1. Introduction

The continuous dramatic growth in data traffic demand has motivated researchers to explore new spectra, techniques, and network architectures. Among the alternative solutions, visible light communication (VLC) has emerged as one of the most promising candidates for fifth generation (5 G) and beyond, owing to the popularity of light-emitting diodes (LEDs) [1]. By modulating the signals onto the LEDs, the VLC realizes illumination and communication simultaneously. VLC networks outperform radio frequency networks [2], in terms of data rates, energy efficiency, battery consumption, and latency [3–5].

Because more than 80% of mobile data traffic is predicted to be consumed indoors, indoor high-speed data transmission is considered to be one of the most important applications of VLC [6]. However, the modulation bandwidth of common commercial LEDs, which ranges from several megabits to tens of megahertz, limits the data rate of VLC systems [7,8]. In addition, multiple users commonly request access to the same LED simultaneously, which causes inevitable multiuser interference. Therefore, multiple access (MA) techniques with high spectral efficiency and reliability are crucial for deploying practical VLC networks. Orthogonal frequency division multiple access (OFDMA) [9–11] and non-orthogonal multiple access (NOMA) [12–15] are two MA techniques widely used in VLC networks. By allocating orthogonal subcarriers to different users, OFDMA can achieve multi-user access without interference, which also benefits from its simplicity in implementation and flexibility in resource allocation. However, the spectral efficiency was low in OFDMA VLC networks. Unlike OFDMA, NOMA, which multiplexes in the power domain, is a non-orthogonal MA technology in which the entire spectrum can be used for all users. NOMA has the advantages of a higher spectral efficiency and better fairness. However, it suffers from inevitable co-channel interference, especially when too many users are served.

Recently, NOMA combined with OFDMA has been proposed to achieve better performance than pure OFDMA or NOMA systems. For the first time, a NOMA-OFDMA scheme for bidirectional VLC systems was proposed and evaluated in [16]. In this scheme, each user can use part or entire subcarriers, where the users allocated to the same subcarriers can be multiplexed in the power domain. In [17], a hybrid MA technique was proposed such that part of the spectrum was reserved for orthogonal access, and the rest was reserved for all users by NOMA. The simulation results validated that the proposed hybrid scheme provided better performance than NOMA in terms of the overall achievable sum rate and coverage probability. Joint power and resource allocation were studied when NOMA and OFDMA were combined for hybrid downlink MA [18]. Based on the concept of minimizing transmit power, an equivalent convex optimization problem was derived and solved. However, error propagation was not considered during multi-user decoding. Chen et al. proposed and experimentally demonstrated a fairness-aware hybrid NOMA/OFDMA scheme for multiuser VLC systems [19]. In the scheme, all users were divided into user pairs, where NOMA and OFDMA were applied for intra-pair and inter-pair MA, respectively. As can be seen, the above studies on hybrid NOMA/OFDMA technology have all focused on single-LED multi-user VLC systems. However, multiple LEDs multi-user VLC systems are considered more common in practice because multiple wide-beam LEDs are required to provide uniform illumination in typical lighting systems, leading to the overlap of the illumination footprints from different LEDs. As a result, multi-user interference becomes more complicated, as users not only interfere with each other within the same cell, but also with users in neighboring cells. To the best of our knowledge, the hybrid NOMA/OFDMA scheme has not been studied for multi-cell multiuser VLC systems.

Moreover, the spectrum resource should be reused as much as possible without interference among different cells to improve spectral efficiency [20,21]. Coordinated multiple-point (CoMP) technology is an effective way to manage inter-cell interference, where the coordination of multiple LEDs turns the problem of overlap and thus interference into an advantage [22]. In existing research on CoMP, multiple LEDs and users are regarded as a multi-user multiple-input multiple-output system, where the interference is eliminated by precoding [23]. However, accurate channel information must be fed back from the receiver to the transmitter. In fact, multiple LEDs and each user located in the overlapping area can also be considered a multiple-input single-output (MISO) system, indicating that the user can receive information from multiple LEDs. Repeated coding is the simplest but most effective method for MISO systems, where the received signal-to-noise ratio (SNR) can be improved while eliminating intercell interference [24].

