final remarks.

2. Channel state information: CSI

The CSI is a metric of the wireless channel. The CSI varies due to random variations in the environment around the transmitter and the receiver, the time variation of the interference and the random movement and rotation of the transmitter and receiver. The slow-varying part of the CSI, for example due to changes in the environment, can be stored in the database, while the fast-varying part of the channel has to be estimated using training symbols. We have identified the following parameters to be stored in the database to represent CSI:

• Power received from a particular BS, which is in the 4G and 5G communication system called reference signal received power (RSRP) [12, 13]. In systems with limited channel estimation functionality, the parameter is rarely available.

• Received signal strength, which is composed of the received interference power, noise and the received power of the respective BS. In many systems, the parameter is referred to as the RSSI. Due to its simple estimation, the RSSI is accessible as a parameter in almost all wireless communication systems [12, 13].

• Carrier frequency is known or can be estimated from wireless communication system parameters.

• Channel bandwidth, the parameter is known from the wireless standard, or if configurable, it can be estimated from the beacons broadcast from the BS.

• CIR, which includes the imaginary and real parts of the CIR coefficients and sampling interval, that is, time between samples.

• Location of a device.

• timestamp of the measurement.

In early wireless communication systems, CIR is not estimated. In fact, for narrowband wireless communication systems, the attenuation of the wireless channel is sufficient for the operation of the system. In today’s broadband wireless communication systems, the CIR is estimated for the bands with an active connection. The exception is the physical layer of 5G communication systems called a new radio (NR) [12], which supports the estimation of CSI even for bands not occupied by an active connection. In order to obtain CIR for the wide frequency bands, the approaches based on the measurements are too expensive in terms of time and accuracy. Currently, a viable solution is based on calibrated computer simulation using ray tracing algorithms.

3. Channel impulse response: CIR

The CIR is the response of the communication channel on the Dirac impulse $\delta \left(t\right)$. Dirac impulse $\delta \left(t\right)$ has a unit power, infinite bandwidth and infinite short time duration. The CIR contains all the information to simulate and analyse a broadband radio channel. In particular, the CIR is convenient for characterising multipath radio channels. If the CIR is time-varying, the channel time variation can be also analysed and reproduced. The CIR of a multipath radio channel can be estimated by ray tracing algorithms. Ray tracing algorithms can be classified into three distinct general approaches [14]:

• ray launching approach, which casts a set of rays from the transmitting source in all directions and traces the rays through the scene,

• ray tube approach, which sends a set of ray tubes from the transmitting source and traces the ray tubes through the scene and

• approach based on the image theory, which by placing virtual receives behind reflective surfaces calculates the radio-ray path through the scene.

The modern wireless communication system consists of a set of BSs and a much higher number of user terminals called user equipment (UE). Generating CIR for all possible communication links, consisting of the BS-UE pair is too complex, due to a high number of UE locations, which would lead also to a huge amount of data being stored in a database. In order to make the problem manageable, the ray launching approach is used to generate radio rays from each BS location and track their paths through space. At each BS location, the set of rays is launched uniformly in all directions in space. When a ray hits an obstacle, the ray is terminated and one or more new rays are generated. The principle of ray generation is shown in Figure 1. The rays, originating from the BS location, are called root rays {1, 4, 5, 6}. The ray${}_3$ penetrates through an obstacle, and it is named a transient ray. The ray${}_2$ is a reflected ray. The programme libraries developed for the computer graphics and running on the graphics processing units (GPUs), for example optix [15], can be used to accelerate the ray generation and tracing. The rays, launched or generated due to ray interaction with the environment, are stored in a database. At least, the following ray information has to be stored in a database, but additional information can also be added:

• Ray id: identification of the ray,

• Ray origin: vector with the coordinate of the ray origin,

• Ray direction: normalised vector showing the direction of the ray,

• Ray type: describing the type of the ray; the following possible types are foreseen: root - ray originated from the transmitter, reflected - reflected rays from the obstacle, diffracted - diffracted ray at the edge of obstacles, transient - passing through the obstacles and scattered - ray scattered at the surface,

