Text from the interactive explanation
Near-Capacity Code Design for Entanglement-Assisted Classical Communication over Quantum Depolarizing Channels
Zunaira Babar, Soon Xin Ng, Lajos Hanzo
Original paperSuperdense coding · near-capacity codes
With superdense coding , one quantum particle (a qubit) can carry two ordinary bits, if sender and receiver share entangled particles in advance. But noise on the link scrambles the message. Physics sets an absolute limit on how much noise a message can survive; this paper asks how close a practical error-correcting code can get to it.
Simulated Live, in your browser: each qubit is kicked with probability p, exactly as in the paper's channel model (its Eq. 7). There is no error correction yet, so a single kick garbles a whole character. The rest of this page is about the fix.
Chapter 1
One particle, two bits
Alice and Bob start by sharing a pair of entangled qubits: two particles whose states are linked, so that only the pair as a whole has a definite state. Bob's half travels to him ahead of time over a clean, noiseless route.
Later, Alice wants to send two bits. She does one of four things to her qubit: nothing, a flip (called X), a phase flip (Z), or both. Each choice moves the pair into a different Bell state , and the four Bell states can be told apart perfectly. She sends just her one qubit. Bob measures both qubits together and reads off the two bits. One qubit sent, two bits received: that is superdense coding, invented by Charles Bennett and Stephen Wiesner in 1992.
Pick the two bits Alice wants to send:
An original diagram in the spirit of the paper's Fig. 1. Colours follow Alice's choice and are used throughout this page.
A three-qubit version. The paper also uses 3-qubit superdense coding (3SD): three entangled qubits, of which Alice encodes and sends two while Bob already holds the third. Bob reads three bits from two qubits sent, so 1.5 bits per qubit. Below, the two-qubit scheme is called 2SD .
Chapter 2
What noise does
The channel studied in the paper is the depolarizing channel (its Eq. 1). With probability 1 − p a qubit passes untouched. With probability p it gets a random kick: an X, a Y or a Z, each equally likely (p/3 each). Y is the same as doing X and Z together.
Here is the key observation: a kick is exactly the kind of operation Alice uses to encode. A kicked qubit is not destroyed. It silently arrives as one of the other three messages. Try it:
Alice sends
The channel applies
So, seen from outside, the quantum link behaves just like an ordinary channel with four symbols: the right symbol arrives with probability 1 − p, and each particular wrong one with probability p/3 (the paper's Eq. 7). This is what lets classical error-correcting codes work on a quantum link. Send a thousand qubits and watch the table fill in:
Simulated Each symbol is scrambled using the paper's transition probabilities: Eq. 7 for 2SD, Eq. 9 for 3SD. In 3SD the two travelling qubits are kicked independently, so the eight-symbol table has uneven odds.
Chapter 3
The speed limit
Every noisy channel has a capacity : the most information per use that any code could ever deliver with vanishingly few errors. For superdense coding over this channel, with entanglement shared in advance, the ceiling is
C = 2 + (1−p)·log₂(1−p) + p·log₂(p/3) bits per qubit sent
(the paper's Eq. 2, from Bennett, Shor, Smolin and Thapliyal, 1999). With no noise it is 2 bits, and it falls as p grows.
Predict first. Suppose we ask for only 1 bit per qubit, half the noiseless maximum. Up to what noise level is that still possible? Tap the chart at your guess for p.
Computed Curves from the paper's Eqs. 2 and 6, and the bit-by-bit limit described with its Fig. 3. The paper reads the 3SD limit at 0.75 bits per qubit as p = 0.11 from that figure; the formula gives 0.107.
Bob's measurement yields a whole 2-bit symbol, and the noise affects the two bits together: an X kick flips one bit, a Z kick flips the other, a Y kick flips both. A decoder that treats the two bits as separate coin-flips forgets that link. Each bit is then simply wrong with probability 2p/3 (two of the three kicks touch it), and the most it can carry is
C_bit = 2 × (1 − h(2p/3)), h(q) = −q·log₂q − (1−q)·log₂(1−q)
The paper's decoders, like most practical ones, pass information around bit by bit, so this slightly lower curve is the honest target. The paper puts it plainly: this loss “cannot be recovered by any bit-based error correcting code”. At p = 0.15 the gap is 1.15 versus 1.06 bits per qubit.
Chapter 4
Redundancy, and two decoders that talk
To survive noise you add redundancy: send more symbols than the bare message, arranged so that errors can be spotted and undone. The transmitter in the paper is a chain of four steps (its Fig. 2):
Outer code (IRCC). Every message bit becomes two coded bits (a “rate-½” code), following fixed rules.
Interleaver π. A fixed shuffle of the coded bits, so errors that strike neighbours get spread apart.
Unity-rate code (URC). A running exclusive-or (XOR) that keeps the length but ties each symbol to the one before.
Superdense encoder. Each pair of bits becomes one of Alice's four operations on a qubit.
