cyclic communication and inseparability in mimo relay...

Post on 11-Aug-2020

3 Views

Category:

Documents

0 Downloads

Preview:

Click to see full reader

TRANSCRIPT

RUB Institute of Digital Communication Systems

Cyclic Communication and Inseparability in MIMORelay Channels

Aydin Sezginjoint work with Anas Chaaban

Institute of Digital Communication SystemsRUB, Bochum

Anas Chaaban, Aydin Sezgin Multi-way Communications 1

RUB Institute of Digital Communication Systems

Outline

1 Motivation: From one-way to multi-way

2 The MIMO Y-channel: From single-antenna to multiple-antennas

3 Main result: From capacity to DoF

4 Insights and ingredientsChannel diagonalization: Separability of communicationstructureAlignment, Compute-and-forwardTransmission strategy(3-users)

5 Extensions and Conclusion

Anas Chaaban, Aydin Sezgin Multi-way Communications 2

RUB Institute of Digital Communication Systems

Outline

1 Motivation: From one-way to multi-way

2 The MIMO Y-channel: From single-antenna to multiple-antennas

3 Main result: From capacity to DoF

4 Insights and ingredientsChannel diagonalization: Separability of communicationstructureAlignment, Compute-and-forwardTransmission strategy(3-users)

5 Extensions and Conclusion

Anas Chaaban, Aydin Sezgin Multi-way Communications 3

RUB Institute of Digital Communication Systems

Towards 50B devices in 2020!

Increasing number of connected devices (IoT, M2M, etc.)

Anas Chaaban, Aydin Sezgin Multi-way Communications 4

RUB Institute of Digital Communication Systems

Towards 50B devices in 2020!

More sophisticated network topologies!

Machine-to-machinePoint↔multi-point

Car-to-carMulti-hop

Device-to-device

X-Channel

Star

Interferencechannel

• Important factor: Relaying!

Question

Is communication mainly one-way?

Anas Chaaban, Aydin Sezgin Multi-way Communications 5

RUB Institute of Digital Communication Systems

Towards 50B devices in 2020!

More sophisticated network topologies!

Machine-to-machinePoint↔multi-point

Car-to-carMulti-hop

Device-to-device

X-Channel

Star

Interferencechannel

• Important factor: Relaying!

Question

Is communication mainly one-way?

Anas Chaaban, Aydin Sezgin Multi-way Communications 5

RUB Institute of Digital Communication Systems

Towards 50B devices in 2020!

More sophisticated network topologies!

Machine-to-machinePoint↔multi-point

Car-to-carMulti-hop

Device-to-device

X-Channel

Star

Interferencechannel

• Important factor: Relaying!

Question

Is communication mainly one-way?

Anas Chaaban, Aydin Sezgin Multi-way Communications 5

RUB Institute of Digital Communication Systems

Towards 50B devices in 2020!

More sophisticated network topologies!

Machine-to-machinePoint↔multi-point

Car-to-carMulti-hop

Device-to-device

X-Channel

Star

Interferencechannel

• Important factor: Relaying!

Question

Is communication mainly one-way?

Anas Chaaban, Aydin Sezgin Multi-way Communications 5

RUB Institute of Digital Communication Systems

Towards 50B devices in 2020!

More sophisticated network topologies!

Machine-to-machinePoint↔multi-point

Car-to-carMulti-hop

Device-to-device

X-Channel

Star

Interferencechannel

• Important factor: Relaying!

Question

Is communication mainly one-way?

Anas Chaaban, Aydin Sezgin Multi-way Communications 5

RUB Institute of Digital Communication Systems

Towards 50B devices in 2020!

More sophisticated network topologies!

Machine-to-machinePoint↔multi-point

Car-to-carMulti-hop

Device-to-device

X-Channel

Star

Interferencechannel

• Important factor: Relaying!

Question

Is communication mainly one-way?

Anas Chaaban, Aydin Sezgin Multi-way Communications 5

RUB Institute of Digital Communication Systems

Towards 50B devices in 2020!

More sophisticated network topologies!

Machine-to-machinePoint↔multi-point

Car-to-carMulti-hop

Device-to-device

X-Channel

Star

Interferencechannel

• Important factor: Relaying!

Question

Is communication mainly one-way?

Anas Chaaban, Aydin Sezgin Multi-way Communications 5

RUB Institute of Digital Communication Systems

Towards 50B devices in 2020!

More sophisticated network topologies!

Machine-to-machinePoint↔multi-point

Car-to-carMulti-hop

Device-to-device

X-Channel

Star

Interferencechannel

• Important factor: Relaying!

Question

Is communication mainly one-way?

Anas Chaaban, Aydin Sezgin Multi-way Communications 5

RUB Institute of Digital Communication Systems

Multi-way Communications

• Part of our daily communication is uni-directional

• A large share is bi-directional (video conferencing e.g.)⇒ Two-way channel

• Distant nodes⇒ Two-way relay channel

• More than 2 nodes⇒ Multi-way relay channel (MRC).

Anas Chaaban, Aydin Sezgin Multi-way Communications 6

RUB Institute of Digital Communication Systems

Multi-way Communications

• Part of our daily communication is uni-directional

• A large share is bi-directional (video conferencing e.g.)⇒ Two-way channel

• Distant nodes⇒ Two-way relay channel

• More than 2 nodes⇒ Multi-way relay channel (MRC).

Anas Chaaban, Aydin Sezgin Multi-way Communications 6

RUB Institute of Digital Communication Systems

Multi-way Communications

• Part of our daily communication is uni-directional

• A large share is bi-directional (video conferencing e.g.)⇒ Two-way channel

• Distant nodes⇒ Two-way relay channel

• More than 2 nodes⇒ Multi-way relay channel (MRC).

Anas Chaaban, Aydin Sezgin Multi-way Communications 6

RUB Institute of Digital Communication Systems

Multi-way Communications

• Part of our daily communication is uni-directional

• A large share is bi-directional (video conferencing e.g.)⇒ Two-way channel

• Distant nodes⇒ Two-way relay channel

• More than 2 nodes⇒ Multi-way relay channel (MRC).

Anas Chaaban, Aydin Sezgin Multi-way Communications 6

RUB Institute of Digital Communication Systems

Multi-way Relaying

• Multi-cast MRC: a message has multiple destinations

• Uni-cast MRC (Y-channel): a message has one destination

Anas Chaaban, Aydin Sezgin Multi-way Communications 7

RUB Institute of Digital Communication Systems

Multi-way Relaying

• Multi-cast MRC: a message has multiple destinations

• Uni-cast MRC (Y-channel): a message has one destination

Anas Chaaban, Aydin Sezgin Multi-way Communications 7

RUB Institute of Digital Communication Systems

Outline

1 Motivation: From one-way to multi-way

2 The MIMO Y-channel: From single-antenna to multiple-antennas

3 Main result: From capacity to DoF

4 Insights and ingredientsChannel diagonalization: Separability of communicationstructureAlignment, Compute-and-forwardTransmission strategy(3-users)

5 Extensions and Conclusion

Anas Chaaban, Aydin Sezgin Multi-way Communications 8

RUB Institute of Digital Communication Systems

MIMO Y-channel

1

· · ·M

R

· · ·N

2

· · ·M

3

· · · M

H1H1x1 H2H2

x2

H3H3

x3

yr

1

· · ·M

R

· · ·N

2

· · ·M

3

· · · M

D1D1y1 D2D2

y2

D3D3

y3

xr

Uplink:

• Tx signal: xi ∈ CM , power P

• Uplink channels: Hi ∈ CN×M

Downlink:

• Relay signal: xr ∈ CN , power P

• Downlink channels: Di ∈ CM×N

Anas Chaaban, Aydin Sezgin Multi-way Communications 9

RUB Institute of Digital Communication Systems

MIMO Y-channel

1

· · ·M

R

· · ·N

2

· · ·M

3

· · · M

H1H1x1 H2H2

x2

H3H3

x3

yr

1

· · ·M

R

· · ·N

2

· · ·M

3

· · · M

D1D1y1 D2D2

y2

D3D3

y3

xr

Uplink:

• Tx signal: xi ∈ CM , power P

• Uplink channels: Hi ∈ CN×M

Downlink:

• Relay signal: xr ∈ CN , power P

• Downlink channels: Di ∈ CM×N

Anas Chaaban, Aydin Sezgin Multi-way Communications 9

RUB Institute of Digital Communication Systems

MIMO Y-channel

1

· · ·M

R

· · ·N

2

· · ·M

3

· · · M

H1H1x1 H2H2

x2

H3H3

x3

yr

1

· · ·M

R

· · ·N

2

· · ·M

3

· · · M

D1D1y1 D2D2

y2

D3D3

y3

xr

Uplink:

• Tx signal: xi ∈ CM , power P

• Uplink channels: Hi ∈ CN×M

Downlink:

• Relay signal: xr ∈ CN , power P

• Downlink channels: Di ∈ CM×N

Anas Chaaban, Aydin Sezgin Multi-way Communications 9

RUB Institute of Digital Communication Systems

MIMO Y-channel

1

· · ·M

R

· · ·N

2

· · ·M

3

· · · M

H1H1x1 H2H2

x2

H3H3

x3

yr

1

· · ·M

R

· · ·N

2

· · ·M

3

· · · M

D1D1y1 D2D2

y2

D3D3

y3

xr

Uplink:

• Tx signal: xi ∈ CM , power P

• Uplink channels: Hi ∈ CN×M

Downlink:

• Relay signal: xr ∈ CN , power P

• Downlink channels: Di ∈ CM×N

Anas Chaaban, Aydin Sezgin Multi-way Communications 9

RUB Institute of Digital Communication Systems

MIMO Y-channel

1

· · ·M

R

· · ·N

2

· · ·M

3

· · · M

H1H1x1 H2H2

x2

H3H3

x3

yr

1

· · ·M

R

· · ·N

2

· · ·M

3

· · · M

D1D1y1 D2D2

y2

D3D3

y3

xr

Uplink:

• Tx signal: xi ∈ CM , power P

• Uplink channels: Hi ∈ CN×M

Downlink:

• Relay signal: xr ∈ CN , power P

• Downlink channels: Di ∈ CM×N

Anas Chaaban, Aydin Sezgin Multi-way Communications 9

RUB Institute of Digital Communication Systems

Capacity to Capacity Region

Single-user (MIMO P2P):

Tx RxH

Input covariance Q, tr(Q) ≤ P

Capacity: C = log |I+HQHH |

R0 C

achievable rate

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Input covariance Qi, i = 1, 2,tr(Qi) ≤ P

Capacity region:

Ri ≤ log |I+HiQiHHi |,

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

R2

R10

achievable

rate region

Anas Chaaban, Aydin Sezgin Multi-way Communications 10

RUB Institute of Digital Communication Systems

Capacity to Capacity Region

Single-user (MIMO P2P):

Tx RxH

Input covariance Q, tr(Q) ≤ P

Capacity: C = log |I+HQHH |

R0 C

achievable rate

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Input covariance Qi, i = 1, 2,tr(Qi) ≤ P

Capacity region:

Ri ≤ log |I+HiQiHHi |,

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

R2

R10

achievable

rate region

Anas Chaaban, Aydin Sezgin Multi-way Communications 10

RUB Institute of Digital Communication Systems

Capacity to Capacity Region

Single-user (MIMO P2P):

Tx RxH

Input covariance Q, tr(Q) ≤ P

Capacity: C = log |I+HQHH |

R0 C

achievable rate

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Input covariance Qi, i = 1, 2,tr(Qi) ≤ P

Capacity region:

Ri ≤ log |I+HiQiHHi |,

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

R2

R10

achievable

rate region

Anas Chaaban, Aydin Sezgin Multi-way Communications 10

RUB Institute of Digital Communication Systems

Capacity to Capacity Region

Single-user (MIMO P2P):

Tx RxH

Input covariance Q, tr(Q) ≤ P

Capacity: C = log |I+HQHH |

R0 C

achievable rate

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Input covariance Qi, i = 1, 2,tr(Qi) ≤ P

Capacity region:

Ri ≤ log |I+HiQiHHi |,

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

R2

R10

achievable

rate region

Anas Chaaban, Aydin Sezgin Multi-way Communications 10

RUB Institute of Digital Communication Systems

Capacity to Capacity Region

Single-user (MIMO P2P):

Tx RxH

Input covariance Q, tr(Q) ≤ P

Capacity: C = log |I+HQHH |

R0 C

achievable rate

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Input covariance Qi, i = 1, 2,tr(Qi) ≤ P

Capacity region:

Ri ≤ log |I+HiQiHHi |,

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

R2

R10

achievable

rate region

Anas Chaaban, Aydin Sezgin Multi-way Communications 10

RUB Institute of Digital Communication Systems

Capacity to Capacity Region

Single-user (MIMO P2P):

Tx RxH

Input covariance Q, tr(Q) ≤ P

Capacity: C = log |I+HQHH |

R0 C

achievable rate

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Input covariance Qi, i = 1, 2,tr(Qi) ≤ P

Capacity region:

Ri ≤ log |I+HiQiHHi |,

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

R2

R10

achievable

rate region

Anas Chaaban, Aydin Sezgin Multi-way Communications 10

RUB Institute of Digital Communication Systems

Outline

1 Motivation: From one-way to multi-way

2 The MIMO Y-channel: From single-antenna to multiple-antennas

3 Main result: From capacity to DoF

4 Insights and ingredientsChannel diagonalization: Separability of communicationstructureAlignment, Compute-and-forwardTransmission strategy(3-users)

5 Extensions and Conclusion

Anas Chaaban, Aydin Sezgin Multi-way Communications 11

RUB Institute of Digital Communication Systems

Capacity to DoF

Single-user (MIMO P2P):

Tx RxH

Capacity:

C = log |I+HQHH |

Optimization: water-filling

DoF:

• M Tx antennas, N Rx antennas

⇒ H ∈ CN×M ⇒ rank(H) = min{M,N}⇒ C ≈ min{M,N} log(P ) at high P

• DoF: d = limP→∞C

log(P )= rank(H) ⇒ d = min{M,N}

⇒ Capacity equivalent to that of d parallel SISO P2P channels!

Tx RxTx RxTx Rx (more on that later)

Anas Chaaban, Aydin Sezgin Multi-way Communications 12

RUB Institute of Digital Communication Systems

Capacity to DoF

Single-user (MIMO P2P):

Tx RxH

Capacity:

C = log |I+HQHH |

Optimization: water-filling

DoF:

• M Tx antennas, N Rx antennas

⇒ H ∈ CN×M ⇒ rank(H) = min{M,N}⇒ C ≈ min{M,N} log(P ) at high P

• DoF: d = limP→∞C

log(P )= rank(H) ⇒ d = min{M,N}

⇒ Capacity equivalent to that of d parallel SISO P2P channels!

Tx RxTx RxTx Rx (more on that later)

Anas Chaaban, Aydin Sezgin Multi-way Communications 12

RUB Institute of Digital Communication Systems

Capacity to DoF

Single-user (MIMO P2P):

Tx RxH

Capacity:

C = log |I+HQHH |

Optimization: water-filling

DoF:

• M Tx antennas, N Rx antennas

⇒ H ∈ CN×M ⇒ rank(H) = min{M,N}⇒ C ≈ min{M,N} log(P ) at high P

• DoF: d = limP→∞C

log(P )= rank(H) ⇒ d = min{M,N}

⇒ Capacity equivalent to that of d parallel SISO P2P channels!

