Dynamic power and frequency domain allocation for dedicated sensing signals in downlink ISAC Authors: Madeeha Aman, Ghani Ur Rehman, Muhammad Zubair, Riad Alharbey, Ali Daud, Sajid Ullah Khan Published: 2025 ABSTRACT AbstractThis paper presents a dynamic joint resource allocation framework for downlink Integrated Sensing and Communication (ISAC) systems. The proposed approach simultaneously optimizes power and frequency assignment using an iterative optimization algorithm that adapts to real-time channel state information and sensing requirements. A composite utility function is introduced to balance sensing accuracy and communication throughput, enabling flexible trade-offs for different application scenarios. Comparative analysis against benchmark schemes demonstrates that the proposed method achieves notable improvements in both sensing accuracy and communication rate, while also providing insights into power distribution across users and sensing tasks. These results highlight the potential of the framework as a practical and adaptive solution for ISAC-enabled 6G networks.Keywords: 6G, Dedicated sensing signals, Downlink communication, Dynamic power allocation, Frequency domain multiplexing, ISAC, Resource allocationSubject terms: Engineering, Mathematics and computing Introduction As 6G networks are evolving, Integrated Sensing and Communication (ISAC) systems attracted more and more research interest, and possible applications include vehicular networks, smart cities, and environmental monitoring. In downlink ISAC scenarios, the signals of interest from the same domain may compete with each other for the same resources with the later usually leading to performance degradation in both domains. The idea of Integrated Sensing and Communications (ISAC) has drawn wide interest from both the academic community and the industry players because of its great potential in promoting the development of 6G and future wireless networks1. ISAC fundamentally distinguishes itself by its capability to share same time, frequency, power, and hardware resources for communication and sensing at the same time2,3. In stark contrast to the conventional approach of Frequency Division Sensing and Communications (FDSAC), where dedicated and separate frequency bands and infrastructure are required for both functions. The spectrum utilization, energy consumption and hardware requirements of ISAC are therefore expected to outperform those of FDSAC2,3. From Information theoretic viewpoint, there are two key system performance metrics, the sensing rate (SR) and communication rate (CR)4–6 which are used to assess the performance of ISAC. The SR reflects the system capacity for estimating environmental information through processing within sensorial events. The CR, from the other hand, measures the capability of the system to efficiently transfer data while performing communication operations. This data will provide researchers with insights into the overall performance and effectiveness of ISAC for dual S&C functionalities. Motivated by this, to more accurately quantify the fundamental performance limits of ISAC and investigate the performance tradeoffs of S&C6 developed a new MI-based framework for ISAC, in which the SR and CR are characterized by the sensing MI and the communication MI, respectively. There has been a significant surge in literature relevant to the performance of integrated sensing and communication (ISAC) recently. Characterization of SR-CR type of regions of ISAC systems with only one communication UT was performed in7 with information-theoretical analysis. Moreover9 discusses the single- UT ISAC systems performance trade-offs in S & C performance, which can be describable in an estimation-theoretical view, exploiting the Cramér-Rao bound metric for sensing and the minimum mean-square error metric for communications. Utilizing the MI-based framework proposed in5, another work8 investigated the fundamental performance limits of ISAC systems involving a single-antenna UT under various beamforming vector designs. Building on this, researchers have expanded their inquiries to multi-UT scenarios where each UT is equipped with a single antenna, as explored in9. In their work, dirty paper coding is employed for interference management among communication users, and it is assumed that the interference from the sensing signal to the communication signal is perfectly mitigated to prioritize communications. These approaches, while effective, demand computationally intensive implementations. To overcome these challenges, this research work10 has investigated downlink multiple-input multiple-output (MIMO)-ISAC systems featuring multiple UTs, each equipped with multiple antennas. User interference is handled using SDMA (space division multiple access) in this mode. However, the number of serviceable UTs is limited by the number of transmit antennas available in the base station (BS) intrinsically. This work has contributed to the existing literature of multi-antenna ISAC systems. That being said, in the case of an overloaded system, both UTs can suffer from severe IUI caused by the limited DoFs of the spatial domain. Such interference may affect the performance of the system by decreasing its effectiveness. Non-orthogonal multiple access, which is capable of reducing IUI, is an approach to remedy the aforementioned problem and enhance the system performance. In general, for IUI, NOMA dynamically assigns power to users based on their channel conditions so that weaker channels will be able to fight off stronger channel interference11,12. Moreover, the use of successive interference cancellation (SIC) at the receiver side decreases the impact of IUI on successive users13,14. NOMA is a significant enhancement for ISAC systems since its superior spectral efficiency enables more simultaneous users than the traditional multiple access techniques e.g., SDMA. Several works15–19 have introduced the concept of NOMA-ISAC. Nevertheless, it is worth noting that the majority of these works have primarily concentrated on the design aspects related to waveform or beamforming. On the other hand, only a few recent works went to the detail of the core performance evaluation of NOMA-ISAC systems20–22. That is, the work of20 studied the behavior of an uplink NOMA-based Semi-ISAC composition under a single-input single-output equivalent, focusing on only one UT pair and thereby ignoring inter-pair interference (IPI). Similarly, the work in21 proposed an uplink NOMA-ISAC framework with single-antenna UTs, exploring the impact of different decoding orders for the S&C signals. On the other hand, the performance of downlink NOMA-ISAC systems has been only studied in the work22, where a basic framework of single-input single-output NOMA-ISAC with a single UT pair is developed. Therefore, research on the performance characterization of NOMA-ISAC is still in its early stages, and the more general NOMA-ISAC within a MIMO framework involving multiple pairs of UTs remains unexplored. In more recent studies, the authors in23 focuses on covert communication protocols for inter-environmental communication, particularly between drones and water-based vehicles. The research builds on the challenge of communication security, specifically in environments with high mobility and reflection-induced interference between air and water. It provides insights into location-aware communication protocols to ensure secure, reliable data exchange. The work references prior studies that explored secure communication in challenging environments, particularly those addressing node mobility and environmental interference. The study in24 presents a covert communication method using spatial modulation (SM) and artificial noise (AN) to enhance secrecy while transmitting data. It examines power allocation strategies for optimal covert capacity, comparing Gaussian and discrete constellation point distributions. This paper is relevant for improving covert communication methods and could inform resource management strategies that enhance security in integrated sensing and communication systems. The paper25 examines the application of orthogonal time frequency space (OTFS) modulation in cell-free massive MIMO ISAC systems. The work emphasizes power allocation strategies for maximizing the communication signal-to-interference-plus-noise ratio (SINR) while satisfying specific sensing SINR constraints. The results demonstrate that OTFS modulation significantly improves spectral efficiency (SE) compared to traditional OFDM signals, particularly in autonomous driving scenarios. This research28 addresses the challenges in vehicular communication networks, particularly focusing on resource allocation through deep reinforcement learning (DRL). It explores the impact of unreliable channel state information and the complexities of large-scale vehicular networks. The use of Proximal On-Policy schemes (POPS) optimizes data transmission, making it highly relevant for systems that require adaptive resource allocation. This paper26 explores the integration of massive MIMO with ISAC systems, focusing on the power allocation and the communication and sensing performance. The study provides a detailed analysis of achievable communication rates and the Cramér–Rao lower bound (CRLB) for sensing. The authors present a power allocation strategy based on successive convex approximation (SCA) to optimize the system’s communication rate while meeting sensing accuracy constraints. The authors in27 introduces an optimization algorithm using a combination of chaos and elite reverse learning with the Sparrow Search Algorithm (SSA) to optimize communication in Industrial Internet of Things (IIoT) systems30,31. It addresses energy consumption, device collaboration, and task offloading in IIoT systems, with a focus on collaborative optimization. This study 28 investigates the joint optimization of beamforming and resource allocation in STAR-IRS-aided SCMA ISAC systems using Meta Deep Reinforcement Learning (MDRL). The paper addresses the challenges posed by non-convex optimization problems in integrated sensing and communication systems. The use of STAR-IRS enhances the system’s energy efficiency and coverage, which is further optimized using MDRL. Simulation results show improvements in energy efficiency compared to traditional methods. While recent works have introduced dynamic resource allocation methods for ISAC, many of these are limited in scope—focusing primarily on beamforming or power allocation in single-user/single-carrier scenarios, or requiring computationally intensive optimization that restricts real-time applicability. Moreover, the joint allocation of power and frequency resources across multiple users and sensing tasks remains underexplored. Consequently, there is still a critical need for adaptive yet lightweight frameworks that can balance communication and sensing objectives efficiently in practical downlink ISAC deployments. Therefore, there is a critical need for an adaptive, low-complexity resource allocation framework that can dynamically assign power and frequency resources based on real-time channel conditions and task demands, while ensuring an effective balance between sensing and communication objectives in downlink ISAC scenarios. This paper proposes a dynamic resource allocation algorithm based on downlink ISAC systems that optimizes the power and frequency jointly allocated to a number of communication users as well as a number of sensing tasks, to overcome the above limitations. Different from conventional fixed schemes, the proposed method utilizes instantaneous channel state information (CSI) and task-specific sensing requirements to reasonably allocate resources in real-time. The algorithm allows for a tunable balance between sensing accuracy and communication rate that can be tailored towards an optimal solution based on the end application by formulating a utility-based optimization problem. Through comparative simulations, the dynamic approach is shown to significantly outperform conventional schemes in terms of sensing mean square error (MSE), communication throughput, and overall utility—demonstrating its potential as a foundational component in the design of adaptive and intelligent ISAC-enabled 6G systems. The main contributions of this paper are summarized as follows: Joint