The expression, familiar to all of us--"They're as alike as two peas in a pod"--is really not based on very careful observations. The truth is, the peas in a pod are not just alike, nor, for that matter, are any of the other myriads of things that go to make up the world around us, whether they are the products of man or nature. Though they may appear to be very similar to many other creations in their particular classification or species, they are never just exactly like them. These variations do not occur in a random fashion either. They follow a very precise mathematical law which anyone can understand.
If you were to carefully collect all of the leaves from one of the trees in your yard and measure their dimensions, you would discover an interesting fact. There would be small, large, and average-sized leaves.
If you should chart all this data with their size measurements along the bottom and the number of leaves you had found which were of each size along the left-hand side, you would find that you had constructed a chart shaped very much like the one in Fig. 4-1. This is known as the curve of normal distribution. In graphic form, it says that most of the product of any process, natural or artificial, will be found to adhere closely to some average characteristic; that only a small percentage will deviate very far from this "norm" and that this percentage is a predictable quantity.
Fig. 4-1. Normal distribution curve.
This curve has certain known mathematical characteristics. The line through the center is the average value of all the samples measured. There is always an upper maximum value and a lower minimum value. Approximately 70 percent of all samples measured will fall within ± 1/3 of the maximum variation from this average. Approximately 95 percent will fall between ±2/3 of the maximum variation from the average.
PREDICTING VARIABLES
The width of the distribution, or the range, will depend upon the particular process or phenomenon being measured. Some processes show very narrow distributions. Others are quite broad. But broad or narrow, the curve will always be of this shape and a known percentage will always fall within limits which are plus or minus a certain percentage of the maximum deviation. Because this will happen with mathematical certainty, it is possible to use this knowledge to make certain very accurate predictions. If you take a few samples out of any process enough to permit you to ascertain the shape of the curve--you can predict with great accuracy the total number of units, or the percentage of units of any given dimensional deviation that will occur in 100, 1,000, 10,000, or 1,000,000 samples. This fact is the basis of all modern mass production quality control techniques.
Vacuum tubes are the end result of a variety of mechanical and chemical manufacturing processes. These processes all have their normal variations which are the characteristic variables noted whenever you measure a large group of tubes. Plate current, screen current, transconductance, and plate resistance are all typical of the characteristics which follow the law of normal distribution. They have an average value which is the one usually published as the rated, or bogey value. This is the value that is more likely to be measured more often than any other value. It is also the value that all the other values will average if they are added together and divided by the number of samples measured.
This last concept is the important one to keep in mind.
It is perfectly possible to measure a large group of tubes and never find even one tube which measures what is given in the manuals as the rated value. They will measure all around this value, but never exactly on it. This is "normal" when you remember that the rated value is really only an average. That means it must be derived from many measurements of tubes that are both above and below this value. Sometimes the mathematical results of such an averaging do not result in a "real" value at all.
The value is real enough if you are considering a very large quantity of tubes, but it may be quite theoretical as far as a few tubes are concerned.
HOW STANDARDS ARE SET
There is another fact about these bogey values for tube characteristics that is very important to keep in mind.
These values are agreed upon by all tube manufacturers only after each one has submitted manufacturing records showing how his product has been running. This data is turned over to the Joint Electron Device Engineering Committee-JEDEC for short-of the EIA. Quite frequently, the several manufacturers of a given type will all submit different averages to the standardizing committee. When this happens, it is the established practice of the committee to average the averages in obtaining the ultimate value which will appear in a tube manual. This is another reason why these values may be very hard to find in a particular manufacturing product.
We have discussed the so-called "normal" distribution and shown that it was a symmetrical curve which always followed known mathematical laws. There is another curve which is just as normal for certain characteristics; however, it is not a symmetrical curve. It is a one-sided curve and is known as the "skewed" distribution. (Fig. 4-2). The skewed distribution is associated with those characteristics which have zero or infinity as one limit of their range. Such characteristics as leakage and plate-current cutoff are typical of the skewed distribution. A maximum number of the samples in any lot should approach the zero or infinite axis. The curve is asymptotic in that it never quite reaches this limiting value.
Fig. 4-2. Skewed distribution curve.
The bogey value for such a characteristic cannot be an average, obviously, since you cannot average either infinity or zero. What is usually done is to set a limit which includes 95 percent of the so-called "normal" product.
The value obtained will be a long way from what a majority of the product will measure. There will be very few tubes in a given lot that will approach this limit most of them will be near the other limit. Yet, these few tubes which are quite different from the bulk of the lot are considered "normal" by definition.
