Energy-Efficient FSO Reconfiguration under User Mobility in Hybrid Fiber-IAB Backhaul 1 , Charitha Madapatha , Carlos Natalino 1 , Tommy Svensson 1 , Paolo Monti 1 Department of Electrical Engineering, Chalmers University of Technology, Gothenburg, Sweden (*) [email protected]
Piotr Lechowicz
*,1
Abstract User mobility creates stochastic, time-varying backhaul demand that static capacity provisioning cannot match. We propose a closed-loop, load-aware hysteresis controller for hybrid fiber-IAB-FSO backhaul and show that energy drops faster than coverage: 8% to 44% energy savings cost only 0.9% to 6.7% coverage. © 2026 The Author(s) This is the authors’ version of this publication.
arXiv:2606.18017v1 [cs.NI] 16 Jun 2026
Introduction Future 6G networks will require massive densification of low-power Small Base Stations (SBSs) for high data rates and coverage. Since providing fiber backhauling to every Base Station (BS) is prohibitively expensive, hybrid architectures combining fiber, Integrated Access and Backhaul (IAB), and free space optics (FSO) have emerged as a practical alternative [1–3]. In standard in-band IAB deployments, high-power Macro Base Stations (MBSs) act as fiber-connected donor nodes, sharing access spectrum to provide wireless backhaul to intermediate IAB-child nodes (SBSs). Deploying FSO as a high-capacity complement to IAB and a cost-effective substitute for fiber yields substantial savings [3] while mitigating the bandwidth starvation inherent in pure in-band IAB deployments [4]. In such multi-technology deployments, a confluent controller [1] can use instantaneous wireless traffic data in a closed loop to dynamically configure both radio and optical links. Conventional network planning assumes static user distributions, disregarding that User Equipment (UE) mobility causes spatial load fluctuations and produces time-varying backhaul demand. The set of SBSs that actually need high-capacity backhaul shifts stochastically with traffic, so no static FSO assignment can match the demand profile at all times. Keeping links active during low-demand intervals wastes energy; keeping them inactive during surges costs coverage. Recent physical-layer optical-to-mmWave switching demonstrations target weather resilience [5, 6], yet their use for system-wide coverage maximization and load-driven energy reduction remains unaddressed. While dynamic FSO reconfiguration under user mobility is explored in flying-platformbased networks [7], it has not been investigated in confluent scenarios operating alongside IAB. Although IAB topology design is studied under static [3, 8–10] or mobile user scenarios [11], the mobility-aware dynamic operation of hybrid fiberIAB-FSO topologies remains an open challenge. We address this gap with a closed-loop heuris-
tic that activates auxiliary FSO links only when short-term average per-SBS user load exceeds a threshold, with hysteresis suppressing switching transients near the decision boundary. To our knowledge, this is the first load-driven reconfiguration scheme for hybrid fiber-IAB-FSO backhaul evaluated under realistic UE mobility conditions. Our results reveal a favorable asymmetry: energy drops faster than coverage. Capping coverage loss at 0.9% yields 8% energy savings, while accepting a 6.7% drop yields up to 44%. Network architecture We consider a hybrid deployment in which fiberconnected MBSs (IAB-donors) provide wireless backhaul to SBSs (i.e., IAB-child nodes) that collectively serve mobile UEs over a metropolitan area. Fig. 1(a) illustrates a topology where 50% of SBSs are fiber-backhauled (yellow squares), 25% rely on IAB (blue triangles), and 25% employ confluent IAB with auxiliary FSO backhaul (red circles). The confluent backhaul between two nodes is depicted in Fig. 1(b), where IAB-child node SBS-1 communicates with IAB-donor MBS-1 via either IAB or FSO. Nodes are equipped with dynamic switching capabilities between the optical and mmWave domains [5, 6]. Fig. 1(c) presents the operational conditions of SBS-1. Operating as a closed-loop system, a centralized controller exploits the monitored wireless load to actively toggle the optical FSO state: exceeding hon activates the FSO link, while dropping below hoff triggers deactivation. The FSO link operates in two energy states: e0 energy units (EU) (standby) and e1 EU (active). In this example, exceeding hon at t1 (event k1 ) activates the FSO link with a slight delay at t2 , necessary to configure the link, align the FSO antennas, and stabilize the channel. The per-user data rate initially drops due to increasing load, but recovers once the FSO backhaul is activated. Conversely, the load drops below hoff at t3 (event k2 ), deactivating FSO at t4 . The per-user data rate temporarily rises due to reduced load before decreasing upon IAB fallback. To formalize the topology, let D, C, S, and U
0.6 0.0 0.0
ac Ob FS
SBS kha
ul
MBS
SBS-1
0.6
1.2
1.8
X Position (km)
2.4
3.0
(a) Network topology.
