Frontiers in Computing and Intelligent Systems 
ISSN: 2832-6024 | Vol. 13, No. 3, 2025 

 

68 

An	Energy‐Efficient	Multi‐Objective	Optimization	
Framework	for	QoS‐Aware	Resource	Allocation	in	5G	
Slicing	Networks	
Li Ding *, Minjia Yu, Zeyuan Du 

Nanjing Normal University of Special Education, Nanjing, Jiangsu, China 

* Corresponding author: Li Ding (Email: 594822001@qq.com) 

 

Abstract: This paper investigates decision optimization for wireless resource allocation and energy-efficient scheduling in 
multi-user, multi-task 5G network slicing scenarios. We propose a hybrid computational framework combining particle swarm 
optimization, temporal scheduling, and genetic algorithms to address interference management, power control, and differentiated 
QoS guarantees across heterogeneous base station deployments. The proposed models incorporate Tikhonov regularization and 
service-type-specific QoS scoring to improve robustness and solution quality. Simulation-based experiments in MATLAB 
demonstrate high resource utilization (75.8%), system utility (1866.0), and significant energy efficiency gains (2.3179 ratio) 
under dynamic and interference-prone environments. Our results validate the framework’s capability to balance spectrum 
utilization, QoS maximization, and energy consumption, offering a scalable solution for intelligent network resource 
management. 

Keywords: Resource Allocation; Genetic Algorithm; QoS Optimization. 
 

1. Introduction 
With the proliferation of 5G and beyond wireless networks, 

ensuring differentiated Quality of Service (QoS) across ultra-
reliable low-latency communication (URLLC), enhanced 
mobile broadband (eMBB), and massive machine-type 
communication (mMTC) slices has become a fundamental 
optimization challenge. Each service type imposes distinct 
requirements in terms of delay sensitivity, data rate, and 
reliability, demanding intelligent resource allocation and 
energy-efficient scheduling strategies [1,2]. 

However, current resource allocation approaches often 
struggle with heterogeneous base station deployments, 
dynamic channel interference, and real-time service 
variability. The interplay between transmission power control, 
spectrum allocation, and delay constraints further complicates 
the optimization landscape [3,4]. Moreover, ensuring robust 
performance under dynamic slicing scenarios requires 
additional mechanisms such as regularization, QoS-aware 
scheduling, and task-aware modeling. 

To address these challenges, we propose an integrated 
multi-objective optimization framework that combines 
particle swarm optimization, genetic algorithms, and 
temporal scheduling to jointly optimize resource blocks 
(RBs), transmission power, and QoS utility. Our approach 
incorporates Tikhonov regularization for stability, task-aware 
traffic modeling, and differentiated scoring for URLLC, 
eMBB, and mMTC services. Experimental results under 
interference-rich conditions demonstrate superior system 
utility, energy efficiency, and service fairness [5,6]. 

2. QoS-Oriented Resource Allocation 
Using PSO in a Single-Cell Network 

2.1. Parameter and Variable Definition 
Users are first divided into three service categories, as 

given by: 

     
UR EB MT1 1 1, , , , , , ,N N NU U e e m m     U    (1) 

Here, URLLCU , eMBBU , and mMTCU  represent the sets of 

users corresponding to URLLC, eMBB, and mMTC slices, 
respectively. For each user iu , the dataset provides large-

scale path loss iL  (in dB), small-scale Rayleigh fading ih 

(unitless), and task data volume iD  (in Mbit). The decision 

variable is defined as: 

0 , ,ux u  U               (2) 

Each RB has a bandwidth of 180 kHz, and all users are 
assumed to have identical transmission power txP . The noise 

power is computed as: 

 0 10174 10 log 7 (dB m )RBN B           (3) 

Minimum RB allocation thresholds are required for each 
service type. The QoS-related penalty parameters are: 

5?ms, 5, 0.95UR URT P             (4) 

100 , 50Mbps, 3EB EB EBT ms R P          (5) 

500?ms, 1MT MTT P                (6) 

2.2. Signal-to-Noise Ratio and Transmission 
Rate Calculation 

For each user iu , the received signal-to-noise ratio (SNR) 

in linear scale (mW/mW) is given by: 

0

210

210 10

10

10 10

u u

v v

u

P L

u
u P L N

v
v

h

h












G

            (7) 



 

69 

The corresponding transmission rate ir  in Mbps is derived 

using the Shannon formula: 

 R B 2log 1 (M bps).u u ur x B          (8) 

2.3. Quality of Service (QoS) Formulation 
Each type of user is associated with a specific QoS 

formulation iQ . 

