How an Electronic Brain Works - Part IV
January 1951 Radio-Electronics

January 1951 Radio-Electronics [Table of Contents] Wax nostalgic about and learn from the history of early electronics. See articles from Radio-Electronics, published 1930-1988. All copyrights hereby acknowledged.

It takes a while - and money - to accumulate issues of the vintage electronics magazines for posting articles here on RF Cafe. Often I can find groups for sale that comprise a full calendar year, but often they are groups of random months and years. That makes getting a complete series of articles like this one on "How an Electronic Brain Works" difficult. A lot of times installments appear every other month, so when a series has more than ten articles, it can run well over a year. For instance Part I of "How an Electronic Brain Works" appeared in the September 1950 issue of Radio-Electronics. The final chapter, Part XIII, appeared in October 1951. Throughout the series, authors Edmund C. Berkeley and Robert A. Jensen describe the workings of "Simon," their compact electronic computer - some even call it the first "desktop computer." Here is an article (with photos) about "Simon" in the November 1950 Scientific American magazine.

See Part IV - Long Division with Relays and Part IX - Some Electronic Circuits for Computers.

Part IV - Long division with relays - our little electric brain learns how to divide and to convert decimal numbers to binary and back again. Simon is getting an education

By Edmund C. Berkeley* and Robert A. Jensen

Previous articles of this series have shown how an electric brain made of relays can add, subtract, and multiply.

Now we shall carry out division. As before, we shall consider the process in binary notation, the scale of two.

As a second topic, we shall consider how to make a relay calculator convert a number from decimal notation to binary notation, and back again. There is every reason in the world why the machine itself should convert any decimal number, say 23, into the corresponding binary number (in this case 10111, one-oh-one-one-one, or one 16 plus no 8's plus one 4 plus one 2 plus one 1).

Addition, subtraction, and multiplication turned out to be very simple in binary notation as compared with decimal. The same is true with division: binary division is simple as can be.

Suppose we divide 1101 (one-one-oh-one, or 8 plus 4 plus 1, or 13 in decimal) into 10000101 (one-oh-oh-oh-oh-one-oh-one, or 128 plus 4 plus 1, or 133.

We do this in the same general way as we do in decimal division, except that we act as if we knew only the two digits 1 and 0:

Only two multiples of the divisor are used, one times the divisor, and zero times the divisor - and the latter is of course zero in every digit. No other multiples of the divisor are needed. If we simply compare the divisor with the partial remainder at any point in the division, we can tell whether the digit of the quotient is 1 or 0.

Circuits for division

As before, to keep the circuits simple, let us ignore a number of fine points, such as: fractions; the binal point (the analogue in the scale of two of the decimal point in the scale of ten); positive and negative numbers; size of numbers; etc. Suppose that we have an eight binary digit dividend, and a four binary digit divisor.

The circuit is on the opposite page. In part 1, terminal T1 is energized at the start, and holds up the relays storing the dividend through their hold contacts. (All current-carrying circuits and relay contacts in the energized state are in red.) The actual number which these relays store, of course, depends on something that happened before the time at which we begin. In the same way, the divisor is stored in relays of part 2 of the circuit, and terminal T2 holds them up.

Now different things have to happen at different stages during the division. So we want to have some relays that will tell us at what stage we are during the process of the division. This is the function of the K relays of part 3 of the circuit. The stages that they detect and report are 0, 1, 2, 3, 4. The time chart in Fig. 1 shows that stage 0 lasts from time 1 to time 8, stage 1 from times 9 to 16, stage 2 from times 17 to 24, etc. At stage 0, we attend to the first quotient digit; at stage 1, we attend to the second quotient digit; etc. The red parts of the circuit apply to the first stage of the division only.

We have to start off the divisions by selecting some digits, which we can call a partial remainder (see part 4). At stage 0, this is the first four digits of the dividend; but at later stages this is the result of a subtraction together with "bringing down" one more digit of the dividend. The circuit of part 4 shows that at each stage of the division, we have just the partial remainder that we desire stored in the E relays. We have to look ahead to part 8, of course, and take on faith that the G relay contacts in part 4 will express the result of a subtraction that we want.

