Signed number representations

The early days of digital computing were marked by competing ideas about both hardware technology and mathematics technology (numbering systems). One of the great debates was the format of negative numbers, with some of the era's top experts expressing very strong and differing opinions. One camp supported two's complement, the system that is dominant today. Another camp supported ones' complement, where a negative value is formed by inverting all of the bits in its positive equivalent. A third group supported sign–magnitude, where a value is changed from positive to negative simply by toggling the word's highest-order bit.

There were arguments for and against each of the systems. Sign–magnitude allowed for easier tracing of memory dumps (a common process in the 1960s) as small numeric values use fewer 1 bits. These systems did ones' complement math internally, so numbers would have to be converted to ones' complement values when they were transmitted from a register to the math unit and then converted back to sign–magnitude when the result was transmitted back to the register. The electronics required more gates than the other systems – a key concern when the cost and packaging of discrete transistors were critical. IBM was one of the early supporters of sign–magnitude, with their 704, 709 and 709x series computers being perhaps the best-known systems to use it.

Ones' complement allowed for somewhat simpler hardware designs, as there was no need to convert values when passed to and from the math unit. But it also shared an undesirable characteristic with sign–magnitude: the ability to represent negative zero (−0). Negative zero behaves exactly like positive zero: when used as an operand in any calculation, the result will be the same whether an operand is positive or negative zero. The disadvantage is that the existence of two forms of the same value necessitates two comparisons when checking for equality with zero. Ones' complement subtraction can also result in an end-around borrow (described below). It can be argued that this makes the addition and subtraction logic more complicated or that it makes it simpler, as a subtraction requires simply inverting the bits of the second operand as it is passed to the adder. The PDP-1, CDC 160 series, CDC 3000 series, CDC 6000 series, UNIVAC 1100 series, and LINC computer use ones' complement representation.

Two's complement is the easiest to implement in hardware, which may be the ultimate reason for its widespread popularity.[1] Processors on the early mainframes often consisted of thousands of transistors, so eliminating a significant number of transistors was a significant cost savings. Mainframes such as the IBM System/360, the GE-600 series,[2] and the PDP-6 and PDP-10 use two's complement, as did minicomputers such as the PDP-5 and PDP-8 and the PDP-11 and VAX machines. The architects of the early integrated-circuit-based CPUs (Intel 8080, etc.) also chose to use two's complement math. As IC technology advanced, two's complement technology was adopted in virtually all processors, including x86,[3] m68k, Power ISA,[4] MIPS, SPARC, ARM, Itanium, PA-RISC, and DEC Alpha.

In the sign–magnitude representation, also called sign-and-magnitude or signed magnitude, a signed number is represented by the bit pattern corresponding to the sign of the number for the sign bit (often the most significant bit, set to 0 for a positive number and to 1 for a negative number), and the magnitude of the number (or absolute value) for the remaining bits. For example, in an eight-bit byte, only seven bits represent the magnitude, which can range from 0000000 (0) to 1111111 (127). Thus numbers ranging from −127₁₀ to +127₁₀ can be represented once the sign bit (the eighth bit) is added. For example, −43₁₀ encoded in an eight-bit byte is 10101011 while 43₁₀ is 00101011. Using sign–magnitude representation has multiple consequences which makes them more intricate to implement:[5]

1. There are two ways to represent zero, 00000000 (0) and 10000000 (−0).
2. Addition and subtraction require different behavior depending on the sign bit, whereas one's complement can ignore the sign bit and just do an end-around carry, and two's complement can ignore the sign bit and depend on the overflow behavior.
3. Comparison also require inspecting the sign bit, whereas in two's complement, one can simply subtract the two numbers, and check if the outcome is positive or negative.
4. The minimum negative number is −127 instead of −128 in the case of two's complement.

This approach is directly comparable to the common way of showing a sign (placing a "+" or "−" next to the number's magnitude). Some early binary computers (e.g., IBM 7090) use this representation, perhaps because of its natural relation to common usage. Sign–magnitude is the most common way of representing the significand in floating-point values.

In the ones' complement representation,[6] a negative number is represented by the bit pattern corresponding to the bitwise NOT (i.e. the "complement") of the positive number. Like sign–magnitude representation, ones' complement has two representations of 0: 00000000 (+0) and 11111111 (−0).[7]

As an example, the ones' complement form of 00101011 (43₁₀) becomes 11010100 (−43₁₀). The range of signed numbers using ones' complement is represented by −(2^N−1 − 1) to (2^N−1 − 1) and ±0. A conventional eight-bit byte is −127₁₀ to +127₁₀ with zero being either 00000000 (+0) or 11111111 (−0).

To add two numbers represented in this system, one does a conventional binary addition, but it is then necessary to do an end-around carry: that is, add any resulting carry back into the resulting sum.[8] To see why this is necessary, consider the following example showing the case of the addition of −1 (11111110) to +2 (00000010):

    binary    decimal
   11111110     −1
+  00000010     +2
───────────     ──
 1 00000000      0   ← Not the correct answer
          1     +1   ← Add carry
───────────     ──
   00000001      1   ← Correct answer

In the previous example, the first binary addition gives 00000000, which is incorrect. The correct result (00000001) only appears when the carry is added back in.

