Get Complete Project Material File(s) Now! »
LED-based luminaires, which are considered as APs in VLC, are based on using diodes that convert electrical signals to light signals. They have been receiving an increasing interest, for their advan-tages in terms of energy and luminance efficiency compared to other types of luminaires (e.g., the incandescent and fluorescent -based lamps) . Because of their relatively fast response, they have been considered for wireless communication applications, in addition to their illumination purposes. For VLC systems, the main types of LEDs used include RGB LEDs, and phosphor-based LEDs.
In RGB LEDs, red, green, and blue LED chips are used to generate the corresponding light-ing, where by controlling and mixing the LED chips’ outputs the overall illumination experi-ence changes. By modulating each colour separately, RGB LEDs can achieve high capacity using wavelength-division multiplexing (WDM) [35–37]. On the other hand, given the balance needed between the red, green, and blue components for white lighting, the overall lighting experience is prone to changes due to components’ colour deviation over time .
For the phosphor-based LEDs, which are commonly referred to as “white LEDs”, an ensemble of blue LED chips and yellow phosphor coating is utilized for an overall white lighting experience. Compared to RGB LEDs, white LEDs offer a simpler structure and a lower cost, which supports their widespread use. However, they suffer from slow response during signal modulation, because of the phosphor part, which results in limiting the modulation bandwidth to few MHz . To boost the VLC link capacity, different solutions have been proposed, such as blue filtering at the Rxs , equalization , and using spectrally efficient modulation (e.g., multi-carrier modula-tion) schemes, such as optical orthogonal frequency-division multiplexing (OFDM) , which will be explained in more detail later in Subsection 2.2.4.
Note that, interest in other types of LEDs such as micro LEDs  and organic LEDs  has been increasing recently. However, for the rest of the thesis, we will consider the widely used white LEDs.
For signal detection at the Rxs, the main types of PDs used include the positive-intrinsic-negative (PIN) photodiodes, the avalanche photodiodes (APDs), and the silicon photo-multipliers (SiPMs), which are in fact arrays of APD pixels operating in Geiger mode [9, 43–45]. For the considered indoor and outdoor VLC systems, due to the relatively strong ambient light at the Rxs, and for Rx simplicity in the user device, we consider PIN PDs in the rest of the thesis. Note, these PDs are non-coherent detectors, thus the converted signal is independent of the phase of the light wave.
To increase the received optical intensity, optical concentrators can be considered at the Rx, to increase the effective area of the PD, which will be further explained in Subsection 2.2.3.
Due to the physical properties of the VLC systems, which allow the transmission of positive and real signals, VLC signalling is based on IM and direct detection (IM/DD).The equivalent VLC link model is given by:
where x(t ) and y(t ) denote the transmitted and the received photocurrents, respectively, ⊗ is the convolution operation, h(t ) represents the channel gain, and z(t ) refers to the Rx noise.
The noise at PIN-based Rxs comprises mainly of shot and thermal noise. Shot noise is caused by the fluctuations of the photo-current at the Rx, which is generated by the incidence of the trans-mitted signals as well as any other ambient radiations, and its power is directly proportional with the strength of the incident radiation. In case of background radiations, shot noise can be domi-nant, which can be modelled as signal-independent additive white Gaussian noise (AWGN) with zero mean [9, 46]. As concerns the thermal noise, it results mainly from the Rx electronic circuits and dominates in case of negligible background radiation. Similar to shot noise, thermal noise is modelled as signal-independent AWGN with zero mean. The overall noise at the Rx is mod-elled by the summation of the variance of the AWGN shot and thermal noises, which can be given by B × N0, where B represents the system bandwidth and N0 denotes the one-sided noise power spectral density.