Tx RxTx RxTx Rx (more on that later)

Anas Chaaban, Aydin Sezgin Multi-way Communications 12

RUB Institute of Digital Communication Systems

Capacity to DoF

Single-user (MIMO P2P):

Tx RxH

Capacity:

C = log |I+HQHH |

Optimization: water-filling

DoF:

• M Tx antennas, N Rx antennas

⇒ H ∈ CN×M ⇒ rank(H) = min{M,N}⇒ C ≈ min{M,N} log(P ) at high P

• DoF: d = limP→∞C

log(P )= rank(H) ⇒ d = min{M,N}

⇒ Capacity equivalent to that of d parallel SISO P2P channels!

Tx RxTx RxTx Rx (more on that later)

Anas Chaaban, Aydin Sezgin Multi-way Communications 12

RUB Institute of Digital Communication Systems

Capacity to DoF

Single-user (MIMO P2P):

Tx RxH

Capacity:

C = log |I+HQHH |

Optimization: water-filling

DoF:

• M Tx antennas, N Rx antennas

⇒ H ∈ CN×M ⇒ rank(H) = min{M,N}

⇒ C ≈ min{M,N} log(P ) at high P

• DoF: d = limP→∞C

log(P )= rank(H) ⇒ d = min{M,N}

⇒ Capacity equivalent to that of d parallel SISO P2P channels!

Tx RxTx RxTx Rx (more on that later)

Anas Chaaban, Aydin Sezgin Multi-way Communications 12

RUB Institute of Digital Communication Systems

Capacity to DoF

Single-user (MIMO P2P):

Tx RxH

Capacity:

C = log |I+HQHH |

Optimization: water-filling

DoF:

• M Tx antennas, N Rx antennas

⇒ H ∈ CN×M ⇒ rank(H) = min{M,N}⇒ C ≈ min{M,N} log(P ) at high P

• DoF: d = limP→∞C

log(P )= rank(H) ⇒ d = min{M,N}

⇒ Capacity equivalent to that of d parallel SISO P2P channels!

Tx RxTx RxTx Rx (more on that later)

Anas Chaaban, Aydin Sezgin Multi-way Communications 12

RUB Institute of Digital Communication Systems

Capacity to DoF

Single-user (MIMO P2P):

Tx RxH

Capacity:

C = log |I+HQHH |

Optimization: water-filling

DoF:

• M Tx antennas, N Rx antennas

⇒ H ∈ CN×M ⇒ rank(H) = min{M,N}⇒ C ≈ min{M,N} log(P ) at high P

• DoF: d = limP→∞C

log(P )= rank(H) ⇒ d = min{M,N}

⇒ Capacity equivalent to that of d parallel SISO P2P channels!

Tx RxTx RxTx Rx (more on that later)

Anas Chaaban, Aydin Sezgin Multi-way Communications 12

RUB Institute of Digital Communication Systems

Capacity to DoF

Single-user (MIMO P2P):

Tx RxH

Capacity:

C = log |I+HQHH |

Optimization: water-filling

DoF:

• M Tx antennas, N Rx antennas

⇒ H ∈ CN×M ⇒ rank(H) = min{M,N}⇒ C ≈ min{M,N} log(P ) at high P

• DoF: d = limP→∞C

log(P )= rank(H) ⇒ d = min{M,N}

⇒ Capacity equivalent to that of d parallel SISO P2P channels!

Tx RxTx RxTx Rx (more on that later)

Anas Chaaban, Aydin Sezgin Multi-way Communications 12

RUB Institute of Digital Communication Systems

Capacity to DoF

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Capacity region:

Ri ≤ log |I+HiQiHHi |

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

Optimization: Iterative water-filling

DoF:

• Mi Tx antennas, N Rx antennas

⇒ Hi ∈ CN×Mi ⇒ rank(Hi) = min{Mi, N}• CΣ ≈ min{M1 +M2, N} log(P ) at high P

• DoF: di = limP→∞Ri

log(P )

⇒ DoF region: di ≤ rank(Hi), d1 + d2 ≤ rank([H1, H2])

⇒ d1 + d2 = min{M1 +M2, N}⇒ Sum-capacity equivalent to that of d1 + d2 parallel SISO P2P channels!

Anas Chaaban, Aydin Sezgin Multi-way Communications 13

RUB Institute of Digital Communication Systems

Capacity to DoF

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Capacity region:

Ri ≤ log |I+HiQiHHi |

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

Optimization: Iterative water-filling

DoF:

• Mi Tx antennas, N Rx antennas

⇒ Hi ∈ CN×Mi ⇒ rank(Hi) = min{Mi, N}• CΣ ≈ min{M1 +M2, N} log(P ) at high P

• DoF: di = limP→∞Ri

log(P )

⇒ DoF region: di ≤ rank(Hi), d1 + d2 ≤ rank([H1, H2])

⇒ d1 + d2 = min{M1 +M2, N}⇒ Sum-capacity equivalent to that of d1 + d2 parallel SISO P2P channels!

Anas Chaaban, Aydin Sezgin Multi-way Communications 13

RUB Institute of Digital Communication Systems

Capacity to DoF

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Capacity region:

Ri ≤ log |I+HiQiHHi |

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

Optimization: Iterative water-filling

DoF:

• Mi Tx antennas, N Rx antennas

⇒ Hi ∈ CN×Mi ⇒ rank(Hi) = min{Mi, N}• CΣ ≈ min{M1 +M2, N} log(P ) at high P

• DoF: di = limP→∞Ri

log(P )

⇒ DoF region: di ≤ rank(Hi), d1 + d2 ≤ rank([H1, H2])

⇒ d1 + d2 = min{M1 +M2, N}⇒ Sum-capacity equivalent to that of d1 + d2 parallel SISO P2P channels!

Anas Chaaban, Aydin Sezgin Multi-way Communications 13

RUB Institute of Digital Communication Systems

Capacity to DoF

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Capacity region:

Ri ≤ log |I+HiQiHHi |

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

Optimization: Iterative water-filling

DoF:

• Mi Tx antennas, N Rx antennas

⇒ Hi ∈ CN×Mi ⇒ rank(Hi) = min{Mi, N}• CΣ ≈ min{M1 +M2, N} log(P ) at high P

• DoF: di = limP→∞Ri

log(P )

⇒ DoF region: di ≤ rank(Hi), d1 + d2 ≤ rank([H1, H2])

⇒ d1 + d2 = min{M1 +M2, N}⇒ Sum-capacity equivalent to that of d1 + d2 parallel SISO P2P channels!

Anas Chaaban, Aydin Sezgin Multi-way Communications 13

RUB Institute of Digital Communication Systems

Capacity to DoF

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Capacity region:

Ri ≤ log |I+HiQiHHi |

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

Optimization: Iterative water-filling

DoF:

• Mi Tx antennas, N Rx antennas

⇒ Hi ∈ CN×Mi ⇒ rank(Hi) = min{Mi, N}

• CΣ ≈ min{M1 +M2, N} log(P ) at high P

• DoF: di = limP→∞Ri

log(P )

⇒ DoF region: di ≤ rank(Hi), d1 + d2 ≤ rank([H1, H2])

⇒ d1 + d2 = min{M1 +M2, N}⇒ Sum-capacity equivalent to that of d1 + d2 parallel SISO P2P channels!

Anas Chaaban, Aydin Sezgin Multi-way Communications 13

RUB Institute of Digital Communication Systems

Capacity to DoF

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Capacity region:

Ri ≤ log |I+HiQiHHi |

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

Optimization: Iterative water-filling

DoF:

• Mi Tx antennas, N Rx antennas

⇒ Hi ∈ CN×Mi ⇒ rank(Hi) = min{Mi, N}• CΣ ≈ min{M1 +M2, N} log(P ) at high P

• DoF: di = limP→∞Ri

log(P )

⇒ DoF region: di ≤ rank(Hi), d1 + d2 ≤ rank([H1, H2])

⇒ d1 + d2 = min{M1 +M2, N}⇒ Sum-capacity equivalent to that of d1 + d2 parallel SISO P2P channels!

Anas Chaaban, Aydin Sezgin Multi-way Communications 13

RUB Institute of Digital Communication Systems

Capacity to DoF

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Capacity region:

Ri ≤ log |I+HiQiHHi |

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

Optimization: Iterative water-filling

DoF:

• Mi Tx antennas, N Rx antennas

⇒ Hi ∈ CN×Mi ⇒ rank(Hi) = min{Mi, N}• CΣ ≈ min{M1 +M2, N} log(P ) at high P

• DoF: di = limP→∞Ri

log(P )

⇒ DoF region: di ≤ rank(Hi), d1 + d2 ≤ rank([H1, H2])

⇒ d1 + d2 = min{M1 +M2, N}⇒ Sum-capacity equivalent to that of d1 + d2 parallel SISO P2P channels!

Anas Chaaban, Aydin Sezgin Multi-way Communications 13

RUB Institute of Digital Communication Systems

Capacity to DoF

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Capacity region:

Ri ≤ log |I+HiQiHHi |

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

Optimization: Iterative water-filling

DoF:

• Mi Tx antennas, N Rx antennas

⇒ Hi ∈ CN×Mi ⇒ rank(Hi) = min{Mi, N}• CΣ ≈ min{M1 +M2, N} log(P ) at high P

• DoF: di = limP→∞Ri

log(P )

⇒ DoF region: di ≤ rank(Hi), d1 + d2 ≤ rank([H1, H2])

⇒ d1 + d2 = min{M1 +M2, N}⇒ Sum-capacity equivalent to that of d1 + d2 parallel SISO P2P channels!

Anas Chaaban, Aydin Sezgin Multi-way Communications 13

RUB Institute of Digital Communication Systems

Capacity to DoF

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Capacity region:

Ri ≤ log |I+HiQiHHi |

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

Optimization: Iterative water-filling

DoF:

• Mi Tx antennas, N Rx antennas

⇒ Hi ∈ CN×Mi ⇒ rank(Hi) = min{Mi, N}• CΣ ≈ min{M1 +M2, N} log(P ) at high P

• DoF: di = limP→∞Ri

log(P )

⇒ DoF region: di ≤ rank(Hi), d1 + d2 ≤ rank([H1, H2])

⇒ d1 + d2 = min{M1 +M2, N}

⇒ Sum-capacity equivalent to that of d1 + d2 parallel SISO P2P channels!

Anas Chaaban, Aydin Sezgin Multi-way Communications 13

RUB Institute of Digital Communication Systems

Capacity to DoF

Multi-user (MIMO MAC):

Tx 1

Tx 2

Rx

H1

H2

Capacity region:

Ri ≤ log |I+HiQiHHi |

R1 +R2 ≤ log |I+H1Q1HH1 +H2Q2H

H2 |

Optimization: Iterative water-filling

DoF:

• Mi Tx antennas, N Rx antennas

⇒ Hi ∈ CN×Mi ⇒ rank(Hi) = min{Mi, N}• CΣ ≈ min{M1 +M2, N} log(P ) at high P

• DoF: di = limP→∞Ri

log(P )

⇒ DoF region: di ≤ rank(Hi), d1 + d2 ≤ rank([H1, H2])

⇒ d1 + d2 = min{M1 +M2, N}⇒ Sum-capacity equivalent to that of d1 + d2 parallel SISO P2P channels!

Anas Chaaban, Aydin Sezgin Multi-way Communications 13

RUB Institute of Digital Communication Systems

Back to the Y-channel

Definition

Rij : Rate of signal from i to j

dij : DoF defined as limP→∞Rij

log(P )

dΣ: sum-DoF =∑

dij

D: Set of simultaneously achievableDoF’s dij

d13

d12

?

??

?

+ dΣ is an overall performance metric for a network

− dΣ doesn’t provide insights on the trade-off between individual DoF’s

Goal

Find the DoF region of the MIMO Y-channel.

Anas Chaaban, Aydin Sezgin Multi-way Communications 14

RUB Institute of Digital Communication Systems

Back to the Y-channel

Definition

Rij : Rate of signal from i to j

dij : DoF defined as limP→∞Rij

log(P )

dΣ: sum-DoF =∑

dij

D: Set of simultaneously achievableDoF’s dij

d13

d12

?

??

?

+ dΣ is an overall performance metric for a network

− dΣ doesn’t provide insights on the trade-off between individual DoF’s

Goal

Find the DoF region of the MIMO Y-channel.

Anas Chaaban, Aydin Sezgin Multi-way Communications 14

RUB Institute of Digital Communication Systems

Back to the Y-channel

Definition

Rij : Rate of signal from i to j

dij : DoF defined as limP→∞Rij

log(P )

dΣ: sum-DoF =∑

dij

D: Set of simultaneously achievableDoF’s dij

d13

d12

?

??

?

+ dΣ is an overall performance metric for a network

− dΣ doesn’t provide insights on the trade-off between individual DoF’s

Goal

Find the DoF region of the MIMO Y-channel.

Anas Chaaban, Aydin Sezgin Multi-way Communications 14

RUB Institute of Digital Communication Systems

Back to the Y-channel

Definition

Rij : Rate of signal from i to j

dij : DoF defined as limP→∞Rij

log(P )

dΣ: sum-DoF =∑

dij

D: Set of simultaneously achievableDoF’s dij

d13

d12

?

??

?

+ dΣ is an overall performance metric for a network

− dΣ doesn’t provide insights on the trade-off between individual DoF’s

Goal

Find the DoF region of the MIMO Y-channel.

Anas Chaaban, Aydin Sezgin Multi-way Communications 14

RUB Institute of Digital Communication Systems

Back to the Y-channel

Definition

Rij : Rate of signal from i to j

dij : DoF defined as limP→∞Rij

log(P )

dΣ: sum-DoF =∑

dij

D: Set of simultaneously achievableDoF’s dij

d13

d12dΣ

?

??

?

+ dΣ is an overall performance metric for a network

− dΣ doesn’t provide insights on the trade-off between individual DoF’s

Goal

Find the DoF region of the MIMO Y-channel.

Anas Chaaban, Aydin Sezgin Multi-way Communications 14

RUB Institute of Digital Communication Systems

Back to the Y-channel

Definition

Rij : Rate of signal from i to j

dij : DoF defined as limP→∞Rij

log(P )

dΣ: sum-DoF =∑

dij

D: Set of simultaneously achievableDoF’s dij

d13

d12dΣ

?

??

?

+ dΣ is an overall performance metric for a network

− dΣ doesn’t provide insights on the trade-off between individual DoF’s

Goal

Find the DoF region of the MIMO Y-channel.

Anas Chaaban, Aydin Sezgin Multi-way Communications 14

RUB Institute of Digital Communication Systems

Back to the Y-channel

Definition

Rij : Rate of signal from i to j

dij : DoF defined as limP→∞Rij

log(P )

dΣ: sum-DoF =∑

dij

D: Set of simultaneously achievableDoF’s dij

d13

d12dΣ

?

??

?

