Edge-enhanced graph neural network for DU-CU placement and lightpath provision in X-Haul networks

Ruikun Wang; Jiawei Zhang; Zhiqun Gu; Shuangyi Yan; Yuming Xiao; Yuefeng Ji; Yuefeng Ji

doi:10.1364/JOCN.465369

Journal of Optical Communications and Networking
Vol. 14,
Issue 10,
pp. 828-839
(2022)
•https://doi.org/10.1364/JOCN.465369

Edge-enhanced graph neural network for DU-CU placement and lightpath provision in X-Haul networks

Ruikun Wang, Jiawei Zhang, Zhiqun Gu, Shuangyi Yan, Yuming Xiao, and Yuefeng Ji

Not Accessible

Your library or personal account may give you access

Get PDF
Email
Share
Get Citation
Copy Citation Text
Ruikun Wang, Jiawei Zhang, Zhiqun Gu, Shuangyi Yan, Yuming Xiao, and Yuefeng Ji, "Edge-enhanced graph neural network for DU-CU placement and lightpath provision in X-Haul networks," J. Opt. Commun. Netw. 14, 828-839 (2022)

Export Citation
- BibTex
- Endnote (RIS)
- HTML
- Plain Text
Citation alert
Save article

Abstract

The emerging mobile services have imposed several new challenges on the radio access network (RAN), which stimulates its evolution from cloud-RAN to next-generation RAN (NG-RAN). In NG-RAN, baseband functions are redefined as the radio unit (RU), distributed unit (DU), and central unit (CU), which are, respectively, connected by optical front/mid/backhaul (X-Haul) networks. The DU-CU placement and lightpath provision need an elaborate strategy for allocating resources tailored to user requests. However, most existing methods fail to fully perceive network conditions and then make inappropriate solutions, which may result in over-consumption of processing and transmission resources. Therefore, we propose a reinforcement-learning-based DU-CU placement and lightpath provision strategy using an edge-enhanced graph neural network, i.e, EGNN-RL. The EGNN is leveraged to adequately exploit graph-structured features in X-Haul networks, while proximal-policy-optimization-based RL is introduced to maintain policy stability. In addition, this problem is formulated as an integer linear programming model to find the optimal solution. We validate the proposed strategy under different service types (i.e., enhanced mobile broadband, ultra-reliable low latency communication, and massive machine connections) and different numbers of services, respectively. The results show that our scheme can achieve higher resource efficiency compared with existing methods. Moreover, it also adapts to new networks with the same nodes through fine-tuning the model. This can decrease nearly 15% of the retaining time and obtain similar performance with retaining.

Full Article | PDF Article

More Like This

Security-aware 5G RAN slice mapping with tiered isolation in physical-layer secured metro-aggregation elastic optical networks using heuristic-assisted DRL

Yunwu Wang, Min Zhu, Jiahua Gu, Xiang Liu, Weidong Tong, Bingchang Hua, Mingzheng Lei, Yuancheng Cai, and Jiao Zhang
J. Opt. Commun. Netw. 15(12) 969-984 (2023)

Latency-Aware CU Placement/Handover in Dynamic WDM Access-Aggregation Networks

Francesco Musumeci, Omran Ayoub, Monica Magoni, and Massimo Tornatore
J. Opt. Commun. Netw. 11(4) B71-B82 (2019)

Techniques for applying reinforcement learning to routing and wavelength assignment problems in optical fiber communication networks

Josh W. Nevin, Sam Nallaperuma, Nikita A. Shevchenko, Zacharaya Shabka, Georgios Zervas, and Seb J. Savory
J. Opt. Commun. Netw. 14(9) 733-748 (2022)

Previous Article Next Article

Cited By

You do not have subscription access to this journal. Cited by links are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Figures (9)

You do not have subscription access to this journal. Figure files are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Tables (4)

You do not have subscription access to this journal. Article tables are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Equations (25)

You do not have subscription access to this journal. Equations are available to subscribers only. You may subscribe either as an Optica member, or as an authorized user of your institution.

