Dynamic bandwidth allocation in time division multiplexed passive optical networks: a dual-standard analysis of ITU-T and IEEE standard algorithms

PON as an optical transport solution

TDM-PON is also widely used for connecting residential and business users and serving wireless access segments like 4G and 5G, as shown in Fig. 3. It has a lot of capacity and bandwidth to be a promising optical transport solution for the fronthaul of Open Radio Access Networks (O-RAN) (Jaffer et al., 2020). It shows multi-tenancy, which lets different service providers or virtual network slices use the same physical infrastructure. DBA mechanisms allow resources to be flexibly distributed based on real-time demand, benefiting the O-RAN fronthaul, where traffic patterns can vary dynamically. This technology can integrate with network slicing, creating virtual networks with specific characteristics to meet diverse O-RAN service requirements (Wong & Ruan, 2023). However, further research is required to develop advanced DBAs that can adapt to the dynamic nature of O-RAN architectures, optimise energy consumption, ensure security, and minimise latency.

PON on the premises

Home internet networks must also be assured that network QoS will provide low latency and little packet loss for applications like online education and live streaming. Wi-Fi is utilised; nevertheless, it is inadequate for emerging services. Wi-Fi mesh technology enhances throughput and backhaul quality and mitigates interference. However, the restricted modulation bandwidth of copper media and air interfaces hinders high throughput and leads to inevitable collisions of packets (Effenberger, 2022). Fiber-to-the-room (FTTR) is an innovative fiber-based framework for home networking designed to address this issue. The goal is to ensure that gigabit connectivity is available in every corner of the home, i.e., PON on the premises. In 2021, the ITU-T Q3/SG15 standardized FTTR and designated it as G.FIN (Cases, 2021). The development of FTTR is now in its early phase, with the introduction of novel ideas for using this technology in homes, residential regions with high population density like big apartments, and smart offices. These new PONs would support diverse applications with different bandwidth, latency, and end-to-end delay requirements.

PON in industry

Manufacturing in industries relies on point-to-point data communications, necessitating a standardized and shared networking system. PONs are an appropriate option because they can accommodate different user interfaces and offer reliable and predictable networking, simplifying coordinated scheduling. PONs are highly effective in facilitating many manufacturing applications due to their TDMA characteristics (Effenberger, 2024). The European Telecommunications Standards Institute (ETSI) formed an Industry Specification Group (ISG) to precisely outline the parameters of Fifth Generation Fixed Network (F5G) to deploy industrial PON in 2021 (Newaz et al., 2024). The ISG provides a detailed description of the necessary specifications for PON networks in industrial environments. This includes configuring a multi-tier spine-leaf OLT architecture, the requirements for ONU interfaces, and the process for allocating bandwidths. AI is regarded as a fundamental component in the F5G architecture for network automation (ETSI Recommendations, 2023). Active research is being conducted on using PON during manufacturing to address the need for low latency and jitter at the millisecond level or below to fulfil industrial needs.

Contributions and layout of the article

Even though the recently proposed DBA/DBWA algorithms are deemed adequate, more new methods will continue to be proposed. We cannot presume that an ideal bandwidth allocation technique can consider every parameter. Some could be less complicated, some might have lower access delays, and others might have better bandwidth consumption. The MAC engine or the frame-level structures used for US and DS communication in any DBA/DBWA reflect the features and complexity of the PON that the algorithm offers. As a result, it is of prime concern to evaluate and understand the differences in features, limitations, and MAC engine designs along with architectural needs/progress in the present and latest PON variants/applications in use for 2.5 decades. We present the following contributions:

• First: This article explains the primary PON variants, their development up till now, architecture, and operating procedures. In continuation, particular emphasis is given to the working of the DBA MAC engines for various TDM-PONs that follow ITU-T and IEEE standards and end the section with the present architectural challenges.

• This article presents a critical review of the well-known DBA algorithms for the TDM PONs, reported in the literature in terms of the key strengths/weaknesses and their quantitative insights.

• We present the analysis on implementing TDM PON as an optical transport solution for 5G mobile fronthaul (MFH) network and present DBA requirements and challenges.

• We also look at the newest PON applications and give DBA algorithm designers goals and steps they can take to deal with future problems related to quality of experience (QoE), allocation of resources, and quality of service (QoS) for time-sensitive applications, supporting industry 4.0 and IoT-driven automation in these areas.

The rest of the article is organized as follows: Methodology presents the methodology, Architectural and Technological Advancements in PONs discusses the PON variants with the latest architectural developments and challenges, and how standard DBA algorithms have changed over the past 25 years under ITU-T and IEEE standards. Dynamic Bandwidth Allocation Algorithms of TDS-PONs thoroughly evaluates the reported DBA schemes for EPON, GPON, 10G-EPON, and NG-PON 2. Next, Evolving PON Applications and Key DBA Challenges presents the potential applications of PONs i.e., TDM PON-based 5G fronthaul network, FTTR, and Industrial PONs, and reports the challenges, opportunities, actionable steps, and targets concerning the DBA algorithms for each, while the Conclusion concludes the article.

APON/BPON-ITU-T Rec. G.983.x (1998/2001)

ATM PON (APON) is the first PON architecture to guarantee quality of service for both voice and data services at the same time (Kani & van Veen, 2020), as shown in Fig. 5.