VLC signals propagate in the communication environment following the physical properties of light. As a result, the received signal may result from LOS or non-LOS (NLOS) propagation compo-nents, where the latter results from the reflection of light signals by the reflecting surfaces within the environment. It has been shown in several works that, NLOS components have small contri-bution to the received signal compared with LOS component in VLC networks [47–50]. For this reason, we consider only LOS propagation in this thesis, assuming that the LOS component is not blocked or shadowed. For modelling the link between APi , and user j (U j ) utilizing a PIN PD and an optical concentrator, the channel gain is denoted by hi j . Figure 2.2 shows the channel gain pa-rameters for a typical LOS VLC link. Assuming a Lambertian pattern for the LED, the channel gain can be written as (see [10,51–54]):
where φi j refers to the angle of emission with respect to APi , θi j refers to the incident angle with respect to U j ( j th Rx, Rxj ), and li j denotes the link distance between APi and U j . Also, R is the PD responsivity, assumed to be the same for all Rxs, S is the LED conversion efficiency, and m is the Lambertian order, assumed to be the same for all LED luminaires. In addition, A j is the j th Rx collection area, which is given by :
with APDj denoting the Rx active photo-detection area, q j the optical concentrator refractive index, and FOVj the Rx FOV corresponding to Rxj . The Lambertian order m is given by m =− ln(2)/ ln(cosψi ) with ψi being the half-power angle of the LED-luminaire at APi . Note that hi j = 0 for θi j > FOVj , where FOVj is the FOV of Rxj . If no concentrator is used at Rxj , we have A j = APDj .
Signal modulation techniques in VLC systems can be classified into serial transmission (usually called “single-carrier”) and multiple-subcarrier (usually called “multi-carrier”) schemes. Serial transmission techniques include on-off keying (OOK), pulse amplitude modulation (PAM), pulse position modulation (PPM), and pulse width modulation (PWM), which encode messages using two intensity levels, multiple intensity levels, position of the pulse in time-domain, and width of the transmitted pulse, respectively. Because of susceptibility of serial transmission techniques to inter-symbol interference due mainly to the limited modulation bandwidth of the white LEDs, multiple-subcarrier modulation techniques, such as multiband carrier-less amplitude and phase modulation (m-CAP) [55–59] and OFDM [60,61], are usually considered to provide high data rates. Generally, use of OFDM offers (i) efficiency in spectrum utilization; (ii) adaptability to the fre-quency selectivity of the channel through bit/power loading; and (iii) simplicity of channel equal-ization . Note, since OFDM signals are complex and bipolar, they cannot be used directly in VLC systems, which rely on IM/DD. To comply with the VLC signal requirements, various tech-niques have proposed to generate real unipolar OFDM signals, such as direct-current biased opti-cal OFDM (DCO-OFDM), and asymmetrically clipped optical OFDM (ACO-OFDM) . In DCO-OFDM, a DC bias is added to the signal before clipping to obtain a unipolar signal [62, 63]. For ACO-OFDM, only odd-numbered subcarriers are modulated before zero-level clipping, which does not result in any loss of information . Given the advantage of DCO-OFDM in terms of spectral efficiency , it is assumed by default as the signalling scheme in the rest of the thesis.
Figure 2.3 shows the block diagram for DCO-OFDM based transmission. At the Tx, firstly, informa-tion bits are mapped (typically using quadrature amplitude modulation (QAM) mapping), before serial to parallel (S/P) conversion. Then, such parallel stream passes by Hermitian symmetry to ensure having real-value signals, followed by an inverse fast Fourier transform (IFFT) block, to generate the time-domain signal. Note, every N2c −1 symbols are grouped to form an OFDM sym-bol X = [X0, X1, …, XNc ], where Nc is the number of OFDM subcarriers, Xk is the symbol trans-mitted on the kth subcarrier, and the Hermitian symmetry is imposed on the OFDM symbol such that:
Here, (.)∗ denotes complex conjugation. After passing the output of the IFFT block to parallel to serial (P/S) conversion, adding a cyclic prefix (CP), and performing DAC, a DC-bias is added to the signal in order to get a positive-value signal. Afterwards, double-sided clipping is applied on the signal to fit it to the dynamic range of the LED. The resulting signal is then used to modulate the LED intensity. After propagation through the channel, the received optical signals are converted to electrical signals by the PD, then filtered and amplified by the TIA, before undergoing ADC and removing the CP of each block. After S/P conversion, fast Fourier transform (FFT) is applied on the received blocks, then one-tap frequency domain equalization is applied to compensate the chan-nel effect. Following this, P/S conversion and demapping are performed to recover the transmitted bits.
Obviously, bidirectional transmission is needed for the considered wireless networks. Whereas VLC is considered as a promising solution for downlink (i.e., from the LED luminaire to a device), implementing the uplink could use IR or RF transmission. Note that, using VLC for uplink could result in unpleasant experience for users. A relatively recent approach considers using short light pulses for uplink, to ensure minimum impact on the user lighting experience, which is referred to as “DarkVLC” . However, this results on limitations on the link performance in terms of the user achievable throughput. On the other hand, the use of RF signals (e.g., WiFi ) is not com-patible with EMI-sensitive scenarios. As a result, the use of IR band for uplink has been considered as a reliable solution to make the balance between link performance, user lighting experience, and compatibility with EMI-sensitive scenarios . To address MA requirements, in the uplink, carrier-sense MA with collision avoidance (CSMA/CA) protocol can be used in the MAC layer, as suggested in , given its advantage in terms of resource utilization efficiency by allowing users to access the channel according to their traffic, after checking the availability of the channel re-sources.
