1. Introduction
Advancements in light emitting diode (LED) technology have caused LEDs to have long life cycles and to overheat less during operation, making them more cost- and energy-efficient. This has resulted in their large-scale use. LEDs are not only used as a light source but can also be used as transmitters for visible light communication (VLC) [
1]. As VLC can provide high speeds and security, it can be an appropriate solution for indoor wireless communications [
2]. VLC can be a solution to overcome the shortage of radio frequency resources, as it adopts an unlicensed spectrum. As such, the potential of VLC can be advantageous for industrial wireless applications. There have also been considerations to apply VLC to industrial wireless communications [
3].
As non-orthogonal multiple access (NOMA) can achieve high spectral efficiency by adopting superposition coding (SC), it can simultaneously transmit signals to both the near-user (NU) and the far-user (FU) using the same frequency resource [
4]. Therefore, NOMA was proposed for VLC [
5,
6,
7,
8].
The signal for FU can be canceled by successive interference cancellation (SIC); then, the signal for NU can be decoded [
9]. Instead, as the signal for FU is with the highest power allocation, the signal can be decoded without SIC. VLC-NOMA systems were evaluated in [
7,
10] in terms of their coverage probability and ergodic sum rate. Kizilirmak et al. evaluated the performance of VLC-NOMA compared with orthogonal frequency-division multiple access in terms of data rate [
11]. Marshoud et al. evaluated the bit error rate (BER) of a NOMA system made up of a user pair including three or more users [
12].
In this study, we describe the VLC-NOMA system applied to factory automation, proposing a deep neural network (DNN)-based power allocation scheme and a priority-based user pairing (PBUP) scheme. As mentioned above, VLC adopts unlicensed bandwidth, is free from radio frequency interference, and has secure transmission by signals not penetrating walls [
8]. Therefore, the advantages of VLC can be considered solutions for factory automation. For more additions, factory automation should handle periodic transmission for each node [
13]. Therefore, handling periodic transmission should be scheduled, and the usage of resources for scheduling should be reduced to avoid missed deadlines of real-time transmission. Adopting NOMA can reduce the usage of resources, as NOMA can transmit multiple signals within the same resource.
We derived the BER performance of both users in a user pair, in terms of both perfect and imperfect SICs, and showed a balanced power ratio that considers both NU and FU in terms of reliability. Reliability is a key performance indicator for defining the stability and accuracy of a system [
14]. To improve reliability, DNN-based power allocation (DBPA) was proposed with the derived BER performance.
Previous studies have applied DNNs to VLC. Lee et al. proposed a deep learning framework for the design of an on–off keying (OOK)-based binary signaling transceiver in dimmable VLC systems [
15]. Ulkar et al. suggested a VLC module that uses a DNN to reduce the complexity caused by handling both the error rate and level of illumination simultaneously [
16]. However, these studies did not consider NOMA. Although Lin et al. suggested a signal DNN-based demodulator in a VLC-NOMA system [
17], their study applied a DNN to a demodulator rather than a power allocation algorithm. In contrast, the proposed DBPA in this study focuses on finding the minimum point of average BER to improve reliability for both users.
In addition, there have been studies for resource allocation in NOMA systems with radio frequency. Ali et al. suggested a power allocation algorithm using DNN in a device-to-device NOMA system to maximize sum capacity [
18]. Kumaresan et al. studied user-clustering and power allocation in downlink NOMA systems to maximize throughput [
19]. Fu et al. suggested dynamic power control for NOMA transmissions to minimize the transmission delay with the consideration of each user’s transmission deadline and the total power constraint [
20]. Manglayev et al. proposed machine-learning- and deep-learning-based power allocation schemes that find near-optimal solutions with regard to sum capacity and provide low computational costs in NOMA systems [
21]. As mentioned in the above studies, DNNs can be used for resource allocation for wireless communication systems due to the low computational complexity [
22].
Moreover, VLC is considered for industrial applications [
3]. There have been reliability issues for industrial applications [
23,
24]. Therefore, in this study, we proposed DBPA and PBUP as solutions to improve reliability and real-time transmission.
Even though the VLC-NOMA system can provide efficient transmissions to reduce the usage of resources, the decoding error is larger than that of the VLC orthogonal multiple access (OMA) system. The VLC-OMA allocates the whole transmit power for one data transmission in a resource. Instead, the VLC-NOMA system transmits superposed signals splitting transmit power; then, the decoding error is larger than that of the VLC orthogonal multiple access (OMA) systems. It can cause significant problems for factory automation. Therefore, the reliability of the VLC-NOMA system should be improved. DBPA can be considered a solution to cover the decoding error caused by splitting transmit power. In this study, DBPA is proposed to improve the reliability of the VLC-NOMA system for factory automation. Moreover, the proposed VLC-NOMA system should handle two signals at a timeslot.
As the VLC-NOMA system should handle a maximum of two signals in a timeslot, each signal for each node should be paired. Instead, the VLC-OMA system has to only schedule transmitting orders in timeslots. However, the VLC-NOMA system should pair two nodes before scheduling. Therefore, in this study, PBUP is considered a way to pair two nodes based on transmission priority.
The contributions of this study are as follows:
Evaluating the reliability of the VLC-NOMA system in terms of the perfect and imperfect SICs. It provides the simulation and numerical results in terms of BER.
Proposing VLC-NOMA system for a factory automation scenario to solve missed deadlines caused by a lack of resources.
Proposing DBPA to solve the degradation of reliability caused by adopting the VLC-NOMA system.