The next thing that we must do is decide whether the divisor "goes" into the partial remainder, or whether it "doesn't go". To make this decision, we must compare two numbers and decide which is the larger. The divisor "goes" and yields 1 as a digit of the quotient if, and only if, the partial remainder is larger. A circuit that does exactly this is shown in part 5. The red contacts show the original partial remainder (stored in the E relays) and the divisor (A relays). We see that there is no path for the quotient relay Q to be energized, and so the first "quotient digit" is 0.

Before we go any further, we want to store that quotient digit, so that we shall know the whole quotient when we get through with the division. This duty is performed by the circuit of part 6, which shows how the digit quotient is routed, according to the time it is obtained, into the right C relay.

We now want to determine the multiple of the divisor that depends on the quotient digit and the divisor. This is the function of part 7 of the circuit, which will give us the divisor itself if the quotient digit is one and zero in all digits if the quotient digit is zero.

In part 8 of the dividing circuit, the subtraction of the divisor multiple from the partial remainder is indicated schematically, because actual circuits for subtraction were discussed previously.

The timing of the circuits, up to the end of the first two quotient digits, is shown in the timing chart of Fig. 1. The same conventions are used here as in the time chart for multiplication in the previous article. Successive time intervals 1,2,3,4, are shown from left to right. In the first column, the different terminals are shown from top to bottom; in the second column, the names of the relays which the terminals energize; in the third column the names of the relay contacts through which the relays are energized. Each horizontal line begins when its terminal is energized, and stops when its terminal ceases to be energized. There are some vertical lines showing X's and O's. X marks the relays energized at a certain time, and the O's mark the contacts through which they are energized.

Now, you may say, it is all very well to be able to add, subtract, multiply and divide in binary notation, but how do we go from decimals to binaries?

In fact, even before we ask this question, we have to ask: how will the machine take in a decimal number? In other words, how will the machine accept it, record it, and store it?

Ordinarily a calculating machine (or some auxiliary part of it) will have a keyboard, containing keys numbered 0, 1, 2 up to 9. Often the keyboard will have a different column for each column of the number to be inserted in the machine. To put in a number like 593, we press down the 5 key in one column, the 9 key in the next column, and the 3 key in the third column.

In many calculating machines, the result of pressing down a key, say 3, is to turn some little counter wheel 3/10 of one complete turn. But in our machine we want the result of pressing down the 3 key to be the energizing of certain relays, so that we can use the information later in the machine.

We would reasonably desire to convert anyone of these ten decimal digits o to 9 into a pure binary number according to Table I.

Fig 2 is a circuit which will do this (using 15 rectifiers and 4 relays).

For example, if we press the 3 key, relays A2 and A1 are energized, but not relays A8 and A4, and so the information produced in the relay register is 0011, which is the binary number three.

In this way the decimal number 593 can be converted into 0101 1001 0011 stored in 12 relays. This form of representing a decimal number by a "code" for each digit is coded decimal rotation.

Now how do we go from 0101 1001 0011 to what this number is in pure binary notation? 593 of course is 5 times 10 times 10, plus 9 times 10, plus 3, and all we have to do is writs this in binary and tell our machine do it:

0101 times 1010 times 1010 plus 1001 times 1010, plus 0011.

And this our machine can do because it has an addition circuit and a multiplication circuit. It will be neater to program this operation with:

5 times 10, plus 9,

all times 10, plus 3.

Thus for a ten-digit decimal number, we shall only need nine multiplications.

Binary to decimal

Now suppose that we have the opposite problem. Given a binary number, we want to find the corresponding decimal number. We divide this number by 1010 (one-oh-one-oh, or 8 plus 2, or 10 in binary) and find the remainder, which will be less than 10, and store it. Then we take the quotient, and divide that by 1010, and store the new remainder. And so on.

For example, suppose we desire to convert the binary number 10000101 into a decimal number.

which is 3 in decimal, and becomes our first decimal digit.

 which is 3 in decimal, and becomes our second decimal digit.

Our relay electric brain has division circuits and registers where we can store remainders; and so we can convert from binary into decimal. In this case we obtain the coded decimal form 0001 0011 0011 which is the same as 133.

How do we get this out of the machine? For example, suppose we have ten typewriter keys, bearing the characters 0, 1, 2, 3, 4, 5, 6, 7, 8, 9. We wish to impulse these keys in order. The circuit in Fig. 3 will do this. When terminal T is energized, the appropriate K relay is energized, depending on the state of the A relays which hold the corresponding binary number.

* Author: Giant Brains

(continued next month)

Posted April 20, 2021