THE BOGEY TUBE
In any discussion of characteristic variables and the normal ranges that must be expected, someone is bound to ask, sooner or later--"What about bogey tubes? Why not supply technical people with tubes that are known to be average. In this way, they could design or adjust circuits so that they would work best with average tubes, and this would insure their proper functioning with a large majority of all tubes encountered in the field." This is a very plausible suggestion and it would seem like the logical thing to do until one has made further examinations. Most tube characteristics are interrelated so that although we can often find a tube having one of its characteristics on bogey, the chances of finding one with two characteristics on bogey are fairly small. As more characteristics are added to the list of requirements, the chances get even more remote until by the time we have added as few as five or six, the chances are almost zero. It is very much like trying to find a living example of the average man. He doesn't really exist except in the realm of statistics. The same is true of the bogey tube.
THE LIMIT TUBE
The next question usually is--"All right, so there are really no bogey tubes; what about limit tubes?" Here, the problem is not quite as bad. Since many limits are fairly broad, tubes that are remote from bogey, but not necessarily on the upper or lower limits, are fairly common.
To get tubes that are lower limit for several characteristics, may be quite a problem, especially since several characteristics are so related that the upper limit of one is produced by the lower limit of another.
But even where limit tubes are obtainable, their value is of some doubt because they tend to be unstable tubes.
By their very nature, they are abnormalities. These abnormalities are often produced by mechanical or chemical deficiencies in their manufacture. These deficiencies result in unstable characteristics so that even though they may have been limit tubes when they were picked, their continued reliability-even as limit tubes-is very poor.
The best method of establishing operating characteristics of a large group of tubes in a particular piece of equipment is simply the statistical sampling procedure used by the manufacturer to control the characteristics of his product in the first place. Obtain a large sample of the tubes in question, test them in the equipment they are to be used, and measure an important characteristic such as gain. By noting the most extreme variations that take place, you will have a measure of what to expect from the majority of tubes that may be encountered in the field.
CORRELATING MEASUREMENTS
Tube characteristics may be classified into two groups; performance and correlated characteristics. Such characteristics as gain and power output are performance characteristics. They are the functions for which tubes are designed to be used. There are very exact means of measuring these characteristics, but they are somewhat involved and require particular knowledge as to the exact method in which the tubes will be used. In an attempt to simplify the testing of tubes and to generalize their characteristics for cataloguing, the methods known as correlative measurements were developed many years ago.
Correlative measurements are those measurements which are supposed to predict performance. They are not of themselves significant except as a means for predicting significant performance characteristics. Because these characteristics are the ones most generally listed in tube manuals and data sheets, their actual significance is some times exaggerated. There is almost never any direct correlation between these measurements and their performance related characteristics. If correlation is 80 percent, this is considered rather good. A correlation of 50 percent is no correlation at all; in fact, a 50 percent correlation would be regarded as a random phenomenon, or one based on chance alone.
Some of the more common correlated measurements are transconductance, electrode currents, and plate resistance. There are many others that will be discussed more fully in a later section on special tests for estimating tube performance. Of these, the most frequently referred to correlated measurement is transconductance. This is the ratio of plate current change to the grid voltage causing that change, and is supposed to be a direct measure of the gain possibilities of the tube. Under highly controlled conditions, it can be, but under the conditions most commonly encountered in most test equipment, it can mean very little. Let us examine some of the reasons why this is so.
Transconductance varies inversely with grid bias, be tween some optimum value near zero bias and a minimum value near cutoff. It varies directly with plate current, over the same range of grid bias voltages. Plate current depends on plate, screen, and grid voltage.
Gain, on the other hand, is dependent upon the change in plate current which takes place across a particular load with a given plate current flowing. The effective value of the plate load is determined by the external load in shunt with the tube resistance or Rw Thus, a knowledge of the transconductance of a given tube without similar knowledge about its plate resistance and plate current can do very little to predict actual performance.
TYPICAL CORRELATION
Fig. 4-3. Typical correlation chart showing gain and transconductance relationships
between various tubes.
A typical correlation chart for a group of tubes is shown in Fig. 4-3. Their measured transconductance is plotted against gain. It will be noted that there is a very good general correlation between the two phenomena.
In most cases, the higher transconductance tubes show higher measured gain and vice versa. But just how good a correlation exists between individual tubes? Taking the tubes that are indicated (A, B, C and D), note that al though they fall nicely within the general grouping for all of the other tubes, specifically they do not correlate at all. Tube A has the same gain as B; yet, the transconductance of tube A is lower than that of B. Tubes B and C have the same transconductance; yet, tube C has considerably more gain. Tube C has more gain than D whose transconductance is somewhat higher than that of C. Such failures to correlate are common experiences for anyone who has taken the time to study tubes as they are, rather than as they are supposed to be. This doesn't mean that there is anything wrong with the engineering formulas which indicate that gain and transconductance are a first-order effect and should correlate. They do, under the right conditions. What we are looking at here is simply the result of not measuring all of the other tube characteristics which affect gain. If this were done, the apparent contradictions would not be contradictions at all.
