reliable, high capacity, multipoint, wireless information networks by matthew bromberg ph.d
TRANSCRIPT
Problem Identification Information Networks Require High Capacity,
Reliable, Data Networking Data networks require the transfer of large data files (e.g.
medical imaging) Bandwidth in Hz is scarce and expensive. (Billions required
for nationwide footprint.) Information transfer must be reliable, especially when
human lives are at stake (e.g. military networks, emergency services, police etc.)
Wired Networks are Costly and Impractical wired infrastructure is costly to build and maintain wiring is infeasible in for mobile units and temporary
structures (e.g field hospitals) Wireless is the Only Solution for Soldiers in the Field
Soldier becomes part of wireless network Can integrate with current Land Warrior program
Problem Identification Continued
Wireless Network faces Severe Multipath unknown terrain effects urban environments indoor multipath
Wireless Network faces Degraded Propagation indoor propagation losses canyons and urban environments
Wireless Network faces Interference co-channel interference from other network nodes hostile interference from jammers interference from other, co-channel networks
Wireless Network Must be Secure must have low probability of intercept (LPI) must be secure against infiltration
Multipath Illustration Multipath: wireless signal bouncing off of
terrain buildings etc. Main path blocked causes severe reduction of
signal strength
Remote Unit
1k
2k
User in group 2
Advantages of Proposed Solution
Maximizes Network Capacity Simulations suggest more than an order of magnitude
improvement ( x 35) Exploits Multipath
Can use multipath diversity to multiply capacity Mitigates Interference
Multi-antenna array excises co-channel interference Optimizes Transmit Beamforming
Permits inexpensive remote transceivers Offers reduced interference profile for LPI increases capacity of network
Network is Optimized Locally Power control only needs local information Entire network performance is optimized
Required Transmit Power is Minimized Total network transmit power can be minimized subject to a
capacity constraint Dramatic reduction in required transmit power observed (factor of
40,000)
Network Objective Function Maximize channel capacity flows through network
Optimize maximum theoretical bit rate achievable in network Minimize transmitted power subject to capacity
constraint alternative network performance formulation can use arbitrary bit rate targets based on quality of service
QoS requirementsMaximize InformationFlow into and out ofCut Sets
Basestation
Remote
Advantages of Time Division Duplex
TDD alternates in time transmission and reception over the same frequencies
Channel on uplink nearly the same as downlink RF components shared for transmission and reception Optimal transmit weights easily obtained
Downlink Transmis-
sion
Uplink Transmis-
sion
Downlink Transmis-
sion
Uplink Transmis-
sion
Time
Freq
uen
cy
Exploitation of Channel Reciprocity
Channel Reciprocity asserts uplink channel response is the same (matrix transpose) as the downlink
Can be achieved in TDD networks after transmit/receive gain compensation
d1k1
d1k2
d1kQ
g1k1g1k1
g1k2g1k2
g1kQg1kQ
...
d1k G1k1k
k,k)
i2k
...
W2k
w2k1*
w2k2*
w2kQ*
d1k1
d1k^
^
d1k2^
d1kQ^
......
Channel ReciprocityH21(k; j) = H12(j; k)T (swap 1 and 2 indices above for downlink)
Remote Transmit
Base Receive
X2klink k has Q sublinks
Receiver Model
Information Theoretic Objective Function
Useful metric is mutual information
Represents maximum achievable throughput
maximize mutual information subject to power constraints
decoupled capacity metric assumes linear receiver weights easier to analyze
data processing inequality
Decoupled Capacity Achieves its Upper Bound Linear Beamforming is the best you can do for Gaussian other user interference Best receiver weights easily computed using local statistics
Reciprocity Theorem Reciprocal channels imply the Reciprocity
Theorem
D21(W,G) = D12(G*,W*)
uplink capacity equals downlink capacity transmit with conjugate of receiver weights uplink sum total power also equals downlink total power
(alternative objective function) Transmit weights are easily obtained from
receive weights. Transmit and receive weights only require local
information. (No Network God) Optimizing the receiver ’globally’ optimizes the
entire network! Network is stable and improves at every iteration.
Illustration of Reciprocity Theorem
Receive Beamformer enhances signal of interest (SOI), suppresses interferer
Transmit beamformer enhances signal of interest, offers minimal interference to other nodes in field of view
km
-1 -0.5 0 0.5 1
-
-
-
-
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
km
SOI
Inter1
Inter2
Network Optimality Using Local Information
Receiver computes optimal Wiener beamforming weights using statistics observed at receiver
Optimal transmit weights are proportional to receive weights: g = w*
Optimal power control only requires post-beamforming interference power estimate from other end of linkL (,g, ) gTlocal model of sum of xmit powers)
q Q(m) log(1 + (q)) m (capacity constraint) g = f( (gradient of total xmit power wrt target SINR)(q) = (q) *(q)/(q) (new xmit power is old power times ratio of optimal target SINR divided by achieved SINR)
gradient can be computed by simply estimating post beamformer interference at both ends of link
Locally Enabled, Globally Optimized (LEGO)
Compute Weightsw2 = Rx
2x2
-1 Rx2s
Estimate Transfer Powerh2 =| w2
H Rx2 s /Rss |2 /1
Estimate Interference Poweri2 = Ry2 y2
- 1 h2 Rss
Set gradient: g( k)= i2 (k) i1 (k)/h2
Optimize local model
= arg min L (,g, )1 =i1 /h2 =i2 /h
Base Station (User 2 Node)
Subscriber Unit (User 1 Node)
Compute Weightsw1 = Rx1 x1
-1Rx
1s
Estimate Transfer Power
h2 =| w1H Rx1 s /Rss |2 /2
Estimate Interference Poweri1 = Ry1y
1 - 2 h2 Rss
i1
Interfering SUs
Interfering BSs
Patented Technique
Computations can be concentrated at basestation
Convergence to Theoretical Maximum Capacity
Simulation Parameters: 19 Cells, 1 km radius, 1800 MHz, Hata cost 231 path loss model,
Rayleigh fading, statistically independent antennas, 128 sample block processing, non-blind max-SINR beamform weights, 4 antennas at base, 2 antennas at remote. 1 remote in network.
