Design Feature: September 28, 1995

Data-interface lines having different potentials relative to earth-ground connections can generate severe ground-current noise when connecting a computer with terminals in remote buildings or in different areas of one building. A few volts of potential difference can interrupt communications, and without current-limiting, this condition can damage equipment.

Noise on the interface cable can also disrupt sensitive analog circuits. Acting as an antenna, the cable couples high-frequency noise through the interface device, through the power supply, and into the analog circuits. Galvanic isolation eliminates this problem and other data-interface problems. Galvanic isolation develops the native potential difference across a controlled, high-impedance isolation barrier.

Designers often include isolation in a data interface when it doesn't require it or omit it when the interface does require it. There are several reasons to include isolation. For example, isolation reduces or eliminates ground-loop noise and common-mode voltage effects and the possible consequent damage to the interface and associated circuitry.

Data lines routed near power lines can inject high-amplitude common-mode signals into the data stream. Worse, the individual lines that originate or terminate near power circuits are subject to high currents and switching transients that cause destructive ground-loop noise and common-mode signals.

Consider, for example, an RS-485 interface between a data logger and a central computer. An adjacent electric motor can develop momentary ground-potential differences between the computer and data logger by drawing a surge of current at start-up. Without isolation in the data-communications path, this potential difference can alter data or damage equipment. Damage is likely if common-mode voltages exceed the maximum specified by the RS-485 interface standard.

Isolation also lowers overall system costs through the use of low-cost twisted-pair wiring rather than costly coaxial cable. Combining electrically isolated terminations with twisted-pair wiring can provide excellent noise immunity and reduce the overall cost of wiring. This cost becomes significant as the line length increases. Shielded pairs, coaxial cable, and fiber optics also provide excellent immunity at progressively higher costs, but comparable noise performance is available with low-cost twisted pairs carrying isolated differential signals.

Another reason to use isolation is to protect interface ICs and adjacent circuitry from damage caused by a catastrophic fault. For example, accidentally connecting data-interface lines to the ac power line or other sources of high voltage can cause damage. Also, lightning or other ESD events can cause voltage transients that damage interface pins and adjacent circuitry.

Isolation also ensures operator safety and complies with the strict patient-safety standards that apply to the electronic equipment in many medical applications. If this equipment has an isolated data interface, it can more easily meet applicable safety rules (see box, "Data-interface standards make designs uniform").

A complete galvanically isolated data-communications interface must convey power as well as digital signals across the isolation barrier. Alternatives for power transmission include using existing ac power if present or batteries on the isolated side of the barrier. The battery approach works only if the power consumption is low, the environmental conditions are mild, and the batteries are easily accessible. Otherwise, the interface must include a transformer to transfer power from the barrier's nonisolated, or logic, side to the isolated side. The common methods of signal transfer include using digital optocouplers, capacitors, and pulse transformers.

Optocouplers are the most common isolation devices for transferring digital signals. Also called optoisolators, photocouplers, and optically coupled pairs, these versatile devices require little design time. Operation is based on the emission and detection of light. The input voltage drives an internal light-emitting source, and an internal photodetector drives the output. Optically transparent insulation separates the emitter from the detector, which reside together in a light-excluding package.

Optocouplers for signal isolation usually combine an LED with a photodiode or phototransistor. The input LED and output photodetector are galvanically isolated from each other, and digital information always crosses the barrier in the same direction.

Inexpensive phototransistors are the most common detectors. They form optocouplers in which the current-transfer ratio (phototransistor output current divided by emitter input current) is often close to or even higher than 100%. Because phototransistor sensitivity is difficult to control, these optocouplers, such as the generic 4N26 family and the Sharp PC357 family, often have loose transfer specifications.

You typically specify timing with a 100Ohms load, although most applications have lighter loads. Phototransistors usually turn on much faster than they turn off, and this difference in response limits the maximum data-rate capability of an optocoupler.

Phototransistor turn-on is faster than turn-off for two major reasons. First, a phototransistor's effective resistance is smaller during turn-on than it is during turn-off and forms a much smaller time constant with the output-load capacitance. The same load capacitance is present at turn-off, but its charging current comes only through the load resistance at that time. Second, charge storage in the phototransistor's base delays its turn-off following the LED turn-off, producing a storage-time delay that adds to the phototransistor's turn-off time.

You can optimize any optocoupler for speed by adjusting its LED drive or by choosing a base-to-emitter resistor to eliminate the storage time. Neither approach is reliable, however, because the optocoupler parameters can change from lot to lot and even from device to device. A better solution is to add a small Schottky diode, such as the 1N5711, across the transistor's collector-base junction with the diode's cathode facing the collector.

