As they operate, digital circuits constantly switch the state of lines between a high-voltage
level and a low-voltage level to represent binary states. As shown in Figure 4.1a, the resulting
time-domain waveform on any single line of a digital circuit can thus be idealized as
a train of trapezoidal pulses of amplitude (either current I or voltage V) A, rise time tr, fall
time tf(between 10 and 90% of the amplitude), pulse width τ (at 50% of the amplitude),
and period T.
The Fourier envelope of all frequency-domain components generated by such a periodic
pulse train can be approximated by the nomogram of Figure 4.1b. The frequency spectrum
is composed mainly of a series of discrete sine-wave harmonics starting at the fundamental
frequency f0 = 1/T and continuing for all integer multiples of f0. The nomogram
identifies two frequencies of interest. The first is f1, above which the locus of the maximum
amplitudes rolls off with a 1/f slope. The second, f2, is the limit above which the locus rolls
off at a more abrupt rate of 1/f2. These frequencies are located at
| |  |
and
where t is the faster of (tr, tf).
The envelope of harmonic amplitude (in either amperes or volts) is then simplified to

For nonperiodic trains, the nomogram must be modified to account for the broadband
nature of the source. To do so, a nomogram of the spectral density envelope of the signal
can be defined for a unity bandwidth of 1 MHz by
Depending on its internal impedance, a circuit carrying such a pulse train will create in its
vicinity a field that is principally electric or magnetic. At a greater distance from the
source, the field becomes electromagnetic, regardless of the source impedance. If there is
a coupling mechanism, which can be either conduction or radiation, some or all of the
frequency components in the digital pulse train’s spectrum will be absorbed by some
“victim” receiver circuit.
To illustrate the magnitude of the problem, imagine a medical instrument’s main circuit
board, consisting of a CPU, some glue logic, and memory ICs, that has been housed in an
unshielded plastic case. Let’s assume that at any given time, a number of these ICs are toggling
states synchronously, at a frequency of 100 MHz, for instance. Furthermore, assume
that the total power switched at any given instant during a synchronous transition is
approximately 10 W. Now, in a real circuit, efficiency is not 100%, and a small fraction of
these 10 W will not do either useful work or be dissipated as heat by the ICs and wiring,
but rather, will be radiated into space. Assuming a reasonable fraction value of 10-6 of the
total switched power at the fundamental frequency, the power radiated is 10 μW.
Now, let’s assume that an FM radio is placed at a distance of 5 m from the device. The
field strength E produced by the 10 μW at this distance may be approximated by the
formula

Considering that the minimum field strength required for good reception quality by a typical
FM receiver is approximately 50 dBμV/m, the radiated computer clock would cause
considerable interference to the reception of a radio station in the same frequency. In fact,
interference caused by the computer of this example may extend up to 50 m or more away!
From the past discussion, it is easy to conclude that a first method for reducing radiated
emissions is to maintain clock speeds low as well as to make rise and fall times as slow as
possible for the specific application. At the same time, it is desirable to maintain the total
power per transition to the bare minimum. Transition times and powers depend primarily
on the technology used. As shown in Table 4.1, the ac parameters of each technology
strongly influence the equivalent radiation bandwidth. In addition, the voltage swing, in
combination with the source impedance and load characteristics of each technology, determines
the amount of power used and thus the power of radiated emissions on each transition.
Figure 4.2 shows how the selection of technology plays a crucial role in establishing
the bandwidth and power levels of radiated emissions that will require control throughout
the design effort.
Another problematic circuit often found in medical devices is the switching power supply.
Here, high-power switching at frequencies of 100 kHz and above produce significant
harmonics up to and above 30 MHz, requiring careful circuit layout and filtering. Fully


integrated filters are available for dc power lines. For example, Figure 4.3 shows the way
in which muRata BNX002 block filters are used to filter the raw dc power outputs produced
by two C&D Technologies’ HB04U15D12 isolating dc/dc converters. In the circuit,
each dc/dc (IC1 and IC2) produces unregulated 24 V (±12 V if the center-tap common is
used), which is isolated from the +15 V dc power input by an isolation barrier rated at
3000 V dc (continuous, tested at 8 kV, 60 Hz for 10 s). The outputs of the dc/dc converters
are filtered via filters FILT1 and FILT2, which internally incorporate multiple EMI filters
implemented with feed-through capacitors, monolithic chip capacitors, and ferrite-bead
inductors. Each of these filters attenuates RF by at least 40 dB in the range 1 MHz to
1 GHz. C1/C4 and C7/C10 are used to reduce ripple, and the circuits following these
capacitors are linear regulators that yield regulated ±24 V at 50 mA to the applied part for
which this isolation power supply was designed.
Another filter worth mentioning is muRata’s PLTxR53C common-mode choke coil.
This family of modules is ideal for suppressing noise from a few megahertz (1 to 5 MHz,
depending on the model) to several hundred megahertz (10 MHz to 1 GHz, depending on
the model) from dc power supplies. This module is useful in suppressing noise radiated
from the cable connecting a device to an external wall-mounted or “brick” ac adapter.
As they operate, digital circuits constantly switch the state of lines between a high-voltage
level and a low-voltage level to represent binary states. As shown in Figure 4.1a, the resulting
time-domain waveform on any single line of a digital circuit can thus be idealized as
a train of trapezoidal pulses of amplitude (either current I or voltage V) A, rise time tr, fall
time tf(between 10 and 90% of the amplitude), pulse width τ (at 50% of the amplitude),
and period T.
The Fourier envelope of all frequency-domain components generated by such a periodic
pulse train can be approximated by the nomogram of Figure 4.1b. The frequency spectrum
is composed mainly of a series of discrete sine-wave harmonics starting at the fundamental
frequency f0 = 1/T and continuing for all integer multiples of f0. The nomogram
identifies two frequencies of interest. The first is f1, above which the locus of the maximum
amplitudes rolls off with a 1/f
More >>