Make your paper interactiveCreate

Text from the interactive explanation

Quantum Information Processing, Sensing and Communications: Their Myths, Realities and Futures

Lajos Hanzo, Zunaira Babar, Zhenyu Cai, Daryus Chandra, Ivan B. Djordjević, Bálint Koczor, Soon Xin Ng, Mohsen Razavi, Osvaldo Simeone

Original paper

Five quantum myths, tested against reality

The Quantum Internet, as the authors picture it Illustrative model

Picture a network of quantum computers, repeaters and satellites where any eavesdropper is caught in the act — something impossible on today’s internet. The authors expect a Quantum Internet to offer both ultimate information security and eavesdropping detection, but say that “quantum-leaps” are required in quantum engineering to make it real.

Nine researchers took stock of what is real and what is still myth.

Can’t copy

No machine can duplicate an unknown qubit (the “no-cloning” rule), so there are no backup copies.

Can’t peek

Looking at a qubit forces its blend into a plain 0 or 1, and the blend is gone for good.

Can’t boost

An amplifier would have to measure the signal, which destroys it, so faint quantum light can’t be boosted on its way.

Every myth below is a battle with these three rules.

Before the myths

The fragile qubit

A qubit is the quantum version of a bit. It can be a single atom, an electron or a particle of light. Three short stations show what makes it powerful — and why it is so easily ruined.

A. A coin spinning in a closed box Illustrative model

The authors compare a qubit to a coin spinning inside a closed box: while it spins it is neither heads nor tails but a superposition — a blend of both. Lifting the lid (measuring) forces a plain 0 or 1, with odds set by the blend. Once you have looked, the blend is gone.

Each lift is one measurement of a freshly prepared coin. Over many lifts the share of 1s approaches the chance you set — that’s the only way to “see” a blend.

B. Looking changes the answer Simulation

Send single photons, one at a time, into a half-silvered mirror (a 50:50 splitter). Unwatched, each photon travels both arms like a wave and the two halves recombine so that every photon lands on detector A. Put an eye on one arm, and the photon is forced to pick a path: now A and B click equally often. This is the paper’s interferometer example (Fig. 31).

Counts reset whenever you start or stop peeking. Without the eye, interference sends every photon to A; with it, the “which-path” knowledge destroys the interference.

C. The decoherence chamber Illustrative model · Eq. 2

Decoherence is the slow leak of a qubit’s quantum character into its surroundings. The paper models it as random errors: a bit-flip (X) swaps 0 and 1, a phase-flip (Z) scrambles the hidden rhythm between them, and Y does both. Two hardware clocks set the pace: T1 , how long until an excited qubit relaxes, and T2 , how long the blend keeps its rhythm. Both depend on the hardware — superconducting circuits, trapped ions or quantum dots.

px = py = (1 − e−t/T1)/4 · pz = (1 + e−t/T1 − 2e−t/T2)/4. Time units are illustrative; each cycle runs to 3×T1 with fresh random draws.

T2 is kept between 0.2× and 2× T1, as physics requires.

Myth 1 · Quantum error-correction codes

Can codes beat decoherence?

The myth

Make a guess, then experiment.

Step 1 · The classical trick is to copy. Quantum can’t.

0 → 000 1 → 111

If noise flips one bit ( 010 ), a majority vote still reads 0.

α|0⟩ + β|1⟩ → α| 000 ⟩ + β| 111 ⟩

No copying allowed, so the blend is spread over three qubits using entanglement. You still can’t look at them — so how do you find an error?

Step 2 · Syndrome detective Simulation

The trick is to ask only comparison questions: “do qubits 1 and 2 agree?” and “do 2 and 3 agree?” These parity checks — together called the syndrome — reveal where a flip happened without ever revealing the stored blend. Let noise strike, read the two lamps, and fix the right qubit.

A lit lamp means that pair disagrees . Which qubit was hit?