The ATM frame, shown in Fig. 6, comprises a concatenation of 56 ATM cells of 53 bytes. The 1^st and 29^th ATM cells are Physical Layer Operation Administration and Management (PLOAM) cells used for cell separation and synchronization in the receiver and support transfer rates up to 1.5 Mb/s to 2.4 Gb/s. In the US, OLT receives the queue size information via PLOAM messages from ONUs and allocates the required bandwidth accordingly. In DS, ONUs receive the grant via PLOAM messages from OLT. The grants act as permissions for the ONUs to send user information to ATM cells; this can impact how bandwidth is allotted to ONU units. APON supported the transmission rate of 4.8 Mb/s in both US/DS for 32 end users, following FBA with a transfer rate of 155 Mb/s.

BPON, an improved version of APON, introduces the DBA concept under G.983.4 recommendations to benefit from the remaining or unused bandwidth of the users. The initial DBA mechanism uses three distinct techniques (Uzunidis et al., 2022):

• I.

  A status report (SR) only;

• II.

  The status report itself;

• III.

  A mixed report.

In the status report only, the OLT observes the traffic continuously and assesses this occurrence as a need to raise the bandwidth if the occupancy of queues rises. The ONU notifies the OLT of its status in the second scenario. The OLT receives a request from the ONU for greater bandwidth (when required). A mixed report combines elements of the two earlier categories. Table 1 compares the PON standards regarding the service, data rates, grant schedule, split ratio, link speed, and many other important factors for critically evaluating architectural and technological developments.

EPON-IEEE 802.3ah (2001)

In 2004, IEEE introduced Ethernet-PON (EPON), using Ethernet as a transport protocol and providing symmetric 1.25 Gb/s with variable-length Ethernet frames (packets) up to 1,518 bytes. Ethernet technology is a simple and low-cost solution (cheap fabrication costs) and has almost covered the network market. In EPON, OLT is connected to the multiple ONUs via a 1:N optical coupler or splitter, as shown in Fig. 7. So, each ONU buffers data obtained from the OLT in the DS. The round-trip time (RTT) between the OLT and each ONU also varies due to the differing link lengths between the OLT and each ONU. In the US, ONUs are unable to communicate with each other directly. Instead, each ONU can only transfer data to the OLT.

Consequently, in the DS, an EPON may be observed as a point-to-multipoint network and in the US as a multipoint-to-point network (Li, Lv & Bi, 2024). Different methods can be applied to avoid data collisions. For instance, wavelength division multiplexing (WDM) PON can be used. However, this approach is deemed costly because the OLT necessitates a dedicated receiver or an array of receivers to receive data across various wavelength channels, and each ONU must also be outfitted with a transceiver tailored to the wavelength. Table 1 presents the technical comparison of EPON with other standards.

Figures 8A and 8B show that the Ethernet frame is converted into an EPON frame by replacing its preamble with a logical link identifier (LLID) ahead of the frame header that comprises the destination address (DA) and source address (SA). Therefore, the MAC mechanism is required to prevent collisions and allocate bandwidth fairly among the shared connection capacity. Users must be in sync and aware of the permitted transmission time window. Each time slot may carry a constant or variable number of Ethernet frames. When the designated time slot for a certain user occurs, data (in the form of Ethernet frames) is delivered in bursts. The user transmits an empty frame if there are no contents. Customers can benefit from upgraded services that require flexible data rates that cannot be legitimately given under FBA, i.e., the PON can serve additional users using a DBA mechanism. It determines how long each user creates a queue to keep data. The OLT is responsible for gathering details on queue length and allocating timeslots to individual users. The approach promotes the effective use of the shared link capacity despite the higher transmission overhead. The multi-point control protocol (MPCP) is a signalling protocol at the MAC layer. It facilitates communication between the OLT and ONUs/ONTs for inter-ONU scheduling and service class support. The three functionalities of MPCP enable PON vendors to apply the preferred DBA system easily:

Discovery processing: PON users are registered and detected through a three-way handshaking process. In the process, the ONU generates the REGISTER REQ message, the OLT responds with the REGISTER message, and the ONU then echoes the registration parameters back to the OLT via the REGISTER ACK message.

MPCP report handling: ONUs report their status or bandwidth requests using the MPCP REPORT message, which contains information regarding the amount of data waiting to transmit a specific LLID.

MPCP gate handling: OLT polls ONUs for their queue status and grants bandwidth using the MPCP GATE message.

Interleaved polling with adaptive cycle time (IPACT) was the first EPON DBA mechanism to evaluate the time slot duration of each ONU. Several variations were also proposed (Logothetis et al., 2013), but overall, till this end, DBAs were not using the priority queues to offer differentiated services.

Gigabit PON (GPON): ITU-T Rec. G.984.x (2004)

The ever-increasing demands for bandwidth by users trigger the development of new PON standards. The GPON is an extended version of the BPON and uses the GPON encapsulation method (GEM) to support both variable-length and fixed-length forms of service, for instance, Ethernet and ATM. By using GEM frame fragmentation at the point of sending, GPON effectively lowers latency and facilitates real-time transmission. The GPON offers service providers the ability to retain conventional services (TDM-based voice, leased lines) without the need to replace customer peripheral equipment (CPE). This standard delivers $1.2$ and $2.4$ Gbps US/DS for a network distance of $20$ km. Table 1 presents the technical comparison with other standards.

