Title: Mitigation of Unknown of Narrowband Interference Using Instantaneous Error Updates Arun Batra1, Take
1Mitigation of Unknown of Narrowband Interference
Using Instantaneous Error UpdatesArun Batra1,
Takeshi Ikuma2, Dr. James R. Zeidler1, Dr. A. A.
(Louis) Beex2, and Dr. John G. Proakis11Departmen
t of Electrical and Computer Engineering,
University of California, San Diego2DSP Research
Laboratory ECE, Virginia Tech
Center for Wireless COMMUNICATIONS
Goal To investigate the BER performance of
adaptive equalizers and to determine the time
required to converge to a specified BER in a
multipath and interference limited environment
Motivation/Previous Work
Convergence Properties
- Two common impairments in many communication
systems are the signal distortion caused by
multipath channels and the presence of narrowband
interference. - Reuter Zeidler demonstrated that an NLMS
equalizer can provide lower MSE than the
corresponding Wiener solution and approaches the
performance a nonlinear DFE in a noise
contaminated environment. - Beex Zeidler demonstrated that the NLMS error
signal produces dynamic weight updating that
causes the filter to track the interference as a
result of the nonlinear characteristics of the
algorithm.
- The convergence of the large-µ DDE adaptive
filter is unacceptably slow (at least 2000
symbols for the case of no multipath), hence
there is a need long training periods. - This limits the use of such filters in practical
applications. - When the filter does converge, the adaptive DDE
filter produces a significant BER advantage over
the non-adaptive Wiener equalizer. - One possible remedy is better initialization of
the NLMS weights.
System Model
TABLE 2MEAN BER FOR ALL STRUCTURES AND ALL A
VALUES
MEAN BER FOR ALL STRUCTURES AND ALL A VALUES
Data-Aided Initialization (DAI)
- sn i.i.d. QPSK signal
- ik pure exponential
interferer - nk additive white Gaussian noise
- a multipath coefficient
- Propose to initialize the decision-directed mode
NLMS algorithm with an estimate of the Wiener
weights over the training period. - Note that there is no adaptation during the
training phase. - With a training period of only 250 symbols, BER
of 10-4 is achieved, which is superior to the
non-adaptive Wiener equalizer.
Equalizer Structures NLMS Nonlinear Effects
NLMS Algorithm
Equalizer Performance
- The DAI-DDE is compared with the DAI-DFE and the
non-adaptive Wiener equalizer. - The DAI-DDE outperforms the non-adaptive Wiener
equalizer and approaches the bound set by the
nonlinear DAI-DFE.
DFE Block Diagram
DDE Block Diagram
- Both structures are adapted via the NLMS
algorithm. - The nonlinear dynamics of the algorithm can be
seen through a recursive expansion of the weight
update equation, where the weights are also a
function of past input vectors and past errors.