Haris Haralambous

2011 Special Issue: Reliable prediction intervals with regression neural networks

This paper proposes an extension to conventional regression neural networks (NNs) for replacing the point predictions they produce with prediction intervals that satisfy a required level of confidence. Our approach follows a novel machine learning framework, called Conformal Prediction (CP), for assigning reliable confidence measures to predictions without assuming anything more than that the data are independent and identically distributed (i.i.d.). We evaluate the proposed method on four benchmark datasets and on the problem of predicting Total Electron Content (TEC), which is an important parameter in trans-ionospheric links; for the latter we use a dataset of more than 60000 TEC measurements collected over a period of 11 years. Our experimental results show that the prediction intervals produced by our method are both well calibrated and tight enough to be useful in practice.

24-Hour Neural Network Congestion Models for High-Frequency Broadcast Users

This paper presents the development of Neural Network models to predict the likelihood of interference experienced by Broadcast users in the HF spectrum (3-30 MHz). The models are based upon several years of measurements recorded at Linkoping (Sweden) across the HF band, covering a substantial part of a sunspot cycle. The dataset used for the model development is a result of a long-term project being undertaken jointly by the University of Manchester and by the Swedish Defence Research Establishment, to measure systematically and to analyze the occupancy of the entire HF spectrum. The measure of occupancy used is congestion, which is defined as the fraction of channels within a certain frequency allocation with signals exceeding a given threshold. The procedures for measuring and modeling congestion as a function of solar activity, time of day, day of year and incident field strength threshold are briefly presented. The accuracy of the predictions produced by the developed models demonstrate their ability to successfully capture the 24-hour, seasonal and long-term trend in the variability of congestion. These models can be used to advise operators on typical interference occupancy levels and assist the HF broadcast service in the planning of frequency usage and management by assessing the interference effect to short-wave broadcasting in an effort to alleviate spectral congestion in the HF broadcast bands.

Study of the effect of 17–18 March 2015 geomagnetic storm on the Indian longitudes using GPS and C/NOFS

The largest geomagnetic storm in solar cycle 24 occurred during March 17-18, 2015 where the main phase of the storm commenced from 07:00 UT of March 17, 2015 and reached the Dst negative minimum at 22:00 UT. The present paper reports observations of TEC, amplitude and phase scintillations from different GPS stations of India during the storm of March 17 and highlights its effects on GPS. It also presents the global ESF occurrence during the storm using total ion density drift measurements from C/NOFS satellite. TEC enhancements were noted from stations along 77oE meridian around 10:00 UT on March 17 compared to March 16 and 18 indicating positive storm effects arising out of equatorward neutral wind in the local morning-noon sector of the main phase. Intense scintillation observations from Calcutta were most extensive during 15:00-16:00 UT, March 17 and the receiver recorded a longitude deviation of 5.2 m during this time. Cycle slips of the order of 8 s could be observed during periods of intense phase scintillations on the same night. Intense scintillation observation from Palampur is an exceptional phenomenon attributed to the dramatic enhancement of the electric field due to PPEF leading to a very high upward ion velocity over the magnetic equator as recorded by C/NOFS. The total ion density measured globally by C/NOFS reveals two distinct longitude regions of ESF occurrence during the storm: i) East Pacific sector and ii) Indian longitude during the storm. The time and longitude of ESF occurrence could be predicted using the time of southward turning of IMF Bz.

A local Total Electron Content neural network model over Cyprus

This paper presents the application of neural networks for the prediction of the Total Electron Content (TEC) over Cyprus. This ionospheric characteristic constitutes an important parameter in trans-ionospheric links since it is used to derive the signal delay imposed by the ionosphere. The model is based on TEC measurements obtained over a period of 11 years.

Neural networks regression inductive conformal predictor and its application to total electron content prediction

In this paper we extend regression Neural Networks (NNs) based on the Conformal Prediction (CP) framework for accompanying predictions with reliable measures of confidence. We follow a modification of the original CP approach, called Inductive Conformal Prediction (ICP), which enables us to overcome the computational inefficiency problem of CP. Unlike the point predictions produced by conventional regression NNs the proposed approach produces predictive intervals that satisfy a given confidence level. We apply it to the problem of predicting Total Electron Content (TEC), which is an important parameter in trans-ionospheric links. Our experimental results on a dataset collected over a period of 11 years show that the resulting predictive intervals are both well-calibrated and tight enough to be useful in practice.