+ dΣ is an overall performance metric for a network

− dΣ doesn’t provide insights on the trade-off between individual DoF’s

Goal

Find the DoF region of the MIMO Y-channel.

Anas Chaaban, Aydin Sezgin Multi-way Communications 14

RUB Institute of Digital Communication Systems

Back to the Y-channel

Definition

Rij : Rate of signal from i to j

dij : DoF defined as limP→∞Rij

log(P )

dΣ: sum-DoF =∑

dij

D: Set of simultaneously achievableDoF’s dij

d13

d12dΣ

?

??

?

+ dΣ is an overall performance metric for a network

− dΣ doesn’t provide insights on the trade-off between individual DoF’s

Goal

Find the DoF region of the MIMO Y-channel.

Anas Chaaban, Aydin Sezgin Multi-way Communications 14

RUB Institute of Digital Communication Systems

Back to the Y-channel

Definition

Rij : Rate of signal from i to j

dij : DoF defined as limP→∞Rij

log(P )

dΣ: sum-DoF =∑

dij

D: Set of simultaneously achievableDoF’s dij

d13

d12dΣ

?

??

?

+ dΣ is an overall performance metric for a network

− dΣ doesn’t provide insights on the trade-off between individual DoF’s

Goal

Find the DoF region of the MIMO Y-channel.

Anas Chaaban, Aydin Sezgin Multi-way Communications 14

RUB Institute of Digital Communication Systems

Main result

The DoF region of a 3-user MIMO Y-channel with N ≤M is described by

d12 + d13 + d23 ≤ N

d12 + d13 + d32 ≤ N

......

... ≤...

Theorem (DoF region)

DoF region for N ≤M described by

dp1p2 + dp1p3 + dp2p3 ≤ N, ∀p

where p is a permutation of (1, 2, 3) and pi is its i-th component.

Anas Chaaban, Aydin Sezgin Multi-way Communications 15

RUB Institute of Digital Communication Systems

Main result

The DoF region of a 3-user MIMO Y-channel with N ≤M is described by

d12 + d13 + d23 ≤ N

d12 + d13 + d32 ≤ N

......

... ≤...

Theorem (DoF region)

DoF region for N ≤M described by

dp1p2 + dp1p3 + dp2p3 ≤ N, ∀p

where p is a permutation of (1, 2, 3) and pi is its i-th component.

Anas Chaaban, Aydin Sezgin Multi-way Communications 15

RUB Institute of Digital Communication Systems

Main result

The DoF region of a 3-user MIMO Y-channel with N ≤M is described by

d12 + d13 + d23 ≤ N

d12 + d13 + d32 ≤ N

......

... ≤...

Theorem (DoF region)

DoF region for N ≤M described by

dp1p2 + dp1p3 + dp2p3 ≤ N, ∀p

where p is a permutation of (1, 2, 3) and pi is its i-th component.

Anas Chaaban, Aydin Sezgin Multi-way Communications 15

RUB Institute of Digital Communication Systems

Main result

The DoF region of a 3-user MIMO Y-channel with N ≤M is described by

d12 + d13 + d23 ≤ N

d12 + d13 + d32 ≤ N

......

... ≤...

Theorem (DoF region)

DoF region for N ≤M described by

dp1p2 + dp1p3 + dp2p3 ≤ N, ∀p

where p is a permutation of (1, 2, 3) and pi is its i-th component.

Anas Chaaban, Aydin Sezgin Multi-way Communications 15

RUB Institute of Digital Communication Systems

Outline

1 Motivation: From one-way to multi-way

2 The MIMO Y-channel: From single-antenna to multiple-antennas

3 Main result: From capacity to DoF

4 Insights and ingredientsChannel diagonalization: Separability of communicationstructureAlignment, Compute-and-forwardTransmission strategy(3-users)

5 Extensions and Conclusion

Anas Chaaban, Aydin Sezgin Multi-way Communications 16

RUB Institute of Digital Communication Systems

Insight from the upper bounds

Message-flow-graph for upperbounds:

dp1p2 + dp1p3 + dp2p3 ≤ N

Message-flow-graph ford = (2, 0, 1, 1, 1, 0) ∈ D:

Message-flow-graph for anyd ∈ D:

p1 p2 p3

dp1p2 dp2p3

dp1p3

No cycles!

1 2 32

1

1

1

2-cycles and 3-cycles!

1 2 3d12

d21

d23

d32

d13

d31

2-cycles and 3-cycles!

Optimal strategy should be able to ‘resolve’ cycles!

Anas Chaaban, Aydin Sezgin Multi-way Communications 17

RUB Institute of Digital Communication Systems

Insight from the upper bounds

Message-flow-graph for upperbounds:

dp1p2 + dp1p3 + dp2p3 ≤ N

Message-flow-graph ford = (2, 0, 1, 1, 1, 0) ∈ D:

Message-flow-graph for anyd ∈ D:

p1 p2 p3

dp1p2 dp2p3

dp1p3

No cycles!

1 2 32

1

1

1

2-cycles and 3-cycles!

1 2 3d12

d21

d23

d32

d13

d31

2-cycles and 3-cycles!

Optimal strategy should be able to ‘resolve’ cycles!

Anas Chaaban, Aydin Sezgin Multi-way Communications 17

RUB Institute of Digital Communication Systems

Insight from the upper bounds

Message-flow-graph for upperbounds:

dp1p2 + dp1p3 + dp2p3 ≤ N

Message-flow-graph ford = (2, 0, 1, 1, 1, 0) ∈ D:

Message-flow-graph for anyd ∈ D:

p1 p2 p3

dp1p2 dp2p3

dp1p3

No cycles!

1 2 32

1

1

1

2-cycles and 3-cycles!

1 2 3d12

d21

d23

d32

d13

d31

2-cycles and 3-cycles!

Optimal strategy should be able to ‘resolve’ cycles!

Anas Chaaban, Aydin Sezgin Multi-way Communications 17

RUB Institute of Digital Communication Systems

Insight from the upper bounds

Message-flow-graph for upperbounds:

dp1p2 + dp1p3 + dp2p3 ≤ N

Message-flow-graph ford = (2, 0, 1, 1, 1, 0) ∈ D:

Message-flow-graph for anyd ∈ D:

p1 p2 p3

dp1p2 dp2p3

dp1p3

No cycles!

1 2 32

1

1

1

2-cycles and 3-cycles!

1 2 3d12

d21

d23

d32

d13

d31

2-cycles and 3-cycles!

Optimal strategy should be able to ‘resolve’ cycles!

Anas Chaaban, Aydin Sezgin Multi-way Communications 17

RUB Institute of Digital Communication Systems

Insight from the upper bounds

Message-flow-graph for upperbounds:

dp1p2 + dp1p3 + dp2p3 ≤ N

Message-flow-graph ford = (2, 0, 1, 1, 1, 0) ∈ D:

Message-flow-graph for anyd ∈ D:

p1 p2 p3

dp1p2 dp2p3

dp1p3No cycles!

1 2 32

1

1

1

2-cycles and 3-cycles!

1 2 3d12

d21

d23

d32

d13

d31

2-cycles and 3-cycles!

Optimal strategy should be able to ‘resolve’ cycles!

Anas Chaaban, Aydin Sezgin Multi-way Communications 17

RUB Institute of Digital Communication Systems

Insight from the upper bounds

Message-flow-graph for upperbounds:

dp1p2 + dp1p3 + dp2p3 ≤ N

Message-flow-graph ford = (2, 0, 1, 1, 1, 0) ∈ D:

Message-flow-graph for anyd ∈ D:

p1 p2 p3

dp1p2 dp2p3

dp1p3No cycles!

1 2 32

1

1

1

2-cycles and 3-cycles!

1 2 3d12

d21

d23

d32

d13

d31

2-cycles and 3-cycles!

Optimal strategy should be able to ‘resolve’ cycles!Anas Chaaban, Aydin Sezgin Multi-way Communications 17

RUB Institute of Digital Communication Systems

Outline

1 Motivation: From one-way to multi-way

2 The MIMO Y-channel: From single-antenna to multiple-antennas

3 Main result: From capacity to DoF

4 Insights and ingredientsChannel diagonalization: Separability of communicationstructureAlignment, Compute-and-forwardTransmission strategy(3-users)

5 Extensions and Conclusion

Anas Chaaban, Aydin Sezgin Multi-way Communications 18

RUB Institute of Digital Communication Systems

Uplink-downlink Separability

Node 1 Node 2Channel

x1

x2y1

y2m1

m2m2

m1

Two-way channels are in general not separable

⇒ uplink and downlink have to be considered jointly

⇒ xi and yi are dependent

Node 1 Node 2

+

+

Tx 1

Tx 2Rx 1

Rx 2

x1

x2y1

y2

n1n1

n2n2

m1

m2m2

m1

Basic Gaussian two-way channel is separable

⇒ uplink and downlink can be considered separately

⇒ xi and yi are independent

Anas Chaaban, Aydin Sezgin Multi-way Communications 19

RUB Institute of Digital Communication Systems

Uplink-downlink Separability

Node 1 Node 2Channel

x1

x2y1

y2m1

m2m2

m1

Two-way channels are in general not separable

⇒ uplink and downlink have to be considered jointly

⇒ xi and yi are dependent

Node 1 Node 2

+

+

Tx 1

Tx 2Rx 1

Rx 2

x1

x2y1

y2

n1n1

n2n2

m1

m2m2

m1

Basic Gaussian two-way channel is separable

⇒ uplink and downlink can be considered separately

⇒ xi and yi are independent

Anas Chaaban, Aydin Sezgin Multi-way Communications 19

RUB Institute of Digital Communication Systems

Uplink-downlink Separability

Node 1 Node 2Channel

x1

x2y1

y2m1

m2m2

m1

Two-way channels are in general not separable

⇒ uplink and downlink have to be considered jointly

⇒ xi and yi are dependent

Node 1 Node 2

+

+

Tx 1

Tx 2Rx 1

Rx 2

x1

x2y1

y2

n1n1

n2n2

m1

m2m2

m1

Basic Gaussian two-way channel is separable

⇒ uplink and downlink can be considered separately

⇒ xi and yi are independent

Anas Chaaban, Aydin Sezgin Multi-way Communications 19

RUB Institute of Digital Communication Systems

Uplink-downlink Separability

Node 1 Node 2Channel

x1

x2y1

y2m1

m2m2

m1

Two-way channels are in general not separable

⇒ uplink and downlink have to be considered jointly

⇒ xi and yi are dependent

Node 1 Node 2+

+

Tx 1

Tx 2Rx 1

Rx 2

x1

x2y1

y2

n1n1

n2n2

m1

m2m2

m1

Basic Gaussian two-way channel is separable

⇒ uplink and downlink can be considered separately

⇒ xi and yi are independent

Anas Chaaban, Aydin Sezgin Multi-way Communications 19

RUB Institute of Digital Communication Systems

Uplink-downlink Separability

Node 1 Node 2Channel

x1

x2y1

y2m1

m2m2

m1

Two-way channels are in general not separable

⇒ uplink and downlink have to be considered jointly

⇒ xi and yi are dependent

Node 1 Node 2

+

+

Tx 1

Tx 2Rx 1

Rx 2x1

x2y1

y2

n1n1

n2n2

m1

m2m2

m1

Basic Gaussian two-way channel is separable

⇒ uplink and downlink can be considered separately

⇒ xi and yi are independent

Anas Chaaban, Aydin Sezgin Multi-way Communications 19

RUB Institute of Digital Communication Systems

Uplink-downlink Separability

Node 1 Node 2Channel

x1

x2y1

y2m1

m2m2

m1

Two-way channels are in general not separable

⇒ uplink and downlink have to be considered jointly

⇒ xi and yi are dependent

Node 1 Node 2

+

+

Tx 1

Tx 2Rx 1

Rx 2x1

x2y1

y2

n1n1

n2n2

m1

m2m2

m1

Basic Gaussian two-way channel is separable

⇒ uplink and downlink can be considered separately

⇒ xi and yi are independent

Anas Chaaban, Aydin Sezgin Multi-way Communications 19

RUB Institute of Digital Communication Systems

Channel Diagonalization

Definition (Channel diagonlization)

Transform an arbitrary MIMO channel matrix H to a diagonal matrix.

MIMO M ×N P2P channel can be diagonalized by zero-forcing (ZF)

M ≥ N : ZF pre-coding:

• Pseudo-inverse:H† = HH [HHH ]−1

• H†H = I

Tx

Rx

x yH†

uu

H†

N parallel SISO P2P channels!

M ≤ N : ZF post-coding:

• Pseudo-inverse:H‡ = [HHH]−1HH

• HH‡ = I

TxRx

x yH‡

yyH‡

M parallel SISO P2P channels!

DoF achievable by treating each sub-channel separately ⇒ Separability!MAC and BC are also separable

Anas Chaaban, Aydin Sezgin Multi-way Communications 20

RUB Institute of Digital Communication Systems

Channel Diagonalization

Definition (Channel diagonlization)

Transform an arbitrary MIMO channel matrix H to a diagonal matrix.

MIMO M ×N P2P channel can be diagonalized by zero-forcing (ZF)

M ≥ N : ZF pre-coding:

• Pseudo-inverse:H† = HH [HHH ]−1

• H†H = I

TxRx

x yH

H†uu

H†

N parallel SISO P2P channels!M ≤ N : ZF post-coding:

• Pseudo-inverse:H‡ = [HHH]−1HH

• HH‡ = I

TxRx

x yH‡

yyH‡

M parallel SISO P2P channels!

DoF achievable by treating each sub-channel separately ⇒ Separability!MAC and BC are also separable

Anas Chaaban, Aydin Sezgin Multi-way Communications 20

RUB Institute of Digital Communication Systems

Channel Diagonalization

Definition (Channel diagonlization)

Transform an arbitrary MIMO channel matrix H to a diagonal matrix.

MIMO M ×N P2P channel can be diagonalized by zero-forcing (ZF)

M ≥ N : ZF pre-coding:

• Pseudo-inverse:H† = HH [HHH ]−1

• H†H = I

TxRx

x yH

H†u

u

H†

N parallel SISO P2P channels!M ≤ N : ZF post-coding:

• Pseudo-inverse:H‡ = [HHH]−1HH

• HH‡ = I

TxRx

x yH‡

yyH‡

M parallel SISO P2P channels!

DoF achievable by treating each sub-channel separately ⇒ Separability!MAC and BC are also separable

Anas Chaaban, Aydin Sezgin Multi-way Communications 20

RUB Institute of Digital Communication Systems

Channel Diagonalization

Definition (Channel diagonlization)

Transform an arbitrary MIMO channel matrix H to a diagonal matrix.

MIMO M ×N P2P channel can be diagonalized by zero-forcing (ZF)

M ≥ N : ZF pre-coding:

• Pseudo-inverse:H† = HH [HHH ]−1

• H†H = I

TxRx

x

yH

H†u

u

H†

N parallel SISO P2P channels!M ≤ N : ZF post-coding:

• Pseudo-inverse:H‡ = [HHH]−1HH

• HH‡ = I

TxRx

x yH‡

yyH‡

M parallel SISO P2P channels!