GPON employs TDM and TDMA mechanisms in DS and the US, respectively. These are executed in the GPON Transmission Convergence (GTC) layer for frames with a constant frame duration of $125$ ms (Thangappan et al., 2020). To guarantee that the transmitting frame aligns with the designated time slot in the TDMA process, synchronization is necessary in the US due to the unequal distances between ONUs and OLT. To achieve this objective, the OLT establishes communication with each ONU and, taking into account the RTT, calculates a delay for each ONU in reverse order of the RTT. This ensures that the propagation delay between the OLT and ONUs is balanced. The DS_GTC frame, as shown in Fig. 9, consists of a physical control block downstream (PCBd) for the synchronisation and payload. Other fields in PCBd include:

• US bandwidth (BW) map-for start and end time information for each allocation interval.

• Physical synchronisation (Psync)-for frame synchronisation

• IDENT-for forward error correction

• Physical-Layer Operations, Administration, and Maintenance for DS (PLOAMd)-contains operation, administration, and maintenance commands.

• Bit-interleaved parity (BIP)-carries FEC parity for detecting error(s) in reception.

• Payload length downstream (PLend)-specifies the length of the preceding US BW map field.

On the other hand, the US_GTC frame consists of a burst of time slots called the transmission container (T-CONT). These are distinguished in the GPON using a specific identifier known as Alloc-IDs. A T-CONT solely comprises either ATM cells or GEM frames, but not both. Each ONU allocates traffic flows with identical attributes to one or more T-CONT types out of five possible T-CONT types, given in Table 2.

The ONU burst structure in Fig. 9 illustrates that the ONU relies on the Dynamic Bandwidth Report US (DBRu) to tell the OLT how much traffic is waiting to be sent in a certain Alloc-ID. Then, OLT checks the reported bandwidths against each Alloc ID to give grants automatically in the next allocation intervals based on the SLA and the DBA that was used (Abbas & Gregory, 2016).

10G-EPON-IEEE 803.2av (2009)

10G-EPON can support the vast number of consumers in metropolitan areas with broadband connectivity, maintaining backward compatibility with EPON. To make 1G E-PON and 10G E-PON work together again through TDM, the US wavelength region for 10G E-PON can be lined up with that of 1G E-PON. 10G-EPON offers 1 and 10 Gbps US and DS data speeds over 20 km. In addition to high reach, the low cost per gigabit per second of bandwidth and providing bandwidth-intensive services such as Triple Play Service, i.e., audio, video, and data, are other pluses. Figure 7 shows the architecture of the US/DS transmissions in 10G-EPON. Table 1 lists the most improved features over time.

In the same way that EPON does, 10G-EPON uses MPCP and the DBA algorithm to support QoS. At the link layer, 802.1Q divides network traffic into eight groups. Each category has a separate queue in each ONU with six priority levels for user traffic (Uzunidis et al., 2022). The system handles time and safety-critical traffic, inter-network control, voice, video, controlled load, excellent effort, best effort, and background traffic. The IEEE 802.3av task force focused on the physical layer, leaving the MAC protocol unchanged. The 10G-EPON MAC-layer control protocol includes enhancements for managing 10G-EPON FEC with Reed-Solomon coding and inter-burst overhead, but it results in up to 12.9% overhead (Laboratory, 2024).

XG-PON-ITU-T Rec. G.987.x (2010)

A total of 10 Gb/s GPON (XG-PON) was developed to accommodate emerging bandwidth-demanding applications, connect more users, and improve QoS and security. XG-PON delivers asymmetric transmission at 10 Gb/s in DS and approximately 2.5 Gb/s in the US. Like GPON, XG(S)PON supports ATM and Ethernet traffic via the XG-PON encapsulation method (XGEM). ONU registration and ranging mechanisms are handled through the DS and US PLOAM sections, and OLT can send several PLOAM messages (specified in the Hlend segment by the PLOAM count field) in a DS frame. Likewise, the bandwidth assignment (Grant-Request) method is incorporated in the DS frame of the OLT scheduler via the bandwidth map (BWmap) segment and the DBRu segment in the US frame. Figure 10 shows the architecture and DBA cycle of the XG-PON, whereas Table 1 compares the specifications with other standards.

NG-EPON-IEEE 803.2ca (2020)

As communication requirements and technology advance, the need for a new PON standard with better specifications arises. An increasing number of users require additional bandwidth for various tasks, such as wireless fronthaul. At the same time, there is a need to increase the number of users per fiber for commercial purposes. Another motivating factor is accomplishing these goals at a more affordable price than NG-PON2. A new EPON standard (NG-EPON) emerged in 2020 (Lam & Yin, 2020). This system would function at combined data rates of 100 Gb/s to offer increased data speeds for each user. Four wavelengths with a combined capacity of 25 GB/s can also be multiplexed to achieve this. The IEEE task group is now focused on developing three distinct generations of the access network, i.e., 25, 50, and 100 Gb/s. The architecture displaying the US/DS transmissions in NG-EPON is shown in Fig. 7, and the comparative advancements of the features are given in Table 1.

The Data Link Layer’s multi-point MAC control (MPMC) sublayer is responsible for adding the NG-EPON to the Ethernet architecture and ensuring it stays compatible with older IEEE standards. It includes protocols for transmission resource allocation, ONU discovery and registration, and DBA. Dynamic wavelength channel bonding is used to achieve high data rates. The Multi-Channel Reconciliation Sublayer (MCRS) facilitates data transmission over numerous signals across multiple MCRS channels. The fundamental transmission unit is the envelope quantum (EQ), which comprises 72 bits. Frames can be categorized as either complete or fragmented. The transmission of a signal involves the interleaving of frames across multiple MCRS channels to obtain higher transmission speeds. The transceiver class utilizes 25 Gb/s avalanche photodiodes (APD) for receivers, employing NRZ modulation. The optical modulation amplitude (OMA) minus penalty approach is used for transmitters. Currently, Optical Network Units (ONUs) require burst-mode laser diodes equipped with a rapid on-off switch and a control loop to regulate the laser bias current (Uzunidis et al., 2022). The NG-PON2 transmission convergence (TC) layer allows for operations with more than one wavelength and wavelength channel mobility, which means it can switch from one TWDM source channel to another target channel.

