Common CRC Standards

CRC isn't just one algorithm—it's a family of algorithms, each tailored to specific requirements. A USB flash drive uses CRC-16. An Ethernet NIC uses CRC-32. A satellite link might use CRC-32C. File archives like ZIP and PNG embed CRC-32. The Bluetooth specification mandates CRC-24.

Each standard represents careful engineering tradeoffs: polynomial selection for optimal error detection, bit ordering for hardware compatibility, initialization values for specific edge case handling. Understanding these standards is essential for implementing protocols correctly and debugging interoperability issues.

This page surveys the most important CRC standards, explaining their specifications, applications, and the engineering rationale behind their design.

A complete CRC standard specifies more than just the polynomial. The full parameter set includes:

1. Width (r) The number of bits in the CRC value. Common values: 8, 16, 24, 32, 64.

2. Polynomial (Poly) The generator polynomial, usually expressed in hexadecimal. May be given with or without the leading 1 bit.

3. Initial Value (Init) The starting value of the CRC register. Often 0x0000...0 or 0xFFFF...F.

4. Input Reflection (RefIn) Whether each input byte is reflected (bit-reversed) before processing.

5. Output Reflection (RefOut) Whether the final CRC value is reflected before output.

6. Final XOR (XorOut) A value XORed with the final CRC before output. Often 0x0000...0 or 0xFFFF...F.

7. Check Value The CRC of the ASCII string "123456789" (0x31 0x32 ... 0x39). Used for implementation verification.

CRC-8 provides 8-bit (1 byte) checksums. With only 256 possible values, CRC-8 is suitable for short messages where space is at a premium and undetected error probability of ~0.4% is acceptable.

Applications:

• Low-level protocols with small packets
• Sensor data in embedded systems
• Protocol header protection
• Memory scrubbing in small systems

Polynomial Analysis:

• 0x07 (x⁸ + x² + x + 1): A primitive polynomial providing maximal period. Simple to implement.

• 0x31 (x⁸ + x⁵ + x⁴ + 1): Used by Dallas/Maxim 1-Wire. Has (x+1) factor for odd-parity detection.

• 0x9B (x⁸ + x⁷ + x⁴ + x³ + x + 1): Strong burst detection properties, used in cellular.

Detection Capabilities:

With 8-bit CRC:

• All single-bit errors: Detected
• All odd-bit-count errors: Detected (if polynomial has (x+1) factor)
• All burst errors ≤ 8 bits: Detected
• Random errors: ~99.6% detection rate

For messages longer than a few hundred bytes, consider CRC-16 or higher.

CRC-16 provides 16-bit (2 byte) checksums with ~99.998% detection rate for random errors. This balance of protection and overhead made CRC-16 the standard for serial protocols, industrial networks, and many embedded systems.

Applications:

• Modbus industrial protocol
• HDLC/SDLC data link protocols
• X.25 packet networks
• USB (CRC-16)
• Bluetooth (CRC-16-CCITT)
• SD/MMC memory cards

The Two Major Polynomials:

CRC-16-CCITT: 0x1021 = x¹⁶ + x¹² + x⁵ + 1

The CCITT (now ITU-T) polynomial has excellent mathematical properties:

• Factors as (x + 1) × (primitive polynomial of degree 15)
• Detects all odd-bit-count errors
• Detects all double-bit errors for messages < 32,767 bits (4KB)
• International standard for telecommunications

CRC-16-IBM: 0x8005 = x¹⁶ + x¹⁵ + x² + 1

The IBM polynomial also excellent:

• Factors as (x + 1) × (x + 1) × (primitive polynomial of degree 14)
• Strong burst detection
• Popular in computing and industrial control
• Used in Modbus, widely deployed in manufacturing

CRC-32 provides 32-bit (4 byte) checksums with approximately 1 in 4 billion undetected error probability. This level of protection is suitable for network frames up to megabytes and storage blocks up to gigabytes.

Applications:

• Ethernet frames (IEEE 802.3)
• ZIP, GZIP, PNG, MPEG-2
• SATA storage
• POSIX cksum utility
• iSCSI, SCTP (CRC-32C)

CRC-32 (Ethernet): The Original Standard

Polynomial: 0x04C11DB7 = x³² + x²⁶ + x²³ + x²² + x¹⁶ + x¹² + x¹¹ + x¹⁰ + x⁸ + x⁷ + x⁵ + x⁴ + x² + x + 1

This polynomial was selected in the 1970s by Wesley Peterson and has remained the dominant choice. Properties:

• Contains (x + 1) factor: Detects all odd-bit-count errors
• Provides Hamming distance 5 for messages up to 268 Mb (very long!)
• Standard for 40+ years across networking and storage

Every Ethernet frame you've ever sent has been protected by this polynomial.

