Channel Estimation

1
 Table of Contents
Introduction to Channel Estimation
Generic Pilot Based Channel Estimator
Uplink Channel Estimator
Downlink Channel Estimator
WCDMA Channel Estimation Techniques
Simple Average
Weighted Multi Slot Averaging (WMSA)
α-Tracker
Interpolation
Fast Channel Estimation

2
Introduction to Channel Estimation

3
 Introduction
Complex channel estimation (i.e., estimation of channel gain,
which includes phase and amplitude) performed for each
individual RAKE fingers is required for coherent detection
(Maximal Ratio Combining).
Complex channel estimation is performed with the assistance of
known transmitted pilot symbols.
The accuracy of the channel estimation is crucial for RAKE
receiver performance, and it depends on the pilot channel energy,
the channel estimation algorithms, and the environment
conditions.
In particular, mobile speed is required for a variety of channel
estimation algorithms.

4
 Pilot Symbol Transmission
The pilot symbols can be transmitted in two basic ways:
In the case of dedicate pilot channel scheme, system has one
physical channel fully dedicated to pilot symbol transmission.
E.g. Common Pilot Channel, CPICH, in downlink of
WCDMA.
Another option is to insert pilot symbols into the data stream
(time multiplexed pilot symbols).
E.g. DPDCH/DPCCH in uplink of WCDMA.

5
 Dedicate Pilot Channel Scheme
One possible phase estimation architecture based on a dedicate
pilot channel is shown in the following figure:

Chip Rate Symbol Rate

Wideband Correlator Narrowband

I/Q signals (Data Channel) Signal
*
Correlator
LPF
(Pilot Channel)
The output of the channel estimation is filtered by a low pass filter
(LPF), whose bandwidth should be made adjustable to the
Doppler frequency.

6
 Time Multiplexed Pilot Symbols
In the case of inserted pilot symbols, there are two well-
known channel estimation techniques:
Decision feedback
Interpolation
Decision feedback scheme
In this approach, hard decisions (decision of ±1) are made
for the non-pilot (data) symbols.
The hard decisions may be made with the help of pilot symbols.
The data symbols are feedback, together with the pilot
symbols, to form a continuous “pilot symbols” for channel
estimation.
The output of channel is again filtered by a LPF which
should be made adaptive to the Doppler frequency.
7
 Time Multiplexed Pilot Symbols
If hard decision results in small error rate, decision
feedback schemes typically lead to better performance.
Interpolation
In this scheme, since data symbols and pilot symbols are
time multiplexed, channel estimations are first obtained for
the pilot symbols and are then interpolated to obtain the
channel gain of data symbols for coherent combining.
If the outputs of hard decision results in a high data error
rate, which might due to small signal energy, severe fading
effect, or high interference level, the decision feedback
performance starts to degrade. Under such scenario,
interpolation scheme might perform better since it uses
only the pilot symbol energy for estimation.
8
Generic Pilot Based Channel Estimator

9
 Generic Pilot Based Channel Estimator
In making channel estimation, we assume that channel
condition is constant over some time interval.
The complex channel estimation can be obtained by
multiplying the “recovered pilot symbols” by the known pilot
pattern.
In uplink direction, channel estimation is made with the help
of pilot symbols within the DPCCH.
Channel estimation is typically based on simple block
averaging.
Interpolation is typically performed between pilot blocks to
estimate the channel status.
In downlink direction, channel estimation is made using the
dedicated pilot channel, called Common Pilot Channel
(CPICH).
10
 Uplink Channel Estimator
Variables are defined as follows:
P : Pilot symbols ( ±1) .
N p : Number of Pilot symbols.
Dd : DPDCH symbols at UE transmitter.
Dc : DPCCH symbols at UE transmitter.
S I + jSQ : Complex scrambling code.
Cd : DPDCH channelization code.
Cc : DPCCH channelization code.
hI + jhQ : Impulse response of complex fading channel.
Rd : Received DPDCH data symbol at BS.
Rc : Received DPCCH control symbol at BS.
11
 Uplink Channel Estimator
At UE, data is BPSK modulated. The spread symbols at the
UE transmitter are:
Dd ⋅ Cd + jDc ⋅ Cc ≡ DI + jDQ
After scrambling:
(D I + jDQ ) ⋅ ( S I + jSQ ) = ( DI ⋅ S I − DQ ⋅ SQ ) + j ( DI ⋅ SQ + DQ ⋅ S I )
At base station, the received baseband signal is:

⎣( DI ⋅ S I − DQ ⋅ SQ ) + j ( DI ⋅ SQ + DQ ⋅ S I ) ⎦ ⋅ ⎣ hI + jhQ ⎦
⎡ ⎤ ⎡ ⎤
At base station, after descrambling:
1
⋅ ( S I − jSQ ) ⋅ ⎣⎡( DI ⋅ S I − DQ ⋅ SQ ) + j ( DI ⋅ SQ + DQ ⋅ S I ) ⎦⎤ ⋅ ⎡⎣ hI + jhQ ⎤⎦
2
= ( DI ⋅ hI − DQ ⋅ hQ ) + j ( DI ⋅ hQ + DQ ⋅ hI ) = ( DI + jDQ ) ⋅ ( hI + jhQ )
12
 Uplink Channel Estimator
After despreading, we obtain:

DPDCH: Rd = Dd ⋅ ( hI + jhQ )
DPCCH: Rc = Dc ⋅ ( −hQ + jhI )
The channel condition information is estimated from the
received pilot symbols:

1
l
hI = ⋅ ∑ Im { Rc } ⋅ P
NP Np
−1
l
hQ = ⋅ ∑ Re { Rc } ⋅ P
NP Np
where average is performed over Np pilot symbols.
13
 Uplink Channel Estimator
After interpolating the channel condition information for the
DPCCH and DPDCH, channel compensation is performed by:
hld ≡ hlI + jhlQ
m
{ l *
} { (
Dd = Re Rd ⋅ hd = Re Rd ⋅ hlI − jhlQ )}
= Re { Rd } ⋅ hlI + Im { Rd } ⋅ hlQ

(
Estimated DPDCH = Dd ⋅ hI ⋅ hlI + hQ ⋅ hlQ )
hlc ≡ − hlQ + jhlI
m
{
l *
} { (
Dc = Re Rc ⋅ hc = Re Rc ⋅ − hlQ − jhlI )}
= − Re { Rc } ⋅ hlQ + Im { Rc } ⋅ hlI

(
Estimated DPCCH = Dc ⋅ hQ ⋅ hlQ + hI ⋅ hlI )
14
 Downlink Channel Estimator
Variables are defined as follows:
Pi + Pq : Complex Pilot symbols at BS ( Pi = Pq ≡ P ) .
N p : Number of Pilot symbols (10 Symbols/Slot).
Di + Dq : Complex DPCH symbols at BS transmitter.
S I + jSQ : Complex scrambling code.
CI + jCQ : Complex channelization code ( CI = CQ ) .
hI + jhQ : Impulse response of complex fading channel.
Rd : Received DPCH data symbol at UE.
RPI = RP + jRP : Received CPICH symbols at UE.
Note that the I and Q branches of either DPCH or CPICH are spread to
the chip rate by the same channelization code.
15
 Downlink Channel Estimator
At BS, data is QPSK modulated. The spread symbols at the
BS transmitter are:
DI + jDQ ≡ ( Di ⋅ CI ) + j ( Dq ⋅ CQ )
After scrambling:
(D I + jDQ ) ⋅ ( S I + jSQ ) = ( DI ⋅ S I − DQ ⋅ SQ ) + j ( DI ⋅ SQ + DQ ⋅ S I )
At UE, the received baseband signal is:
⎡( DI ⋅ S I − DQ ⋅ SQ ) + j ( DI ⋅ SQ + DQ ⋅ S I ) ⎤ ⋅ ⎡ hI + jhQ ⎤
⎣ ⎦ ⎣ ⎦
At UE, after descrambling:
1
⋅ ( S I − jSQ ) ⋅ ⎣⎡( DI ⋅ S I − DQ ⋅ SQ ) + j ( DI ⋅ SQ + DQ ⋅ S I ) ⎦⎤ ⋅ ⎡⎣ hI + jhQ ⎤⎦
2
= ( DI ⋅ hI − DQ ⋅ hQ ) + j ( DI ⋅ hQ + DQ ⋅ hI ) = ( DI + jDQ ) ⋅ ( hI + jhQ )
16
 Downlink Channel Estimator
After despreading, we obtain (note that CI=CQ):
DPCH: Rd = ( Di ⋅ hI − Dq ⋅ hQ ) + j ( Di ⋅ hQ + Dq ⋅ hI )
= ( Di + jDq ) ⋅ ( hI + jhQ )
CPICH: RPI = RP ( hI − hQ ) + jRP ( hQ + hI )
The channel condition information is estimated from the
received pilot symbols:
1
l
hI = ⋅ ∑ {Re { RPI } ⋅ P + Im { RPI } ⋅ P}
2NP Np
1
l
hQ = ⋅ ∑ {Im { RPI } ⋅ P − Re { RPI } ⋅ P}
2NP Np
where average is performed over Np pilot symbols.
17
 Downlink Channel Estimator
After obtaining the channel condition information for the DPCH,
channel compensation for the I branch is performed by:
hl ≡ hl + jhl
d I Q

