Downlink Interference Alignment Changho Suh Wireless Foundations U.C. Berkeley

GLOBECOM 2010 Dec. 8

Joint work with David Tse and Minnie Ho (at Intel)

Two Point-to-Point Links

Two Point-to-Point Links

total # dim. for desired signals

d.o.f. =

total # dim.

(per sec per Hz)

Two Point-to-Point Links

total # dim. for desired signals

d.o.f. =

total # dim.

(per sec per Hz) (no interference)

Two Point-to-Point Links

total # dim. for desired signals

d.o.f. =

total # dim.

(per sec per Hz) (interference)

Two Point-to-Point Links

total # dim. for desired signals

d.o.f. =

total # dim.

(per sec per Hz) (interference)

Two Uplinks

Two Uplinks

Two Uplinks

Suh & Tse 08:

Two Uplinks

Suh & Tse 08:

Interference-free

Two Uplinks

Idea: Interference Alignment

Suh & Tse 08:

Interference-free

Uplink Interference Alignment

Uplink Interference Alignment

Suppose every node has K+1 dimensions (antennas or subcarriers)

Uplink Interference Alignment Transmit a vector

Suppose every node has K+1 dimensions (antennas or subcarriers)

Uplink Interference Alignment

Fix a reference vector, independent of channel gains

Uplink Interference Alignment

Fix a reference vector, independent of channel gains

Uplink Interference Alignment

Uplink Interference Alignment

Uplink Interference Alignment

Uplink Interference Alignment

Uplink Interference Alignment

Uplink Interference Alignment

total # dim =K+1

Uplink Interference Alignment

total # dim =K+1

Downlink Interference Alignment

Downlink Interference Alignment

Uplink-Downlink Duality

Downlink Interference Alignment

Uplink-Downlink Duality

Downlink Interference Alignment

Uplink-Downlink Duality

But HOW?

Downlink Interference Alignment

Fix K-dim reference plane, indep. of channel gains

Downlink Interference Alignment

Fix K-dim reference plane, indep. of channel gains

Downlink Interference Alignment

Downlink Interference Alignment

Downlink Interference Alignment

Downlink Interference Alignment

1

Downlink Interference Alignment

1

Downlink Interference Alignment

1

Downlink Interference Alignment

1

1

Downlink Interference Alignment

1

1

1

Downlink Interference Alignment 1

1

1

1

Downlink Interference Alignment 1

2 1

Downlink Interference Alignment 1 2

2

2

1

2

Downlink Interference Alignment 1

2 1

2 K

Downlink Interference Alignment 1 K

2 1

2 K K

K

Downlink Interference Alignment 1

2 1

2 K K

Downlink Interference Alignment 1 2 K

2 1

K

Downlink Interference Alignment 1

1 2 2 K

K

2 1

K

Downlink Interference Alignment 1

1 2 2 K

K

2 1

K

K out-of-cell int. vectors (K-1) intra-cell int. vectors

Downlink Interference Alignment 1

1 2 2 K

K

2 1

K

K out-of-cell int. vectors (K-1) intra-cell int. vectors  Aligned onto K-dim plane

From Theory to Practice: How to Implement?

Uplink: Implementation Challenge

Uplink: Implementation Challenge

Requires crosschannel knowledge

Uplink: Implementation Challenge

Requires crosschannel knowledge

Uplink: Implementation Challenge

Requires crosschannel knowledge

Requires backhaul cooperation

Backhaul

Uplink: Implementation Challenge

Requires crosschannel knowledge

Requires backhaul cooperation

Backhaul

How about for DOWNLINK?

Uplink: Implementation Challenge

Requires crosschannel knowledge

Requires backhaul cooperation

Backhaul

How about for DOWNLINK? Can achieve interference alignment without using backhaul!

