Medical Electronics Manufacturing Fall 1999
Electromagnetic Interference and Patient Isolation
Radio-frequency interference presents a major challenge to minimizing EMI in patient-connected devices.
William D. Kimmel and Daryl D. Gerke
In patient-connected medical electronics, immunity to external radio-frequency interference (RFI) is one of the toughest electromagnetic interference (EMI) problems to handle. Patient-connected devices are often placed in high-noise environments and often perform life-critical functions. And one of the most difficult EMI issues to address is protecting analog signals from RFI generated by a nearby handheld radio or cellular phone.
Three main elements contribute to the problem. First, physiological processes emit very weak signal levels, which makes the signals vulnerable to RFI. Second, the input signals are impossible to shield adequately because the end of the cable connected to the patient cannot be terminated. Third, leakage current limitations restrict capacitive filtering to the ground, thereby prohibiting the most effective method of RFI filtering.
So the design of a patient-connected device is laced with compromises. Accordingly, designers must identify solutions at each level of the design. This article identifies the problems and the constraints and provides some methodologies to address them.
The problem originates with an RF energy incident on the antenna, which in this case is the patient cable. At the signal pins, the resulting interference can easily be 10 V, which will overload most input amplifiers. Two primary areas of vulnerabilitythe sensor and the input amplifiermust be addressed. The sensor is a problem only if it is nonlinear, which would be the case with a pulse oximeter (Figure 1). The diode acts as a detector, rectifying the RF and producing a demodulated output. The input amplifier, being quite sensitive and always subject to RF overload, either results in signal demodulation or drives the amplifier out of its dynamic range (Figure 2). When these two situations are compensated for, the problem is pushed back to the isolation barrier where the signal levels are much higher.
Figure 1. Effect of rectifying junction at patient sensor.
Figure 2. Effect of high-frequency common mode on input amplifier.
No matter which circuit is affected, the solution inevitably involves either shielding or filtering, both of which are standard electromagnetic compatibility techniques. Shielding all of the circuits would certainly solve the problem, but that solution would involve putting the patient inside the shield, a generally undesirable approach. As long as the patient is exposed to the interference, the cable ends are also exposed. Figure 3 shows the cable: the open loop area produces differential mode (DM) interference, and the open cable end produces common mode (CM) interference. In principle, the cable can be shielded until it reaches the point where the cable splits into individual wires, but shielding beyond that point is useless. The loop that is formed acts like a good antenna, whether or not the individual wires are shielded. This is the best shielding that can be achieved, even assuming the cable shield can be grounded to the enclosure (usually not an option). So, any cable shielding inevitably involves some compromises.
Figure 3. Patient-to-equipment connection.
The next line of defense is to filter the interference. Because the frequencies measured in physiological processes are much lower than the RFI, filtering doesn't compromise the signal quality. There are two ways to achieve effective filtering: high-impedance series elements (inductors or resistors) block interference currents, and low-impedance shunts (capacitors) divert interference currents from the circuit. Here, the terms high impedance and low impedance are relative to the impedance of the circuit in question. The greater the impedance mismatch between the filter element and the load or source, the more effective the filtering is. Because input amplifiers generally have high input impedance, low-impedance shunt capacitors make ideal filter elements. However, getting high series impedances relative to the amplifier inputs is impossible. It is usually necessary to use a combination of shunt and series elements.
It is preferable to divert the currents to the enclosure, which keeps them off the circuit board entirely, but leakage current limitations prevent adequate filtering. Although filtering to the circuit board ground is still recommended, doing so effectively diverts the currents to the circuit ground, which presents another set of problems.
So, lacking the ability to shunt the interference currents, designers are left to put in a high-impedance block. Considering that this block drives a high-impedance amplifier, it takes a lot of series inductance to filter anything. Realistically, an impedance block does not even provide enough inductance to be marginally effective.
The solutions involve cable-shield grounding and signal-line filtering. The first option addresses the cable shield alone. The second solution focuses on filtering. Because the two methods do not function independently, this article also discusses combining both to obtain the most effective control.
