Soft-Input-Soft-Output decoder

Description

A Soft-Input-Soft-Output (SISO) decoder is a fundamental component in the receiver chain for modern error-correcting codes like Turbo Codes and Low-Density Parity-Check (LDPC) codes, which are pivotal in 3GPP standards from HSPA+ onwards. Its operation is based on probabilistic or 'soft' information, as opposed to traditional 'hard-decision' decoders that only process binary 1s and 0s. The input to a SISO decoder consists of Log-Likelihood Ratios (LLRs) for each received bit. An LLR is a soft value that indicates both the likely binary value (sign) and the confidence or reliability (magnitude) of that decision based on the received signal and noise.

The internal architecture of a SISO decoder is tailored to the specific code structure. For a Turbo code decoder, it typically comprises two constituent SISO decoders (often based on the BCJR or MAP algorithm) that operate on the two convolutional encoders used in the Turbo encoder. The decoding process is iterative. In the first iteration, the first SISO decoder processes the systematic bits and the first set of parity bits, using initial a priori LLRs (often set to zero). It produces extrinsic information, which is new probabilistic information gleaned from the code's constraints. This extrinsic output is interleaved and passed as a priori input to the second SISO decoder, which processes the interleaved systematic bits and the second set of parity bits. The second decoder also generates extrinsic information, which is de-interleaved and fed back to the first decoder for the next iteration. This loop continues for a fixed number of iterations or until a convergence criterion is met.

After the final iteration, the SISO decoder(s) combine the intrinsic channel LLRs and the accumulated extrinsic information to produce final soft-output LLRs for each bit. These outputs can then be used for hard decision (by taking the sign) to recover the original bits. The power of the SISO decoder lies in this iterative exchange of soft information, which allows it to correct errors that are far beyond the capability of a single-pass hard decoder. In 3GPP, SISO decoding principles are essential for achieving the high spectral efficiency and near-Shannon-limit performance required for high-speed data services in LTE and 5G NR, where LDPC codes for data channels also employ iterative belief propagation decoding, a form of SISO decoding across a factor graph.

Purpose & Motivation

The SISO decoder was developed to unlock the performance potential of iterative error-correcting codes, primarily Turbo codes, which were a breakthrough in channel coding theory. Traditional hard-decision decoders (e.g., Viterbi decoders) made irreversible decisions early in the process, discarding valuable reliability information. This limited the performance gain achievable by more complex codes.

The fundamental problem SISO decoding solves is how to efficiently decode codes with long block lengths and complex interdependencies between bits, such as Turbo and LDPC codes. These codes are constructed to have a probabilistic graphical structure where bits are related through parity constraints. A SISO decoder's purpose is to navigate this graph iteratively, propagating and refining probabilistic beliefs about each bit's value. By preserving and processing soft information throughout the decoding process, it allows the decoder to correct errors that appear uncorrectable in early iterations, as later iterations use the growing consensus from the code's structure to resolve ambiguities.

Its adoption in 3GPP, starting with the inclusion of Turbo codes for high-speed data in Release 5 (HSDPA) and solidifying in LTE and NR, was motivated by the relentless demand for higher data rates within limited bandwidth. SISO-enabled iterative decoding allows systems to operate at much lower Signal-to-Noise Ratios (SNR) for a given error rate, or to use higher-order modulation for a given SNR. This directly translates to higher user throughput and better cell-edge coverage. Without SISO decoders, the high spectral efficiency targets of 4G and 5G would be unattainable, making them a cornerstone technology for modern mobile broadband.

Key Features

• Processes and generates Log-Likelihood Ratios (LLRs), preserving probabilistic reliability information.
• Enables iterative decoding algorithms crucial for Turbo codes and LDPC codes.
• Core component of the BCJR/MAP algorithm for Turbo decoding and belief propagation for LDPC.
• Exchanges extrinsic information between constituent decoders (in Turbo codes) or check/variable nodes (in LDPC) to refine bit estimates.
• Allows decoding to approach the theoretical Shannon capacity limit of a channel.
• Fundamental to achieving the high spectral efficiency and link reliability in LTE and 5G NR data channels.

Evolution Across Releases

Defining Specifications