DoF achievable by treating each sub-channel separately ⇒ Separability!MAC and BC are also separable

Anas Chaaban, Aydin Sezgin Multi-way Communications 20

RUB Institute of Digital Communication Systems

Channel Diagonalization

Definition (Channel diagonlization)

Transform an arbitrary MIMO channel matrix H to a diagonal matrix.

MIMO M ×N P2P channel can be diagonalized by zero-forcing (ZF)

M ≥ N : ZF pre-coding:

• Pseudo-inverse:H† = HH [HHH ]−1

• H†H = I

Tx

Rx

x

y

H†u

u

H†

N parallel SISO P2P channels!

M ≤ N : ZF post-coding:

• Pseudo-inverse:H‡ = [HHH]−1HH

• HH‡ = I

TxRx

x yH‡

yyH‡

M parallel SISO P2P channels!

DoF achievable by treating each sub-channel separately ⇒ Separability!MAC and BC are also separable

Anas Chaaban, Aydin Sezgin Multi-way Communications 20

RUB Institute of Digital Communication Systems

Channel Diagonalization

Definition (Channel diagonlization)

Transform an arbitrary MIMO channel matrix H to a diagonal matrix.

MIMO M ×N P2P channel can be diagonalized by zero-forcing (ZF)

M ≥ N : ZF pre-coding:

• Pseudo-inverse:H† = HH [HHH ]−1

• H†H = I

Tx

Rx

x

y

H†u

u

H†

N parallel SISO P2P channels!M ≤ N : ZF post-coding:

• Pseudo-inverse:H‡ = [HHH]−1HH

• HH‡ = I

TxRx

x yH

H‡yy

H‡

M parallel SISO P2P channels!

DoF achievable by treating each sub-channel separately ⇒ Separability!MAC and BC are also separable

Anas Chaaban, Aydin Sezgin Multi-way Communications 20

RUB Institute of Digital Communication Systems

Channel Diagonalization

Definition (Channel diagonlization)

Transform an arbitrary MIMO channel matrix H to a diagonal matrix.

MIMO M ×N P2P channel can be diagonalized by zero-forcing (ZF)

M ≥ N : ZF pre-coding:

• Pseudo-inverse:H† = HH [HHH ]−1

• H†H = I

Tx

Rx

x

y

H†u

u

H†

N parallel SISO P2P channels!M ≤ N : ZF post-coding:

• Pseudo-inverse:H‡ = [HHH]−1HH

• HH‡ = I

TxRx

x yH

H‡y

yH‡

M parallel SISO P2P channels!

DoF achievable by treating each sub-channel separately ⇒ Separability!MAC and BC are also separable

Anas Chaaban, Aydin Sezgin Multi-way Communications 20

RUB Institute of Digital Communication Systems

Channel Diagonalization

Definition (Channel diagonlization)

Transform an arbitrary MIMO channel matrix H to a diagonal matrix.

MIMO M ×N P2P channel can be diagonalized by zero-forcing (ZF)

M ≥ N : ZF pre-coding:

• Pseudo-inverse:H† = HH [HHH ]−1

• H†H = I

Tx

Rx

x

y

H†u

u

H†

N parallel SISO P2P channels!M ≤ N : ZF post-coding:

• Pseudo-inverse:H‡ = [HHH]−1HH

• HH‡ = I

TxRx

x

y

H

H‡y

yH‡

M parallel SISO P2P channels!

DoF achievable by treating each sub-channel separately ⇒ Separability!MAC and BC are also separable

Anas Chaaban, Aydin Sezgin Multi-way Communications 20

RUB Institute of Digital Communication Systems

Channel Diagonalization

Definition (Channel diagonlization)

Transform an arbitrary MIMO channel matrix H to a diagonal matrix.

MIMO M ×N P2P channel can be diagonalized by zero-forcing (ZF)

M ≥ N : ZF pre-coding:

• Pseudo-inverse:H† = HH [HHH ]−1

• H†H = I

Tx

Rx

x

y

H†u

u

H†

N parallel SISO P2P channels!M ≤ N : ZF post-coding:

• Pseudo-inverse:H‡ = [HHH]−1HH

• HH‡ = I

Tx

Rx

x

yH‡

y

y

H‡

M parallel SISO P2P channels!

DoF achievable by treating each sub-channel separately ⇒ Separability!MAC and BC are also separable

Anas Chaaban, Aydin Sezgin Multi-way Communications 20

RUB Institute of Digital Communication Systems

Channel Diagonalization

Definition (Channel diagonlization)

Transform an arbitrary MIMO channel matrix H to a diagonal matrix.

MIMO M ×N P2P channel can be diagonalized by zero-forcing (ZF)

M ≥ N : ZF pre-coding:

• Pseudo-inverse:H† = HH [HHH ]−1

• H†H = I

Tx

Rx

x

y

H†u

u

H†

N parallel SISO P2P channels!M ≤ N : ZF post-coding:

• Pseudo-inverse:H‡ = [HHH]−1HH

• HH‡ = I

Tx

Rx

x

yH‡

y

y

H‡

M parallel SISO P2P channels!

DoF achievable by treating each sub-channel separately ⇒ Separability!

MAC and BC are also separable

Anas Chaaban, Aydin Sezgin Multi-way Communications 20

RUB Institute of Digital Communication Systems

Channel Diagonalization

Definition (Channel diagonlization)

Transform an arbitrary MIMO channel matrix H to a diagonal matrix.

MIMO M ×N P2P channel can be diagonalized by zero-forcing (ZF)

M ≥ N : ZF pre-coding:

• Pseudo-inverse:H† = HH [HHH ]−1

• H†H = I

Tx

Rx

x

y

H†u

u

H†

N parallel SISO P2P channels!M ≤ N : ZF post-coding:

• Pseudo-inverse:H‡ = [HHH]−1HH

• HH‡ = I

Tx

Rx

x

yH‡

y

y

H‡

M parallel SISO P2P channels!

DoF achievable by treating each sub-channel separately ⇒ Separability!MAC and BC are also separable

Anas Chaaban, Aydin Sezgin Multi-way Communications 20

RUB Institute of Digital Communication Systems

Outline

1 Motivation: From one-way to multi-way

2 The MIMO Y-channel: From single-antenna to multiple-antennas

3 Main result: From capacity to DoF

4 Insights and ingredientsChannel diagonalization: Separability of communicationstructureAlignment, Compute-and-forwardTransmission strategy(3-users)

5 Extensions and Conclusion

Anas Chaaban, Aydin Sezgin Multi-way Communications 21

RUB Institute of Digital Communication Systems

Signal Alignment

Definition (Signal alignment)

Placing two signals x1 and x2 in signal space so that span(x1) = span(x2).

Two signals can be aligned bypre-coding:x1 = V1u1

x2 = V2u2

• V1, V2 arbitrary

• H1V1 = H2V2

Node 1

Node 2

Node 3

H1

H2

x1

x2

H1x1

H2x2

x1

x2

H1x1H2x2

Anas Chaaban, Aydin Sezgin Multi-way Communications 22

RUB Institute of Digital Communication Systems

Signal Alignment

Definition (Signal alignment)

Placing two signals x1 and x2 in signal space so that span(x1) = span(x2).

Two signals can be aligned bypre-coding:x1 = V1u1

x2 = V2u2

• V1, V2 arbitrary

• H1V1 = H2V2

Node 1

Node 2

Node 3

H1

H2

x1

x2

H1x1

H2x2

x1

x2

H1x1H2x2

Anas Chaaban, Aydin Sezgin Multi-way Communications 22

RUB Institute of Digital Communication Systems

Signal Alignment

Definition (Signal alignment)

Placing two signals x1 and x2 in signal space so that span(x1) = span(x2).

Two signals can be aligned bypre-coding:x1 = V1u1

x2 = V2u2

• V1, V2 arbitrary

• H1V1 = H2V2

Node 1

Node 2

Node 3

H1

H2

x1

x2

H1x1

H2x2

x1

x2

H1x1H2x2

Anas Chaaban, Aydin Sezgin Multi-way Communications 22

RUB Institute of Digital Communication Systems

Signal Alignment

Definition (Signal alignment)

Placing two signals x1 and x2 in signal space so that span(x1) = span(x2).

Two signals can be aligned bypre-coding:x1 = V1u1

x2 = V2u2

• V1, V2 arbitrary

• H1V1 = H2V2

Node 1

Node 2

Node 3

H1

H2

x1

x2

H1x1

H2x2

x1

x2

H1x1H2x2

Anas Chaaban, Aydin Sezgin Multi-way Communications 22

RUB Institute of Digital Communication Systems

Signal Alignment

Definition (Signal alignment)

Placing two signals x1 and x2 in signal space so that span(x1) = span(x2).

Two signals can be aligned bypre-coding:x1 = V1u1

x2 = V2u2

• V1, V2 arbitrary

• H1V1 = H2V2

Node 1

Node 2

Node 3

H1

H2

x1

x2

H1x1

H2x2

x1

x2

H1x1H2x2

Anas Chaaban, Aydin Sezgin Multi-way Communications 22

RUB Institute of Digital Communication Systems

Signal Alignment

Definition (Signal alignment)

Placing two signals x1 and x2 in signal space so that span(x1) = span(x2).

Two signals can be aligned bypre-coding:x1 = V1u1

x2 = V2u2

• V1, V2 arbitrary

• H1V1 = H2V2

Node 1

Node 2

Node 3

H1

H2

x1

x2

H1x1

H2x2

x1

x2

H1x1H2x2

Anas Chaaban, Aydin Sezgin Multi-way Communications 22

RUB Institute of Digital Communication Systems

Compute-forward

Definition (Compute-forward (CF))

Computing and forwarding a linear combination of two signals a1x1 + a2x2.

CF can be accomplished byusing lattice codes.

• Property: u1 and u2

lattice codes ⇒ u1 + u2

lattice code!

• x1 = V1u1,

x2 = V2u2,

H1V1 = H2V2 = [1, 1]T

(e.g.)

• receivey3 = (u1 +u2)[1, 1]T +n

• compute u1 + u2 from[1, 1]y3 = 2(u1+u2)+n′

0 1 2−1−2

u1u2 u1 + u2xx x

Node 1

Node 2

Node 3

H1

H2

x1

x2

H1x1

H2x2

Anas Chaaban, Aydin Sezgin Multi-way Communications 23

RUB Institute of Digital Communication Systems

Compute-forward

Definition (Compute-forward (CF))

Computing and forwarding a linear combination of two signals a1x1 + a2x2.

CF can be accomplished byusing lattice codes.

• Property: u1 and u2

lattice codes ⇒ u1 + u2

lattice code!

• x1 = V1u1,

x2 = V2u2,

H1V1 = H2V2 = [1, 1]T

(e.g.)

• receivey3 = (u1 +u2)[1, 1]T +n

• compute u1 + u2 from[1, 1]y3 = 2(u1+u2)+n′

0 1 2−1−2

u1u2 u1 + u2xx x

Node 1

Node 2

Node 3

H1

H2

x1

x2

H1x1

H2x2

Anas Chaaban, Aydin Sezgin Multi-way Communications 23

RUB Institute of Digital Communication Systems

Compute-forward

Definition (Compute-forward (CF))

Computing and forwarding a linear combination of two signals a1x1 + a2x2.

CF can be accomplished byusing lattice codes.

• Property: u1 and u2

lattice codes ⇒ u1 + u2

lattice code!

• x1 = V1u1,

x2 = V2u2,

H1V1 = H2V2 = [1, 1]T

(e.g.)

• receivey3 = (u1 +u2)[1, 1]T +n

• compute u1 + u2 from[1, 1]y3 = 2(u1+u2)+n′

0 1 2−1−2u1u2 u1 + u2xx x

Node 1

Node 2

Node 3

H1

H2

x1

x2

H1x1

H2x2

Anas Chaaban, Aydin Sezgin Multi-way Communications 23

RUB Institute of Digital Communication Systems

Compute-forward

Definition (Compute-forward (CF))

Computing and forwarding a linear combination of two signals a1x1 + a2x2.

CF can be accomplished byusing lattice codes.

• Property: u1 and u2

lattice codes ⇒ u1 + u2

lattice code!

• x1 = V1u1,

x2 = V2u2,

H1V1 = H2V2 = [1, 1]T

(e.g.)

• receivey3 = (u1 +u2)[1, 1]T +n

• compute u1 + u2 from[1, 1]y3 = 2(u1+u2)+n′

0 1 2−1−2u1u2 u1 + u2xx x

Node 1

Node 2

Node 3

H1

H2

x1

x2

H1x1

H2x2

Anas Chaaban, Aydin Sezgin Multi-way Communications 23

RUB Institute of Digital Communication Systems

Compute-forward

Definition (Compute-forward (CF))

Computing and forwarding a linear combination of two signals a1x1 + a2x2.

CF can be accomplished byusing lattice codes.

• Property: u1 and u2

lattice codes ⇒ u1 + u2

lattice code!

• x1 = V1u1,

x2 = V2u2,

H1V1 = H2V2 = [1, 1]T

(e.g.)

• receivey3 = (u1 +u2)[1, 1]T +n

• compute u1 + u2 from[1, 1]y3 = 2(u1+u2)+n′

0 1 2−1−2u1u2 u1 + u2xx x

Node 1

Node 2

Node 3

H1

H2

x1

x2

H1x1

H2x2

Anas Chaaban, Aydin Sezgin Multi-way Communications 23

RUB Institute of Digital Communication Systems

Compute-forward

Definition (Compute-forward (CF))

Computing and forwarding a linear combination of two signals a1x1 + a2x2.

CF can be accomplished byusing lattice codes.

• Property: u1 and u2

lattice codes ⇒ u1 + u2

lattice code!

• x1 = V1u1,

x2 = V2u2,

H1V1 = H2V2 = [1, 1]T

(e.g.)