Ground the Cable Shield. The cable shield is a key component in successful design. It is difficultperhaps impossibleto meet RF immunity requirements without a cable shield. More importantly, an ungrounded cable shield is useless for protecting against RFI. If it is impossible to ground the shield to a metal member, then there is no shield. In this article, grounding always refers to a local reference, which may or may not be connected to earth ground.
There are two basic options for terminating the shield: to the enclosure ground or to the isolated ground. The optimal termination is to the enclosure, assuming the enclosure is conductive (Figure 3). This practice inserts capacitance between the patient lead and the ground. However, this could produce an excessive amount of capacitance. If the enclosure is not metallic, a metal plate or film can be used at the bottom of the enclosure to form a partial shield, which can then be isolated. This method is not highly effective, but it does provide some relief for devices that are close to meeting RF immunity requirements. Although terminating the shield through a capacitor can retain much of the effectiveness, the voltage rating necessary is likely to result in a minimally effective capacitor. Subject to leakage limitations, the capacitor should be as large as possible.
Figure 4. Shield terminated to isolated ground.
The cable shield can also be terminated to the isolated ground (Figure 4). With this technique, addressing leakage currents is not an issue. Such termination, however, works only if the device has a ground surface under the entire isolated area. The greatest effectiveness occurs when the isolated area is enclosed in a six-sided shield, leaving only openings that are necessary to cross the isolation barrier. Using a metal cover for five sides and the isolated ground plane for the sixth side is the easiest way to provide this shield. If the design uses a two-sided circuit board (no ground plane), it definitely requires a metal cover on the bottom of the circuit board to provide the ground plane. It will probably also need a cover on the top to complete the shield.
The cable shield termination must be connected directly to its ground, and connection must surround the perimeter. Using a connector wire seriously degrades the performance of the shield and can even convert the shield into an efficient antenna under resonant conditions. Sadly, a single-wire termination is the rule, rather than the exception, and it usually creates problems.
This termination technique, along with the filtering techniques described below, protects the sensitive input amplifier as well as all of the circuits on the isolation area, moving the problem downstream to the isolation barrier. A small capacitor from the isolated ground to the enclosure ground helps divert interference currents, minimizing the interference at the isolation barrier.
With or without a cable shield, a patient-connected device still requires plenty of filtering. The presence of the cable shield merely makes the filtering process less onerous. As a minimum, the input amplifier must be filtered. If the sensor is nonlinear, it needs filtering as well. Starting with the input amplifier, both CM and DM interference must be suppressed because generous quantities of both will be present. CM interference, however, is much more troublesome to handle.
Figure 5. Filtering common-mode and differential-mode interference.
Typically, op-amps are used to suppress considerable low-frequency CM interference for input devices. At higher frequencies, the CM rejection ratio degrades, largely due to asymmetries in the chip, the printed circuit board (PCB), and the cable. The best strategy is to attack the CM currents first, then address DM filtering at the input to the chip (Figure 5). Filtering DM currents last enables the filter to function for both the original DM noise and any converted CM noise.
Beginning with the cable end, it is best to implement a CM choke, which inserts high impedance in line to minimize the interference current on the signal wires. It is much easier to achieve high impedances with CM chokes than with individual DM chokes. The choke should be followed by an individual capacitor on each line to the ground to divert the remaining current away from the amplifier. The capacitors should be terminated to the enclosure ground whenever possible, but they can be terminated to the isolated ground as a second option. Finally, a DM choke should be inserted using a pair of ferrite inductors or resistors and a line-to-line capacitor.
The principles involved in filtering are straightforward, but successful execution requires that designers consider the hidden schematic, which includes the parasitic elements in each component and the coupling factors on the PCB. All capacitors have some series inductance, and all inductors have some parallel capacitance, commonly with resonant frequencies of 10100 MHz. In addition, all components (including PCB traces) have electric and magnetic fields that extend beyond the footprint of the component itself, coupling to other elements. These factors cannot be ignored at high frequencies.
Therefore, for a filter to work, designers must assess the parasitic elements in the filter components to ensure that they are working as intended over the entire frequency range of interest. This could require using multiple elements to obtain the necessary frequency range, as well as laying out the components so as to minimize coupling and asymmetries.
