4.3 HARD- AND SOFT-DECISION DECODING

When a digital signal arrives at the receiver, it has usually been corrupted to some extent by noise, interference, fading, and other channel impairments. As a result, the receiver rarely observes perfectly formed binary symbols. Instead, each received signal lies somewhere between the ideal values corresponding to a binary 0 and a binary 1. The decoder must therefore determine not only which bit was most likely transmitted, but also how confident it is in that decision.

Two broad approaches are used for this purpose: hard-decision decoding and soft-decision decoding. The distinction lies in how much information the demodulator passes to the channel decoder.

In hard-decision decoding, the demodulator converts each received symbol directly into a binary value by comparing it with a decision threshold. If the received signal lies on one side of the threshold, the output is declared to be a 0; if it lies on the other side, it is declared to be a 1. Any additional information about how close the received signal was to the threshold is discarded. The decoder therefore operates using only binary values.

For example, suppose a binary modulation scheme represents a transmitted 0 by a voltage of –1 V and a transmitted 1 by +1 V. If noise causes the receiver to observe +0.9 V, +0.3 V, or +0.01 V, all three values are converted to exactly the same hard decision of "1", even though the receiver is much less confident about the last measurement than the first.

Soft-decision decoding retains this additional information. Rather than producing only a binary decision, the demodulator also provides a measure of the reliability of each received bit. This reliability information is commonly expressed as a quantized confidence value or, in modern iterative decoders, as a log-likelihood ratio (LLR). A large positive LLR indicates that a 1 is highly probable, a large negative value indicates that a 0 is highly probable, while values close to zero indicate considerable uncertainty.

The decoder uses these reliability measures to make better decisions. Bits received with high confidence are treated differently from those received near the decision threshold, allowing the decoder to distinguish between highly reliable and potentially erroneous symbols. By exploiting this additional information, the decoder can often recover the original codeword even when several received bits appear ambiguous.

The improvement achieved by soft-decision decoding can be substantial. For many practical coding schemes, including convolutional, turbo, LDPC, and polar codes, soft-decision decoding provides a coding gain of up to approximately 2 dB compared with hard-decision decoding. In practical terms, this means that the same error performance can often be achieved with a lower received signal power or over a greater transmission distance.

This improved performance comes at the cost of increased implementation complexity. Soft-decision decoders must store and process reliability information for every received symbol, requiring more memory and significantly greater computational capability than hard-decision decoders. Historically, this complexity limited their use to specialized applications. However, advances in digital signal processing and integrated circuits have made soft-decision decoding practical for almost all modern communication systems.

Today, hard-decision decoding remains attractive for simple, low-cost systems where implementation complexity and power consumption are more important than achieving the lowest possible error rate. Soft-decision decoding, however, has become the standard approach in high-performance communication systems, including contemporary cellular networks, Wi-Fi, optical fibre systems, satellite communications, digital television, and deep-space communications, where its significant improvement in reliability more than justifies the additional processing required.