NG-PON2-ITU-T Rec. G.989 (2015)

NGPON2 is a hybrid approach based on time and wavelength division multiplexing (TWDM) techniques. The ITU-T and FSAN jointly developed this hybrid PON standard, providing over 40 Gbps to meet such relentless growth in traffic demand with low-cost TDM equipment. The NG-PON2 can easily co-exist with the existing ODN infrastructure of the previous generation of PON, such as the EPON, GPON, and XG(S)-PON, attracting the telecom operators for quick deployment in 2017. Moreover, the PON industry has also suggested NGPON2 technology by using the TWDM approach, not only to support fixed broadband user data but also to provide 5G wireless service. Therefore, NGPON2 is considered the ideal technology for connecting the fronthaul/backhaul link of the small cell architecture of 5G.

Figure 11 presents the architecture of TWDM-PON, wherein an OLT has optical subscriber units (OSUs), a wavelength multiplexer (to keep multiple wavelengths in a single ODN featuring tunable transceivers), a passive splitter, and several ONUs. Table 1 presents the comparison with other PON standards in terms of the specifications. The tunable transceivers offer high flexibility and reconfigurability, allowing them to communicate with any OLT transceiver at any wavelength. This colorless design simplifies ONU installation and management (Memon et al., 2020). Under the MAC layer, this standard provides two channels, i.e., TWDA channels (TDM/TDMA access—P2MP) and WDM channels (P2P access), i.e., development of the dynamic bandwidth and wavelength algorithms (DBWA) (ITU-T Recommendations, 2021).

This is possible with OLT channel terminations (OLT CTs) communicating via the inter-channel termination protocol (ICTP). Referring to Figs. S1 and S2 (ITU-T Recommendations, 2021), it can be seen that the TWDM transmission convergence (TC) layer is made up of three sublayers: the service adaptation (SAS) layer, the framing (FS) layer, and the PHY adaptation (PS) layer.

Every ONU begins a US PHY frame by adjusting the starting position of the DS frame. The OLT CT and all ONUs on the PON share a common timing reference. In the US, each ONU transmits short PHY bursts with PSBu and payload (as Framing Sublayer), remaining silent between them. The OLT CT uses guard time (TX enable/disable times, 64-bit) and BWmap to control the timing and duration of these bursts to avoid overlap, collision, or jamming in US transmissions of different ONUs. The SAS uses XGEM for service data unit (SDU) encapsulation, representing its protocol data unit (PDU) as an XGEM frame. The XGEM header consists of an eight-byte encrypted field that specifies the logical connection or traffic pattern, whereas the XGEM payload might include a full TC layer SDU. In TWDM-PON, US bandwidth is assigned using a BWmap, transmitted every 125 $\mu$s and containing allocation structures (AS) for a particular ONU. Alloc-ID serves as the grant recipient’s identification, and T-CONTs are in charge of managing it. A 1-bit forced wakeup indication (FWI) flag is used for energy-efficient scheme operations. The MAC layer uses an extra auxiliary management and control channel (AMCC) for dynamic wavelength assignment, reusing the TWDM TC’s three layers. FEC is subject to ON/OFF control for TWDM channels and is integral to PtP WDM channels. NG-PON2 uses the PLOAM messaging channel like XG-PON, with additional features like wavelength handover, protection, and power control.

Recent advancements

ITU-T released the most recent PON specifications under the Higher Speed PON (HS-PON) project, which defines a 50G-PON system with a DS of 50 G and a US of 25 G. In February 2023, two revisions to the HS-PON standards were officially authorized, signifying that the system specifications have reached a level of maturity suitable for commercial development and implementation by the telecom operators at the beginning of 2025 (ITU-T Recommendations, 2023). HS-PON employs electronic equalization to counteract chromatic dispersion (CD) and utilizes the O-band, the region with the least dispersion. Nevertheless, the challenge is observing forthcoming transmitter advancements due to the high signaling speeds above 50 Gb/s.

Latest architectural challenges

ITU-T began a project on very high-speed PON (VHSP) systems to achieve above 50 Gb/s per wavelength. This project intends to collect the VHSP system requirements, characteristics, and prospective technologies and pave the way for the standardization of future PON systems after the successful implementation of 50G-PON (ITU-T, 2024b). IM-DD, which utilizes power budget improvement technologies to increase performance, and coherent technologies (CTs), which use coherence detection to address fiber dispersion, are two candidate technologies that are now the subject of considerable debate over their potential applications.

• The most significant obstacle that IM-DD must overcome to achieve high-speed transmission is fiber dispersion, which results in the deterioration of optical signals and a decrease in receiver sensitivity. At 100 Gb/s, the receiver’s sensitivity would drop from −27 dBm at 50 Gb/s (Rosales et al., 2022) to −20 dBm (Yi et al., 2019).

• CTs can use DSPs to mitigate these degradations but at the cost of increased system complexity, higher costs, and power consumption (Rizzelli, Torres-Ferrera & Gaudino, 2023). Companies such as British Telecom, Huawei, ZTE, and others are now participating as members and providing support.