In this study, we propose a hybrid NOMA/OFDMA scheme with CoMP transmission for multi-cell multi-user VLC systems. In this scheme, the illuminated coverage of an LED is defined as an optical cell, where each cell is further divided into multiple sub-cells. The proposed hybrid NOMA/OFDMA scheme is described as follows. First, multiple sub-cells are multiplexed by OFDMA to avoid performance degradation caused by co-channel interference in NOMA systems. Each sub-cell is allocated to a subset of subcarriers considering user fairness. Second, within the same sub-cell, multiple users are multiplexed by NOMA to share the spectrum, which greatly improves the spectral efficiency. Third, CoMP transmission based on repeated coding is introduced for the users of the overlapping areas to eliminate multi-user interference and improve the received SNR. Benefitting from the CoMP transmission, each cell can use all spectra with a frequency reuse (FR) factor equal to 1. Furthermore, a two-dimensional (2D) power allocation algorithm is proposed for hybrid NOMA/OFDMA VLC systems, where power is allocated between sub-cells and users within sub-cells. Targeting the minimization of the average symbol error rate (SER), optimal 2D power ratios were derived based on the fixed power allocation strategy. In particular, the impact of error propagation owing to NOMA decoding is considered in the derivation of the SER expression. The performance of the proposed system was thoroughly investigated by computer simulations, where the SER performance was evaluated on different VLC networks. The simulation results also confirm the SER performance improvement of the proposed hybrid NOMA/OFDMA system compared with the traditional NOMA and OFDMA systems. Finally, the results of the proof-of-concept experiment further prove the superiority of the proposed system.

2. Hybrid NOMA/OFDMA VLC system

2.1 Principle of the hybrid NOMA/OFDMA scheme

A typical indoor VLC network is shown in Fig. 1(a). The red dots denote LEDs in the network. Each circle represents an optical cell, and the cell size is determined by the user’s field of view (FOV), vertical distance, attenuation coefficient, etc. [21]. The coverage area consists of four types, denoted as L₁, L₂, L₃, and L₄, as shown in Fig. 1(b), where the subscript represents the number of LEDs that can receive line-of-sight signals in the corresponding area. According to the concept of the proposed hybrid NOMA/OFDMA scheme, the sub-cells are multiplexed by OFDMA, where each sub-cell is allocated a part of the bandwidth. Figure 1(b) and (c) depict the spectrum allocation among the different sub-cells. It is worth noting that the sub-cells do not strictly correspond to the four types of coverage areas. As shown in Fig. 1(b), the areas colored pink and yellow are assigned to different bandwidths, although they belong to the same coverage area type that received signals from two LEDs. Users in area L₃ share bandwidth with the users in area L₂ of the same cell. Because area L₄ is small and the probability of user existence is reasonably small, an individual bandwidth is allocated to users in area L₄ to avoid interference. As can be seen, the FR factor is equal to 1 because each cell can use all bandwidths. In addition, benefitting from OFDMA, the subcarriers are flexibly assigned to each sub-cell; that is, the subcarriers allocated to each sub-cell can be discontinuous.

 

Fig. 1. Typical indoor VLC network model (a) Indoor layout of illumination (b) Sub-cell division (c) Frequency-power domain multiplexing.

Download Full Size | PDF

In general, sub-cells can be categorized into two types: sub-cells of overlapping areas (L₂, L₃, and L₄) and sub-cells of non-overlapping areas (L₁). For users located in the sub-cells of overlapping areas, each user is served by multiple LEDs. Therefore, one user and multiple LEDs form an MISO system that cooperates based on the concept of repeated coding. For users in the areas only served by one LED, one user and one LED form a simple single-input single-output system. Once there are multiple users in the same sub-cell, they are multiplexed by NOMA, where signals are superposed in the power domain, as shown in Fig. 1 (c).

An example of a hybrid NOMA/OFDMA VLC network with 4 optical cells and 11 users is shown in Fig. 2 to better explain the proposed scheme's principle. In the figure, the user is denoted by a square and the LED is marked by a red circle. As can be seen, users numbered 2, 3, 5, 8, and 10 are located in the non-overlapping area served by the LEDs of their own cells. User2 and User3 are multiplexed in the power domain because they belong to the same cell. Users located in the overlapping areas colored pink and yellow cooperate with two adjacent

 

Fig. 2. Example of the proposed hybrid NOMA/OFDMA VLC network.

