Viterbi IP Core User Guide

ID 683280
Date 11/06/2017
Public
Document Table of Contents

3.3.4. Trellis Termination

Block decoders must properly decode the last bits of the block and adapt to the convolutional encoder.

Tail-biting feeds the convolutional encoder with a block and terminates it with (L – 1) unknown bits taken from the end of the block. Tail-biting sets the initial state of the convolutional encoder with the last (L – 1) information bits. Tail-biting is decoded by replicating the block at the decoder or double feeding the block into the decoder. By decoding in the middle point, the trellis is forced into the state that is both the initial state and the end state. From the first decoding block, you can take the last half of the block; from the second decoded block (or second pass through the decoder), you can obtain the first half of the bits of the block.

Note: In tail-biting, the block size must be large enough to train the decoder, otherwise you may see BER loss.

Alternatively, if you initialize the convolutional encoder to zero, the initial state of the trellis is zero. The decoder knows the last (L – 1) bits to the convolutional encoder. They bring the convolutional encoder to a known end state. The decoder then uses this information to set the end state of the trellis with tr_init_state, which is derived from the last (L – 1) bits of the block in reverse order. For example, for a block that ends in: ...000101 If L = 5 and the decoder knows the last (L – 1) = 4 bits, it sets tr_init_state as 0101, which reversed and in binary is 1010, or 10 in decimal. The wizard generates tr_init_state as if it knows the last (L – 1) bits of each block.