Research On Traceability Of Emitted Interference Based On Underdetermined Blind Source Separation And Support Vector Machine

As the mini-computer and large scale integrated circuits are more and more popular today, electromagnetic interference is becoming an increasingly prominent issue in people’s lives, countries have also set standards to restrict noise radiation of products. Therefore, it’s necessary to strictly control its radiation power in its beginning of product design. Once the radiation problem occurs, it is necessary to quickly find the cause out, corrective actions and re-designed to shorten product design cycles.Usually, however, the technical staff often rely on their experience of the problem product positioning, efficiency is often very low, and therefore the need for a standardized method for rapid tracing of electromagnetic interference noise. This paper presents this objective noise traceable method based on blind source separation and support vector machines. First used in the anechoic chamber several observational approach sensors for electronic products radiation signal acquisition, access to several groups of mixed-signal components of radiation; then radiated signals using blind source separation algorithm to separate the collected signals, break out each individual element; Finally, the empirical mode decomposition method based on the radiation signal feature extraction, to provide a reference for the electromagnetic compatibility problems traceable.In the step of signal decomposition, considering we don’t know the number of the source signal, so using a sparse decomposition based on statistical underdetermined blind source separation method to decompose the observed signal. This paper presents a new hybrid matrix estimation method, the algorithm first potential function by repeatedly setting the unit interval estimate for the number of source signals were observed, the longest observed steady intervals to determine the number of source signals, then the effective within the parameter interval, confidence by setting samples were screened for samples, retaining only reliable samples for mixing matrix estimation.After the screening of the samples, using a regression algorithm based on support vector regression, regression operation on the samples to obtain the highest degree of all samples fit a straight line on a direction of a single cluster, the cluster as a cluster clustering direction represented, thereby obtaining an estimated mixing matrix column vector. Experimental results show that, compared with the traditional blind source separation algorithm, the mixing matrix is estimated that the proposed algorithm can effectively improve the estimation accuracy, and can more accurately separate the lab actual signal.