Resource Allocation Framework:We propose a dynamic resource allocation scheme for downlink ISAC systems that jointly optimizes power and frequency allocation to balance the trade-off between sensing accuracy and communication performance. A composite utility function is constructed that captures the sensing and communication objectives, controlled by a tunable trade-off parameter λ, allowing adaptive prioritization in accordance with system specifications. Joint Resource Allocation Framework:We propose a dynamic resource allocation scheme for downlink ISAC systems that jointly optimizes power and frequency allocation to balance the trade-off between sensing accuracy and communication performance. A composite utility function is constructed that captures the sensing and communication objectives, controlled by a tunable trade-off parameter λ, allowing adaptive prioritization in accordance with system specifications. Comparative Evaluation with Benchmark Schemes: The performance of the proposed dynamic scheme is evaluated against benchmark NOMA-ISAC approaches in terms of key performance metrics, including sensing MSE, communication rate, and total utility. To examine the efficacy of the proposed algorithm under varying user specifications, MATLAB-based simulations were conducted under Rayleigh fading and AWGN noise, validating the performance of the algorithm across diverse λ values and user/task configurations, clearly demonstrating its adaptability and superiority. Comparative Evaluation with Benchmark Schemes: The performance of the proposed dynamic scheme is evaluated against benchmark NOMA-ISAC approaches in terms of key performance metrics, including sensing MSE, communication rate, and total utility. To examine the efficacy of the proposed algorithm under varying user specifications, MATLAB-based simulations were conducted under Rayleigh fading and AWGN noise, validating the performance of the algorithm across diverse λ values and user/task configurations, clearly demonstrating its adaptability and superiority. Power Distribution Insights: We provide detailed analysis of power distribution patterns across users and sensing tasks under each scheme, offering practical insights into resource utilization and efficiency in ISAC deployments. Power Distribution Insights: We provide detailed analysis of power distribution patterns across users and sensing tasks under each scheme, offering practical insights into resource utilization and efficiency in ISAC deployments. Existing solutions often rely on fixed allocation strategies or shared resource schemes or NOMA-ISAC schemes that do not fully leverage the distinct characteristics of sensing and communication. This paper proposes a novel dynamic resource allocation algorithm for dedicated sensing signals in downlink ISAC that optimizes power and frequency resources to maximize sensing accuracy while maintaining communication throughput. In summary, while dynamic resource allocation methods for ISAC have been studied, most are restricted to beamforming- or power-only optimization, simplified single-user scenarios, or computationally intensive algorithms unsuitable for real-time operation. In contrast, our work introduces a lightweight dynamic framework that jointly allocates power and frequency resources across multiple users and sensing tasks, scales to complex downlink settings, and explicitly balances dual-function objectives through a tunable utility formulation. These contributions distinguish our approach from existing methods and demonstrate its practical potential for ISAC-enabled 6G networks. The rest of this paper is organized as follows, section II outlines the system model, section III explains the proposed dynamic resource allocation algorithm for downlink ISAC, Section IV defines the methodology and the simulation setup, Section V shows and discusses the simulation results. Section VI concludes the paper and section VII discusses the future work. System model Network model Consider a downlink ISAC system where a base station (BS) serves multiple user equipment (UEs) while simultaneously performing sensing tasks, such as radar detection or environmental monitoring. The BS is equipped with a multi-antenna system capable of beamforming for spatial diversity. Let Nu​ be the number of UEs, and Ns​ be the number of sensing tasks. The system operates over a bandwidth B composed of multiple subcarriers. Each UE is allocated a subset of the available subcarriers, and dedicated sensing signals are transmitted in parallel with communication signals. The architecture diagram in Fig. 1 shows the integration of dynamic ISAC resource allocation algorithm into the 6G network stack. It features three core layers: Application Layer – Represents smart services like autonomous driving, smart surveillance, and environment-aware applications. Application Layer – Represents smart services like autonomous driving, smart surveillance, and environment-aware applications. Network Layer – Hosts the dynamic resource allocation engine and control functions to manage communication and sensing trade-offs in real time. Network Layer – Hosts the dynamic resource allocation engine and control functions to manage communication and sensing trade-offs in real time. Physical Layer – Implements the joint downlink ISAC functionality using shared RF chains and dynamically allocated power and frequency resources. Physical Layer – Implements the joint downlink ISAC functionality using shared RF chains and dynamically allocated power and frequency resources. Architecture diagram showing integration into network layers for this ISAC scenario. Signal model Let xu(t) represent the communication signal for user u and xs(t) the dedicated sensing signal. The transmitted signal from the BS can be expressed as: where, x(t): The total received signal at time t. x(t): The total received signal at time t. Nu : Number of desired users. Nu : Number of desired users. Ns​: Number of interferers (or secondary users, depending on the context). Ns​: Number of interferers (or secondary users, depending on the context). Pu​: Power of the u-th desired user. Pu​: Power of the u-th desired user. xu(t): Signal transmitted by the u-th desired user. xu(t): Signal transmitted by the u-th desired user. Ps​: Power of the s-th interferer/source. Ps​: Power of the s-th interferer/source. xs(t): Signal transmitted by the s-th interferer/source. xs(t): Signal transmitted by the s-th interferer/source. where Pu​ and Ps​ are the power allocated to the communication and sensing signals, respectively. The BS allocates subcarriers fu​ and fs​ to the communication and sensing signals based on the channel conditions and the sensing requirements. Channel model The downlink channels for communication and sensing are modeled as frequency-selective Rayleigh fading channels. The received signal at UE u is given by: where, yu(t): The received signal at user u at time t. yu(t): The received signal at user u at time t. hu(t): The channel impulse response for user u — this represents the effect of the wireless channel. hu(t): The channel impulse response for user u — this represents the effect of the wireless channel. *: Convolution operation — indicating that the transmitted signal passes through the channel. *: Convolution operation — indicating that the transmitted signal passes through the channel. xu(t): The transmitted signal from user u. xu(t): The transmitted signal from user u. nu(t): The noise at the receiver for user u, typically modeled as AWGN (Additive White Gaussian Noise). nu(t): The noise at the receiver for user u, typically modeled as AWGN (Additive White Gaussian Noise). The sensing receiver, such as a radar target, receives: where hs(t) is the sensing channel and ns(t) is the AWGN. ys(t): Received signal from source/interferer s at time t. ys(t): Received signal from source/interferer s at time t. hs(t): Channel impulse response for source s. hs(t): Channel impulse response for source s. xs(t): Transmitted signal from source s. xs(t): Transmitted signal from source s. ns(t): Noise at the receiver for source s. ns(t): Noise at the receiver for source s. *: Convolution, representing the effect of the channel on the transmitted signal. *: Convolution, representing the effect of the channel on the transmitted signal. To optimize both communication and sensing, we introduce a dynamic resource allocation algorithm. The algorithm considers the channel state information (CSI) for both communication and sensing and dynamically allocates power and frequency resources to optimize the performance metrics. Dynamic resource allocation algorithm To optimize both communication and sensing, we introduce a dynamic resource allocation algorithm. The algorithm considers the channel state information (CSI) for both communication and sensing and dynamically allocates power and frequency resources to optimize the performance metrics. Objective function The objective is to maximize the sensing accuracy while maintaining a minimum communication throughput for each UE. The formulated problem is non-convex due to the joint optimization of power and frequency variables under interference constraints, making closed-form optimal solutions infeasible. To address this, we propose a low-complexity iterative gradient ascent algorithm that converges to a stationary near-optimal solution suitable for real-time deployment. The optimization problem is formulated as: Maximize: where, Decision Variables:oPu,Ps​: Transmit powers for communication users and sensing nodes.oƒu,ƒs​: Frequencies or subcarriers allocated to users and sensing functions. Decision Variables:oPu,Ps​: Transmit powers for communication users and sensing nodes.oƒu,ƒs​: Frequencies or subcarriers allocated to users and sensing functions. oPu,Ps​: Transmit powers for communication users and sensing nodes. Pu,Ps​: Transmit powers for communication users and sensing nodes. oƒu,ƒs​: Frequencies or subcarriers allocated to users and sensing functions. ƒu,ƒs​: Frequencies or subcarriers allocated to users and sensing functions. Objective: Maximize a weighted sum of:oƒs​(Ps​, ƒs​) represents the sensing accuracy as a function of the sensing power Ps​ and the frequency ƒs allocated to sensing nodes.oƒu​(Pu​, ƒu​) represents the communication rate (or throughput) as a function of the user power Pu​ and the frequency ƒu allocated to communication users.oS and U are the sets of sensing nodes and communication users, respectively. Objective: Maximize a weighted sum of:oƒs​(Ps​, ƒs​) represents the sensing accuracy as a function of the sensing power Ps​ and the frequency ƒs allocated to sensing nodes.oƒu​(Pu​, ƒu​) represents the communication rate (or throughput) as a function of the user power Pu​ and the frequency ƒu allocated to communication users.oS and U are the sets of sensing nodes and communication users, respectively. oƒs​(Ps​, ƒs​) represents the sensing accuracy as a function of the sensing power Ps​ and the frequency ƒs allocated to sensing nodes. ƒs​(Ps​, ƒs​) represents the sensing accuracy as a function of the sensing power Ps​ and the frequency ƒs allocated to sensing nodes. oƒu​(Pu​, ƒu​) represents the communication rate (or throughput) as a function of the user power Pu​ and the frequency ƒu allocated to communication users. ƒu​(Pu​, ƒu​) represents the communication rate (or throughput) as a function of the user power Pu​ and the frequency ƒu allocated to communication users. oS and U are the sets of sensing nodes and communication users, respectively. S and U are the sets of sensing nodes and communication users, respectively. λ: Trade-off parameter (with 0 ≤ λ ≤ 1) controlling the balance between sensing and communication objectives. λ: Trade-off parameter (with 0 ≤ λ ≤ 1) controlling the balance between sensing and communication objectives. The first term in the objective function maximizes the total sensing accuracy, weighted by the factor λ. The first term in the objective function maximizes the total sensing accuracy, weighted by the factor λ. The second term in the objective function maximizes the total communication rate, weighted by (1 − λ), to reflect the goal of jointly enhancing communication throughput and sensing accuracy, with λ controlling the trade-off between the two objectives. The second term in the objective function maximizes the total communication