• receivey3 = (u1 +u2)[1, 1]T +n

• compute u1 + u2 from[1, 1]y3 = 2(u1+u2)+n′

0 1 2−1−2u1u2 u1 + u2xx x

Node 1

Node 2

Node 3

H1

H2

x1

x2

H1x1

H2x2

Anas Chaaban, Aydin Sezgin Multi-way Communications 23

RUB Institute of Digital Communication Systems

Outline

1 Motivation: From one-way to multi-way

2 The MIMO Y-channel: From single-antenna to multiple-antennas

3 Main result: From capacity to DoF

4 Insights and ingredientsChannel diagonalization: Separability of communicationstructureAlignment, Compute-and-forwardTransmission strategy(3-users)

5 Extensions and Conclusion

Anas Chaaban, Aydin Sezgin Multi-way Communications 24

RUB Institute of Digital Communication Systems

Overview

Achievability of D is proved using:

Channel diagonalization:

MIMOY-channel

→N SISO

Y-channels(sub-channels)

U1

U3

R

U2

U1

U3

R

U2U1

U3

R

U2

Information exchange:

• Bi-directional: signal-alignment/compute-forward

• Cyclic: signal-alignment/compute-forward

• Uni-directional: decode-forward

Resource allocation: distribute sub-channels over users

Anas Chaaban, Aydin Sezgin Multi-way Communications 25

RUB Institute of Digital Communication Systems

Overview

Achievability of D is proved using:

Channel diagonalization:

MIMOY-channel

→N SISO

Y-channels(sub-channels)

U1

U3

R

U2

U1

U3

R

U2U1

U3

R

U2

Information exchange:

• Bi-directional: signal-alignment/compute-forward

• Cyclic: signal-alignment/compute-forward

• Uni-directional: decode-forward

Resource allocation: distribute sub-channels over users

Anas Chaaban, Aydin Sezgin Multi-way Communications 25

RUB Institute of Digital Communication Systems

Overview

Achievability of D is proved using:

Channel diagonalization:

MIMOY-channel

→N SISO

Y-channels(sub-channels)

U1

U3

R

U2U1

U3

R

U2U1

U3

R

U2

Information exchange:

• Bi-directional: signal-alignment/compute-forward

• Cyclic: signal-alignment/compute-forward

• Uni-directional: decode-forward

Resource allocation: distribute sub-channels over users

Anas Chaaban, Aydin Sezgin Multi-way Communications 25

RUB Institute of Digital Communication Systems

Overview

Achievability of D is proved using:

Channel diagonalization:

MIMOY-channel

→N SISO

Y-channels(sub-channels)

U1

U3

R

U2U1

U3

R

U2U1

U3

R

U2

Information exchange:

• Bi-directional: signal-alignment/compute-forward

• Cyclic: signal-alignment/compute-forward

• Uni-directional: decode-forward

Resource allocation: distribute sub-channels over users

Anas Chaaban, Aydin Sezgin Multi-way Communications 25

RUB Institute of Digital Communication Systems

Overview

Achievability of D is proved using:

Channel diagonalization:

MIMOY-channel

→N SISO

Y-channels(sub-channels)

U1

U3

R

U2U1

U3

R

U2U1

U3

R

U2

Information exchange:

• Bi-directional: signal-alignment/compute-forward

• Cyclic: signal-alignment/compute-forward

• Uni-directional: decode-forward

Resource allocation: distribute sub-channels over users

Anas Chaaban, Aydin Sezgin Multi-way Communications 25

RUB Institute of Digital Communication Systems

Overview

Achievability of D is proved using:

Channel diagonalization:

MIMOY-channel

→N SISO

Y-channels(sub-channels)

U1

U3

R

U2U1

U3

R

U2U1

U3

R

U2

Information exchange:

• Bi-directional: signal-alignment/compute-forward

• Cyclic: signal-alignment/compute-forward

• Uni-directional: decode-forward

Resource allocation: distribute sub-channels over users

Anas Chaaban, Aydin Sezgin Multi-way Communications 25

RUB Institute of Digital Communication Systems

Overview

Achievability of D is proved using:

Channel diagonalization:

MIMOY-channel

→N SISO

Y-channels(sub-channels)

U1

U3

R

U2U1

U3

R

U2U1

U3

R

U2

Information exchange:

• Bi-directional: signal-alignment/compute-forward

• Cyclic: signal-alignment/compute-forward

• Uni-directional: decode-forward

Resource allocation: distribute sub-channels over users

Anas Chaaban, Aydin Sezgin Multi-way Communications 25

RUB Institute of Digital Communication Systems

Overview

Achievability of D is proved using:

Channel diagonalization:

MIMOY-channel

→N SISO

Y-channels(sub-channels)

U1

U3

R

U2U1

U3

R

U2U1

U3

R

U2

Information exchange:

• Bi-directional: signal-alignment/compute-forward

• Cyclic: signal-alignment/compute-forward

• Uni-directional: decode-forward

Resource allocation: distribute sub-channels over users

Anas Chaaban, Aydin Sezgin Multi-way Communications 25

RUB Institute of Digital Communication Systems

Channel Diagonalization

• a MIMO Y-channelwith M = N = 3

• actually looks likethis!

• Pre- and post-codeusing theMoore-Penrosepseudo inverse

• ChannelDiagonalization ⇒ Nsub-channels

User 1

User 2

Relay User 3

Uplink

x1

x2

x3

H1H1

H2H2

H3H3

User 1

User 2

Relay

xr

User 3

Downlink

y1

y2

y3

D1D1

D2D2

D3D3

Anas Chaaban, Aydin Sezgin Multi-way Communications 26

RUB Institute of Digital Communication Systems

Channel Diagonalization

• a MIMO Y-channelwith M = N = 3

• actually looks likethis!

• Pre- and post-codeusing theMoore-Penrosepseudo inverse

• ChannelDiagonalization ⇒ Nsub-channels

User 1

User 2

Relay User 3

Uplink

x1

x2

x3

H1

H2

H3

User 1

User 2

Relay

xr

User 3

Downlink

y1

y2

y3

D1

D2

D3

Anas Chaaban, Aydin Sezgin Multi-way Communications 26

RUB Institute of Digital Communication Systems

Channel Diagonalization

• a MIMO Y-channelwith M = N = 3

• actually looks likethis!

• Pre- and post-codeusing theMoore-Penrosepseudo inverse

• ChannelDiagonalization ⇒ Nsub-channels

User 1

User 2

Relay User 3

Uplink

x1

x2

x3

H1

H2

H3

HR1

u1

HR2

u2

HR3

u3

User 1

User 2

Relay

xr

User 3

Downlink

y1

y2

y3

D1

D2

D3

DL1

xr1

DL2

xr2

DL3

xr3

Anas Chaaban, Aydin Sezgin Multi-way Communications 26

RUB Institute of Digital Communication Systems

Channel Diagonalization

• a MIMO Y-channelwith M = N = 3

• actually looks likethis!

• Pre- and post-codeusing theMoore-Penrosepseudo inverse

• ChannelDiagonalization ⇒ Nsub-channels

User 1

User 2

Relay User 3

Uplink

x1

x2

x3

u1

u2

u3

User 1

User 2

Relay

xr

User 3

Downlink

y1

y2

y3

xr1

xr2

xr3

Anas Chaaban, Aydin Sezgin Multi-way Communications 26

RUB Institute of Digital Communication Systems

Information transfer

Bi-directional:

• signal-alignment

• compute-forward

• exchanges 2 symbols

• requires 1 sub-channel(up- and down-link)

• efficiency 2DoF/dimension

User 1

User 2

Relay User 3

Uplink

u12

u21

0

u12+u21

User 1

User 2

Relay User 3

Downlink

u21

u12

0

u12+u21

Anas Chaaban, Aydin Sezgin Multi-way Communications 27

RUB Institute of Digital Communication Systems

Information transfer

Cyclic:

• signal-alignment

• compute-forward

• exchanges 3 symbols

• requires 2 sub-channels(up- and down-link)

• efficiency 3/2DoF/dimension

User 1

User 2

Relay User 3

Uplink

u12

0

u23

u23

0

u31

u12+u23

u31+u23

User 1

User 2

Relay User 3

Downlink

u23

u31

u12

u31

u12

u23

u12+u23

u31+u23

Anas Chaaban, Aydin Sezgin Multi-way Communications 28

RUB Institute of Digital Communication Systems

Information transfer

Uni-directional:

• decode-forward

• exchanges 1 symbols

• requires 1 sub-channel(up- and down-link)

• efficiency 1DoF/dimension

User 1

User 2

Relay User 3

Uplink

u12

0

0

u12

User 1

User 2

Relay User 3

Downlink

0

u12

0

u12

Anas Chaaban, Aydin Sezgin Multi-way Communications 29

RUB Institute of Digital Communication Systems

Example

DoF tuple d = (d12, d13, d21, d23, d31, d32) = (2, 0, 1, 1, 1, 0), Y-channel with3 = N ≤M

Uni-directional only:

• 5 sub-channels > N !

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Bi-directional + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• uni-directional needs 3 more sub-channels

• total number of sub-channels 4 > N !

Bi-directional + cyclic + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• cyclic achieves dc12 = dc23 = dc31 = 1 over 2 sub-channels

• residual DoF (0, 0, 0, 0, 0, 0)

• total number of sub-channels 3 = N !

Anas Chaaban, Aydin Sezgin Multi-way Communications 30

RUB Institute of Digital Communication Systems

Example

DoF tuple d = (d12, d13, d21, d23, d31, d32) = (2, 0, 1, 1, 1, 0), Y-channel with3 = N ≤M

Uni-directional only:

• 5 sub-channels > N !

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Bi-directional + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• uni-directional needs 3 more sub-channels

• total number of sub-channels 4 > N !

Bi-directional + cyclic + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• cyclic achieves dc12 = dc23 = dc31 = 1 over 2 sub-channels

• residual DoF (0, 0, 0, 0, 0, 0)

• total number of sub-channels 3 = N !

Anas Chaaban, Aydin Sezgin Multi-way Communications 30

RUB Institute of Digital Communication Systems

Example

DoF tuple d = (d12, d13, d21, d23, d31, d32) = (2, 0, 1, 1, 1, 0), Y-channel with3 = N ≤M

Uni-directional only:

• 5 sub-channels > N !

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Bi-directional + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• uni-directional needs 3 more sub-channels

• total number of sub-channels 4 > N !

Bi-directional + cyclic + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• cyclic achieves dc12 = dc23 = dc31 = 1 over 2 sub-channels

• residual DoF (0, 0, 0, 0, 0, 0)

• total number of sub-channels 3 = N !

Anas Chaaban, Aydin Sezgin Multi-way Communications 30

RUB Institute of Digital Communication Systems

Example

DoF tuple d = (d12, d13, d21, d23, d31, d32) = (2, 0, 1, 1, 1, 0), Y-channel with3 = N ≤M

Uni-directional only:

• 5 sub-channels > N !

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Bi-directional + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• uni-directional needs 3 more sub-channels

• total number of sub-channels 4 > N !

Bi-directional + cyclic + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• cyclic achieves dc12 = dc23 = dc31 = 1 over 2 sub-channels

• residual DoF (0, 0, 0, 0, 0, 0)

• total number of sub-channels 3 = N !

Anas Chaaban, Aydin Sezgin Multi-way Communications 30

RUB Institute of Digital Communication Systems

Example

DoF tuple d = (d12, d13, d21, d23, d31, d32) = (2, 0, 1, 1, 1, 0), Y-channel with3 = N ≤M

Uni-directional only:

• 5 sub-channels > N !

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Bi-directional + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• uni-directional needs 3 more sub-channels

• total number of sub-channels 4 > N !

Bi-directional + cyclic + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• cyclic achieves dc12 = dc23 = dc31 = 1 over 2 sub-channels

• residual DoF (0, 0, 0, 0, 0, 0)

• total number of sub-channels 3 = N !

Anas Chaaban, Aydin Sezgin Multi-way Communications 30

RUB Institute of Digital Communication Systems

Example

DoF tuple d = (d12, d13, d21, d23, d31, d32) = (2, 0, 1, 1, 1, 0), Y-channel with3 = N ≤M

Uni-directional only:

• 5 sub-channels > N !

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Bi-directional + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• uni-directional needs 3 more sub-channels

• total number of sub-channels 4 > N !

Bi-directional + cyclic + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• cyclic achieves dc12 = dc23 = dc31 = 1 over 2 sub-channels

• residual DoF (0, 0, 0, 0, 0, 0)

• total number of sub-channels 3 = N !

Anas Chaaban, Aydin Sezgin Multi-way Communications 30

RUB Institute of Digital Communication Systems

Example

DoF tuple d = (d12, d13, d21, d23, d31, d32) = (2, 0, 1, 1, 1, 0), Y-channel with3 = N ≤M

Uni-directional only:

• 5 sub-channels > N !

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Bi-directional + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• uni-directional needs 3 more sub-channels

• total number of sub-channels 4 > N !

Bi-directional + cyclic + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• cyclic achieves dc12 = dc23 = dc31 = 1 over 2 sub-channels

• residual DoF (0, 0, 0, 0, 0, 0)

• total number of sub-channels 3 = N !

Anas Chaaban, Aydin Sezgin Multi-way Communications 30

RUB Institute of Digital Communication Systems

Example

DoF tuple d = (d12, d13, d21, d23, d31, d32) = (2, 0, 1, 1, 1, 0), Y-channel with3 = N ≤M

Uni-directional only:

• 5 sub-channels > N !

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Bi-directional + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• uni-directional needs 3 more sub-channels

• total number of sub-channels 4 > N !

Bi-directional + cyclic + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• cyclic achieves dc12 = dc23 = dc31 = 1 over 2 sub-channels

• residual DoF (0, 0, 0, 0, 0, 0)

• total number of sub-channels 3 = N !

Anas Chaaban, Aydin Sezgin Multi-way Communications 30

RUB Institute of Digital Communication Systems

Example

DoF tuple d = (d12, d13, d21, d23, d31, d32) = (2, 0, 1, 1, 1, 0), Y-channel with3 = N ≤M

Uni-directional only:

• 5 sub-channels > N !

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Bi-directional + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• uni-directional needs 3 more sub-channels

• total number of sub-channels 4 > N !

Bi-directional + cyclic + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• cyclic achieves dc12 = dc23 = dc31 = 1 over 2 sub-channels

• residual DoF (0, 0, 0, 0, 0, 0)

• total number of sub-channels 3 = N !

Anas Chaaban, Aydin Sezgin Multi-way Communications 30

RUB Institute of Digital Communication Systems

Example

DoF tuple d = (d12, d13, d21, d23, d31, d32) = (2, 0, 1, 1, 1, 0), Y-channel with3 = N ≤M

Uni-directional only:

• 5 sub-channels > N !

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Bi-directional + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• uni-directional needs 3 more sub-channels

• total number of sub-channels 4 > N !

Bi-directional + cyclic + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• cyclic achieves dc12 = dc23 = dc31 = 1 over 2 sub-channels

• residual DoF (0, 0, 0, 0, 0, 0)

• total number of sub-channels 3 = N !

Anas Chaaban, Aydin Sezgin Multi-way Communications 30

RUB Institute of Digital Communication Systems

Example

DoF tuple d = (d12, d13, d21, d23, d31, d32) = (2, 0, 1, 1, 1, 0), Y-channel with3 = N ≤M

Uni-directional only:

• 5 sub-channels > N !

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Bi-directional + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• uni-directional needs 3 more sub-channels

• total number of sub-channels 4 > N !

Bi-directional + cyclic + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• cyclic achieves dc12 = dc23 = dc31 = 1 over 2 sub-channels

• residual DoF (0, 0, 0, 0, 0, 0)

• total number of sub-channels 3 = N !

Anas Chaaban, Aydin Sezgin Multi-way Communications 30

RUB Institute of Digital Communication Systems

Example

DoF tuple d = (d12, d13, d21, d23, d31, d32) = (2, 0, 1, 1, 1, 0), Y-channel with3 = N ≤M

Uni-directional only:

• 5 sub-channels > N !

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Bi-directional + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• uni-directional needs 3 more sub-channels

• total number of sub-channels 4 > N !

Bi-directional + cyclic + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• cyclic achieves dc12 = dc23 = dc31 = 1 over 2 sub-channels

• residual DoF (0, 0, 0, 0, 0, 0)

• total number of sub-channels 3 = N !

