ITLinQ: A New Approach for Spectrum Sharing in Device-to-Device Networks

Transcription

1 ITLinQ: A New Approach for Spectrum Sharing in Device-to-Device Networks Salman Avestimehr In collaboration with Navid Naderializadeh ITA 2/10/14

2 D2D Communication Device-to-Device (D2D) communication is expected to play a key role in future wireless networks Will be the backbone for Ø Communication Ø Control and Sensing Ø Proximal Interaction

3 Key Bottleneck (at PHY) Interference control (or spectrum sharing) and coordination Main issues as the network becomes large and complex

4 How to Deal with the Interference? Fully-coordinated Cellular-type Hard to implement in Practice! Fully Distributed WiFi-type Degraded Performance for large number of users

5 Another Approach Rely on a small level of coordination for scheduling links Ø Use a conflict graph to describe when interference among a set of users is low enough Ø Schedule a set of non-conflicting users at each time Ø Treat the interference among them as noise Two key challenges Ø How to define the conflict graph? Ø How to design a scheduler with minimal coordination?

6 How to define the conflict graph? Two common approaches 1. Geometric : Comparing Interference-to-noise ratio (INR) to a fixed threshold If INR between two users is below threshold, they are considered non-conflicting (i.e., independent) Criteria: INR γ IS In a path-loss fading, this corresponds to a fixed distance-guard

7 How to define the conflict graph? Qualcomm s FlashLinQ : Comparing Signal-to-interference ratio (SIR) to a fixed threshold If SIR between two users is above threshold, they are considered non-conflicting (i.e., independent) Criteria: SIR γ In a path-loss fading, this corresponds to an adaptive distance guard * X. Wu, S. Tavildar, S. Shakkottai, T. Richardson, J. Li, R. Laroia, and A. Jovicic, FlashLinQ: A Synchronous Distributed Scheduler for Peer-to-Peer Ad Hoc Networks, IEEE/ACM Transactions on Networking, vol. 21, no. 4, pp , Aug

8 How to define the conflict graph? Geometric Metric: INR γ Why not, for example, p SNR INR FlashLinQ Metric: SIR γ A theoreticallyjustified criteria?

9 Our proposal: Information-Theoretic Conflict Graph A set of users in a wireless network form an informationtheoretic independent set (ITIS) (i.e., are non-conflicting), if Ø treating interference as noise (TIN) is informationtheoretically optimal for the sub-network created by them

10 Condition for Optimality of TIN Question: Under what condition, power control and treating interference as noise (TIN) is optimal? Not much is known about its op2mality (except for some symmetric and very low interference regimes)! TIN region is hard to analyze analy2cally Includes a (hard) op2miza2on problem for power control (e.g. Foschini- Milijanic 1993, Tan- Chiang- Srikant 2013) Can be characterized through solving a sequence of GP s (Mahdavi et al 2008) Non- explicit and non- convex region in general Very few general bounds on the capacity region of K- user IC

11 Optimality condition for TIN Theorem: In a K- user interference channel, if SNR i max INR ji max INR ij, j:j6=i j:j6=i 8i 2 {1, 2,...,K} Ø TIN achieves the capacity region within a constant gap of log 2 (3K) bits, Ø TIN region is approximated by a polyhedron. In words, the condi2on is at each user, the desired channel strength is at least the sum of the strengths of the strongest interference from this user and the strongest interference to this user T 1 # R 1 # T 2 # R 2 # T K # R K # *C. Geng, N. Naderializadeh, A. S. Aves2mehr, and S. A. Jafar On the Op2mality of Trea2ng Interference as Noise, submi\ed to IT.

12 Optimality condition for TIN Theorem: In a K- user interference channel, if SNR i max INR ji max INR ij, j:j6=i j:j6=i 8i 2 {1, 2,...,K} Ø TIN achieves the capacity region within a constant gap of log 2 (3K) bits, Ø TIN region is approximated by a polyhedron. In words, the condi2on is at each user, the desired channel strength is at least the sum of the strengths of the strongest interference from this user and the strongest interference to this user

13 Definition of ITIS and ITLinQ In a network of n users, S {1,,n} is called an informationtheoretic independent set (ITIS) if for any user i S Information-theoretic link scheduling (ITLinQ) Identify an ITIS and schedule the users in ITIS to transmit together Each destination treats all its incoming interference as noise

14 A Simpler Sufficient Condition S forms an ITIS if for any user i S Geometric ITLinQ FlashLinQ Metric: INR γ Metric: p SNR INR Metric: SNR INR

15 How good is ITLinQ? n source-destination pairs Sources located uniformly at random in a circle of radius R Each destination located within a distance r n =r 0 n -β of its corresponding source, β >0 Channel gain at distance r is equal to h 0 r -α (path loss) r n n-β R

16 Capacity Analysis of ITLinQ Theorem*: In the described setting, when n, ITLinQ can almost-surely achieve a fraction λ of the capacity region within a gap of k bits, where r n n-β R * N. Naderializadeh and A. S. Avestimehr, ITLinQ: A New Approach for Spectrum Sharing in Device-to-Device Communication Systems, submitted to IEEE JSAC Special Issue on 5G Wireless Communications

17 Comparison with Geometric approach r n n-β R

18 Closest AP-Selection Model All sources and destinations located randomly and uniformly inside a circle of radius R Each destination gets associated with its closest source Corollary: In the above model, ITLinQ can achieve a fraction ( 1 p n ) of the capacity region ( p n) gain over the geometric approach

19 Proof Sketch of the Theorem If the distance between sources in S {1,,n} is greater than d th,n, where then S is an ITIS. Convert the network to information-theoretic conflict graph G n Nodes: {1,,n} Edges: i and j are connected if the distance between S i and S j is not greater than d th,n G n is a random geometric graph ITLinQ can achieve 1/χ(G n ) fraction of capacity region > d th,n d th,n

20 Comparing with FlashLinQ Links (S-D pairs) dropped uniformly in a 1km 1km square Length of each link ~ u(0,40m) Two-way training is used for estimating local channels at each node 1km At each phase a (random) priority order is assigned to all links in the network Link j will be active in that phase if 1) At destination j: 2) At source j: p SNRj 1, 8i <j INR ij p SNRj 1, 8i <j INR ji Transmit power: 20 dbm Noise PSD: -174 dbm/hz Channel model: ITU-1411 LoS Log-normal shadowing with 10 db STD

21 Comparing with FlashLinQ Links (S-D pairs) dropped uniformly in a 1km 1km square Length of each link ~ u(0,40m) Two-way training is used for estimating local channels at each node 1km At each phase a (random) priority order is assigned to all links in the network Link j will be active in that phase if 1) At destination j: 2) At source j: SNR j INR ji 1, 8i <j SNR j INR ij 1, 8i <j Transmit power: 20 dbm Noise PSD: -174 dbm/hz Channel model: ITU-1411 LoS Log-normal shadowing with 10 db STD

22 Comparison with FlashLinQ: Sum-rate 110% Gain

23 Comparison with FlashLinQ: Average Rate Distribution

24 Summary and Concluding Remarks Interference management and coordination is a key bottleneck in D2D networks We introduced ITLinQ We characterized a sufficient condition for optimality of treating interference as noise We used this condition to define information-theoretic conflict graphs for scheduling Demonstrated significant gains over similar state-of-the-art schemes, such as FlashLinQ Impact on utility maximization? An approach to jointly address Interference management and coordination?

25 Questions?