The external diode's forward voltage is lower than that of the phototransistor's collector-base junction, so the diode absorbs nearly all of the excess photocurrent. During turn-off, the diode's negligible storage time reduces the circuit's turn-off delay to nearly zero.

Linear photodiodes have excellent high-frequency response, and, in most cases, an integrated linear photodiode produces an optocoupler whose speed is limited only by the LED. Linear photodiodes' main drawback is poor sensitivity, so the chip usually includes an amplifier to drive logic gates. Discrete linear photodiodes, on the other hand, have little application in optocouplers for RS-232C/422/485 interface circuitry.

Photo-IC optocouplers are easy to apply. They typically include a linear photodiode receiver and a matching amplifier on the same silicon. Some, such as the 6N135 and PC417, are simply large positive-intrinsic-negative (PIN) photodiodes that connect directly to the base of a relatively small npn transistor. You might expect photo-IC optocouplers to cause problems similar to those of a phototransistor, but two factors improve the photodiode's performance. First, the photodiode's capacitance is isolated from the output circuit, and, more important, the transistor prevents Miller multiplication of that capacitance. PIN photodiodes still introduce a storage delay when overdriven, but that delay is smaller than the that of a phototransistor because the output transistor is much smaller.

True photo-ICs, such as the generic 6N137 and the Sharp PC410, have small photodiodes, so they produce fairly low signal levels. These signals are amplified on the chip and presented to a digital output comparator that develops a nearly ideal logic signal. The only problems with these devices are their high cost and large amount of power they require.

The main drawback to optoisolation is the requirement for power on the barrier's isolated side. On the other hand, optoisolation provides superior bandwidth, ease of design, and the ability to transmit signals of arbitrarily low frequency, including dc. Compared with capacitive couplers, the optoisolator has dc capability, a smaller package, and lower cost for comparable levels of bandwidth and isolation.

Capacitively coupled digital isolators can provide extremely high bandwidths and very low power consumption. The model ISO150 from Burr-Brown, for instance, is an ultrahigh-speed, low-power, bidirectional device in a 24-pin DIP. It has a maximum data-transfer rate of 50 MHz and a quiescent current of 5 mA in receive mode. Two 0.4-pF capacitors per channel provide an isolation barrier of 1500V rms. The ISO150 maintains CMOS and TTL compatibility with no external components and lets you select the direction of transmission by toggling the receive/transmit pin.

The biggest drawback of capacitively coupled devices is their large package, which is necessary to accommodate the physical separation between capacitors that provides isolation. The ISO150 may be a good choice, however, if its low power and ultrahigh speed outweigh size considerations. It passes only logic levels across the barrier, so it isn't a complete interface. You must provide power on the far side of the barrier, along with level translation as needed.

When possible, it is usually cost-effective to pass digital signals through a transformer with no active devices at all. Transformers, however, have a low-frequency limit that prevents the transmission of dc. This limitation requires that the logic signal be ac symmetrical; that is, its dwell time at the positive limit must equal that at the negative limit. The coding schemes that meet this requirement are Manchester and Return to Zero.

The low-frequency limit for a transformer usually depends on the required signal level, so an important specification for pulse transformers is the volt-microsecond rating or ET product. The volt-microsecond rating represents the number of volts that you can apply to the winding for 1 µsec without causing the core to saturate. This parameter is primarily a measure of the number of turns in the winding and the amount and quality of iron in the transformer.

You can freely trade voltage and time. For example, you can use transformers such as the 500-1488 from BH Electronics in 5V systems with pulse widths as long as 5 µsec (10-µsec period=2x(25V-µsec/5V)). This device has an ET product of 25V-µsec. You can also use this transformer in 3.3V systems with pulses as wide as 7.58 µsec (a period of 15.15 µsec). These conditions correspond to minimum data rates of 200 kbps (100 kHz) and 132 kbps (66 kHz), respectively.

Stray capacitance in the windings usually dominates the high-frequency performance in a transformer. Core material can also be a limitation. Higher harmonics of the data frequency must pass with minimal distortion.

Transformers are passive components that cannot produce a power gain. Thus, the driver must overcome transformer loss by providing more power than the load demands. Therefore, maintaining worst-case logic levels through a transformer can be a problem, because manufacturers specify, test, and guarantee all driver ICs for a nominal resistive load.

The signal levels for RS-422 and RS-485 transmission easily pass through a transformer at relatively high data rates. Both of these standards define differential signals between pairs of signal lines, so the normal practice is to use a 1-to-1 center-tapped (CT) winding. The primary in this configuration connects between a pair of signal lines on the digital side, and the secondary connects between a pair on the isolated side. The secondary's center tap can optionally connect to isolated ground.