For URLLC users, the end-to-end latency id  must not 

exceed the strict delay threshold. The QoS is defined as: 

1 , ifi
i i UR

UR

d
Q d 


              (9) 

Otherwise,
U RiU

Q P  . 

For mMTC users, only access success is required within 
the time frame: 

EB EB1 ,j

j j

EB e

e e j
EB

R r
Q if r R and d T

R


      (10) 

Otherwise, 
EBje

Q P   

For mMTC users, only access success is required within 
the time frame: 

,  if 
| |

,  others. 

k

j
j M

j MT
m

MT

TQ M

P




  








       (11) 

2.4. Optimization Objective and Constraints 
The model seeks to maximize the total system QoS across 

all users, subject to bandwidth constraints and user-type 
allocations. The objective function is: 

   2
max ; , ,

u u u u u u ux
u

w Q x L h D



U

       (12) 

Subject to: 

 

 
 
 

  queue 
0 2

, ,

 s.t. , , , , log 1 ,

, ,

U i

u
u RB u e j u u u RB u u u

u u

m k

n u U
D

x N z n u e x r x B d d
r

n u m




 


       
 


U

          (13)

 

Here, uw  represents the importance weight for each user. 

The input parameters 
2

, ,u u uL h D  are data-driven and 

provided by the system dataset. 

2.5. Solution Strategy and Numerical Results 
This section applies the Particle Swarm Optimization (PSO) 

algorithm under a fixed transmission power of 30 dBm. Based 

on MATLAB simulations, the system generates user-specific 
RB allocations and performance metrics. 

As shown in Figure 1, the convergence curve of the PSO 
algorithm confirms its stability and optimization 
effectiveness. The average QoS achieved across the URLLC, 
eMBB, and mMTC slices are 0.4415, 0.5190, and 1.0000 
respectively, demonstrating the model's ability to enforce 
differentiated service guarantees while optimizing spectrum 
utilization. 

 

 
Figure 1. RB allocation performance under PSO optimization 

 

3. Matrix-Based Modeling for QoS-
Aware Scheduling Optimization 

3.1. Matrix Symbol Definition 
To support the scheduling process, the time slot set is 

defined as  1, 2, ,10 T , and the user set as 

 1, 2, , N U , where users are categorized into URLLC, 

eMBB, and mMTC. Let urllc embb mmtcN N N N    denote 

the total number of users. A delay tolerance matrix is first 
initialized. The large-scale fading matrix is defined as: 

', ,10
(?dB)  Large-scale fading table u t u tN

L L L


     (14) 

Based on this, the small-scale gain matrix is constructed as: 

', ,10
 Small-scale gain table u t u tN

H h h


      (15) 

The position matrix of users is defined as: 

, , 2 10u t u t N
x y  
   P              (16) 

Task matrix representing data demand is expressed as: 

, ,10
 Task demand in Mbit u t u tN

T T
   T      (17) 

Resource allocation across users and time slots is defined 
by: 



 

70 

, ,10
{0,1, ,20}u t u tN

R R
    R          (18) 

The system assumes parameters tx 30dBmP  , 

RB 360kHzB  , 0 Noise Power DensityN  , tot 50RB   as 

the total available resource blocks per time slot. 

3.2. Channel Quality and Transmission Rate 
Calculation 

Based on wireless channel propagation, large-scale fading 
values in dB are first converted to linear scale as , /1010 u tL . 
Given transmission power txP  and system noise 

0 RBN N B  , the instantaneous SNR is calculated by: 

,

,
,

1010
u t

tx u t
u t L

P h

N








                 (19) 

This yields the SNR matrix , 10u t N



   Γ  

The corresponding spectral efficiency is calculated as: 

 , 2 ,log 1u t RB u tc B             (20) 

Given the number of allocated RBs and channel conditions, 
the user transmission rate is then: 

, , ,u t u t u tD R c                (21) 

The rate matrix is then defined as 
, 10[ ]u t ND R  . 

3.3. User QoS Function 
The QoS satisfaction matrix 

,u tQ  is determined according 

to user type and delay-rate tradeoff. 
For URLLC users, the QoS utility is defined by: 

   ,
, , ,

,

max 1 , 0 , .u t
u t u t u t

u t

T
Q ms

D
         (22) 

where 0   is the delay penalty coefficient. 
For eMBB users, the satisfaction function is based on rate 

thresholds: 

,
, min ,1u t
u t

SLA

D
Q

R

 
  

 
             (23) 

For mMTC users, both delay and rate are jointly considered: 

, ,

,,
, ,

,

1,

,

u t u t

u tu t
u t u t

u t

D T

DQ
D T

T



   


          (24) 

where (0,1)   represents the penalty factor for service 

mismatch. Thus, the complete QoS matrix is denoted as 

, 10[ ]u t NQ Q  . 

