Milan Bjelica

Energy savings for cellular network with evaluation of impact on data traffic performance

We present a concrete methodology for saving energy in future and contemporary cellular networks. It is based on re-arranging the user-cell association so as to allow shutting down under-utilized parts of the network. We consider a hypothetical static case where we have complete knowledge of stationary user locations and thus the results represent an upper bound of potential energy savings. We formulate the problem as a binary integer programming problem, thus it is NP-hard, and we present a heuristic approximation method. We simulate the methodology on an example real cellular network topology with traffic-and user distribution generated according to recently measured patterns. Further, we evaluate the energy savings, using realistic energy profiles, and the impact on the user-perceived network performance, represented by delay and throughput, at various times of day. The general findings conclude that up to 50% energy may be saved in less busy periods, while the performance effects remain limited. We conclude that practical, real-time user-cell re-allocation methodology, taking into account user mobility predictions, may thus be feasible and bring significant energy savings at acceptable performance impact.

Towards TV recommender system: experiments with user modeling

As the number of cable TV programs grows, it becomes more difficult for the viewers to find the right one. This calls for specialized recommender systems, often in a form of electronic program guides, which should provide unobtrusive assistance. In this paper, we analyze such recommender system design under the broadcast scenario, where uplink connection to the network center is not available. We put special emphasis on user modeling algorithm that would be able to efficiently learn the users interests. Our proposal applies the elements of machine learning and pattern recognition, as well as the information retrieval theory, like vector spaces and cluster hypothesis. The derived algorithm is computationally simple, while experimental results show high acceptance ratio of the proposed recommendations.

Energy consumption comparison between macro-micro and public femto deployment in a plausible LTE network

We study the energy consumptions of two strategies that increase the capacity of an LTE network: (1) the deployment of redundant macro and micro base stations by the operator at locations where the traffic is high, and (2) the deployment of publicly accessible femto base stations by home users. Previous studies show the deployment of publicly accessible residential femto base stations is considerably more energy efficient; however, the results are proposed using an abstracted model of LTE networks, where the coverage constraint was neglected in the study, as well as some other important physical and traffic layer specifications of LTE networks. We study a realistic scenario where coverage is provided by a set of non-redundant macro-micro base stations and additional capacity is provided by redundant macro-micro base stations or by femto base stations. We quantify the energy consumption of macro-micro and femto deployment strategies by using a simulation of a plausible LTE deployment in a mid-size metropolitan area, based on data obtained from an operator and using detailed models of heterogeneous devices, traffic, and physical layers. The metrics of interest are operator-energy-consumption/total-energy-consumption per unit of network capacity. For the scenarios we studied, we observe the following: (1) There is no significant difference between operator energy consumption of femto and macro-micro deployment strategies. From the point of view of society, i.e. total energy consumption, macro-micro deployment is even more energy efficient in some cases. This differs from the previous findings, which compared the energy consumption of femto and macro-micro deployment strategies, and found that femto deployment is considerably more energy efficient. (2) The deployment of femto base stations has a positive effect on mobile-terminal energy consumption; however, it is not significant compared to the macro-micro deployment strategy. (3) The energy saving that could be obtained by making macro and micro base stations more energy proportional is much higher than that of femto deployment.

Unobtrusive relevance feedback for personalized TV program guides

Paradoxically, a growing number of available channels in digital cable TV systems brings discomfort to the viewers who now experience difficulties in finding a content that would hold their attention. In such an environment, personalized program guides are needed to assist the viewers in retrieving the preferred programs in reasonable time. The design of these systems is bounded by the demand of unobtrusiveness and the limitations of broadcast infrastructure, with the lack of return (uplink) connection to the network center being the most significant one. In this paper, we investigate learning of users viewing preferences through mechanism known as relevance feedback. Our goal is to develop a system that would efficiently track the patterns of users interests without disturbing her viewing habits. Our proposal applies the elements of machine learning and information retrieval theory. We consider three different schemes and validate their performances by series of computer simulations.

Impact of class imbalance on personalized program guide performance

When TV recommender systems perform well, number of interactions in which their users expressed positive feedback on the recommended content is expected to be greater than the number of negative ones. This is known as class imbalance and, paradoxically, it degrades the system performance by making the identification of the programs the user will dislike increasingly difficult. As the misclassification of the unwanted content is easily perceived by TV viewers, it should be avoided by all means. In this paper, a personalized TV program guide based on neural network is described. It is shown how class imbalance information can be exploited in learning the user preferences. This not only improves the system performance, but increases the user satisfaction as well1.

Adaptive feedback schemes for personalized content retrieval

We consider personalized content retrieval in a resource-constrained multiservice environment with broadcast TV acting as a model usage scenario. We propose personalized recommender system that captures the users viewing habits without obstructing the usual way TV is watched. Our proposal describes program representation and retrieval, user modeling, and aggregation of her/his estimated interests by adaptive feedback schemes. Through series of experiments with TV viewing application, we show that our proposal promptly learns the users preferences and delivers valued recommendations.

COST-EFFICIENT PHASE NOISE MEASUREMENT

Abstract In this paper, an automated system for oscillator phase noise measurement is described. The system is primarily intended for use in academic institutions, such as smaller university or research laboratories, as it deploys standard spectrum analyzer and free software. A method to compensate the effect of instrument intrinsic noise is proposed. Through series of experimental tests, good performances of our system are verified and compliance to theoretical expectations is demonstrated.

Experiences with remote laboratory

A case study in implementation of a remote laboratory for teaching telecommunication measurements is described. The proposal is simple, uses standard equipment, and could be deployed for engineering courses in similar economically challenged institutions as well. It provided a valuable support to both the students and the instructors, and helped us perceive some flaws in the curriculum. To the best of our knowledge, this was the first case of using a remote laboratory on a university level in our country.

Personalized program guide based on one-class classifier

As TV viewers inherently tend to avoid contents they might dislike, they do not provide equal amounts of positive and negative feedback on their viewing preferences. At the same time, mobile devices are becoming target platforms for multimedia content delivery. Personalized program guides need to cope with these challenges. They must not only properly recognize the undesired content, despite the lack of the learning data, but also provide valuable recommendations without compromising the users privacy. Moreover, the complexity of the applied algorithms has to be low enough to match the limited hardware resources of the mobile terminals. In this paper, the design of such program guide is described. Several system architectures are developed and compared. The best performance is achieved for a single hidden layer autoencoder neural network trained with the FISTA algorithm1.

Performance metrics for personalized program guides

The performance of TV recommender system which uses machine learning techniques is degraded due to imbalanced distribution of collected viewing preferences. As users have a tendency to provide positive feedbacks much more than the negative ones, the system that does not use methods to deal with class imbalance provides poor prediction of the contents that the user does not like to watch. Thus undesirable contents can often be recommended, which is perceived by users as bad recommendation. The probability of bad recommendations can be significantly decreased if the information about class imbalance is incorporated into the machine learning algorithm. However, this improvement comes at expense of degraded prediction of contents that user likes to watch; thus the quality of recommendations is decreased. In addition to learning algorithm, the choice of performance metric that is maximized influences user perception of the system performance. In this paper, it is shown that using the adjusted G-mean instead of G-mean metric can increase the quality of recommendations provided by the system based on neural network, without significant increase in the probability of bad recommendations. This further results in the increase of the recommendation diversity and, consequently, in the increase of user satisfaction.