In contrast, RS-232 signals travel on a single wire referenced to ground. The voltage levels would seem to allow use of a transformer for isolation, but the asynchronous signaling allows arbitrarily long periods in the space state (logic low). These periods without a data transition produce quasi-dc voltages that cannot pass through a transformer, so you cannot use transformers for isolating an RS-232C interface.

Designing a complete isolated interface

Most interface designs entail trade-offs. You should compare the shorter design cycle and higher reliability of chips and chip sets, for example, with the lower hardware cost of a design-intensive circuit built with discrete components. A designer should also compare trade-offs among the costs of design, production, testing, agency approval, and reliability.

Figs 1 and 2 feature a transformer driver (IC₁) designed as the heart of an isolated RS-485 or RS-232 data interface. These fully tested circuits handle RS-485 data (Fig 1) and RS-232C data (Fig 2). A single 5V supply powers both circuits, which can withstand 1800V rms typ for 1 sec.

IC₁ enables galvanically isolated communications when combined with a linear regulator, CT transformer, several optocouplers, and an interface device. IC₁ comprises an RC oscillator followed by a toggle flip-flop that generates two 50%-duty-cycle square waves out of phase at half the oscillator frequency. The two signals then drive internal ground-referenced output switches.

The RS-485 interface device in Fig 1 (IC₃) should be a Maxim MAX483 for data-transfer rates as high as 250 kbps or a MAX485 for data rates as high as 2.5 Mbps. The application circuit then forms a complete isolated RS-485 interface for half-duplex communications. You can also implement full-duplex communications (simultaneous trans- mission and reception) by substituting a MAX488 for the MAX483 or a MAX491 for the MAX485. The optocouplers and interface device limit the resulting data-transfer rate. Table 1 lists the optocoupler and corresponding interface device suggested for data-transfer rates as fast as 2.5 Mbps.

IC₁ is also ideal for an isolated RS-232C data interface that requires more than four transceivers. IC₁'s 1W power-output capability can drive more than 10 transceivers simultaneously. Fig 2, for example, shows a complete 120-kbps, isolated RS-232C data interface with five transmitters and five receivers. The interface includes schematics for replacing the Sharp PC417 optocouplers with lower cost, generic 4N26 devices to achieve data-transfer rates to 19.2 kbps.

The CT transformer primary must have a sufficient ET product to prevent saturation at the worst-case combination of lowest selected frequency and highest supply voltage. The required ET product for half the primary is simply the product of maximum supply voltage and half the maximum period. The guaranteed minimum frequency for IC₁ (150 kHz with FS held low) translates to a maximum period of 6.67 µsec. With FS high, the guaranteed minimum frequency is 250 kHz, giving a maximum period of 4 µsec.

The transformer turns ratio must provide the minimum required output voltage at the maximum anticipated load with the minimum expected input voltage, making an allowance for worst-case losses in the rectifiers. Determining the turns ratio in this manner usually produces a much higher secondary voltage under conditions of high input voltage or light loading (or both), so an overvoltage condition is possible. The linear regulator in Figs 1 and 2 prevents overvoltage, but you can omit it the regulator in some applications.

Transformers used with IC₁ are usually wound on high-permeability magnetic material. To minimize radiated noise, use the shapes that provide a closed magnetic path (pot cores, toroids, and E/I/U cores). A typical core is the Philips 213CT0503B7--a toroid 0.190 in. in diameter and 0.05 in. thick. When operating this core at the 5.5V max supply voltage, its primary should have about 22 turns on each side of the center tap for a total of 44 turns. The result is a nominal primary inductance of approximately 832 µH and an ET product (with FS high) of approximately 18.3V-µsec. You can scale the secondary to produce any reasonable dc output.

Table 2 summarizes the transformer characteristics required in Figs 1, 2, and 3. In addition, Table 3 lists suggested manufacturers of transformers, transformer cores, and optocouplers.

Two ICs, a transformer, and four optocouplers form a complete isolated, dual RS-232C transceiver that can operate at data rates as high as 19.2 kbps (Fig 3). Replacing the 4N26 optocouplers with 6N136 devices boosts the maximum data rate to 90 kbps.

IC₁ connects to the nonisolated side of the interface, translating logic signals to and from the optocouplers, and IC₂ resides on the isolated, or cable, side, translating data between the optocouplers and the RS-232C line drivers and receivers. IC₁ also contains drive circuitry for the isolation transformer, which supplies power to IC₂ and the isolated side of the barrier.

Open-drain n-channel MOSFETs (D1 and D2) provide 150-kHz push-pull drive with 50% duty cycle to the external isolation transformer. The transformer's 1-to-1 CT turns ratio provides a 10V p-p output at the secondary. >Tables 2 and 3 list specificationsand suitable manufacturers for this transformer.

