Full-duplex without Strings: Enabling Full- duplex with Half-duplex - - PowerPoint PPT Presentation

Full-duplex without Strings: Enabling Full- duplex with Half-duplex Clients

Karthikeyan Sundaresan, Mohammad Khojastepour, Eugene Chai, Sampath Rangarajan

NEC Labs America MobiCom 2014

Full-duplex

• Transmitting + receiving on same time-

frequency resource

• Key challenge: Self-interference
• Several advancements in full-duplex design

– Antenna + RF + digital cancelation – Three, two and single antenna designs – Co-existence with MIMO

– Focus on peer-peer FD networks

2

FD Base Station FD Client

SI

• 1. “Achieving single channel, full duplex wireless communication”, J. Choi et. al. MobiCom, 2010.
• 2. “Experiment-driven characterization of full-duplex wireless systems”, M. Duarte et. al. IEEE Transactions on Wireless Communications, 2012.
• 3. “MIDU: Enabling MIMO Full-duplex”, E. Aryafar et al. MobiCom 2012.
• 4. “Full-duplex radios”, D. Bharadia et. al., Sigcomm 2013.

Distributed Full-duplex

• Can we enable FD communication

(2x multiplexing gain) with HD clients in a single cell?

– Easier to embed FD functionality in BS/AP

• Distributed FD

– Uplink from one client and downlink to another client

• Key challenge: uplink-downlink

interference (UDI)

3

FD Base Station HD Client HD Client

UDI SI

Potential Solutions for UDI

• Impact of UDI depends on topology
• Large impact for comparable distances

4

UDI

(d – distance between BS and DL client)

d

Potential Solutions for UDI

• Impact of UDI depends on topology
• Implicit: leverage client separation
• Explicit: use side channels [Bai-Arxiv’12]
• Explicit: time-based interference alignment [Sahai-ITW’13]
• Explicitly address UDI in the same channel in a scalable manner

5

UDI

(d – distance between BS and DL client)

Scaling to MIMO?

d

Approach

• Leverage spatial interference

alignment to address UDI between HD clients

– Use multiple antennas at HD clients – Pack interference in lesser dimensions

• Efficient: same channel
• Scalable: co-exist with MIMO
• Deployable: only as challenging

as MU-MIMO systems

6 y1 y2

x1,x2 x3,x4 y1,y2 y3,y4

x3 x2 x4 x1 y4 y3 x4 x2 x1 x3

DL UL

Challenges

• CSI overhead for UDI

– More clients (dimensions), easier IA, but more overhead

• Constructing a feasible IA solution

– MIMO precoders (V), receiver filers (U) at clients and AP

• Handling clients with

heterogeneous antenna capabilities

• Optimizing rate for the FD streams

7

• FDoS: System that addresses above challenges to enable

FD with HD clients

....

(N)

.... .... .... ....

(N) (N) N streams N streams

V1 V2 U1 U2 V0 U0

(1) Applying IA to FD Networks

• Results

– N even: 4 clients necessary to address UDI and enable 2N streams – N odd: 6 clients necessary (symmetric) – N odd: 5 clients necessary (asymmetric)

• Focus on symmetric FD networks

– Constant overhead: CSI between 4

• r 6 clients

– Does not scale with N

8

.... V1 V3 V2 U1 U3 U2 .... .... .... .... .... ....

N/2 N/2 N/2 N/2 1 1

• Receiver spatial dimensions

– Desired (1:1) – Interference suppression (1:1) – IA (1:many)

(2) Constructing IA Solution

9

.....

• At most one cycle in IAN
• Closed-form IA solution
• 2N streams achievable even with UDI

for symmetric FD networks

– With 4 (6) clients for N even (odd)

(2) Constructing IA Solution

10

... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...

q K-q

p+1 p+1 p+1 p+1 p+1 p+1 p p p p p p

cyclic acyclic

Construct a feasible IAN Determine IA solution Select IA solution for cyclic part Find resulting IA solution for acyclic part

cyclic acyclic

Example: N=5, 6 clients, 10 streams

11

2 1 2 2 1 2 V1 V3 V2

H11v11 H11v12 H12v21 H12v22 H13v31 H10v01 H10v02 H21v11 H22v21 H22v22 H21v12 H23v31 H20v04 H20v03 H32v22 H31v12 H31v11 H33v31 H32v21 H30v05

(3) Heterogeneous Clients

• Clients with different number of

antennas

– Affects number of FD streams supported

• M+N streams achievable with FD
• Different IA construction required

– Combination of symmetric and asymmetric FD networks

12

....

(N)

.... .... .... ....

(M) (N) ? streams ? streams (M) (N)

Evaluation

• Testbed

– One AP and four clients (WARP nodes) with 2 or 4 antennas each – FD: SI cancelation based on prior works – Focus on UDI cancelation between UL and DL clients

• Cancelation over 64 sub-carrier OFDM, 10 MHz channel

– Experiments in indoor office environment

• Baselines

– HD system MU-MIMO (zero-forcing beamforming) – FD without UDI cancelation

• Metric

– SINR measurements, rate translation from SINR

13

Results (1) – UDI Suppression

• 10-20 dB of median UDI suppression out of 30 dB

15-20 dB 10-15 dB

Results (2) – Rate Performance

• 1.75-2x FD rate gain
• 1.5-2x gain over schemes not addressing UDI
• Not addressing UDI can degrade performance to

worse than HD

Conclusions

• FD has potential to increase system capacity by 2x

– All the more powerful if HD clients can be used

• UDI is a key challenge in distributed FD networks
• FDoS: a system that leverages spatial IA to address UDI

– Theory and design of applying spatial IA for distributed FD – Incorporates practical considerations (overhead, rate, heterogeneity) – Demonstrates 1.5-2x gain in presence of UDI in practice

• Next steps…
• FD with HD clients in multi-cell networks

Thanks!

17

(3) Rate Optimization

• Very challenging problem

– MU-MIMO precoding on DL and UL coupled through IA between DL-UL

• Modular design

– De-couple IA from MU-MIMO precoder (rate) optimization – Retain structure of IA solution for UDI – Optimize DL and UL MU-MIMO precoders given IA solution

• Distributed realization of IA solution

– Overhead reduced further by half

18 Jointly pick N/2 vectors each for V1,V2 that maximize rate of N UL streams subject to IA

Fix receive filter U0

for AP from UL

• ptimization

Given V1,V2, pick receiver filters U1,U2 orthogonal to sub-space spanned by interference Pick precoder V0 at AP to maximize rate

• f N DL streams

UL clients (V1,V2) AP (U0) DL clients (U1,U2) AP (V0)

FDoS Operations

19 Client and mode (FD

• vs. HD) selection based
• n multiplexing gain,

scheduling policy Estimate CSI for UL, DL and UDI channels with reduced feedback Distributed computation

• f solution (AP

broadcasts only one precoder) AP coordinates joint UL and DL transmissions during FD AP solicits/delivers block ACKs similar to MU-MIMO

Results (3) – Heterogeneity

....

(4)

.... .... .... ....

(2) (4) 4 streams 2 streams (2) (4)

• 6 streams sent in heterogeneous set-up
• Leverages heterogeneous antenna capabilities

effectively

Results (4) - Scalability

21

• Evaluated FDoS design for larger (even/odd) N
• FD gains scale and more pronounced with rate
• ptimization

(a) With rate optimization (b) Without rate optimization