Cyclic Redundancy Check (CRC)

A cyclic redundancy check adds parity bits to detect transmission errors. The following code snippets show how to add CRC parity bits to a bit sequence and how to verify that the check is valid.

First, we need to create instances of CRCEncoder and CRCDecoder:

encoder = CRCEncoder(crc_degree="CRC24A") # the crc_degree denotes the number of added parity bits and is taken from the 3GPP 5G NR standard.

decoder = CRCDecoder(crc_encoder=encoder) # the decoder must be associated to a specific encoder

We can now run the CRC encoder and test if the CRC holds:

# u contains the information bits to be encoded and has shape [...,k].
# c contains u and the CRC parity bits. It has shape [...,k+k_crc].
c = encoder(u)

# u_hat contains the information bits without parity bits and has shape [...,k].
# crc_valid contains a boolean per codeword that indicates if the CRC validation was successful.
# It has shape [...,1].
u_hat, crc_valid = decoder(c)

CRCEncoder

class sionna.fec.crc.CRCEncoder(crc_degree, output_dtype=tf.float32, **kwargs)[source]

Adds cyclic redundancy check to input sequence.

The CRC polynomials from Sec. 5.1 in [3GPPTS38212_CRC] are available: {CRC24A, CRC24B, CRC24C, CRC16, CRC11, CRC6}.

The class inherits from the Keras layer class and can be used as layer in a Keras model.

Parameters:

crc_degree (str) – Defining the CRC polynomial to be used. Can be any value from {CRC24A, CRC24B, CRC24C, CRC16, CRC11, CRC6}.
dtype (tf.DType) – Defaults to tf.float32. Defines the output dtype.

Input:

inputs ([…,k], tf.float32) – 2+D tensor of arbitrary shape where the last dimension is […,k]. Must have at least rank two.

Output:

x_crc ([…,k+crc_degree], tf.float32) – 2+D tensor containing CRC encoded bits of same shape as inputs except the last dimension changes to […,k+crc_degree].

Raises:

AssertionError – If crc_degree is not str.
ValueError – If requested CRC polynomial is not available in [3GPPTS38212_CRC].
InvalidArgumentError – When rank(inputs)<2.

Note

For performance enhancements, we implement a generator-matrix based implementation for fixed k instead of the more common shift register-based operations. Thus, the encoder need to trigger an (internal) rebuild if k changes.

property crc_degree: CRC degree as string.

property crc_length: Length of CRC. Equals number of CRC parity bits.

property crc_pol: CRC polynomial in binary representation.

property k: Number of information bits per codeword.

property n: Number of codeword bits.

CRCDecoder

class sionna.fec.crc.CRCDecoder(crc_encoder, dtype=None, **kwargs)[source]

Allows cyclic redundancy check verification and removes parity-bits.