Combining functions into one package greatly reduces the cost and complexity of an isolated interface. Several such complete single-chip, isolated RS-232C and RS-485 interfaces are available in standard DIP packages. In a standard 40-pin plastic DIP, for example, two ICs, four optocouplers, four capacitors, two diodes, and a small transformer form a complete isolated RS-232C data interface (Fig 4). A single 5V supply powers both sides of the isolation barrier in this 9.6-kbps dual transceiver. Digital signals transfer through optocouplers, and power magnetically couples through the internal transformer. The UL-recognized MAX252A withstands 130V rms continuously or 1520V rms for 1 sec. For less severe environments, choose the MAX252B, which withstands 600V rms for 1 sec.

Another complete low-cost communications interface is available in a standard 28-pin DIP (Fig 5). The MAX1480A and MAX1480B are electrically isolated data transceivers that contain the necessary receivers, transmitters, optocouplers, power driver, and transformer for RS-485 or RS-422 systems. Each device transports signals and power across an internal isolation barrier. Power is transferred from the logic side (nonisolated side) to the isolated side by means of a center-tapped transformer, and signals cross the barrier through high-speed optocouplers.

A 5V supply on the logic side powers both sides of the interface in Fig 5. The ICs typically withstand 1600V rms for 2 min or 2000V rms for 1 sec, and their isolated outputs meet all RS-485/RS-422 specifications. Series resistors enable TTL/CMOS logic to drive the logic inputs, and resistive pull-ups enable the received-data output to directly drive any TTL/CMOS logic.

Full-speed slew rates in the MAX1480A drivers allow transmission rates as fast as 2.5 Mbps, but large-amplitude, high-frequency harmonics are evident at 150 kHz in an FFT plot of its driver-output waveform. Reduced-slew-rate drivers in the MAX1480B, on the other hand, minimize EMI and reduce the reflections that improperly terminated cables cause, allowing error-free transmission as fast as 250 kbps.

The RS-422/RS-485 standard specifies cable lengths as long as 4000 ft. At and beyond this maximum length, the ground potentials at either end can easily differ by several tens of volts. The potential can have any imaginable waveform, including dc, power-line ac, noise, and impulses of voltage and current. Source impedances are typically low, so connecting the two grounds can cause large, unstable, and unpredictable currents to flow. These currents can inject noise into sensitive instruments and damage the equipment in severe cases.

Isolation in a half-duplex, two-wire, bidirectional, party-line repeater system eliminates the possibility of interference and damage from ground potential differences (Fig 5). The two RS-485 transceivers (IC₁ and IC₄) isolate the repeater and its electrical environment from each segment of the network. The ICs also regenerate bus signals degraded by line attenuation or dispersion.

The idle state disables both transmitters and enables all receivers. Any device can send data to another device on the bus. Each data transition retriggers the one shot (IC₂ or IC₃), which ensures that the sending device remains enabled until no more transitions are detected. All receivers "hear" all data. If that is undesirable, the protocol must include an address field that allows the receivers to ignore data not intended for them.

Each node has the responsibility not to transmit when data is active on the bus and to resend any data that collisions corrupt. These situations are common in any party-line system. Moreover, you can separate the transmitters in Fig 5 by as much as 8000 ft, which represents more than 8 µsec in which the two transmitters can unwittingly transmit simultaneously.

The slew-rate-limited MAX1480B is useful for data rates as fast as 250 kbps, and the MAX1480A is useful for data rates as fast as 2.5 Mbps. If you don't need dual-port isolation, you can replace IC₁/IC₄ with a MAX485 for 2.5-Mbps operation or a MAX483 for 250-kbps operation.

Many data-interface circuits require little if any protection from I/O-line transients but may require protection if the bus represents an extended network. The first line of defense should be a four-diode bridge rectifier with zener-diode clamps, which prevents a line transient from pulling A or B outside the part's specified common-mode range. The zeners can be ordinary zener diodes or any of the Transzorb or MOV types.

If a prolonged short between an I/O line and the power source is possible, place positive-temperature-coefficient (PTC) resistors between IC₁ and the network. Their normal cold resistance (5 to 10Ohms) has a negligible effect on data transmission. In the event of a fault, high current in the diode bridge causes the PTC resistors to heat and dramatically increase their resistance, effectively isolating IC₁ from the fault.

For all isolated interfaces, Maxim recommends a pc layout in which the IC connections are optimized for minimal lengths and crossovers. For maximum isolation, nothing but the interface device itself should breach the "isolation barrier." All components and pc lines that connect to one side of the barrier should be physically separated from those that connect to the other side.

A shield trace between grounds on each side of the barrier can help intercept capacitive currents that might otherwise couple into the signal path. These shield traces should be present on all conductor layers for double-sided and multilayer boards. Maximize the isolation barrier's width wherever possible; Maxim suggests a clear space of at least 0.25 in. between ground and isolated ground.