Cooperative MIMO
Communications
Hsin-Yi Shen
January 23, 2009
1
Outline
 Introduction
 Cooperative Diversity
 Our Contribution
 Asynchronous cooperative MIMO communication
 Overhead Analysis
 Cooperative MIMO systems with space-time
block codes (STBC) and code combining
 Cooperative MIMO systems with Multiple Carrier
Frequency Offset
 Conclusion
2
Introduction
 Fading effects and channel variation often degrade data
transmission in wireless environments
 MIMO: degree-of-freedom gain & diversity gains
 However, MIMO requires multiple antennas at
transmitter and receiver

Cooperative diversity => achieve spatial diversity with even
one antenna per-node (eg: MISO, SIMO, MIMO)
 Main idea: recruit nearby idle nodes to assist
transmitting and receiving data

Cooperative MIMO: special case of coop. diversity
 Achieve MIMO gains even with one antenna pernode.
 Eg: open-spectrum meshed/ad-hoc networks, sensor
networks, backhaul from rural areas
3
MIMO vs Cooperative MIMO
Tx
.
Rx
8x8 MIMO => 64 channel estimations required
Source node
Destination node
Transmitting
cluster
Receiving cluster
4
We decompose 8x8 MIMO into a 8x1 MISO problems + soft combining.
=> only 8 per-node channel estimations required
Cooperative MIMO Communication
with Meshed Backhaul Networks
Inter-cluster transmission
Intra-cluster
transmission
BS
BS
BS
BS
user
BS
5
Outline
 Introduction
 Cooperative Diversity
 Our Contribution
 Asynchronous cooperative MIMO communication
 Overhead Analysis
 Cooperative MIMO systems with space-time
block codes (STBC) and code combining
 Cooperative MIMO systems with Multiple Carrier
Frequency Offset
 Conclusion
6
Cooperative Diversity
 Motivation
 In MIMO, size of the antenna array must be several
times the wavelength of the RF carrier
 unattractive choice to achieve receiver diversity in
small handsets/cellular phones
 Cooperative diversity: Transmitting nodes
use idle nodes as relays to reduce multi-path
fading effect in wireless channels
 Methods
 Amplify and forward
 Decode and forward
 Coded Cooperation
 Application: Virtual MIMO
7
Cooperative Diversity Schemes
Amplify and forward
Decode and forward
decode
amplify
0101…
forward
Relay node
Source Signal
N1 bits
Source node
Source Signal
N2 bits
Frame 1
Relay node
Destination node
Source node
Coded cooperation
Frame 2
Relay node
Destination node
Source node
N1 bits
N2 bits
Frame 1
Frame 2
8
Our Design for Cooperative
MIMO
Amplify and forward
amplify
Coded cooperation
Decode and forward
decode
0101…
N1 bits
Frame 2
N2 bits
Frame 1
forward
Relay node
Relay node
Relay node
Tx
Rx
Tx
Tx
Rx
N1 bits
Frame 1
N2 bits
Frame 2
Source node
Tx cluster
Rx cluster
Destination node
9
Outline
 Introduction
 Cooperative Diversity
 Our Contribution
 Asynchronous cooperative MIMO communication
 Overhead Analysis
 Cooperative MIMO systems with space-time
block codes (STBC) and code combining
 Cooperative MIMO systems with Multiple Carrier
Frequency Offset
 Conclusion
10
Cooperative MIMO: Phase 1 & 2
PHASE 1: Source node broadcasts symbol
(to ALL cluster members and destination)
Source node
Tx cluster
Destination node
Rx cluster
PHASE 2: Inter-cluster: Tx-cluster detects & rexmits symbol
… to Rx cluster AND destination.
Source node
Rx cluster
Tx cluster
Destination node
11
Cooperative MIMO: Phase 3
Source node
PHASE 3: Rx-cluster & destination do MISO soft-symbol detection.
Rx-cluster transmits soft symbols sequentially to destination
Tx cluster
Rx cluster
Destination node
Destination: combines the soft symbols from Rx-cluster
Source node
Tx cluster
Rx cluster
Destination node
12
Handling Asynchrony…

Synchronization techniques are required in most current
cooperative schemes



The lack of synchronization may result in inter symbol interference
(ISI) and dispersive channels
Different propagation delays due to distance variation
We allow 1-symbol asynchrony… (see next slide)
13
1-symbol asynchrony
Rx cluster node has to do MISO
soft-detection while tolerating this max
multi-path delay!

Difference between the direct & 1-hop path is at most 1symbol time


Why? the send cluster member is at most ½ symbol time away
from sender
The detect-and-rexmit step is assumed to be near-instantaneous
14
Asynchronous MISO detection
yri(t)
h(t)
DFE
n
Sampling rate
= n*data rate

Multi-path requires equalization. We choose the DFE equalizer.


Soft
Quantization
Handles fast & deep fades well. But linear equalizers can also be
used (instead of DFE)
We tap the channel at n-times the symbol rate.




Even though we control max-delay spread…
… with more randomly positioned Tx cluster nodes,
… more taps allows us to resolve indirect paths in the equalization…
Tradeoff: need to tap faster if many Tx cluster nodes
15
Overall Receiver Structure
(Rx-cluster & destination)
Destination node
Receiving
cluster
16
Bit error rate with different SNR
 Coop MIMO
increase diversity
gain and degrees
of freedom
compared to SISO
 Cluster sizes > 3
 BER curve for
proposed system is
better than 3x3
MIMO system
17
Outline
 Introduction
 Cooperative Diversity
 Our Contribution
 Asynchronous cooperative MIMO communication
 Overhead Analysis
 Cooperative MIMO systems with space-time
block codes (STBC) and code combining
 Cooperative MIMO systems with Multiple Carrier
Frequency Offset
 Conclusion
18
Overhead analysis
 Analysis starts from AWGN channel capacity formula
 Three phases with transmission times t1, t2, t3.
 Total time = t1+t2+t3 or capacity = 1/(t1+t2+t3)
 Then compute capacity ratio with respect to direct Tx
capacity
 Assumptions:
 The size of transmitting cluster is M+1 and the size of
receiving cluster is N+1 (including the source node
and destination node).
 Each node in source cluster transmits with equal
power P/(M+1)
19
Analysis of capacity ratio -Phase I
 Phase I: Broadcasting
Transmission time for Phase I
Source node transmits to cluster
members and destination
Destination node
Source node
20
Analysis of capacity ratio -Phase II
 Phase II: Inter-Cluster Transmission
Transmission time for Phase II
Inter-cluster transmission between transmitting
cluster and receiving cluster
Source node
Transmitting
cluster
Receiving
cluster
Destination node
21
Analysis of capacity ratio -Phase III
 Phase III: intra-cluster transmission in
destination cluster
Transmission time for Phase III
Source node
Intra-cluster transmission for soft symbols
Transmitting
cluster
Receiving
cluster
Destination node
Note: Q is # of bits to represent a hard symbol as soft symbol
22
Capacity ratio
 Total transmission time and the
capacity is
 Thus the system capacity ratio is
23
The relation of capacity ratio and
major system factors
5x5 MIMO
N=4
4x4 MIMO
5x5 Coop
N=3
The size of receiving cluster (N+1)
more important factor for capacity ratio
(than Tx cluster size M+1)
Capacity ratio decreases as SNR
increases.
4x4 Coop
3x3 MIMO
Compared to the equivalent MIMO
case, the capacity ratio is smaller due to
node cooperation overheads
24
Note: Tx cluster size (M+1) & Rx cluster size (N+1), incl. of src/dest
Outline
 Introduction
 Cooperative Diversity
 Our Contribution
 Asynchronous cooperative MIMO communication
 Overhead Analysis
 Cooperative MIMO systems with STBC and code
combining
 Cooperative MIMO systems with Multiple Carrier
Frequency Offset
 Conclusion
25
Cooperative MIMO system with
STBC and code combining

Key Challenges in Cooperative MIMO
 node coordination in sending and receiving group
=>cluster recruiting algorithm and asynchronous scheme


Achieve distributed MIMO gain by utilizing both transmitter
and receiver diversity


distributed space-time coding in senders
data combining in the destination

STBC only change the order of information bits
=> suitable for distributed implementation
Solution
 distributed implementation of space-time block codes
(STBC) in sending group