CRC-32C (Castagnoli): The Modern Improvement

Polynomial: 0x1EDC6F41 = x³² + x²⁸ + x²⁷ + x²⁶ + x²⁵ + x²³ + x²² + x²⁰ + x¹⁹ + x¹⁸ + x¹⁴ + x¹³ + x¹¹ + x¹⁰ + x⁹ + x⁸ + x⁶ + 1

Developed by Guy Castagnoli, this polynomial provides better error detection in several scenarios:

• Better burst detection for certain burst lengths
• Slightly better random error detection
• Has (x + 1) factor: Detects all odd-bit-count errors
• Hardware accelerated: Intel SSE4.2 includes CRC32C instruction

Used by iSCSI (storage over IP), SCTP (reliable transport), and modern file systems like Btrfs.

CRC-64 provides 64-bit (8 byte) checksums with approximately 1 in 10¹⁹ undetected error probability. This extreme reliability is required for large-scale storage systems and high-reliability applications where even CRC-32's 1-in-4-billion miss rate is insufficient.

Applications:

• ECMA-182 tape storage
• XZ compression format
• High-reliability databases
• Distributed storage systems
• DNA sequence analysis (for error detection in sequence data)

The ECMA-182 Polynomial:

0x42F0E1EBA9EA3693 = x⁶⁴ + x⁶² + x⁵⁷ + x⁵⁵ + x⁵⁴ + x⁵³ + x⁵² + x⁴⁷ + x⁴⁶ + x⁴⁵ + x⁴⁰ + x³⁹ + x³⁸ + x³⁷ + x³⁵ + x³³ + x³² + x³¹ + x²⁹ + x²⁷ + x²⁴ + x²³ + x²² + x²¹ + x¹⁹ + x¹⁷ + x¹³ + x¹² + x¹⁰ + x⁹ + x⁷ + x⁴ + x + 1

This polynomial was carefully selected for tape storage where:

• Data blocks are very large (megabytes)
• Physical error mechanisms can cause specific burst patterns
• Archival reliability is critical (decades of storage)

When Is CRC-64 Necessary?

CRC-64 is overkill for most applications. Consider it when:

• Data blocks exceed gigabytes
• System requires multiple nines of reliability (99.9999%+)
• Storage duration extends beyond years
• Error rate multiplied by data volume makes CRC-32 inadequate

Example Calculation:

A petabyte storage system with 4KB blocks:

• 250 million blocks
• CRC-32 miss rate ≈ 2.3 × 10⁻¹⁰ per block
• Expected undetected errors: 250M × 2.3 × 10⁻¹⁰ ≈ 0.06 per petabyte

With CRC-64:

• Miss rate ≈ 5 × 10⁻²⁰ per block
• Expected undetected errors: 250M × 5 × 10⁻²⁰ ≈ 10⁻¹¹ per petabyte

For truly massive storage, CRC-64 provides meaningful improvement.

Beyond the main families, several specialized CRCs serve specific protocols and applications:

CRC-5 (USB Token Packets) Polynomial: 0x05, 5 bits, used in USB for short token packets (handshake, address). Minimal overhead for very small packets.

CRC-7 (SD/MMC Cards) Polynomial: 0x09, 7 bits, used in memory card command packets. Optimized for short command sequences.

CRC-10 (ATM AAL) Polynomial: 0x233, 10 bits, used in ATM Adaptation Layer for cell payloads.

CRC-12 (Telecommunications) Polynomial: 0x80F, 12 bits, used in telecommunications for 6-bit character streams.

CRC-24 (Bluetooth, OpenPGP) Polynomial: 0x864CFB, 24 bits. Strong enough for Bluetooth enhanced data rate packets and PGP encrypted messages.

Automotive CRCs:

Automotive applications have strict CRC requirements due to safety criticality:

• CAN bus (CRC-15): Every CAN frame includes a 15-bit CRC to protect the identifier and data. The polynomial 0x4599 was chosen for excellent burst detection in the noisy automotive electrical environment.

• FlexRay (CRC-11): Used in high-speed automotive networks for powertrain control. The polynomial provides guaranteed detection of specific error patterns matching EMI characteristics.

• LIN (CRC-8): Lighter-weight protocol for subsystems uses CRC-8 to minimize overhead on slow serial links.

Protocol Stack CRCs:

A single packet may traverse multiple CRC-protected layers:

Application Data
    └── CRC-32 (if file transfer with integrity check)
         └── TCP Segment (TCP checksum, not CRC)
              └── IP Packet (IP header checksum)
                   └── Ethernet Frame (CRC-32 FCS)
                        └── Physical Layer (possibly FEC)

Each layer provides independent error protection optimized for its scope.

Given the variety of CRC standards, how do you choose the right one for a new application?

Decision Framework: