Troubleshooting Analog Circuits--Preventing Material and Assembly Problems: PC Boards and Connectors, Relays and Switches (part 2)

Home | Articles | Forum | Glossary | Books

(cont. from part 1)

Kelvin Connections Improve Measurement Accuracy

The usefulness of Kelvin contacts and connections is not widely appreciated. In fact when Julie Schofield, an Associate Editor at EDN, asked me some questions about them recently, I was surprised to find very little printed reference material on the subject. I looked in a few dozen reference books and text books and didn't find a decent definition or explanation anywhere. I did find some "Kelvin clips," which facilitate Kelvin-connected measurements, in a Keithley catalog. I also found some Textool socket data sheets, which mentioned, in a matter-of-fact way, the advantages

of Kelvin contacts. I'll try to explain their usefulness in more detail here, since neither Julie nor I could find a good description in any technical encyclopedia.

Perhaps the most common use of Kelvin connections is the remote-sense technique.

Kelvin connections and sockets let you bring a precise voltage right to the terminal or pin of your circuit under test. If you don't control the supply voltage precisely, you might start failing parts that are actually meeting their specs.

For example, let's say we want to test the load regulation of an LM323, a 5-V regulator, when Vi" is held at exactly +8.00 V, and the load changes from 5 mA to 3.00 A (Fig 4). In this circuit, there are four pairs of Kelvin connections at work.

The first pair is located at the power supply's output. This programmable supply's remote-sense terminals permit it to maintain an accurate 8.00-V output right up to the pin of the DUT. This is commonly called "remote sense," when you are in the power-supply business, but actually it represents a Kelvin connection. This is important because if the 8.00-V supply dropped to 7.9 or 7.8, that would be an unfair test.

The second Kelvin connection in Figure 4 is located at the output of the DUT. In order for you to observe the changes in V,,, as you apply various loads, the Kelvin contacts provide Force leads for the 3A of output current. They also provide Sense leads, so you can observe the DUT's output with a high-impedance voltmeter. Note that there are two Sense and two Force connections to the ground pin of the DUT.

You don't really need all four contacts--you can tie both. Force leads together and also both Sense leads together. You can do that because there is no significant current flowing in the Sense lead; and in the Force lead, we don't care how much current flows, nor do we care exactly what the voltage drop is.

The op amp in Figure 4 forces the DUT's output current through the Darlington transistor and then through a 0.1 ohm precision resistor. The only way to use a 0.1 –ohm resistor with any reasonable accuracy and repeatability is to use the 4-wire (Kelvin) connections as shown. The op amp can force the upper Sense lead to be precisely 300 mV above the lower Sense level, even if the lower end of the resistor does rise above ground due to various IR drops in wires or connectors.

There are several places in this circuit that we could call ground, but the only ground we can connect to the 300-ohm resistor and get good results is the Sense lead at the bottom of the 0.1 ohm resistor. If you connect the bottom of the 300-ohm resistor to any other "ground," shifts in the IR drops would cause relatively large and unpredictable and unacceptable shifts in the value of the 3-A current-in other words, Inaccuracy and Trouble. So, when you're running large currents through circuits, think about the effects of IR drops in various connectors and cables. If you think IR drops will cause trouble, maybe Kelvin connections can get you out of it.

Figure 4's fourth Kelvin connection is hidden inside the LM323 5-V regulator, which has separate Force and Sense connections to the output terminal. A fifth Kelvin connection is also concealed inside the current-limit circuitry of the regulator. Here, the device senses the load current with a 4-wire, Kelvin-connected resistor and sends that voltage to the current-limit sense amplifier.

The use of Kelvin sockets is not confined to large power transistors or high-current circuits. Consider a voltage reference with 2 mA of quiescent current. If you're trying to observe a 1 -ppm stable reference and the ground connection changes by 5 m-Ohm (which most socket manufacturers do not consider disastrous), the 10-uV shift that results from this change in ground impedance could confound your measurements. If you want to avoid trouble in precision measurements, avoid sockets or at least avoid sockets that do not have Kelvin contacts. Lord Kelvin-William Thomson before he was appointed a baron-did indeed leave us with a bag full of useful tools.

Figure 4. By using Kelvin connections you can avoid measurement errors caused by IR drops in the circuit that you're trying to measure, and in its connections. In this circuit, there are (at least) four pairs of Kelvin connections.

Avoid Cold-Soldered Joints

I have a few comments to add about solder; most of the time we take it for granted.

You'll normally use ordinary rosin-core tin-lead solder. If you avoid jiggling the soldered joint as it is cooling, you won't get a cold-soldered joint. But you should know what a cold-soldered joint looks like and how much trouble one can cause in a critical circuit. I feel sort of sad that today's young people aren't building kits for electronic equipment. In the old days, you could learn all about cold-soldered joints before you got into industry, by building a "Heathkit" or a "Knightkit." I built several of each. I made a few cold-soldered joints and I had to fix them. With modem wave soldering equipment, it's fairly easy to avoid cold-soldered joints in your production line. But on hand-soldered circuits, it's always a possibility to have cold-soldered joints, so, if you have a nasty problem, don't forget the old solution: Re-solder every joint. Once in a while you'll find a joint that never got any solder at all! If for some reason you have acid-core solder around-it's mainly used for plumbing and is not found in most electronics labs, for good reason-keep it strictly labeled and segregated from ordinary rosin-core solder. Acid will badly corrode conductors. Also, keep specialty solders such as high-temperature solder, low-temperature solder, silver solder, and aluminum solder in a separate place, to avoid confusion.

There is also solder for stainless steel, which requires special flux.

Recently I have heard people promoting silver-solder as a kind of superior solder for splicing speaker cables. The "Golden Ear" set claim that this solder makes the audio sound better. However, I must caution you that silver-solder requires rather high temperatures, such that you need a small torch, and some messy borax flux, and I suspect that the high temperatures will do a lot more damage to the insulation and to the copper wire (by oxidizing it excessively) than any advantage you might get from a "superior soldered joint."

Make Good Connections

Printed-circuit boards aren't the only assembled component you'll have to contend with while trying to make circuits work. In Tracy Kidder's Pulitzer-Prize-winning book, The Soul of a New Machine (Ref. 2). one of the crucial moments occurs when the engineers explain to a management team that their new computer has a flaw that occurs only rarely but is driving them to distraction. One of the managers stands thoughtfully for a while and then reaches over and warps the main board: Scrunch, scrunch. To the horror and shock of the project engineers, the scrunching correlates with the terrible intermittent problem. When the main board's DIP sockets were replaced, the problem disappeared.

Like a faithful dog, a socket or connector is expected to do its job without question, and it usually does. However, on the rare occasion when one does go bad, the connector usually becomes intermittent before it fails utterly. Fortunately, many engineers and technicians learn early on that the way to check an intermittent problem is to make it reveal itself when the board's connector is wiggled and jiggled and plugged in and out while the power is on.

But don't all the instruction books say that you shouldn't plug in the board with the power on? Sure, a lot of them do. But I've never gotten into more trouble plugging a board into a hot connector than the trouble I've gotten out of. There may be some boards that are destroyed or damaged by this method, but they are in the distinct minority and should be studied. One way to help avoid problems is to make the ground fingers on a printed-circuit edge connector stick out longer than the other fingers.

Thus, ground will be established before any other connection. Still, if you have a board that tends to latch up because the power-supply sequencing may be improper, you have to be prepared to stop plugging the board into that hot socket-fast.

Learn by Fiddling and Tweaking

There are many situations that can foul things up, but we all tend to learn more from fiddling around with things, tweaking and unplugging, than by purely cerebral processes.

Once I had a technician who thought that DIP sockets should not be secured in place by tack soldering, but by glue. This technique worked fine for a while, but occasionally the sockets would act like an open circuit on one pin or another. To solve the problem, we used an old technique: We traced the circuit coming into the IC, and it was fine. We traced the signal coming out of the IC; nothing. Then, we traced the signals on the pins of the DIP itself; the signals were not the same as the signals on the socket, not at all. I finally realized that the glue was getting into the internal voids of the socket and preventing the IC's pin from making a true connection.

We banned the glue from that task, and the problem went away, mostly. Still, both before and after that time, we have seen sockets that just failed to connect to an IC's pin. You merely have to probe to the pin of the IC itself, not just to the socket, to nail down this possibility. Sometimes, the pin goes into the socket and actually fails to connect; but, more often than not, the pin is simply bent under the package.

There is one other kind of problem you can have with a socket, as a friend of mine recounted. He was trying to troubleshoot a very basic op-amp circuit, but its waveforms did not make sense. After several minutes, he turned his circuit over and realized he had forgotten to plug an op amp into the socket. This example leads us to McKenna's Law (named after an old friend, Dan McKenna): "You can't see it if you don't look at it." We invoke this law when we discover that we forgot to plug in a line cord or connect something. A vital part of troubleshooting is realizing that we are all at the mercy of McKenna's Law when we get absent-minded.