Impedances. Because they are primarily capacitive, inputs to amplifiers are quite high impedance. Cable impedances range from near 0 to a maximum of about 100 , depending on resonance conditions. The biggest problems come at resonance, where currents can get very large. Inserting a series impedance (CM choke) of between 100 and 500 should provide the right impedance level. An impedance of less than 100 is insufficient, and an impedance of more than 500 is probably excessive because parasitic capacitance prevents impedances from going above that level anyway.
In the frequency range of interest, capacitive impedances should be sized to have an impedance of much less than 100 , preferably of about 1 . This can usually be satisfied with a 0.0010.01-µF capacitor, which provides an impedance that can result in excessive leakage currents if placed between the signal and the enclosure ground.
The amount of attenuation needed varies widely, so experimentation is necessary to determine the correct level (it will almost assuredly take more than estimated). An attenuation of 40 dB (100:1) should be appropriate. Lesser attenuation will probably be inadequate, and greater attenuation will require multiple-stage filters and meticulous layout.
It is important to arrange components to minimize stray capacitance and inductance. This means placing them in line, avoiding a serpentine alignment with either the filter elements or other nearby elements. CM chokes are best placed away from the ground plane. The goal is to get high impedances, so close proximity to the circuit ground should be avoided. Wound inductors placed close together in parallel will couple to reduce the CM rejection. Components placed close to other components, traces, or ground planes will increase the stray coupling.
Patient sensor. In pulse oximeters, the sensor is a highly nonlinear photodiode. Such cases require inserting additional DM filtering to minimize the DM currents. DM currents in the patient cable arise primarily from loop areas in the patient cable. Fortunately, in pulse oximeters, the two wires can be closely spaced all the way to the sensor, so it is feasible to keep loop areas small. Ideally, the cable shield should be extended to the sensor. Eliminating DM currents is one thing an ungrounded shield does accomplish. An alternative would be to twist the wires to achieve loop-area cancellation.
Figure 6. Differential-mode filtering for a nonlinear device.
Even after applying these techniques, a certain measure of DM currents must be blocked by filtering to prevent rectification. Basically, this involves inserting filter elements to linearize the impedance facing the diode sensor (Figure 6).
Combining Cable Shields and Filters
Most patient-connected applications require both shields and filters to provide adequate immunity to RFI. In addition, some devices contain more wires than just the sensitive input signal lines in the cable.
Mixed input and output lines. As a rule, output lines (whether power or signal source) are less susceptible to RFI than input lines; however, they still need to be filtered. They are susceptible to interference to some extent, so they cannot be left unattended. Outputs can also inadvertently serve as antennas, coupling to input lines inside the shield.
CM chokes and output signal lines. This article has addressed the use of CM chokes for the input signal lines on a stand-alone basis. The question is, should CM chokes be combined, putting both input and output lines on the same choke? CM chokes function more efficiently with more wires included; however, with the significant differences in input and output impedances, such a practice is risky because it can result in transforming interference onto the input lines rather than blocking it. Therefore, separate CM chokes should be used.
CM chokes and cable shield termination. One final issue is to determine whether to include the cable shield in the CM choke. After all, that is effectively what is done when a ferrite is clamped onto the cable. This practice reduces the overall CM interferenceperhaps protecting nonisolated circuitsbut as noted previously, it can also result in transferring interference from the shield onto the signal lines. The shield should be excluded from the choke to avoid this unwanted outcome.
Low-level analog circuits are vulnerable to RFI, a problem that is exacerbated by leakage current limitations. Suppressing EMI at the patient connection requires considerable care and experimentation. Following the priorities listed below generally provides the shortest route to successful test results:.
- Terminate the cable shield ground to the enclosure, if possible, or to an isolated ground. If the cable shield is terminated to an isolated ground, however, a small capacitor from the isolated ground to the enclosure ground should be used to help reduce CM interference.
- Use a shield over the isolated area for best results.
- Filter CM currents first; then filter DM currents.
- Select filter components that are functional over the desired frequency range. In particular, be alert for component resonances.
- Lay out the circuit boards to minimize coupling from component to component.
William D. Kimmel, PE, and Daryl D. Gerke, PE, are principals in Kimmel Gerke Associates Ltd., an electrical engineering consulting firm specializing in EMI/EMC issues with offices in Phoenix and St. Paul, MN. They can be reached at www.emiguru.com.