215 IEEE Hong Kong-Taiwan Joint Workshop on Information Theory and Communications Harnessing Interaction in Bursty Interference Networks I-Hsiang Wang NIC Lab, NTU GICE 1/19, 215
Modern Wireless: Grand Challenges Global mobile data traffic is expected to increase nearly 11-fold from 213 218 (source: Cisco VNI mobile forecast) What drives this increase? - Growing # of users - Data-hungrier (smarter) mobile devices Grand challenges: Support higher data rate Accommodate more users Increase spectral efficiency 2
More Users + Growing Demand more interference along with intermittence @!@#! abed ㄅㄆ ø kklkle #$# πø $@#ßå µ orzz Toy example: people talking in a crowded room 3
Also, more chances for interaction 4
Interference vs. Intermittence Interference: - Born-nature of wireless medium - Limits the throughput of modern wireless networks Intermittence: - Lack of coordination across distributed terminals - Born-nature of data traffic - Not a critical issue if users do not interfere with one another Major bottleneck: interference 5
Modern View on Interference Traditional techniques: orthogonalization (2G), treating interference as noise (3G, 4G) Recent advances in network information theory shed light on how to break the interference barrier: Two-user IC: approximate capacity to within 1 bit/s/hz - Partial interference cancellation [Etkin et al IT8] K-user IC: total # of degrees of freedom = K/2 - Interference alignment [Maddah-Ali et al IT8] [Cadambe & Jafar IT8] Caveat: (in most information theoretic literature) Assume that interference is always present 6
Interaction Helps Manage Interference Cooperation among Tx/Rx terminals - Cloud RAN, cooperative MIMO, etc. - Capacity increase is bounded by the cooperation capacity [W & Tse IT11] Feedback from Rx to Tx - With perfect output feedback, capacity increase is multiplicative [Suh & Tse IT11] However, interaction cannot increase the total multiplexing gain (degrees of freedom) [Cadambe & Jafar IT9] Caveat: (in most information theoretic literature) Assume that interference is always present 7
Interference can be Intermittent At PHY, interference may not be always present Medium access control mechanism - Decentralized resource allocation Time 1 Time 2 Time 3 User 1 f1 f1 f1 User 2 f1 f2 f1 Network layer protocol/demand - Bursty network traffic Time 1 Time 2 Time 3 User 1 On Off On User 2 Off On On 8
Exploit Intermittence of Interference Naive idea: send more when there is no interference but may not know if there is interference beforehand Not completely hopeless: we may have delayed information about the presence of interference, thanks to interaction (feedback/cooperation) If interferer s activity can be predicted from the past, then obviously such delayed information can help If not (in the extreme case when coherence time = 1 symbol time), it seems such delayed information will not help 9
Question 1: Two Major Questions In bursty/intermittent environment, does delayed state information help mitigate interference? Question 2: In bursty/intermittent environment, does interaction play a more important role in interference management, compared with the static one? Both answers are YES. 1
Two Case Studies in this Talk Bursty Interference Channel with Feedback Bursty Interference Channel with a Relay Intermittent Interference Bursty Traffic With delayed state information, in both cases interaction increases the multiplexing gain degrees of freedom (DoF) 11
Major Findings and Contributions Bursty IC with feedback: - One-bit feedback increases DoF - Characterize the approximate capacity - An adaptive coding scheme - Novel outer bounds Intermittent Interference Bursty (MIMO) IC with a relay: - In-band cooperation increases DoF - DoF increase linearly with the # of relay antennas - Derive conditions for interference-free DoF performances Bursty Traffic 12
Gaussian Interference Channel X 1 [t] Y 1 [t] W 1 Tx1 h 11 h 12 Rx1 cw 1 X 2 [t] h 21 Y 2 [t] W 2 Tx2 h 22 Rx2 cw 2 Unit power constraint CN(, 1) Y 1 [t] =h 11 X 1 [t]+h 12 X 2 [t]+z 1 [t] Y 2 [t] =h 22 X 2 [t]+h 21 X 1 [t]+z 2 [t] 13
Modeling Burstiness/Intermittence X 1 [t] Y 1 [t] W 1 Tx1 h 11 h 12 Rx1 cw 1 X 2 [t] Interference Channel h 21 Y 2 [t] W 2 Tx2 h 22 Rx2 cw 2 Random Activity State Controls the presence of interference 14
Bursty Interference Channel: Two Models Intermittent Interference: motivated by the MAC layer Bursty Traffic: motivated by the Network layer 15
Part I: Bursty Interference Channel with Feedback Joint work with S. Diggavi, C. Suh, P. Viswanath Intermittent Interference 16
Model Y 1 [1 : t 1],S[1 : t 1] X 1 [t] Y 1 [t] h 11 W 1 Tx1 h 12 Rx1 cw 1 S[t] S[t] X 2 [t] h 21 Y 2 [t] W 2 Tx2 h 22 Rx2 cw 2 Y 2 [1 : t 1],S[1 : t 1] {S [t ]}: i.i.d. Bernoulli Ber(p) process Y 1 [t] =h 11 X 1 [t]+h 12 (S[t]X 2 [t]) + Z 1 [t] Y 2 [t] =h 22 X 2 [t]+h 21 (S[t]X 1 [t]) + Z 2 [t] Encoding: X i [t] =(W f i, Y i [1 : t 1],S[1 : t 1]) 17
Generalized Degrees of Freedom For better illustration, let us focus on the symmetric setting: h 11 2 = h 22 2 = SNR, h 21 2 = h 12 2 = INR INR INR SNR SNR Generalized degrees of freedom: d := Achievable Rate under Interference Interference Free Capacity R 2 (R sym,r sym ) d( ) = lim SNR!1 R sym log INR, with fixed := log SNR log SNR capacity region R 1 GDoF: a function of - strength of interference - probability of interference p Gain in GDoF multiplicative gain in capacity 18
Baseline d( ) Non-bursty, no-feedback Non-bursty, feedback Bursty, p=.5, no-feedback 1 No gain in the regime 2 2 for 3, 2 1 p 2 Can Feedback Really Help? DoF? [Suh & Tse IT11] [Etkin et al IT8] [Khude et al ISIT9] 1 2 1 2 2 3 1 2 19
In the regime 2 2, 2 3 (moderate to strong interference) Feedback in non-bursty case No gain in GDoF and DoF Exploiting intermittence without feedback No gain in GDoF and DoF 2
Baseline d( ) Non-bursty, no-feedback Non-bursty, feedback Bursty, p=.5, no-feedback 1 No gain in the regime 2 2 for 3, 2 1 p 2 Can Feedback Really Help? DoF? [Suh & Tse IT11] [Etkin et al IT8] [Khude et al ISIT9] 1 2 1 2 2 3 1 2 21
One- Bit Feedback Increases DoF Phase Fresh Transmit as if there is no interference X1 L1(X1,X2) State No interference Rx Operation Both Decode Tx Operation Stay in Phase F Interference Hold Phase Retransmit Retransmit the unresolved symbol State No interference Interference Rx Operation Both Decode Both Decode (exploit side info.) Jump to Phase R Tx Operation Back to Phase F Back to Phase F X2 DoF = 1 1 p F L2(X1,X2) DoF = p 1 DoF = 1 R Stationary Distribution: (F) = 1 1+p, (R) = p 1+p =) DoF = 1 1+p 22
Synergetic Bene_it of Feedback d( ) Non-bursty, no-feedback Non-bursty, feedback Bursty, p=.5, no-feedback Bursty, p=.5, feedback 1 No gain in the regime 2 2 for 3, 2 1 p 2 [Khude et al ISIT9] 1 1+p [Suh & Tse IT11] [Etkin et al IT8] 1 2 1 2 2 3 1 2 23
Harnessing Output Feedback Weak interference regime Analog interference refinement Strong interference regime Interference refinement with structured codes Very strong interference regime Opportunistic relaying with structured codes 24
Part II: Bursty Interference Channel with a Relay Joint work with S. Kim, C. Suh Bursty Traffic 25
Model M Tx antennas S t 1 Y R (t) Z R (t) S t L Relay antennas Enc Relay R X R (t) N Rx antennas W 1 Enc Tx11 X 1 (t) Y 1 (t) Dec Rx11 Ŵ 1 S t 1 S 1 (t) Z 1 (t) W 2 Enc Tx22 X 2 (t) Y 2 (t) Dec Rx22 Ŵ 2 S 2 (t) S := (S 1,S 2 ) Z 2 (t) Y 1 (t) =H 11 (S 1 (t)x 1 (t)) + H 12 (S 2 (t)x 2 (t)) + H 1R X R + Z 1, Y 2 (t) =H 21 (S 1 (t)x 1 (t)) + H 22 (S 2 (t)x 2 (t)) + H 2R X R + Z 2, Y R (t) =H R1 (S 1 (t)x 1 (t)) + H R2 (S 2 (t)x 2 (t)) + Z R. {S1 [t]} {S 2 [t]}: i.i.d. Bernoulli Ber(p) processes 26
In- Band Relay Increases DoF Bursty P2P M Ber(p) N Tx Rx d = p min (M,N) = pn for M sufficiently large Adding a relay L Relay Ber(p) M N Tx Rx d =min{p min (M,N + L),pmin (M + L, N)+(1 p)min(l, N)} = pn + min{pl, (1 p)min(l, N)} for M sufficiently large 27
d N pn N L DoF increase is linear in L, # of relay antennas DoF is boost by up to p 1 times 28
Interference- Free Communication For two-user bursty interference channel with a relay, main focus is on getting interference-free DoF p [,1) Sufficient condition$ $ $ $ Necessary condition 2M apple N or M 2N + L and L 2N 2M apple N or M 2N + L and L 2N or or M 2N + L and 3L apple N M 2N and 3L apple N 29
Cooperative Zero- Forcing 4 a 5 + b 5 3 2 1 t 1 2 3 sent received a 1 + b 1 a 1 + b 1 a 1 + b 1 b 5 a 5 Relay 4 L =1 4 3 2 1 1 2 3 4 a 9 a 5 a 5 a 1 2 + b 1 b 5 a 6 6 6 +a 1 + b 1 +a 1 + b 1 11 7 a 7 12 8 8 Tx 1 Rx1 a 3 b 1 a 7 +a 1 + b 1 +a 1 + b a 1 5 a 5 a 9 b 9 b 1 b 11 b 12 a 1 b 9 sent b 1 b 11 b 12 a 1 b 9 b 9 b 5 b 6 b 7 b 8 a 1 2 3 4 a 1 b 5 b 1 b 1 b1 b 2 b 3 b 4 Tx 2 t (S 1,S 2 ) 1 (1, 1) 2 (, 1) 3 (1, ) 4 (, ) Rx 2 M =7 N =3 a 4 b 2 + a 1 b 3 b 4 a 1 + b 1 b 6 received +a 1 + b 1 b 7 +a 1 + b 1 b 8 +a 1 + b 1 a 8 +a 1 + b 1 a 5 +a 1 + b 1 a 1 +a 1 + b 1 a 1 + b 1 a 5 + b 5 a 5 + b 5 a 5 + b 5 a 5 + b 5 a 5 + b 5 a 5 + b 5 3
Takeaway 1 Interaction (feedback and cooperation) is more helpful in bursty communication 2 Harnessing delayed state information and interaction boost up capacity greatly 31
Since interaction is so useful, please give us feedback and let us collaborate! Email: ihwang@ntu.edu.tw Website: http://cc.ee.ntu.edu.tw/~ihsiangw 32
Thank You! 33