DBA without differentiated QoS support

Luo & Ansari (2005) in proposed the DBA algorithm based on the volume of each ONU buffer or request. After the ONUs request, the OLT plans the time slot in which ONUs can communicate several bytes. The authors considered the tree architecture in the proposed setup, where the three ONUs are connected to the single power splitter. The interleaved polling with adaptive cycle time (IPACT) complete operation steps are shown in Fig. 13 (Zheng & Mouftah, 2009).

The algorithm’s first step is to know the waiting number of bytes in every ONU and round trip time (RTT) and the total number of bytes in ONUs, where a polling table is stored at OLT. The OLT also sends a GRANT message to ONU1 at some time T1 to start the 3,200-byte transmission. In the next step, after receiving the GRANT message from OLT, ONU1 will start uploading its data; hence, the ONU has to continue buffering and receiving packets from end subscribers, so ONU1 will produce a REQUEST noting how many bytes are kept in its buffer. After that, OLT keeps tracking the REQUEST data, and after its arrival, it updates the polling table and calculates the succeeding time slot. After that, 3200 bytes were changed to 1,200 at time T1.

Now, at time T2, the ONU2 will start data transmission, sending and receiving the frame from the OLT after receiving the GRANT note, so from the above explanation, it is understood how IPACT repeats the same process to ONU3 at time T3. According to the explanations and the steps above, we can conclude that the DBA algorithm will be defined as following (Zheng & Mouftah, 2009).

(1) $$T_k^{i+1}=max\{\begin{array}{c}RT{T}_i+T_k^i+T_{guard}+\frac{W_{grant}}{R},\hfill \\ -RT{T}_{i+1}{T}_{i+1}^{k-1}+RT{T}_{i+1}\hfill \end{array}$$where $T_k^i$ represents the start time of ${\mathrm{ONU}}_{i+1}$ transmission in the $k$-th cycle, $W_{grant}$ represents the grant window for each ONU, R is the transmission bit rate (bits/sec), $T_{guard}$ is the guard time between transmissions and $RT{T}_{i+1}{T}_{i+1}^{k-1}$ represents a correction factor.

Bandwidth guaranteed polling algorithm

There are many deficiencies in IPACT. It cannot make differences between numerous classes of services, and it does not support priority scheduling schemes because it follows first-in-first-out transmission. A new algorithm has been proposed to overcome these issues, which joins the novel IPACT restricted-service system named inter-ONU scheduling and priority queuing scheme named as (intra-ONU scheduling (Ma, Zhu & Cheng, 2003)). They are supposed to support delays in varied sensitive services. Dynamic Bandwidth allocation can be considered in the internal and external ONU scheduling as shown in Fig. 14. The OLT always deals with high-priority messages first, conferring strict priority scheduling so that the OLT ignores the QoS guarantee of the low priority, leading to increased packet delay and packet drop rate (Luo & Ansari, 2005).

To sum up, the benefit of this method is that the downlink channel grants the bandwidth to the ONUs as per the service level agreement. On the other hand, upstream link transmission has one downside: it grants static bandwidth that reduces the bandwidth and the throughput of uplink transmission.

Efficient DBA

Based on the MPCP mechanism, this algorithm assigned the extra bandwidth of lightly loaded ONU to heavily loaded ONU. In this way, it significantly improved the network’s performance, minimised the packet delay and length of the queue, and increased the throughput of heavily loaded traffic. To enhance bandwidth efficiency during periods of heavy network traffic, an operational scheduling control system is implemented to identify and report instances of idle time (Zheng, 2006).

IPACT with grant estimation DBA algorithm

This algorithm is a development of IPACT. It guesses the recently arrived packets between two successive polling and granting ONUs sideways with the further assessed amount. In small traffic situations, the data packets communicate the consecutive cycle, thus dropping the waiting delay (Zheng, 2006).

Two-layer bandwidth allocation DBA algorithm

This algorithm works in a double layer, as the name suggests. It divides the transmission into two parts: allocate bandwidth to upstream works in the primary layer and, in second, allocate the bandwidth to the whole ONU. Below the high traffic load, it assures the lowest bandwidth (Choi & Park, 2010). The comparison of all DBAs in this subsection is given in Table 3.

GIANT DBA algorithm

The first DBA algorithm for GPON is Giga PON access network (GIANT). While distributing the bandwidth as per TCONT priority, the scheduling is done round-robin. When bandwidth is distributed to a queue, it only uses a down counter for every frame duration and is down by one for each frame duration. Upon expiration, the increased idle time generates bandwidth allocation for each service class, while some unused bandwidth remains (Leligou et al., 2006).

Immediate allocation with colorless grant DBA algorithm

This algorithm is proposed to improve GIANT by minimising idle time and utilizing the unallocated bandwidth. In each ONU, a Byte counter (VB) is issued for immediate bandwidth allocation. When the down counter expires, VB needs to be recharged again by expiry; unallocated bandwidth also gets wasted, and ONU works through TCONT5 (Han et al., 2008).

GPON redundancy eraser algorithm for long reach DBA

This algorithm overcomes long-distance networks’ issues if the sent report is subtracted from the previous grant and leads to a real updated report form. At loads of lower traffic, bandwidth wastage is greater because of additional requests of high-priority users (Sales, Segarra & Prat, 2014). The comparison of all DBAs in this subsection is given in Table 4.

Challenges at this stage: GPON and EPON were successful, but they had major issues that prompted XG-PON and 10G-EPON. GPON’s 2.5/1.25 Gbps bandwidth and EPON’s 1 Gbps symmetric bandwidth were insufficient for 4K video, cloud services, and AR/VR. Static bandwidth allocation and ineffective DBA methods caused real-time service latency. Scalability issues hindered both protocols, and limited ONU support restricted urban installations. The lack of sophisticated QoS differentiation makes prioritizing latency-sensitive or high-priority traffic challenging.

