We have developed a new framework using time-series analysis for dynamically assigning mobile network traffic prediction models in previously unseen wireless environments. Our framework selectively employs learned behaviors, outperforming any single model with over a 50% improvement relative to current studies. More importantly, it surpasses traditional approaches without needing prior knowledge of a cell. While this paper focuses on network traffic prediction using our adaptive forecasting framework, this framework can also be applied to other machine learning applications in uncertain environments.
The framework begins with unsupervised clustering of time-series data to identify unique trends and seasonal patterns. Subsequently, we apply supervised learning for traffic volume prediction within each cluster. This specialization towards specific traffic behaviors occurs without penalties from spatial and temporal variations. Finally, the framework adaptively assigns trained models to new, previously unseen cells. By analyzing real-time measurements of a cell, our framework intelligently selects the most suitable cluster for that cell at any given time, with cluster assignment dynamically adjusting to spatio-temporal fluctuations.

C-24

Adversarial Attacks and Defenses for Wireless Signal Classifiers using CDI-aware GANs

We introduce a Channel Distribution Information (CDI)-aware Generative Adversarial Network (GAN), designed to address the unique challenges of adversarial attacks in wireless communication systems. The generator in this CDI-aware GAN maps random input noise to the feature space, generating perturbations intended to deceive a target modulation classifier. Its discriminators play a dual role: one enforces that the perturbations follow a Gaussian distribution, making them indistinguishable from Gaussian noise, while the other ensures these perturbations account for realistic channel effects and resemble no-channel perturbations.
Our proposed CDI-aware GAN can be used as an attacker and a defender. In attack scenarios, the CDI-aware GAN demonstrates its prowess by generating robust adversarial perturbations that effectively deceive the target classifier, outperforming known methods. Furthermore, CDI-aware GAN as a defender significantly improves the target classifier's resilience against adversarial attacks.

C-23

Channel Aware Adversarial Attacks are Not Robust

Sujata Sinha, and Alkan Soysal

In IEEE Military Communications Conference (MILCOM) Workshops, Nov 2023

Adversarial Machine Learning (AML) has shown significant success when applied to deep learning models across various domains. This paper delves into channel-aware adversarial attacks on DNN-based modulation classification models within wireless environments. Our investigation centers on the robustness of these attacks concerning channel distribution and path-loss parameters. We examine two scenarios: one where the attacker has instantaneous channel knowledge and the other where the attacker is informed by statistical channel data. In both cases, we study channels under conditions of Rayleigh fading alone, Rayleigh fading combined with shadowing, and Rayleigh fading with both shadowing and path loss. Our findings reveal that the success of an AML attack is largely dictated by the distance between the attacker and the legitimate receiver. Without precise distance estimation, adversarial attacks are likely to fail.

J-11

Multivariate, Multi-step, and Spatiotemporal Traffic Prediction for NextG Network Slicing under SLA Constraints

This study presents a spatiotemporal traffic prediction approach for NextG mobile networks, ensuring the service-level agreements (SLAs) of each network slice. Our approach is multivariate, multi-step, and spatiotemporal. Leveraging 20 radio access network (RAN) features, peak traffic hour data, and mobility-based clustering, we propose a parametric SLA-based loss function to guarantee an SLA violation rate. We focus on single-cell, multi-cell, and slice-based prediction approaches and present a detailed comparative analysis of their performances, strengths, and limitations. First, we address the application of single-cell and multi-cell training architectures. While single-cell training offers individual cell-level prediction, multi-cell training involves training a model using traffic from multiple cells from the same or different base stations. We show that the single-cell approach outperforms the multi-cell approach and results in test loss improvements of 11.4\% and 38.1\% compared to baseline SLA-based and MAE-based models, respectively. Next, we explore slice-based traffic prediction. We present single-slice and multi-slice methods for slice-based downlink traffic volume prediction, arguing that multi-slice prediction offers a more accurate forecast. The slice-based model we introduce offers substantial test loss improvements of 28.2\%, 36.4\%, and 55.6\% compared to our cell-based model, the baseline SLA-based model, and the baseline MAE-based model, respectively.