HF Spectral Occupancy Time Series Models Over the Eastern Mediterranean Region

The high frequency (HF) radio channel is highly variable and unpredictable. On the other hand, being largely utilized around the globe, the HF spectrum is densely occupied, characterized by significant levels of interference. Therefore, spectral congestion is always a severe issue to be addressed. One year data (June 2012-June 2013) of HF spectral occupancy measurements collected by a dedicated measurement system installed in Cyprus have been examined in order to characterize and model the HF interference environment over the eastern Mediterranean region. An investigation into the short-term temporal variability of HF spectral occupancy for selected indicative allocations has been conducted. With a view to extend the data analysis from descriptive to predictive and from qualitative to quantitative, time series analysis has been applied to model the HF spectral occupancy of the indicative ranges in the HF spectrum. We categorized different time series models for different occupancy patterns. Analysis for three broadcast allocations covering three different portions of the HF spectrum namely, lower (5.950-6.200 MHZ), middle (11.650-12.050 MHZ), and upper (17.550-17.900 MHZ), is presented. Results of this investigation are expected to be useful for radio system design and frequency management strategies.

A Neural Network Model for the Critical Frequency of the F2 Ionospheric Layer over Cyprus

This paper presents the application of Neural Networks for the prediction of the critical frequency foF2 of the ionospheric F2 layer over Cyprus. This ionospheric characteristic (foF2) constitutes the most important parameter in HF (High Frequency) communications since it is used to derive the optimum operating frequency in HF links. The model is based on ionosonde measurements obtained over a period of 10 years. The developed model successfully captures the variability of the foF2 parameter.

Short-term forecasting of the likelihood of interference to groundwave users in the lowest part of the HF spectrum

In the design and performance evaluation of practical HF communication systems, it is essential to use procedures that assess the detrimental effect of interference from other users in a near real time mode. These procedures can extend system capability to estimate interference background, in the context of real time channel evaluation (RTCE) in order to advise operators on typical interference occupancy levels and to improve the quality and reliability of radio communication services through adaptation of communication parameters. In this study a Neural Network approach is proposed for the short-term forecasting of the likelihood of interference experienced by HF groundwave communication systems. In particular this paper describes the development of neural network models to indicate the degree of spectral congestion in frequency allocations in the lowest part of the HF spectrum (1.6 to 4 MHz) 1 hour in advance, as a function of the present congestion level, time of day, season, and field strength threshold. The modeled parameter, congestion, is defined as the relative number of narrow frequency channels (1 kHz wide) within a frequency allocation that have signals above a given threshold.

Characteristics of postmidnight L band scintillation in the transition region from the equatorial to midlatitudes over the Indian longitude sector using COSMIC, C/NOFS, and GPS measurements

Occurrence of L band scintillations around midnight and postmidnight hours have not been well studied and reported from the higher equatorial latitudes in the transition region from the equatorial to midlatitudes over the Indian longitude sector. The present paper reports cases of postmidnight L band scintillation observations by COSMIC during March 2014 over the Indian longitude sector. GPS S4 measurements from the International Global Navigation Satellite Systems Service station at Lucknow (26.91°N, 80.96°E geographic; magnetic dip: 39.75°N) corroborate occurrence of postmidnight scintillations. The F region vertical upward velocities around the magnetic equator during evening hours have been used to understand the possibility of impact of irregularities generated over the magnetic equator at latitudes north of 30°N. Postmidnight L band scintillations at latitudes greater than 30°N without corresponding premidnight scintillations present interesting scientific scenario and give rise to suggestions of (1) any coupling mechanism between the equatorial and midlatitudes through which irregularities seeded in the midlatitudes may affect transionospheric satellite links at low latitudes or (2) irregularity generation at midlatitudes not connected with equatorial instabilities. Long-term analysis of S4 at L band measured by COSMIC over the Indian longitudes during March 2007–2014 exhibits a well-defined longitude swath around 75–83°E of reduced (0.2 < S4 < 0.4) or no scintillations which may be attributed to the longitudinal variability of scintillation occurrence following the global four-cell pattern of ionospheric activity.

A Neural Network Tool for the Interpolation of foF2 Data in the Presence of Sporadic E Layer

This paper presents the application of Neural Networks for the interpolation of (critical frequency) foF2 data over Cyprus in the presence of sporadic E layer which is a frequent phenomenon during summer months causing inevitable gaps in the foF2 data series. This ionospheric characteristic (foF2) constitutes the most important parameter in HF (High Frequency) communications since it is used to derive the optimum operating frequency in HF links and therefore interpolating missing data is very important in preserving the data series which is used in long-term prediction procedures and models.