Search

The paper aims to highlight relative strengths and weaknesses of some of the recently proposed architectures for hardware implementation of analog-to-information converters based on Compressive Sensing. To do so, the most common architectures are analyzed when saturation of some building blocks is taken into account, and when measurements are subject to quantization to produce a digital stream. Furthermore, the signal reconstruction is performed by established and novel algorithms (one based on linear programming and the other based on iterative guessing of the support of the target signal), as well as their specialization to the particular architecture producing the measurements. Performance is assessed both as the probability of correct support reconstruction and as the final reconstruction error. Our results help highlighting pros and cons of various architectures and giving quantitative answers to some typical design-oriented questions. Among these, we show: 1) that the (Random Modulation Pre-Integration) RMPI architecture and its recently proposed adjustments are probably the most versatile approach though not always the most economic to implement; 2) that when 1-bit quantization is sought, dynamically mixing quantization and integration in a randomized ΔΣ architecture help bringing the performance much closer to that of multi-bit approaches; 3) for each architecture, the trade-off between number of measurements and number of bits per measurements (given a fixed bit-budget); and 4) pros and cons of the use of Gaussian versus binary random variables for signal acquisition.

Results of recent theoretical investigations highlighted that the use of chaos in direct-sequence code-division multiple access (DS-CDMA) systems may lead to nonnegligible improvements in communication quality for several scenarios. We here briefly review the main steps in this derivation and report the experimental verifications of the corresponding theoretical prediction. With this, we confirm that chaos-based spreading sequences outperform classical pseudo-random sequences in at least two important cases. Over nonselective channels, the ability of chaos-based spreading of minimizing multiple-access interference leads to a measured 60% improvement in P/sub err/ with respect to classical spreading. Over selective channels, the possibility of jointly optimizing chaos-based spreading and Rake receiver profiles leads to improvements of up to 38% in P/sub err/ with respect to systems with either conventional spreading or conventional Rake policies.

A simple form of cooperation between the k-nearest neighbors (NN) approach to classification and the neural-like property of adaptation is explored. A tunable, high level k-nearest neighbors decision rule is defined that comprehends most previous generalizations of the common majority rule. A learning procedure is developed that applies to this rule and exploits those statistical features that can be induced from the training set. The overall approach is tested on a problem of handwritten character recognition. Experiments show that adaptivity in the decision rule may improve the recognition and rejection capability of standard k-NN classifiers.

The theoretical limits of asynchronous DS-CDMA systems over AWGN channels are investigated by computing the statistics of the associated Shannon capacity. A comparison is reported between chaos-based spreading and ideal random spreading Gold codes and maximum-length codes. It is shown that chaos-based spreading results in a nonnegligible increase of the capacity.

A novel estimation of the minimum achievable interference in direct sequence-code division multiple access (DS-CDMA) systems is introduced, which holds when spreading sequences with exponentially vanishing autocorrelation are employed. This can be applied to many of the recently proposed improvements to classical maximum-length or Gold sequences, such as chaos-based spreading. Asymptotic, infinite-bandwidth results are also provided, clarifying the maximum attainable gain. Empirical evidence shows that this theoretical maximum is achieved by some chaos-based sequences which are therefore optimal.

The problem of improving the capability of statistical character classifiers based on finite and sparse training sets is addressed. A significant improvement is obtained coupling standard classifiers based on the k-nearest neighbours technique with a second higher level classification stage. This method has been applied to three existing classifiers reducing the error rate at zero rejection of ˜ 17%.

A receiver for synchronous DS-CDMA systems is proposed with a decision window that is aligned to encompass two half bits of each user. Such a receiver is compared with the classical one in terms of Shannon capacity showing a definite improvement. Based on this, traditional upper bounds and optimality criteria on the capacity of synchronous DS-CDMA systems are revised.

The periodic signals used in electronic circuits (e.g. as timings) are a major cause of interference which can be reduced by introducing suitable modulations. Chaotic sources have so far been proposed for producing the modulating signals only as cheap replacements for random generators. Focusing on fast modulations it is shown that chaotic signals can also outperform random ones.

Recent measures on LANs have highlighted the self-similar nature of the traffic. A systematic procedure to design a 1D chaotic map, which generates a self-similar process characterised by a polynomial OFF time distribution, is considered and reviewed. This polynomial law allows the performance of a queue system to be investigated by extending the G/M/1 theory to the case of discrete arrival and service processes. Analytical results are reported highlighting the impact of the traffic self-similar degree and simulations show the validity of the developed theory.

The classical estimation of co-channel interference in asynchronous direct sequence-code division multiple access (DS-CDMA) systems, obtained for a conventional receiver under the standard Gaussian approximation (SGA), is analytically minimised with respect to the auto-correlation function of spreading sequences. The optimum auto-correlation function is found to be real, with alternating signs, and extremely closely follows an exponential trend. A family of chaotic maps is presented which is able to generate sequences with exponential auto-correlation function with alternating signs. The resulting DS-CDMA system is an extremely good approximation of the optimum achievable system.