Normalisation of floating point form — AQA A-Level Computer Science
Normalisation of floating-point numbers is the process of adjusting the mantissa and exponent so that a positive number starts with '01' and a negative number starts with '10'. This format maximises precision for a given bit length and ensures each distinct real value has a single, unique binary representation.
Normalisation of floating point form explained
Floating-point numbers in computer systems are normalised to maximise precision and ensure each non-zero value has a unique binary representation.
Read the full explanation
For a positive number using two's complement, normalisation requires the first two bits to be 01. To normalise an un-normalised positive number such as mantissa 0.0001101 with exponent 3 (0.0001101 * 2^3), the mantissa is shifted left by 3 places to produce 0.1101000. Shifting left by 3 places doubles the value three times, so the exponent must be decreased by 3 to keep the magnitude unchanged. Calculating 3 - 3 gives exponent 0. The normalised representation is mantissa 0.1101000 with exponent 0 (0.1101000 * 2^0).
Your focus
- Know why floating point numbers are normalised and be able to normalise un- normalised floating point numbers with positive
Which command word is this?
Provide a brief, factual answer without explanation or calculation, such as identifying if a number is normalised or quoting the starting bits.
Normalisation of floating point form exam tips
Topic Overview
Normalisation in computer science refers to the standardisation of floating-point representations such that the most significant non-sign bit immediately follows the binary point. In two's complement notation, this requires positive numbers to begin with '01' and negative numbers to begin with '10'.
This process is essential for computer systems as it maximises the precision of real numbers stored within a fixed bit allocation by eliminating redundant leading sign bits. Furthermore, normalisation guarantees a unique binary representation for each non-zero real value, simplifying numerical comparisons and hardware arithmetic operations across the CPU.
Key Concepts
- →Normalised Bit Patterns: Positive numbers start with 01; negative numbers start with 10 (assuming binary point is between the first and second bits).
- →Maximising Precision: By shifting the mantissa so that significant bits occupy the highest available positions, quantization errors and rounding loss are minimised.
- →Exponent Compensation: Shifting the mantissa left by n places requires subtracting n from the exponent; shifting right requires adding n to the exponent.
- →Unique Representation: Prevents multiple representations of the same real number (e.g. 0.5 can otherwise be represented with varying exponents and leading zeros).
Marking Points
- States the reasons for normalisation: maximising precision and providing a unique binary representation.
- Identifies that a normalised positive floating-point number starts with 01.
- Shifts the mantissa left by the required number of positions until 01 is achieved.
- Decreases the exponent by the exact number of places the mantissa was shifted left.
- Calculates the normalised form of 0.0001101 * 2^3 as mantissa 0.1101000 and exponent 0 (2^0).
Examiner Tips
- 💡Remember the golden rule: shift mantissa left, decrease exponent; shift mantissa right, increase exponent.
- 💡Always check the first two bits: positive normalised must be 01, negative normalised must be 10 in two's complement.
- 💡Always convert both the initial and final representations into denary as a sanity check to verify they evaluate to the exact same real value.
- 💡When asked for two reasons why floating-point numbers are normalised, state: 'To provide maximum precision for the given number of bits' and 'To ensure each real number has a single, unique representation'.
- 💡Pay close attention to whether the question defines the binary point after the first bit (e.g. 0.1...) or after the sign bit; AQA standard practice assumes the binary point lies directly between the first and second bits of the mantissa.
Common Mistakes
- Increases the exponent when shifting the mantissa left instead of decreasing it; correct by subtracting the shift count from the exponent.
- Calculates the new exponent incorrectly, for example subtracting 2 instead of 3 when shifting three places.
- Leaves trailing bits unpadded when shifting left; pad newly opened lower-order bits with zeros.
- Assuming negative numbers start with 01: Students frequently apply the rule for positive numbers to negative numbers, forgetting that a negative two's complement value must begin with 10.
- Adding to the exponent during a left shift: Students often confuse left shifts with multiplication and incorrectly add to the exponent, rather than decrementing it to maintain balance.
- Underflow during normalisation: Forgetting that decrementing the exponent can cause an exponent underflow if the exponent is already at its minimum representable negative value.
Revision Plan
- 1Day 1: Review two's complement and practice converting fractional denary numbers into fixed-point binary.
- 2Day 2: Master the structure of unnormalised floating-point numbers and understand how mantissa and exponent combine via 2^exponent.
- 3Day 3: Practice normalising positive numbers (shifting left until '01' is obtained and decrementing exponent).
- 4Day 4: Practice normalising negative numbers (shifting left until '10' is obtained and decrementing exponent).
- 5Day 5: Complete timed past-paper exam questions covering precision trade-offs, normalisation steps, and edge cases like zero.
Exam Question Types
- 📋Direct Normalisation: You are given an unnormalised mantissa and exponent and asked to produce the normalised form with working.
- 📋Calculation followed by Normalisation: Perform binary addition or subtraction on two floating-point numbers, then normalise the resulting sum or difference.
- 📋Theory and Justification: State advantages of normalisation or explain why a particular binary pattern is not normalised.
Command Word Expectations (AQA)
Provide a brief, factual answer without explanation or calculation, such as identifying if a number is normalised or quoting the starting bits.
Provide a detailed account of reasons or causes, linking computational mechanism (e.g. eliminating leading 0s/1s) to the benefit (e.g. maximising precision).
Write out full step-by-step arithmetic operations and shift directions to demonstrate clearly how the final normalised format is derived.
How Students Lose Marks (Examiner Pitfalls)
Step-by-Step Worked Solutions
Question: A real number is represented in floating-point form using an 8-bit two's complement mantissa and a 4-bit two's complement exponent. Normalise the unnormalised representation: Mantissa: 00010100, Exponent: 0101.
- 1.Step 1: Identify sign and current state. The mantissa is positive because its sign bit is 0. A normalised positive mantissa must begin with '01'. The current mantissa begins with '0001', indicating two redundant leading zeros.
- 2.Step 2: Shift the mantissa left until it begins with '01'. Shifting left by 2 positions gives: 01010000.
- 3.Step 3: Adjust the exponent to preserve the value. Shifting the mantissa left by 2 positions multiplies it by 2^2, so the exponent must be decremented by 2. The original exponent is 0101 (denary 5). Subtracting 2 gives 5 - 2 = 3, which in 4-bit two's complement is 0011.
- 4.Step 4: Combine the normalised components. Normalised Mantissa: 01010000, Normalised Exponent: 0011.
Question: Normalise the negative floating-point value with an 8-bit mantissa and 4-bit exponent: Mantissa: 11101000, Exponent: 0010.
- 1.Step 1: Check sign and normalisation condition. The sign bit is 1, indicating a negative value. A normalised negative mantissa in two's complement must start with '10'. The current mantissa starts with '1110'.
- 2.Step 2: Count redundant sign bits and shift. The mantissa has two redundant leading '1' bits. Shift the mantissa left by 2 positions to achieve the '10' prefix, padding with 0s on the right: 10100000.
- 3.Step 3: Decrement the exponent by the number of left shifts. Original exponent is 0010 (denary 2). Decrementing by 2 gives 2 - 2 = 0, which is 0000 in 4-bit two's complement.
- 4.Step 4: Write down final bit patterns. Mantissa = 10100000, Exponent = 0000.