C-22

Multivariate and Multi-step Traffic Prediction for NextG Networks with SLA Violation Constraints

Evren Tuna, and Alkan Soysal

In International Balkan Conference on Communications and Networking (BalkanCom), Jun 2023

This paper focuses on predicting downlink (DL) traffic volume in mobile networks while minimizing overprovisioning and meeting a given service-level agreement (SLA) violation rate. We present a multivariate, multi-step, and SLA-driven approach that incorporates 20 different radio access network (RAN) features, a custom feature set based on peak traffic hours, and handover-based clustering to leverage the spatiotemporal effects. In addition, we propose a custom loss function that ensures the SLA violation rate constraint is satisfied while minimizing overprovisioning. We also perform multi-step prediction up to 24 steps ahead and evaluate performance under both single-step and multi-step prediction conditions. Our study makes several contributions, including the analysis of RAN features, the custom feature set design, a custom loss function, and a parametric method to satisfy SLA constraints.

2022

C-21

Multivariate Spatio-temporal Cellular Traffic Prediction with Handover Based Clustering

Evren Tuna, and Alkan Soysal

In Conference on Information Sciences and Systems (CISS), Mar 2022

We consider an RNN-based traffic volume prediction, which is a critical problem for network slice management and resource allocation in slicing-enabled next generation cellular networks. We propose to use a novel cost function that takes SLA violations into account. Our approach is multivariate and spatio-temporal in three aspects. First, we consider the effects of several other RAN features in a cell besides the traffic volume. Second, we introduce feature vectors based on peak hours of the day and days of the week. Third, we introduce feature vectors based on incoming handover statistics from the neighboring cells. Our results show about 60\% improvement over MAE-based univariate LSTM models and about 20\% improvement over SLA-based univariate models.

2021

J-10

Age of Information in G/G/1/1 Systems: Age Expressions, Bounds, Special Cases, and Optimization

We consider the average age of information in G/G/1/1 systems under two service discipline models. In the first model, if a new update arrives when the service is busy, it is blocked; in the second model, a new update preempts the current update in service. For the blocking model, we first derive an exact age expression for G/G/1/1 systems. Then, using the age expression for G/G/1/1 systems, we calculate average age expressions for special cases, i.e., M/G/1/1 and G/M/1/1 systems. We observe that deterministic interarrivals minimize the average age of G/M/1/1 systems for a given mean interarrival time. Next, for the preemption in service model, we first derive an exact average age expression for G/G/1/1 systems. Then, similar to blocking discipline, using the age expression for G/G/1/1 systems, we calculate average age expressions for special cases, i.e., M/G/1/1 and G/M/1/1 systems. Average age for G/M/1/1 can be written as a summation of two terms, the first of which depends only on the first and second moments of interarrival times and the second of which depends only on the service rate. In other words, interarrival and service times are decoupled. We prove that deterministic interarrivals are optimum for G/M/1/1 systems for a given mean interarrival time. On the other hand, we observe for non-exponential service times that the optimal distribution of interarrival times depends on the relative values of the mean interarrival time and the mean service time. Finally, we propose a simple to calculate upper bound to the average age for the preemption in service discipline.

J-9

Scaling Laws for Age of Information in Wireless Networks

We study age of information in a multiple source-multiple destination setting with a focus on its scaling in large wireless networks. There are \(n\) nodes uniformly and independently distributed on a fixed area that are randomly paired with each other to form \(n\) source-destination (S-D) pairs. Each source node wants to keep its destination node as up-to-date as possible. To accommodate successful communication between all \(n\) S-D pairs, we first propose a three-phase transmission scheme which utilizes local cooperation between the nodes along with what we call mega update packets to serve multiple S-D pairs at once. We show that under the proposed scheme average age of an S-D pair scales as \( O(n^{\frac{1}{4}} \log n) \) as the number of users, \(n\), in the network grows. Next, we observe that communications that take place in Phases I and III of the proposed scheme are scaled-down versions of network-level communications. With this along with scale-invariance of the system, we introduce hierarchy to improve this scaling result and show that when hierarchical cooperation between users is utilized, an average age scaling of \( O(n^{\alpha(h)} \log n) \) per-user is achievable, where \(h\) denotes the number of hierarchy levels and \( \alpha(h) = \frac{1}{3·2^h + 1} \). We note that \( \alpha (h) \) tends to \(0\) as \(h\) increases, and asymptotically, the average age scaling of the proposed hierarchical scheme is \( O(\log n) \). To the best of our knowledge, this is the best average age scaling result in a status update system with multiple S-D pairs.