Proposing PBUP to pair two nodes based on priority and channel state for the VLC-NOMA system.
The remainder of this paper is structured as follows: The system model is described in
Section 2;
Section 3 proposes DBPA and PBUP;
Section 4 explains the evaluation and analysis of the downlink VLC-NOMA system in terms of reliability and of DBPA compared with other power allocation algorithms, and evaluates PBUP in a factory automation scenario;
Section 5 concludes the paper and describes future work.
2. System Model
This study is on the VLC-NOMA system for a factory automation scenario. Therefore, the first subsection is the system model for the VLC-NOMA system and the second subsection is the scenario of factory automation.
2.1. The VLC-NOMA System
In this subsection, a downlink VLC-NOMA system is described for the BER performance. The system is made up of a single LED as a transmitter, and two users with photodiodes (PDs) as receivers in an NOMA user pair in an indoor environment. The modulation technique adopts OOK, and the OOK signals for NU and FU are superposed for transmission from the transmitter.
The line-of-sight VLC channel model was averaged between the LED and users [
25]. The channel model is expressed as follows:
The notations of the above equations are described in
Table 1. The channel gain between the LED and a user can be calculated by (
1).
NOMA superimposes and transmits signals to users. As described in
Figure 1, the system is made up of one LED as a transmitter and two users, including NU and FU. An equation for it is as follows:
As the system superimposes signals, NU signal and FU signal are superimposed.
The signal for NU is regarded as interference for FU. Therefore, the signal
is as follows:
The signal
for NU after SIC is as follows:
As the signal for FU is canceled by SIC, only the other signal for NU is residual and then decoded by NU. However, in cases where SIC fails, decoding the signal is hard for NU. Therefore, the success ratio of decoding signals to FU affects the success of SIC.
To calculate the error mentioned above, the BER equations for both users are going to be dealt with. There are four cases for bits of NU and FU using OOK in the downlink VLC-NOMA system.
where
denotes the probability of bits,
x for FU, and
y for NU.
The constellation for the received signal is described in
Figure 2. We assume
. In the case of received signals, the constellation of the receiver is affected by channel gain
h. Therefore, the larger
h is, the lower BER is. The threshold for signals for FU is
, because FU should decode the signal regardless of the existence of the signal for NU, as shown in
Figure 2. This means that irrespective of whether the bit for NU is 0 or 1, the range of signals for FU starts at the same point,
.
Therefore, the BER for FU, where
x is 1 and
y is 0, is as below:
means energy per bit;
is noise power; and as the bit for FU is 1 and the bit for NU is 0, the signal for (
7) will be converged at
in the constellation shown in
Figure 2. The reason why
was not expressed in (
7) is that the bit 0 for NU means no power in OOK. When the bit is 0, no power is allocated for transmitting the signal.
As shown in
Figure 2, the boundary of the signal for FU is from
regardless of the signal for NU. This means that the threshold for FU will not be changed by the existence of signals for NU. Then, as
, the cases
, which are
and
, are considered as simple OOK BERs. As drawn in
Figure 2, the decision threshold is
. Therefore, the equation for the cases
and
is as follows:
Finally, the BER for FU is calculated by (
14).
Next, the BER for NU must be calculated. First, perfect SIC is assumed. The BER equation simply adds the power ratio for NU into a simple OOK BER equation. The BER equation for an NU with a perfect SIC is as follows:
Based on (
14) and (
15), the BER equation for an NU with imperfect SIC can be derived. There are two cases of error: one wherein SIC succeeds and decoding fails, and the other wherein both SIC and decoding fail. The first case is as follows:
The second case is as follows:
The reason for the subtraction of (
14) in (
17) is to take into account removing the cases wherein NU can decode its data even though the SIC fails. To continue the derivation of the equations,
is derived below:
Then, the final equation for the failed SIC is
Finally, the BER of NU with imperfect SIC is
Using (
14) and (
26), the datasets for the training model can be generated. A detailed explanation for generating the datasets is provided in
Section 3.1.
2.2. Factory Automation Scenario
In this study, we applied the VLC-NOMA system for the factory automation scenario [
13,
26]. The scenario is established based on time-division multiple access (TDMA) and superframe structure. The scenario includes periodic real-time messages sent by dedicated time slots considering beacon enable mode. The message characteristics are described in
Figure 3.
Computation time c is less or equal to deadline d, d is less or equal to period p. After the released time, a message will be generated with cycle p and should be transmitted by d. Every message for the VLC nodes has the characteristic mentioned above. Therefore, the message transmissions should be managed to be finished within their deadlines.
Figure 4 shows frame structures for VLC-OMA and VLC-NOMA systems. VLC-OMA and VLC-NOMA systems have the same frame duration consisting of 25 timeslots. The beacon starts at the beginning of a frame. Each timeslot is 10 ms. The difference between VLC-OMA and VLC-NOMA is whether to superpose signals or not.
Figure 4a shows that the VLC-OMA system transfers signal for a node in a timeslot. Instead, shown in
Figure 4b, the VLC-NOMA system can transfer two signals for two nodes in a timeslot by allocating different power levels.
Arrangements of nodes and LED are described in
Figure 5. There is one LED, as a base station, and multiple nodes. Messages for the nodes are generated periodically from the LED. The messages have the characteristics described in
Figure 3. Although a previous work [
1] considered one message in one timeslot, the proposed system in this study considers two messages in one timeslot as it adopts the NOMA system with the superposition coding. Next,
Section 3 describes resource allocation schemes for the VLC-NOMA system in the factory automation scenario.