• Error margin: determines the interaction of the ray with the element of the environment,

• Root of the ray: identification of the ray root, that is. the radio transmitter,

• Ray depth: a number of interactions of a ray with the environment from ray root,

• Ray polarisation: ray polarisation stored as a vector,

• Cumulative distance: the cumulative distance the ray has passed from its root,

• Cumulative delay: the cumulative delay of the ray from its root,

• Cumulative phase shift: the cumulative phase shift of the ray from its root,

• Parent ray: the pointer to the parent ray used to identify the previous rays in the chain and

• Distance of the ray: this is the distance of the ray if it hits the scene, if the distance is infinite it means the ray does not hit any obstacle in the scene.

Figure 1.

The principle of ray generation.

In Figure 1, the parent ray of ray${}_2$ is ray${}_1$. The distance of rays {2, 3, 4, 5, and 6} is infinite, while the distance of ray${}_1$ is finite because it hits the scene. The child rays inherit parent ray parameters. The rays are stored in a database, to be later used for the generation of the CIR. The proposed approach is similar to the CPM model proposed in Ref. [10], but in our case, only rays are stored in the database, not the complete path from the transmitter to the receiver.

The receiver is represented by the reception sphere. The rays, which hit the reception sphere, contribute to the CIR. The size of the reception sphere depends on ray path length and the number of rays originating from the transmitter. Some advanced techniques to prevent ray double counting and inconsistent rays in propagation prediction are applied to determine rays, which hit the transmitter [14]. If the rays hit the receiver, we can backtrack and estimate the channel path information $z$, which includes the number of paths and their losses, phase shifts, and ray directions at the transmitter $p_l$ and at the receiver $r_l$ [10]:

$$z=\left\{L,{\left\{{\alpha }_l,{\tau }_l,p_l,r_l\right\}}_{l=1}^L\right\}$$E1

where $L$ is the number of channel paths, that hit the receiver, ${\alpha }_l$ is the complex gain of the $l^{\mathit{th}}$ channel path, ${\tau }_l$ is the excess delay of $l^{\mathit{th}}$ channel path, $p_l$ is the direction of the $l^{\mathit{th}}$ channel path at the transmitter and is the direction of $l^{\mathit{th}}$ channel path at the receiver. The CIR can be obtained from channel paths $z\left(\mathit{\alpha },\mathit{\tau },\mathit{p},\mathit{r}\right)$:

$$h\left(t\right)=\sum _{l=1}^L{\alpha }_l{e}^{j2\pi \left(t-{\tau }_l\right)f_c}{\mathit{a}}_t\left(p_l\right){\mathit{a}}_r\left(r_l\right)sinc\left(2\mathit{\pi B}\left(t-{\tau }_l\right)\right),$$E2

where $a_t\left(r\right)$ and $a_r\left(r\right)$ are antenna gains in the direction of vector $r$ at the transmitter and receiver respectively, $sinc\left(x\right)$ is sinc function $\frac{\mathit{\text{sinx}}}{x}$, while $B$ denotes the bandwidth of the system. By increasing system bandwidth $B$ a sinc function becomes more similar to the Dirac impulse, and consequently by increasing the system bandwidth the resolution of the paths is improved. While decreasing the bandwidth, it leads to a narrowband system, which has a low time resolution, and different channel paths are merged. The impact of the system bandwidth on the CIR is illustrated in Figure 2. The CIR assuming infinite bandwidth is generated by Saleh channel model [16]. The impact of bandwidth limitation to 500 MHz, 50 MHz, and 10 MHz results in a sparser CIR, where a significant part of environmental information is lost. The dotted line in Figure 2 represents the power delay profile (PDP) for different frequency bandwidths. Only the reflection from the main building structures, such as room walls and huge windows, can be deduced from the CIR if the frequency bandwidth is narrow.