Advanced sleep-aware (ASDBA)

This algorithm exploits ONU’s energy efficacy and DS and US scheduling, both done at the same transmission slot with its sleep mode. The ASDBA modifies the typical sequence of control message exchange used in the 10G-EPON system to adjust the time an ONU waits between sending a REPORT message and responding to a GATE message by converting it into the ONU’s sleep time transmission. This feature allows the ONU to enter a state of continuous sleep after transmitting a REPORT message, remaining inactive until the start of the next transmission slot. This enhances energy efficiency in the ONU but also leads to delays in frame queuing (Van et al., 2014).

DBA-gated algorithm

This is the first algorithm projected. It works with a very simple scheduling method. It grants the window size according to the demand of ONU without any composite calculation. It grants at first request maximum according to the demand of 50,000 bytes, and if asked for further bytes, ONU must resend the request of GRANT the window (Rout, 2016a).

DBA-linear algorithm

As the name suggests, this algorithm grants window size to the user with a linear factor more than the demand; considering that may be due to delay, more bytes could be used. This shows that this algorithm is proportional to the requesting and grant window sizes. The window is assigned to ONU until the full data is transferred. In this regard, this algorithm offers good QoS (Rout, 2016a).

DBA-MAX algorithm

In this algorithm, a threshold level is fixed for the GRANT window size for every user; if any user needs more windows than the threshold, it provides more to that user. Other than that, a fixed threshold value is assigned to every user. The advantage of this scheme is that other users will block no OLT, but the drawback is that GRANT window size is independent of REQUEST size, generating too much overhead (Rout, 2016a).

TD-optimum algorithm

This algorithm eliminates the downside of the above DBA GATE algorithm; if the throughput and delay of the system vary, this algorithm can monitor the performance of the network well. It supports delay-sensitive traffic like voice and video over IP and non-delay sensitive traffic like massive data and video transfer applications, so it is the triple play. So, it did not drop data, voice, and video packets and communicate well (Rout, 2016b).

Sub-MOS-IPACT DBA algorithm

At different granularities, this scheme promises bandwidth, individual ONUs, multi-ONU clients, and subgroups of ONUs. A Sub-group is a set of ONUs belonging to the same network. Customer multi-ONU bandwidth was also prioritized by sub-MOS-IPACT distribution of the same multi-ONU client among subgroups. The simulation results indicate that the proposed mechanism enhances the efficiency of the network (Ciceri, Astudillo & da Fonseca, 2019). The comparison of all DBAs in this subsection is given in Table 5.

Immediate allocation with reallocation DBA algorithm

In 2012, the immediate allocation with reallocation (IAR) algorithm was introduced to overcome the unused bandwidth issue. In this algorithm, polling is introduced iteratively. Scheduling is achieved in four steps: First, from GPA to TCONT2 and TCONT3. Second, scheduling will repeat for overloaded queues for TCONT2 and TCONT3 so that under-load queues get bandwidth. Third, SPA to TCONT3 and TCONT4. Finally, TCONT5 gets allocation from CG (Han, 2012).

Efficient bandwidth utilization

As repeating the scheduling process results in a delay to some level, this algorithm proposed that if we use the Borrow and Refund (BR) concept while updating the operations, this issue could be resolved. If the request is more than VB, it gets negative results in over-allocation. The unused bandwidth will compensate for this over-allocation in the previous queue. But this method makes low priority class get fewer assignments (Han, Yoo & Lee, 2013).

Simple and feasible (SF) DBA algorithm

This algorithm proposed a simple method to utilize the unused bandwidth compared to the BR operation. The author proposed a service class using a down counter and a single available 448-byte counter and allocated the bandwidth at request to every queue. By this method, each service class ignores the unused bandwidth, but the limitation is one service class; it cannot use the bandwidth of other service classes to avoid depletion (Han, 2013).

DBA with high utilization (DBAHU) algorithm

The updated version of SFDBA is DBAHU. This algorithm used BR in the updated operation of SFDBA. This can fix the problems in the above algorithm. But it will disturb the next cycle if there is no unused bandwidth. As round-trip time (RTT) is large in long-distance PON, this OLT will receive an extra report and waste the bandwidth (Han, 2014).

Improved bandwidth utilization DBA algorithm

It improves the scheduling and polling mechanism by assigning a slot three times to DBRu when the counter range is down to 2, 5, and 8. When it expires, TCONT4 starts execution. Because of the high priority of TCONT3 and TCONT2, it mostly focused on them. But the issue is that as TCON4 is executed after the down counter expires, this quickly increases the delay and ignores the unused bandwidth (Butt et al., 2017).

IRIS DBA algorithm

After the execution of SPA and GPA, IRIS will stand by for the bandwidth phase. Standby bandwidth starts from the finish of the last allocation and flinches of the fresh grant cycle. It performs better in TCONT4 and TCONT3 than DBAHU. IRIS later used the standby bandwidth by predicting the queue size and allocating its percentage to the queue. It used only 8 ONUs to show the results (Kyriakopoulos & Papadimitriou, 2016).