Downlink: Implementation Benefits 1 2 K

2 1

K

Fix K-dim reference plane, indep. of channel gains

Downlink: Implementation Benefits 1 2 K

(K+1)-by-K

Fix K-dim reference plane, indep. of channel gains

Downlink: Implementation Benefits 1 2 K

(K+1)-by-K

Fix K-dim reference plane, indep. of channel gains

Can estimate it using pilots or preamble

Downlink: Implementation Benefits 1 2 K

(K+1)-by-K

Can estimate it using pilots or preamble

Downlink: Implementation Benefits 1 2 K

(K+1)-by-K

Can estimate it using pilots or preamble

Downlink: Implementation Benefits 1 2 K

(K+1)-by-K

Downlink: Implementation Benefits 1 2 K

(K+1)-by-K

Same as the feedback in standard MU-MIMO techniques

Downlink: Implementation Benefits 1 2 K

(K+1)-by-K

Same as the feedback in standard MU-MIMO techniques

Can be implemented with little change to an existing cellular system supporting multi-user MIMO

Modification

Modification

Reality: 1. Finite SNR

Modification

Reality: 1. Finite SNR 2. Residual out-of-cell interference

Modification

Reality: MMSE-like Modification

1. Finite SNR 2. Residual out-of-cell interference

Cell-Edge Performance: Two Cells

Cell-Edge Performance: Two Cells 3 cell-edge users per cell

Cell-Edge Performance: Two Cells 3 cell-edge users per cell 4-by-4 antenna configuration

Cell-Edge Performance: Two Cells 3 cell-edge users per cell 4-by-4 antenna configuration 30

Sum Rate (bits/s/Hz)

25

20

15

10

Baseline: Matched Filtering

5

0 -20

-10

0 10 SNR (dB)

20

30

Cell-Edge Performance: Two Cells 3 cell-edge users per cell 4-by-4 antenna configuration 30

Proposed IA

Sum Rate (bits/s/Hz)

25

20

15

300% improvement 10

Baseline: Matched Filtering

5

0 -20

-10

0 10 SNR (dB)

20

30

19 Hexagonal Cells 3 cell-edge users per cell 4-by-4 antenna configuration

19 Hexagonal Cells 3 cell-edge users per cell 4-by-4 antenna configuration

Non-negligible residual interference

19 Hexagonal Cells 3 cell-edge users per cell 4-by-4 antenna configuration Proposed IA 28% gain

Sum Rate (bits/s/Hz)

4

Baseline: Matched Filtering

3

2

Non-negligible residual interference

1

0 -20

-10

0

10 20 SNR (dB)

30

40

Performance Gain Residual interference

negligible

non-negligible

Significant gain Gain reduced; but still respectable

Performance Gain Residual interference

negligible

non-negligible

Significant gain Gain reduced; but still respectable

Great potential to heterogeneous networks

Performance Gain Residual interference

negligible

non-negligible

Significant gain Gain reduced; but still respectable

Great potential to heterogeneous networks E.g., macro-pico cellular networks

Pico-BS

Performance Gain Residual interference

Significant gain

negligible

non-negligible

Gain reduced; but still respectable

Great potential to heterogeneous networks E.g., macro-pico cellular networks

Pico-user significantly interfered with by the nearby macro-BS

Pico-BS

Conclusion • Developed a new downlink IA technique

Conclusion • Developed a new downlink IA technique • Requires only localized within-the-cell feedback mechanism

Conclusion • Developed a new downlink IA technique • Requires only localized within-the-cell feedback mechanism • Can be implemented with small changes to multiuser MIMO systems

Conclusion • Developed a new downlink IA technique • Requires only localized within-the-cell feedback mechanism • Can be implemented with small changes to multiuser MIMO systems

• Shows ~30% cell-edge throughput gain in 19 hexagonal cellular layout

Conclusion • Developed a new downlink IA technique • Requires only localized within-the-cell feedback mechanism • Can be implemented with small changes to multiuser MIMO systems

• Shows ~30% cell-edge throughput gain in 19 hexagonal cellular layout • Has great potential to heterogeneous networks