J-8

Toward an integrative professional and personal competency-based learning model for inclusive workforce development

Amy J Arnold, Jared Keyel, Alkan Soysal, and
3 more authors

The Journal on Systemics, Cybernetics and Informatics (JSCI), Apr 2021

We consider the age of information in a multihop multicast network where there is a single source node sending timesensitive updates to \(n^L\) end nodes, and \(L\) denotes the number of hops. In the first hop, the source node sends updates to n first-hop receiver nodes, and in the second hop each first-hop receiver node relays the update packets that it has received to \(n\) further users that are connected to it. This network architecture continues in further hops such that each receiver node in hop \(\ell\) is connected to \(n\) further receiver nodes in hop \(\ell + 1\). We study the age of information experienced by the end nodes, and in particular, its scaling as a function of \(n\). We show that, using an earliest \(k\) transmission scheme in each hop, the age of information at the end nodes can be made a constant independent of \(n\). In particular, the source node transmits each update packet to the earliest \(k_1\) of the \(n\) first-hop nodes, and each first-hop node that receives the update relays it to the earliest \(k_2\) out of \(n\) second-hop nodes that are connected to it and so on. We determine the optimum \(k_1\) stopping value for each hop \(\ell\) for arbitrary shifted exponential link delays.

In this paper, we introduce nonoverlay microcell/macrocell planning that is optimally designed for improving energy efficiency of the overall heterogeneous cellular network. We consider two deployment strategies. The first one is based on a fixed hexagonal grid and the second one is based on a stochastic geometry. In both of our models, microcells are placed in those areas where the received signal power levels of macrocell common pilot channels are below a certain threshold. Thus, interference between microcells and macrocells is minimized. As a result, addition of microcells increases the achieved number of bits per unit energy. Under such deployment assumptions, we investigate the effects of certain parameters on the energy efficiency. These parameters include the user traffic, the Intersite Distance (ISD), the size of microcells and the number of microcells per macrocell for the grid model, and macrocell density and microcell density for the stochastic model. The results of our performance analyses show that utilizing microcells in a sparse user scenario is worse for the energy efficiency whereas it significantly improves both energy and spectral efficiencies in a dense user scenario. Another interesting observation is that it is possible to choose an optimum number of microcells for a given macrocell density.

2015

C-14

Energy and Spectral Efficient Microcell Deployment in Heterogeneous Cellular Networks

Mahmut Demirtas, and Alkan Soysal

In IEEE Vehicular Technology Conference (VTC), May 2015

In this paper, the source and relay transmit covariance matrices are jointly optimized for a fading multiple-antenna relay channel when the transmitters only have partial channel state information (CSI) in the form of covariance feedback. For both full-duplex and half-duplex transmissions, we evaluate lower and upper bounds on the ergodic channel capacity. These bounds require joint optimization over the source and relay transmit covariance matrices. The methods utilized in the previous literature cannot handle this joint optimization over the transmit covariance matrices for the system model considered in this paper. Therefore, we utilize matrix differential calculus and propose iterative algorithms that find the transmit covariance matrices to solve the joint optimization problem. In this method, there is no need to specify first the eigenvectors of the transmit covariance matrices. The algorithm updates both the eigenvectors and the eigenvalues at each iteration. Through simulations, we observe that lower and upper bounds are close to each other. However, the distance between the lower and upper bounds depends on the channel conditions. If the mutual information on the source-to-relay channel and the broadcast channel get closer to each other, the bounds on capacity also get closer.

C-11

Fading MIMO Relay Channels with Channel Estimation Error

Bengi Aygün, and Alkan Soysal

In IEEE International Conference on Communications (ICC), Jun 2013

This is the second part of a two-part paper on the joint channel estimation and resource allocation problem in MIMO systems with noisy channel estimation at the receiver side and partial CSI, in the form of covariance feedback, available at the transmitter side. We consider transmit-side correlated MIMO channels with block fading, where each block is divided into training and data transmission phases. In this paper, we extend the single-user results of Part I to the multiple access channel. For the data transmission phase, we propose an iterative algorithm to solve for the optimum system resources such as time, power and spatial dimensions. This algorithm updates the parameters of the users in a round-robin fashion. In particular, the algorithm updates the training and data transmission parameters of a user, when those of the rest of the users are fixed, in a way to maximize the achievable sum-rate in a multiple access channel; and iterates over users in a round-robin fashion. Finally, we provide a detailed numerical analysis to support the analytical results of both parts of this two-part paper.