Multi-cell VLC system
For VLC systems, as the area of the illumination area increases, the number of LED luminaires required to satisfy the lighting and the communications needs increases. As mentioned earlier, such LEDs serve as APs, each one potentially handling several users in its coverage area (i.e., cell). This is the case, for example, for a relatively large office or a conference hall. Figure 2.4 illustrates the principle of such a multi-cell network. Therein, users are handled differently depending on whether they are located at “cell centers” or at “cell edges”. A central control unit connects the APs and has the task of information exchange and coordination between them. It also defines the cell boundaries, and consequently, specifies cell-center users (CCUs) and cell-edge users (CEUs). The APs, in turn, are responsible for identifying the channel information for all users, and forwarding it to the central control unit. In such a cellular scenario, there are two primary types of interference: the inter-user interference (IUI) resulting from the interference between the users within a cell, and the ICI resulting from the interference arising from the users in neighbouring cells. In partic-ular, handling CEUs that are subject to intense ICI is a delicate task. In the next sections, different techniques for handling IUI and ICI will be discussed.
Note, the ensemble of Txs (APs) and Rxs (users) can be regarded as either a MU multiple-input single-output (MU-MISO) or a MU multiple-input multiple-output (MU-MIMO) system, depend-ing on using a single PD or multiple PDs per Rx, respectively. Throughout this thesis we consider a single PD at each Rx for the reason of reduced implementation complexity. Consequently, the MU-MISO architecture is considered in the sequel.
Interference mitigation in multi-cell VLC networks
To handle multiple users in multi-cell VLC networks while mitigating IUI and ICI, two main ap-proaches could be considered. The first is exploiting the spatial degrees of freedom at the Txs and the Rxs (e.g., the Tx and the Rx locations), by using linear pre-coding techniques, to fully use the time and the frequency resources in the network. However, this comes at the expense of con-straints on the number of users that can be handled, and increased complexity of the network. Higher flexibility on the number of users can be achieved by distributing the network resources (e.g., time, frequency) among the users using MA techniques, which come at the expense of a lower spectral efficiency.
The idea of linear pre-coding is based on pre-processing the user data, based on the channel gain, to improve the user performance according to a pre-defined criterion. Indeed, this requires the availability of the user channel state information (CSI) at the Tx.
For the context of VLC, different pre-coding schemes have been proposed, such as those based on zero forcing (ZF) pre-coding [69–71] and mean-squared error (MSE) [50, 72, 73] criteria. For MU-MISO VLC systems, linear ZF pre-coding has been widely used, due to its simplicity, and its good performance at high signal-to-noise ratio (SNR) levels. This is mostly the case for indoor VLC networks because of the limited link distance in the case of existing LOS [74–77]. Due to the merits of ZF pre-coding, it is further considered in the following chapters.
For the given MU-MISO network structure, linear ZF pre-coding is performed at the APs on the users’ signals, which results in the suppression of interference from the other users at each Rx. To do this, the CSI of all underlying channels must be estimated at each Rx (for instance, based on the transmission of some pilot symbols from the APs in the downlink), and sent back to the APs (in the uplink). This CSI is then used by the APs to calculate the appropriate ZF pre-coding matrix . Figure 2.5 shows a simplification for the idea of ZF pre-coding.
Let Nt denotes the number of APs, located at the VLC network, that handle a total number of Nr Rxs within their coverage area, such that Nr ≤ Nt . At APi , the Rxj symbol d j is multiplied by a pre-coding weight wi j .
The diagonal entries γj > 0 (that can be considered as the coefficients of parallel sub-channels) de-termine the SNR at user Rxj , and are determined based on a performance metric (e.g., throughput maximization, as considered in ). For simplicity, we consider the max-min fairness criterion proposed in , where the goal is to maximize the minimum achievable throughput. Lets define the vector µ as:
with abs(.) denoting element-wise absolute value operation. Also, 1 in (2.12) is a vector with all entries equal to one and Pe is the transmit electrical power, assumed to be the same for all APs.