Step 3 · The speed limit: the hashing bound Eq. 7 Fig. 4 data, approximate

Every code spends extra qubits on protection. The coding rate is the share of each block that carries real quantum information. For noise that randomly hits qubits with X, Y or Z errors with total probability p (a “depolarising” channel), the hashing bound CQ(p) = 1 − H2(p) − p·log23 is the best rate that random codes can reach — but only with infinitely long code words. (H2 is the binary entropy, a measure of randomness.)

Code positions read approximately from the paper’s Fig. 4 (word error rate 10−3). Dot area grows with log10 of the code-word length n. Hover or tap a dot for details.

The codes that get close to the curve need thousands to hundreds of thousands of qubits per code word. Today’s devices have “a hundred plus” qubits, so only short, simple codes — such as topological (surface or toric) codes — can run now.

Stabilizer codes borrow the machinery of classical codes. Each error I, X, Z, Y on a qubit is written as a pair of bits (00, 10, 01, 11; Eq. 3), so syndrome checks become ordinary parity checks.

CSS codes are built from two classical codes: one catches bit-flips, the other phase-flips.

Entanglement-assisted codes use entangled pairs shared in advance between sender and receiver. They raise the bound, opening the paper’s “hashing region” of Fig. 5.

Quantum degeneracy: different errors can need the same fix, a uniquely quantum effect that can push the true capacity above the hashing bound.

A race against the clock: encode, measure the checks and decode, all before the qubits decohere.

Milestones , in the order of the paper’s timeline: Shor’s 9-qubit code (1995) → CSS codes and Steane’s 7-qubit code → the optimum 5-qubit code → Gottesman’s stabilizer framework → Lloyd’s hashing bound → Kitaev’s toric codes → surface codes → entanglement-assisted codes → quantum turbo codes → hypergraph- and lifted-product codes → GRAND universal decoders.

Myth 2 · Error mitigation

Repairing the average, not the run

The myth

Make a guess, then experiment.

Today’s machines are too small and noisy to run full error correction. Error mitigation is the clever stop-gap: it doesn’t fix any single run. Instead you run the same circuit many times — each run is a shot that ends in a measurement of +1 or −1 — and repair the average afterwards.

The paper’s Fig. 9 makes the trade-off clear: noise shifts the average away from the truth (a bias ); mitigation removes the bias but widens the spread, so you need more shots to pin the answer down.

Zero-noise extrapolation lab Illustrative model

You can’t turn the noise down, but you can turn it up — for example by stretching the control pulses so every gate is exposed to noise for longer. Measure at 1×, 2× and 3× the natural noise, fit a curve through the points, and follow it back to zero noise. In this model the ideal answer is ⟨Z⟩ = 0.80 and noise λ shrinks it to 0.80·e−λ.

Error bars show ±σ/√N for N shots. The lab runs a demo on its own until you press one of its buttons.

Fig. 9 replay: bias versus spread Simulation

One experiment gives one estimate. Repeat the whole experiment many times and look at the spread of answers, using the settings above.

The price tag Illustrative model

Why not mitigate everything? Add up the error rates of all the gates and you get the circuit fault rate λ = N·p (N gates, each failing with probability p). Errors arrive like raindrops (a Poisson process), so only a fraction e−λ of runs is error-free — and mitigation needs roughly eλ times more runs to make up for it.

Practical when λ is about 1 or lower. Because the cost grows exponentially with circuit size, mitigation — unlike error correction — does not scale to arbitrarily large computations.

Probabilistic error cancellation: randomly insert extra gates and flip the sign of some results so that, on average, the noise cancels. For a simple bit-flip error with probability p the weights are γ1 = (1−p)/(1−2p) and γ2 = −p/(1−2p) (Eq. 8).

Purification with copies: compare several copies of the noisy state; like a high-pass filter it keeps the dominant, error-free part and suppresses the error components as (pk)n for n copies.

