Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems Sean Huberman Department of Electrical and Computer Engineering McGill University Montreal, Quebec, Canada
October 20, 2014
S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
1/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Outline
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
2/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Background and Motivation • Demand for data-intensive services is increasing • Higher data-rates are required • Improve the spectral efficiency (bits/s/Hz/Area)
c Nokia Siemens Networks 2012
=⇒ Dynamic resource allocation techniques are needed S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
3/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Point-to-Point and Point-to-Multi-Point Channels
Point-to-point
Point-to-multi-point
Multi-link point-to-point
• Solid links: intended signals • Dashed links: interference signals • Links: single or multiple spatial dimensions
S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
4/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Interference Issues in Multi-User Systems
Conventional approach: • Different frequency bands • Avoid co-channel interference
S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
5/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Interference Issues in Multi-User Systems
Conventional approach: • Different frequency bands
Universal frequency reuse: • Same frequency band
• Avoid co-channel interference
S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
5/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Interference Issues in Multi-User Systems
• Different frequency bands
Universal frequency reuse: • Same frequency band
• Avoid co-channel interference
• Interference must be managed
Conventional approach:
S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
5/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Half-Duplex vs. Full-Duplex Transmission
Half-Duplex (HD)
Full-Duplex (FD)
• Conventional approach
• Increase sharing (up/downlink)
• Separate time or frequency slots
• “Double” capacity • Creates self-interference
S. Huberman
Slot 1
Slot 1
Slot 2
Slot 1
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
6/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Half-Duplex vs. Full-Duplex Transmission
Half-Duplex (HD)
Full-Duplex (FD)
• Conventional approach
• Increase sharing (up/downlink)
• Separate time or frequency slots
• “Double” capacity • Creates self-interference
Slot 1
Slot 1
SelfInterference
Slot 2
S. Huberman
Slot 1
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
6/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Environments Under Consideration: Digital Subscriber Line (DSL) • Uses twisted-pair copper telephone wire • Lines buried underground in shared binders (e.g., Binder B) • Interference known as crosstalk • Practical example of a multi-link point-to-point channel
Central Office
A
B
DSLAM Fiber Links
JWI: Junction Wire Interface
DSLAM: Digital Subscriber Line Access Multiplexer
DSLAM
C
Fiber Links
S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
7/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Environments Under Consideration: Cellular Networks • Radio network distributed over a cell (operated by base station) • Concept of small cells emerged • Share same frequency • Low-powered nodes =⇒ reduced interference • Shorter transmission =⇒ better channels
=⇒ Small cells are excellent candidates for full-duplex transmission S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
8/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Dynamic Power Allocation (DPA) Techniques • Standard transmission equation:
y = Hx + z Scalar DPA: • Treat each line/antenna, k, separately • Optimize the power, pk = E[|xk |2 ], of each line/antenna • Maximize rate in the presence of interference
Vectored DPA: • Jointly process multiple lines/antennas • Apply precoding to preprocess transmitted signal • Re-write x = V˜ x, resulting in:
x+z y = |{z} HV ˜
• Optimize V (subject to power constraints) to maximize rate • Can “modify” channel conditions (i.e., H) S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
9/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Objectives
To design scalar and vectored DPA transmission techniques: • Manage interference in severely interference-limited
