520 IEEE TRANSACTIONS ON ELECTROMAGNETIC COMPATIBILITY, VOL. 43, NO. 4, NOVEMBER 2001 A Genetic Algorithm Based Method for Source Identification and Far-Field Radiated Emissions Prediction From Near-Field Measurements for PCB Characterization Joan-Ramon Regué, Member, IEEE, Miquel Ribó, Associate Member, IEEE, Josep-Maria Garrell, and Antonio Martín Abstract—In this paper, a new method for predicting the far- field radiated emissions and for finding the radiation sources of a device from near-field measurements is presented. It is based on the substitution of the original device by an equivalent set of elemental dipoles, placed over the main radiating sources, which radiate the same near-field (and therefore, far-field). This equivalent set of el- emental dipoles is generated using a genetic algorithm. From the position and type of the equivalent elemental dipoles, the position of the actual radiating sources is determined. Since the field pro- duced by an elemental dipole is known, the far-field radiation of the actual radiating source can be calculated. The new method has been tested using synthetic data and real measurements from the radiation generated by a modem PCB demonstrating its viability and usefulness. Index Terms—Electromagnetic inverse problem, genetic algo- rithms, measurement techniques, near-field/far-field transforma- tion, radiated emissions, source identification. I. INTRODUCTION I N the measurement of radiated emissions produced by a PCB or, in general, by any electric or electronic device, two problems typically arise: identifying the radiating elements and performing the measurement according to the standard EMC regulations. Detecting the radiating elements is essential in order to perform any EMC trouble-shooting, and can be a very time-consuming process. On the other hand, most radiated EMI regulations specify that measurements must be performed in an open-area test site (OATS) or, since the OATS is a very noisy environment, in a semi-anechoic chamber (SAC), the best environment for measuring radiated emissions. Since SACs have to be large (and therefore very costly) in order to fulfill standard EMC regulations (i.e., CISPR 22, FCC Part 15, etc., specify that measurements of radiated emissions must be Manuscript received July 3, 2000; revised January 31, 2001. This work was supported in part by the Direcció General de Recerca de la Generalitat de Catalunya (D.O.G.C. 30/12/1997) and by the Ministerio de Ciencia y Tecnologia of the Spanish Government under Grant DPI2001-1529-C02-01 and Grant DPI2001-1529-C02-02. J.-R. Regue a ´ nd M. Ribó are with the Department of Communications and Signal Theory at Enginyeria i Arquitectura la Salle, Ramon Llul University, Barcelona 08022, Spain. J.-M. Garrell is with the Department of Computer Science at Enginyeria i Arquitectura la Salle, Ramon Llul University, Barcelona 08022, Spain. A. Martín is with the LGAI Technological Center (LGAI), Bellaterra, Barcelona 08193, Spain Publisher Item Identifier S 0018-9375(01)10217-6. performed at a distance of 10 or 30 meters from the DUT), it is interesting to predict the far-field radiation of a DUT from measurements performed in a smaller (and cheaper) SAC, from near-field measurements. Both problems (source identification and far-field prediction from near-field measurements) have received considerable attention from the EMC community in the last few years, and a number of different approaches can be found in the literature. The problem of radiating source identification, or electro- magnetic inverse problem, consists of finding a current distri- bution that radiates an electromagnetic field equal to the mea- sured one. In the literature, approaches based on mathematical minimization techniques are usually found. In [1] the norm (minimum energy) is used, while in [2] and [3] the source iden- tification is performed using Lagrange Multipliers. In [4] a dif- ferent approach for determining the total radiated power and the radiation pattern of an electrically small source is presented. The method consists of finding three orthogonal magnetic and electric equivalent dipole moments using a three-loop antenna in whose center is placed the unknown small radiating source. Each loop is terminated symmetrically by two identical loads whose current is measured. This current is used to calculate the dipole moments, which are used to determine the total radiated power. In this paper a similar method useful for TEM cell mea- surements [5] instead of three-loop antenna is also reviewed. The technique developed in [4] and [5] is used in [6] to sim- ulate OATS measurements from GTEM measurements; an ad- ditional feature presented in this paper is the consideration of the multi-pole expansion over an infinite ground plane. The ap- proach is taken a step further in [7], [8], where the sources of radiation of a DUT are modeled by a set of electric and mag- netic dipoles which radiate the same near-field components in an anechoic environment. Both amplitude and phase of electric and magnetic fields are needed to find the set of dipoles. Their position, orientation and intensity is found in [8] by evolutionary strategies. In [9], where a method aimed at the characterization of areas of brain activity is described, a similar approach is used. The method is based on the substitution of those areas by dipole sources which generate the same potential as that measured by an electroencephalograph. The position, magnitude and orien- tation of the dipoles are found by a genetic algorithm. A number of different approaches are described in the litera- ture concerning the problem of far-field prediction from near- 0018–9375/01$10.00 © 2001 IEEE