A. Bernardini

Analysis of different optimization criteria for IF predistortion in digital radio links with nonlinear amplifiers

A compensation technique for nonlinearities is usually necessary in QAM digital satellite links. Analog waveform compensation has the advantage of reducing out-of-band spectral emissions with respect to end-to-end digital predistortion, in addition to its easy realization. Nevertheless, it has been often neglected in the literature because of its lack of adaptivity; we propose making the predistorter fully adaptive. Different control schemes applied to fifth-order IF predistortion with different optimization criteria have been investigated. A simulation-based analysis shows that adaptive IF compensation achieves excellent performance if compared with static predistortion of the same order. Comparisons with upper bounds related to digital predistortion with memory, moreover, denote that IF analog predistortion can be truly considered an alternative approach, especially for narrow pulse shaping.

Application of neural waveform predistortion to experimental TWT data

An evaluation is made of the predistorters achievable performance using HPA (high-power amplifier) models matched to experimental TWT data. How the neural net capability for inverse modeling, and then as predistorter, is related to the various TWTs that must be fitted is discussed. A supervised neural net is used with one internal neuron and no more than 10 internal unit, and the backpropagation algorithm for the learning process. The results related to TWT data obtained confirm the performances achievable with the generic TWT model: an average gain of 3 dB for the 64-QAM and an average gain of 5.5 dB for the 256-QAM systems, with respect to a baseband predistorter.<<ETX>>

A new predistortion technique using neural nets

Abstract The present paper deals with the nonlinear deconvolution problem incurring in the design of a predistorter with memory applied to digital transmissions over nonlinear channels. We propose the employment of neural nets for the inversion of nonlinearity because of the basic attitude of parallel structures to cope with nonlinear problems. The task of the neural net consists in the implementation of an input-output mapping as required by the specific inversion or deconvolution problem. In the present case we exploit the circumstance that the nonlinear channel is constituted by a nonlinear memoryless block and two linear blocks with memory which involves that the complex predistorter function can be split into simpler ones. The approximation task is therefore reduced to surface (only two input variables) reconstruction: it can easily be accomplished by a multilayer perceptron since the spatial representation of the function to be mapped suggests geometrical interpretation of the nets parameters, thus providing a method to get a starting configuration of the net itself.

Linear prediction methods for interference elimination in spread spectrum systems

The performance of Pseudonoise (PN) Spread Spectrum System in the presence of narrow-band interference can be improved by using various types of interference rejection filters. In this paper two different approaches are considered, the first based on Surface Acoustic Wave (SAW) devices and the second based on digital whitening using transversal filters. The reasons for the convenience of digital whitening are discussed and techniques for determining the coefficients of a linear interference suppression filter are described. Numerical results are presented on the characteristics of the linear filter and on its effectiveness in suppressing the interference when many different jamming and signaling conditions are considered. It is found that a critical point is the choice of the filter order which is related to a number of different design parameters: the signal-to-jammer power ratio, the noise power, the number of bands in which the jammer tones are splitted and the number of signal samples used in the coefficient estimation.

Performance Analysis of new Techniques of Predistortion with Memory in Digital Radio Links

This paper presents an efficient data predistortion technique with memory applied to digital transmissions over nonlinear channels, specifically satellite links, employing QAM signal formats. The proposed technique is aimed at reducing the high complexity of a conventional pre-distorter with memory by splitting the function to be realised into simpler blocks. This approach has been suggested by the circumstance that the numerical channel is constituted, in this case, by a nonlinear memoryless block and by two linear blocks with memory so that the predistorter itself can be realised by two FIR filters separated by a memoryless nonlinear element. As for the nonlinear block, we propose the employment of a neural net performing a typical surface reconstruction task, that is the approximation of the function inverting the nonlinearity source (in this case the high power amplifier). Performance analysis has shown a significant improvement with respect to memoryless predistortion (up to 4.5 dB); our proposal, moreover, outperforms conventional 3-stage-memory predistortion as the modulation levels increase and as the spectral shaping gets narrower.

Semi-analytical performance evaluation in satellite digital links in the presence of interference

A model for the semi-analytic performance evaluation of digital satellite radio links in the presence of interference on both the uplink and the downlink is presented. Error probability on the linear portion of the link is estimated using simulation to determine the moments of the interfering signal samples and analyzing the effect of an undetermined phase difference among carriers. The nonlinear portion of the link is modeled using a series expansion of the nonlinearity; the output terms are then separated, allowing construction of the conditional probability densities required in the error probability computation. Results are carried out for M-QAM (M-ary quadrature amplitude modulation) and M-PSK (M-ary phase-shift keying) modulation systems. >

Optimization of an adaptive cubic predistorter for multilevel quadrature amplitude modulation

The performance of an optimized waveform predistorter to compensate for the nonlinear distortions in M-QAM (quadrature amplitude modulation) digital radio links is analyzed. An optimization technique for cubic predistorters is proposed. The optimization is carried out by varying the predistorter characteristic against the working point of the power amplifier, whereas conventional cubic distorters are aimed at approximating the HPA (high power amplifier) characteristic only in a fixed point (the origin or the saturation point). The receiver is supposed to provide amplitude and phase control. Simulation results show that the proposed solution performs much better than not only the conventional cubic predistorter but also the baseband digital predistorter.<<ETX>>

The use of a Neural Net for Copeing with Nonlinear Distortions

In this paper a predistorter realized with a neural net is proposed. The predistortion acts on the whole input waveform : it provides a continuous time prewarping where convenctional base band predistortion acts only on the discrete time input sequence. The overall linearity of the transmitter is so recovered and the nonlinear ISI is removed at sampling point The work is a development of a previous one [10] in which a neural net was employed to realize a base-band memoryless predistortion. The results obtained for 16, 64 and 256 QAM modulation systems show excellent performance if compared with base-band memoryless predistortion expecially if small roll-off factors are employed. The major improvement, however, is represented by the extreme simplicity of the proposed solution compared with convenctional predistorters with memory. In this work we have assumed some simplification concerning the neural net architecture; future works, however, will be aimed to removing these reductive assumptions.

Letter: Neural classifiers as optimal decision devices for warped M-QAM signal formats

The paper is focused on the application of perceptron-based neural networks to optimal decision for M-QAM signals transmitted over nonlinear channels. The decision is performed by two cascaded blocks: the first denotes one of the 4 quadrants and the second selects one of the M/4 symbols. A specific initialization procedure has been adopted in order to avoid local minima. The classifier performance, for the specific application, has been evaluated by the CNR (Carrier to Noise Ratio) degradation due to nonlinearity (ΔC/N) for a target error rate P e = 10 -3 . The results, obtained by MonteCarlo simulations, denote optimal matching with respect to upper bounds obtained with some minor simplifying hypothesis, even if the overall methods effectiveness can be adequate only for mild nonlinearity conditions.

A calculated distribution for attenuation due to atmospheric absorption in the millimeter wave range

Based on the Liebes model for the atmospheric absorption dependance on temperature, relative humidity and pressure, a statistical distribution of absorption for seven Italian cities is calculated from data collected over a six years period. Results are statistically analyzed and relevant issues such as climatic dependence and time variability are investigated.