Anas Chaaban, Aydin Sezgin Multi-way Communications 30

RUB Institute of Digital Communication Systems

Example

DoF tuple d = (d12, d13, d21, d23, d31, d32) = (2, 0, 1, 1, 1, 0), Y-channel with3 = N ≤M

Uni-directional only:

• 5 sub-channels > N !

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Bi-directional + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• uni-directional needs 3 more sub-channels

• total number of sub-channels 4 > N !

Bi-directional + cyclic + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• cyclic achieves dc12 = dc23 = dc31 = 1 over 2 sub-channels

• residual DoF (0, 0, 0, 0, 0, 0)

• total number of sub-channels 3 = N !

Anas Chaaban, Aydin Sezgin Multi-way Communications 30

RUB Institute of Digital Communication Systems

Example

DoF tuple d = (d12, d13, d21, d23, d31, d32) = (2, 0, 1, 1, 1, 0), Y-channel with3 = N ≤M

Uni-directional only:

• 5 sub-channels > N !

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Bi-directional + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• uni-directional needs 3 more sub-channels

• total number of sub-channels 4 > N !

Bi-directional + cyclic + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• cyclic achieves dc12 = dc23 = dc31 = 1 over 2 sub-channels

• residual DoF (0, 0, 0, 0, 0, 0)

• total number of sub-channels 3 = N !

Anas Chaaban, Aydin Sezgin Multi-way Communications 30

RUB Institute of Digital Communication Systems

Example

DoF tuple d = (d12, d13, d21, d23, d31, d32) = (2, 0, 1, 1, 1, 0), Y-channel with3 = N ≤M

Uni-directional only:

• 5 sub-channels > N !

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Bi-directional + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• uni-directional needs 3 more sub-channels

• total number of sub-channels 4 > N !

Bi-directional + cyclic + uni-directional:

• bi-directional achieves db12 = db21 = 1 over 1 sub-channel

• residual DoF (1, 0, 0, 1, 1, 0)

• cyclic achieves dc12 = dc23 = dc31 = 1 over 2 sub-channels

• residual DoF (0, 0, 0, 0, 0, 0)

• total number of sub-channels 3 = N !

Anas Chaaban, Aydin Sezgin Multi-way Communications 30

RUB Institute of Digital Communication Systems

Example

User 1

User 2

User 3

Relay

User 1

User 2

User 3

u12

u21

u12

u21

u12 + u21

u12 + u21

u12 + u21

v12

v23

v12

v23

v12 + v23

v12 + v23

v12 + v23

v23

v31

v31v23

v23 + v31

v23 + v31

v23 + v31

H1H1

H2H2

H3H3

D1D1

D2D2

D3D3

Anas Chaaban, Aydin Sezgin Multi-way Communications 31

RUB Institute of Digital Communication Systems

Example

User 1

User 2

User 3

Relay

User 1

User 2

User 3

u12

u21u12

u21

u12 + u21

u12 + u21

u12 + u21

v12

v23

v12

v23

v12 + v23

v12 + v23

v12 + v23

v23

v31

v31v23

v23 + v31

v23 + v31

v23 + v31

H1H1

H2H2

H3H3

D1D1

D2D2

D3D3

Anas Chaaban, Aydin Sezgin Multi-way Communications 31

RUB Institute of Digital Communication Systems

Example

User 1

User 2

User 3

Relay

User 1

User 2

User 3

u12

u21u12

u21

u12 + u21

u12 + u21

u12 + u21

v12

v23

v12

v23

v12 + v23

v12 + v23

v12 + v23

v23

v31

v31v23

v23 + v31

v23 + v31

v23 + v31

H1H1

H2H2

H3H3

D1D1

D2D2

D3D3

Anas Chaaban, Aydin Sezgin Multi-way Communications 31

RUB Institute of Digital Communication Systems

Example

User 1

User 2

User 3

Relay

User 1

User 2

User 3

u12

u21u12

u21

u12 + u21

u12 + u21

u12 + u21

v12

v23

v12

v23

v12 + v23

v12 + v23

v12 + v23

v23

v31

v31v23

v23 + v31

v23 + v31

v23 + v31

H1H1

H2H2

H3H3

D1D1

D2D2

D3D3

Anas Chaaban, Aydin Sezgin Multi-way Communications 31

RUB Institute of Digital Communication Systems

Example

User 1

User 2

User 3

Relay

User 1

User 2

User 3

u12

u21u12

u21

u12 + u21

u12 + u21

u12 + u21

v12

v23

v12

v23

v12 + v23

v12 + v23

v12 + v23

v23

v31

v31v23

v23 + v31

v23 + v31

v23 + v31

H1H1

H2H2

H3H3

D1D1

D2D2

D3D3

Anas Chaaban, Aydin Sezgin Multi-way Communications 31

RUB Institute of Digital Communication Systems

Example

User 1

User 2

User 3

Relay

User 1

User 2

User 3

u12

u21u12

u21

u12 + u21

u12 + u21

u12 + u21

v12

v23

v12

v23

v12 + v23

v12 + v23

v12 + v23

v23

v31

v31v23

v23 + v31

v23 + v31

v23 + v31

H1H1

H2H2

H3H3

D1D1

D2D2

D3D3

Anas Chaaban, Aydin Sezgin Multi-way Communications 31

RUB Institute of Digital Communication Systems

Resource allocation

Consider a DoF tuple d = (d12, d13, d21, d23, d31, d32)

⇒ 2-cycles and 3-cycles!

Bi-directional:1) set dbij = dbji = min{dij , dji}2) requires dbij sub-channels3) resolves 2-cycles4) residual DoF d′ij = dij − dbij

1 2 3

d12

d21

d21

d23

d32

d32

d13

d13

d31

d31

Residual DoF tuple (e.g.) d′ = (d′12, 0, 0, d′23, d

′31, 0) ⇒ 3-cycle!

Cyclic:

1) set dcij = dcjk = dcki = min{d′ij , d′jk, d′ki}2) requires 2dcij sub-channels3) resolves 3-cycles4) residual DoF d′′ij = d′ij − dcij

Uni-directional:

1) set duij = d′′ij

2) requires duij sub-channels

d achieved!

Anas Chaaban, Aydin Sezgin Multi-way Communications 32

RUB Institute of Digital Communication Systems

Resource allocation

Consider a DoF tuple d = (d12, d13, d21, d23, d31, d32) ⇒ 2-cycles and 3-cycles!

Bi-directional:1) set dbij = dbji = min{dij , dji}2) requires dbij sub-channels3) resolves 2-cycles4) residual DoF d′ij = dij − dbij

1 2 3

d12

d21

d21

d23

d32

d32

d13

d13

d31

d31

Residual DoF tuple (e.g.) d′ = (d′12, 0, 0, d′23, d

′31, 0) ⇒ 3-cycle!

Cyclic:

1) set dcij = dcjk = dcki = min{d′ij , d′jk, d′ki}2) requires 2dcij sub-channels3) resolves 3-cycles4) residual DoF d′′ij = d′ij − dcij

Uni-directional:

1) set duij = d′′ij

2) requires duij sub-channels

d achieved!

Anas Chaaban, Aydin Sezgin Multi-way Communications 32

RUB Institute of Digital Communication Systems

Resource allocation

Consider a DoF tuple d = (d12, d13, d21, d23, d31, d32) ⇒ 2-cycles and 3-cycles!

Bi-directional:1) set dbij = dbji = min{dij , dji}

2) requires dbij sub-channels3) resolves 2-cycles4) residual DoF d′ij = dij − dbij

1 2 3

d12

d21

d21

d23

d32

d32

d13

d13

d31

d31

Residual DoF tuple (e.g.) d′ = (d′12, 0, 0, d′23, d

′31, 0) ⇒ 3-cycle!

Cyclic:

1) set dcij = dcjk = dcki = min{d′ij , d′jk, d′ki}2) requires 2dcij sub-channels3) resolves 3-cycles4) residual DoF d′′ij = d′ij − dcij

Uni-directional:

1) set duij = d′′ij

2) requires duij sub-channels

d achieved!

Anas Chaaban, Aydin Sezgin Multi-way Communications 32

RUB Institute of Digital Communication Systems

Resource allocation

Consider a DoF tuple d = (d12, d13, d21, d23, d31, d32) ⇒ 2-cycles and 3-cycles!

Bi-directional:1) set dbij = dbji = min{dij , dji}2) requires dbij sub-channels

3) resolves 2-cycles4) residual DoF d′ij = dij − dbij

1 2 3

d12

d21

d21

d23

d32

d32

d13

d13

d31

d31

Residual DoF tuple (e.g.) d′ = (d′12, 0, 0, d′23, d

′31, 0) ⇒ 3-cycle!

Cyclic:

1) set dcij = dcjk = dcki = min{d′ij , d′jk, d′ki}2) requires 2dcij sub-channels3) resolves 3-cycles4) residual DoF d′′ij = d′ij − dcij

Uni-directional:

1) set duij = d′′ij

2) requires duij sub-channels

d achieved!

Anas Chaaban, Aydin Sezgin Multi-way Communications 32

RUB Institute of Digital Communication Systems

Resource allocation

Consider a DoF tuple d = (d12, d13, d21, d23, d31, d32) ⇒ 2-cycles and 3-cycles!

Bi-directional:1) set dbij = dbji = min{dij , dji}2) requires dbij sub-channels3) resolves 2-cycles

4) residual DoF d′ij = dij − dbij

1 2 3

d12

d21

d21

d23

d32

d32

d13

d13

d31

d31

Residual DoF tuple (e.g.) d′ = (d′12, 0, 0, d′23, d

′31, 0) ⇒ 3-cycle!

Cyclic:

1) set dcij = dcjk = dcki = min{d′ij , d′jk, d′ki}2) requires 2dcij sub-channels3) resolves 3-cycles4) residual DoF d′′ij = d′ij − dcij

Uni-directional:

1) set duij = d′′ij

2) requires duij sub-channels

d achieved!

Anas Chaaban, Aydin Sezgin Multi-way Communications 32

RUB Institute of Digital Communication Systems

Resource allocation

Consider a DoF tuple d = (d12, d13, d21, d23, d31, d32) ⇒ 2-cycles and 3-cycles!

Bi-directional:1) set dbij = dbji = min{dij , dji}2) requires dbij sub-channels3) resolves 2-cycles4) residual DoF d′ij = dij − dbij

1 2 3

d12

d21

d21

d23

d32

d32

d13

d13

d31

d31

Residual DoF tuple (e.g.) d′ = (d′12, 0, 0, d′23, d

′31, 0) ⇒ 3-cycle!

Cyclic:

1) set dcij = dcjk = dcki = min{d′ij , d′jk, d′ki}2) requires 2dcij sub-channels3) resolves 3-cycles4) residual DoF d′′ij = d′ij − dcij

Uni-directional:

1) set duij = d′′ij

2) requires duij sub-channels

d achieved!

Anas Chaaban, Aydin Sezgin Multi-way Communications 32

RUB Institute of Digital Communication Systems

Resource allocation

Consider a DoF tuple d = (d12, d13, d21, d23, d31, d32) ⇒ 2-cycles and 3-cycles!

Bi-directional:1) set dbij = dbji = min{dij , dji}2) requires dbij sub-channels3) resolves 2-cycles4) residual DoF d′ij = dij − dbij

1 2 3

d12

d21

d21

d23

d32

d32

d13

d13

d31

d31

Residual DoF tuple (e.g.) d′ = (d′12, 0, 0, d′23, d

′31, 0) ⇒ 3-cycle!

Cyclic:

1) set dcij = dcjk = dcki = min{d′ij , d′jk, d′ki}2) requires 2dcij sub-channels3) resolves 3-cycles4) residual DoF d′′ij = d′ij − dcij

Uni-directional:

1) set duij = d′′ij

2) requires duij sub-channels

d achieved!

Anas Chaaban, Aydin Sezgin Multi-way Communications 32

RUB Institute of Digital Communication Systems

Resource allocation

Consider a DoF tuple d = (d12, d13, d21, d23, d31, d32) ⇒ 2-cycles and 3-cycles!

Bi-directional:1) set dbij = dbji = min{dij , dji}2) requires dbij sub-channels3) resolves 2-cycles4) residual DoF d′ij = dij − dbij

1 2 3

d12

d21

d21

d23

d32

d32

d13

d13

d31

d31

Residual DoF tuple (e.g.) d′ = (d′12, 0, 0, d′23, d

′31, 0) ⇒ 3-cycle!

Cyclic:1) set dcij = dcjk = dcki = min{d′ij , d′jk, d′ki}

2) requires 2dcij sub-channels3) resolves 3-cycles4) residual DoF d′′ij = d′ij − dcij

Uni-directional:

1) set duij = d′′ij

2) requires duij sub-channels

d achieved!

Anas Chaaban, Aydin Sezgin Multi-way Communications 32

RUB Institute of Digital Communication Systems

Resource allocation

Consider a DoF tuple d = (d12, d13, d21, d23, d31, d32) ⇒ 2-cycles and 3-cycles!

Bi-directional:1) set dbij = dbji = min{dij , dji}2) requires dbij sub-channels3) resolves 2-cycles4) residual DoF d′ij = dij − dbij

1 2 3

d12

d21

d21

d23

d32

d32

d13

d13

d31

d31

Residual DoF tuple (e.g.) d′ = (d′12, 0, 0, d′23, d

′31, 0) ⇒ 3-cycle!

Cyclic:1) set dcij = dcjk = dcki = min{d′ij , d′jk, d′ki}2) requires 2dcij sub-channels

3) resolves 3-cycles4) residual DoF d′′ij = d′ij − dcij

Uni-directional:

1) set duij = d′′ij

2) requires duij sub-channels

d achieved!

Anas Chaaban, Aydin Sezgin Multi-way Communications 32

RUB Institute of Digital Communication Systems

Resource allocation

Consider a DoF tuple d = (d12, d13, d21, d23, d31, d32) ⇒ 2-cycles and 3-cycles!

Bi-directional:1) set dbij = dbji = min{dij , dji}2) requires dbij sub-channels3) resolves 2-cycles4) residual DoF d′ij = dij − dbij

1 2 3

d12

d21

d21

d23

d32

d32

d13

d13

d31

d31

Residual DoF tuple (e.g.) d′ = (d′12, 0, 0, d′23, d

′31, 0) ⇒ 3-cycle!

Cyclic:1) set dcij = dcjk = dcki = min{d′ij , d′jk, d′ki}2) requires 2dcij sub-channels3) resolves 3-cycles

4) residual DoF d′′ij = d′ij − dcij

Uni-directional:

1) set duij = d′′ij

2) requires duij sub-channels

d achieved!

Anas Chaaban, Aydin Sezgin Multi-way Communications 32

RUB Institute of Digital Communication Systems

Resource allocation

Consider a DoF tuple d = (d12, d13, d21, d23, d31, d32) ⇒ 2-cycles and 3-cycles!

Bi-directional:1) set dbij = dbji = min{dij , dji}2) requires dbij sub-channels3) resolves 2-cycles4) residual DoF d′ij = dij − dbij

1 2 3

d12

d21

d21

d23

d32

d32

d13

d13

d31

d31

Residual DoF tuple (e.g.) d′ = (d′12, 0, 0, d′23, d

′31, 0) ⇒ 3-cycle!