UE
optical fiber
e0
k2
on
k1
off
t1 t2
t3 t4
User load
k1
h off
k2
(b) Confluent backhaul between SBS-1
t3 t4
time
k2
h on
t1 t2
t5 t6 t7 t8
Achieveable data rate per user
t5 t6 t7 t8
time
data rate
1.2
ele wir
FSO operation e1
state
Y Position (km)
1.8
MBS-1
l
hau
ack ss b
operational energy consumption [EU]
MBS SBS Fiber SBS IAB SBS IAB & FSO MBS ring Fiber link FSO link
2.4
# connected users
3.0
k1 t1 t2
t3 t4
t5 t6 t7 t8
time
(c) Operational values of SBS-1.
and MBS-1.
Fig. 1: Illustrative example of hybrid topology operation.
denote the sets of IAB-donors, child nodes, fiberbackhauled SBSs, and UEs, respectively. Let F ⊂ C be the subset of child nodes equipped with FSO. We define zf ∈ {0, 1} as the binary operational state of node f ∈ F (1 = FSO active, 0 = IAB fallback). For UE u, bu denotes the serving BS and SINRu the experienced signal-to-interferenceplus-noise ratio (SINR). Let D(c) denote the donor serving child node c. For any c ∈ C, SINRc is the backhaul link SINR between child c and its donor D(c). We define βd ∈ [0, 1] as the accessbackhaul sharing ratio for d ∈ D. The access bandwidth for node d ∈ D is Wac,d = W · (1 − βd ). The remaining spectrum is shared among the associated child nodes proportionally to their load: βD(c) Nu,c W Wbh,c = , c∈C (1) Nc,d where Nu,c is the number of UEs at child c ∈ C, and Nc,d is the total UEs served via all children of donor d. For each child node, the access spectrum is Wac,c = W − Wbh,c . We define SEx = log2 (1 + SINRx ) as the spectral efficiency. The downlink rate Ru for UE u is: Wac,d SEu , bu = d ∈ D Nu,d( ) Wac,x min Nu,x SEu , bu ∈ {x ∈ C \ F} ∪ Wbh,x Ru = {x ∈ F : zx = 0} . Nx,d SEc bu ∈ {x ∈ S} ∪ W SEu , Nu,x {x ∈ F : zx = 1} (2) The backhaul bandwidth is shared between the IAB-donor and its children proportionally to the number of UEs connected to each child BS, and each child equally divides its access bandwidth among associated UEs. Specifically, Nu,d is the number of UEs served by donor d ∈ D, and Nu,x is the number of UEs connected to node x ∈ C ∪ S. We formulate a bi-objective problem that maximizes coverage probability C while minimizing energy consumption E under a fixed topology. Coverage is defined as the average fraction of users over time T achieving a data rate above threshold η. The FSO operational energy is modeled as a two-state system: standby (e0 EU) and active (e1 EU), enabling E to be normalized between 0 (all inactive) and 1 (all active). The specific objectives are defined as:
C=
1 |U| · T
X u∈U,t∈T
I (Ru (t) ≥ η) ,
E=
1 X zf , |F| f ∈F
(3) where I(·) denotes the indicator function. We propose a hysteresis-based heuristic HH that toggles FSO links based on the average load N̄u,f over a monitoring window τ . Two thresholds, hon > hoff , create a dead-band that prevents rapid switching transients. For each node f ∈ F, the algorithm: • turns FSO on if N̄u,f ≥ hon , and sets zf = 1, • turns FSO off if N̄u,f ≤ hoff , and sets zf = 0. Let HH(a, b) denote the HH algorithm executed with thresholds hon = a, hoff = b. As baselines, we consider All On, where all FSO links are permanently active, and All Off, where all FSO links are permanently deactivated. Numerical Results The topology (Fig. 1(a)) comprises 14 MBSs and 201 SBSs across a 3 km2 area. Specifically, 100 SBSs are fiber-backhauled, 50 use IAB with auxiliary FSO, and the remainder are pure IAB-child nodes. The network operates at 28 GHz with a system bandwidth W = 1 GHz [3, 11]. Channel characteristics follow the 5GCM Urban Macro (UMa) model [12] with 3GPP TR 38.901 blocking probability [13]. Transmit powers are 40 dBm for MBSs and 24 dBm for SBSs. All BS sites use three-sector antennas (60◦ half-power beamwidth, 24 dBi main-lobe gain, −2 dBi side-lobe gain). We simulate 1000 UEs using a random waypoint model (1–15 m/s, 0–10 s pauses). The 3 km2 area is partitioned into a 3×3 grid; UEs select destinations equally between their current and the central square, inducing temporary congestion in the center of the network. UEs associate with the BS offering the highest received power (computed as a product of transmit power, antenna gain, and path loss). FSO operational energy consumption is normalized between all-active and all-inactive states. Performance is evaluated over a 1-hour window (T ) with 1-second granularity. We set τ = 30 seconds to balance responsiveness and stability. We empirically select HH thresholds to evaluate the full operational trade-off from conservative (coverage-
85
f45 f41
FSO state
75 83 82
70
f21
80
2100
2200
2300
2400
2500
2600
2700
f1
HH(2,1): 81.1% HH(3,1): 80.6% HH(3,2): 79.3%
0
600
HH(5,2): 76.4% All Off: 61.9% All On: 81.8%
1200
1800 2400 3000 Time [s] Fig. 2: User coverage over time.
