Make your paper interactiveCreate

Text from the interactive explanation

Two Decades of MIMO Design Tradeoffs and Reduced-Complexity MIMO Detection in Near-Capacity Systems

Chao Xu, Shinya Sugiura, Soon Xin Ng, Peichang Zhang, Li Wang, Lajos Hanzo

Original paper

Two decades of multi-antenna radio have traded speed against safety, and accuracy against computing effort. This interactive tour lets you play the receiver, from a single noisy signal to Spatial Modulation and receivers that check fewer points.

Millimetre-wave radio makes it practical to put many antennas on one mast, and the paper looks ahead to systems with “tens to hundreds of transmit/receive antennas” . Every extra antenna can carry more data, but for the best possible receiver the number of combinations to weigh grows exponentially with the number of transmitting antennas. How can a receiver listen to many antennas at once without checking every possibility?

Simulated in your browserSimulated receiver view (one receive antenna, Rayleigh fading, QPSK). Dots are every possible noiseless received signal; the white dot is what actually arrived. The ring is the receiver’s pick (green = right, red = wrong), made using all of its antennas.

Controls

The scope draws every candidate it can; above a few thousand it shows a haze.

Everyone talks · V-BLAST

One talks · Spatial Modulation

Everyone-talks carries more bits per send — but its search grows exponentially. That tension is the story.

01A message is a point

02The air twists signals

03Two voices at once

04Three ways to untangle

05Fast or safe?

06Let the antenna speak

07Near the limit

08Check fewer points

09No single winner

→The hundred-antenna tower

A message is a point

Digital radio sends bits by picking one of a few agreed points on a flat map of “signal shapes”, called a constellation. Two bits need 4 points (QPSK), three bits need 8 (8PSK), four bits need 16 (16QAM).

Noise nudges every signal on its way, so it lands near — not on — the point that was sent. The receiver simply picks the nearest point (each point owns a tinted “decision region”). Points are labelled so that neighbours differ in just one bit (a Gray label), so a near-miss usually costs one bit, not all of them.

Hollow ring = the point that was sent. Filled ring = the receiver’s decision (green right, red wrong). Fading dots are recent landings.

What you just saw More points = more bits per send, but less room for noise.

The air twists every signal

Radio waves bounce off buildings and arrive by many paths at once. The result: each antenna-to-antenna link multiplies the signal by a random complex number h — it rotates and scales the whole constellation. When |h| gets small (a deep fade) the points collapse toward the centre and errors spike.

The receiver estimates h (“channel knowledge”) and undoes the twist. A second receive antenna helps a lot: the chance that both links fade at the same moment is much smaller, so combining them (maximum-ratio combining) slashes errors. That is diversity.

QPSK at Eb/N0 = 10 dB. Dots turn red when the decision was wrong.

What you just saw Diversity = several independent looks at the same message → reliability.

Two voices at once

Now flip it around: give the transmitter two antennas and let each send its own QPSK symbol at the same time, on the same frequency. That doubles the data — V-BLAST(2,2)-QPSK carries 4 bits per send. This is multiplexing.

The catch: each receive antenna hears the sum h₁·s₁ + h₂·s₂ plus noise. The voices overlap (inter-antenna interference). Your turn: you are the receiver. Guess what each antenna sent so that your predicted points (rings) land on the white dots on both scopes.

Attempts: 0

Look closely: on one scope two candidates can sit almost on top of each other. The second receive antenna sees them differently and pulls them apart — that is why the receiver uses both.

Bars use a logarithmic scale: each step right is ×10. Doubling the bits per send squares the search.

What you just saw The optimal V-BLAST receiver must weigh MNT combinations — in the paper’s words, a complexity “that grows exponentially with NT”.

Three ways to untangle

Check everything — the maximum-likelihood (ML) receiver tests every combination. Best accuracy, exponential cost.

Search smart — a sphere decoder builds the guess one antenna at a time, like walking down a tree, and abandons any branch whose partial distance already exceeds the best complete answer found so far (the “sphere radius”).

Undo the mixing — a linear filter (zero-forcing or MMSE) multiplies by a matrix to separate the streams, then decides each antenna on its own. Cheap, but leftover interference and amplified noise cost accuracy.