rate, weighted by (1 − λ), to reflect the goal of jointly enhancing communication throughput and sensing accuracy, with λ controlling the trade-off between the two objectives. Subject to: Power constraint: The total power (for both communication and sensing) must not exceed the available power at the base station.5 Power constraint: The total power (for both communication and sensing) must not exceed the available power at the base station.5 Frequency allocation: Frequencies must be allocated such that there is no overlap between sensing and communication functions.6 Frequency allocation: Frequencies must be allocated such that there is no overlap between sensing and communication functions.6 Non-negativity: All powers must be non-negative.7 Non-negativity: All powers must be non-negative.7 where Ptotal​ is the total available power at the BS. Algorithm steps Since OFDM subcarriers are inherently discrete, we employ a greedy assignment step after the continuous power optimization. Each subcarrier is assigned to the user or sensing task that provides the maximum incremental utility, ensuring that the allocation remains consistent with the optimized power distribution. The process is outlined in Algorithm 2. Power and frequency allocation for UEs and sensing tasks. Greedy subcarrier assignment. Detailed algorithm description for dynamic resource allocation in ISAC systems The dynamic resource allocation algorithm proposed for Integrated Sensing and Communication (ISAC) systems is central to optimizing the performance of the system by jointly allocating power and frequency resources to communication users (UEs) and sensing tasks. The gradient ascent–based allocation proposed here is heuristic and guarantees convergence only to a local stationary solution. While it lacks closed-form optimality guarantees, this design achieves the dual goals of low complexity and scalability to multi-user, multi-task ISAC scenarios, making it more practical for real-time 6G applications than computationally intensive convex or learning-based alternatives. To make this process more accessible to readers unfamiliar with the approach, we provide a clearer breakdown of the algorithm, focusing on the application of the gradient ascent method for power updates, the criteria for convergence, and the overall optimization process. Step 1: Initialization Description: In this initial step, the algorithm sets up the power and frequency allocations for both communication users (UEs) and sensing tasks. The allocations are made uniformly, meaning that each UE and sensing task is assigned a fair share of power and frequency at the outset. Description: In this initial step, the algorithm sets up the power and frequency allocations for both communication users (UEs) and sensing tasks. The allocations are made uniformly, meaning that each UE and sensing task is assigned a fair share of power and frequency at the outset. Mathematical Formulation:oLet and denote the initial power allocations for the u-th communication user and s-th sensing task, respectively.oLet and denote the initial frequency allocations for the u-th communication user and s-th sensing task, respectively. Mathematical Formulation:oLet and denote the initial power allocations for the u-th communication user and s-th sensing task, respectively.oLet and denote the initial frequency allocations for the u-th communication user and s-th sensing task, respectively. oLet and denote the initial power allocations for the u-th communication user and s-th sensing task, respectively. Let and denote the initial power allocations for the u-th communication user and s-th sensing task, respectively. oLet and denote the initial frequency allocations for the u-th communication user and s-th sensing task, respectively. Let and denote the initial frequency allocations for the u-th communication user and s-th sensing task, respectively. Uniform Allocation: The power and are initialized based on the total available power and the number of UEs and sensing tasks. Similarly, the frequency allocations and are set based on the total bandwidth and the number of UEs and sensing tasks. Uniform Allocation: The power and are initialized based on the total available power and the number of UEs and sensing tasks. Similarly, the frequency allocations and are set based on the total bandwidth and the number of UEs and sensing tasks. Step 2: Collecting Channel State Information (CSI) Description: The BS collects CSI from UEs and the sensing environment. In practice, CSI for communication links can be acquired through uplink pilots in TDD systems (leveraging reciprocity) or via downlink training and feedback in FDD systems. For sensing tasks, CSI can be estimated from echoes of dedicated probing signals or derived from communication pilots. In this work, we assume perfect instantaneous CSI at the BS to provide an upper-bound evaluation of the proposed scheme. We acknowledge that CSI in real-world ISAC systems is subject to estimation errors, delays, and mobility-induced variations. Extending the framework to handle imperfect CSI through robust optimization or learning-based approaches is an important direction for future work. Description: The BS collects CSI from UEs and the sensing environment. In practice, CSI for communication links can be acquired through uplink pilots in TDD systems (leveraging reciprocity) or via downlink training and feedback in FDD systems. For sensing tasks, CSI can be estimated from echoes of dedicated probing signals or derived from communication pilots. In this work, we assume perfect instantaneous CSI at the BS to provide an upper-bound evaluation of the proposed scheme. We acknowledge that CSI in real-world ISAC systems is subject to estimation errors, delays, and mobility-induced variations. Extending the framework to handle imperfect CSI through robust optimization or learning-based approaches is an important direction for future work. Mathematical Formulation:oLet hu​ represent the CSI for the u-th communication user and hs​ represent the CSI for the s-th sensing task.oThe CSI data is collected from each UE and sensing task, which is used to adjust power and frequency allocations based on the channel conditions. Mathematical Formulation:oLet hu​ represent the CSI for the u-th communication user and hs​ represent the CSI for the s-th sensing task.oThe CSI data is collected from each UE and sensing task, which is used to adjust power and frequency allocations based on the channel conditions. oLet hu​ represent the CSI for the u-th communication user and hs​ represent the CSI for the s-th sensing task. Let hu​ represent the CSI for the u-th communication user and hs​ represent the CSI for the s-th sensing task. oThe CSI data is collected from each UE and sensing task, which is used to adjust power and frequency allocations based on the channel conditions. The CSI data is collected from each UE and sensing task, which is used to adjust power and frequency allocations based on the channel conditions. Complexity: The collection of CSI is assumed to take constant time, O(1), for each user and sensing task. Therefore, the total complexity for this step is O(Nu + Ns), where Nu ​ is the number of UEs and Ns is the number of sensing tasks. Complexity: The collection of CSI is assumed to take constant time, O(1), for each user and sensing task. Therefore, the total complexity for this step is O(Nu + Ns), where Nu ​ is the number of UEs and Ns is the number of sensing tasks. Step 3: Iterative optimization (Gradient Ascent) Description: This is the core step of the algorithm, where the gradient ascent method is applied to iteratively update the power and frequency allocations in order to maximize the utility function. The gradient ascent method is used to iteratively adjust the power and as well as the frequency allocations and to achieve the optimal trade-off between communication and sensing objectives.While the optimization step treats subcarrier allocation as continuous for tractability, in practice subcarriers are discrete. After each iteration, we apply a greedy rounding strategy as mentioned above in algorithm 2, where each subcarrier is assigned to the user or sensing task that yields the maximum utility gain, consistent with the optimized power distribution. This ensures feasible subcarrier allocation in OFDM systems while maintaining low complexity. Description: This is the core step of the algorithm, where the gradient ascent method is applied to iteratively update the power and frequency allocations in order to maximize the utility function. The gradient ascent method is used to iteratively adjust the power and as well as the frequency allocations and to achieve the optimal trade-off between communication and sensing objectives. While the optimization step treats subcarrier allocation as continuous for tractability, in practice subcarriers are discrete. After each iteration, we apply a greedy rounding strategy as mentioned above in algorithm 2, where each subcarrier is assigned to the user or sensing task that yields the maximum utility gain, consistent with the optimized power distribution. This ensures feasible subcarrier allocation in OFDM systems while maintaining low complexity. Mathematical Formulation: The utility function U(Pu, Ps, ƒu, ƒs) combines the communication rate and sensing accuracy as follows:8where:o is the communication rate for the u-th user, calculated using Shannon’s capacity formula:9where B is the bandwidth, Pu​ is the power allocated to the u-th user, hu​ is the CSI, and N0​ is the noise power, Pj​ is the power allocated to the j-th user, where j ≠ u represents the power of all other users interfering with the u-th user’s signal. hj​ is the CSI of the j-th user.oMSE(Ps,fs) is the mean squared error (MSE) of the sensing accuracy for the s-th task, which quantifies the difference between the estimated and true environmental data. The sensing performance is quantified by the mean square error (MSE) of target parameter estimation. Following radar estimation theory, the sensing MSE is inversely proportional to the effective sensing SNR. For a given sensing task s, we model:10where ​ is the AWGN variance, Ps(f) is the allocated power on subcarrier f, and hs(f) is the sensing channel coefficient. This formulation explicitly links sensing accuracy to power and frequency allocation, thereby aligning the sensing objective with the optimization framework. Mathematical Formulation: The utility function U(Pu, Ps, ƒu, ƒs) combines the communication rate and sensing accuracy as follows:8 where:o is the communication rate for the u-th user, calculated using Shannon’s capacity formula:9where B is the bandwidth, Pu​ is the power allocated to the u-th user, hu​ is the CSI, and N0​ is the noise power, Pj​ is the power allocated to the j-th user, where j ≠ u represents the power of all other users interfering with the u-th user’s signal. hj​ is the CSI of the j-th user.oMSE(Ps,fs) is the mean squared error (MSE) of the sensing accuracy for the s-th task, which quantifies the difference between the estimated and true environmental data. The sensing performance is quantified by the mean square error (MSE) of target parameter estimation. Following radar estimation theory, the sensing MSE is inversely proportional to the effective sensing SNR. For a given sensing task s, we model:10where ​ is the AWGN variance, Ps(f) is the allocated power on subcarrier f, and hs(f) is the sensing channel coefficient. This formulation explicitly links sensing accuracy to power and frequency allocation, thereby aligning the sensing objective with the optimization framework. o is the communication rate for the u-th user, calculated using Shannon’s capacity formula:9where B is the bandwidth, Pu​ is the power allocated to the u-th user, hu​ is the CSI, and N0​ is the noise power, Pj​ is the power allocated to the j-th user, where j ≠ u represents the power of all other users interfering with the u-th user’s signal. hj​ is the CSI of the j-th user. is the communication rate for the u-th user, calculated using Shannon’s capacity formula:9where B is the bandwidth, Pu​ is the power allocated to the u-th user, hu​ is the CSI, and N0​ is the noise power, Pj​ is the power allocated to the j-th user, where j ≠ u represents the power of all other users interfering with the u-th user’s signal. hj​ is the CSI of the