Connectors and sockets usually do more good than harm. They permit you to check options and perform experiments that may seem absurd and preposterous, yet are Instructional and life-saving. Once a friend was in the throes of a knock-down-drag-out struggle to troubleshoot a fast AD converter. He had tried many experiments, but a speed problem eluded him. He asked me if he should try a socket for a critical high-speed component. At first, I was aghast. But, after I thought about it and realized that the socket would add barely 1 pF of capacitance, I said, "Well, OK. it may not do much harm." The addition of the socket led to the realization that the speed problem was critically correlated with that component, and the problem was soon solved. The socket that might have caused terrible strays actually caused almost no harm and, in fact, facilitated the real troubleshooting process. If nothing you do leads in an encouraging direction, and you have a half-baked notion to tell your technician to install a socket, that may be the best idea you have all day. The socket may do very little harm and could lead to many experiments, which might give you the vital clue that puts you on the track of the real culprit.

When is a Connector Not a Connector?

When it's a relay. A relay is an electrically-controlled connector, and though relays are not as popular as they used to be, there are many times and many ways to use a relay to get a job done exactly right. Conversely, you can use a relay to get things done wrong, and you don't want to do that. Let's discuss.

Some relays are made with gold-plated or other precious-metal contacts, for the highest reliability in low-level circuits. What is the definition of low-level? As with every relay application, you have to refer to the manufacturer's data sheets. Very important. Because if you try to use a low-level relay for a high-power use, the contacts can erode, wear, or maybe even weld together. Conversely, if you try to use a heavy-duty relay, with its special metallurgy and tungsten contacts, they can run dry and refuse to contact at all in a high-impedance circuit. Gold is a wonderful (if expensive) material, and it can prevent contacts from running "dry," but it won't be found in high-current contacts.

Reed relays are hermetically sealed, and have good proven reliability if properly applied. I evaluated some recently that had less than 5 fA of leakage, when I guarded them carefully. I was impressed. Also they can be engineered to have low thermocouples and fast response; but they need a lot of ampere-turns as there is no iron in their magnetic circuit, so they are not normally considered low-power devices.

Oh, yes, I must add in one more comment on reed relays. If you install a well-built coil around a clean glass reed, you can get those low leakages. But if you BUY a reed relay, complete with coil, all packaged in a neat package, I bet I can tell you what the package is made out of: nylon. At room temperature, nylon is a fair insulator, but at 35 degrees, under conditions of high humidity, nylon is a LOUSY insulator. You can't even get 109 0 of leakage. So, if you want decent low leakage, you may well have to "roll your own" reed relays.

All mechanical relays have contact bounce. This may last 2 or 20 ms, and the duration can vary. When you want relays or switches to talk to digital logic circuits, anti-bounce circuits are de-rigeur. Also, when there is even a moderate amount of voltage or current, the manufacturer will usually spell out the need for some series RC network to put across the contacts, to help minimize arcing or "burning." You gotta read the data sheet.

But, these days, not all relays are mechanical. First of all, there are mercury-wetted relays which are credited with giving no contact bounce. Most of these must be held upright or they do not work. There are mercury-wetted reed relays that do not bounce and can be used in any position. But even they will not work at temperatures colder than -38 "C, where mercury freezes.

Then there are the solid-state relays. Some of them can switch many amperes-but they often use SCRs, and thus have a loss of more than a volt. You can't use that for a little signal. Others have low-ohm MOSFETs, and can handle a few amperes with low losses and low offset voltage. But the big ones have a lot of leakage and capacitance (which is not always mentioned). The little ones are nice and delicate, for precision switching, but cannot carry many milliamperes.

So, for high reliability, you have to be pretty knowledgeable and thoughtful, and selective, when you choose a relay, or you'll pick one that's inappropriate for your application, and you'll be embarrassed when some of them provide poor performance, or fail sooner than expected.

When Is a Relay Not a Relay?

When it is just a switch, mechanically operated by hand. But when you choose a switch, the contacts have almost exactly the same limitations as the relay's contacts.

There are high-current ones, there are delicate ones, there are hermetically-sealed ones. So in the same way, be thoughtful when you make your choice. You do have the advantage here that if you try to wear out a relay, a few million operations can cause failures in just a few weeks; but most people couldn't wear out a switch by hand fast enough to get in trouble! As with relays, if you aren't sure what the data sheet is trying to tell you, talk to the manufacturer's good people for advice and interpretation.

They may have a switch in their "back room" that is just what you need.

Weird Wired World

Now, I'll add a few pithy comments about wire and cable. Not all wire is the same.

For example, when I first got a job in electronics, I was having a lot of trouble with Teflon-insulated wire. The wires would often break right at the point where the solder stopped. After several engineers assured me that all wire was the same and suggested that I was just imagining things, I was ready to scream.

Finally, I found an engineer who admitted that cable manufacturers couldn't put individually tinned wires into a Teflon insulator, as they do with plastic-insulated wire. At the temperatures at which the Teflon is extruded, the solder would all flow together, thus making the stranded wire a solid wire. Instead, cable manufacturers use silver-plated wire strands for Teflon-insulated wire. With this type of wire, solder tends to wick up into the strands, thus making the wire quite brittle. Once I understood the wire's structure, I was able to solve my problems by adding strain relief for any bends or pulling stresses.

