Text from the interactive explanation
Cooperative Overlay Spectrum Access in Cognitive Radio Networks
Wei Liang, Soon Xin Ng, Lajos Hanzo
Original paperLiang, Ng & Hanzo · IEEE Communications Surveys & Tutorials, 2017
Licensed radio bands sit mostly silent, yet newcomers aren't allowed in. This survey maps a bargain in which an unlicensed radio relays a licensed user's signal and is paid in the airtime it frees.
Licensed sender & receiver
Helper (unlicensed radio)
Helper relaying the licensed message
Helper's own link
The bar along the bottom of the scene is one frame of the licensed user's airtime. Worked example from the paper: helper halfway, 9 dB signal, path-loss exponent 3. Other helper positions use our simplified model.
01 · The hook
Crowded licences, quiet airwaves
Almost every slice of the radio spectrum between 30 MHz and 3 GHz, where TV, mobile phones and emergency radio live, is licensed to an owner. The paper calls these owners primary users (PUs).
Owning a licence is not the same as using it. Before you look, make a guess.
Each column is a slice of frequency, grouped into licensed bands; time flows downward. Illustrative waterfall generated at the occupancy figures the paper cites.
A cognitive user (CU) is a smart unlicensed radio: it senses its surroundings and adapts how it transmits. Three rules can let it in; each is drawn on the waterfall above.
Pick a sharing rule to see where an unlicensed cognitive radio could fit in.
The measurement the paper cites found average occupancy of just 5.2 % across six cities, with New York the busiest at 13.1 % . Back in 2002 the US regulator, the FCC, had already found licensed spectrum largely unoccupied most of the time.
Meanwhile, newcomers have nowhere to go: a cognitive user may only borrow spectrum without getting in the owner's way.
The survey focuses on overlay . The authors describe it as a “win-win”: the owner gets help, and the helper gets airtime in return.
Why would helping make the owner faster?
02 · Signal & symbols lab
Why a helper makes the message faster
Radio signals fade fast with distance. In the paper's example the received power falls as 1/d³: halve the distance and the signal arrives 2³ = 8 times stronger.
Engineers measure signal quality as SNR , the signal-to-noise ratio, in decibels (dB). Every +3 dB roughly doubles the signal, so +9 dB is about 8×. A helper halfway along turns one long, weak link into two short, strong hops .
Distances are fractions of the sender–receiver gap; hop SNR = transmitter SNR + 30·log₁₀(1/d). The 9 dB (8PSK) and 18 dB (64QAM) thresholds come from the paper; thresholds marked * (6, 12, 15 dB) are evenly spaced placeholders, so off-midpoint results are illustrative. Two hops share the time, so the overall rate is 1 ÷ (1/h₁ + 1/h₂).
The coral bars show the sweet zone : the helper earns airtime only near the middle, where both hops are short. Near either end one hop stays long and weak.
Turn the transmitter up to 12 dB or more and the bars vanish: the direct link is already good, and a relay “would only double the transmission period”, as the authors put it when explaining their simulations.
Symbols: how many bits fit in one pulse
A radio sends a stream of symbols : short pulses, each one a point picked from a pattern of possible shapes. Choosing the pattern is called modulation . 8PSK has 8 points; 64QAM has 64.
Denser patterns pack more bits per symbol but put points closer together; noise blurs them; the radio picks the densest pattern its signal can support. Error-correcting coding cleans up occasional raw slips; the paper's thresholds mark where the cleaned-up error rate stays below 1 bit in 100,000.
The paper's adaptive scheme (turbo trellis coded modulation) spends one bit of every symbol on that protection, so 8PSK delivers 2 useful bits and 64QAM delivers 5.
Rings are the ideal points; dots are received symbols with noise set by each link's SNR (shared with the lab above). Pink dots landed nearer a wrong neighbour: raw slips before error correction. Near a threshold roughly one symbol in eight still slips, which the code repairs; force a denser pattern on a weak link and the clouds merge.
03 · The trade
Spend the gain: speed or spectrum?
With a helper, the licensed user moves 2.5 useful bits per symbol instead of 2, a 25 % gain. It can keep that gain as speed, or hand it over to the helper as freed time or bandwidth.
analytical example From the paper's worked example (Fig. 8, Table V): 9 dB transmit SNR, helper at the midpoint, path-loss exponent 3. Released share = 1 − 0.8 × (bit-rate multiplier).
Time: the owner broadcasts for βT₁, the helper relays for βT₂, and the rest of the frame, (1 − β)T, belongs to the helper. Bandwidth: owner and helper share a narrower band W₁, and the leftover band W₂ = W₀ − W₁ is released for the whole frame.
The authors pick System C : the licensed user keeps exactly its old bit rate and releases 20 % of its bandwidth. System D keeps a little of both: ×1.10 the bit rate and 12 % released.
04 · Two-way relaying
Two conversations, one helper
Often two licensed users want to swap messages, like the two ends of a call. Relaying each message separately costs two turns per message: in to the helper, then out.
Two licensed users want to swap messages through one helper. How few turns can it take?
A's message
B's message
helper forwarding
dashed slots: turns saved vs one-way
TDBC (time-division broadcast) lets A and B send in turn, then one broadcast from the helper delivers both messages. MABC (multiple-access broadcast) has A and B transmit at the same moment; the helper must untangle the two overlapping signals (the paper uses multi-user detection) before broadcasting.
Fewer turns, more spectrum to give away
simulation Simulated in the paper (flat Rayleigh fading, 10⁶ frames); curves redrawn approximately from Fig. 11. Faint dashed lines are idealised one-way schemes (CCMC, DCMC). Values at 5 dB are quoted in the text: 81 %, 88 % and 92 %.
At 5 dB, two-way relaying saves 7 points more bandwidth than one-way with TDBC and 11 points more with MABC. Savings fall as the signal strengthens because the direct link becomes good enough on its own. At any given SNR two-way beats one-way, and MABC stays above zero across the whole 0–30 dB range.
05 · The market
Who helps whom? The pairing game
Real networks have many licensed users and many helpers. Each pair strikes a deal over a time slot: the licensed user keeps a share β , the helper gets ξ = 1 − β .
Each would reach a rate C with the whole slot, so the licensed user gets CPU·β and the helper CCU·ξ. Both have minimum rates. The numbers below are the paper's worked example.
Should the licensed users compete for the best helper, or take turns?
Blue numbers: CPU, the licensed user's rate with the whole slot. Coral numbers: CCU, the helper's. Paired links carry a β | ξ time bar.
Why: both licensed users rank CU2 first. In the bidding war they keep cutting β by 0.1 to outbid each other, paying CU2 more and more airtime, until PU1 gives up and settles for CU1. Taking turns stops the war: a round-robin priority list lets one user pick first each round, so nobody overpays, and every helper still meets its minimum. A short, limited punishment discourages anyone from breaking the rota.
Edit the rates and both runs are recomputed. Try “Different favourites”: when the licensed users don't want the same helper there is no war, and the two approaches tie.
Does it hold up with more users?
simulation Simulated in the paper; curves redrawn approximately from Fig. 14 (values read from the plot). The upper bound is a centralised exhaustive search; the lower bound is random pairing.
Taking turns closely tracks the centralised upper bound and beats the bidding war at every size. The paper's further simulations (Fig. 16) show the advantage is largest when licensed users outnumber helpers, exactly when competition for helpers is fiercest.
06 · Evidence
What the survey found
Relaying with adaptive coded modulation
20 % released
A midpoint helper lifts the licensed link from 2 to 2.5 useful bits per symbol. System C keeps the bit rate and releases 20 % of the bandwidth; System D gains 10 % speed and releases 12 %.
Two-way relaying
81 → 88 → 92 %
Bandwidth saving at 5 dB for one-way, TDBC and MABC: +7 and +11 points over one-way. Two-way relaying beats one-way at every SNR tested.
Taking turns beats a bidding war
6.3 vs 4.95
Average licensed-user rate in the paper's two-by-two example. In simulation the turn-taking algorithm (PDA) tracks the centralised upper bound, especially when licensed users outnumber helpers. It is sub-optimal, and it reaches an equilibrium only when the game is repeated.
The wider landscape
The survey also reviews auctions where helpers bid relaying power for leased time, leader–follower (Stackelberg) games, coalitions of cooperating users, full-duplex relaying that sends and receives at once, and helpers that harvest energy from the licensed signal.
Mode switching (Eq. 2). The radio compares the received SNR γR with thresholds γ0 … γ4 and picks the densest mode that clears its threshold:
Frequency division (Eq. 5–7). The helper's own rate on the released band W₂ = W₀ − W₁, where ψ is the share of its power spent relaying:
Time division (Eq. 8–11). Licensed user l with helper k at time share βl,k:
Round-robin priority (Eq. 17–18). In round i the licensed transmitters pick helpers in the order
Bandwidth saving in the worked example: Bs = 1 − ηA/ηC = 1 − 2/2.5 = 0.2.
07 · Coda
Back to the quiet airwaves
owner transmits
helper relays the owner's message
helper's paid airtime
Instead of fencing spectrum off or sneaking into gaps, overlay sharing turns capacity into a trade both sides want: the owner's messages get through faster or more cheaply, and the helper is paid in airtime.
Much of the band is still quiet, so the deals sit alongside the other ways of sharing rather than replacing them. The authors' stated aim is to stimulate further research on spectrally efficient cognitive radio networks.
Sources & credits
Wei Liang, Soon Xin Ng and Lajos Hanzo, “Cooperative Overlay Spectrum Access in Cognitive Radio Networks,” IEEE Communications Surveys & Tutorials, vol. 19, no. 3, pp. 1924–1944, 2017. DOI: 10.1109/COMST.2017.2690866. Author-hosted copy: southampton.ac.uk/~sxn/paper/2017_survey.pdf.
All visuals on this page are original illustrations and simulations made for it; no figures from the paper are reproduced. The charts are approximate redraws of the paper's Figs. 11 and 14, read from the published plots. Worked-example numbers come from the paper's Figs. 8, 12 and 13 and Table V. The spectrum waterfall and the off-midpoint relay positions (including the 6, 12 and 15 dB mode thresholds) are illustrative models. Occupancy figures (5.2 % six-city average, 13.1 % in New York) are those cited by the paper from its reference [10].
Typeface: IBM Plex Sans and IBM Plex Mono, SIL Open Font License.
Topics in this iPaper
Engineering
the appealing concepts of cooperative communication techniques and cognitive radio (CR) networks have been combined
Abstract, page 1924
Computer science
provide a solution in the form of a distributed algorithm that can be shown to be convergent but sub-optimal
Section VI (Game Model for Overlay SAS in CCR Network), page 1935
cognitive radio
We mainly survey the overlay spectrum access scheme in this novel cooperative CR (CCR) network context
Abstract, page 1924
cooperative communication
Cooperative relaying is widely regarded as the key technology in CR networks
Section I (Introduction), page 1925
game theory
the PUs negotiate with CUs concerning the TS allocation by using game-theoretic techniques
Section VII (Conclusion), page 1940
matching theory
Matching theory [153], [154] has been widely used as an efficient technique of solving the combinatorial problem
Section VI-B (Matching Game Model), page 1936
adaptive coded modulation
a lower-throughput but more robust TTCM mode is used
Section IV (Adaptive Trellis Coded Modulation), page 1930
two way relaying
Liang et al. [127] have employed the TWR in the overlay SAS
Section V-D (TWR Aided Overlay SAS in CCR Network), page 1933