j-th user. oMSE(Ps,fs) is the mean squared error (MSE) of the sensing accuracy for the s-th task, which quantifies the difference between the estimated and true environmental data. The sensing performance is quantified by the mean square error (MSE) of target parameter estimation. Following radar estimation theory, the sensing MSE is inversely proportional to the effective sensing SNR. For a given sensing task s, we model:10where ​ is the AWGN variance, Ps(f) is the allocated power on subcarrier f, and hs(f) is the sensing channel coefficient. This formulation explicitly links sensing accuracy to power and frequency allocation, thereby aligning the sensing objective with the optimization framework. MSE(Ps,fs) is the mean squared error (MSE) of the sensing accuracy for the s-th task, which quantifies the difference between the estimated and true environmental data. The sensing performance is quantified by the mean square error (MSE) of target parameter estimation. Following radar estimation theory, the sensing MSE is inversely proportional to the effective sensing SNR. For a given sensing task s, we model:10where ​ is the AWGN variance, Ps(f) is the allocated power on subcarrier f, and hs(f) is the sensing channel coefficient. This formulation explicitly links sensing accuracy to power and frequency allocation, thereby aligning the sensing objective with the optimization framework. Gradient Ascent Method:oIn each iteration, the algorithm computes the gradient of the utility function with respect to the power and frequency allocations for each user and sensing task.oThe gradient of the utility function with respect to power for the u-th user is given by:11oThe gradient with respect to the power for the s-th sensing task is:12where ​ represents the optimal power for the sensing task based on previous iterations.oThe gradient with respect to frequency fu​ for the u-th user is computed by taking the derivative of the communication rate with respect to frequency. Similarly, the gradient with respect to fs​ for the s-th sensing task is computed by taking the derivative of the sensing MSE with respect to frequency. Gradient Ascent Method:oIn each iteration, the algorithm computes the gradient of the utility function with respect to the power and frequency allocations for each user and sensing task.oThe gradient of the utility function with respect to power for the u-th user is given by:11oThe gradient with respect to the power for the s-th sensing task is:12where ​ represents the optimal power for the sensing task based on previous iterations.oThe gradient with respect to frequency fu​ for the u-th user is computed by taking the derivative of the communication rate with respect to frequency. Similarly, the gradient with respect to fs​ for the s-th sensing task is computed by taking the derivative of the sensing MSE with respect to frequency. oIn each iteration, the algorithm computes the gradient of the utility function with respect to the power and frequency allocations for each user and sensing task. In each iteration, the algorithm computes the gradient of the utility function with respect to the power and frequency allocations for each user and sensing task. oThe gradient of the utility function with respect to power for the u-th user is given by:11 The gradient of the utility function with respect to power for the u-th user is given by:11 oThe gradient with respect to the power for the s-th sensing task is:12where ​ represents the optimal power for the sensing task based on previous iterations. The gradient with respect to the power for the s-th sensing task is:12where ​ represents the optimal power for the sensing task based on previous iterations. oThe gradient with respect to frequency fu​ for the u-th user is computed by taking the derivative of the communication rate with respect to frequency. Similarly, the gradient with respect to fs​ for the s-th sensing task is computed by taking the derivative of the sensing MSE with respect to frequency. The gradient with respect to frequency fu​ for the u-th user is computed by taking the derivative of the communication rate with respect to frequency. Similarly, the gradient with respect to fs​ for the s-th sensing task is computed by taking the derivative of the sensing MSE with respect to frequency. Update Rule: After computing the gradients, the power and frequency allocations are updated using the following rule:13141516where η is the step size or learning rate, which controls the size of the updates in each iteration. The step size is a critical parameter that affects convergence speed and stability. Update Rule: After computing the gradients, the power and frequency allocations are updated using the following rule:13141516where η is the step size or learning rate, which controls the size of the updates in each iteration. The step size is a critical parameter that affects convergence speed and stability. Step 4: Convergence check Description: After each iteration, the algorithm checks whether the change in the utility function is below a predefined threshold, indicating that the algorithm has converged to a near-optimal solution. Description: After each iteration, the algorithm checks whether the change in the utility function is below a predefined threshold, indicating that the algorithm has converged to a near-optimal solution. Mathematical Formulation: The convergence criterion is given by:17where ε is a small positive number representing the convergence threshold. If the change in the utility function is below this threshold, the algorithm stops; otherwise, it proceeds to the next iteration. Mathematical Formulation: The convergence criterion is given by:17where ε is a small positive number representing the convergence threshold. If the change in the utility function is below this threshold, the algorithm stops; otherwise, it proceeds to the next iteration. Step 5: Final Output Description: Once the convergence criterion is met, the algorithm outputs the final power and frequency allocations for the communication users and sensing tasks. Description: Once the convergence criterion is met, the algorithm outputs the final power and frequency allocations for the communication users and sensing tasks. Final Output:The algorithm outputs the optimized power Pu​ and frequency fu​ allocations for each UE, and the optimized power Ps​ and frequency fs​ allocations for each sensing task. These allocations represent the optimal trade-off between communication performance and sensing accuracy, tailored to the current network conditions and sensing requirements. Final Output: The algorithm outputs the optimized power Pu​ and frequency fu​ allocations for each UE, and the optimized power Ps​ and frequency fs​ allocations for each sensing task. These allocations represent the optimal trade-off between communication performance and sensing accuracy, tailored to the current network conditions and sensing requirements. Summary of the optimization process The dynamic resource allocation algorithm uses a gradient ascent approach to iteratively optimize the power and frequency allocations for communication users and sensing tasks. The key steps involve: Initializing the allocations uniformly. Initializing the allocations uniformly. Collecting CSI from users and sensing tasks. Collecting CSI from users and sensing tasks. Iteratively updating the allocations using the gradient ascent method. Iteratively updating the allocations using the gradient ascent method. Checking for convergence based on a predefined threshold. Checking for convergence based on a predefined threshold. Outputting the final optimized allocations once the algorithm converges. Outputting the final optimized allocations once the algorithm converges. By dynamically adjusting the power and frequency allocations based on real-time channel conditions and task-specific requirements, the algorithm efficiently balances the competing demands of communication throughput and sensing accuracy, making it suitable for adaptive and intelligent 6G networks. Complexity analysis of the dynamic resource allocation algorithm for ISAC systems In this section, we present a detailed analysis of the time and space complexity associated with the dynamic resource allocation algorithm proposed for Integrated Sensing and Communication (ISAC) systems. Given the importance of real-time adaptation in 6G applications, understanding the computational overhead of this algorithm is critical for evaluating its feasibility in resource-constrained environments. Time complexity The dynamic resource allocation algorithm involves several key steps, including: Initialization: In the initialization step, power and frequency are allocated uniformly across communication users (UEs) and sensing tasks. This step has a time complexity of O(Nu + Ns), where Nu​ is the number of UEs and Ns is the number of sensing tasks. This step involves setting initial values for each task, which is a linear operation. Initialization: In the initialization step, power and frequency are allocated uniformly across communication users (UEs) and sensing tasks. This step has a time complexity of O(Nu + Ns), where Nu​ is the number of UEs and Ns is the number of sensing tasks. This step involves setting initial values for each task, which is a linear operation. CSI Collection: The algorithm collects Channel State Information (CSI) from the UEs and sensing tasks. Assuming that the CSI for each user and sensing task can be collected in constant time O(1), the overall complexity for this step is O(Nu + Ns). CSI Collection: The algorithm collects Channel State Information (CSI) from the UEs and sensing tasks. Assuming that the CSI for each user and sensing task can be collected in constant time O(1), the overall complexity for this step is O(Nu + Ns). Iterative Optimization (Gradient Ascent): This is the core of the dynamic algorithm and involves the iterative optimization of the resource allocation. The optimization process is carried out using a gradient ascent method, which computes the gradient of the utility function with respect to power and frequency allocation, and then updates the allocation values.oGradient Calculation: In each iteration, the algorithm computes the gradient for Nu UEs and Ns​ sensing tasks. The gradient calculation for a single user or sensing task involves O(1) operations. Thus, for all users and sensing tasks, the gradient calculation has a complexity of O(Nu + Ns).oUpdating Allocations: The gradient ascent algorithm updates the power and frequency allocations for both communication and sensing resources. The update for each user and sensing task is a constant-time operation, so the total update complexity per iteration is also O(Nu + Ns).oConvergence Check: After each iteration, the algorithm checks if the change in the objective function is below a predefined threshold to determine if the optimization has converged. This check involves a comparison that takes constant time, O(1). However, since this step needs to be performed after each iteration, the overall time complexity is O(T), where T is the maximum number of iterations before convergence. Iterative Optimization (Gradient Ascent): This is the core of the dynamic algorithm and involves the iterative optimization of the resource allocation. The optimization process is carried out using a gradient ascent method, which computes the gradient of the utility function with respect to power and frequency allocation, and then updates the allocation values.oGradient Calculation: In each iteration, the algorithm computes the gradient for Nu UEs and Ns​ sensing tasks. The gradient calculation for a single user or sensing task involves O(1) operations. Thus, for all users and sensing tasks, the gradient calculation has a complexity of O(Nu + Ns).oUpdating Allocations: The gradient ascent algorithm updates the power and frequency allocations for both communication and sensing resources. The update for each user and sensing task is a constant-time operation, so the total update complexity per iteration is also O(Nu + Ns).oConvergence Check: After each iteration, the algorithm checks if the change in the objective function is below a predefined threshold to determine if the optimization has converged. This check involves a comparison that takes constant time, O(1). However, since this step needs to be performed