COMPUTERS and their interface problems

Dec 1, 1998 12:00 PM, Bill Whitlock

Let me start by saying that I have mixed feelings about computers. There is no doubt that they can be truly wonderful productivity tools when they work properly, but when they fail, they quickly become a ball and chain, resulting in user frustration and unproductive downtime, unless you always wanted to be a computer consultant. At this point, the finger-pointing game usually begins, and the hardware guy says it is the software and the software guy says it is the hardware (or a power-line problem). It makes me wonder if automobiles would ever have become popular if their makers had the same relationship with gasoline providers. Anyway, my point is that the cause of a computer crash is usually a mystery to everyone involved, and a reboot seems to be the universal cure.

Although I think an awful lot of software is rushed to market before it is thoroughly tested (witness the steady stream of updates to fix it), I am convinced that many computer systems have interfacing problems serious enough to cause many, if not most, unexplained problems. Computer users and designers alike seem to think that digital signals are somehow exempt from the laws of physics, therefore immune to ground loops and other interference that is routinely dealt with in analog interfaces. The only thing different about interference in digital signal interfaces is that there is a threshold effect. Up to this threshold level, interference has no effect at all, but at a level ever so slightly higher, errors occur.

Ground noise Many engineers believe that all system noise is delivered with the incoming AC power and that simply cleaning it up will solve all the noise problems. For new facilities, they will specify special, expensive "technical" power distribution systems using every known technique in an attempt to ward off future problems. For existing facilities, they may modify power wiring with no regard as to just how (or if) it will actually solve the noise problem. It seems to me that selecting an effective cure requires some knowledge of how and where noise is actually entering the signal path.

As I have said before, significant noise voltage will always exist between the chassis grounds of any two devices operating from AC power, whether the devices are safety grounded or not. This must be accepted as a fact of life. This inter-chassis noise voltage is the dominant problem in most systems, not noise "picked up" by cables, as is so widely believed.

Power line noise current is coupled through equipment power transformers and flows in system ground conductors, creating ground noise voltages. This is true whether the power supply is conventional or switching or whether it is inside the equipment or outside. If power line RFI/EMI filters are used to keep noise generated inside the equipment from getting out, they further increase the noise coupling to the chassis. Because the coupling is capacitive, high-frequency components of power line noise are coupled much better than the 60 Hz line voltage itself. This "buzz-and-beyond," high-frequency power line noise is created by power supplies in electronic equipment, fluorescent or dimmer-controlled lights and intermittent or sparking loads such as switches, relays or brush-type motors.

When this power line noise current flows in the grounded conductor of an unbalanced (single-ended) signal cable between two devices, it causes a voltage drop, which directly adds to the received signal. This results in hum and buzz in audio signals, hum bars or sparkle bands in video signals, and glitches or otherwise unexplained crashes in digital systems. Because of this common-impedance coupling in the ground conductor, an unbalanced interface of any kind is simply a problem waiting to happen.

Although it is appealing to think that a big fat wire between the two chassis would short out any interfering noise voltage between them, it is simply not possible because even huge wires have a high impedance at high frequencies due to their unavoidable inductance. To make matters worse, at high frequencies, a building's power wiring behaves like a system of misterminated transmission lines gone berserk, reflecting high-frequency energy back and forth throughout the building's power wiring until it is eventually absorbed or radiated. The noise does not just follow the green ground wire back to the earth ground rod and disappear.

Power line isolation transformers, so-called balanced power, filters and conditioners presumably divert noise currents to ground. In most cases, however, there is no suitable low-impedance ground to dump the noise currents into (copper sheet flooring would make a wonderful ground plane for this purpose). Diverting them into the outlet's safety ground is not a good solution. Because equipment usually ties safety ground to signal reference ground and because the building's safety ground wiring has high impedance (inductance again), these power-line treatments often aggravate high-frequency system noise problems. If, however, any of these treatments are installed at the service entry panel where a low-impedance earth ground is available (and star connected to the building ground system), these treatments can substantially eliminate ingress of external power line noise.

For the same reasons, divert the power surge or spike caused by a nearby lightning strike with a hefty MOV suppressor installed at the service entry panel. If you feel it is necessary to protect against spikes or surges caused by other equipment on the user side of the service entry panel, use only a series-mode surge suppressor at the point-of-use outlet. Series-mode suppressors, such as the Surge-X line from New Frontier Electronics (www.frontierelec.com), do not dump spike currents into the outlet's safety ground. Such dumping by widely overused shunt mode MOV suppressors can cause severe ground voltage transients between the grounds of interconnected equipment. Reports of computer interface boards going up in flames when a real surge actually occurs are not uncommon. Ground-related noise problems of some sort are virtually guaranteed once the distance between system devices gets large enough or multiple devices are operated from different AC branch circuits. Computers and their peripherals are no exception.

Data interfaces Today, computer slots can provide analog audio and video I/O ports in addition to data I/O, and I am amazed at how often these are unbalanced connections. The oldest of the data transmission standards, RS-232, was first used in Teletype machines at a time when electromagnetic interference (EMI) was hardly the concern it is today. It eventually migrated into computers and is in surprisingly widespread use today. RS-232 is an unbalanced interface where, typically, receiver thresholds for a one and a zero are separated by as little as 300 mV, and response times are well under 1 ms.

More modern standards have tended toward smaller voltage swings and higher data rates. A good example is RS-422, which was developed by the telecommunications industry for short-haul modems. Perhaps its single most important feature is the differential signaling technique. This was hardly a surprise because decades earlier, telephone companies recognized the important advantages of differential (balanced) lines at rejecting noise, but even these interfaces are not bulletproof. Although, like balanced audio inputs, they are differential responding, typical RS-422 receivers cannot reject common-mode noise over about +/- 7 V.

I spoke recently with a fellow who does video post-production work and was experimenting with power-line conditioners to reduce noise. He said that he compares the file sizes of the digitized and compressed video for the indirect measurement of noise in the source analog video. In this case, I strongly suspect that low-level, high-frequency noise, which may be barely visible, is being coupled into the unbalanced video coaxial cable via common-impedance coupling, which is then treated by the digitizer and compressor as fine picture detail.

Guidelines Do not add unnecessary grounds. Additional grounding of equipment will generally increase circulating interference currents rather than reduce them. Of course, never disconnect or lift a safety ground or lightning protection ground to solve a problem; the practice is both illegal and dangerous.

Keep cables as short as possible. Longer cables increase the coupling impedance and the likelihood of interference. Serious noise coupling is common with 50 foot (15.2 m) or 100 foot (30.4 m) cables. Even much shorter cables can produce severe problems if there are multiple grounds. Never coil excess cable length.

Use surge suppressors and filters carefully and sparingly. Simply putting them everywhere can create more problems than it solves. Choose the type and location after sketching out signal and data flow. Power line paranoia and the shotgun problem-solving approach may result in costly but ineffective solutions.

Install signal or data line isolation devices. Isolation is the silver bullet that provides a barrier between different ground points and completely stops noise current flow in signal cables. It allows equipment that is grounded at widely separated points to exchange noise-free signals. I highly recommend them as low-cost data error insurance. Telebyte Technology (www.telebyteusa.com) makes a wide variety of data line isolators. Its model 268, for example, is an RS-232 opto-isolation module that costs about $100.

Even as expanding interconnections between audio, video, security, telephone and computer systems create more ground noise problems, most equipment manufacturers and so-called industry experts will continue to blame bad grounding and power-line noise for the shortcomings of their I/O interfaces. A ray of hope is the fact that many, if not most, LAN interface cards have built-in transformer isolation at the I/O connection.