Make your paper interactiveCreate

Text from the interactive explanation

Near-Capacity Cooperative Space-Time Coding Employing Irregular Design and Successive Relaying

Lingkun Kong, Soon Xin Ng, Robert G. Maunder, Lajos Hanzo

Original paper

Successive relaying · Kong, Ng, Maunder & Hanzo · 2010

A radio can't listen while it talks. So when a nearby phone helps relay your signal, everything slows to half speed. Could two helpers, taking turns, carry a faint signal at full speed—and almost as efficiently as physics allows?

Illustration of the paper's successive-relaying protocol (its Fig. 1). In the strip below the scene, each column is one time slot: a solid block means sending, a dashed block means listening, a dotted outline means resting; colours are frame numbers.

1 · Distance and fading

Why ask for help?

Radio waves spread out and weaken with distance, and obstacles and movement make their strength flicker at random, an effect engineers call fading . What decides whether a message gets through is the signal-to-noise ratio (SNR) : how strong the signal is compared with the random hiss, or noise , that every receiver picks up.

In the paper's example the two helpers sit much nearer the source than the destination does. Each helper hears the source 8 times more strongly than the destination hears it directly, and the destination hears a helper 2 times more strongly than it hears the source.

Three receivers, one transmission

One antenna, no error-correcting code, 4QAM symbols; error rates from 24 000 random symbols per receiver. "Equivalent SNR" is the transmit power compared with the receiver noise; each receiver then gains its link's factor (×1, ×8 or ×2). The bar beside each plot is the signal strength at this moment; with fading on it flickers, and the receiver is assumed to know each fade.

A helper close to the source hears it far more cleanly than the distant destination does, and the destination hears a helper better than it hears the source. That is why the paper routes each message through helpers.

2 · Half-duplex radios

The catch: nobody can talk and listen at once

A helper has to receive a message before it can pass it on. Why not do both at once? Because a radio that transmits and receives on the same frequency at the same time hears itself. As the paper puts it, the received signal "may be almost 100 dB lower than the transmit signal": ten billion times weaker. The smallest leak of the helper's own transmission swamps its receiver, so practical radios are half-duplex : they either talk or listen in any one moment.

Inside a helper's receiver

Receiver: listening. The faint incoming signal is clear.

A helper must alternate: listen to a frame in one time slot, forward it in the next. With a single helper, the source has to fall silent while the helper speaks, so only half the time slots carry new data.

3 · Successive relaying

One helper halves the speed. Two take turns.

Chop the message into frames (blocks of bits) and time into equal slots . With one helper, every frame costs two slots: the source sends it, then the helper forwards it while the source waits. The paper calls this the "factor 0.5 multiplexing loss".

Successive relaying adds a second helper. While helper 1 forwards frame 1, helper 2 listens to frame 2; then they swap. The source sends a fresh frame in every slot and never pauses.

Two helpers take turns. How many time slots does it take to deliver 5 frames?

Make a guess, then check it against the timeline below.

Who does what, slot by slot

Two helpers deliver N frames in N + 1 slots instead of 2N. For long batches that is almost the full speed of a direct link: the half-speed penalty is recovered.

4 · Untangling the overlap

Two voices at once

Taking turns has a price. In every middle slot the destination hears two frames on top of each other: the previous frame, forwarded by a helper, and the new frame, straight from the source. The helper is closer, so its copy arrives twice as strong (the paper's Grd = 2).

The destination uses successive interference cancellation (SIC) : read the louder frame first, treating the quieter one as noise; subtract it; then read the quieter frame from what is left. Each received point below is the sum of one helper symbol and one source symbol, so there are 4 × 4 = 16 clusters.

Peel off the louder frame

Read the louder frame

Subtract it

Read the quieter frame

Single antenna, no error-correcting code, helper frame twice the power of the source frame; error rates from 20 000 random symbol pairs. In the paper the destination does this iteratively with 4 receive antennas and strong codes, and also uses the copy of the earlier frame that it heard directly in the previous slot.

Reading the louder frame first and subtracting it leaves the quieter frame almost clean. Skip that step and the quieter frame is buried: about half its bits come out wrong.

5 · The speed limit

How close to the limit?

Every noisy channel has a capacity : the most information that can get through per use with vanishingly few errors, a limit established by Claude Shannon. The authors calculated it for successive relaying with their antennas and 4QAM symbols, which can carry at most 2 bits per channel use.

The horizontal axis is the paper's equivalent SNR : the transmit power compared with the receiver noise, before any boost from the helpers being closer. Drag across the chart or use the slider.

Achievable rate vs. power

Redrawn (approximately) from Fig. 5 of Kong et al. 2010: constrained information rates for 4QAM with Alamouti's two-antenna code, helpers with 8× and 2× link gains. The band spans the paper's lower and upper bounds. The dashed line at 1 bit is an illustration of a halved ceiling, not a curve from the paper.

With two helpers the ceiling stays at 2 bits per channel use instead of being halved, and 1 bit per use needs only about −11.2 dB, roughly 2.6 dB less power than the direct link's ≈ −8.6 dB.