Download Full Size | PDF

LEDs. For example, LED1 and LED2 send the same signals to User4, indicating that User4 receives the superposed signals from the two transmitters. Users located in areas colored blue are illuminated by three LEDs, which can use the bandwidth of users in the yellow and pink areas. For example, User7 receives two independent data streams simultaneously. One data stream is transmitted cooperatively by LED3 and LED4, which occupy the same bandwidth as the data of User9. The other data stream sent by LED2 and LED3 concurrently is multiplexed with the data of User6 in the power domain. User1 in the green area, receives the signals sent from the four LEDs. Thus, extra bandwidth was assigned to avoid interference.

2.2 System model

Without loss of generality, a simplified VLC network consisting of two LEDs and five users is presented in Fig. 3, where both sub-cell types are included. According to the concept of the hybrid NOMA/OFDMA scheme, the multiplexed manners of different users in the power and frequency dimensions are depicted in Fig. 3.

 

Fig. 3. Simplified VLC network.

Download Full Size | PDF

Figure 4 illustrates the detailed system block diagram, where the system model for users in the non-overlapping area is shown in Fig. 4(a) and that in the overlapping area is shown in Fig. 4(b). After M-quadrature amplitude modulation (M-QAM) modulation and power allocation, the signals of different users within the same sub-cell are superposed in the power domain, which can be expressed as

 

Fig. 4. System block diagram of users (a) in non-overlapping sub-cell (b) in overlapping sub-cell.

Download Full Size | PDF

The superposed signals are then mapped to the subcarriers. Because only positive and real signals can be transmitted in VLC systems, the modulated signals in the frequency domain must be symmetrically conjugated to ensure that the signals in the time domain are real. After the inverse fast Fourier transform (IFFT) and cyclic prefix (CP) attachment, direct current (DC) is supplied to guarantee that the transmitted signals are nonnegative. Finally, signals are transmitted from one LED or multiple LEDs in the form of optical power, depending on the number of LEDs that the user can receive.

At the receiver, each user can receive signals from one or multiple LEDs. Frame synchronization is performed to detect the starting positions of the data streams. After the CP removal and fast Fourier transform (FFT), the received signals of the k^th user in the frequency domain can be expressed as

Channel was estimated in the frequency domain based on the least square criterion and then the signal distortion induced by the channel was eliminated by channel equalization. Subsequently, the serial interference cancellation (SIC) algorithm was employed for NOMA decoding, where the user with higher power was decoded first. Finally, the original bit data stream is recovered after QAM demodulation.

3. Proposed 2D power allocation algorithm

The performance of the NOMA system depends mainly on power allocation. Therefore, power should be allocated to different users within the same sub-cells according to the channel gains [25]. Moreover, the power of the signals received by users in the overlapping area is higher because of the superposition of the signals sent from multiple LEDs. Therefore, more power tends to be allocated to users in the non-overlapping area to achieve better performance, whereas users in overlapping areas require less power. Consequently, we propose a 2D power-allocation algorithm for the proposed hybrid NOMA/OFDMA VLC system. To minimize the average SER of the system, the power is allocated in two dimensions of the sub-cells and the users within the sub-cells concurrently, which is based on the fixed power allocation strategy. In particular, the impact of error propagation owing to SIC decoding is considered to derive a more accurate expression of the SER.

Next, we derived the closed form of the SER expression. In each sub-cell, multiple users are multiplexed by NOMA, whose signals are decoded based on the SIC algorithm at the receiver. Obviously, error propagation occurs during SIC, leading to imperfect SIC. Therefore, there are two cases for solving the SER. In Case1, the SER of the first decoding user is solved without considering the error propagation (the single-user system can also be regarded as this case). While in Case2, the impact of error propagation should be considered when solving the SER of the remaining users in a multiuser system.