At the receiver, two decoders take turns. The inner decoder (superdense decoder plus URC) knows how the channel behaves. The outer decoder knows the code's rules. They pass each other soft hints , which are probabilities rather than hard yes/no verdicts, and with every round both grow more certain. This back-and-forth is called iterative decoding .
Illustration Original animation of the paper's Fig. 2. The certainty carried by the hint packets follows the decoding path predicted for this page's 2SD code at p = 0.135 (see Chapter 6).
Hard verdict or soft hint?
Bob's measurement returns 10.
A conventional superdense decoder announces a hard verdict and that is the end of it. The paper's new ingredient is a soft-decision superdense decoder : it reports how likely each message is, given what Bob saw (its Eq. 10, E(x) ≈ P(y|x)). These odds come straight from the channel: the received symbol is right with probability 1 − p, each other symbol p/3.
No room for doubt, nothing to revise.
Soft hints can be weighed against each other and revised as evidence comes in. That is what lets the two decoders hold a conversation at all.
Chapter 5
Watch a real decoder
Here is a small picture: 96 × 80 pixels in four colours, so two bits per pixel, 15,360 bits in all. On the left it is sent with no protection over the superdense link, one pixel per qubit. On the right it rides inside a 30,000-bit coded frame through the paper's full chain, and a real iterative decoder works on it, round by round, in your browser.
15,360 bits
–
–
Find the edge. Nudge the noise up in small steps from 0.13. Around 0.14–0.145 the decoder needs many more rounds; a little higher it stalls and the picture stays speckled. Watch the dots on the chart get stuck where the curves pinch together.
Swap the code. At p = 0.13, switch to one rate-½ code. Same rate, same channel, but its curve has the wrong shape, so the conversation dies early.
Three-qubit coding. Switch to 3SD at about p = 0.085. It carries 1.5 bits per qubit instead of 2, so it reaches its limit at a lower noise level.
This re-creation Real encoding, channel and log-MAP iterative decoding (up to 40 rounds). The dots on the chart are this run's measured information levels after each round, like the trajectory in the paper's Fig. 5. Expect the 17-code mix to clean the picture completely up to p ≈ 0.14 for 2SD, and to stall above that, while the single code gives up earlier.
Chapter 6
The tunnel
How do you know in advance whether the decoders' conversation will succeed? The paper's tool is an EXIT chart , a map of the conversation. Across the bottom: how much the inner decoder is told by its partner (from 0, nothing, to 1, everything). Up the side: how much it then knows. The cyan curve is the inner decoder; more noise pushes it down. The amber curve is the outer decoder, drawn with its axes swapped so the two can share one chart.
Decoding is a staircase bouncing between the curves. Each vertical step is the inner decoder's turn, each horizontal step the outer decoder's. If a tunnel stays open between them, the staircase climbs to the top-right corner, where every bit is known. If the curves touch, it gets stuck.
This re-creation Curves measured with this page's decoder; staircase as predicted by the chart. Illustration The picture's noise follows the predicted bit error rate after each step.
Why the URC is there. Switch to “SD alone”. A superdense decoder on its own cannot use hints: a measured symbol is what it is, so its curve is a flat line and the staircase stalls at once. The unity-rate code bends that line upward so it reaches the corner, while keeping the area under it the same (the paper's Fig. 5). Push the noise up and watch the cyan curve sink until it touches the amber one and the tunnel closes.
One caution the paper's own numbers show: the chart assumes an endlessly long frame and unlimited rounds. A real, finite frame needs a little extra room. The paper's tunnel stays open up to 0.4 dB from the limit, while its 30,000-bit simulations reach an error rate of 10⁻⁴ at 0.6 dB.
Chapter 7
Build the code
EXIT charts have a remarkable area property . The area under the inner curve is about the channel's capacity (per coded bit). The area under the flipped outer curve equals the code's rate. The sliver between them is capacity thrown away. So a near-capacity code needs a tunnel that is narrow, yet still open: the outer curve has to hug the inner curve's shape.
A single convolutional code has the wrong shape. An irregular code mixes 17 subcodes with rates from 0.10 to 0.90; you choose what fraction of the coded bits each one handles. Your turn at the mixing desk:
Each fader sets how much of the coded frame a subcode of that rate handles; shares are normalised to add up to 100%.
Goal: push your threshold (the highest noise at which the tunnel stays open) as close to the limit as you can. The paper reached 0.4 dB. Tip: raise p and press “Optimise for this p”. The closer to the limit you design, the narrower the tunnel, until no mix can keep it open.
This re-creation Threshold = highest p at which your mixed curve stays clear of the inner curve. Distance in decibels = 10·log₁₀(limit ÷ threshold); 0 dB would mean touching the limit.
Chapter 8
What the researchers found
Babar, Ng and Hanzo designed rate-½ irregular codes this way for both schemes and tested them by computer simulation, with a 30,000-bit interleaver and up to 32 decoding rounds. Everything in this chapter is reported in the paper.
Within 0.4 dB of the limit
Highest noise at which the EXIT tunnel stays open, against the bit-by-bit limit for the same throughput.
Distance from the limit at the EXIT threshold, in dB. The single-code value is computed here from the reported thresholds.
Against a turbo code
Distance from the limit at a bit error rate of 10⁻⁴ (the paper's Table I). The turbo code used the same number of trellis states per round, 16.
Lower is better. The irregular design is 1.3 to 1.75 dB closer to the limit.
Capacity left on the table
From the area property: area under the inner curve minus the code rate, scaled to bits per qubit sent.
Extra noise tolerated when the number of decoding rounds is doubled. Reaching the threshold itself takes many rounds, and that costs computation.
The winning mixes
Fraction of coded bits given to each of the 17 subcodes (rates 0.10 … 0.90).
The weight vectors ρ reported in the paper. The biggest share goes to the rate-0.40 subcode in both.
The paper's error-rate simulations (its Fig. 7) confirm the EXIT predictions: below the threshold, every extra round lowers the error rate; above it, more rounds don't help. For comparison, this page's re-creation (with stand-in subcodes) finds thresholds of – for 2SD and – for 3SD, and – for the single rate-½ code.
Chapter 9
Back to the link
Now send your message from the top of the page again, at noise p = 0.13 . Uncoded, as before, it arrives in tatters. With the full chain (17-code mix, interleaver, URC, soft superdense decoder) it rides inside a 30,000-bit frame, padded with filler bits, and the two decoders talk it back into shape.
This re-creation Both versions cross the same simulated channel at the same p. The coded version sends twice as many qubits, since the rate is ½.
How much noise can a quantum message survive? The ceiling for sending ordinary data over a noisy quantum link with shared entanglement was worked out in 1999. This paper shows that mature classical tools (irregular convolutional codes, a unity-rate precoder and EXIT-chart curve matching), together with a soft superdense decoder, get within 0.4 dB of it.
The paper
Near-Capacity Code Design for Entanglement-Assisted Classical Communication over Quantum Depolarizing Channels Zunaira Babar, Soon Xin Ng, Lajos Hanzo IEEE Transactions on Communications (accepted for publication), 2013 DOI: 10.1109/TCOMM.2013.111013.130035 Author copy: southampton.ac.uk/~sxn/paper/tcom-sdc.pdf
Also cited above: C. H. Bennett and S. J. Wiesner, Phys. Rev. Lett. 69, 2881 (1992), superdense coding; C. H. Bennett, P. W. Shor, J. A. Smolin, A. V. Thapliyal, Phys. Rev. Lett. 83, 3081 (1999), entanglement-assisted capacity.
How this page computes
Everything interactive is computed live in your browser. The quantum link is simulated through its equivalent classical channel, scrambling each symbol with the exact probabilities of the paper's Eqs. 7 and 9. Capacity curves use Eqs. 2 and 6 and the bit-by-bit limit. The image lab and the final message run a real decoder: rate-½ irregular convolutional code, interleaver, 1/(1+D) unity-rate code, the soft superdense decoder of Eq. 10 (its symbol probabilities turned into bit probabilities, treating the bits of a symbol as independent, as in the paper), and log-MAP iterative decoding on a 30,000-bit coded frame. EXIT curves were measured with the same decoder by Monte Carlo simulation.
One difference: the paper's 17 subcodes come from earlier work (Tüchler, 2004) and their exact generators are not listed in the paper. This page uses a stand-in family instead: one memory-4 recursive convolutional code (feedback 1 + D³ + D⁴) punctured to the same 17 rates. So this re-creation's thresholds differ slightly from the paper's. Numbers marked this re-creation come from this page; numbers marked reported in the paper are the authors' simulation results.
All visuals are original; no figures reproduced.
Topics in this iPaper
Engineering
the architecture of our proposed classical-quantum communication system
Section III, page 2
Physics
reliable transmission of classical information over a noisy quantum channel
Section I, page 1
Computer science
EXIT charts [7], [12], [13] are capable of visualizing the convergence behaviour of iterative decoding schemes
Section IV-A, page 3
superdense coding
a soft-decision aided Superdense Code (SD)
Abstract, page 1
quantum depolarizing channel
entanglement-assisted classical communication over the quantum depolarizing channel
Abstract, page 1
entanglement assisted capacity
ENTANGLEMENT-ASSISTED classical capacity [1] sets the ultimate capacity limit on the reliable transmission of classical information over a noisy quantum channel
Section I, page 1
irregular convolutional codes
we have employed an IRCC relying on a set of 17 memory-4 convolutional subcodes
Section IV-B, page 4
exit charts
We have exploited the area property of EXIT charts [17] for designing a near-capacity classical error correction code
Section IV-B, page 3
iterative decoding
iterative decoding is invoked for exchanging extrinsic information between the inner (URC-SD) and outer (IRCC) decoders
Section III, page 3