0.6
HH(2,1): 0.92 HH(3,1): 0.87 HH(3,2): 0.75
HH(5,2): 0.56 All Off: 0.00 All On: 1.00
0.2
e0
0.0
1200
1800 Time [s]
2400
3000
3600
f45 f41
0.8
0.4
600
Fig. 4: FSO state for HH(3,2) under traffic pattern A.
e1
1.0
0
3600
0
600 1200 1800 2400 3000 3600 Time [s] Fig. 3: Normalized number of active FSO links over time, which also corresponds to the normalized FSO operational energy consumption.
prioritizing) to aggressive (energy-saving) configurations: (hon , hoff ) ∈ {(2, 1), (3, 1), (3, 2), (5, 2)}. Results are averaged over 10 independent UE initial positions and mobility patterns. To ensure visual clarity, 95% confidence intervals are omitted from all plots; the time-averaged (and absolute maximum per-time-step) error margins are 0.3 p.p. (2.7 p.p.) for user coverage and 0.012 (0.06) for normalized FSO energy consumption. Fig. 2 presents user coverage over time (dashed lines indicate averages). The All On strategy achieves 81.8% coverage, while All Off yields 61.9%. HH(2,1) achieves coverage only 0.9% below the All On benchmark, while HH(5,2) results in a reduction of 6.7%. Normalizing to the achievable range (All On minus All Off ), the HH variants attain 96.5%, 94.0%, 87.4%, and 72.9% of the All On highest performance for threshold pairs (2, 1), (3, 1), (3, 2), and (5, 2), respectively. Fig. 3 shows FSO operational energy consumption over time. HH(2,1) reduces energy con-
FSO state
55
f11
79
65 60
Normalized active FSO links
f32
f31
81
FSO operational energy consumption [EU]
User coverage [%]
80
f32
f31
f21
f11
f1
0
600
1200
1800 Time [s]
2400
3000
3600
Fig. 5: FSO state for HH(3,2) under traffic pattern B.
sumption by 8% relative to All On, while HH(5,2) achieves a 44% reduction. Crucially, energy drops faster than coverage across HH configurations, confirming that substantial energy savings are attainable at the cost of smaller coverage reductions. Figs. 4 and 5 show the per-SBS FSO state over time under two distinct traffic patterns. Dark shading denotes active FSO; light shading denotes inactive. The set of SBSs benefiting from additional FSO capacity varies both temporally and across traffic patterns. Two representative nodes, f32 and f45 , illustrate this clearly. Node f32 remains active throughout the simulation under pattern A but is mostly inactive under pattern B. Conversely, f45 is predominantly inactive under pattern A yet active under pattern B. This confirms that the need for high-capacity backhaul is inherently stochastic: no static a priori assignment can match the adaptivity of dynamic closed-loop control. Conclusion Dynamic operation of hybrid fiber-IAB-FSO topologies yields meaningful energy savings under UE mobility. By selectively activating auxiliary FSO links based on per-SBS user load, our closed-loop control mechanism cuts energy by 8% with a 0.9% coverage loss. Accepting a 6.7% coverage drop raises energy savings up to 44%. A per-FSOlink state analysis confirms that nodes requiring high-capacity backhaul vary stochastically, high-
lighting the necessity of dynamic strategies over static planning. Future work will incorporate FSO channel-state monitoring into the control loop to maintain reliability under adverse weather. Acknowledgements This work has been supported by the Horizon Europe ECO-eNET project, funded by the SNS JU under grant agreement No. 101139133.