Comprehensive bandwidth assignment scheme for long reach DBA algorithm

Comprehensive bandwidth assignment scheme for long reach (CBA-LR) enhances the scheduling and polling mechanisms used in algorithms such as GREAL, EBU, and IACG by efficiently assigning residual bandwidth to the appropriate traffic class. The residual bandwidth can be utilised more effectively by eliminating the BR concept in the EBU. However, this comes at the cost of increased delay for traffic classes T2 and T3 compared to the EBU algorithm (Butt et al., 2019).

Comprehensive bandwidth utilization DBA algorithm

It is better to overcome the downsides in polling and scheduling IACG and EBU by allocating the whole bandwidth by ONU and OLT to TCONT and avoiding bandwidth wastage. BR concept is not used such that extra bandwidth will be moved to TCONT solitary (Butt et al., 2018b).

Sleep assistive DBA algorithm

It works over ONU on its sleep mode cycle to save energy because polling only happens on ONU during SI in odd cycles. The GPA is according to SLA limits for T4, T3, and T2, but it is assigned to active state ONUs when near RBW (Butt et al., 2018a).

Demand forecasting (DFDBA) algorithm

Using statistical modeling, forecast the demand in future stores’ last 100 demand cycles using a circular buffer. By this, a significant amount of delay gets reduced (Memon et al., 2019). The comparison of all DBAs in this subsection is given in Table 6.

Challenges at this stage: Comparing the different versions of DBAs from both standards, it is observed that the fixed designs and restricted wavelength support were unsuitable for quickly expanding traffic from 4K streaming and cloud services, limiting bandwidth scalability. Asymmetric upstream bandwidth limits real-time gaming, video conferencing, and 5G fronthaul. Latency and QoS control added to the challenges. Emerging applications like AR/VR and mission-critical 5G services need minimal latency and jitter, but DBA techniques were too inflexible. Fairness and resource allocation under bursty traffic remained unsolved, especially in different service settings. Energy efficiency was another issue. As the number of connected devices multiplied, both standards’ high power consumption, especially at ONUs, became unsustainable. Scalability and flexibility issues prevented them from adapting to different traffic patterns and multi-service networks, which are essential for digital infrastructure.

Fair DWBA algorithm

Hussain, Hu & Li (2017) proposed a fair algorithm to reduce the re-sequencing problem of frames that happened when a single grant ONU transmits data on a single wavelength. Controlling this issue increased the fairness of scheduling. It improved bandwidth usage efficiency by assigning the allocated bandwidth, thus minimising bandwidth wastage. In general, this results in a minimal packet delay and loss ratio for all levels of network traffic. The comparison of all DBAs in this subsection is given in Table 7.

DWBA algorithm

In addition to the fair algorithm, this first-fit (FF) algorithm also minimises the transmission latency along with the mitigation of the re-sequencing problem, so by increasing the efficiency of bandwidth by decreasing the disproportionate use of guard time, it also reduced the wastage on a single wavelength, just like fair scheme do in results, giving low packet delay (Hussain et al., 2017).

Agile wavelength (AW) DWBA algorithm

This scheme schedules the ONU in ascending order of their RTT and tries to grant ONU on a single wavelength channel to avoid the guard time from waste of excessive bandwidth. This also adjusts some ONU on multiple wavelength channels and ensures the transmission is finished simultaneously. This scenario maximised the upstream bandwidth of channels, which is the best-proposed scenario for better utilization of bandwidths. This also lowered the average delay and satisfied the high-load ONUs (Wu, Wang & Guo, 2018).

Flexible wavelength DBA algorithm

This scheme reduces the transmission delay. It grants bandwidth to ONU according to load traffic and is flexible in scheduling. A high traffic load assigns more bandwidth to those ONUs and vice versa. Bandwidth allocation also depends on the number of total ONUs in the network and the mitigation of re-sequencing delay for light-loaded ONUs. It also does this for heavy-load ONUs. It also introduced a parameter alpha by which flexible bandwidths are decided. It also depends on the number of ONUs in a network (Hussain et al., 2018).

Multi-type users and multi-type services DBA algorithm

This algorithm is proposed to facilitate residential and business users at the same OLT and supports QoS parameters. It comprises numerous sub-algorithms like WoF and ToF grant sizing and scheduling to resolve the sub-problems of the transceiver ONUs DBA synthetically. Firstly, a two-dimensional priority multi-type users and multi-type services (MUMS) queue model is designed to determine grant scheduling among MUMS. Later, grant sizing is introduced, and further ToF and WoF too. It also reduced the average packet delay and packet loss ratio as well as the jitter delay (Yan et al., 2018).

Quality of service DWBA algorithm

This scheme can fulfil the demands of network users in the future. This QoS-based algorithm handles LP (low priority) and HP (high priority) traffic simultaneously. It minimises the delay that ONU faces in the online manner. The HP traffic case for LP works offline because it does not need a constant bit rate and can bear delay in data transmission (Rafiq & Hayat, 2019).

Single channel as possible DWBA algorithm

The single channel as possible (SCAP) algorithm reduces frame reordering issues, which occur simultaneously with parallel transmissions. If among multiple wavelengths, the grant distributes without considering grant size, and the ONU transmits data among all channels, resulting in frame reordering. This algorithm controls this issue by granting bandwidth in single channels. Along with this, packet delay also improved (Wang, Guo & Hu, 2018).

First in first out DBA algorithm

The network capacity in the NG-EPON standard increased by 100 GB/s, and strict requirements must be applied to delay packets. New scheduling criteria were introduced, and the grant assigned upstream is now CDMA-based on assuring the meetup of the new required network throughput. This algorithm improved network capacity by reducing the mean waiting time for received packets and jitter delay. In first in first out (FIFO), ONU ordered the request according to the arrival time of requests (Raad, Inaty & Maier, 2019).