Rapid convergence to theoretical maximum capacity
0 5 10 15 20 25 300
510
1520
25
Bits
per
Sam
ple
Reverse Link Capacity
0 5 10 15 20 25 305
10
15
20
25
Bits
per
Sam
ple
Forward Link Capacity
Iteration
Max Capacity
Max Capacity
Convergence Example 19 Cell network 1 remote per
cell (per band) Each remote
has links to 2 basestations
8 antennas at each basestation
2 antennas at each remote
2 independent channels
-5 -4 -3 -2 -1 0 1 2 3 4 5
-4
-3
-2
-1
0
1
2
3
4
km
km
LEGO Convergence Easily achieves 5 bps per Hz (could have achieved a lot more) Converges in 15 iterations (40 msec or so) Node transmit power < 20 dBm (Well under unlicensed band
spec.)
0 10 20 30 40 50 60 70 800
10
20
30
40
50
dB
m
Forward Link Xmit Gains
0 10 20 30 40 50 60 70 80-5
0
5
10
15
20
Iterations
dB
m
Reverse Link Xmit Gains
0 10 20 30 40 50 60 70 800
5
10
15
Bit
s p
er
Sa
mp
le
Reverse Link Capacity
0 10 20 30 40 50 60 70 800
5
10
15
20
Bit
s p
er
Sa
mp
le
Forward Link Capacity
Iteration
5 bps/Hz = 10 bits/samp * 50 ksamps/100kHz
Minimize power subject tocapacity constraint metric
Multiplying Capacity by Exploiting Diversity
7 Cell Network.(Reuse pattern of 3).
Max Xmit pow. = 15 dBm.
Equal number of antennas per base and remote.
Number of antennas varied.
LEGO Exploits Multipath, vs Single Path Transmission, Conventional Power Management
-5 -4 -3 -2 -1 0 1 2 3 4 5
-4
-3
-2
-1
0
1
2
3
4
km
km
0 2 4 6 8 10 120
5
10
15
20
25
Number of antennas
Bits
pe
r S
amp
leCapacity vs Num. antennas
LEGO
Single Mode Equal Power
Channel rank allowed to growLEGO never uses more than 5 modes
Multiplying Capacity by Optimal Power Management
19 cell network. Number of users per cell varied.
Maximum achievable worst-case capacity plotted.
0 1 2 3 4 51
2
3
4
5
6
7
8
9
10
Ca
pac
ity, B
its P
er
Sa
mp
le
Users Per Cell (1/Reuse)
Channel Capacity vs. Cell Capacity
LEGO
Const. Power Single antenna
2.5 × capacity of standard power management and 35 × an algorithm that does not exploit reciprocity
Multiplying Capacity Using MultiPoint Networks
4 Cell Network Max Xmit Power 53 dBm Star and Ad-Hoc
Topologies 8 antennas at base, 2 at
each remote
km
-2 -1 0 1 2
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
km-2 -1 0 1 2
-2.5
-2
-1.5
-1
-0.5
0
0.5
1
km
km
0
5
10
15
20
25
30
35B
its
Per
Sam
ple
Star Topology
Ad-Hoc Topology
65% capacity increase
(100% Asymptotic increase)
Increased connectivity multiplies capacity.
Minimizing Transmit Power Experiment Setup
5 Antennas at each base station * 2 Antennas at each remote unit 3 Basestations, 6 Remotes, 2 links per remote LEGO power control, vs Standard vs 1 antenna comparison Transmit power varied, max remote bit rate plotted 6 independent (50 kHz) frequency
channels
-2.5 -2 -1.5 -1 -0.5 0 0.5 1-2.5
-2
-1.5
-1
-0.5
0
0.5
km
km
Standard power control:Constant link transmit powerand constant link receive power at basestation. (similar to CDMA)
1 Antenna case can only use a single link at each remote, and FDMA for co-channel interference
LEGO is 40,000times better.
Reducing Required Transmit Power
To achieve 8.6 bps per Hz requires 25.3 dB or 339 times more power for Standard Power Management
To achieve2.9 bps per Hzrequires 46.1dBor 41 thousandtimes morepower for thesingle antennacase.
Cost of poweramplifiersincreases by thepower squared. -20 -10 0 10 20 30 40 50
0
2
4
6
8
10
12
14
16
18
20
dBm
bps
per
Hz
LEGO
Standard
Single Antenna
Compared to LEGO Performance
COTS Implementation
LEGO permits network operation in the presence of co-channel interference
Network could operate in unlicensed band, at 2.4 GHz 5 GHz and 900 MHz are other possibilities
A large amount of commercial off the shelf hardware (COTS) exists for the unlicensed bands.
Hardware costs can be kept down, following the philosophy of the Army’s Land Warrior program.