The SNR at Rxj can be written as:
where the data rate is measured in units of bits per second (bps), and due to the Hermitian sym-metry constraint in DCO-OFDM, we have a loss of factor two in the spectral efficiency .
Although ZF pre-coding offers an efficient solution for handling IUI and ICI, it limits the num-ber of users. Indeed the number of users cannot exceed the number of APs. To allow handling a larger number of users, other MA scheme should be used, as discussed in the next subsection.
To serve a number of users larger than the number of APs, another approach for IUI and ICI mitiga-tion in VLC networks is to use MA techniques. We present here different techniques, namely, time-division MA (TDMA), space-division MA (SDMA), optical code-division MA (O-CDMA), OFDMA, and power-domain NOMA (that we will simply refer to as “NOMA”), while discussing their suit-ability for a multi-cell VLC network with specific features including high data rate, large commu-nication area, and user mobility. Figure 2.6 shows a comparison between different MA techniques with respect to the time, code, frequency, and power domains.
By TDMA, the channel temporal resources are shared among users by allocating to each one a certain time slots [9, 78]. Although in indoor applications, we are concerned with a relatively low mobility of users, the accurate synchronization required at the user terminals and at the AP could become a challenge due to very high data-rates, especially for increased number of users. Also, in a multi-cell network, TDMA suffers from increased ICI at overlapping areas between the cells .
Table of contents :
1.1 Overview of visible-light communication
1.1.1 Wireless communications
1.1.2 Optical communications
1.2 VisIoN project
1.3 Thesis objectives
1.4 Thesis overview and contributions
1.4.1 Thesis outline
1.4.2 Author’s contributions
1.4.3 Author’s publications
2 Multi-cell VLC Netwotks
2.2 VLC system
2.2.3 VLC channel
2.2.4 Signal modulation
2.2.5 VLC uplink
2.2.6 Multi-cell VLC system
2.3 Interference mitigation in multi-cell VLC networks
2.3.1 Linear pre-coding
2.3.2 MA techniques
2.4 Optimization of VLC systems
2.5 Chapter summary
3 UAV Locations Optimization inMISO ZF Pre-coded VLC Networks
3.1.1 UAV-based BSs
3.2 System model
3.3 UAV locations optimization
3.4 Performance analysis
3.4.1 Performance over different optimization parameters
3.4.2 Case study of UAVs with optimized and non-optimized locations
3.4.3 Performance of optimized UAV locations in case of mobility
3.5 Chapter summary
4 Rx FOV Optimization inMISO ZF Pre-coded VLC Networks
4.2 System model
4.3 FOV optimization
4.4 Performance analysis
4.4.1 Performance over different optimization parameters
4.4.2 Case study of optimized and non-optimized FOVs
4.4.3 Performance over random Rx locations
4.4.4 Performance in the case of user mobility
4.5 Chapter summary
5 Performance Comparison Between OFDMA and NOMA
5.2 Performance analysis
5.2.1 OFDMA-based signal transmission
5.2.2 NOMA-based signal transmission
5.3 Performance comparison
5.4 Chapter summary
6 Hybrid NOMA-ZF pre-coding for VLC Networks
6.2 Hybrid NOMA-ZFP
6.2.2 NOMA-ZFP schemes
6.3 Performance study of hybrid NOMA-ZFP schemes
6.3.1 Main assumptions and considered scenarios
6.3.2 Performance study
6.3.3 Computational complexity and network latency
6.4 Chapter summary
7 Time-Sliced NOMA forMulti-Cell VLC Networks
7.2 System model
7.3 TS-NOMA signaling
7.3.2 TS-NOMA schemes
7.3.3 Time-slot fixing strategies
7.4 Performance analysis
7.4.1 Main assumptions and considered scenarios
7.4.2 Numerical results
7.5 Chapter summary
8 Addressing User Handover inMulti-Cell VLC Networks
8.2 Handover-aware scheduling
8.2.1 Single cluster network
8.2.2 Multiple cluster network
8.3 Time scheduling strategies study
8.3.1 Time scheduling strategies
8.4 Performance analysis
8.4.1 Main assumptions and considered scenarios
8.4.2 Numerical results
vi TABLE OF CONTENTS
8.5 Chapter summary
9 Experimental Investigation of Effect of PA in NOMA
9.2 Experiment setup
9.3 Effect of varying PA coefficient in NOMA
9.4 Chapter summary
10 Conclusions and Perspectives