Digital Communications

Compress-and-Forward

Q: When does compress-and-forward outperform decode-and-forward?

Compress-and-forward (CF) wins when the relay is positioned closer to the destination than to the source. Decode-and-forward (DF) requires the relay to fully decode the source message, so its rate is capped by the source-to-relay link capacity; if that link is weak, DF throttles the whole route. CF never decodes, so it is not limited by the source-to-relay channel. Instead it forwards a compressed description of its noisy observation, and the destination exploits its own direct-link signal as Wyner-Ziv side information. When the relay-to-destination link is strong (relay near the destination), CF can approach the cut-set upper bound to within a constant gap, while DF falls short. Conversely, when the relay is near the source and can decode reliably, DF is the better choice.

Q: Why does compress-and-forward use Wyner-Ziv coding instead of plain quantization?

The destination already has a correlated, noisy copy of the source signal from the direct link. Wyner-Ziv (lossy source coding with decoder side information) lets the relay compress its observation at a lower rate than ordinary quantization would require, because it only needs to convey the information the destination does not already have. The relay bins the quantization codewords and transmits only the bin index, similar to Slepian-Wolf binning for lossless coding. The destination uses its direct-link observation to resolve which codeword within the bin was sent. This binning gain reduces the required relay-to-destination rate by the mutual information between the relay observation and the destination side information, which directly raises the end-to-end achievable rate.

Q: How does the relay quantization distortion affect achievable rate in compress-and-forward?

There is a fundamental trade-off set by the relay-to-destination link capacity. Finer quantization (lower distortion, larger quantization noise variance N_q approaching zero) preserves more of the relay observation but demands a higher compression rate that may exceed the relay-to-destination link, forcing a quantization-rate constraint violation. Coarser quantization fits the link budget but injects more quantization noise into the signal the destination combines. The optimum balances the compression index rate against the relay-to-destination capacity C_RD, typically setting the quantization noise so that the compressed description rate exactly equals the side-information-aided Wyner-Ziv rate that C_RD can support. In Gaussian channels the optimal quantization noise variance scales with the destination SNR and the direct-link SNR.

/kuhm-PRES and FOR-werd/

A cooperative relaying strategy for the three-node relay channel in which the relay never decodes the source message. Instead it quantizes its noisy received observation, compresses the quantization using Wyner-Ziv coding against the destination's correlated direct-link signal, and forwards only the compression bin index. The destination then combines the recovered relay description with its own observation to decode. Because it sidesteps the decoding bottleneck of decode-and-forward, compress-and-forward (CF) excels when the relay sits near the destination, approaching the cut-set bound within a constant gap of 0.5 bit per dimension in the Gaussian relay channel.

Category: Digital Communications

Relay role: Quantize, no decoding

Gaussian gap: ≤ 0.5 bit/dim

How the Relay Compresses Instead of Decoding

Compress-and-forward emerged from Cover and El Gamal's 1979 framework for the relay channel, where it was originally called the "Theorem 6" or estimate-and-forward strategy. Unlike amplify-and-forward, which simply scales and retransmits the analog received signal (and with it the relay's own noise), CF performs a digital quantization of the relay observation Y_R into a reproduction Ŷ_R. The relay does not attempt to recover the transmitted bits at all. This is the defining contrast with decode-and-forward, whose achievable rate is hard-limited by the source-to-relay link capacity.

The clever part is exploiting the destination's prior knowledge. The destination already receives a degraded, correlated copy of the source signal over the direct link. CF treats that direct-link observation as decoder side information and applies Wyner-Ziv lossy source coding: the relay bins its quantization codewords (Slepian-Wolf style) and transmits only the bin index, at a rate reduced by the mutual information the destination can supply on its own. The destination first decodes the bin index, uses its side information to pick the correct codeword within the bin, then performs joint maximum-likelihood decoding of the original message using both signal paths.

The performance hinges on a budget equation: the Wyner-Ziv description rate must fit within the relay-to-destination link capacity C_RD. Choosing the quantization noise variance N_q too small produces a description too rich to transmit; too large, and excessive quantization noise corrupts the combined estimate. The optimum sets N_q so the compression rate exactly saturates C_RD, which is why CF rewards a strong relay-to-destination channel.

Achievable Rate and the Wyner-Ziv Constraint

CF achievable rate (Gaussian relay channel):
R_CF = I(X; Y_D, Ŷ_R | X_R)

Wyner-Ziv compression constraint:
I(Y_R; Ŷ_R | X_R, Y_D) ≤ I(X_R; Y_D | X) ≡ C_RD

Optimal Gaussian quantization noise:
N_q = N_R × (1 + SNR_SR + SNR_SD) / (2^C_RD − 1)

Where Y_R = relay observation, Ŷ_R = quantized reproduction, Y_D = destination direct-link signal, X / X_R = source / relay transmit symbols, N_R = relay noise power, SNR_SR and SNR_SD = source-to-relay and source-to-destination SNRs. The CF rate stays within 0.5 bit of the cut-set bound for all channel gains.

Relaying Strategy Comparison

Strategy	Relay processing	Limited by S→R link?	Best relay position	Noise behavior	Complexity
Compress-and-Forward	Quantize + Wyner-Ziv compress	No	Near destination	Adds quantization noise N_q	High (binning + joint decode)
Decode-and-Forward	Full decode + re-encode	Yes (rate ≤ C_SR)	Near source	Noise-free re-transmission	Medium
Amplify-and-Forward	Scale analog signal	No	Midpoint	Amplifies relay noise	Low
Quantize-and-Forward	Quantize, no side-info binning	No	Near destination	Higher N_q than CF	Medium

Common Questions

Frequently Asked Questions

When does compress-and-forward outperform decode-and-forward?

CF wins when the relay sits closer to the destination than to the source. Decode-and-forward must fully decode the source message, so its rate is capped by the source-to-relay capacity C_SR; a weak first hop throttles the entire route. CF never decodes and so escapes that limit, instead forwarding a compressed description that the destination resolves using its direct-link signal as side information. With a strong relay-to-destination link, CF approaches the cut-set bound within 0.5 bit. When the relay is near the source and can decode reliably, DF is preferable.

Why does compress-and-forward use Wyner-Ziv coding instead of plain quantization?

The destination already holds a correlated noisy copy of the source from the direct link. Wyner-Ziv lossy coding with decoder side information lets the relay compress at a lower rate than ordinary quantization, conveying only what the destination lacks. The relay bins quantization codewords (Slepian-Wolf style) and sends just the bin index; the destination uses its direct-link observation to pick the correct codeword inside the bin. This binning gain reduces the required relay-to-destination rate by I(Y_R; Y_D), directly raising end-to-end rate.

How does the relay quantization distortion affect achievable rate in compress-and-forward?

It is a budget trade-off against C_RD. Fine quantization (small N_q) preserves the observation but needs a description rate that may exceed the relay-to-destination link. Coarse quantization fits the budget but injects more quantization noise into the combined estimate. The optimum sets N_q so the Wyner-Ziv compression rate exactly equals C_RD; in Gaussian channels N_q scales with the relay noise and the source-link SNRs, which is why a high-capacity relay-to-destination channel lets the relay quantize finely and deliver near-optimal rate.

Related Terms

← Compound Slot Compressed Block ACK →