after each iteration, the overall time complexity is O(T), where T is the maximum number of iterations before convergence. oGradient Calculation: In each iteration, the algorithm computes the gradient for Nu UEs and Ns​ sensing tasks. The gradient calculation for a single user or sensing task involves O(1) operations. Thus, for all users and sensing tasks, the gradient calculation has a complexity of O(Nu + Ns). Gradient Calculation: In each iteration, the algorithm computes the gradient for Nu UEs and Ns​ sensing tasks. The gradient calculation for a single user or sensing task involves O(1) operations. Thus, for all users and sensing tasks, the gradient calculation has a complexity of O(Nu + Ns). oUpdating Allocations: The gradient ascent algorithm updates the power and frequency allocations for both communication and sensing resources. The update for each user and sensing task is a constant-time operation, so the total update complexity per iteration is also O(Nu + Ns). Updating Allocations: The gradient ascent algorithm updates the power and frequency allocations for both communication and sensing resources. The update for each user and sensing task is a constant-time operation, so the total update complexity per iteration is also O(Nu + Ns). oConvergence Check: After each iteration, the algorithm checks if the change in the objective function is below a predefined threshold to determine if the optimization has converged. This check involves a comparison that takes constant time, O(1). However, since this step needs to be performed after each iteration, the overall time complexity is O(T), where T is the maximum number of iterations before convergence. Convergence Check: After each iteration, the algorithm checks if the change in the objective function is below a predefined threshold to determine if the optimization has converged. This check involves a comparison that takes constant time, O(1). However, since this step needs to be performed after each iteration, the overall time complexity is O(T), where T is the maximum number of iterations before convergence. Therefore, the total time complexity for the iterative optimization process, considering the number of iterations T, is O(T⋅(Nu + Ns)). The total time complexity for the entire algorithm, including initialization, CSI collection, and iterative optimization, is: Here, T depends on the convergence criteria and the specific characteristics of the wireless environment (e.g., the channel conditions and sensing requirements). Space complexity The space complexity is primarily determined by the storage requirements for the following components: Power and Frequency Allocations: The algorithm needs to store the power and frequency allocations for each UE and sensing task. This requires O(Nu + Ns) space. Power and Frequency Allocations: The algorithm needs to store the power and frequency allocations for each UE and sensing task. This requires O(Nu + Ns) space. CSI Data: The algorithm needs to store the CSI for each user and sensing task. Assuming that the CSI data for each UE and sensing task can be stored in constant space, the total space required to store the CSI is O(Nu + Ns). CSI Data: The algorithm needs to store the CSI for each user and sensing task. Assuming that the CSI data for each UE and sensing task can be stored in constant space, the total space required to store the CSI is O(Nu + Ns). Optimization Variables: The gradient ascent algorithm requires storage for the gradients and updated allocations for each iteration. Since the gradients and allocations for each user and sensing task are stored, this requires O(Nu + Ns) space. Optimization Variables: The gradient ascent algorithm requires storage for the gradients and updated allocations for each iteration. Since the gradients and allocations for each user and sensing task are stored, this requires O(Nu + Ns) space. Thus, the space complexity of the dynamic resource allocation algorithm is dominated by the storage required for power, frequency allocations, CSI data, and optimization variables. Therefore, the space complexity is: O(Nu + Ns). Scalability considerations for 6G applications In the context of 6G applications, the dynamic resource allocation algorithm’s scalability is an important consideration. The algorithm’s time complexity of O(T⋅(Nu + Ns)) suggests that the computational overhead grows linearly with the number of UEs and sensing tasks. However, the number of UEs and sensing tasks in a real-world 6G deployment could be very large, especially in scenarios such as autonomous driving, smart cities, or large-scale environmental monitoring. Time Complexity and Real-Time Adaptation: The convergence time T plays a critical role in real-time adaptation. If T is too large, the algorithm may not adapt quickly enough to dynamic channel conditions. Techniques such as adaptive step-size methods in gradient ascent or leveraging machine learning-based methods to predict the optimal allocation could be considered to reduce the number of iterations and speed up convergence. Time Complexity and Real-Time Adaptation: The convergence time T plays a critical role in real-time adaptation. If T is too large, the algorithm may not adapt quickly enough to dynamic channel conditions. Techniques such as adaptive step-size methods in gradient ascent or leveraging machine learning-based methods to predict the optimal allocation could be considered to reduce the number of iterations and speed up convergence. Space Complexity and Edge Computing: The space complexity O(Nu + Ns) indicates that the algorithm’s memory requirements grow with the number of UEs and sensing tasks. In 6G systems, edge computing could be leveraged to distribute the computation and memory requirements across edge devices, ensuring that the algorithm remains scalable as the number of connected devices increases. Space Complexity and Edge Computing: The space complexity O(Nu + Ns) indicates that the algorithm’s memory requirements grow with the number of UEs and sensing tasks. In 6G systems, edge computing could be leveraged to distribute the computation and memory requirements across edge devices, ensuring that the algorithm remains scalable as the number of connected devices increases. Optimizations for real-time applications To ensure that the dynamic resource allocation algorithm can be efficiently deployed in 6G networks, several optimization techniques could be considered: Parallelization: The gradient computation and update steps for each user and sensing task are independent and could be parallelized to reduce the time complexity, especially in large-scale networks. Parallelization: The gradient computation and update steps for each user and sensing task are independent and could be parallelized to reduce the time complexity, especially in large-scale networks. Approximate Algorithms: Instead of solving the optimization problem to exact precision, approximate methods could be used to speed up convergence. For instance, heuristic or greedy algorithms could be used for initial allocations, followed by refinement using the dynamic approach. Approximate Algorithms: Instead of solving the optimization problem to exact precision, approximate methods could be used to speed up convergence. For instance, heuristic or greedy algorithms could be used for initial allocations, followed by refinement using the dynamic approach. Online Learning: Machine learning techniques, such as reinforcement learning or deep learning, could be used to adaptively learn the optimal resource allocation policy based on real-time network conditions, further reducing the time complexity associated with the iterative optimization process. Online Learning: Machine learning techniques, such as reinforcement learning or deep learning, could be used to adaptively learn the optimal resource allocation policy based on real-time network conditions, further reducing the time complexity associated with the iterative optimization process. The dynamic resource allocation algorithm for ISAC systems has a time complexity of O(T⋅(Nu + Ns)) and a space complexity of O(Nu + Ns). While these complexities are linear with respect to the number of UEs and sensing tasks, optimizations such as parallelization, approximate solutions, and machine learning-based adaptation could help make the algorithm more efficient and scalable for real-time 6G applications. Additionally, leveraging edge computing for distributed computation could help alleviate the memory and computational burden, ensuring that the algorithm can handle the large-scale, resource-constrained environments typical of 6G networks. Note that the proposed iterative reallocation framework assumes quasi-static or slowly varying channel conditions, such that the acquired CSI remains valid throughout the optimization process. In scenarios with fast channel variations, convergence delays and CSI acquisition overhead may limit the practicality of the algorithm. Addressing such fast-varying environments is left as an important direction for future research, where predictive CSI estimation and low-complexity update strategies could be integrated into the framework. Methodology Simulation comparison against benchmark NOMA-ISAC approaches To ensure a fair comparison, we consider the following simulation scenario in comparison with other works: 10 downlink user equipments (UEs) and 3 sensing targets, with 1 W total transmit power spread across 100 subcarriers in a frequency-selective Rayleigh fading channel and AWGN noise of σ2 = 0.01. Table 1 presents the all the simulation metrics. Under this setup, we compare the proposed dynamic joint power & frequency allocation algorithm against state-of-the-art NOMA-ISAC schemes 15,17–20 on key metrics – sensing accuracy, communication throughput, and combined utility. The baseline for improvement calculations is a conventional static allocation (e.g. equal or fixed splitting of power/subcarriers). Table 2 (at the end of section V) summarizes the quantitative performance outcomes. Below, we discuss each metric in turn, highlighting both quantitative results and qualitative insights. Table 1Simulation metrics.MetricDescriptionSensing AccuracyUse MSE of target localization (lower = better)Communication RateShannon capacity:Total UtilityWeighted sum: Accuracy + λ . RatePower DistributionDivision of power among UEs/ sensing tasks. It is part of the decision variables in our dynamic resource allocation algorithm:Optimizing how to allocate Pu and Ps across UEs and sensing Simulation metrics. Use MSE of target localization (lower = better) Division of power among UEs/ sensing tasks. It is part of the decision variables in our dynamic resource allocation algorithm: Optimizing how to allocate Pu and Ps across UEs and sensing Table 2Performance comparison of proposed dynamic allocation vs. state-of-the-art NOMA-ISAC schemes (Downlink Scenario) ​Scheme (Reference)Sensing Performance (Accuracy/MSE)Communication Performance (Rate/Throughput)Overall Utility / Trade-offNotes / HighlightsProposed Dynamic Allocation (Power + Frequency allocation for 10 UEs, 3 targets)MSE reduced by 17.65% (vs. static baseline) Significantly lower sensing error due to adaptive resource tuningThroughput increased by 22.86% (vs. baseline) ​Higher sum rate by exploiting multi-user diversity on 100 subcarriersTotal utility + 13.49% (vs. baseline)Balanced S&C improvement, expanding the Pareto frontierDynamic per-subcarrier power allocation. Achieves joint gains in sensing and comm concurrently. Uses CSI and task demands for real-time optimizationConventional Static Allocation (Baseline: equal or fixed resource split)Baseline MSE (no improvement). Sensing limited by interference and no adaptationBaseline sum rate. Many resources under-utilized (no dynamic optimization)Baseline utility (treat as 0% improvement). Rigid trade-off, far from optimalOrthogonal or fixed allocation. Serves as reference; suffers from inefficient resource use and suboptimal S&C trade-offNOMA-ISAC BeamformingImproved sensing power vs OMA: achieves near-ideal radar power in high-load cases​(MSE not given, but higher reflected power implies lower MSE.)Higher sum rate vs conventional ISAC under interference​E.g. at K = 6, NOMA-ISAC delivers comm throughput close to dual-function ideal, far above orthogonal scheme​Better weighted sum utility achieved in