Text from the interactive explanation
Space-Time Coding-Assisted Double-Spread Rake Receiver-Based CDMA for Dispersive Rayleigh Fading Environments
S. X. Ng, L.-L. Yang, T. H. Liew, L. Hanzo
Original paperInteractive paper · IEEE VTC Spring 2002
Your phone hears every message several times over — bounced off buildings, arriving late, fading in and out, tangled with everyone else's. In 2002 a Southampton team showed that a simple receiver could untangle it, needing roughly 3,000× less energy per bit in their simulations.
Live simplified simulation at Eb/N0 = 12 dB, 4 phones sharing the channel, no error-correcting code.
In the paper's own simulations, adding turbo coding to the scheme cut errors to 10 per million (BER 10−5) at only ≈4.3 dB in 5-path Channel A.
How to read this: each glowing square is one chip of a spreading code. Longer paths arrive later. At the phone every echo lands on its own Rake finger; the merged result becomes a dot in the constellation, and a dot inside its ring means the two bits were read correctly.
01 Multipath & fading
Echoes that fade
In a city there is rarely a clear line to the tower. What reaches your phone is a crowd of reflections — and they don't arrive in step.
Think of each reflected wave as a little arrow: its length is the wave's strength, its angle is where the wave is in its cycle when it arrives. Your phone feels the sum , the arrows laid tip to tail. Move a few centimetres and every arrow turns by a different amount. Sometimes they line up and the signal is strong; sometimes they cancel and it plunges. This random rise and fall from many reflections adding in and out of step is called Rayleigh fading .
Suppose your phone could add up 4 copies of the signal that fade independently. How often would the combined strength fall below the red dropout line?
This is diversity : copies that fade independently rarely all fade at once. The paper collects such copies from two places — the two tower antennas and the L echoes — giving 2L branches . With 5 echoes, that's 10.
The paper tested two kinds of city channel, with echoes spaced one chip apart: Channel A , where all echoes are equally strong, and Channel B , where each later echo is weaker.
Outlines: average echo strength read from Fig. 2 of the paper (normalised so total power = 1; approximate). Glowing fill: one possible moment of fading (illustrative). Animates with the fading scope above.
02 Codes & crosstalk
Sharing the air with codes
The tower talks to everyone at once, on the same frequency. To keep the conversations apart, each phone gets its own pattern of + and − pulses, called a code.
Here four phones share the air. Each sends one bit (+1 or −1) multiplied by its own 4-chip Walsh code . The tower adds everything up and transmits the sum. Your phone multiplies what it hears by its own code, chip by chip, and adds: your bit comes back four times larger and every other phone's contribution is exactly zero .
The echo arrives at 0.6 strength. In this toy, its first chips are empty and its tail runs into the next bit, which we leave out.
Predict: once an echo arrives late, will the crosstalk still cancel to zero?
Do: slide the echo delay to 1, 2 and 3 chips, and flip a few bits.
Notice: other phones now leak into yours (red bars). The codes still exist, but shifted copies no longer cancel. This leak is multiple-access interference .
That is the problem in one picture. Walsh codes are perfect only when everything arrives in step. Random codes behave the other way round: they barely resemble a shifted copy of themselves, which is great for separating echoes, but different users' random codes always overlap a little, so phones interfere. The heavyweight cure is multi-user detection (MUD): the receiver works out everyone's signal to subtract it, at a high computing cost.
Walsh codes of length Q are orthogonal (Eq. 1):
The paper's Eq. 2 concerns autocorrelation, how well a code matches a shifted copy of itself, Ri(τ) = Σl ci[l] · ci[l+τ]. For Walsh codes this stays high at non-zero shifts, so a delayed copy resembles other codes and orthogonality is lost in a multipath channel.
03 The central idea
The double-spreading trick
Keep a short Walsh code to tell phones apart, but wrap every one of its chips in the same short random pattern.
Conventional spreading uses a 32-chip Walsh code, scrambled by a random code at the same chip rate: every chip gets its own random sign. Double spreading uses a 4-chip Walsh code (Q = 4) and spreads each Walsh chip with the same 8-chip random code (Q = 8). That is 4 × 8 = 32 chips per symbol, the same length as before.
Why it helps: an echo a few chips late mostly stays inside the same 8-chip block. Because every block carries the identical random pattern and the channel barely changes within a symbol, the echo smears every Walsh chip the same way : one common interference factor I (Eqs. 3–4). A factor common to all four Walsh chips simply scales them, so other phones still cancel (Eq. 5). Only the first few chips of each block, which spill over from the block before, break the pattern.
power leaking in from other phones, relative to your signal
—
Meters: exact crosstalk power averaged over 300 random codes and random data, for a direct path plus one echo at 0.6 strength, with the receiver locked onto the direct path. The looms show one example code.
Predict: as the echo gets later (1 → 4 chips), which meter suffers more?
Do: move the echo delay, then the number of phones; press “New random code”.
Notice: conventional crosstalk is high at once. Double spreading starts about 10 dB lower and creeps up as more chips spill between blocks — the paper's point that a longer delay spread, relative to the 8-chip code, means more interference.
Walsh orthogonality (Eq. 1) is what keeps users apart:
Walsh codes have high autocorrelation at non-zero shifts (Eq. 2), which is why a plain echo destroys orthogonality. With double spreading, every Walsh chip l of user k arrives multiplied by the same multipath interference factor I (Eqs. 3–4). When the receiver despreads with user i's Walsh code, that common factor comes out of the sum (Eq. 5):
Total spreading factor: Q = Q × Q = 4 × 8 = 32.
04 Diversity
Catching every echo, from two antennas
Double spreading stops other phones from leaking in. Two more ideas turn the echoes — and a second tower antenna — from a nuisance into extra strength.
One finger per echo
A Rake receiver has one "finger" per echo. Each finger slides the 8-chip random code to line up with one echo's delay. A random code matches itself strongly only when perfectly aligned — a sharp spike — so each finger picks out its own echo and largely ignores the rest. A Walsh code can't do this: shifted, it still looks a lot like itself.
—
Two antennas, two time slots
With two antennas the tower uses Alamouti's G2 space-time block code . It sends two symbols over two time slots: in slot 1, antenna 1 sends x1 while antenna 2 sends x2; in slot 2, antenna 1 sends −x2* and antenna 2 sends x1* (the star flips the phase). In this scheme one slot is one Walsh chip — 8 random chips — and the fading is assumed constant across both slots. Combining the two slots cancels the cross-talk between x1 and x2 and leaves each multiplied by |h1|² + |h2|². One antenna can fade away completely; the other still carries the message.
Both set-ups transmit the same total power: with two antennas each sends half, hence the factor ½. Dots: received symbols after combining, drawn relative to the noise level; red = wrong decision. Keyboard: use the sliders, or focus the plane and use the arrow keys to rotate h1.
Predict: antenna 1 is in a deep fade (h1 near zero). Will the message survive?
Do: drag h1 right into the centre, then switch between 1 and 2 antennas.
Notice: with two antennas the dots stay near their rings, carried by h2; with one antenna they collapse into the noise.
The paper puts both ideas together: every Rake finger runs its own space-time decoder , and the fingers' outputs are added. The result is multiplied by Σi(|α1i|² + |α2i|²): two antennas × L echoes = diversity order 2L (Eq. 10). On top, a turbo convolutional code (rate ½, constraint length 3, 8 decoding iterations, 1000-bit interleaver) adds time diversity and error correction. The live simulations on this page leave turbo coding out.
G2 transmission over two slots (Eq. 6); in each slot, (antenna 1, antenna 2) send:
Received in the two slots, with gains h1, h2 constant over both, and the receiver's combining step (Eqs. 7–9, written here in the standard Alamouti form):
With an STBC decoder in each of the L Rake fingers, summed over fingers (Eq. 10):
05 Your turn
The virtual lab
Run the whole downlink yourself: a chip-by-chip simulation that spreads, fades, echoes and adds noise to every symbol, then decodes every phone with a Rake receiver — as fast as your device allows.
Simplified simulation running in your browser: uncoded QPSK, perfect channel knowledge as in the paper, no turbo coding. Errors are counted over every phone sharing the channel; the constellation shows phone 1. Expect the paper's trends; exact numbers differ, since this is a simplified re-creation, not the authors' simulator. In particular, with many echoes the chips that spill between neighbouring 8-chip blocks leave a little residual crosstalk here, so the double-spread curves level off sooner than the paper's.
06 Findings
What the paper found
The headline figure puts every ingredient side by side: how much signal energy each one saves at a fixed error rate of 1 wrong bit in 100,000.
—
Gain relative to an uncoded, single-antenna narrowband benchmark that needs Eb/N0 = 39.3 dB to reach BER 10−5; 4 users. Values read from the published plot (approximate); hover or tap points for values.
To reach 1 wrong bit in 100,000, the basic benchmark link needs Eb/N0 = 39.3 dB : every bit must carry about 8,500 times the noise energy. Each amber square here is one unit of noise energy.
The turbo-coded, two-antenna scheme in 5-path Channel A gains 35.05 dB , so it needs only about 39.3 − 35.05 ≈ 4.3 dB : about 2.7 squares . A 35 dB gain is a factor of about 3,200 — roughly 3,000 times less energy per bit , in the paper's simulations.
Fig. 3 · More phones, no penalty
Two equal echoes, two antennas. Conventional spreading gets worse as phones are added — the echoes destroy orthogonality. Double spreading shows no degradation all the way to Q = 4 phones.
Redrawn from Fig. 3 of the paper; values approximate. Simulated in the paper.
Fig. 4 · Echoes become allies
Four phones, Channel A. With conventional spreading, more echoes mean more interference and higher error floors. With double spreading, more echoes mean more diversity — the curves improve, then saturate.
Redrawn from Fig. 4 of the paper; values approximate. Simulated in the paper.
07 The trade-off
The price — and what it buys
Nothing is free. Double spreading pays in capacity: a 32-chip symbol carries up to Q = 4 phones instead of 32, a factor Q = 8 fewer. In exchange, a simple Rake receiver delivers near-single-user quality with no multi-user detection.
— Ng, Yang, Liew & Hanzo, conclusion of the paper
Conventional 32-chip up to 32 phones
More phones per symbol, but in a city full of echoes they leak into each other, so the receiver needs complex multi-user detection to stay clean.
Double-spread 4 × 8 up to 4 phones
Fewer phones per symbol, but the echoes stop being a source of interference and become extra copies a simple Rake receiver can add up.
Scroll back up and flip the switch again — now you know what each glowing piece is doing: the chips, the echoes, the two antennas, the Rake fingers and the constellation.
The paper
S. X. Ng, L.-L. Yang, T. H. Liew and L. Hanzo, “Space-Time Coding-Assisted Double-Spread Rake Receiver-Based CDMA for Dispersive Rayleigh Fading Environments,” IEEE 55th Vehicular Technology Conference (VTC Spring 2002), Birmingham, AL, USA, pp. 1670–1674, 2002.
Published version: doi.org/10.1109/VTC.2002.1002904
This page is based on the authors' copy of the conference paper, hosted by the University of Southampton: southampton.ac.uk/~sxn/paper/09-vtc02s.pdf
All charts are redrawn from the paper's figures (values approximate). The live simulations are simplified re-creations, not the authors' code. Visual geometry of the city is illustrative. The paper's figures are © IEEE and are not reproduced here.
Topics in this iPaper
Engineering
we propose a novel Space-Time Coding-Assisted Double-Spread Rake Receiver (STC-DS-RR) based CDMA scheme for downlink (base to mobile) transmissions over dispersive Rayleigh fading channels
Abstract, page 1
Computer science
soft decision trellis decoding utilising the binary Log-Maximum A Posteriori (Log-MAP) algorithm [9] was invoked for turbo decoding
Section II-B, page 3
code division multiple access
Double-Spread Rake Receiver-Based CDMA for Dispersive Rayleigh Fading Environments
Title and abstract, page 1
space time block coding
Here Alamouti's G2 code associated with two transmit antennas is used for STBC.
Section II, page 1
rake receiver
The main idea behind Rake receivers [7] is that of capturing the energy of several multi-path components
Section II-B, page 3
rayleigh fading
Each path was faded according to independent Rayleigh fading statistics
Section II-C, page 4
turbo codes
The STC-DS-RR scheme invokes Turbo Convolutional (TC) channel coding and Space-Time Block Coding (STBC)
Abstract, page 1
walsh codes
Walsh codes are well-known for their attractive zero cross-correlation property
Section II-A, page 2