3.4. Constraints 
To ensure feasibility of the scheduling, the model 

introduces several constraints. 
First, the total RB allocation must not exceed the system 

capacity: 

, tot
1

,
N

u t
u

R RB t


   T          (25) 

Second, integer constraints for RB allocation: 

, {0,1, 2, }u tR                   (26) 

Third, user-type-specific minimum RB guarantees are 
enforced: 

, , {URLLC, eM BB, mM TC}.
s

u t s
u

R m s


 
U

   (27) 

3.5. Objective Function 
Based on the modeling and constraints, this study defines 

the objective as maximizing the cumulative QoS utility over 
all users and time slots: 

 ,

10

,
1 1

max
u t

N

u u tR
t u

J w Q
 

            (28) 

where uW  represents the priority weight of user u , which 

is assigned based on user class preferences. Therefore, the 
complete optimization problem can be formulated as a 
constrained integer program: 

, tot ,
1

, ,
t

N

u t u t s
u u

R RB R m
 

  
U

        (29) 

subject to: 

 10
0

10

, , , ,
1 1

,
, 0 ,

, ,

max , ,

 s.t.  

,

N

N

u u t u t u t u t
t u

u t
u t u t

u t u t

w Q R c T

T
R

R c





 







 




R

   (30) 

3.6. Dynamic Analysis of Resource Slicing 
Results 

To evaluate the dynamic slicing performance of the 
proposed resource allocation strategy, simulation 
experiments were conducted to monitor real-time variations 
in key indicators such as resource utilization, slice-wise QoS 
levels, and throughput. 

This section presents the model validation results derived 
from MATLAB simulations under dynamic slicing scenarios. 
The corresponding performance metrics are illustrated in 
Figure 2. 

As shown in Figure 2, the QoS value for URLLC remains 
at 0.07, which aligns with its prioritized service requirement. 
The mMTC slice demonstrates a low-speed transmission of 
5.10 Mbps, falling within acceptable bounds for its service 
class. The peak slice proportion reaches 28.2%, while 
resource utilization fluctuates between 46% and 100%, 
suggesting a fair and adaptive resource allocation mechanism. 
The observation that QoS remains below 0.5 further indicates 
the overall stability of the model. 

4. Optimization Framework for Multi-
Slice Downlink Resource Allocation 

4.1. System Structure and Notations 
This section considers a multi-slice network consisting of 

one macro base station and multiple small base stations, with 
users categorized into URLLC, eMBB, and mMTC slices. 



 

71 

Each slice includes 10 users. The time is discretized into 
frames  0,1, ..., 1T T , and the bandwidth is allocated 

based on resource blocks (RBs). Let ( )ijx t  denote the 

allocation indicator and ( )jp t  the transmission power: 

 , , m in m ax( ) {0,1} , ( ) , ,i j r jx t p t P P        (31) 

 

 
Figure 2. QoS, RB usage, and throughput dynamics across service slices 

 

4.2. Channel and Interference Model 
The signal propagation incorporates geometric path loss 

and Rayleigh fading. The SINR between user i  and base 
station j  is: 

,
, ,

, 0 RB
,

( ) ( )
( )

( ) ( )
j i j

i j r
k i k

k k j

p t G t
t

p t G t N B


 




B

      (32) 

All user-to-base station links are summarized in the gain 
matrix ( )t . 

4.3. Task Arrival and Service Process 
For each user, task arrivals are assumed continuous, and the 

total delivered traffic is computed using transmission rate: 

 , , RB 2 , ,( ) log 1 ( )i j r i j rR t B t           (33) 

The aggregated user throughput is: 

max

, , , ,
 B 1

( ) ( ) ( ).
R

i i j r i j r
j r

t x t R t
 

             (34) 

4.4. Objective and Constraints 
To optimize differentiated QoS requirements, we construct 

a utility function for each slice: 

 

th

exp , URLLC

( , ) log 1 , eMBB

, mMTC

i

i
i

i

D i

f D i

i







 


  
      

 

 



      (35) 

Power and RB smoothing terms are added to penalize 
excessive fluctuations. The final objective is: 

{ , } 0
max ( )d

T

x p p xU t t    R R      (36) 

4.5. Model Solving and Result Analysis 
This study employs a randomized search approach to solve 

the optimization model. As shown in Figure 3, the 
optimization stabilizes at a system utility of −283232.7391. 
The final total throughput reaches 5.97 × 10⁶ bps, and the 
average user throughput is 8.53 × 10⁴ bps. 