overloaded regime​NOMA-ISAC outperforms conventional ISAC on the S&C trade-off curve (especially when channels are correlated or system is overloaded)Joint beamforming optimization (SCA). Focused on MIMO gains (ULA with 4 antennas). Did not consider subcarrier allocation (single wideband channel). Shows NOMA enables more users and pushes trade-off boundarySecure Precoding (NOMA-ISAC) No sensing degradation despite security enforcement​Sensing MSE maintained at baseline level (or improved slightly) while serving more users securelyHigher secure comm rate than TDMA-based ISAC​NOMA scheme improves secrecy sum-rate (SSR) compared to orthogonal access, thanks to parallel serving of usersCombined S&C utility not explicitly given, but scheme ensures secure comm + robust sensing simultaneouslySecure MIMO precoding design. Demonstrated NOMA can guarantee security and serve multiple UEs without sacrificing sensing quality. Baseline (TDMA) could not achieve as high SSR for same sensing performanceRIS-aided NOMA-ISACStronger sensing signals: Higher minimum beampattern gain to all targets with NOMA​This improves worst-case target SNR (reducing MSE for the weakest target)Interference mitigated by NOMA, enabling more UEs to be served under radar constraints​Communication requirements are met alongside enhanced sensing (exact rate gain not stated, but no throughput loss with NOMA)No single utility metric; design maximizes radar illumination while satisfying comm SINR. Achieves better S&C simultaneity than benchmarks​Joint active beamforming, power allocation, and RIS phase design. Focus on maximizing sensing coverage (min beampattern). Found NOMA an efficient way to reduce inter-user interference in both overloaded & underloaded cases​. Comm performance maintained as a constraintSensing-Efficient SchedulingHigher sensing accuracy vs benchmark schemes​By scheduling sensing tasks intelligently, the scheme lowers estimation error for prioritized targets (relative to static or round-robin scheduling)Communication performance preserved or improved​Despite allocating more resources to sensing tasks at times, overall throughput remains on par with (or above) baseline due to NOMA serving users concurrentlyImplicitly achieves a better trade-off: both sensing and comm metrics see gains over baseline​. (No explicit “utility” value given, but dual improvement implies higher overall utility)Joint sensing-task scheduling and transmit beamforming. Prioritizes important sensing tasks without compromising user QoS. Shows NOMA-aided ISAC can outperform static allocation in both domains​. Scenario: multi-target sensing with beamforming; did not allocate subcarriers (single carrier assumed)Semi-ISaC (OMA vs NOMA) Ergodic Radar Estimation Rate similar or slightly improved with NOMA(Semi-ISaC dedicates part spectrum to radar, so sensing largely unaffected; NOMA allows more comm users without hurting radar’s portion.)NOMA yields higher channel capacity than OMA​For the communication portion, two-user NOMA Semi-ISaC achieves better outage and ergodic rates than one-user-per-band OMA Semi-ISAC​Overall, NOMA-Semi-ISaC > OMA-Semi-ISaC in spectral efficiency. The joint S&C utility is improved because unused resources are minimized (NOMA serves an extra user on ISAC band)Partial spectrum sharing approach. Uplink single-target scenario (analysis + simulation). Highlights that introducing NOMA increases total system capacity without degrading sensing (which uses its exclusive band)​. Trade-off is less dynamic here (split fixed), but NOMA still shows clear comm gainSIC Decoding Strategies (Analysis-focused) Showed sensing rate (SR) in NOMA-ISAC can surpass orthogonal schemesDifferent SIC orders yield same SR at high SNR, with differences in power offset​Communication rate (CR) similarly benefits from ISAC integration​The derived SR–CR region for NOMA-ISAC entirely covers that of an FDSAC (separated) system, confirming comm performance is never worse and often better with integration​Identified that ISAC provides more DoF than FDSACNo single metric, but overall performance region is expanded (better trade-off frontier)Theoretical performance analysis (two SIC order cases compared). Emphasizes that proper interference cancellation in NOMA-ISAC can achieve joint optimality – reaching points unattainable by static division of spectrum​. Validates conceptually the utility gains we realize in practiceDownlink NOMA-ISAC (single pair)Basic downlink NOMA-ISAC framework (single UE + single target) – likely showed lower sensing error than separate allocation for that simple case (details not publicly available)Likely higher comm rate than separate baseline in single-pair caseNo multi-user utility considered (single pair scenario). Provided initial validation that NOMA-ISAC is beneficial, but not a full algorithmAnalyzed performance limits for one communication UT and one sensing task (SISO). Demonstrated fundamental S&C trade-off, but did not involve dynamic allocation or multiple users Performance comparison of proposed dynamic allocation vs. state-of-the-art NOMA-ISAC schemes (Downlink Scenario) ​ Throughput increased by 22.86% (vs. baseline) ​ Higher sum rate by exploiting multi-user diversity on 100 subcarriers Total utility + 13.49% (vs. baseline) Balanced S&C improvement, expanding the Pareto frontier Improved sensing power vs OMA: achieves near-ideal radar power in high-load cases​ (MSE not given, but higher reflected power implies lower MSE.) Higher sum rate vs conventional ISAC under interference​ E.g. at K = 6, NOMA-ISAC delivers comm throughput close to dual-function ideal, far above orthogonal scheme​ Better weighted sum utility achieved in overloaded regime​ NOMA-ISAC outperforms conventional ISAC on the S&C trade-off curve (especially when channels are correlated or system is overloaded) No sensing degradation despite security enforcement​ Sensing MSE maintained at baseline level (or improved slightly) while serving more users securely Higher secure comm rate than TDMA-based ISAC​ NOMA scheme improves secrecy sum-rate (SSR) compared to orthogonal access, thanks to parallel serving of users Stronger sensing signals: Higher minimum beampattern gain to all targets with NOMA​ This improves worst-case target SNR (reducing MSE for the weakest target) Interference mitigated by NOMA, enabling more UEs to be served under radar constraints​ Communication requirements are met alongside enhanced sensing (exact rate gain not stated, but no throughput loss with NOMA) Higher sensing accuracy vs benchmark schemes​ By scheduling sensing tasks intelligently, the scheme lowers estimation error for prioritized targets (relative to static or round-robin scheduling) Communication performance preserved or improved​ Despite allocating more resources to sensing tasks at times, overall throughput remains on par with (or above) baseline due to NOMA serving users concurrently Ergodic Radar Estimation Rate similar or slightly improved with NOMA (Semi-ISaC dedicates part spectrum to radar, so sensing largely unaffected; NOMA allows more comm users without hurting radar’s portion.) NOMA yields higher channel capacity than OMA​ For the communication portion, two-user NOMA Semi-ISaC achieves better outage and ergodic rates than one-user-per-band OMA Semi-ISAC​ (Analysis-focused) Showed sensing rate (SR) in NOMA-ISAC can surpass orthogonal schemes Different SIC orders yield same SR at high SNR, with differences in power offset​ Communication rate (CR) similarly benefits from ISAC integration​ The derived SR–CR region for NOMA-ISAC entirely covers that of an FDSAC (separated) system, confirming comm performance is never worse and often better with integration​ Identified that ISAC provides more DoF than FDSAC No single metric, but overall performance region is expanded (better trade-off frontier) In implementing the NOMA-ISAC baseline, we adopted the conventional downlink NOMA model with two-user superposition coding per subcarrier. Power-domain multiplexing was applied, and successive interference cancellation (SIC) was performed at the stronger user according to channel conditions. For sensing tasks, the echo of the superimposed signal was used to compute MSE. Both our orthogonal allocation framework and the NOMA baseline were simulated under identical system parameters, including total transmit power, fading channels, noise variance, and number of subcarriers. This ensures a fair and credible comparison despite the difference in multiple access paradigms. Simulation setup The simulation is conducted using MATLAB with a multi-user ISAC downlink system. The number of UEs is set to Nu = 10, and the number of sensing tasks is Ns = 3. The total available bandwidth is divided into 100 subcarriers, and the total transmission power is set to 1 W. The channel model follows a frequency-selective Rayleigh fading model, and the AWGN variance is set to σ2 = 0.01. Performance metrics Sensing Accuracy: Measured by the mean squared error (MSE) of the estimated target position. Sensing Accuracy: Measured by the mean squared error (MSE) of the estimated target position. Communication Rate: Measured by the achievable data rate for each UE, calculated using Shannon’s formula. Communication Rate: Measured by the achievable data rate for each UE, calculated using Shannon’s formula. Total Utility: Total utility reflects how well the chosen power/frequency allocations perform, not the allocation itself. Total Utility: Total utility reflects how well the chosen power/frequency allocations perform, not the allocation itself. Power Distribution: This refers to how the total available transmit power (e.g., 1W in our setup) is divided among users (UEs) and sensing tasks. Power Distribution: This refers to how the total available transmit power (e.g., 1W in our setup) is divided among users (UEs) and sensing tasks. Simulation results – comparison against benchmark NOMA-ISAC approaches. Sensing accuracy (MSE Performance) Proposed dynamic allocation Our algorithm achieves a 17.65% reduction in sensing mean-squared error (MSE) compared to static resource allocation (at balanced weighting). This indicates a significantly more accurate sensing performance (lower estimation error) due to adaptive resource tuning. The dynamic scheme’s ability to cut MSE by ~ 17.7% underscores its effectiveness in prioritizing sensing when needed, without wasting power on unneeded frequencies. NOMA-ISAC benchmarks Prior NOMA-ISAC works also report improved sensing performance, though often measured in indirect ways (e.g. radar beampattern gain or estimation bounds rather than explicit MSE). For example, Zuo et al. 18 incorporate a RIS and NOMA to boost radar sensing power – their NOMA-enhanced RIS-ISAC design achieves a higher minimum beampattern gain toward targets than a comparable non-NOMA scheme. This improvement in beam pattern gain translates to stronger reflected signals and thus lower sensing error (better MSE) for all targets in overloaded scenarios. Similarly, Dou et al. 19 focus on sensing-efficient scheduling: by smartly selecting which targets to sense and optimizing beamforming, they ensure better sensing performance than baseline scheduling strategies​. While 19 does not directly quote an MSE percentage, their numerical results verify a clear sensing accuracy advantage over conventional equal allocation (fewer missed targets and lower estimation error)​. Ouyang et al. 21 analytically show that a fully integrated NOMA-ISAC system can achieve strictly superior sensing capability relative to orthogonal resource use – in fact, the achievable sensing-vs-communication region under NOMA-ISAC entirely covers that of a separated (FDSAC) system​. In other words, for any required communication rate, the joint ISAC approach can attain a sensing accuracy at least as good as (and often better than) a frequency-division scheme. This aligns with our findings: integrating resources (especially via dynamic allocation) yields tangible sensing accuracy gains. Notably, Yang et al. 17 demonstrate that even when adding security constraints, a NOMA-aided ISAC precoding design can “ensure the sensing performance” – i.e. maintain low sensing MSE – while serving more users securely. Comparison As can be seen in Fig. 2, the proposed scheme’s ~ 17.7% MSE reduction is competitive with and in many cases exceeds the sensing improvements reported in