Motivation
BER depends on bit energy—not on the
CDMA bandwidth
Large bandwidth signals are
less sensitive to multipath fading
Instructor: Mary Ann Ingram
less vulnerable to jamming
ECE 4823 can be concealed
can share a common bandwidth without
interfering with each other
Direct Sequence Spread
Code-Division Multiple Access Spectrum (DS-SS)
Allows multiple users to share same DS-SS is one popular way to make the
bandwidth at the same time noise-like waveforms for CDMA
Each user’s waveform is like an Maximal-length shift registers make
independent noise random process binary sequences that have noise-like
Interference appears as white noise properties
Matched filter pulls out desired user’s m-stage shift register produces a sequence
with a period of length 2m-1
waveform, suppresses interference
DS-SS Baseband Waveform Chips
Binary noise sequence is mapped to a chip The spreading sequence comprises
spreading sequence of +/- 1’s chips (very short pulses) with width TC
Each user gets a different spreading
There are an integer number of chips
sequence
Baseband for each data symbol
Waveform
data symbol
0 TS 2TS 3TS 0 TS 2TS 3TS
1
Information
-1 chip width
waveform TS
Short code
example
1
Codes for Different Users M-Sequences
Their cross-correlation is nearly zero: “Maximal-length” or m-sequences are a
well-known class of spreading
sequences
user 1
Generated with a linear feedback shift
∫ (•)dt
TS
≈0
0 register
user 1
A register of length m generates a code
N long, where N=2m-1
Autocorrelation of the M-
Generating M-Sequences Sequence
The pi’s are the coefficients of a primitive Very much like an impulse
polynomial
0 +1 1
ai ai-1 ai-2 ai-3 ai-4 ai-m+1 ai-m
1 -1
p1 p2 p3 p4 p1m-1
-1/N
Processing Gain Signal Model for k-th User
The number of chips per symbol is the 2 ES
processing gain (PG) sk (t ) = mk (t ) pk (t ) cos(2πf c t + θ k )
This is also
TS
BSS ES = symbol energy
PG =
B mk(t) = information waveform for k-th user
where BSS and B are the bandwidths of pk(t) = spreading sequence for k-th user
the chips and the data symbols, B = bandwidth of mk(t)
respectively. Usually, BSS >> B BSS = bandwidth of pk(t)
2
Received Signal Model—No
Multipath Correlator Receiver
K Assume user 1’s delay, τ1, is known
r (t ) = ∑ sk (t − τ k ) + n(t )
k =1
To receive the signal of user 1,
correlate the received signal with user
Assume K users and that the k-th user’s 1’s spreading sequence delayed by τ1
signal is delayed by τk iTS +τ 1
Z i(1) = ∫ r (t ) p1 (t − τ 1 ) cos(2πf c (t − τ 1 ) + θ1 )dt
( i −1)TS +τ 1
iTS +τ 1 K
=∫
( i −1)TS +τ 1 ∑ sk (t − τ k ) + n(t ) p1 (t − τ 1 ) cos(2πf c (t − τ 1 ) + θ1 )dt
k =1
Multi-user Interference Simplified Correlator Output
The receiver correlates to the code of the Z i(1) = I i(1) + ξ
desired user
Every undesired user’s code has a small = I i(1) + Y + η
amount of residue because of imperfect
orthogonality Y is the multiple access interference
The multi-user interference from each user is part
approximated as a Gaussian RV based on a Y is N (0, σ ξ ) , based on Central Limit
2
Central Limit Theorem argument Theorem
Contribution of lots of chips
η is the thermal noise part
Random Sequence Model BER for BPSK Assuming AWGN
The analysis of the BER for DS-SS assumes
that the K-1 interfering spreading sequences 1
are random and N chips long Pe = Q
The BER is obtained by averaging over all K − 1 No
3N + 2 E
possible spreading sequences, including the
b
desired sequence the “3” comes from an assumption of
Therefore, σ ξ > 0 , σ ξ is proportional to K-
2 2 chip and phase asynchrony; this is
dropped if synchronous
1, and σ ξ is inversely proportional to N
2
Reduces to standard BPSK BER
expression when K=1
3
Graceful Degradation Narrowband Interference
Stuber 2000
Unlike TDMA, Interference signal is spread and then filtered
CDMA BER Complex envelope of desired signal Filter
increases matched
to TS-long
0 TS 2TS 3TS 0 TS 2TS 3TS
gradually as 1
pulse
more users TS
-1
are added Complex envelope of desired signal Filter
matched
0 TS 2TS 3TS TS 2TS 3TS to TS-long
0
pulse
1
Interference bandwidth
-1 spreads out to BSS, but
TS bandwidth of filter is only B
Baseband Tapped-Delay Line Model
of Received Complex Envelope Statistical Models of Tap Gains
Under the wide-sense stationary uncorrelated
~
sk (t ) 1
BSS
1
BSS
1
BSS L scattering (WSSUS) assumption, the tap
gains are uncorrelated complex RVs
Complex
tap gains
g1 g2 g3 L gL A reasonable model for the tap gain
magnitudes, αl, is Rayleigh with exponentially
Complex decreasing mean square values
noise z (t ) Σ
jφ l
~
{ }
E α l2 = Ce −l / β
Small β , small
delay spread
gl = α l e r (t ) L
r (t ) = Re ~ {
r (t )e j 2πf ct } s.t. ∑ Ce
l =1
−l / β
=1
Correlator (RAKE) Receiver The Decision Variable
autocorrelation
of spreading sequence
~ The RAKE receiver output is
sk* (t ) 1
BSS
1
BSS
1
BSS L 1
BSS L L −1 L −1− k
µk = 2ε ∑ α l2 + 4ε ∑ ∑Y i ,i + k φ p ( k ) + n~k
Conjugate
path gains
g1* g*2 g *3 L g *L l =1 k =1 i =1
not needed if
Σ modulation is
where the self-interference is
εk
equal energy
{
Ym ,l = Re g m g l* }
~
r (t ) X Decision
- Variable
Re {∫ (•)dt }
0
T +
Σ µk and generally non-Gaussian and correlated
4
BER BPSK DS-SS BER Curves
[Stuber, 2001]
For DS-SS-BPSK, and assuming ideal Channel has 4
speading sequence (impulse autocorrelation), taps β
then L
1 γm RAKE has 4 β
Pb = ∑ Am 1 −
β
taps
2 l =1 1+ γ m
L
γm
Am = ∏
l =1,l ≠ m m − γ l
γ
γm =
(1 − e )e
−1 / β −m / β
γb
e −1/ β − e −( L +1) / β
Compare to 4-th Order Spatial Sliding Correlator RAKE
Diversity Receiver
[Stuber, 2001]
The received signal is split into M
β
β
branches (M could be less than L)
β
Each branch signal is weighted with the
small delay spread conjugate of the path gain (like MRC)
yields no diversity
Each weighted branch signal is
β=1 not
as good
correlated with a differently delayed
as L=3 version of the spreading sequence
Sliding Correlator RAKE
Receiver; Example: M=3 Summary
CDMA allows efficient use of spectrum
~
r (t ) τ 2 −τ1 τ 3 −τ 2 by putting all users on top of each other
in time and frequency
g a* g*b g *c 3 RAKE
“fingers” Graceful degradation as traffic increases
Σ Robust against interference
RAKE receiver provides some fading
pk (t )
•Correlators “hunt” for
best delays mitigation that depends on delay spread
•for M<L, performance
won’t be as good