Case1:

According to [26], the user with the lowest channel gain is assigned the highest power and is decoded first. Assume that there are K users in a sub-cell, and they are sorted in order from far to near. User1 is the farthest from the access point with the lowest channel gain. Therefore, User1 is decoded first at the receiver, whose SER can be expressed as

Taking User1 in Fig. 3 as an example. Assuming that the symbols are modulated by 4QAM, SER can be solved as follows:

Case2:

For the remaining users except User1, the SER of the k^th user can be solved by

The Eq. (6) represents the SER of the k^th user, assuming that the decoding has been completed from the 1^st user to the (k-1)^th user, where k = 2…K, v denotes all possible error combinations, and $P({e_k^{(\nu )}|e_1^{(\nu )},e_2^{(\nu )},\ldots ,e_{k - 1}^{(\nu )}} )$ denotes the SER of the k^th user under the v^th error combination. e_k^(v) was used to determine whether an error had occurred. When ${\hat{S}_k} = {S_k}$, e_k^(v) equals zero, indicating that the k^th user is decoding correctly. Otherwise, e_k^(v)= 1 indicates that an error arises, that is, ${\hat{S}_k} \ne {S_k}$.

Furthermore, the SER of the k^th user has two possibilities. One possibility is the case without error propagation, where ${\hat{S}_m} = {S_m}$ is used for $1 \le m \le k - 1$. Then, the SER of the k^th user can be expressed as

The other is the case with error propagation, that is, an error occurs for at least one user among-k-1 users. In this case, the SER of the k^th user can be expressed as:

Finally, the SER of the k^th user can be given by

Taking User2 in Fig. 3 as an example, the SER of User2 can be expressed as

The second term is solved by

The detailed expression of Eq. (13) is denoted as

After deriving the SER, the optimal power ratios were obtained by formulating an optimization equation. Based on the fixed power allocation strategy, power is allocated jointly among sub-cells and users within sub-cells to minimize the average SER of the system, which can be expressed as the following optimization problem:

According to the formulated optimization equation, the 2D optimal power ratios can be solved based on the genetic algorithm, which can be summarized as follows:

4. Simulation results and discussion

In this section, we investigate the performance of the proposed hybrid NOMA/OFDMA VLC system using computer simulations, which are also compared with the traditional OFDMA and NOMA VLC systems. Considering the channel frequency response of exponential attenuation, odd sub-carriers are assigned to the non-overlapping sub-cells, and even sub-carriers are assigned to the overlapping sub-cells to achieve user fairness. The channel was modeled as described in [24], where the channel gain of an optical propagation link can be calculated as follows:

The parameters used in the simulations are presented in Table 1.

Table 1. System Parameters

View Table

Figure 5 depicts the first VLC network for simulation, where User1 and User2 are located in the non-overlapping sub-cell illuminated by LED1, User3 is located in the overlapping area of LED1 and LED2, and User4 is in the non-overlapping sub-cell of LED2. The detailed coordinates of the users are shown in the figure. As defined above, the sub-carriers colored in blue represent the subcarriers allocated to the non-overlapping sub-cells, whereas the red ones represent the subcarriers allocated to the overlapping sub-cell. In the simulation, the modulation format for User1 and User2 is 4QAM, and User3 and User4 use 16QAM modulation. The transmitted power of each LED is equal to 1. The noise variance for each user is set as 0.0009. According to the 2D power allocation algorithm proposed in this study, the optimal power ratio between User1 and User2 denoted as α, is equal to 4.42, and the power ratio between non-overlapping sub-cells and overlapping sub-cells, denoted as β, is equal to 1.78.

 

Fig. 5. The first VLC network.

Download Full Size | PDF

The SER performance versus different power ratios is presented in Fig. 6. The curve named “average” stands for the average SER of the system. In the figures, the solid curves represent the results of the Monte Carlo simulation, whereas the dotted curves express the theoretical results solved by the SER equations derived in this study. The results of the Monte Carlo simulation are consistent with the theoretical results, verifying the correctness of the SER expressions. Figure 6(a) shows the SER performance of different users with increasing β when α is set as the optimal value. As can be observed, except for User3, the SERs of the remaining users decrease with increasing β. This is because the power of the overlapping sub-cells decreases when β increases, leading to a worse SER for User3. Instead, users located in non-overlapping areas achieved better performance. When β was approximately 1.8, the average SER was the lowest, which was consistent with the theoretical value obtained using the 2D power allocation algorithm. In Fig. 6(b), the SERs of different users varying with α are shown under the condition of optimal β. As User3 and User4 are located in their individual sub-cells, the SERs are independent of α. With an increase in α, the power of Uer1 also increased, resulting in a better SER performance. While for User2, the SER first increased and then decreased. Initially, the SER of User2 decreases because the impact of the error propagation caused by User1 reduces. However, when α continues to increase, the power allocated to User2 becomes lower, and thus the SER of User2 rises again owing to the lower SNR. The best average SER is achieved when α is approximately 4.5, which is consistent with the optimal theoretical value.

