EMI Potential Of Non-periodic Signals

Increasing speeds of LANs and microprvcessors require that non-periodic signals such as data line signals, address busses, data busses and control lines be analyzed when assessing the EMI potential of equipment. The power spectrum of these signals consists of the narrowband component which may related to the intrinsic timing information and the broadband component. These are characterized theoretically. The relative magnitude of the spectral components is dependent on the shape of the waveform and its rate relative to the 120 kHz bandwidth of the quasi-peak detector. Return-to-Zero (RZ) and NonReturn-to-Zero (NRZ) signals are presented at 1 Mbitls and 100 Mbitls rates. I. hX’RODUCrrON requires knowledge of the stochastic properties, or equivalently the energy spectrum, of the data stream in addition to the repetition rate, risetime, duty cycle and amplitude of the waveform. The resulting spectral signature is made up of both “broadband” and “narrowband” components as shown in Figure 1. This presents an additional factor in EMI modelThe increasing speeds of electronics technology necessitate the continuous development of analytic techniques to assess the EMI potential of new designs. The analysis of the EMI potential of non-periodic signals is important due to the increasing data rates over twisted pair LAN systems and the increasing operating speeds of microprocessors. This implies extending the complexity of signal modelling beyond the basic level of trapezoidal clock waveforms into the area of random variables and cyclostationary processes [ll. These various non-periodic signals such as data line signals, address busses, data busses and control lines will be referred to generally as “data signals” to distinguish them from clock signals which are usually analyzed in EMI modelling. Non-periodic signals have generally been ignored in EMT analysis. Their intrinsic randomness reduces the interference potential and the consequent threat to compliance with regulatory emissions requirements when compared with the clock signals usually in the same circuitry. Increasing data rates and processing speeds, accompanied by the success of design strategies to reduce the emissions potential of clock signals, have required that this source of emissions be more carefully examined. The spectral signature of a random or quasi-random signal is more complex than that of a clock signal. For the purposes of analysis it may be treated as two distinct parts; one which is due to the intended or idealized data stream and another which is directly attributable to the physical implementation in a circuit. In the latter category parasitic clock signals superimposed on the data line are the most common and significant spectral elements. Either or both of these contributions may be significant in an EMI analysis. The a priori analysis of an idealized non-periodic signal prCq.Sspn:4OMHz-6ohWz M.p.lOdB/div RClBW:l2&Hz Pa.Puk swP2oms Fig. 1. Emissions spectrum of data bus with and without f e d bead ling due to the particular response of the quasi-peak detector to these different signal components. When modelling systems for regulatory compliance it is important to take into account the response of the CISPR quasi-peak detector (QPD) [2] as is shown later in this paper. The second part of the analysis, that of the parasitic clock contributions, is based on accurate modelling of the electrical and physical properties of integrated circuits (ICs) [3,4]. This will not be dealt with in detail in this paper. The evaluation of the narrowband spectrum of an ideal data line signal is related to the study of clock recovery in data transmission. It is the intrinsic timing information in the data line signal which is found in the narrowband spectral components. For this reason the work of Bennett [51 and Bylanski [ 11 formed the theoretical background to this investigation. 11. BROADBAND SPECTRAL CONTENT The broadband component of the data signal is dependent on the power spectral density (pdf) of the data signal. Unlike the clock signal, this is a continuous function of frequency. If the data signal can be modelled as a stationary random process then the power spectral density may be related to its CH3169-0/92/0000-0066 $3.00 01992 IEEE 334 autocorrelation function R(T) [6]. S(O) = /R(7)ejord7 (1)