A remark on terminology: The system is referred to as "ones' complement" because the negation of a positive value x (represented as the bitwise NOT of x) can also be formed by subtracting x from the ones' complement representation of zero that is a long sequence of ones (−0). Two's complement arithmetic, on the other hand, forms the negation of x by subtracting x from a single large power of two that is congruent to +0.[9] Therefore, ones' complement and two's complement representations of the same negative value will differ by one.

Note that the ones' complement representation of a negative number can be obtained from the sign–magnitude representation merely by bitwise complementing the magnitude (inverting all the bits after the first). For example, the decimal number −125 with its sign–magnitude representation 11111101 can be represented in ones' complement form as 10000010.

In the two's complement representation, a negative number is represented by the bit pattern corresponding to the bitwise NOT (i.e. the "complement") of the positive number plus one, i.e. to the ones' complement plus one. It circumvents the problems of multiple representations of 0 and the need for the end-around carry of the ones' complement representation. This can also be thought of as the most significant bit representing the inverse of its value in an unsigned integer; in an 8-bit unsigned byte, the most significant bit represents the 128ths place, where in two's complement that bit would represent −128.

In two's-complement, there is only one zero, represented as 00000000. Negating a number (whether negative or positive) is done by inverting all the bits and then adding one to that result.[10] This actually reflects the ring structure on all integers modulo 2^N: ${\displaystyle \mathbb {Z} /2^{N}\mathbb {Z} }$. Addition of a pair of two's-complement integers is the same as addition of a pair of unsigned numbers (except for detection of overflow, if that is done); the same is true for subtraction and even for N lowest significant bits of a product (value of multiplication). For instance, a two's-complement addition of 127 and −128 gives the same binary bit pattern as an unsigned addition of 127 and 128, as can be seen from the 8-bit two's complement table.

An easier method to get the negation of a number in two's complement is as follows:

Method two:

1. Invert all the bits through the number
2. Add one

Example: for +2, which is 00000010 in binary (the ~ character is the C bitwise NOT operator, so ~X means "invert all the bits in X"):

1. ~00000010 → 11111101
2. 11111101 + 1 → 11111110 (−2 in two's complement)

In the offset binary representation, also called excess-K or biased, a signed number is represented by the bit pattern corresponding to the unsigned number plus K, with K being the biasing value or offset. Thus 0 is represented by K, and −K is represented by an all-zero bit pattern. This can be seen as a slight modification and generalization of the aforementioned two's-complement, which is virtually the excess-(2^N−1) representation with negated most significant bit.

Biased representations are now primarily used for the exponent of floating-point numbers. The IEEE 754 floating-point standard defines the exponent field of a single-precision (32-bit) number as an 8-bit excess-127 field. The double-precision (64-bit) exponent field is an 11-bit excess-1023 field; see exponent bias. It also had use for binary-coded decimal numbers as excess-3.

In the base −2 representation, a signed number is represented using a number system with base −2. In conventional binary number systems, the base, or radix, is 2; thus the rightmost bit represents 2⁰, the next bit represents 2¹, the next bit 2², and so on. However, a binary number system with base −2 is also possible. The rightmost bit represents (−2)⁰ = +1, the next bit represents (−2)¹ = −2, the next bit (−2)² = +4 and so on, with alternating sign. The numbers that can be represented with four bits are shown in the comparison table below.

The range of numbers that can be represented is asymmetric. If the word has an even number of bits, the magnitude of the largest negative number that can be represented is twice as large as the largest positive number that can be represented, and vice versa if the word has an odd number of bits.

The following table shows the positive and negative integers that can be represented using four bits.

Same table, as viewed from "given these binary bits, what is the number as interpreted by the representation system":

Google's Protocol Buffers "zig-zag encoding" is a system similar to sign–magnitude, but uses the least significant bit to represent the sign and has a single representation of zero. This allows a variable-length quantity encoding intended for nonnegative (unsigned) integers to be used efficiently for signed integers.[11]

A similar method is used in the Advanced Video Coding/H.264 and High Efficiency Video Coding/H.265 video compression standards to extend exponential-Golomb coding to negative numbers. In that extension, the least significant bit is almost a sign bit; zero has the same least significant bit (0) as all the negative numbers. This choice results in the largest magnitude representable positive number being one higher than the largest magnitude negative number, unlike in two's complement or the Protocol Buffers zig-zag encoding.

Another approach is to give each digit a sign, yielding the signed-digit representation. For instance, in 1726, John Colson advocated reducing expressions to "small numbers", numerals 1, 2, 3, 4, and 5. In 1840, Augustin Cauchy also expressed preference for such modified decimal numbers to reduce errors in computation.

• Balanced ternary
• Binary-coded decimal
• Computer number format
• Method of complements
• Signedness

• Ivan Flores, The Logic of Computer Arithmetic, Prentice-Hall (1963)
• Israel Koren, Computer Arithmetic Algorithms, A.K. Peters (2002), ISBN 1-56881-160-8