literature. Prior works confirm that NOMA techniques help reduce sensing error (e.g. via higher radar SNR or information rate), but they often focus on specific scenarios (like two-user cases or single-target beamforming). Our algorithm’s advantage is that it generalizes this benefit to a broader multi-user, multi-target setting with dynamic frequency allocation. In summary, the dynamic power/frequency allocation offers sharper sensing accuracy gains than static or orthogonal schemes, and performs on par with the best NOMA-ISAC designs in the literature – all while simultaneously supporting many UEs and tasks. This is evidenced by its 17.65% MSE improvement, achieved concurrently with high communication load as can be seen in Fig. 2, whereas earlier works typically reported qualitative improvements (e.g. improved beampattern or lower Cramér–Rao bound) rather than a double-digit percentage drop in MSE. Sensing MSE vs λ. Communication throughput (Rate performance) Proposed dynamic allocation In terms of communication service, our algorithm provides a 22.86% increase in total downlink rate over the static allocation benchmarks (again at balanced sensing/communication priority). This nearly 23% throughput gain comes from intelligent power and subcarrier assignment that maximizes each UE’s channel conditions. Importantly, this improvement is achieved without sacrificing sensing – the algorithm finds extra degrees of freedom in frequency domain to boost sum rate while still meeting sensing needs. The dynamic scheme effectively mitigates intra-system interference and ensures more users can be served at high rates. NOMA-ISAC benchmarks Many state-of-the-art works have highlighted NOMA’s spectral efficiency advantage for ISAC, which translates to higher communication capacity or rate. Zhang et al. 20 introduced a Semi-ISAC framework (part of the bandwidth dedicated to either sensing or communication, rest for ISAC) and analytically showed that NOMA-based Semi-ISAC achieves higher channel capacity than OMA-based Semi-ISAC​. In simulations, they observed that using NOMA to multiplex two users on the ISAC portion yields a notable boost in ergodic sum rate compared to strictly orthogonal access​. Likewise, Mu et al. 16 demonstrated that a NOMA beamforming design can deliver superior throughput under interference-limited conditions: their NOMA-ISAC scheme outperforms a conventional (orthogonal) ISAC in high-load scenarios, achieving a better communication–sensing trade-off (much higher sum rate for a given sensing power level in overloaded cases)​. In fact, at heavy load (K = 6 users in 16), the NOMA approach’s throughput comes very close to an “ideal” case (as if sensing and communication had no trade-off), whereas the conventional scheme had to severely curtail throughput to accommodate sensing​. Yang et al. 17 report that NOMA can also improve secure communication rate – their secure precoding scheme yielded higher secrecy sum-rate (SSR) than a traditional TDMA-based ISAC, while still meeting sensing requirements. This implies NOMA allowed them to serve multiple users simultaneously (improving spectral usage) rather than orthogonal time slots, boosting the effective data rate. Similarly, Dou et al. 19 found that with optimized scheduling, communication throughput is maintained or improved alongside sensing gains​ – their approach avoids the communication-rate degradation that naive sensing-first strategies might cause. Overall, these benchmarks consistently show 15–25% (approx.) throughput improvements when moving from orthogonal or static allocations to NOMA-powered schemes (depending on scenario). For instance, the semi-ISaC study 20 effectively noted NOMA provides a higher capacity region, and 16 saw the NOMA scheme supporting roughly ~ 2 × the sum-rate of the orthogonal scheme under certain conditions (e.g. when spatial interference is high). Comparison: Fig. 3 shows the comparison. The 22.86% rate enhancement achieved by our dynamic algorithm is in line with or better than the gains in prior NOMA-ISAC studies. Notably, our approach realizes this improvement together with a sensing boost, whereas some literature solutions emphasize communication gains predominantly. Communication Rate vs λ. The ability to adaptively allocate subcarriers means our scheme can opportunistically use favorable frequencies for data transmission, something most beamforming-centric NOMA-ISAC works (which often assume a single channel or single-carrier) could not exploit. Consequently, the proposed method can match state-of-the-art sum rate performance (e.g. reaching the high throughput region of NOMA-ISAC observed in 16 and 20) while still reserving resources for sensing tasks. It essentially combines the multi-user efficiency of NOMA (serving 10 UEs concurrently) with fine-grained frequency-domain scheduling, yielding a throughput advantage that outstrips static and OMA benchmarks and rivals the best known NOMA-ISAC schemes. In quantitative terms, a ~ 23% rate increase over static allocation is a significant leap, ensuring that communication QoS is greatly improved in our ISAC system – comparable to the leap seen when introducing NOMA in prior works. Total utility and tradeoff Proposed dynamic allocation A key strength of our approach is its composite utility optimization, balancing sensing and communication. Using a tunable weight λ, we optimized a combined utility = (–MSE) + λ·(throughput). Under balanced weighting (λ = 0.5 in our tests), the dynamic scheme achieved a 13.49% higher total utility than the benchmark algorithms. This reflects an overall superior trade-off: the proposed resource allocation pushes the Pareto frontier outward, simultaneously improving sensing and throughput to yield a larger aggregate benefit. The 13.5% utility gain encapsulates the net improvement in dual-function performance. We emphasize that this gain comes from joint optimization of power and frequency, which allows the algorithm to find solutions that static frequency assignment or pure power allocation approaches cannot reach. NOMA-ISAC benchmarks Few prior works explicitly define a single “utility” metric combining sensing and communication, but many evaluated the S&C trade-off. Mu et al. 16 did formulate a weighted sum of communication throughput and sensing power (analogous to our utility) and solved it via beamforming design. Their results show that NOMA-ISAC achieves a better communication–sensing trade-off than conventional ISAC – for example, in 16 the NOMA scheme could simultaneously obtain high throughput and high radar sensing power, whereas the conventional scheme had to sacrifice one for the other. In the extreme case of system overload, they found the NOMA-ISAC’s joint performance was close to an “ideal” ISAC with no trade-off, underscoring a dramatically improved utility region. Dou et al. 19 similarly targeted dual-objective performance: they ensured that their scheduling/beamforming solution improves both sensing and communication metrics relative to benchmarks​, effectively moving toward a higher overall utility (though it doesn’t report a single utility value, the qualitative improvement in both aspects implies a better combined outcome). Ouyang et al. 21 provide insight that ISAC (joint) systems have more degrees of freedom than separated ones, leading to an expanded S&C capability region​. This theoretical result supports the idea that an optimal joint design (like our utility maximization) can dominate simpler strategies. In summary, the literature suggests that NOMA-based joint designs yield a noticeably superior S&C trade-off – e.g., one study noted a 29% sensing CRB reduction at equal communication rate using an advanced multiple access scheme (RSMA) over NOMA​, highlighting the headroom for improving utility in multi-user ISAC. Most NOMA-ISAC works report qualitative or case-by-case trade-off improvements (curves showing one metric vs the other). Our work goes further by quantifying a single utility increase. Comparison As can be seen in Fig. 4 the proposed dynamic allocation is unique in providing a quantifiable total utility gain (~ 13.5%) across a multi-user ISAC system. This indicates a more balanced and efficient trade-off than any static or heuristic approach, and it stands on par with the best optimized schemes in literature. For instance, our utility gain encapsulates simultaneous ~ 17% better sensing and ~ 23% better rate – a combination that is difficult to achieve with one-dimensional optimizations. Prior NOMA-ISAC algorithms usually excel in one domain (e.g. maximize radar power or communication sum-rate) while maintaining the other, whereas our approach actively optimizes both together. The result is that no compromise is needed: we improve sensing and communication concurrently (as evidenced by positive gains in both MSE and rate), which translates to a higher composite utility. This holistic improvement is a clear advantage of the dynamic joint allocation framework. It underscores the uniqueness of our solution: by leveraging instantaneous CSI and task requirements across subcarriers, we tap into performance gains that isolated power allocation or single-carrier NOMA methods left on the table. The dynamic scheme effectively outperforms conventional methods in all metrics simultaneously – a trait that few prior works explicitly demonstrated with numerical gains. Overall, the balanced utility improvement highlights the practical benefit of our method for ISAC: it can adapt to different λ priorities to trade off S&C, but under a balanced scenario it simply outclasses the benchmarks across the board. Total Utility vs λ. Power distribution Figure 5 shows the Power Distribution plots: Power Distribution Across UEs (Communication): This plot shows how power is allocated to each communication user (UE). The dynamic allocation scheme distributes power more efficiently, assigning varying power levels based on the channel conditions, while the static allocation distributes power equally across all UEs. The NOMA-ISAC scheme also distributes power across UEs, but in a different manner based on its unique approach. Power Distribution Across UEs (Communication): This plot shows how power is allocated to each communication user (UE). The dynamic allocation scheme distributes power more efficiently, assigning varying power levels based on the channel conditions, while the static allocation distributes power equally across all UEs. The NOMA-ISAC scheme also distributes power across UEs, but in a different manner based on its unique approach. Power Distribution Across Sensing Tasks: This plot illustrates how power is allocated across the sensing tasks. The dynamic allocation gives more power to certain sensing tasks, optimizing for accuracy, while the static allocation assigns equal power to all tasks. The NOMA-ISAC approach uses a different allocation, balancing the power between sensing tasks. Power Distribution Across Sensing Tasks: This plot illustrates how power is allocated across the sensing tasks. The dynamic allocation gives more power to certain sensing tasks, optimizing for accuracy, while the static allocation assigns equal power to all tasks. The NOMA-ISAC approach uses a different allocation, balancing the power between sensing tasks. Power distribution. These plots visually compare how power is distributed across both UEs and sensing tasks for each of the schemes, showcasing the dynamic allocation’s advantage in efficiently optimizing power distribution for both communication and sensing. Comparative performance summary To synthesize the above findings, Table 2 compares the proposed dynamic joint allocation algorithm with representative benchmark algorithms from 15,17,22,23 on key performance metrics. We list the sensing accuracy, communication rate, and overall utility/trade-off outcomes, using either reported quantitative values or qualitative indicators from each work. All schemes are considered under a comparable downlink scenario (with differences noted). The proposed algorithm’s improvements (in bold) are given relative to the baseline static allocation. As shown, our dynamic scheme provides significant gains in all metrics, whereas other NOMA-ISAC schemes typically improved one metric considerably while maintaining (or modestly improving) the other. The table highlights how the proposed approach uniquely