environments • Using HD or FD transmission • With applications to wireline and wireless systems
S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
10/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Contributions for Half-Duplex (HD) Systems
Interference control in a multi-link environment: • Constant offset Autonomous Spectrum Balancing using Multiple
Reference Users (ASB-MRU) algorithm • Method for estimating spectral efficiency using channel statistics
(prior to allocating resources) • Omitted due to time constraints
=⇒ HD publications: 4 journal papers, 5 conference papers, 1 book chapter
S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
11/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Interference Control in a Multi-Link Environment Scalar Dynamic Power Allocation
Limitations of existing approaches: • Semi-centralized & global channel knowledge • Many iterations (i.e., interference measurements)
Objectives of proposed approach: • Comparable performance (i.e., sum-rate) • Reduce computational complexity • Distributed, local channel knowledge, reduce number of iterations
max
pk , k∈K
subject to:
X
wk Rk
k∈K
X
pkf ≤ Pk ,
∀ k
f ∈F
0 ≤ pkf ≤ pk,mask , f
S. Huberman
∀
f,k
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
12/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Autonomous Spectrum Balancing Using Multiple Reference Users (ASB-MRU) • Construct a virtual network of reference users • Provides approximate global channel knowledge • Users self-optimize: maximize the virtual network rate & their own • Key considerations: • How many reference users should there be? • How should we select the KR reference users? • What should the reference user parameters be? Virtual Network
For each user k: User 1
max wk Rk + pk
User 4
User 2
subject to: Virtual Network
Virtual Network
X
˜r w ˜rR
r∈R
X
pkf ≤ Pk
f ∈F
0 ≤ pkf ≤ pk,mask , f
User 3
∀ f
Virtual Network S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
13/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Constant Offset ASB-MRU • Many existing algorithms have a similar power-update structure:
"
Γintkf wk pkf = − 2 λk + ∆λkf |hk,k f |
#pfk,mask 0
• Offsets: per-sub-carrier quotas for each user • Existing algorithms tune the offset, ∆λkf , on a per-iteration basis
=⇒ leading to slow convergence. • Using the virtual network: approximate the final tuned offset • Pre-compute offset during initialization phase • Offset becomes constant =⇒ fast convergence
S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
14/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Constant Offset ASB-MRU (Con’t.) 1. Setup virtual network (cluster users based on pay-off function) fpay (k) ,
F X f =1
log 1 +
σfk +
P
k,l 2 fixed l6=k |hf | pf 2 fixed |hk,k f | pf
!
=
F X
log
f =1
(recpay )kf 2 fixed |hk,k f | pf
!
2 Select parameters for each reference user • Optimize reference user power allocation assuming no crosstalk • Compute pay-off function for reference users • Scale reference user power by ratio: fpay (r)/ maxr∈R {fpay (r)}
=⇒ Weaker reference users use more power (more accurate)
3 Determine offset for each user • Same formula as varying offset ASB-MRU using virtual network
4 Each user iteratively updates power until convergence • Modified water-filling procedure with constant offsets S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
15/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Constant Offset ASB-MRU (Con’t.) Summary of VDSL Illustrative Examples
IWF SCALE & DSB Constant Offset ASB-MRU
O(FKKR )
–
User O(F) O(F) O(F)
Messages/User Initial Per Iter – – – 2F 2F
–
48 User FTTN Uplink - VDSL - 100 to 1000 m Avg. DSB SCALE IWF ASB-MRU PoM 100.0 100.0 66.0 98.4 # iters 267.7 225.5 5.1 10.8 iters (98%) 35.9 10.9 2.0 1.0 # Refs – – – 15.9
500 480 460 440 420 Sum Rate (Mbps)
Complexity SMC / Iter – O(FK 2 )
Initial – –
400 380 360 340
• 98.4 PoM with ∼16 reference users
320
DSB SCALE IWF ASB−MRU
300 280 0
5
S. Huberman
10
15 20 Number of Iterations
25
30
• Sensitivity to users entering/leaving the system • Only generate offsets for new users • Maintains 96–99 PoM
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
16/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Sufficient Conditions for the Convergence, Existence, and Efficiency Theorem 3.1: Existence and Convergence k,l 2 Γ|h | 1 If maxf ,k,l6=k |hk,kf |2 < (K−1) , the constant offset algorithm using fixed f
weights will converge to a fixed point.