Figure 2.

System bandwidth impact on the channel impulse response (CIR).

4. Estimation and calibration of the channel state information

The CSI can be obtained either by measurements or predicted by computer simulations using calibrated radio channel models. The channel measurements are expensive in time and in equipment and frequency limited, while the simulated CSI is prone to errors due to errors in environment geometrical description and unknown material electromagnetic properties and surface roughness. In the design of the system architecture, we have to consider both mentioned approaches, so we first examined the critical factors that influence the simulated CSI, and based on this preliminary study, we have identified, which information has to be included and stored in the database or accessed via the internet.

We have identified three categories of factors that have an important impact on CIR prediction:

• Environmental information that may influence or induce changes in the wireless channel. The environment is mostly static or changes slowly over time, but the movement and rotation of some objects can introduce rapid CSI variations.

• Characteristics of the equipment used for communication and measurement of the CSI. The characteristic of radio equipment does not vary with time, but with the fast development of the technology, timely maintenance of the database is mandatory.

• Calibration of channel models.

4.1 Environmental information

As radio waves propagate through the environment, they engage and interact with the surrounding, which has a huge impact on their characteristics. Various factors within the environment, including structures, obstacles and atmospheric conditions, contribute to the dynamic nature of CIR and CSI. Furthermore, high-frequency radio waves are highly dependent on meteorological data, such as rain, humidity.

In software tools for radio coverage prediction [17], the digital elevation model (DEM) is the basic input information. A DEM is usually represented as a raster, that is. a grid of squares with different heights. The raster DEM format is applied in the majority of radio frequency planning tools for outdoor radio planning. The global DEM is available with a resolution of 30 m. A higher resolution and quality DEMs are available from national mapping agencies. There exist several standard DEM formats, the most popular are the United States Geological Survey for storing a raster-based digital elevation model standard [18, 19], the Digital Terrain Elevation Data (DTED) and the Earth Gravitational Model 1996 (EGM96). In order to take into account additional attenuation caused by trees, buildings, etc., the usage of the terrain has to be considered for outdoor radio planning. In this respect, the additional attenuation in the form of a raster clutter file is also a necessary input for modern planning tools [17]. The increase of computational of graphical processing units and speed up of ray tracing algorithms enables usage of deterministic channel models also for the rural areas [20]. Due to the inappropriateness of the raster DEM format for ray tracing, in this case, the DEM is represented as a vector-based triangular irregular network.

An urban environment is a particularly challenging area for radio coverage prediction. Initially, the urban propagation models were based using the raster representation of the building similar to DEM, but today ray tracing is a popular approach to model radio wave propagation in an urban environment, and urban areas are represented by the vector-based triangular irregular network. A popular format for storing DEM as a triangular irregular network is the ESRI TIN format. It describes elevation information, including breaking edge features. Each point and triangle can carry tag information [21]. To use a ray tracing approach in an urban environment described with a vector-based triangular description the electromagnetic properties in terms of surface conductivity, permeability, permittivity and surface roughness should be known. While the geometrical properties of urban areas are well represented in the available local GIS databases, the electromagnetic properties of surfaces are rarely available, and that limits the use of ray tracing in urban environments.

In the last decade, a significant quantity of wireless communications occurs indoors. The indoor propagation environment significantly differs from the urban and rural ones. The straightforward multi-wall propagation model [22] is nearly completely replaced by the ray tracing approach. The indoor geometry is easily estimated from blueprints, electronic plans of buildings and infrared scanning. The lack of information about the electromagnetic characteristics of the used material and additional objects in rooms may introduce errors in radio wave propagation prediction. There exist several commercial [23, 24, 25] and open-source [26] tools for the prediction of indoor radio coverage. These tools usually include a subsystem for drawing or importing the indoor environment.