[10]
C. Madapatha, B. Makki, A. Muhammad, E. Dahlman, M.-S. Alouini, and T. Svensson, “On topology optimization and routing in integrated access and backhaul networks: A genetic algorithm-based approach”, IEEE Open Journal of the Communications Society, vol. 2, pp. 2273–2291, 2021. DOI: 10 . 1109 / OJCOMS . 2021 . 3114669
[11]
P. Lechowicz, C. Madapatha, C. Natalino, T. Svensson, and P. Monti, “User-mobility-aware optimization of fiber placement in hybrid fiber–IAB networks”, in 2026 Optical Fiber Communications Conference and Exhibition (OFC), Mar. 2026, W1H.7.
[12]
T. S. Rappaport, Y. Xing, G. R. MacCartney, A. F. Molisch, E. Mellios, and J. Zhang, “Overview of millimeter wave communications for fifth-generation (5G) wireless networks—with a focus on propagation models”, IEEE Transactions on Antennas and Propagation, vol. 65, no. 12, pp. 6213–6230, Dec. 2017. DOI: 10 . 1109/TAP.2017.2734243
[13]
ETSI, “5G; study on channel model for frequencies from 0.5 to 100 GHz (3GPP TR 38.901 version 16.1.0 release 16)”, Tech. Rep. TR 38 901 V16.1.0, Nov. 2020, 103 pp. [Online]. Available: https://www.etsi.org/deliver/ etsi_tr/138900_138999/138901/16.01.00_60/tr_ 138901v160100p.pdf
References [1]
R. Raj, D. Dass, K. Kaeval, et al., “Towards efficient confluent edge networks”, in 2024 Joint European Conference on Networks and Communications & 6G Summit (EuCNC/6G Summit), Antwerp, Belgium, Jun. 2024, pp. 1163–1168. DOI: 10.1109/EuCNC/6GSummit60053. 2024.10597093
[2]
J. Zhang, J. Wang, Y. Xu, M. Xu, F. Lu, L. Cheng, J. Yu, and G.-k. Chang, “Fiber-wireless integrated mobile backhaul network based on a hybrid millimeter-wave and free-space-optics architecture with an adaptive diversity combining technique”, Optics Letters, vol. 41, no. 9, pp. 1909–1912, May 2016. DOI: 10.1364/OL.41. 001909
[3]
C. Madapatha, P. Lechowicz, C. Natalino, P. Monti, and T. Svensson, “Joint fiber and free space optical infrastructure planning for hybrid integrated access and backhaul networks”, in IEEE International Symposium on Personal, Indoor and Mobile Radio Communications (PIMRC), Sep. 2025. DOI: 10 . 48550 / arXiv . 2507 . 20367
[4]
M. Polese, M. Giordani, T. Zugno, A. Roy, S. Goyal, D. Castor, and M. Zorzi, “Integrated access and backhaul in 5G mmwave networks: Potential and challenges”, IEEE Communications Magazine, vol. 58, no. 3, pp. 62–68, Mar. 2020. DOI: 10.1109/MCOM.001.1900346
[5]
M. Vargemidou, C. Vagionas, A. Kokkinis, A. Ntanos, A. Stathis, P. Kourelias, G. Giannoulis, H. Avramopoulos, K. Siozios, N. Pleros, and A. Miliou, “Field trial of an SDNcontrolled, hybrid fiwi FSO/mmwave x-haul with zerotouch handovers and record-high data rate for 6G”, Journal of Lightwave Technology, vol. 43, no. 18, pp. 8761– 8769, Sep. 2025. DOI: 10.1109/JLT.2025.3592851
[6]
E. Kyriazi, P. Toumasis, G. Brestas, et al., “Demonstration of an SDN-enabled, resilient, rate-adaptable hybrid fiber/outdoor FSO/mmwave transport for a 6G xhaul architecture”, Journal of Optical Communications and Networking, vol. 17, no. 7, p. 638, Jul. 2025. DOI: 10.1364/JOCN.551852
[7]
Z. Gu, J. Zhang, Y. Ji, L. Bai, and X. Sun, “Network topology reconfiguration for FSO-based fronthaul/backhaul in 5G+ wireless networks”, IEEE Access, vol. 6, pp. 69 426–69 437, 2018. DOI: 10.1109/ACCESS.2018. 2880880
[8]
J. Zhang, S. Chetty, Q. Wang, C. Sun, P. D. Mitchell, D. Grace, and H. Ahmadi, “Optimizing 6G dense network deployment for the metaverse using deep reinforcement learning”, in IEEE INFOCOM 2025 - IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), London, United Kingdom, May 2025, pp. 1–6. DOI: 10.1109/INFOCOMWKSHPS65812.2025.11152919
[9]
C. Huang and X. Wang, “A bayesian approach to the design of backhauling topology for 5G IAB networks”, IEEE Transactions on Mobile Computing, vol. 22, no. 4, pp. 1867–1879, Apr. 2023. DOI: 10.1109/TMC.2021. 3118958