As I mentioned in Section 2, the ordinary plastic-insulated single-conductor wire that is used in telephones has just the right stiffness to make good twisted-pair wire for making capacitors with values of 1, 2.1, or 4.95 pF. The wire doesn't have a Teflon dielectric, but it's good enough for most applications.

Consider Your Wire Type

Shielded or coaxial cables, such as RG-58U, RG-174, shielded twisted pairs, and other special flat cables, all have their place in the job of getting signals from here to there without undue attenuation or crosstalk. When you have a large number of wires mindlessly bundled together and you don't have any bad crosstalk, you're witnessing a miracle. Often you have to unbundle the wires and separate the offending ones or the sensitive, delicate signals from the rest. Also, you may end up rewiring some or all of the wires into shielded cables.

Remember, Teflon is a good insulator, but air is even better. If you have to add struts, standoffs, or spacers to make sure that the critical wires stay put, go right ahead. If you have problems, the wire manufacturers can give you some advice.

Conversely, just as you use Teflon or air when you need a superior insulator, you have to be careful to get your best conductances. A friend who is an amateur radio operator says that many kinds of problems in RF circuits arise because nuts and bolts are used to make ground connections. If a lock washer or star washer is not included, the mechanical connection can loosen, the ground impedances will change with every little stress or strain, and nasty intermittent electrical problems will result. So. a major factor in the reliability of these circuits is ensuring the integrity of all bolted joints by always including star washers. And, make sure that wires and connectors do not get so loose as to hurt the reliability of your circuit or system. (See also comments on star washers in Section 13.) When you use shielded cable, should you ground the shield at one end or at both ends? Many cases call for a ground at the receiving end of the cable, but there are cases in which the shield is the main ground return. Neither way is necessarily bad, but be consistent. Likewise, in the design and the execution of the design, avoid ground loops, which can cause weird noise problems. In my systems, I build my analog ground system completely separate from the digital ground and make sure that the case or package ground is also strictly divorced. Then, after I use an ohmmeter to confirm that these grounds are really separate, I add one link from the analog ground to the digital ground and another link to the case or chassis. This technique works well for me, and I recommend it.

It is a little-known fact that some coaxial cable can degrade just sitting on the shelf. (Well, that's true, but it degrades faster if the shelf is sitting in the sunshine, or out in the rain. . . .) Some specialty types of cable whose codes and specifications are nominally similar can have an outer jacket that is not guaranteed to have good chemical stability. The jacket may be especially resistant to some chemicals, but less resistant to others. Specifically, in the 1950s there was a lot of military-surplus cable similar to RG-58 and RG-74 that did not have good stability. As the outer jacket degraded, the inner insulator was chemically degraded and the cable's UHF attenuation was degraded. In other cases, the outer shield was chemically attacked and corroded, and its conductivity got worse and its UHF attenuation was also impaired.

Most of that old cable has died and you can't even find it in junk-piles any more. But there are still specialty cables being made and sold now that do not last as well or age as gracefully as you would expect a good wire to do. If you select a cable to be especially resistant to one kind of chemical, it may be less resistant than normal to the attack of other ordinary chemicals. So, you should be aware that, even in something as simple as a wire, there may be more problems than meet the eye.

Heck, I just had a couple yards of Teflon-insulated wire sitting outside my kitchen window, running over to a sensor for an electronic thermometer. The wire would only get, on the average, 1 hour per day of direct sunlight. After 10 years, the yellow wire was still in good shape, but the white insulation had just about died utterly. Who wants to explain that one? Recently an engineer showed me the results of a study of wire for loudspeakers.

He showed that the inductance of ordinary two-conductor wire (per 20 feet) can cause a small but noticeable phase shift-perhaps 10 degrees at 20 kHz, even with the large and ultra-expensive speaker cables ($10 per foot and up). But when he took flat "ribbon" cable with 40 conductors (which is typically used to bus digital signals around), and tied every other wire in parallel, as the hot side, and every other wire in parallel as the return side, the inductance was lower by a factor of 10. When you think of it, if you take some cables that are supposed to look like 75 ohm, and you parallel 10 of these pairs, it should have pretty flat response, quite suitable for driving an 8 ohm load. Much better than some of the monstrous huge cables, and less bulky, and much cheaper, too.


1. Pease, Robert A., "Programmable Pulser Takes 5-nsec Steps," EDN, May 10, 1990, p. 150.

2. Kidder, Tracy, The Soul of a New Machine, Avon Books, New York, NY, 1981, p. 265.

Top of Page

PREV.   NEXT Related Articles HOME