The trends in wireless communications heading toward using frequencies above 10 GHz, which are sensitive to atmospheric distortions. Some of the well-recognised factors, which affect radio wave propagation at high frequencies are:

• Precipitation: It greatly influences the channel attenuation. We quantify different types of weather {sunny, partly cloudy, cloudy, foggy, light rain, medium rain, heavy rain, light snow, medium snow and heavy snow} to be able to take the different weather types into the radio channel prediction.

• Temperature: The temperature of the ambient air relates to the density of the atmosphere, which affects the wireless channel in terms of scattering and fading.

• Humidity: It has a similar impact on the effects of scattering and fading on the wireless signal.

• Season of the year and part of the day. They also influence the CSI since the atmosphere density is different for different periods of the day. Therefore, the timestamp of the measured/estimated CSI has to be stored in the ISO 8601 standard format, containing year, month day and hours with minutes and seconds.

The information (e.g. temperature and humidity) can be obtained from the device itself, or from the publicly available weather websites, which include all of the described factors. Since they are related to the time when the CSI is to be estimated, we can categorise them as temporal information. Regarding the location of the device and corresponding spatial information, there are a lot of additional factors that influence the CSI.

4.2 Equipment characteristics

The nodes providing services in modern wireless communication networks have to obey standards to enable the interworking of nodes from different vendors. The standards specify some basic characteristics of wireless nodes, but detailed node characteristic is under the control of the network provider or the owner of the access point. In order to estimate the CSI, the transmitter and receiver characteristics have to be known, including the radio node position, antenna type, elevation, orientation, transmit power, bandwidth and carrier frequency of the link, etc. The majority of the mentioned features do not vary with time, but some of them are selected for each link establishment, or they even change during communication. Three wireless mobile terrestrial communication networks are foreseen for future public communications, namely: cellular wireless networks, wireless local area networks (WLANs) and wireless sensor networks (WSNs).

In cellular wireless networks, also called mobile wireless networks, the geographical area is divided into cells. Each cell is covered by the BS. The mobile wireless networks are completely under the control of the mobile operator, which also owns all information about the BS locations, parameters, technology in use, allocated frequencies and transmit powers. Historically, cellular wireless technology has evolved from the analogue first generation to the second generation, which applies narrowband digital technology to the latest fifth broadband wireless technology. The mobile operators developed operator-specific databases to store the static parameters of their networks. The database consists of separated tables of Ref. [17]:

• antennas, with the links to antenna diagrams and other parameters,

• locations of the BS and other GIS data related to BS location and

• BS parameters, including antenna height, mechanical and electrical antenna tilt, antenna azimuth, antenna type, technology, transmit power, etc.

These parameters do not vary with time or they change from time to time and can be accessed via the application programme interface (API). Similarly, a database with the characteristics of UEs; that is., mobile phones has to be established. The ability of CSI estimation varies from generation to generation, from RSSI to RSRP and RSRQ to CIR in orthogonal frequency division multiplex (OFDM)-based systems [1]. This information is at BSs but, in general, not stored in any database.

The WLANs are used to cover local areas, and thus they are installed ad hoc, without any central register or database with data about the installed access points. However, at university campuses, factories, schools, hotels or other organisations occupying large areas, the installation of access points is under the responsibility of the special department or is outsourced. For efficient maintenance, the setup has to be documented and optionally put in the database. In general, the API to the database is usually not implemented. CSI measured in WLANs was typically not reported to the access points and could not be directly retrieved by the user, so this information was not available for further use. In the last decade, however, commercially available chipsets [27] have been upgraded with an API to extract the estimated CSI of a wireless link due to advantages that CSI offered over RSSI for various wireless sensing applications [28] and to enable advanced signal processing techniques for optimised WLAN performance.

At the moment, large area coverage WSNs are set up and maintained in industrial objects. The network is under the control company’s specialised department, but access to the sensors, information about their locations and other data is available within the company. The standardised channel sensing is limited [1], but some estimation of the CSI in the complete ISM frequency band is possible by taking advantage of channel hopping specified in some ISM band operating wireless communication systems [29].