 

Fig. 6. SER performance versus different power ratios for the first VLC network (a) β (b) α.

Download Full Size | PDF

Furthermore, we compared the proposed hybrid NOMA/OFDMA scheme with the traditional OFDMA and NOMA schemes. In the OFDMA scheme being compared, the sub-cell division and spectrum allocated to sub-cells are the same as those in the proposed scheme, where the users located in the overlapping areas still cooperate based on the concept of repeated coding. The difference is that users within the same sub-cell are multiplexed using OFDMA. The multi-cell NOMA scheme for comparison is referred to [14], where adjacent cells use different bandwidths to avoid inter-cell interference. Thus, the FR factor was reduced to 1/3. Users in the same cell are multiplexed using NOMA. To ensure fairness in the comparison, the data rates of the different systems are kept the same. In addition, the power ratios involved in traditional NOMA and hybrid NOMA/OFDMA systems are assumed to be optimal.

An SER performance comparison of the different schemes is shown in Fig. 7. In the simulation, the SNR is defined as the transmitted SNR of the whole system, which can be solved by

 

Fig. 7. SER comparisons of the different schemes for the first VLC network.

Download Full Size | PDF

As mentioned above, the transmitted power is the same for each LED, which is kept fixed in the simulation. The noise variance is also assumed the same for each user. Therefore, the SNR is changed by adjusting the noise variance. The hybrid NOMA/OFDMA system achieved the best SER performance. Compared with the NOMA scheme, the FR factor of the hybrid NOMA/OFDMA scheme is much higher, indicating that the spectral efficiency is significantly improved. As a result, higher-order modulation is required for the NOMA scheme to maintain the same data rate, resulting in the degradation of the SER performance. Moreover, OFDMA was introduced to realize the multiplexing of multiple sub-cells in the hybrid NOMA/OFDMA system, which alleviates the co-channel interference of NOMA caused by too many users. As for the OFDMA scheme, although the FR factor is also equal to 1, multiple users within the same sub-cell are multiplexed by OFDMA, which reduces the spectral efficiency as well. Thus, high-order modulation is necessary to maintain the data rate constant.

Figures 8 and 9 present the simulation results when User1 moves within the same sub-cell, where the VLC network is the same as the first network and the simulation parameters are kept the same as well. After User1 moving from the center of the sub-cell to the edge, the coordinates of User1 is changed to (0.14 m, 0 m, 0.8 m). In this case, the solved theoretical optimal power ratios α and β are equal to 5.95 and 2.10 respectively. Due to the movement of User1, the corresponding channel gain of User1 changes. As a result, the power ratio between User1 and User2 changes obviously, which also leads to the difference of the power ratio between non-overlapping and overlapping sub-cells. The curves in Fig. 8 (a) and (b) are similar to those in Fig. 6 (a) and (b), except that the optimal values of the power ratios are different. Compared with the traditional schemes, the simulation results in Fig. 9 show that the proposed scheme still achieves the best SER performance.

 

Fig. 8. SER performance versus different power ratios for the first VLC network when User 1 moves (a) β (b) α.

Download Full Size | PDF

 

Fig. 9. SER comparisons of the different schemes for the first VLC network when User 1 moves.

Download Full Size | PDF

Figure 10 provides the second VLC network for simulation, which can be considered as the scenario that User2 moves from the non-overlapping sub-cell of LED1 to the overlapping sub-cell. Thus, User2 and User3 are multiplexed by NOMA in the overlapping sub-cell. In the simulation, the modulation format for User1 and User4 is 16QAM, and the modulation format of User2 and User3 data is 4QAM. The transmitted power of each LED is set as 1. The noise variance of each user is equal to 0.0009. According to the proposed 2D power allocation algorithm, the optimal 2D power ratios α and β are equal to 4.11 and 1.61 separately.