Reservation pattern DBA algorithm

This is also used for CDMA-based NG-EPON networks. It differed from FIFO in the scheduling sense that ONU transmits the queue according to the predefined reserved pattern. Its objective is the same as that of FIFO (Raad, Inaty & Maier, 2019).

Deep learning DBA

This DBA analyzes previous use patterns and real-time network metrics to estimate ONU traffic demands using deep neural networks. The model allocates bandwidth based on expected demand, guaranteeing effective resource distribution. Training the system on big datasets allows it to dynamically adapt to changing traffic scenarios, decreasing packet delays and optimizing network resources (Hatem, Dhaini & Elbassuoni, 2019).

Reinforcemnt leaning based DBA

This DBA uses reinforcement learning to dynamically adjust bandwidth allocations by interacting with the network environment. It learns optimal strategies based on feedback, ensuring effective bandwidth distribution in unpredictable situations and continuous improvement through experimentation (Zhou et al., 2020).

Mobile CPRI predictive (MCP)-DWBA algorithm

It predicts how many time slots ONU will need in the future, after which there will be no need for grant messages. In this way, it saves the propagation time spent on grant messages so that time can now be used to transmit data. Its objective is the same as that of the MC-DWDBA algorithm. The MCPDWBA method optimizes the utilization of time slots, decreases the number of active channels, and increases the pace at which data is transmitted, hence facilitating the deployment of NG-EPON as the MF networks (Araujo et al., 2019).

Efficient bandwidth management DWDBA algorithm

It is designed to bring about expanded traffic supplies for changed ONUs. EBMA routine is estimated moderately with adapted first fit (FF-DWBA) and IPACT (M-IPACT) algorithms. It is better in packet delay and drop ratio; it segregates both low-loaded (LL) and heavy-loaded (HL) users (Rafiq & Hayat, 2020). The comparison of all DBAs in this subsection is given in Table 7.

DBA algorithms for 5G TDMA PON based fronthaul traffic and challenges

The 5G fronthaul network can support many types of traffic, notably high-priority traffic for real-time services like audio and video and lower-priority traffic for data exchanges. NG-MFH networks need to handle many different services and types of traffic. This makes it even more important to have a DBA that can handle traffic demand, quality of service needs, and service level agreements (SLAs) for 5G customers. Implementing PON technology for MFH solutions and resource sharing in convergent networks poses challenges. TDM involves sharing US bandwidth, leading to contention for uplink capacity and collisions in the RAN payload. Allocating the appropriate bandwidth and determining the transmission timing is critical in the US to prevent congestion and fulfil latency requirements. RAN scheduling involves the user equipment (UE) requesting transmission by submitting a scheduling request (SR) to the centralized/distributed unit (CU/DU). The CU/DU subsequently plans and allocates radio resources by returning a scheduling grant (SG) to the UE. In Ethernet PONs, the standard baseline DBA (Kramer, Mukherjee & Pesavento, 2002b) uses report and grant messages to ask for and give bandwidth to the MFH, which causes poor latency. The extra delay is shown in the timing diagram in Fig. 16A. The cooperative DBA approaches utilize fronthaul traffic data via a cooperative transport interface message to tackle this problem (Tashiro et al., 2014). The CU/DU collaborates with the OLT by sharing previous knowledge about UE traffic, which is determined by analyzing how packets arrive, i.e., traffic patterns at the ONU-RU. If one looks at CPRI traffic with a fixed bit rate and packet size, as well as the patterns of packet arrival at ONU-RUs or the inter-packet gaps (IPGs) in MFH transmissions, one can find out how much bandwidth each ONU-RU is given. This is called $T_{grant}$. The interval between two successive transmissions by an ONU RU is known as a grant cycle, denoted as $T_{cycle}$. By coordinating the transfer of $T_{grant}$ with SG, the grant-report procedure of the standard baseline DBA can be removed, as shown in Fig. 16B.

Wong & Ruan (2023) and Zhang et al. (2018) presented ML-driven and Cell-Average Traffic (CAT)-based DBA techniques to enhance bandwidth allocation and maintain QoS in MFH networks. The former DBA made a fast and self-adaptive bandwidth (FSA) decision in response to traffic patterns, network loads, and line rates. This enhanced the DBA scheme’s ability to handle various applications in future MFH networks. The latter DBWA approach uses CAT to guarantee QoS for different types of cells in MFH. The work of Lagkas et al. (2021) described a complete method for allocating radio, optical, and MEC resources together. This method uses advanced technologies to make resource allocation dynamic and energy-efficient. Kalfas et al. (2019) discussed the unique constraints of 5G mmWave networks and suggested a fiber wireless fronthaul design with a DBA protocol to address capacity and latency needs. All these studies emphasize the significance of DBA in NG-MFH networks and the ability of several methods to meet this requirement.

In industries that host DOCSIS, PON, and 5G systems, a converged environment needs unique and interoperable (AI/ML) DBA design solutions (Poletti, 2024), even though it reduces uplink traffic (Garima, Jha & Singh, 2024) and gives too much fronthaul support (Feng et al., 2023). Keeping AI/ML developments in focus for the two different optical and wireless domains and self-optimizing networks that adapt to link access and network requirements can increase the efficiency of resource sharing. It is expected that 6G would use SDN’s current enabling technologies to improve networks by attempting to render them more modular, accessible, and self-driving. Table 8 presents the challenges and opportunities in DBAs for 5G/6G fronthaul networks.