4.3 Calibration of channel models

The radio propagation channel models have to be calibrated before usage. The network planning of the second-generation mobile communication system was mainly based on the Okumura-Hata channel model [30] or its deviates, such as Ericsson 9999 channel model [31]. In the Ericsson model 9999, a path loss is specified as:

$$\begin{array}{ll}L& =a_0+a_{1}log\left(d\right)+a_{2}log\left(h_{\mathit{BS}}\right)+a_{3}log\left(h_{\mathit{BS}}\right)log\left(d\right)+\\ & -3.2{\left(log\left(11.75h_{\mathit{UE}}\right)\right)}^2+44.49log\left(f\right)-4.78{\left(log\left(f\right)\right)}^2\end{array}$$E3

Where $d$ is the distance between the BS and UE in metres, $f$ is the frequency in MHz and $h_{\mathit{BS}}$ and $h_{\mathit{UE}}$ are effective heights of the BS and UE antennas in metres, respectively. Parameters $a_0$, $a_1$, $a_2$, and $a_3$ are constants, which are used in the calibration process. The default values are $a_0=36.2$, $a_1=30.2$, $a_2=-12.0$ and $a_3=0.1$. The constants can be calibrated for each BS separately or for the entire wireless network [32]. The Ericsson 9999 model applies a digital elevation model and clutter file to introduce shadowing, diffraction and impact of terrain usage into prediction.

4.3.1 Calibration of the ray tracer

A ray tracing radio propagation models are more general than empirical channel models, but their prediction accuracy highly depends upon measurements for the purpose of tuning, that is. calibrating their parameters, such as the material electromagnetic material characteristics and roughness of the surfaces [33]. The electromagnetic material parameters can be obtained by

• checking the scientific literature and using the estimated parameters in the simulations [34]. This straightforward approach lacks data for all materials and their properties at the target frequencies. Furthermore, a facet is rarely built from a single material and thus cannot be classified into a set of known materials. The facets from the same material differ in the roughness, which has a significant impact on the diffuse scattering in particular at millimetre frequencies and above.

• estimating the material parameters from measurements of the reflection from the single facet and using the particular wireless communication system operating in specified frequencies [35]. The method is expensive and time-consuming due to the huge number of existing parameters in the scenario and the variety of wireless communication systems and their operational frequencies.

• finding the environment parameters by minimising the difference between measured and simulated CIR [36], Different optimisation methods can be applied, starting from the brute force approach to simulated annealing. The problem is in general non-convex, thus optimisation may end in local minima.

5. Frequency bands allocated for wireless communication systems

The channel properties strongly depend on the wavelength of the radio waves, and the bandwidth of the radio signal. In this respect, we look at the frequency bands allocated for modern wireless communication systems. The frequency bands allocated for public wireless technologies are listed in Table 1. The upper five rows in Table 1 include the most important frequency bands for WiFi wireless technology, while the rest of Table 1 provides the important frequency bands for NR wireless technologies, which comprises also the re-farmed frequency bands from the previous generation of mobile wireless networks. The allocated frequency bands depend on the region in terms of allocated bandwidth and allowed transmitted power.

${\mathbf{f}}_{\mathbf{c}}$ [MHz]	BW [MHz]	Wireless technology	Chanel BW [MHz]
2400	100	802.11 b/g/n/axe	20–40
5000	725	802.11 a/n/ac/axe	20–320
6000	480	802.11 axe	20–320
60,000	1296	802.11 ad/ay	1080 - 8640
600	35	FDD	5–35
700	10	FDD	5–30
800	30	FDD	5–20
850	25	FDD	5–50
900	35	FDD	5–35
1500	85	TDD	5–80
1700	90	FDD	5–35
1800	75	FDD,	5–50
1900	60	FDD, TDD	5–45
2000	25	FDD	5–25
2100	90	FDD, TDD	5–50
2300	100	FDD, TDD	5–100
2500	194	TDD	5–100
2600	70	FDD, TDD	5–50
3500	500	TDD	5–100
3700	900	TDD	10–100
4700	70	TDD	10–100
26,000	3250	TDD	50–400
28,000	3000	TDD	50–400
39,000	3000	TDD	50–400
41,000	4000	TDD	50–400
47,000	1000	TDD	50–400
60,000	1400	TDD	100–2000

Table 1.