 

Fig. 10. The second VLC network.

Download Full Size | PDF

The evaluations of the SER performance versus different power ratios are provided in Fig. 11, where the power ratio α is set as the optimal value in Fig. 11(a), and the power ratio β is optimal, as shown in Fig. 11(b). The Monte Carlo simulation results, denoted by solid curves, are also consistent with the theoretical SER results, depicted by dotted curves. In Fig. 11(a), the SERs of User2 and User3 increase with the increase in β, as the power allocated to the overlapping sub-cell decreases when β is increased. The SERs of User1 and User4 both decrease with the increase in β, where the two SER curves almost coincide. This is because the two users are assigned the same power, and the distance between the users and access points is the same as well. When β is approximately 1.6, the average SER of the system is the lowest, which is consistent with the theoretical value obtained using the 2D power allocation algorithm. When β is fixed, the power allocated to User1 and User4 remains unchanged, and thus the SERs remain constant. As the power ratio α increases, the SER performance of User3 improves, whereas the SER of User2 decreases first and then increases, which is similar to the case of User1 and User2 in Fig. 6(b) for the same reason. Figure 12 shows the average SERs of the three schemes versus the SNR. Similar to the simulation results for the first VLC network, the hybrid NOMA/OFDMA scheme achieved the best SER performance.

 

Fig. 11. SER performance versus different power ratios for the second VLC network (a) β (b) α.

Download Full Size | PDF

 

Fig. 12. SER comparisons of the different schemes for the second VLC network.

Download Full Size | PDF

In Fig. 13, the third VLC network for simulation is illustrated, where one user is added compared with the first network. User1 and User2 are in the non-overlapping sub-cell illuminated by LED1, User5 is located in the non-overlapping sub-cell of LED2, and User3 and User4 are located in the overlapping sub-cell of the two LEDs. In the simulation, the data of User5 is modulated with 16QAM format, and the remaining users all employ 4QAM modulation. The transmitted power of each LED and the noise variance for each user are still set as 1 and 0.0009 separately. Since User1 and User2, and User3 and User4 are multiplexed by NOMA in their own sub-cells, there are two power ratios: the power ratio α₁ between User1 and User2 and the other is the power ratio α₂ between User3 and User4. Based on the 2D power allocation algorithm, the optimal power ratios α₁, α₂, and β were 4.24, 4.13, and 1.66, respectively.

 

Fig. 13. Third VLC network.

Download Full Size | PDF

When α₁ and α₂ are set as optimal values, the SER performance of different users is evaluated when β is varied. As shown in Fig. 14(a), the SERs of User3 and User4 become higher with an increase in β because of the decrease in the allocated power to the overlapping area. Instead, the SER performance of User1, User2 and User5 in the non-overlapping sub-cells is improved. The average SER is lowest when β is approximately 1.6, which is consistent with the theoretical value. When β reaches the optimal value, the SER performance of User1 and User2 versus the power ratio α₁ is shown in Fig. 14(b), and the SERs of User3 and User4 versus the power ratio α₂ are depicted in Fig. 14(c). The “average” in the two figures indicates the average SER of users within the same sub-cell. The simulation results of the two figures are similar, where the SER of the decoded user first decreases with an increase in the power ratio. This is because the power allocated to users increases as the power ratio increases. However, the SER of the other user first decreases and then increases, indicating that the SER is not only related to the allocated power, but is also affected by error propagation.

 

Fig. 14. SER performance versus different power ratios for the third VLC network (a) β (b) α₁ (c) α₂.

Download Full Size | PDF

Figure 15 compares the system SER of the different schemes. As can be seen, the hybrid NOMA/OFDMA scheme has the best SER performance. Compared with the results in Figs. 7 and 12, the SER performance improvement of the hybrid NOMA/OFDMA scheme appears more significant. This is because there are more users in the VLC network. Therefore, the co-channel interference of the NOMA scheme becomes more serious, and the spectral efficiency of OFDMA is reduced further.