One of the most common causes for the apparent lack of correlation which sometimes exists between transconductance measurements and gain lies in the fact that the majority of applications use cathode bias, whereas most instruments used to measure transconductance have fixed bias. With fixed bias, it is assumed that all tubes have the same transfer characteristics which, of course, is not true. Transfer characteristics vary, especially in the region where the tube is most critical to bias changes. This is shown in Fig. 4-4. Note that these curves coincide at points X and Y. These are frequently the commercial test points because more than likely they are the points used in rating the tubes. If these tubes are used in equipment, they will more likely operate in the region of A, B, or C, which are all within 0.5 volt of each other in the minus 1 volt region. Over this very small voltage range, the transconductance may vary as much as 20 percent, and with it, the effective gain. Here are three tubes which, according to standard measurement techniques, have the same transconductance; yet, which in use, will be found to have three different gain measurements.
Fig. 4-4. Curves showing that transfer characteristics vary, especially in
the region where the tube is most critical to bias changes.
Similar examples could be shown for tubes having the same transconductance, but with differing plate resistances. These too would show a gross miscorrelation. The purpose in making this clear is to point out to the reader that correlated measurements are not in themselves overly significant. When applied to a single tube, they may have absolutely no significance. They are a characteristic which, on the average, shows a definite relationship to a performance characteristic, but which are based on averages and are not, therefore, applicable to individual tubes as such. This fact should be borne in mind as we go on to discuss characteristic ranges.
CHARACTERISTIC SPREADS
We have said in earlier paragraphs that all tube characteristics follow the known laws of variation, and distribute themselves around an average value known as the bogey value. The degree for which each characteristic is allowed to deviate from this central value is known as the spread of that characteristic. Characteristic spreads are not given in most tube specifications which include, for the most part, only the average or bogey figure. Although the exact limits are an individual company's own prerogative to choose, they may vary from tube type to tube type and from company to company. There is, however, fairly general agreement throughout the industry as to what constitutes good engineering practice, and so, commercial practice. These practices apply to tubes sold for general renewal use and do not apply to specially selected tubes sold to some equipment manufacturers, or to those premium quality tubes sold either commercially or to the military. These special groups will be dealt with in sub sequent sections.
Transconductance is usually permitted to vary about ±40 percent of the published bogey value. This means that if the registered value for a given tube type is 2,000 micro mhos, you can expect to find tubes reading anywhere from 1,200 micromhos to 2,800 micromhos in any sample you measure. This does not mean that there will necessarily be a similar spread in gain when these tubes are compared in typical applications. It is probable that the gain spread will be narrowed by at least half this amount because of the foregoing reasons.
Plate current cutoff is a published characteristic of considerable significance in many applications. This characteristic is normally controlled at about twice the published rating. In other words, if the rating sheet for the tube states that the plate current will be 50 microamperes at minus 10 volts, you can expect to find some tubes that do not reduce to this level until the grid voltage is raised to -20 volts. Once again, this is not a serious departure from specifications as far as performance is concerned because the plate current of some tubes may be only 100 microamperes at -10 volts, but it takes that extra 10 volts on the grid to reduce it the additional 50 microamperes.
Either current is quite inconsequential for all practical purposes.
Plate current is usually controlled within limits of ±20 percent and plate resistance, because of its intimate relation to transconductance, is likewise a ±40 percent characteristic. The correlation between either or both of these characteristics and any measure of performance is rather low, except in the case of certain unique applications.
For these applications, the limits may be quite often tightened.
Screen current is one that is controlled on the high side only, being usually about twice the published rating.
The reasoning here is that you can't have too little screen current, since it is only a loss current anyhow. If a tube has a better than average plate-to-screen current ratio, it is just a more efficient tube, and no one is going to find anything wrong with it. High screen current, on the other hand, will waste power; reduce the effective screen volt age and, hence, cause the tube to cut off sooner and have lower transconductance at a given grid bias. High screen current will also contribute to excessive screen dissipation with the danger of screen emission and runaway.
Grid current is controlled at about 1 microampere for most small tubes, but is allowed to get up to as high as 5 microamperes for some power tubes. This is important to recognize when attempting to measure this characteristic on commercial type testers. There isn't any one gas current test that will apply to all tubes.
This is also true of heater-to-cathode leakage. In a later section devoted to tube testing and tube testers, this will be dealt with more completely; however, it is normal for tubes with different powered heaters to vary from as little as 5 microamperes of leakage to as much as 100 micro amperes. Again, there isn't any universal test, such as a neon bulb checker, that can evaluate all tubes for this characteristic. Such a test is either too sensitive or not sensitive enough.