Theorem 3.2: Efficiency k,l 2 Γ|h | 1 < (K−1) and there exists a point satisfying the If maxf ,k,l6=k |hk,kf |2 f
KKT conditions of the rate adaptive optimization problem, then there exists a set of constant offsets, {∆λnk }, such that the constant offset algorithm will converge to that point. • With appropriate choice of constant offsets, convergence to a
KKT point can be achieved • The constant offset ASB-MRU algorithm provides an
approximation to this KKT point S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
17/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Contributions for Full-Duplex (FD) Systems • FD Precoding (FDP) transceiver structure proposed • Jointly control forward precoding & self-interference cancellation • FD MIMO-OFDM point-to-point & point-to-multi-point systems • Allows for different algorithms and optimization objectives
• Various separate and joint FDP algorithm designs • Focus on sum-rate maximization objective • Sequential Convex Approximations for Matrix-variable
Programming (SCAMP) • Analytical expressions for loss in sum-rate due to residual
self-interference (omitted due to time constraints) =⇒ FD publications: 3 patent applications, 4 journal papers, 2 conference papers S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
18/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Full-Duplex System Models FD MIMO Point-to-Point
FD MIMO Point-to-Multi-Point [HDL]1
H1
UE
DL
[HDL]K UE
Node 1 (Tx / Rx)
G1
G2
Node 2 (Tx / Rx)
BS (Tx / Rx)
GBS
GUE [HUL]1 [HUL]k
H2
UL UE UE
For example, for the FD MIMO point-to-point system: yi = Hj Vj xj + Gi Vi xi + zi −1 ˆ i,i , ˆ i = log2 INR + Σi + C C R i,j
Ci,j = Hj Qj H†j ,
ˆ i,i = G ˆ i Qi G ˆ †, C i
max
Q1 ,Q2
ˆ1 + R ˆ2 R
subject to: Tr{Qi } ≤ Pmax,i , i = 1, 2, Qi < 0, i = 1, 2
Qi = Vi Si V†i =⇒ Non-convex matrix-variable optimization problems! S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
19/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Full-Duplex Precoding (FDP) Structure
M antennas
H1
Node 1 (Tx / Rx)
G1
Rx
G2
Node 2 (Tx / Rx)
?
Inside the nodes
Rx
H2
M Tx signals
• Applies to both FD MIMO point-to-point and point-to-multi-point
S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
20/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Full-Duplex Precoding Structure (Con’t.) • Jointly controls forward precoding
H
& self-interference cancellation
M antennas LNA
ADC
Rx
G M receivers
• Additional dimensionality at Tx LNA
ADC
Rx
• Various precoding algorithms and
different optimization criteria
PA
PA 2M paths
• Forward paths affect forward Tx
DAC
DAC
DAC
DAC
and cancellation
H and G channel information
Precoder (2M x M)
• Auxiliary paths only affect
cancellation
M Tx signals
Hj =
˜j H
0M
,
Gi =
˜ i,a G
αIM
,
=⇒ Can “trade-off” forward Tx to assist in cancellation S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
21/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Separate Full-Duplex Precoding (FDP) • Forward precoding and cancellation designed separately
Vi =
Vi,F Vi,S
• Optimize Vi,F assuming no self-interference =⇒ convex:
max
˜ 1 ,Q ˜2 Q
˜1 + R ˜2 R
˜ i } ≤ Pmax,i , i = 1, 2 subject to: Tr{Q ˜ i < 0, i = 1, 2, Q ˜ j = Vi,F Si V† and R ˜ i is the self-interference-free rate where Q i,F • Then, optimize Vi,S to minimize the residual self-interference: ˆ 2 min G V i i Vi,S
F
subject to: Tr{Vi,S V†i,S } ≤ Pmax,i
S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
22/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Joint Full-Duplex Precoding (FDP) • Directly optimize Vi
• Allows for forward transmission to assist in cancellation
• Joint optimization =⇒ Non-convex • Apply Sequential Convex Programming (SCP)
• Other approaches: • FDP Self-Interference Pricing (FDP-SIP) • FDP Self-Interference Threshold (FDP-SIT)
S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
23/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Sequential Convex Programming (SCP) Non-convex optimization problem
Update current matrices
Apply local convex approximation
Convex optimization problem
Convergence?