 

Fig. 15. SER comparisons of the different schemes for the third VLC network.

Download Full Size | PDF

5. Experimental set-up and results

Further, an experimental demonstration is implemented to study the performance of the proposed system. In Fig. 16, the experimental set-up is illustrated. According to the system block diagram in Fig. 4, the signal processing is programmed using MATLAB, and the offline transmitted signals are generated, which are uploaded to an arbitrary function generator (AFG: Tektronix AFG3252C). Then, the electrical signals are amplified by an electrical amplifier (EA: Mini-Circuit ZHL-6A-S+), and coupled with DC through a DC bias (Mini-Circuits ZFBT-4R2GWFT+) to drive the red-light LED (Cree XLamp XP-E), where the signals are transmitted in the form of optical power. There are two LEDs representing two optical cells at the transmitter. In the receiver, four APD modules (Hamamatsu C 12702-11) are used to represent four users. Two APDs are placed in the non-overlapping area, and two APDs are placed in the overlapping area of two LEDs to simulate the second VLC network. Once the APDs convert the optical signals to electrical signals, the electrical signals are recorded by a high-speed oscilloscope (OSC: Tektronix MDO4104C) and forwarded for offline processing.

 

Fig. 16. Experimental set-up.

Download Full Size | PDF

In Fig. 17, the measured SERs versus different power ratios are depicted. In the experiment, the subcarrier number was equal to 256 and the length of CP was 8. The modulation format for User1 and User4 was 16QAM, and the modulation format of User2 and User3 data was 4QAM.The transmission rate of the AFG was set to 100 Mb/s with an up-sampling rate of 4; thus, the available bandwidth was approximately 25 MHz. The DC bias current was set as 50 mA. To emulate different positions of the receiver, we adjust the driving peak-to-peak voltages (Vpps) of transmitted signals instead of physically moving the position of the receiver [14]. As can be seen, the trends of SER curves are consistent with the simulation results in Fig. 11. In Fig. 17(a), the SER performance of User1 and User4 is improved while the SERs of User2 and User3 increase when β is increased. The average SER of the system is the lowest when β is around 1.6. When α is increased in Fig. 17(b), the SER of User3 is decreased, while the SER of User2 is decreased first and then increased. The SERs of User1 and User4 remains unchanged roughly. When α is about 5, the system SER reaches the lowest. Figure 18 evaluates the SER performance of the different schemes through experiment, where the SNR is changed by adjusting the value of Vpp. Experimental results also confirm that the proposed hybrid NOMA/OFDMA scheme achieves the best performance, which is consistent with the simulation result.

 

Fig. 17. Measured SERs versus different power ratios (a) β (b) α.

Download Full Size | PDF

 

Fig. 18. Measured SERs of the different schemes.

Download Full Size | PDF

6. Conclusion

In this study, we propose a hybrid NOMA/OFDMA scheme for multi-cell multi-user VLC systems. In this scheme, an optical cell is divided into multiple sub-cells. Different sub-cells are multiplexed by OFDMA, and users within the sub-cell are multiplexed by NOMA. Thus, the proposed hybrid NOMA/OFDMA system achieves higher spectral efficiency than the OFDMA system and reduces the co-channel interference of the NOMA system. Meanwhile, CoMP transmission based on repeated coding is introduced, which eliminates multi-cell interference and improves the received SNR of users in the overlapping area. Furthermore, a 2D power allocation algorithm is proposed for the hybrid NOMA/OFDMA VLC system. Taking the average SER minimum as the optimization objective, the optimal power ratios are solved in the two dimensions of sub-cells and users within sub-cells simultaneously, where fixed power allocation is applied as the basic strategy. In particular, the closed-form expression of SER is derived by considering imperfect SIC. The theoretical values of the SER and optimal power ratios were verified using Monte Carlo simulations. In addition, the simulation results show that the SER performance of the proposed hybrid NOMA/OFDMA VLC system is significantly improved compared with the traditional NOMA and OFDMA VLC systems in different VLC networks, considering the user mobility and user number. Finally, the results of the concept-of-proof experiment further validate the superiority of the proposed system.