6 · Near-capacity codes

Decoders that pass notes

To get near that limit, each frame is protected by an error-correcting code : extra, cleverly chosen bits that let mistakes be fixed. The receiver runs two decoders that trade hints back and forth, called extrinsic information . An EXIT chart tracks how much each decoder knows, from 0 (nothing) to 1 (certain).

The teal curve shows what the inner decoder knows after being told a given amount; the amber curve is the outer decoder's answer, drawn flipped. Decoding climbs a staircase between them. If an open tunnel separates the curves, the staircase reaches the top-right corner and errors all but vanish; if the curves touch, it stalls.

An irregular code blends 17 sub-codes, from strong ones (rate 0.10: one data bit per ten sent) to light ones (rate 0.90). Choosing the blend so its curve hugs the teal one keeps the tunnel narrow but open, which is how the paper works close to capacity.

Open the tunnel

Model curves are simplified stand-ins calibrated to two curves in Fig. 6; the paper's curves come from Monte-Carlo simulation. In this model a single rate-½ code needs about −1.0 dB, a matched blend works down to about −2.0 dB, and nothing can work below about −2.4 dB, where the area under the teal curve drops below ½. Bit-error figures here are estimates from the final information level.

The authors' real recipes

A tunnel can only open if the area under the teal curve beats the code rate (½). A single ordinary code wastes part of that margin; a tailored blend hugs the curve and keeps working at lower power.

7 · The design recipe

Designing the source and helpers together

Every link in the network needs its own code, and they all interact. The paper's contribution is a step-by-step joint design that matches each code to its link with EXIT charts. Step through it:

Step 1

By matching the source's code to the helpers, then each helper's code to what the destination already knows, the whole cooperative scheme can work close to its capacity limit. The authors say the procedure applies to "virtually any network configuration".

8 · Results

What they found

The authors simulated the complete system on a computer: fading links, multiple antennas, their codes and 250 000-bit frames. They measured the bit-error ratio (BER) , the fraction of bits that arrive wrong. Each curve drops off a cliff once the power is high enough. The further left the cliff, the less power the scheme needs.

Bit-error ratio vs. power

Redrawn (approximately) from Fig. 8; Monte-Carlo simulations, 250 000-bit frames. The pictures flip each pixel (one bit) at random with the BER read off the redrawn curves (log-linear interpolation). Past the end of a curve the BER is below one error per million bits in the simulation, shown as a clean picture.

In simulation, the jointly designed two-helper scheme reaches error-free operation at about −9.7 dB, within 1.5 dB of its capacity and about 2.2 dB ahead of transmitting without helpers.

9 · Why it matters

Idle phones, full speed

Back to the opening question. A second helper lets a half-duplex network recover the half-speed penalty without full-duplex radios, which the authors call "more practical and feasible". Extra helpers need not mean extra infrastructure "when the relays are constituted by inactive mobile users": phones that happen to be nearby and idle.

The design recipe is not tied to this network: the authors say it applies to "virtually any network configuration" and to other irregular codes, such as low-density parity-check (LDPC) codes. Their stated next step is the case set aside here, interference between the two helpers, which would require suppressing interference at the helpers as well as at the destination.

The paper

Near-Capacity Cooperative Space-Time Coding Employing Irregular Design and Successive Relaying

Lingkun Kong, Soon Xin Ng, Robert G. Maunder and Lajos Hanzo

IEEE Transactions on Communications, vol. 58, no. 8, pp. 2232–2241, August 2010. DOI: 10.1109/TCOMM.2010.08.090646

Credits: the relay animation and timelines illustrate the protocol of Fig. 1; the capacity and bit-error charts are approximate redrawings of Figs. 5 and 8; the EXIT game uses a simplified model calibrated to Fig. 6, with the sub-code weights printed in Figs. 6 and 7; the constellation and cancellation demos are simplified simulations. All figures are redrawn from the paper's data; no figure images are reproduced.

Topics in this iPaper

Engineering
such as cellular networks, wireless ad-hoc and sensor networks, and wireless local area networks

Section I-A (Background), page 2232

Computer science
we first derive both the upper and lower bounds of the Continuous-input Continuous-output

Section I-B (Contribution), page 2233

cooperative communication
is capable of near-capacity cooperative communications in the twin-relay successive relaying aided network

Section VI (Conclusions), page 2240

successive relaying
a successive relaying protocol in this paper, where an additional relay node is activated

Abstract, page 2232

space time block code
we employ Alamouti’s 4QAM-based G2 scheme at both the source and relay nodes

Section IV-A, page 2236

irregular convolutional code
irregular convolutional code, EXIT charts

Index Terms, page 2232

extrinsic information transfer chart
jointly designed with the aid of EXtrinsic Information Transfer (EXIT) charts

Abstract, page 2232

successive interference cancellation
frame-by-frame Successive Interference Cancellation (SIC) aided

Abstract, page 2232

iterative decoding
A frame-by-frame SIC aided iterative decoder of the Ir-CSTC scheme

Fig. 3 caption, page 2235

channel capacity
CSTC scheme is capable of performing within 1.5 dB of the

Section V, page 2240

Near-Capacity Cooperative Space-Time Coding Employing Irregular Design and Successive Relaying — Interactive explanation | iPaper.ai