Symmetry verification: if the right answer must have, say, even parity, throw away the runs that come out odd.

Subspace expansion: improve a rough answer by combining it with a few slightly altered versions.

Learning-based mitigation: train a correction α0 + α1·E on circuits that a classical computer can check, then apply it to the real one.

Myth 3 · Quantum machine learning

Does quantum supercharge machine learning?

The myth

Make a guess, then experiment.

A parameterised quantum circuit is a row of quantum gates with adjustable rotation knobs θ — a “quantum neural network”. A classical computer turns the knobs to make the circuit’s output match some data. One promising use is a Born machine : it uses the built-in randomness of measurement to generate new samples that look like the data.

Train a two-qubit Born machine Simulation

The circuit below has three knobs. Its four possible outputs (00, 01, 10, 11) should come out with the target odds shown as outlines: mostly 00 and 11. But you never see the true odds — only counts from a limited number of shots , so the bars jitter. That jitter is honest shot noise.

Training uses the parameter-shift rule : run the circuit with each knob nudged by ±π/2 and compare. There is no backpropagation inside a quantum circuit, so the cost grows with the square of the number of knobs. Pause stops the shot-noise resampling.

Barren plateaus: when the landscape goes flat Illustrative model

Training means walking downhill on a “loss landscape”. The paper explains that for generic circuits the landscape’s variation shrinks exponentially as qubits are added. The slope then hides beneath shot noise, and telling uphill from downhill needs exponentially many measurements.

Wiggle size here scales as 2−n/2; the grey band is the ±1/√1000 noise of a 1,000-shot estimate.

Four ways to mix quantum and classical

Myth 4 · Quantum radar

Spotting a faint echo in the glare

The myth

Make a guess, then experiment.

The idea is called quantum illumination . An entangled source makes twin photons. The signal twin flies out towards a possible target; the idler twin stays at home in a memory. Echoes come back very faint and buried in a glare of background photons — but the receiver can check each arrival against its stored twin. Background photons have no twin, so they are discounted.

Error probabilities use the paper’s bounds (Eqs. 16–17) with transmissivity × signal brightness fixed at 0.01 Illustrative model

Reality check: how big is the advantage?

Bar scale: 0–6 dB. More hurdles: no long-lived quantum memory exists yet (optical delay lines stand in for it); optical light struggles through clouds and fog; microwave versions need better entangled sources, detectors and converters.

Myth 5 · Quantum key distribution

Keys guarded by physics

The myth

Make a guess, then experiment.

Two ways to protect secrets from tomorrow’s quantum computers. Post-quantum cryptography (PQC) builds new locks from maths problems believed to be hard even for quantum machines. Quantum key distribution (QKD) instead lets two people, Alice and Bob, grow a shared secret key from single photons — and any eavesdropper (Eve) who touches the photons leaves fingerprints, because of the “can’t peek” rule.

Why care? An attacker can record encrypted traffic today and wait for a machine that cracks it later: “harvest now, decrypt later” . Some data, such as medical records, may have to stay private for a whole lifetime and beyond.

The BB84 lab: catch Eve in the act Simulation

Alice sends each photon with a random bit, polarised in a random “basis”: + (↑ = 0, → = 1) or × (↗ = 0, ↘ = 1). Bob measures in his own random basis. Where their bases match he reads her bit exactly; where they don’t, his result is a coin toss and the bit is thrown away (“sifting”). Eve can’t know Alice’s basis, so when she measures and resends she guesses wrong half the time — and that shows up as errors.

Raw key: Alice picks random bits.

Bases: she encodes each in a random basis; Bob measures in his own random basis.

Sifting: over an ordinary channel they compare bases (not bits) and keep only matches.

Error estimation: they publicly compare some “check bits”. Too many errors means Eve — abort.

Reconciliation: the remaining small errors are corrected.

Privacy amplification: the key is shortened so whatever Eve learned becomes useless.