achieves across-the-board enhancements (simultaneous MSE reduction and rate increase), distinguishing it from the state-of-the-art. Discussion As seen above, the proposed dynamic scheme outperforms traditional static allocation by a wide margin in all three metrics (sensing MSE, communication rate, and total utility). Compared to other NOMA-ISAC algorithms, our approach achieves a more holistic improvement: whereas others often focus on one aspect (e.g. maximizing sensing or ensuring comm capacity), our joint power–frequency allocation delivers strong gains on both simultaneously. This is attributed to its fine-grained resource control (per-subcarrier power tuning) and adaptability to instantaneous channel conditions and sensing needs. In qualitative terms, the proposed algorithm combines the strengths of prior NOMA-ISAC strategies – aggressive spectral reuse for high throughput 15,16,20, interference management for sensing fidelity 17,18, and adaptive scheduling for efficiency 19 – within a single framework. Its advantages and uniqueness can be summarized as follows: Simultaneous Sensing & Communication Gains: Unlike many benchmarks that trade one for the other, the dynamic scheme boosts sensing accuracy and communication rate together, as evidenced by the 17.65% MSE drop alongside a 22.86% rate increase​. Prior works ensured one domain didn’t degrade (or improved one at a time), but our results show concurrent enhancements in both domains, indicating a superior joint optimization. Simultaneous Sensing & Communication Gains: Unlike many benchmarks that trade one for the other, the dynamic scheme boosts sensing accuracy and communication rate together, as evidenced by the 17.65% MSE drop alongside a 22.86% rate increase​. Prior works ensured one domain didn’t degrade (or improved one at a time), but our results show concurrent enhancements in both domains, indicating a superior joint optimization. Dynamic Frequency-Domain Allocation: The use of 100 subcarriers with dynamic allocation is novel in NOMA-ISAC literature. Most existing works 16,19,21,22 optimize power or beamforming on a single carrier (or assume a fixed split like 20. By distributing power across subcarriers based on channel gains, our algorithm exploits frequency diversity and multi-user water-filling gains. This yields extra performance improvements (particularly in throughput) that static or single-carrier approaches cannot attain. Essentially, we unlock the benefit of frequency-selective scheduling in an ISAC context, which is a new contribution on top of NOMA’s known spatial/power domain benefits​. Dynamic Frequency-Domain Allocation: The use of 100 subcarriers with dynamic allocation is novel in NOMA-ISAC literature. Most existing works 16,19,21,22 optimize power or beamforming on a single carrier (or assume a fixed split like 20. By distributing power across subcarriers based on channel gains, our algorithm exploits frequency diversity and multi-user water-filling gains. This yields extra performance improvements (particularly in throughput) that static or single-carrier approaches cannot attain. Essentially, we unlock the benefit of frequency-selective scheduling in an ISAC context, which is a new contribution on top of NOMA’s known spatial/power domain benefits​. Multi-User, Multi-Task Scalability: The proposed method is evaluated with 10 UEs and 3 sensing tasks simultaneously, a considerably more complex scenario than many prior studies (e.g. 17 considered up to 6 UEs/2 targets; 20,21 focus on 2 UEs; 22 only 1 UE). The results show the algorithm scales to many users/tasks, dynamically balancing their needs. This is particularly important for 6G use cases (e.g. vehicular networks or smart factories) with dense devices and sensing demands. Our approach provides a practical way to handle such complexity, whereas earlier algorithms often simplified the setting (single pair or no inter-user interference between multiple pairs) ​. The ability to manage inter-user and inter-function interference jointly in a large system is a key advantage. Multi-User, Multi-Task Scalability: The proposed method is evaluated with 10 UEs and 3 sensing tasks simultaneously, a considerably more complex scenario than many prior studies (e.g. 17 considered up to 6 UEs/2 targets; 20,21 focus on 2 UEs; 22 only 1 UE). The results show the algorithm scales to many users/tasks, dynamically balancing their needs. This is particularly important for 6G use cases (e.g. vehicular networks or smart factories) with dense devices and sensing demands. Our approach provides a practical way to handle such complexity, whereas earlier algorithms often simplified the setting (single pair or no inter-user interference between multiple pairs) ​. The ability to manage inter-user and inter-function interference jointly in a large system is a key advantage. Adaptive and Context-Aware: Thanks to its utility-based formulation with a weight λ, our scheme is flexible – it can prioritize sensing or communications as needed, while still optimizing both. This adaptability is not explicitly present in most benchmarks, which typically fix a mode (either a sensing-centric design or a comm-centric design). Our simulations under a “balanced” λ already outperform static heuristics, and the framework can easily shift to other λ values for different mission profiles. This aligns with the trend of AI-driven, context-aware resource management envisioned for 6G, making our solution a strong candidate for real-time implementation. Adaptive and Context-Aware: Thanks to its utility-based formulation with a weight λ, our scheme is flexible – it can prioritize sensing or communications as needed, while still optimizing both. This adaptability is not explicitly present in most benchmarks, which typically fix a mode (either a sensing-centric design or a comm-centric design). Our simulations under a “balanced” λ already outperform static heuristics, and the framework can easily shift to other λ values for different mission profiles. This aligns with the trend of AI-driven, context-aware resource management envisioned for 6G, making our solution a strong candidate for real-time implementation. The proposed dynamic joint power/frequency allocation algorithm offers quantitatively superior performance on sensing MSE, communication rate, and total utility when compared to the state-of-the-art NOMA-ISAC schemes. It leverages the insights of prior works – NOMA’s spectral efficiency and interference cancellation benefits – and extends them with a broader resource allocation strategy. The result is an enhanced ISAC system that can sense more accurately (MSE ↓17.65%) and communicate faster (rate ↑22.86%), yielding an overall utility gain of ~ 13.5% over conventional methods​. These improvements underscore the advantage and uniqueness of our approach in achieving an effective balance between sensing and communications in downlink ISAC, thereby highlighting its potential as a benchmark for future 6G ISAC resource allocation designs. Conclusion In this paper, we propose a dynamic resource allocation algorithm for dedicated sensing signals in this context of downlink ISAC systems. The proposed algorithm enhances the accuracy of sensing while the communication performance is not deteriorated, by optimizing the resources power and frequency resources. Results from simulations comprehensively show that the proposed downlink ISAC system adopting the dynamic resource allocation scheme is highly efficient. The dynamic scheme achieves superior allocation performance over the current NOMA-ISAC schemes over diverse metrics by simultaneously optimizing power and frequency-domain resources according to channel instantaneous state information and sensing task requirements. In specific, it reduces the sensing MSE by up to 17.65% at the balanced λ value, improves the communication rate by up to 22.86% at the balanced λ value, and increases the total utility by 13.49% at the same time, indicating its superior ability of joint optimization under different system priorities. Static and heuristic-based schemes, and other NOMA-ISAC schemes have the merit of easier implementation but do not come close to achieving a dynamic balance of the conflicting decision objectives to optimize the tradeoff between sensing and communication. In contrast, the dynamic scheme exhibits strong adaptability and represents a powerful candidate for real-world ISAC applications in next-generation wireless systems where flexibility and performance co-optimization is paramount. ISAC systems that adopt this strategy may improve the future 6G networks, that is, high-precision sensing, like autonomous driving and smart city applications. Finally, we note that the methodological contribution of this paper lies in integrating subcarrier and power allocation into a utility-driven dynamic framework, rather than in introducing a new optimization theory. Future work will focus on enhancing theoretical depth by incorporating convex relaxation, stochastic optimization, or hybrid learning-based approaches that can provide provable performance bounds without sacrificing real-time applicability. Future work This work will be extended in future research to more sophisticated multi-antenna scenarios and the use of machine learning methods for real-time adaptation of resource allocation strategies. Also, the generalization of this algorithm to full-duplex ISAC systems will be studied. Another important direction is to relax the assumption of perfect CSI. In practical ISAC systems, CSI acquisition is imperfect due to estimation errors, feedback delays, and mobility. These imperfections can significantly affect both communication throughput and sensing accuracy. Future work will therefore extend the proposed algorithm to incorporate robust resource allocation under imperfect CSI, for example by employing stochastic optimization, chance-constrained formulations, or machine learning–based CSI prediction and compensation techniques. This will further enhance the practicality of the proposed scheme in real-time 6G deployments. Future research will explore extensions of the algorithm toward provable optimality through convex approximation or hybrid learning-based approaches, while further tailoring it to application domains such as vehicular networks, smart cities, and industrial automation. Another important extension concerns the integration of security into ISAC resource allocation. While this paper considers performance optimization under ideal secure conditions, real-world deployments face challenges such as eavesdropping, spoofing, and intentional jamming. Recent works 29–31 have highlighted the potential of secure ISAC designs, dual-functional communication–jamming frameworks, and RSMA-based methods to counter adversarial environments. Incorporating such security-aware mechanisms into our dynamic resource allocation framework will be a key focus of future research, enabling ISAC systems to operate reliably in hostile or mission-critical scenarios. Future extensions will also target high-mobility scenarios by incorporating predictive CSI models and lightweight allocation updates, ensuring robustness in rapidly time-varying wireless environments such as vehicular or UAV networks. In addition, we also intend to refine the discrete subcarrier allocation step using combinatorial optimization or machine learning–based assignment techniques, providing tighter optimality guarantees while respecting the discrete nature of OFDM resources. Acknowledgements This study is supported via funding from Prince sattam bin Abdulaziz University project number (PSAU/2025/R/1446). Author contributions This work was carried out in collaboration between all authors. Detailed methodology: Madeeha Aman and Ghani Ur Rehman. Formal analysis: Ali Daud and Riad Alharbey. Investigations: Sajid Ullah Kahn and Muhammad Zubair. Writing, review, and editing: Madeeha Aman and Muhammad Zubair. Writing original draft: Ali Daud. Data availability All data generated or analyzed during this study are included in this published article. Declarations Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Ghani Ur Rehman, Email: Ghani.rehman@kkkuk.edu.pk. Ali Daud, Email: alimsdb@gmail.com. References Associated Data Data Availability Statement All data generated or analyzed during this study are included in this published article.