code combining in receiving group


Use convolution code and Viterbi decoder
provide not only spatial diversity but the MIMO diversity
26
Proposed Design
Step 1: Broadcasting
 Before transmission, the sending and receiving
group have been formed
 The source node encodes information bits by FEC
and broadcasts to select neighbor nodes
 Number of nodes required by STBC is selected
 Gives order for selected helper nodes so each helper
node will choose the corresponding row in space-time
block code (STBC) matrix.
Sending
Group
Source node broadcasts data and sends control
message to destination node to forms receiving group
Receiving
Group
27
Proposed Design
Step 2: STBC MIMO transmission
 The helper nodes in sending group use the
corresponding row in STBC code matrix to change the
permutation of data bits
 Transmit space-time coded data to the receiving group
 Note: STBC is applied properly with distributed
implementation because of knowing exact sending group
size and assigning order to each node
Sending
group
(b) MIMO transmission
Receiving
Group
28
Proposed Design
Step 3: Data Collection & Combining
 Each node in the receiving group decodes the space-time
block coded (STBC) data.
 After decoding for STBC, the helper nodes in receiving
group relay their copies to the destination node.
 The destination detects them as soft symbols.
 Then the destination uses code combining and chooses
the most possible codeword based on received soft
symbols.
Sending
group
(c) Data Collection and Code Combining
Receiving
group
29
BER and energy consumption
As sending/receiving groups increase,
BER decreases faster because of
transmitter and receiver diversity
Although cooperative MIMO communication
has more control-message overhead, the
total power consumption is low due to low
BER and fewer retransmissions
30
Energy Consumption
 Energy for unsuccessful
attempt
 Energy for successful
attempt
Total Energy
Proposed system utilizes both
transmitter and receiver diversity =>
lowest power consumption when
transmission power is the same
31
Outline
 Introduction
 Cooperative Diversity
 Our Contribution
 Asynchronous cooperative MIMO communication
 Overhead Analysis
 Cooperative MIMO systems with STBC and code
combining
 Cooperative MIMO systems with Multiple Carrier
Frequency Offset
 Conclusion
32
Cooperative MIMO Systems with
Multiple Carrier Frequency Offsets

Key challenges in proposed cooperative MIMO system design

Each sending node has individual electronic circuit for
carrier frequency generation



The sending group implements space-time block codes
(STBC) in a distributed manner


Distortion from multiple carrier frequency offsets
Most of current techniques consider single carrier frequency
offset, such Phase Lock Loop (PLL)=> Multiple CFO estimation
is desired
Each receiving node will receive STBC-coded signal under the
distortion of multiple carrier frequency offsets
Solution

Estimation of the multiple carrier frequency offsets


Use uncorrelated pilot symbols
MMSE Detection of space-time block coded (STBC) data under
multiple carrier frequency offsets
33
Using PN sequence as uncorrelated
pilot symbols


Each receiver needs to estimate the multiple carrier
frequency offsets (CFO) from senders
We propose to use pseudo-random noise (PN) sequence as
uncorrelated pilot symbols for multiple CFO estimation




Use shift register to generate PN sequence
Different initial state in shift register=> generate uncorrelated
sequence
Thus each receiver only requires information on shift register
length and initial state of shift register in each sender to obtain
uncorrelated pilot symbols =>suitable for distributed
implementation
To send pilot symbols, source node first decides the length
of shift register and assigns the initial state of the shift
register for each sending node

Include this information in MIMO RTS so receiving nodes can
obtain pilot symbol information
34
Design of estimation algorithm for
multiple CFOs




Sending group starts pilot symbol transmission and all receiving nodes use the
received mixed signal of pilot symbols for multiple CFO estimation
Assume M sending nodes and N receiving nodes
Denote pilot symbols and carrier frequency offset in sending node i as pi and fi
Thus receiving signal at receiving node r is
where n is the symbol index

Then compute the discrete-time Fourier Transform (DTFT) of the received signal

The cross-correlation of the DTFT of receiving signal and the DTFT of pilot
symbols is

The above function has maximum at lag 0 and can use to estimate CFO
35
Iterative estimation algorithm for
multiple CFOs


In each iteration, use estimated information from last iteration and
estimate the multiple CFOs sequentially
Algorithm stops when small estimation error or large # of iterations
36
STBC decoding under Multiple
Carrier Frequency Offset





After obtaining the information of multiple CFOs, the receiving nodes
need to detect receiving signals.
With STBC-coded data x, the received signal at receiving node r, yr, is
given by
,while N is noise and Hr is the matrix of path
gain
The element in position (t,τt(i)) of Hr is the path gain of symbol xi
transmitted at time t by sending node τt(i),
where λ is path-loss component and αi,r is fading gain
Hr become non-orthogonal and time-variant matrix due to impefect
carriers
We propose to use a linear MMSE detector to detect the STBC-coded
data under multiple carrier frequency offsets
37
Detection algorithm of STBC
decoding under multiple CFOs

At time kc, the signal received at receiving node r is

To simplify the computational complexity in receiving node r ,use

The mean square value of detection error is

So the MMSE detector can be rewritten as

Thus the linear MMSE detector is applied to received signal,
and the ith element in the output vector above is detected as xi
38
BER Simulation result


Compare proposed system
with a) Cooperative code
combining without STBC
and b) Cooperative MIMO
systems without code
combining
Proposed system has best
performance


No full transmitter
diversity guaranteed due
to multiple CFO and nonorthogonal path gain
matrix
But STBC coding, FEC
code combining and the
proposed linear MMSE
detector still improves
BER performance.
39
Comparison of simulation result
with different estimation algorithms