Distance and the no-boost rule Simulation · 0.2 dB/km

Glass fibre loses light: about 0.2 dB per kilometre , so after 100 km only 1% of the photons are left. Ordinary internet links fix this with amplifiers. A quantum link can’t — an amplifier would have to measure the photons, returning them to the classical world. Today’s workaround is a chain of trusted relays : secure stations that receive the key, then pass it on with fresh photons. Each relay must be guarded, because it holds the key.

Back to the network

The long road to a Quantum Internet

Remember the glowing globe at the top? The authors sketch how quantum-secured networks could grow in three phases, each removing a little more of the trust that today’s relays demand. All timelines are the paper’s own speculative estimates, in years from now.

Same globe, same cities. Illustrative model

Commercial quantum repeaters are envisaged in “15+ years”. The earlier labs matter here: error mitigation and error-correction codes are expected to speed this up by stopping errors from propagating from hop to hop along a repeater chain.

Your scorecard

Beyond: quantum ideas for next-generation wireless

Handle with care — but keep building

Real now: QKD products you can buy, trusted-node networks spanning thousands of kilometres, and error mitigation running on machines with 100+ qubits.

Next: untrusted relays, satellites linking ground networks, and longer, error-corrected repeater chains — each step fighting the same three rules: can’t copy, can’t peek, can’t boost.

Source & credits

Lajos Hanzo, Zunaira Babar, Zhenyu Cai, Daryus Chandra, Ivan B. Djordjević, Bálint Koczor, Soon Xin Ng, Mohsen Razavi and Osvaldo Simeone, “Quantum Information Processing, Sensing, and Communications: Their Myths, Realities, and Futures.”

Read from arXiv:2412.00987v1 (submitted version, 1 Dec 2024), registry-linked to the published Proceedings of the IEEE article: arxiv.org/abs/2412.00987 · doi.org/10.1109/JPROC.2024.3510394

All visuals are original code-drawn reconstructions; Fig. 4 code positions are read approximately; simulations are illustrative models built from the paper’s equations (Eqs. 2, 7, 15–17) and statements. The globe’s land outline uses public-domain Natural Earth data. arXiv version licensed CC BY 4.0.

Topics in this iPaper

Physics
Quantum radar is a particular quantum sensing technique that exploits quantum-mechanical features of the electromagnetic fields

Section V-A (Quantum Radars, State-of-the-Art), page 23

Computer science
quantum machine learning (QML) is emerging as a promising paradigm to program gate-based quantum computers

Section IV (Quantum Machine Learning), abstract, page 17

Engineering
it is becoming realistic to incorporate QKD-based key negotiation in next-generation wireless systems

Section I-A2 (Quantum Key Distribution), page 3

quantum error correction
Quantum error-correction codes (QECCs) can be constructed from the known classical coding paradigm

Section II (Mitigation of Decoherence by Quantum Coding), abstract, page 4

quantum error mitigation
we have to resort to quantum error mitigation (QEM) techniques

Section III (Mitigation of Decoherence Without Coding), abstract, page 11

quantum machine learning
quantum machine learning (QML) is emerging as a promising paradigm to program gate-based quantum computers

Section IV, abstract, page 17

quantum illumination
The quantum illumination technique proposed in [245] is capable of improving the sensitivity, when operating in a hostile noisy regime.

Section V-A2 (Quantum Illumination Based Radars), page 24

quantum key distribution
The key objective of a QKD protocol is to exchange, in a secure way, a secret cryptographic key between a pair of authenticated network users

Section VI (Quantum Key Distribution Networks), abstract, page 30

quantum internet
Their broad vision is to make the creation of the Quantum Internet (Qinternet) [16] a reality

Section I-A3 (A Vision of the Quantum Internet), page 3

Quantum Information Processing, Sensing and Communications: Their Myths, Realities and Futures — Interactive explanation | iPaper.ai