Radiated EMI Recognition and Identification from PCB Configuration Using Neural Network

Transcription

1 PIERS ONLINE, VOL. 3, NO., Radiated EMI Recognition and Identification from PCB Configuration Using Neural Network P. Sujintanarat, P. Dangkham, S. Chaichana, K. Aunchaleevarapan, and P. Teekaput Faculty of Engineering, Chulalongkorn University, Bangkok, Thailand Electrical and Electronic Products Testing Center, NECTEC, NSTDA, KMITL Bangkok, Thailand Abstract In this paper the application of neural network (NN) in EMI radiated measurement is proposed to recognize and identify the basic PCB configuration. The emission of electromagnetic (EMI) noise form printed circuit board (PCB) is studied. The different kinds of PCB shape are used for produce electromagnetic field. The fields are measured using near-field probe with termination load and computation using finite element method (FEM). After measurement by near-field probe and simulation by FEM then, is a thinning process. The images acquired from thinning process will be analyzed to calculate number of junction point, end point, cross point, ratio of black and white, position of end point, vertical line and horizontal line, respectively then pass to NN. The trained NN can identify emission pattern successfully. DOI: 0.59/PIERS INTRODUCTION This paper presents the radiated EMI recognition and identification for PCB configuration measured by near-filed scanner and simulated by Finite Element Method (FEM). Applying by used electromagnetic field radiated from PCB that has many difference shapes to generate electromagnetic field. Then used the neural network to identify the kind of PCB shape. The learning process is achieved by using FEM of the radiated field from difference shapes of PCB.. PCB CONFIGURATION In this paper, six kinds of basic PCBs shape are used to produce radiated EMI. PCB configurations (Fig. ) included. Strip line(a), radius-shape(b), L-shape(c), H-shape(d), X-shape(e), 7-shape(f) are prepared and used in this research. The trace of the PCB is adjusted to match a 50 Ω resistance at the end of the terminal. Ferrite clamps minimize the influence of the signal cable entering the test-site ensuring that the signal is greater than 0 db, which is below the emissions measured from the test PCB. The ferrite clamps are included in the test boards. (a) (b) (c) (d) (e) (f) Figure : Shown PCB configurations. 3. FINITE ELEMENT METHOD (FEM) This paper used Finite Element Method (FEM) based on Maxwell s equation including the eddy current and displacement current terms []. Finite conductivity, arbitrary configuration and arbitrary dielectric constant may be considered. In this analysis of emission from PCB, the required input data is only physical data, such as conductivity, configuration, dielectric constant, permeability, etc., and no equivalent circuit or characteristic impedance values are required as input

2 PIERS ONLINE, VOL. 3, NO., data. The analysis based on Maxwell s equations. For electrolysis and computation of resistances of grounding plates we have So E = grad(v ) () J = σe () divj = q (3) div(σ grad(v )) = q (4) where E is the electric field, V is the electric potential, σ is the conductivity and q is the current source. For discussing the design for PCB trace on ground structures including frame ground. The method of FEM for calculating E-field on PCB by separating a PCB area become a very small elements (Fig. ()). The flux density of strip line PCB is shown in Fig. (), the maximum of flux density appeared at the center of the strip line, then there decrease, continuously when the distance between the strip line and measured point in y-direction increase. Then, the flux density increase again at the edge of the PCB. This phenomenon occurred because the coupling mechanism of the ground plane to the strip line PCB. Fig. (3) shown that the flux density of the radius shape is highest amplitude at the edge of the PCB trace and it induce into inside of the loop. In case of L-shape, H-shape and 7-shape of PCB, the flux density appeared at the trace and induced into the inner side of the shape, the highest amplitude occurred at the each corner of the trace (Fig. (4), Fig. (5) and Fig. (7)). Fig. (6) shown the flux density of the X-shape PCB, the maximum E-field appeared at he cross point of the trace. () (3) (4) () (5) (6) (7) Figure : Simulated by Finite Element Method (FEM). 4. NEAR-FIELD SCANNER MEASUREMENT The near-field measurements are performed in a shielded room with the approximate dimensions 3 5 m 3. The kinds of experimental equipment are an EMI receiver, pulse generator, dipole antenna, ferrite clamps and a 4 db pre-amplifier. The measurement is carried out using a Quasi- Figure 3: The near-field measurement configuration. Peak detector with a 00 khz resolution bandwidth. A generator producing a 30 MHz square-wave with a 50 ns rise time, 5 V peak-to-peak amplitude, 0.5 duty cycle, and a 50 Ω input resistance