Frequency bands of public wireless technologies.

At the early stage of wireless communications technologies, the sub-gigahertz frequency bands were applied due to the provision of good outdoor coverage. The request for high throughput of the wireless technologies in the wireless communication systems resulted in the usage of microwave frequency bands, starting at L-band (1–2 GHz) and S-band (2–4 GHz) for outdoors and C-band for indoors. Recently, the millimetre frequency bands, namely Q, U and V bands, are foreseen for future wireless networks due to the availability of the wide bandwidth.

The first WiFi wireless technology, which was widely used, IEEE 802.11b operates in 20 MHz bandwidth offering a data rate of 11 Mbit/s. However, the provided data rate was not sufficient for the emerging wireless services; thus, the next generation of IEEE standards IEEE 802.11 g/a was based on OFDM technology, offering data rates up to 54 Mbits/s. The 40 MHz bandwidth is introduced by IEEE standard 802.11n, which bonds two 20 MHz channels, and in addition, exploits null OFDM carriers between bands for data transmission. The same principle is applied in IEEE 802.11 ac to achieve the 80 MHz bandwidth, while for achieving 160 MHz bandwidth two dis-joined or neighbouring 80 MHz bands. The WiFI 7 based on the IEEE 802.11be standard operating in the 2.4, 5 and 6 GHz frequency bands supports frequency bandwidths up to 320 MHz. In the V-band, the gigahertz WiFi is specified at IEEE 802.11ad and IEEE 802.11ay standards using the bandwidth of several gigahertz, namely 1.080, 2.160, 4.320, 6.480 and 8.640 GHz.

The second part of Table 1, bottom rows, listed the frequency bands allocated for the NR. The NR specifies the two frequency ranges, namely frequency range FR1, which includes frequencies below 6 GHz and frequency range FR2 above 6 GHz. The FR1 is also called cm-wave frequency band, while FR2 mm-wave frequency band. The bandwidth in FR1 can be 5, 10, 15,20, 25, 30, 35, 40, 50, 60, 70, 80, 90 and 100 MHz, while in FR2 50, 100, 200, 400, 800, 1600 and 2000 MHz.

6. Architecture of the intelligent network

The environment-aware communications shall be supported by adequate network architecture, which will enable CSI storage, calculation, estimation and prediction for the selected BS and UE. A BS shall include an intelligent local agent with an instance of a local database or access to the remote database. The main purpose of the agent is to store measured CSI values and map them with corresponding locations, calculate the CSI, calibrate the channel model, fuse the measured CSI and relay the data to the global agent. CSI is location-specific; thus, the storage of CSI at the network edge at each BS or for several BSs is a natural choice compared to keeping all CSI data in a centralised cloud. The calculation of the ray paths requires a powerful computer, which is not available at the network edge, and thus we propose to be partially centralised. The database elements, computational resources and their distribution are plotted in Figure 3. We should distinguish between indoor wireless networks, which provide services within shopping malls, big industrial complexes, skyscrapers, sport halls, etc., and are isolated from outside wireless networks due to metallic surfaces used for outdoor walls and outdoor networks, which provide outdoor coverage as well indoor coverage by penetrating radio waves inside buildings. Former, more or less isolated islands can be managed by a single local agent per building or several buildings, while in the case of outdoor wireless networks, we propose a local and a global database and distributed mesh local agents, which can communicate with each other.

Figure 3.

Database elements, computational resource and their distribution.

