Unformatted text preview: G3 • + 4 •
• + + = Take every other symbol + out 1b + + ••
••
••
RATE 1/2
RATE 1/3
RATE 1/4
RATE 1/6 ENCODERb •
• •• •• Not used + out 3b • ••
• Figure 73: Turbo Encoder Block Diagram
The two convolutional encoders in the Recommended Standard (reference [3]) are recursive
with constraint length K = 5, and are realized by feedback shift registers. However, unlike the
encoder for the recommended plain convolutional code in section 4, the turbo codeblock is
terminated by running each encoder for an additional K1 bit times beyond the end of the
information bit frame. After encoding the last bit in the frame, the leftmost adder in each
component encoder receives two copies of the same feedback bit, causing it to zero its
output. After K1 more bit times, all 4 memory cells become filled with zeros, but in the
interim the encoder continues to output nonzero encoded symbols.
The Recommended Standard (reference [3]) allows options for nonpunctured codes with
rates between 1/3, 1/4, and 1/6. The puncturer is used only for code rate 1/2.
The interleaver in the Recommended Standard (reference [3]) is based on a permutation rule
which can be computed onthefly or precomputed and stored in a lookup table, for all
allowable frame lengths (1784 to 16384 bits).
In figure 72, CLK indicates the frame clock. It is used: (1) by the input buffer to determine
when to empty and refill the buffer; (2) by the output buffer/multiplexer to determine when
to insert the frame sync marker; (3) by each of the convolutional encoders to determine when
to terminate the codeblock. Note that an entire information block of k bits must be read in
before the encoding can proceed, because some of the bits in the tail end of block will be
permuted to the front and need to be encoded first. Thus, there is a fundamental encoding
latency of at least k bits in the encoding process. CCSDS 130.1G1 Page 73 June 2006 TM SYNCHRONIZATION AND CHANNEL CODING —SUMMARY OF CONCEPT AND RATIONALE The turbo code introduces a couple of unique encoder complexity issues. The information
block needs to be buffered and read out in a permuted order as part of the encoding process.
This buffering has no analog in the plain convolutional encoder, but the size of this buffer is
comparable to that required for an interleaved ReedSolomon codeblock of the same size.
The difference is that the traditional concatenated coding architecture completely separates
the ReedSolomon encoder (with its associated buffer) from the convolutional encoder. Thus,
the turbo encoder cannot be regarded as a plugin replacement for the convolutional encoder
hardware. The turbo encoder actually replaces the ReedSolomon/convolutional encoder
combination.
Another complexity consideration is how to implement the permutation. The best
permutations for turbo codes look very random, but this requires specifying a randomlooking readout order via a ROM (Table lookup). An alternative is to use a permutation that
can be generated by a simple rule rather than from a lookup table, with minor performance
sacrifice. The Recommended Standard (reference [3]) specifies a permutation based on a
simple rule, because it was preferred in terms of implementation on the spacecraft.
7.3 TURBO DECODER A turbo decoder uses an iterative decoding algorithm based on simple decoders individually
matched to the two simple constituent codes. Each constituent decoder makes likelihood
estimates derived initially without using any received parity symbols not encoded by its
corresponding constituent encoder. The (noisy) received uncoded information symbols are
available to both decoders for making these estimates. Each decoder sends its likelihood
estimates to the other decoder, and uses the corresponding estimates from the other decoder
to determine new likelihoods by extracting the ‘extrinsic information’ contained in the other
decoder’s estimates based on the parity symbols available only to it. Both decoders use the
‘a posteriori probability’ (APP) bitwise decoding algorithm, which requires the same number
of states as the wellknown Viterbi algorithm. The turbo decoder iterates between the outputs
of the two constituent decoders until reaching satisfactory convergence. The final output is a
hardquantized version of the likelihood estimates of either of the decoders.
The Recommended Standard (reference [3]) does not include a detailed description of the
specific turbo decoding algorithm. However, the performance curves in 7.4 for the turbo
code family in the Recommended Standard (reference [3]) were obtained using a decoding
algorithm with the following characteristics:
a) decoder type: Iterative ‘turbo’ decoding using two 16state component decoders (see
reference [18]);
b) type of component decoders: Softinput, softoutput APP decoders (see reference [19]);
c) quantization of channel symbols: At least 6 bits/symbol;
d) quantization of decoder metrics: At least 8 bits;
e) number...
View
Full Document
 Spring '14
 Information Theory, ........., Error detection and correction, CCSDS

Click to edit the document details