Contact your librarian or system administrator
or
Login to access Optica Member Subscription

Constant	Description
$R, D, U$	Set of PPs, 5GC, and network nodes
$WS, PS$	Set of wavelengths, and GPPs
$S$	Set of service requests
$J$	Set of service segments (i.e., $j = 1$ for RU, $j = 2$ for DU, $j = 3$ for CU, $j = 4$ for UPF)
$L_{s, r}$	Binary, 1 if AAU of placing RU of request $s$ is directly connected to node $r$
$V_{r}$	Binary, 1 if 5GC is placed at node $r$
$M^{m, n}$	Binary, 1 if node $m$ is connected to node $n$ , $m, n \in U$
$sp$	Length of RAN service chaining (i.e., 4)
$T^{m, n}$	Propagation latency of link $(m, n)$ , $m, n \in U$
$T_{o e o}$	Latency of OEO and electrical switching (i.e., 20 µs)
$C_{r}$	Processing capacity of each GPP $r$ , $r \in PS$
$C_{w}$	Bandwidth capacity of each wavelength $w$ , $w \in WS$
$A_{s}$	Access latency of service request $s$ between AAU and its directly connected PP
$B_{s}$	Fronthaul latency demand of request $s, s \in S$
$T_{s}$	Total latency demand of request $s, s \in S$
$Ψ_{s, j}$	Processing demand of segment $j$ of request $s$
$Ω_{s, j}$	Bandwidth demand of requests $s$ after segment $j$ being processed $(j < sp)$
$Δ$	A large positive integer

Variable	Description
$Q_{r}$	Binary, 1 if node $r$ is activated
$G_{r, p}$	Binary, 1 if GPP $p$ of PP $r$ is activated
$O_{s, j, r, p}$	Binary, 1 if segment $j$ of service request $s$ is placed in GPP $p$ of node $r$
$Y_{s, r}$	Binary, 1 if service request $s$ experiences OEO and electrical switching in PP $r$
$Z_{s, j}^{m, n, w}$	Binary, 1 if service request $s$ is carried on wavelength $w$ of link $(m, n)$ after segment $j$ being processed $(j < sp)$
$H_{m, n, w}$	Binary, 1 if wavelength $w$ of link $(m, n)$ is occupied

Service Types	Wireless RBs	Fronthaul Latency (µs)	Total Latency (µs)
eMBB	450	100	1000
uRLLC	50	50	450
mMTC	150	250	1500

Parameter	Value
Number of layers in EGNN	2
Number of neurons in each layer of EGNN	20
Number of layers in FCNN	2
Number of neurons in each layer of FCNN	256 and 128
Size of action space	158 or 120
Number of the shortest paths $\hat{K}$ in KSP	5
Learning rate of new actor network	0.001
Learning rate of critic network	0.003
Discounted factor $μ$ for long-term return	0.9
Penalty $r_{max}$ of DU-CU placement failure	4

Constant	Description
$R, D, U$	Set of PPs, 5GC, and network nodes
$WS, PS$	Set of wavelengths, and GPPs
$S$	Set of service requests
$J$	Set of service segments (i.e., $j = 1$ for RU, $j = 2$ for DU, $j = 3$ for CU, $j = 4$ for UPF)
$L_{s, r}$	Binary, 1 if AAU of placing RU of request $s$ is directly connected to node $r$
$V_{r}$	Binary, 1 if 5GC is placed at node $r$
$M^{m, n}$	Binary, 1 if node $m$ is connected to node $n$ , $m, n \in U$
$sp$	Length of RAN service chaining (i.e., 4)
$T^{m, n}$	Propagation latency of link $(m, n)$ , $m, n \in U$
$T_{o e o}$	Latency of OEO and electrical switching (i.e., 20 µs)
$C_{r}$	Processing capacity of each GPP $r$ , $r \in PS$
$C_{w}$	Bandwidth capacity of each wavelength $w$ , $w \in WS$
$A_{s}$	Access latency of service request $s$ between AAU and its directly connected PP
$B_{s}$	Fronthaul latency demand of request $s, s \in S$
$T_{s}$	Total latency demand of request $s, s \in S$
$Ψ_{s, j}$	Processing demand of segment $j$ of request $s$
$Ω_{s, j}$	Bandwidth demand of requests $s$ after segment $j$ being processed $(j < sp)$
$Δ$	A large positive integer

Abstract

Cited By

Figures (9)

Tables (4)

Equations (25)

Journal of Optical Communications and Networking