Cyclic:1) set dcij = dcjk = dcki = min{d′ij , d′jk, d′ki}2) requires 2dcij sub-channels3) resolves 3-cycles4) residual DoF d′′ij = d′ij − dcij

Uni-directional:

1) set duij = d′′ij

2) requires duij sub-channels

d achieved!

Anas Chaaban, Aydin Sezgin Multi-way Communications 32

RUB Institute of Digital Communication Systems

Resource allocation

Consider a DoF tuple d = (d12, d13, d21, d23, d31, d32) ⇒ 2-cycles and 3-cycles!

Bi-directional:1) set dbij = dbji = min{dij , dji}2) requires dbij sub-channels3) resolves 2-cycles4) residual DoF d′ij = dij − dbij

1 2 3

d12

d21

d21

d23

d32

d32

d13

d13

d31

d31

Residual DoF tuple (e.g.) d′ = (d′12, 0, 0, d′23, d

′31, 0) ⇒ 3-cycle!

Cyclic:1) set dcij = dcjk = dcki = min{d′ij , d′jk, d′ki}2) requires 2dcij sub-channels3) resolves 3-cycles4) residual DoF d′′ij = d′ij − dcij

Uni-directional:1) set duij = d′′ij

2) requires duij sub-channels

d achieved!

Anas Chaaban, Aydin Sezgin Multi-way Communications 32

RUB Institute of Digital Communication Systems

Resource allocation

Consider a DoF tuple d = (d12, d13, d21, d23, d31, d32) ⇒ 2-cycles and 3-cycles!

Bi-directional:1) set dbij = dbji = min{dij , dji}2) requires dbij sub-channels3) resolves 2-cycles4) residual DoF d′ij = dij − dbij

1 2 3

d12

d21

d21

d23

d32

d32

d13

d13

d31

d31

Residual DoF tuple (e.g.) d′ = (d′12, 0, 0, d′23, d

′31, 0) ⇒ 3-cycle!

Cyclic:1) set dcij = dcjk = dcki = min{d′ij , d′jk, d′ki}2) requires 2dcij sub-channels3) resolves 3-cycles4) residual DoF d′′ij = d′ij − dcij

Uni-directional:1) set duij = d′′ij

2) requires duij sub-channels

d achieved!

Anas Chaaban, Aydin Sezgin Multi-way Communications 32

RUB Institute of Digital Communication Systems

Resource allocation

Consider a DoF tuple d = (d12, d13, d21, d23, d31, d32) ⇒ 2-cycles and 3-cycles!

Bi-directional:1) set dbij = dbji = min{dij , dji}2) requires dbij sub-channels3) resolves 2-cycles4) residual DoF d′ij = dij − dbij

1 2 3

d12

d21

d21

d23

d32

d32

d13

d13

d31

d31

Residual DoF tuple (e.g.) d′ = (d′12, 0, 0, d′23, d

′31, 0) ⇒ 3-cycle!

Cyclic:1) set dcij = dcjk = dcki = min{d′ij , d′jk, d′ki}2) requires 2dcij sub-channels3) resolves 3-cycles4) residual DoF d′′ij = d′ij − dcij

Uni-directional:1) set duij = d′′ij

2) requires duij sub-channels d achieved!

Anas Chaaban, Aydin Sezgin Multi-way Communications 32

RUB Institute of Digital Communication Systems

Outline

1 Motivation: From one-way to multi-way

2 The MIMO Y-channel: From single-antenna to multiple-antennas

3 Main result: From capacity to DoF

4 Insights and ingredientsChannel diagonalization: Separability of communicationstructureAlignment, Compute-and-forwardTransmission strategy(3-users)

5 Extensions and Conclusion

Anas Chaaban, Aydin Sezgin Multi-way Communications 33

RUB Institute of Digital Communication Systems

K-user Case

For the K-user Y-channel with N ≤M :

• 2-cycles up to K-cycles,

• `-cycles resolved by an `-cyclic strategy

• exchanges ` symbols

• requires `− 1dimensions

• efficiency `/(`− 1)

1 2 3 · · · K

• DoF region described by

K−1∑i=1

K∑j=i+1

dpipj ≤ N, ∀p

where p is a permutation of (1, 2, · · · ,K).

Anas Chaaban, Aydin Sezgin Multi-way Communications 34

RUB Institute of Digital Communication Systems

K-user Case

For the K-user Y-channel with N ≤M :

• 2-cycles up to K-cycles,

• `-cycles resolved by an `-cyclic strategy

• exchanges ` symbols

• requires `− 1dimensions

• efficiency `/(`− 1)

1 2 3 · · · K

• DoF region described by

K−1∑i=1

K∑j=i+1

dpipj ≤ N, ∀p

where p is a permutation of (1, 2, · · · ,K).

Anas Chaaban, Aydin Sezgin Multi-way Communications 34

RUB Institute of Digital Communication Systems

K-user Case

For the K-user Y-channel with N ≤M :

• 2-cycles up to K-cycles,

• `-cycles resolved by an `-cyclic strategy

• exchanges ` symbols

• requires `− 1dimensions

• efficiency `/(`− 1)

1 2 3 · · · K

• DoF region described by

K−1∑i=1

K∑j=i+1

dpipj ≤ N, ∀p

where p is a permutation of (1, 2, · · · ,K).

Anas Chaaban, Aydin Sezgin Multi-way Communications 34

RUB Institute of Digital Communication Systems

K-user Case

For the K-user Y-channel with N ≤M :

• 2-cycles up to K-cycles,

• `-cycles resolved by an `-cyclic strategy

• exchanges ` symbols

• requires `− 1dimensions

• efficiency `/(`− 1)

1 2 3 · · · K

• DoF region described by

K−1∑i=1

K∑j=i+1

dpipj ≤ N, ∀p

where p is a permutation of (1, 2, · · · ,K).

Anas Chaaban, Aydin Sezgin Multi-way Communications 34

RUB Institute of Digital Communication Systems

K-user Case

For the K-user Y-channel with N ≤M :

• 2-cycles up to K-cycles,

• `-cycles resolved by an `-cyclic strategy

• exchanges ` symbols

• requires `− 1dimensions

• efficiency `/(`− 1)

1 2 3 · · · K

• DoF region described by

K−1∑i=1

K∑j=i+1

dpipj ≤ N, ∀p

where p is a permutation of (1, 2, · · · ,K).

Anas Chaaban, Aydin Sezgin Multi-way Communications 34

RUB Institute of Digital Communication Systems

K-user Case

For the K-user Y-channel with N ≤M :

• 2-cycles up to K-cycles,

• `-cycles resolved by an `-cyclic strategy

• exchanges ` symbols

• requires `− 1dimensions

• efficiency `/(`− 1)

1 2 3 · · · K

• DoF region described by

K−1∑i=1

K∑j=i+1

dpipj ≤ N, ∀p

where p is a permutation of (1, 2, · · · ,K).

Anas Chaaban, Aydin Sezgin Multi-way Communications 34

RUB Institute of Digital Communication Systems

K-user Case

For the K-user Y-channel with N ≤M :

• 2-cycles up to K-cycles,

• `-cycles resolved by an `-cyclic strategy

• exchanges ` symbols

• requires `− 1dimensions

• efficiency `/(`− 1)

1 2 3 · · · K

• DoF region described by

K−1∑i=1

K∑j=i+1

dpipj ≤ N, ∀p

where p is a permutation of (1, 2, · · · ,K).

Anas Chaaban, Aydin Sezgin Multi-way Communications 34

RUB Institute of Digital Communication Systems

Conclusion

• Studied the K-user MIMO Y-channel

• Combination of:

• Channel diagonalization• Signal-alignment with compute-forward

• decode-forward

• DoF region characterized

• Cyclic strategy required joint encoding over multiple sub-channels:

• Downlink and uplink are seperable• Subchannels are not decomposable (seperable)

Thank You

Anas Chaaban, Aydin Sezgin Multi-way Communications 35

RUB Institute of Digital Communication Systems

Conclusion

• Studied the K-user MIMO Y-channel

• Combination of:

• Channel diagonalization

• Signal-alignment with compute-forward

• decode-forward

• DoF region characterized

• Cyclic strategy required joint encoding over multiple sub-channels:

• Downlink and uplink are seperable• Subchannels are not decomposable (seperable)

Thank You

Anas Chaaban, Aydin Sezgin Multi-way Communications 35

RUB Institute of Digital Communication Systems

Conclusion

• Studied the K-user MIMO Y-channel

• Combination of:

• Channel diagonalization• Signal-alignment with compute-forward

• decode-forward

• DoF region characterized

• Cyclic strategy required joint encoding over multiple sub-channels:

• Downlink and uplink are seperable• Subchannels are not decomposable (seperable)

Thank You

Anas Chaaban, Aydin Sezgin Multi-way Communications 35

RUB Institute of Digital Communication Systems

Conclusion

• Studied the K-user MIMO Y-channel

• Combination of:

• Channel diagonalization• Signal-alignment with compute-forward

• decode-forward

• DoF region characterized

• Cyclic strategy required joint encoding over multiple sub-channels:

• Downlink and uplink are seperable• Subchannels are not decomposable (seperable)

Thank You

Anas Chaaban, Aydin Sezgin Multi-way Communications 35

RUB Institute of Digital Communication Systems

Conclusion

• Studied the K-user MIMO Y-channel

• Combination of:

• Channel diagonalization• Signal-alignment with compute-forward

• decode-forward

• DoF region characterized

• Cyclic strategy required joint encoding over multiple sub-channels:

• Downlink and uplink are seperable• Subchannels are not decomposable (seperable)

Thank You

Anas Chaaban, Aydin Sezgin Multi-way Communications 35

RUB Institute of Digital Communication Systems

Conclusion

• Studied the K-user MIMO Y-channel

• Combination of:

• Channel diagonalization• Signal-alignment with compute-forward

• decode-forward

• DoF region characterized

• Cyclic strategy required joint encoding over multiple sub-channels:

• Downlink and uplink are seperable

• Subchannels are not decomposable (seperable)

Thank You

Anas Chaaban, Aydin Sezgin Multi-way Communications 35

RUB Institute of Digital Communication Systems

Conclusion

• Studied the K-user MIMO Y-channel

• Combination of:

• Channel diagonalization• Signal-alignment with compute-forward

• decode-forward

• DoF region characterized

• Cyclic strategy required joint encoding over multiple sub-channels:

• Downlink and uplink are seperable• Subchannels are not decomposable (seperable)

Thank You

Anas Chaaban, Aydin Sezgin Multi-way Communications 35

RUB Institute of Digital Communication Systems

Conclusion

• Studied the K-user MIMO Y-channel

• Combination of:

• Channel diagonalization• Signal-alignment with compute-forward

• decode-forward

• DoF region characterized

• Cyclic strategy required joint encoding over multiple sub-channels:

• Downlink and uplink are seperable• Subchannels are not decomposable (seperable)

Thank You

Anas Chaaban, Aydin Sezgin Multi-way Communications 35

RUB Institute of Digital Communication Systems

Related work

1 C. Shannon, Two-way communication channels, Proc. of Fourth BerkeleySymposium on Mathematics, Statistics, and Probability, 1961.

2 B. Rankov and A. Wittneben, Spectral efficient signaling for half-duplex relaychannels, Proc. of the Asilomar Conference on Signals, Systems, andComputers, 2005.

3 D. Gunduz, A. Yener, A. Goldsmith, and H. V. Poor, The multi-way relaychannel, IEEE Transactions on Information Theory, Vol. 59(1), pp. 51-63.

4 N. Lee, J.-B. Lim, and J. Chun, Degrees of freedom of the MIMO Y channel:Signal space alignment for network coding, IEEE Trans. on Info. Theory, 2010.

5 A. Chaaban and A. Sezgin,, Approximate Sum-Capacity of the Y-channel, IEEETransactions on Information Theory, Vol.59(9), pp.5723-5740.

6 A. Chaaban and A. Sezgin,, Multi-way communications, Foundations andTrends in Communications and Information Theory, now publishers, in review.

Anas Chaaban, Aydin Sezgin Multi-way Communications 36

RUB Institute of Digital Communication Systems

Optimality

Total number ofdimensions required toachieve d ∈ D:

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Ns =

bi-directional︷ ︸︸ ︷2∑

i=1

3∑j=i+1

dbij +

cyclic︷ ︸︸ ︷3∑

j=2

2dc1j +

uni-directional︷ ︸︸ ︷3∑

i=1

3∑j=1, j 6=i

duij

(duij = dij − dbij − dcij)

=3∑

i=1

3∑j=1, j 6=i

dij −2∑

i=1

3∑j=i+1

dbij −3∑

j=2

dc1j

(dij + dji − dbij = max{dij , dji})

= max{d12, d21}+ max{d13, d31}+ max{d23, d32}︸ ︷︷ ︸d12+d23+d31 e.g.⇒dc

13=0, db

12=d21

−dc12 − dc13

= d12 + d23 + d31 − dc12

(dc12 = d12 − db12 e.g.)

= db12 + d23 + d31

= d21 + d23 + d31

No cycles ⇒ Ns ≤ N ⇒ All d ∈ D are achievable

Anas Chaaban, Aydin Sezgin Multi-way Communications 37

RUB Institute of Digital Communication Systems

Optimality

Total number ofdimensions required toachieve d ∈ D:

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Ns =

bi-directional︷ ︸︸ ︷2∑

i=1

3∑j=i+1

dbij +

cyclic︷ ︸︸ ︷3∑

j=2

2dc1j +

uni-directional︷ ︸︸ ︷3∑

i=1

3∑j=1, j 6=i

duij

(duij = dij − dbij − dcij)

=3∑

i=1

3∑j=1, j 6=i

dij −2∑

i=1

3∑j=i+1

dbij −3∑

j=2

dc1j

(dij + dji − dbij = max{dij , dji})

= max{d12, d21}+ max{d13, d31}+ max{d23, d32}︸ ︷︷ ︸d12+d23+d31 e.g.⇒dc

13=0, db

12=d21

−dc12 − dc13

= d12 + d23 + d31 − dc12

(dc12 = d12 − db12 e.g.)

= db12 + d23 + d31

= d21 + d23 + d31

No cycles ⇒ Ns ≤ N ⇒ All d ∈ D are achievable

Anas Chaaban, Aydin Sezgin Multi-way Communications 37

RUB Institute of Digital Communication Systems

Optimality

Total number ofdimensions required toachieve d ∈ D:

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Ns =

bi-directional︷ ︸︸ ︷2∑

i=1

3∑j=i+1

dbij +

cyclic︷ ︸︸ ︷3∑

j=2

2dc1j +

uni-directional︷ ︸︸ ︷3∑

i=1

3∑j=1, j 6=i

duij (duij = dij − dbij − dcij)

=3∑

i=1

3∑j=1, j 6=i

dij −2∑

i=1

3∑j=i+1

dbij −3∑

j=2

dc1j

(dij + dji − dbij = max{dij , dji})

= max{d12, d21}+ max{d13, d31}+ max{d23, d32}︸ ︷︷ ︸d12+d23+d31 e.g.⇒dc

13=0, db

12=d21

−dc12 − dc13

= d12 + d23 + d31 − dc12

(dc12 = d12 − db12 e.g.)