3 PIERS ONLINE, VOL. 3, NO., constitutes the digital source. The magnetic near-field measurement setup is illustrated in Fig. 3. The probe or loop antenna is positioned by a computer controller []. The data are acquired at intervals of cm in the X-direction and Y -direction, respectively. Figure 4: The measured results. The E-field of 30 MHz frequency of the input signal are measured and plotted in 3D method are depicted in Fig. 4. The measured results also takes into account the averaging effect of the electric field over the PCB trace. The measured results are good in agreement to the FEM calculated results. 5. THINNING PROCESS The thinning of set A by a structuring element B, denote A B A B = (A B) (A c B) (5) where B = W B, W is the local background, A is an image, B is sequence element and x is don t care x x x x x x x x x x x x x x x x (B) (B) (B3) (B4) (B5) (B6) (B7) (B8) Figure 5: Sequence of element used for thinning. A B = A (A B) = A (A B) c (6) {B} = {B, B, B3,..., B8} (7) A {B} = ((... ((A B) B)...) B8) (8) The process is to thin by one pass with B, then thin the result with one pass of B, and so Figure 6: Thinning from simulation. Figure 7: Thinning from measurement.

4 PIERS ONLINE, VOL. 3, NO., on, until A is thinned with one pass of B8. The entire process is repeated until no further changes occur. The results of thinning from simulation by FEM and measuring by near field scanner shown in Fig. 6 and Fig. 7, respectively [4]. 6. THE NEURAL NETWORK The recognition method used in this research is using features of PCB (number of junction point, end point, cross point, ratio of black and white, position of end point, vertical line and horizontal line) for classify PCB. The two layer feed-forward backpropagation network is created. The first layer has 00 log sigmoid neurons, the second layer has 6 purelin neurons, log sigmoid is a transfer functions calculate a layers output from its network input and return value between 0 and. Bias node weight Linear function Target Input data n n weight Input layer (i) Hidden layer (j) Output layer (k) Sigmoid function 0.5 Threshold Figure 8: The Neural Network architecture. The Neural Network consists an input (i layer), a hidden (j layer) and an output (k layer) is adapted to implement the proposed application. The capability and accuracy in the estimation depends on the number of input, hidden, output node, etc. The backpropagation network can be thought of as a converter having many inputs and outputs. The learning process begins with feeding the input data into the NN input layer and assigning the NN target for the output layer. The network converts the input data according to connection weights. The calculated output in each hidden node is converted to the output layer using the sigmoid function. The summation of each sigmoid function in the hidden layer is the calculated output node. The calculated result from the output layer is converted to the output data and used for comparing with the NN target using the linear function. From this point, the sum-square error is obtained and used for stopping the learning process. The backpropagation processes begin when the sum-square error is greater than the maximum error. The output data in the output node is back propagated to the hidden layer and the input layer, respectively. During propagation, connection weights are adjusted until the network sum-square error is less then maximum error. When the learning process is finished, the weights are obtained and the NN architecture is defined. The trained NN is ready to identify or predict outputs related to the input data. Table : The training data sets of the NN for learning process. Input data for NN NN Target PCB Position Junction End Cross Vertical Horizontal Conf. Ratio of end point point point Line Line point T T T 3 T 4 T 5 T 6 E I L O X

5 PIERS ONLINE, VOL. 3, NO., Spectrum Near-field probe PCB Pulse generator Near-field scanner Spectrum Thinning Your type is 7-shape FEM Thinning PCB indentify input hidden output Figure 9: Step of process. 7. RESULTS AND CONCLUSIONS The aims of this paper are recognize and identify the PCB configurations from emission spectrum. After measurement by near-field scanner (30 MHz) and FEM is thinning process then pass to neural network. (Fig. 9). Two layer backpropagation neural network is created and trained by used data from Table, the six bits digital output is considered. When the training process is finished the unknown source of radiated emission are fed to the NN input layer for identify PCB configurations. The NN successfully to identify emission pattern from PCB. ACKNOWLEDGMENT This work was supported by National Science and Technology Development Agency (NSTDA), Thailand. REFERENCES. Roczniak, A., E. M. Petriu, and G. I. Costache, 3D electromagnetic field modeling base on near-field measurements, IEEE Instrumentation and Measurement Technology Conf., Brussels, Belgium, , June Laurin, J. J., Z. Ouardhiri, and J. Colinas, Near-field imageing of radiated emission source on printed-circuit boards, IEEE International Symposium on Electromagnetic Compatibility, , Aunchaleevarapan, K., K. Paitoonwatanakij, W. Khan-Ngern, and S. Nitta, Novel method for predicting PCB configurations for near-field and far-field radiated EMI using a neural network, IEICE Transactions on Communication, Vol. E86-B, No. 4, , April Gonzalez, C. R. and E. R. Woods, Digital Image Processing, Prentice Hall, New Jersey, 00.