Finally, heater current is a variable and usually falls between ±10 percent of the ratio value. For example, 600 milliampere heaters that read as low as 530, or as high as 670 milliamperes, would not be outside of specifications. This applies to either series-string or parallel heater types. The significance of a 10 percent heater current variation is probably not too great, although the lower limit tubes will probably fail to operate if they are used where the line voltage is 10 percent below normal as well. The reasoning that backs up these limits is the fact that far more situations exist where line voltages are high than low. Also, the statistical number of tubes falling in the lower limit is very few; probably less than 1 percent.
QUALITY CONTROL
Quality is a word that is used by many people to de scribe many different things. There is no other word that can be used as an adequate substitute, and this probably accounts for its great use and even its frequent misuse.
Regardless of what the advertising men may have made it appear to mean, the term "quality control" to a manufacturing man means just one thing-uniformity control.
The model or prototype as released by the development group, has certain characteristics. These have been established as the optimum or most desirable ones that the products should have in order to comply with a majority of the anticipated customers' needs. As we have noted earlier in this section, all manufacturing processes result in variations from the established design. The problem is to control all the individual variations so that their combined effect on the end product will result in gross variations that are not unacceptable in the market. It follows, therefore, that the successful control of quality is an economic necessity for any manufacturer in order that he may stay in business.
It has been correctly stated by competent authorities in the field of industrial management that it is cheaper to make a quality (uniform) product than it is to make a poor quality (non-uniform) product. This may seem like a paradox, but it is not. When all of the minor variables in a complex process are carefully controlled and minimized, the end product goes together with fewer mis fits and, consequently, fewer rejects. Rejects add enormously to cost and, therefore, the fewer the rejects, the lower the manufacturing costs. It would be hard to find a manufacturer who is not continually striving to lower his manufacturing costs, and for this reason, no manufacturer can afford to produce any product that doesn't meet good quality standards. The foregoing facts are pointed out in order to make it clear to the reader that what follows is not a surrender to penny pinching on the part of the manufacturer, but it is, in fact, a highly scientific activity, motivated by the soundest of engineering and management principles.
SAMPLING
The complete testing of any vacuum tube in order to insure its conformance to an original design is a very complicated and detailed matter. Over 50 individual tests and measurements are frequently called for. Naturally, it becomes impossible to do this on 100 percent of the product. Even if time and cost were no consideration at all, some of the tests, such as life tests, are self destructive. Therefore, if 100 percent of the product were tested, there wouldn't be any product left to ship. That is why the practice of sampling is of such a necessity.
Sampling consists of taking a given number of tubes from every hour's production and submitting them to various tests. These tests are designed to disclose process changes or dimensional changes that might indicate a trend. One of the cornerstones of all mass production quality controls is the fact that a process cannot vary beyond certain known limits except through a long pro gram of change. In other words, it can drift, but it cannot jump from one state to another. Knowing this, quality controls are designed to watch for trends as when tools begin to wear, or machines start to go out of adjustment.
Noting these trends permits corrective action to be taken before serious departures from normal have occurred.
So, the whole philosophy of quality control is aimed at producing a large quantity of the most uniform product possible and at the most economical cost. Because of the many tests performed and the manner in which one test cross checks another, it becomes virtually impossible for more than a very insignificant number of true defects to actually slip through to the customer. When they do, they are never in so-called "runs." The statistical probability of one individual obtaining two defects in a row is quite low. When it comes to getting three or four such defects in a row, the chances are almost zero. When such a situation seems to have been encountered, it is almost certain that one of the following circumstances is the real culprit.
Either the test device is at fault, or the equipment is designed around tubes having a different set of bogey characteristics. This leads us into a new topic-that of matching tubes and equipment for optimum performance.
DESIGN TOLERANCES
In the design of electronic equipment, one of the principle responsibilities of the design engineer is an under standing of the normal variations which will exist in all of the components which will go into the mass-produced article. A well designed piece of apparatus, like a well designed machine, will accept all standard replacement parts which are normally available on the market. Most designs do follow this common sense rule; however, there are a few, and it is unfortunately true that their numbers have been increasing in recent years, that do not. These latter designs are developed as a result of price squeezing short cuts which, while they may save cost for the equipment manufacturer, add much to the problems of the service industry and the public. There are varying degrees to which this practice is followed by various equipment manufacturers. In general, it has been severely dealt with by the military services wherever it has been possible to prove its existence. But private industry continues to practice it wherever it will gain for them some temporary advantage over their competition. This practice is known as "tube selection" and will be discussed in the next section.