= db12 + d23 + d31

= d21 + d23 + d31

No cycles ⇒ Ns ≤ N ⇒ All d ∈ D are achievable

Anas Chaaban, Aydin Sezgin Multi-way Communications 37

RUB Institute of Digital Communication Systems

Optimality

Total number ofdimensions required toachieve d ∈ D:

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Ns =

bi-directional︷ ︸︸ ︷2∑

i=1

3∑j=i+1

dbij +

cyclic︷ ︸︸ ︷3∑

j=2

2dc1j +

uni-directional︷ ︸︸ ︷3∑

i=1

3∑j=1, j 6=i

duij (duij = dij − dbij − dcij)

=3∑

i=1

3∑j=1, j 6=i

dij −2∑

i=1

3∑j=i+1

dbij −3∑

j=2

dc1j

(dij + dji − dbij = max{dij , dji})

= max{d12, d21}+ max{d13, d31}+ max{d23, d32}︸ ︷︷ ︸d12+d23+d31 e.g.⇒dc

13=0, db

12=d21

−dc12 − dc13

= d12 + d23 + d31 − dc12

(dc12 = d12 − db12 e.g.)

= db12 + d23 + d31

= d21 + d23 + d31

No cycles ⇒ Ns ≤ N ⇒ All d ∈ D are achievable

Anas Chaaban, Aydin Sezgin Multi-way Communications 37

RUB Institute of Digital Communication Systems

Optimality

Total number ofdimensions required toachieve d ∈ D:

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Ns =

bi-directional︷ ︸︸ ︷2∑

i=1

3∑j=i+1

dbij +

cyclic︷ ︸︸ ︷3∑

j=2

2dc1j +

uni-directional︷ ︸︸ ︷3∑

i=1

3∑j=1, j 6=i

duij (duij = dij − dbij − dcij)

=3∑

i=1

3∑j=1, j 6=i

dij −2∑

i=1

3∑j=i+1

dbij −3∑

j=2

dc1j (dij + dji − dbij = max{dij , dji})

= max{d12, d21}+ max{d13, d31}+ max{d23, d32}︸ ︷︷ ︸d12+d23+d31 e.g.⇒dc

13=0, db

12=d21

−dc12 − dc13

= d12 + d23 + d31 − dc12

(dc12 = d12 − db12 e.g.)

= db12 + d23 + d31

= d21 + d23 + d31

No cycles ⇒ Ns ≤ N ⇒ All d ∈ D are achievable

Anas Chaaban, Aydin Sezgin Multi-way Communications 37

RUB Institute of Digital Communication Systems

Optimality

Total number ofdimensions required toachieve d ∈ D:

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Ns =

bi-directional︷ ︸︸ ︷2∑

i=1

3∑j=i+1

dbij +

cyclic︷ ︸︸ ︷3∑

j=2

2dc1j +

uni-directional︷ ︸︸ ︷3∑

i=1

3∑j=1, j 6=i

duij (duij = dij − dbij − dcij)

=3∑

i=1

3∑j=1, j 6=i

dij −2∑

i=1

3∑j=i+1

dbij −3∑

j=2

dc1j (dij + dji − dbij = max{dij , dji})

= max{d12, d21}+ max{d13, d31}+ max{d23, d32}︸ ︷︷ ︸d12+d23+d31 e.g.⇒dc

13=0, db

12=d21

−dc12 − dc13

= d12 + d23 + d31 − dc12

(dc12 = d12 − db12 e.g.)

= db12 + d23 + d31

= d21 + d23 + d31

No cycles ⇒ Ns ≤ N ⇒ All d ∈ D are achievable

Anas Chaaban, Aydin Sezgin Multi-way Communications 37

RUB Institute of Digital Communication Systems

Optimality

Total number ofdimensions required toachieve d ∈ D:

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Ns =

bi-directional︷ ︸︸ ︷2∑

i=1

3∑j=i+1

dbij +

cyclic︷ ︸︸ ︷3∑

j=2

2dc1j +

uni-directional︷ ︸︸ ︷3∑

i=1

3∑j=1, j 6=i

duij (duij = dij − dbij − dcij)

=3∑

i=1

3∑j=1, j 6=i

dij −2∑

i=1

3∑j=i+1

dbij −3∑

j=2

dc1j (dij + dji − dbij = max{dij , dji})

= max{d12, d21}+ max{d13, d31}+ max{d23, d32}︸ ︷︷ ︸d12+d23+d31 e.g.⇒dc

13=0, db

12=d21

−dc12 − dc13

= d12 + d23 + d31 − dc12

(dc12 = d12 − db12 e.g.)

= db12 + d23 + d31

= d21 + d23 + d31

No cycles ⇒ Ns ≤ N ⇒ All d ∈ D are achievable

Anas Chaaban, Aydin Sezgin Multi-way Communications 37

RUB Institute of Digital Communication Systems

Optimality

Total number ofdimensions required toachieve d ∈ D:

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Ns =

bi-directional︷ ︸︸ ︷2∑

i=1

3∑j=i+1

dbij +

cyclic︷ ︸︸ ︷3∑

j=2

2dc1j +

uni-directional︷ ︸︸ ︷3∑

i=1

3∑j=1, j 6=i

duij (duij = dij − dbij − dcij)

=3∑

i=1

3∑j=1, j 6=i

dij −2∑

i=1

3∑j=i+1

dbij −3∑

j=2

dc1j (dij + dji − dbij = max{dij , dji})

= max{d12, d21}+ max{d13, d31}+ max{d23, d32}︸ ︷︷ ︸d12+d23+d31 e.g.⇒dc

13=0, db

12=d21

−dc12 − dc13

= d12 + d23 + d31 − dc12 (dc12 = d12 − db12 e.g.)

= db12 + d23 + d31

= d21 + d23 + d31

No cycles ⇒ Ns ≤ N ⇒ All d ∈ D are achievable

Anas Chaaban, Aydin Sezgin Multi-way Communications 37

RUB Institute of Digital Communication Systems

Optimality

Total number ofdimensions required toachieve d ∈ D:

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Ns =

bi-directional︷ ︸︸ ︷2∑

i=1

3∑j=i+1

dbij +

cyclic︷ ︸︸ ︷3∑

j=2

2dc1j +

uni-directional︷ ︸︸ ︷3∑

i=1

3∑j=1, j 6=i

duij (duij = dij − dbij − dcij)

=3∑

i=1

3∑j=1, j 6=i

dij −2∑

i=1

3∑j=i+1

dbij −3∑

j=2

dc1j (dij + dji − dbij = max{dij , dji})

= max{d12, d21}+ max{d13, d31}+ max{d23, d32}︸ ︷︷ ︸d12+d23+d31 e.g.⇒dc

13=0, db

12=d21

−dc12 − dc13

= d12 + d23 + d31 − dc12 (dc12 = d12 − db12 e.g.)

= db12 + d23 + d31

= d21 + d23 + d31

No cycles ⇒ Ns ≤ N ⇒ All d ∈ D are achievable

Anas Chaaban, Aydin Sezgin Multi-way Communications 37

RUB Institute of Digital Communication Systems

Optimality

Total number ofdimensions required toachieve d ∈ D:

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Ns =

bi-directional︷ ︸︸ ︷2∑

i=1

3∑j=i+1

dbij +

cyclic︷ ︸︸ ︷3∑

j=2

2dc1j +

uni-directional︷ ︸︸ ︷3∑

i=1

3∑j=1, j 6=i

duij (duij = dij − dbij − dcij)

=3∑

i=1

3∑j=1, j 6=i

dij −2∑

i=1

3∑j=i+1

dbij −3∑

j=2

dc1j (dij + dji − dbij = max{dij , dji})

= max{d12, d21}+ max{d13, d31}+ max{d23, d32}︸ ︷︷ ︸d12+d23+d31 e.g.⇒dc

13=0, db

12=d21

−dc12 − dc13

= d12 + d23 + d31 − dc12 (dc12 = d12 − db12 e.g.)

= db12 + d23 + d31

= d21 + d23 + d31

No cycles ⇒ Ns ≤ N ⇒ All d ∈ D are achievable

Anas Chaaban, Aydin Sezgin Multi-way Communications 37

RUB Institute of Digital Communication Systems

Optimality

Total number ofdimensions required toachieve d ∈ D:

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Ns =

bi-directional︷ ︸︸ ︷2∑

i=1

3∑j=i+1

dbij +

cyclic︷ ︸︸ ︷3∑

j=2

2dc1j +

uni-directional︷ ︸︸ ︷3∑

i=1

3∑j=1, j 6=i

duij (duij = dij − dbij − dcij)

=3∑

i=1

3∑j=1, j 6=i

dij −2∑

i=1

3∑j=i+1

dbij −3∑

j=2

dc1j (dij + dji − dbij = max{dij , dji})

= max{d12, d21}+ max{d13, d31}+ max{d23, d32}︸ ︷︷ ︸d12+d23+d31 e.g.⇒dc

13=0, db

12=d21

−dc12 − dc13

= d12 + d23 + d31 − dc12 (dc12 = d12 − db12 e.g.)

= db12 + d23 + d31

= d21 + d23 + d31

No cycles ⇒ Ns ≤ N

⇒ All d ∈ D are achievable

Anas Chaaban, Aydin Sezgin Multi-way Communications 37

RUB Institute of Digital Communication Systems

Optimality

Total number ofdimensions required toachieve d ∈ D:

Bi-directional 2 symbols 1 sub-channelCyclic 3 symbols 2 sub-channels

Uni-directional 1 symbol 1 sub-channel

Ns =

bi-directional︷ ︸︸ ︷2∑

i=1

3∑j=i+1

dbij +

cyclic︷ ︸︸ ︷3∑

j=2

2dc1j +

uni-directional︷ ︸︸ ︷3∑

i=1

3∑j=1, j 6=i

duij (duij = dij − dbij − dcij)

=3∑

i=1

3∑j=1, j 6=i

dij −2∑

i=1

3∑j=i+1

dbij −3∑

j=2

dc1j (dij + dji − dbij = max{dij , dji})

= max{d12, d21}+ max{d13, d31}+ max{d23, d32}︸ ︷︷ ︸d12+d23+d31 e.g.⇒dc

13=0, db

12=d21

−dc12 − dc13

= d12 + d23 + d31 − dc12 (dc12 = d12 − db12 e.g.)

= db12 + d23 + d31

= d21 + d23 + d31

No cycles ⇒ Ns ≤ N ⇒ All d ∈ D are achievable

Anas Chaaban, Aydin Sezgin Multi-way Communications 37

RUB Institute of Digital Communication Systems

Outer bound

Dec1 y1

Dec2 y2

Relay Dec3y3

m12, m13

m21, m23

m32, m31

D1D1

D2D2

D3D3

Consider any reliable scheme for the 4-user MIMO MRC

Anas Chaaban, Aydin Sezgin Multi-way Communications 38

RUB Institute of Digital Communication Systems

Outer bound

Dec1 y1

Dec2 y2

Relay Dec3y3

m12, m13

m21, m23

m32, m31

D1D1

D2D2

D3D3

m21, m31

m12, m32

m13, m23

Users can decode their desired signals

Anas Chaaban, Aydin Sezgin Multi-way Communications 38

RUB Institute of Digital Communication Systems

Outer bound

Dec1 y1

Dec2 y2

Relay Dec3y3

m12, m13

m21, m23

m32, m31

D1D1

D2D2

D3D3

m21, m31

m12, m32

m13, m23

m23, y2

Give m23 and y2 to user 1 as side info.

Anas Chaaban, Aydin Sezgin Multi-way Communications 38

RUB Institute of Digital Communication Systems

Outer bound

Dec1 y1

Dec2 y2

Relay Dec3y3

m12, m13

m21, m23

m32, m31

D1D1

D2D2

D3D3

m21, m31

m12, m32

m13, m23

m23, y2

Now, user 1 has the info. available at user 2

Anas Chaaban, Aydin Sezgin Multi-way Communications 38

RUB Institute of Digital Communication Systems

Outer bound

Dec1 y1

Dec2 y2

Relay Dec3y3

m12, m13

m21, m23

m32, m31

D1D1

D2D2

D3D3

m21, m31

m12, m32

m13, m23

m23, y2

m32

⇒ User 1 can decode m32

Anas Chaaban, Aydin Sezgin Multi-way Communications 38

RUB Institute of Digital Communication Systems

Upper bound

User 1 can decode (m21,m31,m32) from (m12,m13,y1,

side info.︷ ︸︸ ︷m23,y2)

Dec1y1

y2Relay

m12, m13, m23

D1D1

D2D2

m21, m31, m32

⇒ R21 +R31 +R32 ≤ I (xr;y1,y2) = I

(xr;

[D1

D2

]xr +

[z1

z2

])P2P Channel

⇒ d21 + d31 + d32 ≤ rank

([D1

D2

])= N

Considering different combinations of users gives the desired outer bound

2∑i=1

3∑j=i+1

dpipj ≤ N, ∀p

Anas Chaaban, Aydin Sezgin Multi-way Communications 39

RUB Institute of Digital Communication Systems

Upper bound

User 1 can decode (m21,m31,m32) from (m12,m13,y1,

side info.︷ ︸︸ ︷m23,y2)

Dec1y1

y2Relay

m12, m13, m23

D1D1

D2D2

m21, m31, m32

⇒ R21 +R31 +R32 ≤ I (xr;y1,y2) = I

(xr;

[D1

D2

]xr +

[z1

z2

])P2P Channel

⇒ d21 + d31 + d32 ≤ rank

([D1

D2

])= N

Considering different combinations of users gives the desired outer bound

2∑i=1

3∑j=i+1

dpipj ≤ N, ∀p

Anas Chaaban, Aydin Sezgin Multi-way Communications 39

RUB Institute of Digital Communication Systems

Upper bound

User 1 can decode (m21,m31,m32) from (m12,m13,y1,

side info.︷ ︸︸ ︷m23,y2)

Dec1y1

y2Relay

m12, m13, m23

D1D1

D2D2

m21, m31, m32

⇒ R21 +R31 +R32 ≤ I (xr;y1,y2) = I

(xr;

[D1

D2

]xr +

[z1

z2

])P2P Channel

⇒ d21 + d31 + d32 ≤ rank

([D1

D2

])= N

Considering different combinations of users gives the desired outer bound

2∑i=1

3∑j=i+1

dpipj ≤ N, ∀p

Anas Chaaban, Aydin Sezgin Multi-way Communications 39

RUB Institute of Digital Communication Systems

Upper bound

User 1 can decode (m21,m31,m32) from (m12,m13,y1,

side info.︷ ︸︸ ︷m23,y2)

Dec1y1

y2Relay

m12, m13, m23

D1D1

D2D2

m21, m31, m32

⇒ R21 +R31 +R32 ≤ I (xr;y1,y2) = I

(xr;

[D1

D2

]xr +

[z1

z2

])P2P Channel

⇒ d21 + d31 + d32 ≤ rank

([D1

D2

])= N

Considering different combinations of users gives the desired outer bound

2∑i=1

3∑j=i+1

dpipj ≤ N, ∀p

Anas Chaaban, Aydin Sezgin Multi-way Communications 39

top related