V-BLAST(2,2)-QPSK at Eb/N0 = 0 dB. Level v = 2 picks antenna 2’s symbol, level v = 1 picks antenna 1’s. Numbers are partial squared distances. Tree values from Fig. 17 of the paper.

Press run: each detector receives exactly the same noisy signals.

Work per send is a rough operation count: a tree node is one partial-distance calculation.

What you just saw The sphere decoder matches ML accuracy while visiting fewer tree nodes; MMSE is cheapest but makes clearly more errors at high signal-to-noise ratio.

Fast or safe? The first tradeoff

Alamouti’s code spends the antennas on safety. Over two time slots, antenna 1 sends s₁ then −s₂*, antenna 2 sends s₂ then s₁* (the star means “mirror image”). The pattern is orthogonal, so the receiver separates the two symbols with simple arithmetic — no interference — and each symbol enjoys two independent paths. The price: 2 symbols in 2 slots, no faster than one antenna.

V-BLAST spends them on speed: 2 new symbols every slot, but no transmit diversity. Zheng and Tse (2003) captured the tradeoff in one curve: with multiplexing gain r, the best diversity is d = (NT − r)(NR − r). Linear dispersion codes (Heath and Paulraj, 2002) spread symbols across space and time and can reach both full capacity and full diversity — but their receiver must untangle interference again.

What you just saw The first era of MIMO was multiplexing versus diversity. Linear dispersion codes resolve it on paper — but their receiver must handle interference again.

Let the antenna speak

Spatial Modulation (Song et al. 2004; Mesleh et al. 2008) lights only one transmit antenna per send. The bits are split: log₂NT bits choose which antenna, log₂M bits choose the symbol it sends. The transmitter needs just one radio chain.

With a single voice there is no interference, so the receiver can find the antenna and the symbol almost separately — a “reduced-scope” detector with optimal accuracy at much lower cost. The price: the antenna index carries only log₂NT bits, so SM’s capacity grows only logarithmically with antennas while V-BLAST’s grows linearly. (At 30 dB, Fig. 33a shows roughly 17.5 bits per channel use for V-BLAST(2,2) versus about 11.5 for SM(2,2) — approximate values read from the figure.) Space-Time Shift Keying (STSK, Sugiura et al. 2010) applies the same trick to “dispersion matrices” that spread one symbol over antennas and time, regaining diversity.

Key result In the paper’s simulations, SM matches the best V-BLAST receiver almost exactly while costing about as little as the cheap filter — the performance gap between SM and ML V-BLAST is tiny compared with the loss from using MMSE.

With good channel coding SM also performs close to V-BLAST: in the authors’ simulations the IRCC-URC coded SM(2,2)-8PSK is only about 0.3 dB from coded V-BLAST(2,2)-QPSK (Fig. 42).

What you just saw Lighting one antenna at a time removes interference, so the receiver searches far less for nearly the same accuracy — at the cost of fewer bits per antenna added.

Getting close to the limit: a conversation

Shannon (1948) proved every channel has a maximum error-free data rate, its capacity. Since turbo codes (1993), receivers get close to it by letting two parts talk: the MIMO detector and the channel decoder pass each other soft beliefs about every bit — a number meaning “probably 1, fairly sure”, called a log-likelihood ratio.

Each round, both improve. An EXIT chart (ten Brink, 1999) plots how much information each part puts out given what it receives. The two curves form a tunnel, and decoding climbs a staircase through it to the top-right corner: error-free. Too much noise and the tunnel closes — the staircase gets stuck. Crucially, the detector runs again every round , so its cost is multiplied. That is why cheap soft detection matters.

What you just saw Near-capacity receivers re-run the detector many times, so every saving in the detector is multiplied.

Check fewer points

The paper’s recipe for cheaper soft detection (its Fig. 46) has four moves: (1) separate the streams so each constellation is visited alone; (2) attach the decoder’s prior beliefs to the right parts of the signal; (3) use the symmetry of Gray-labelled constellations to find the best candidate for each bit by visiting only a small subset; (4) finish the calculation by comparing terms already computed.

Move 3 is the neat one. For rotated 8PSK, only M/4 = 2 points — one quadrant — need checking: fold the received point into the first quadrant and every other distance follows by symmetry. For square 16QAM, the sign of each axis is settled separately and only the magnitudes need checking. Drag the white dot and compare.

Drag the white dot, or focus the scope and use the arrow keys.

Complexity reduction ratio of the reduced-complexity soft demodulator versus the conventional one, in multiplications per bit. Upper bounds: 50% for square MQAM, 75% for MPSK and star/cross MQAM.

What you just saw Same soft answer, a fraction of the points: symmetry lets the receiver skip most of the constellation.

No single winner

The paper ranks five families of design — V-BLAST, linear dispersion codes (LDC), Spatial Modulation (SM), Space-Time Shift Keying (STSK) and space-time block codes (STBC) — in six categories (its Fig. 45). Pick a category and watch the order change.

Design guidelines (Fig. 47). MIMO factors: multiplexing, diversity, complexity. Near-capacity factors: reduced-complexity algorithms; LLR accuracy (soft beliefs must be honest — over-confident beliefs mislead the turbo conversation); EXIT charts.

What you just saw Every design wins somewhere and loses somewhere. Choosing one means choosing what to give up.

Back to the hundred-antenna tower

Remember the question: how can a receiver listen to many antennas without checking every possibility? Count it out for a large base station.

V-BLAST · QPSK · exhaustive

Spatial Modulation · QPSK

Counts of candidate transmissions an exhaustive receiver would weigh. The paper also notes that in massive MIMO, simple linear MMSE detection may become near-optimum when the number of antennas grows very large.

What comes next, according to the paper. Generalised SM lights several antennas at once — choosing nt of NT antennas gives many more patterns, so more bits — while sending the same symbol on all of them, so there is still no interference (though its capacity gain stays limited). Differential MIMO skips channel estimation altogether, which matters when there are too many links to measure. And the reduced-complexity recipe of chapter 08 still has to be carried over to these designs.

The answer. Receivers can get near the limit without checking every possibility — by choosing signal designs, and detectors, whose structure lets them search less. But every choice gives something up.

Source & credits

Chao Xu, Shinya Sugiura, Soon Xin Ng, Peichang Zhang, Li Wang, Lajos Hanzo, “Two Decades of MIMO Design Tradeoffs and Reduced-Complexity MIMO Detection in Near-Capacity Systems”, IEEE Access, 2017. DOI: 10.1109/ACCESS.2017.2707182.

Read from the accepted manuscript hosted by the University of Southampton (PDF). Published version: doi.org/10.1109/ACCESS.2017.2707182.

The paper is published under CC BY 3.0. Figures 17, 19, 34, 44 and 45 are shown as cropped excerpts; interactive charts redraw values read approximately from the paper’s figures; simulations on this page are simplified re-creations run in your browser. Results in the paper are simulations and complexity counts, not field measurements.

Fonts: Space Grotesk and IBM Plex Mono, both under the SIL Open Font License. Made with iPaper.ai.

Topics in this iPaper

Engineering
Fig. 1 offers a glimpse of a few key factors that directly affect the design of wireless communications systems.

Section I Introduction, page 1

Computer science
the reduced-complexity soft-decision MIMO detection algorithms may be carried out in two stages

Section IV-B (Design Guidelines, item 6), page 56

multiple input multiple output
Multiple-Input Multiple Output (MIMO) techniques have been one of the most vibrant areas in communications

Section I-B, page 7

spatial modulation
the following BPST = log2 NT source information bits are assigned to activate a single TA out of a total of NT TAs.

Section III-A, page 36

space time block coding
Alamouti’s G2 STBC is the only full unity-rate code

Section I-B, page 9-10

sphere decoding
Examples of the SD tree-search strategies

Section II-A3 and Fig. 17 caption, page 19

turbo detection
the terminologies of iterative decoding and turbo detection are used interchangeably

Section I-A, page 5

extrinsic information transfer charts
ten Brink [96] proposed the powerful tools of EXtrinsic Information Transfer (EXIT) charts in 1999, which visualized the convergence of turbo detection.

Section I-A, page 6

wireless communications
wireless communications systems have gradually become more and more complex.

Section I Introduction, page 1

Two Decades of MIMO Design Tradeoffs and Reduced-Complexity MIMO Detection in Near-Capacity Systems — Interactive explanation | iPaper.ai