J-3

Joint Channel Estimation and Resource Allocation for MIMO Systems-Part I: Single-User Analysis

Multiple antenna systems are known to provide very large data rates, when the perfect channel state information (CSI) is available at the receiver. However, this requires the receiver to perform a noise-free, multi-dimensional channel estimation, without using communication resources. In practice, any channel estimation is noisy and uses system resources. We shall examine the trade-off between improving channel estimation and increasing the achievable data rate. We consider transmit side correlated multi-input multi-output (MIMO) channels with block fading, where each block is divided into training and data transmission phases. The receiver has a noisy CSI that it obtains through a channel estimation process, while the transmitter has partial CSI in the form of covariance feedback. In Part I of this two-part paper, we consider the single-user case, and optimize the achievable rate jointly over parameters associated with the training phase and data transmission phase. In particular, we first choose the training signal to minimize the channel estimation error, and then, develop an iterative algorithm to solve for the optimum system resources such as time, power and spatial dimensions. Specifically, the algorithm finds the optimum training duration, the optimum allocation of power between training and data transmission phases, the optimum allocation of power over the antennas during the data transmission phase.

2009

J-2

Optimality of Beamforming in Fading MIMO Multiple Access Channels

We consider the sum capacity of a multi-input multi-output (MIMO) multiple access channel (MAC) where the receiver has the perfect channel state information (CSI), while the transmitters have either no or partial CSI. When the transmitters have partial CSI, it is in the form of either the covariance matrix of the channel or the mean matrix of the channel. For the covariance feedback case, we mainly consider physical models that result in single-sided correlation structures. For the mean feedback case, we consider physical models that result in in-phase received signals. Under these assumptions, we analyze the MIMO-MAC from three different viewpoints. First, we consider a finite-sized system. We show that the optimum transmit directions of each user are the eigenvectors of its own channel covariance and mean feedback matrices, in the covariance and mean feedback models, respectively. Also, we find the conditions under which beamforming is optimal for all users. Second, in the covariance feedback case, we prove that the region where beamforming is optimal for all users gets larger with the addition of new users into the system. In the mean feedback case, we show through simulations that this is not necessarily true. Third, we consider the asymptotic case where the number of users is large. We show that in both no and partial CSI cases, beamforming is asymptotically optimal. In particular, in the case of no CSI, we show that a simple form of beamforming, which may be characterized as an arbitrary antenna selection scheme, achieves the sum capacity. In the case of partial CSI, we show that beamforming in the direction of the strongest eigenvector of the channel feedback matrix achieves the sum capacity. Finally, we generalize our covariance feedback results to double-sided correlation structures in the Appendix.

2008

C-6

MIMO Multiple Access Channels with Noisy Channel Estimation and Partial CSI Feedback

We consider both the single-user and the multi-user power allocation problems in MIMO systems, where the receiver side has the perfect channel state information (CSI), and the transmitter side has partial CSI, which is in the form of covariance feedback. In a single-user MIMO system, we consider an iterative algorithm that solves for the eigenvalues of the optimum transmit covariance matrix that maximizes the rate. The algorithm is based on enforcing the Karush-Kuhn-Tucker (KKT) optimality conditions of the optimization problem at each iteration. We prove that this algorithm converges to the unique global optimum power allocation when initiated at an arbitrary point. We, then, consider the multi-user generalization of the problem, which is to find the eigenvalues of the optimum transmit covariance matrices of all users that maximize the sum rate of the MIMO multiple access channel (MIMO-MAC). For this problem, we propose an algorithm that finds the unique optimum power allocation policies of all users. At a given iteration, the multi-user algorithm updates the power allocation of one user, given the power allocations of the rest of the users, and iterates over all users in a round-robin fashion. Finally, we make several suggestions that significantly improve the convergence rate of the proposed algorithms.

2006

C-3

Optimum Power Allocation in Fading MIMO Multiple Access Channels with Partial CSI at the Transmitters