Compare with results of
no CFO estimation and
non-iterative estimation
No CFO estimation:
cannot detect symbols
due to distortion of
multiple carrier
frequency offsets
Non-iterative estimation:
cannot precisely
estimate when multiple
senders
Iterative estimation:
estimate precisely even
under multiple senders.
=> Performance of
proposed iterative
algorithm is significantly
better.
40
Simulation result of energy
consumption



We compare the energy
consumption in different
system design
Energy consumption in
cooperative FEC system is
lower than it in cooperative
relay because cooperative
code combining improves
BER performance and
require less retransmission.
The proposed system has
lower energy consumption



Proposed system uses more
control messages in node
coordination
But it also has better BER
and require less
retransmission
Thus proposed system
provides reliable low-power
transmission
41
Conclusion


Cooperative communication systems achieve lower
transmission power, extend battery life, and improve network
connectivity and throughput
Our works consider to design the cooperative MIMO
communication step by step






Cluster recruiting algorithm: form clusters for cooperative MIMO
communication (shown in the thesis)
Asynchronous cooperative transmission: deal with the
synchronization problem in sending nodes
Overhead analysis: Consider the system overhead and
performance analysis
Cooperative MIMO with STBC and code combining: fully utilize
both transmitter and receiver diversity to achieve MIMO gain
Cooperative MIMO system with multiple carrier frequency offset:
Consider the distributed senders and provide CFO estimation and
signal detection scheme for proposed system design
Theoretical analysis and formulas are provided in dissertation
42
Related Publications






IEEE DCDIS, Guelph, Canada, July27-29, 2005, “Cluster Recruiting for Ad
Hoc Cooperative Networks,” Hsin-Yi Shen, Babak Azimi-Sadjadi, and
Alejandra Mercado
IEEE WiOPT, April 16-20, 2007, Limassol, Cyprus, “Asynchronous
Cooperative MIMO Communications,” Hsin-Yi Shen and Shivkumar
Kalyaraman
IEEE Globecom 2007 Ad-hoc and Sensor Networking Symposium Globecom 2007 Ad-hoc and Sensor Networking Symposium "A MAC
Protocol for Cooperative MIMO Transmissions in Sensor Networks"
Haiming Yang, Hsin-Yi Shen, Biplab Sikdar
IEEE Globecom 2008 Ad Hoc, Sensor and Mesh Networking Symposium IEEE Globecom 2008 Ad Hoc, Sensor and Mesh Networking Symposium "A
Distributed System for Cooperative MIMO Transmissions" Hsin-Yi
Shen, Haiming Yang, Biplab Sikdar, Shivkumar Kalyanaraman
The 28th IEEE International Conference on Computer Communications INFOCOM Mini-Conference, "A Threshold Based MAC Protocol for
Cooperative MIMO Transmissions", Haiming Yang, Hsin-Yi Shen, Biplab
Sikdar, Shivkumar Kalyanaraman
Hsin-Yi Shen, Shivkumar Kalyanaraman and Biplab Sikdar, “Asynchronous
Cooperative MIMO Communications: System Design and Overhead
Analysis”, submitted to IEEE IEEE Transactions on Wireless
Communications
43
Comparison of Cooperative
Diversity Scheme

Decode and Forward

Simple and adaptable to channel condition (power allocation)
If detection in relay node unsuccessful => detrimental for detection in
receiver (adaptive algorithm can fix the problem)
Receiver need CSI between source and relay for optimum decoding



Achieve full diversity
Performance better than direct transmission and decode-and-forward
achieve the capacity when number of relays tend to infinity





transmit incremental redundancy for partner
Automatic manage through code design
no feedback required between the source and relay
Rely on full decoding at the relay => cannot achieve full diversity!
Not scalable to large cooperating groups.





Amplify and Forward
Coded Cooperation
Other methods are proposed to use spatial diversity by node
cooperation
44
Energy consumption analysis

The energy consumption for an unsuccessful transmission attempt is
And energy consumption for a successful transmission is
where Emrts, Emcts, Eack, Errts and Escts are the energy spent on sending
MRTS, MCTS, ACK, RRTS and SCTS packets.
Ecol: energy spent by each receiving node during data collection.
M, N :size of source and destination clusters, respectively
Ebr: energy spent on broadcasting data to nodes in the sending group.
Edata: energy spent on data transmission between sending/receiving
groups.
 Assume the length of all control messages is Lc and the size of a data
packet is L.
 The data rate is R and a convolutional code with rate Rc
45
Energy consumption analysis- Cont’

Thus the equation of unsuccessful and successful transmission can be
rewritten as

And the total energy consumption for transmission in cooperative
MIMO system is
where Pe is the packet error probability.
46