 

 
Figure 3. Convergence and resource allocation visualization for the optimized slicing model 

 

5. Energy-Efficient Multi-Objective 
Optimization for QoS-Aware 
Resource Allocation 

5.1. Modeling Framework 
In this section, we construct a resource optimization model 

that maximizes user service quality while minimizing total 
energy consumption. The system objective is to maximize the 

utility function U , defined as the weighted sum of user QoS 
scores: 

 Users 
i i

i

U w Q


                  (37) 

Here, iw  denotes the priority weight of user i , and iQ  

is the corresponding QoS level. 
The total energy consumption is comprised of fixed station 

energy, RB activation energy, and transmission energy: 



 

72 

total static activation transmission 
RBs BS

( ) ( )
j k

P P P j P k
 

      (38) 

The final optimization objective is formulated to balance 
utility and energy: 

 Users 

total 

max
i i

i

w Q

P


 
 
 
 
 


             (39) 

5.2. Optimization Results and Evaluation 
To solve the above model, this section adopts a genetic 

algorithm-based approach. Parameters are set to ensure 
balance between energy efficiency and service quality. As 
shown in Figure 4, the proposed optimization method 
effectively improves multi-base-station coordination metrics 
across various performance dimensions. 

 

 
Figure 4. Multi-base-station performance metrics after optimization 

 

6. Conclusion 
This paper proposes an integrated multi-objective 

optimization framework to jointly address delay-aware 
scheduling, differentiated QoS satisfaction, and energy-
efficient transmission under multi-slice wireless service 
scenarios. The proposed models incorporate Tikhonov 
regularization, task-aware arrival modeling, and QoS utility 
scoring to construct solvable optimization objectives, which 
are effectively handled by meta-heuristic algorithms such as 
particle swarm optimization and genetic algorithms. 
Experimental evaluations demonstrate that the proposed 
approach significantly improves system utility, reduces energy 
consumption, and ensures fairness across URLLC, eMBB, and 
mMTC services under dynamic channel and interference 
conditions. Compared to baseline strategies, the method 
achieves higher robustness and adaptability in multi-service 
environments. However, the current framework relies on 
simplified traffic arrival assumptions and fixed 
hyperparameter configurations, which may limit its 
generalizability in highly dynamic real-time systems. Future 
work will focus on integrating deep reinforcement learning for 
adaptive policy generation and extending the framework to 
support distributed scheduling in edge-enabled heterogeneous 
architectures.. 

Acknowledgments 
The authors gratefully acknowledge the financial support 

from xxx funds. 

References 
[1] Mei J, Wang X, Zheng K, et al. Intelligent radio access network 

slicing for service provisioning in 6G: A hierarchical deep 
reinforcement learning approach[J]. IEEE Transactions on 
Communications, 2021, 69(9): 6063-6078. 

[2] Si P, Zhang Q, Yu F R, et al. QoS-aware dynamic resource 
management in heterogeneous mobile cloud computing 
networks[J]. China Communications, 2014, 11(5): 144-159. 

[3] Attar H, Issa H, Ababneh J, et al. A Review of 6G Conceptual 
Components, Its Ultra-Dense Networks, and Research 
Challenges towards Cyber-Physical-Social Systems[J]. 
International Journal of Crowd Science, 2024. 

[4] Azimi Y, Yousefi S, Kalbkhani H, et al. Energy-efficient deep 
reinforcement learning assisted resource allocation for 5G-
RAN slicing[J]. IEEE Transactions on Vehicular Technology, 
2021, 71(1): 856-871. 

[5] Tian J, Liu Q, Zhang H, et al. Multiagent deep-reinforcement-
learning-based resource allocation for heterogeneous QoS 
guarantees for vehicular networks[J]. IEEE Internet of Things 
Journal, 2021, 9(3): 1683-1695. 

[6] Mhatre S, Adelantado F, Ramantas K, et al. Intelligent QoS-
aware slice resource allocation with user association 
parameterization for beyond 5G O-RAN-based architecture 
using DRL[J]. IEEE Transactions on Vehicular Technology, 
2024.