Solve convex optimization
No
Yes Apply optimal matrices
DC-based approach (conventional): • Re-write objective function as: f = g − h, g and h convex • Apply first-order approximation to h (to convexify f ) h i 1 log2 |A + X| ≥ log2 |A + X0 | + ln(2) Tr (A + X0 )−1 (X − X0 ) • Approximation only affects self-interference and noise terms S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
24/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Sequential Convex Approximations for Matrix-Variable Programming (SCAMP) • Combines a lower-bound with an upper-bound • Non-convex approximation that can be more easily convexified • SCAMP approximation:
log2 |A + X| ≈ log2 |X| −
1 ln(2) Tr [ΦX]
+ β,
• Affects forward, self-interference, and noise terms but spreads
out the approximation ˆ i ≈ − log2 |Ci,j | + log2 |Υi | + 1 Tr Φi Υ−1 Ci,j −βi , −R i ln(2) | {z } | {z } ϕi
ti
• Apply Taylor’s series to ϕi and ti =⇒ convex! S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
25/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
FDP Self-Interference Pricing (FDP-SIP) • Utility function optimization based approach
ˆ 2 max ψi ||Hi Vi ||2F − πi G V i i Vi
subject to:
F
V†i Vi
= IM
• Solve using subspace maximization algorithm for fixed ψi and πi • Optimize ψi and πi using pricing (using matrix differentials):
2 ∂R i , ψi , † † ∂ Hj Vj Sj Vj Hj F
S. Huberman
2 −∂R i πi , † † ˆ ˆ i Vi Si V G ∂ G i i
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
F
26/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
FDP Self-Interference Threshold (FDP-SIT) • Apply threshold on self-interference:
max
Q1 ,Q2
th Rth 1 + R2
subject to: Tr{Qi } ≤ Pmax,i , i = 1, 2, ˆ i Qi G ˆ † } ≤ θi , i = 1, 2, Tr{G i
Qi < 0, i = 1, 2, where θi is a fixed threshold value • Component-wise upper-bound the self-interference covariance
matrix:
−1 Ci,j Rth i = log2 IM + (Σi + Θi )
where Θi is the M × M matrix with each entry as θi /M • For fixed threshold values =⇒ Convex • Outer-loop adjusts thresholds using nested bisection search S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
27/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Separate vs. Joint Full-Duplex Precoding (FDP) Theorem 4.1 p ˆ Separate FDP is optimal if and only if α ≥ G i,a Vi,F xi / Pmax,i . 2
• α represents the gain of the auxiliary paths
• Intuition: if we can cancel “all” the self-interference using
auxiliary paths alone then separate if optimal 1.8 1.6
• MIMO point-to-point • FD-to-HD sum-rate ratio vs. SIR. 2 • SNR = 10 dB, α = 15 dB, σerr =1
1.4 FD−to−HD sum−rate ratio
Wireless simulation setup:
1.2 1 0.8 0.6
Joint FDP Separate FDP FDP−SIP FDP−SIT Half−duplex
0.4 0.2 −60
S. Huberman
−55
−50
−45
−40 −35 SIR (dB)
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
−30
−25
28/30
−20
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Full-Duplex Illustrative Results MIMO Point-to-Point
2
35
1.9 30
25
1.7 Sum−rate (b/s/Hz)
FD−to−HD sum−rate ratio
1.8
1.6 1.5 1.4 1.3
1.1 1 −2 10
15
10
Joint FDP Separate FDP FDP−SIP FDP−SIT Half−duplex
1.2
20
FDP−SCAMP FDP−DC FDP−SIP FDP−SIT Half−Duplex
5
−1
0
10
10 2
σerr
• Simulation results 2 • FD-to-HD sum-rate ratio vs. σerr • SNR = 10 dB, SIR = −45 dB, α = 18
0
0
5
10
15
20 25 30 Number of Iterations
35
40
45
• Measured data (5 m, 2.5 GHz) • Convergence comparison 2 =1 • α = 19 dB and σerr
dB
S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
29/30
Background and Motivation
Contributions for Half-Duplex (HD) Systems
Contributions for Full-Duplex (FD) Systems
Concluding Remarks
Concluding Remarks Half-Duplex Systems: • Constant offset ASB-MRU • Achieves approx. 98% sum-rate of state-of-the-art • Sufficient conditions for convergence, existence, and efficiency • Method for estimating spectral efficiency using channel statistics
(omitted due to time constraints) Full-Duplex Systems: • MIMO-OFDM FDP transceiver structure • Various algorithm designs (e.g., SCAMP, DC, separate, SIP, SIT) • Point-to-point: 1.6 to 1.8 times spectral efficiency of HD • Point-to-multi-point: 1.2 to 1.3 times spectral efficiency of HD • Analytical expressions for loss in sum-rate due to residual
self-interference (omitted due to time constraints) =⇒ 3 patent applications, 8 journal papers, 7 conference papers, 1 book chapter S. Huberman
Dynamic Resource Allocation Techniques for Half- and Full-Duplex Systems
30/30