6.1 Local agent and local database

When a UE requests a new service, the local agent can use its local database to find CSI if it has been previously estimated at the same location in the near past. The local database can be collocated with the particular BS and corresponding agent, or it can serve many BSs at the network edge, for example where two BSs cover the same area. In addition, the local database can be also a part of a radio cloud. Each local agent and its database are connected with one or more global agents. The local agent also requires internet access to obtain weather information since the temperature and humidity can influence the CSI measurement and should be stored alongside the measured CSIs.

6.2 Global agent and global database

The global intelligent agents and their corresponding global databases shall be distributed over wide regions such as separate city districts, town areas, or countryside. Due to their scalable nature, they can be added and removed in case of additional needs/requirements. Their main purpose is the calculation and estimation of the CSI values using the information obtained from the location and service requirements of the device, therefore a fast connection with the local agents is required. Since they are querying the existing publicly available sources of information, they also require access to the internet.

6.3 Call set up in environment-aware wireless networks

In wireless networks, the BS broadcasts its access information at the predefined frequency band and the radio technology. The UE joining the network sends a service request (step 1 depicted in Figure 4), which at least includes:

• terminals unique ID,

• desired service requirements and

• precise or approximate location.

Figure 4.

Call setup in environment-aware wireless networks.

The UE location can be estimated from the received broadcasting signal of several BSs. To get even more precise location information, the UE can use other sensors available (WiFi, RFID, ACC/GYRO sensor, …) or from global positioning systems (GPS, Galileo, Glonass or BeiDou).

Upon receiving the link/access request, the BS has to select the most appropriate technology (frequency band, modulation, coding schemes, etc.) to meet the requirements of the UE. The technology can be selected based on the characteristics of the wireless channel between the BS and the UE, which is at this point still not known with the required accuracy. Here, the BS can use its local agent to obtain the CSI based on the approaches described in Section 3.

6.3.1 The role of local agent

The local agent can query its local database and check if there were any comparable requirements created in the near past. For example, if the device made a similar request from (approximately) the same location, the database repository can quickly return the CSI value to the BS. The database must also consider all other parameters that can influence the CSI value, such as weather information with humidity and temperature, timestamp or part of the day (day/night), as discussed in Subsection 4.1. A detailed description of the query procedure and corresponding API calls is given in Section 7.

By using the retrieved CSI value, the local agent can then select the most appropriate technology and frequency band for terminal service requirements. With this information, the BS can open the communication link and provide access to the terminal. Since people are usually following a similar everyday routine, we expect that this first step will be very beneficial. Over time, the local database will include many samples and most of the area will be covered. At the same time, we have to be aware of the ageing of CSI due to changes in the radio environment and people habits.

6.3.2 The role of global agent

In case no related CSI exists for service requests, the local agent can request the global agent to estimate a new CSI parameter for the new radio terminal location. Since the estimation of the CSI can be sped up, if the initial CSI is close to the real one, the BS can provide the closest one from its local database (e.g. the same device at the same location but with different weather conditions, at different seasons). The BS must also hand over the information about the UE (device ID, antenna characteristics, location and speed of device movement), while the detailed information of the BS is already stored in the global databases repository. Furthermore, the local agent can predict the trajectory and velocity of the UE based on its previous locations and historical data. This could further increase the speed of CSI retrieving.

The global agent then calculates the CSI estimate/prediction by using sophisticated methods as discussed in Section 3 and returns the value to the local agent. Applying the received value, the BS can then provide access to the UE and allocate a frequency band.

6.3.3 User equipment feedback

After receiving a connection to the network, the UE performs the CSI estimation, which is sent back to the BS (step 5 depicted in Figure 4). The estimated CSI is stored in the local database with the corresponding timestamp, location and weather information and is also sent to the global database as feedback, where it is applied for the calibration of the deterministic channel models.

