Upload
others
View
7
Download
3
Embed Size (px)
Citation preview
Blind Interference Cancellation for the Downlink of CDMA Systems
Dragan Samardzija, Narayan Mandayam, Ivan Seskar
Wireless Information Network Laboratory (WINLAB), Rutgers, the State University of New Jersey, 73 Brett Road, Piscataway NJ 08854, USA
Downlink of CDMA Systems
• Downlink of CDMA Systems• Getting more important in the case of data traffic
• Asymmetric nature of internet traffic requires higher data rates for the downlink
• Mobile does not have knowledge of number of users (interferers) or their parameters
• Multiple-access interference (MAI) • Intercell MAI (usually a few strong interferers)• Intracell MAI (considered orthogonal)
Multiple-access Interference
• How to combat the MAIa. Resource management: power control and
admission controlb. Receiver beamforming: requires multiple-
antenna receiverc. Mitigate interference: multiuser detection
(MUD) and interference cancellation (IC)
Our Approach
• Blind Successive Interference Cancellation (BIC)
• Blind approach: the absence of any prior information about the interfering signals
• Based on maximum mean energy criterion (MME), using 2nd order statistics
Successive IC - flow chart
Start detection, K=1
Detect contribution of the component K in the received baseband signal
Reconstruct the contribution of the component K
Cancel the reconstructed contribution
Stopping rule
K=K+1
Interference parameters: prior knowledge or estimates
no
Detect the desired useryes
Maximum Mean Energy (MME) Optimization Criterion
• Find a vector v that maximizes the mean energy (ME) as
v = argmax{ME = E[(rTu)2]}u
Received vector: r = Akbksk + n
Ak,bk,sk are amplitude, info. bit and signature sequence of the kth user, respectively. n is additive noise.
• Constrain the vector v such that vTv=1
∑=
L
1k
MME Optimization Criterion, contd.
• Maximizing the ME, with respect to v, results in solution that satisfies
E(rrT) v = λ v
• Note that v and λ are an eigenvector and an eigenvalueof the matrix Rr = E(rrT)
• Remove the desired user contribution from Rr as
Ri =Rr – (A1)2 s1s1T
Note that no knowledge is required of the desired user’s bit decision. Now, the MME criterion is applied for Ri
MME Optimization Criterion, contd.
• The eigenvector of Ri that corresponds to the maximumeigenvalue (λmax) is the vector that maximizes the ME (mean energy). That vector is denoted as vmax (the maximizer of ME).
• If the contribution of vmax is removed from the matrix Ri as follows: R’i= Ri - λmax vmaxvT
max , then the eigenvector v’maxthat corresponds to the maximum eigenvalue of R’i is the same as the eigenvector that corresponds to the second largest eigenvalue of Ri.
Application of the MME CriterionBlind Interference Cancellation
• Receiver executes the following steps:1. Estimate Rr (input covariance matrix)2. Remove the desired user contribution in Rr i.e.,
determine Ri
3. Find the maximizer vmax for Ri
4. Remove the contribution of vmax in Ri
5. Cancel vmax from the input vector6. Execute steps 3-5 repeatedly for successive
cancellation of interference components or perform detection of desired user
Estimation of the covariance matrix Rr
Reconstruction of the desired user contribution A1
2s1s1T
+
Ri = Rr – A12 s1 s1
T
–
+
r
Matrix flow
Vector flow
IC stage 1 …
Detection of the desired userb1= sgn(s1
T r’L )
r r’1 r’L
R’i 1 R’i L-1
r’L-1
Evaluation of themaximizer vmax(using the ED)
+– λmax vmax vmax
T
+
Reconstruction of themaximizer contribution in the received signal
+
– (rT vmax) vmax
+
r
vmax vmax
Prevention of excessive cancellation of the desired user
r
r’
r’1
Ri R’i 1
IC stage 1
S1
1
2
Optional block
IC stage L
Signal Space Interpretation
Example: 5 users in 3D space
v1…3 – Orthogonal basis for interference subspace
s2…5 – Interferer signature sequencesA2...5 – Interferer amplitudesr – Received vectorsmax ~ v1 – ME maximizersmin ~ v2 – ME minimizer
The maximization scheme determines a vector which captures most (on average) of the energy of the interference.
V1
V2
V3
A2S2
A3S3
A5S5
A4S4
r
Smax
Smin
Simulation Results
• Synchronous CDMA system, randomly generated signature sequences, processing gain G=64
• Case 1: Number of users L=16 , interferers have the same energy: Ai
2/A12=25, i=2...16
• Case 2: Number of users is L=4, the energy of interferers is 16 times higher than in Case 1: Ai
2/A12=400, i=2...4
• Covariance matrix is estimated as
Rr=E(rrT) ~ r(i) rT(i) /n
• Comparison with blind MMSE where copt = s1
∑=
=n
1ir
^R
^
rR1−
Why Does It Work Better?
§ It is still based on the second order statistics, but it outperforms the blind MMSE
§ EstimationRr=E(rrT) ~ r(i) rT(i) /n
Introduces specific distribution of eigenvector and eigenvalue estimation errors
§ BIC-MME receiver deals with the estimation errors more efficiently than other well known blind receivers
∑=
n
1i
Conclusions
• A blind SIC receiver was derived using the MME criterion
• Performance gains over the blind MMSE scheme are greater with smaller number of samples (for estimation of the covariance matrix)
• Particularly effective for a system with a few very strong interferers
• This may be a very viable solution for implementation on the downlink
• Future study: fading channels, packet traffic
Ongoing Research: Implementation of the Advanced Receivers
Blind SIC Implementation Issues
a. Iterative solutionsb. Suitable for the pipelining (i.e., parallel
executions)c. Number of the SIC stages directly
corresponds to the performance, i.e., it is easy to control tradeoff between performance vs. complexity
Ongoing Research: Implementation of the Advanced Receivers
WINLAB Wireless Testbed: Core Technologies1. DSP
a. Multiprocessor systems (parallelism, pipelining…)
§ Processor communication: comm. ports vs. global access memory
b. DSP as a controller for reprogrammable hardware2. FPGA
a. Parallelism, pipelining…
b. Adoption of DSP functionalities (MPY+ADD):
§ Implementation of vector and matrix operations (vector multiplications, matrix inverse, eigen-decomposition)
§ Performance vs. complexity evaluations
Winlab Testbed-SW
System Level Simulations ( SPW, Matlab…)
HW/SW partitioning
DSP Simulation (SPW-CGS)
Partitioning (SPW-Muliprox)
DSP Compiler
DSP Boards
VHDL Description (SPW-HDS)
Synthesis (VSS)
FPGA Place and Route (Xilinx Foundation)
APTIX/FPGA Board
Winlab Testbed-HW
VME Bus Adapter
MIX Baseboard CPU (MC 68020)
Quad DSP TMS320C40 Processor
MIX
Bu
s
VM
E B
us
Dual A/D & Multibandreceiver (Pentek 6472 &
4272)
PIOFPGA
(Aptix/Xilinx)
Dual A/D (AD9042ST)
Dual D/A (AD9742XR)
RF Frontend
n Problems that have not been efficiently solved in the case of conventional linear detectors
a. Presence of highly correlated and/or linearly non-independent interferer signature sequences (case of highly loaded system)
b. Multipath c. Mismatch
§ Our solution: Nonlinear blind interference cancellation using higher order statistics.
a. Blind equivalent to nonlinear centralized SIC b. Related to blind source separation
Ongoing Research: Nonlinear Blind SIC schemes