7. Information exchange protocols

Both instances of global and local databases are created using modern web development tools as an application in a microservice architecture [37]. The data is stored in PostgreSQL server, which can be managed through NodeJS server. PostgreSQL was chosen because of its support JavaScript object notation (JSON) and full-text search over measurement descriptions, which fastens the querying process. An angular web client with a graphical user interface is added to the database to support the maintenance and configuration of the database and for viewing the measurements. To enable in-depth research, the database is extended with additional services for plotting the measured data graphs using Python scripts. All of the interactions with the database are done via the Nginix reverse proxy server. For each received request, the Nginx server decides to which running service should it pass it on. This mechanism is transparent to the agents and allows for much easier interaction. Proxying also enables us great flexibility in choosing different languages and technologies for different tasks, providing an option to add new features while exposing a single unified entry point for the agents.

All the interactions with the database are done using easy-to-use REST protocol, which enables the agents to access a resource within the database or other supported services. Besides HTTP, the data can be stored using the MQTT protocol.

All records in the database are hierarchically organised from high-level ideas to low-level measurement details. Separation is done in the following fields: database stage, base station, location and CSI estimate.

The database stage field is used to separate the measurements from different stages of the database’s operation (development, production and archive measurements). In the scope of any database stage, different measurements are performed. The base station field corresponds to the different BSs, which can use the same local database. The locations are the lowest organisational level and include different CSI estimates, which contain the actual measured data. The location level can represent an actual UE position using longitude and latitude, or it can represent a smaller region around a precise location. To represent the CSI estimate in a database, the JSON format is used with the following fields: deviceID, timestamp, CIR, RSSI, RSRP, frequency, bandwidth, temperature, humidity and weather.

8. Existing publicly available information sources

There are multiple publicly available information sources, where accurate environmental information can be obtained. Most of the sites enable an API to download the data in various file formats (HTML/XML/RSS/CSV).

An example of a dynamic dataset is weather information. The accurate weather information for Slovenia can be obtained from the official ARSO website. There is data available for 20 areas, and it is updated repeatedly each hour, while the data can be also accessed for the separate weather station, which is updated at intervals of 10 minutes. In addition, the site offers a two-day archive. By using an HTTP request, an agent can obtain desired information from the website in different file formats. The global weather data can be obtained from the open weather map [38] or global climate and weather data [39]. The ozone and UV radiation data can be found in [40].

The digital elevation models are gathered from different sources, namely:

• US Geo Science Shuttle Radar Topography Mission (SRTM) data with a resolution of 1 arc-second for the United States and 3 arc-seconds for global coverage at [18],

• US Geo Science GTOPO30 global 30 Arc-Second resolution in [19],

• SRTM Tile Grabber [41],

• SRTM 90 m DEM Digital Elevation Database [42],

• The Copernicus Global Land Service (CGLS) [42],

• NASA’s Land Processes Distributed Active Archive Center (LP DAAC) data [43],

• Deutsches Zentrum für Luft- und Raumfahrt (DLR) Earth Observation Center [44] and

• European Digital Elevation Model (EU-DEM) [45].

9. Conclusion

Intelligence in wireless networks can be introduced based on reliable past and current information about the radio environment. In this context, we have proposed, in this chapter, a system architecture that enables the introduction of intelligent environmentally aware wireless communications that will pave the way for digital transformation to reach its full potential. Special attention has been given to the description of the wireless channel in the database with a set of radio rays, generated by the calibrated ray tracing based on the ray bouncing method. The proposed approach enables straightforward calculation of CIR enriched by the information of the AoA and AoD information for the band-limited multiple antenna wireless communication systems. The proposed architecture introduces a concept of a local and global agent, whose roles are not described in detail, and remains an open research problem, which will be solved in our next contributions.

The concept of intelligent environment-aware communications is a new concept that will hopefully be standardised and implemented in the 6G wireless networks. People are becoming aware of radio pollution, and the proposed concept increases the bandwidth efficiency of wireless communication links and wireless systems as a whole.

Acknowledgments

This research was funded by the Slovenian Research Agency under grant no. P2-0016 and grant no. J2-2507.