l
{ l *
} { (
Di = Re Rd ⋅ hd = Re Rd ⋅ hlI − jhlQ )}
= Re { Rd } ⋅ hlI + Im { Rd } ⋅ hlQ
Estimated DPCH I = ( Di ⋅ hI − Dq ⋅ hQ ) ⋅ hlI + ( Di ⋅ hQ + Dq ⋅ hI ) ⋅ hlQ

( ( )
= Di ⋅ hI ⋅ hlI + hQ ⋅ hlQ + Dq ⋅ hI ⋅ hlQ − hQ ⋅ hlI )
For perfect channel estimation ( hl = h , hl = h ) :
I I Q Q

Estimated DPCH I ≈ Di ⋅ ( hI2 + hQ2 )

18
 Downlink Channel Estimator
After obtaining the channel condition information for the
DPCH, channel compensation for the Q branch is performed by:
hld ≡ hlI + jhlQ
m
{ l *
} { (
Dq = Im Rd ⋅ hd = Im Rd ⋅ hlI − jhlQ )}
= Im { Rd } ⋅ hlI − Re { Rd } ⋅ hlQ
Estimated DPCH Q = ( Di ⋅ hQ + Dq ⋅ hI ) ⋅ hlI − ( Di ⋅ hI − Dq ⋅ hQ ) ⋅ hlQ

( ) (
= Dq ⋅ hI ⋅ hlI + hQ ⋅ hlQ + Di ⋅ hQ ⋅ hlI − hI ⋅ hlQ )
(
For perfect channel estimation hlI = hI , hlQ = hQ : )
Estimated DPCH Q ≈ Dq ⋅ ( hI2 + hQ2 )
19
WCDMA Channel Estimation Techniques

20
 Channel Estimation Techniques
Numerous approaches can be used for pilot aided channel
estimation.
We need to take into account at least the following factors:
The performance of the algorithms.
The complexity relative to the current DSP processing power.
In WCDMA, channel estimation typically consists of two steps:
Periodic channel sampling by virtue of known pilot symbols:
the instantaneous channel estimate in each time slot is
obtained by averaging the channel estimates calculated from
each pilot symbol.
Channel tracking (interpolation): the tracking of the time-
varying fading channel is realized through a filter to obtain
the channel estimates from the data symbols.
21
 Channel Estimation Techniques
In general, the channel estimate from each pilot symbol is
provided by multiplying the received pilot symbols rl(n,k) by
the complex conjugate of the known pilot symbols:

hil = rl ( n, k ) ⋅ p* ( n, k ) n = 1, 2,..., N p
where l is the multi-path index, Np is the number of pilot
symbols and time slot index is k.
To minimize the effects of noise and interference, an average
is taken over the number of pilot symbols within a time slot:
Np
1
hl ( k ) =
Np
∑l
i ( n, k )
h
n =1

22
 Simple Average
This is a simple but effective method to estimate the channel
complex gain.
Simple average technique averages the received pilot symbols
within one time slot.
Simple average technique reduces the sampling rate from
symbol rate to slot rate.
0 Pilot Pattern

Pilot Field Slot

Timing Timing

Channel
DPCCH Symbols ∫ Estimate

23
 Weighted Multi Slot Average (WMSA)
H. Andoh, M. Sawahashi, and F. Adachi, “Channel Estimation
Using Time Multiplexed Pilot Symbols for Coherent Rake
Combining for DS-CDMA Mobile Radio”, pp. 954—958, IEEE
1997.
Pilot symbols (n+1)th Slot (n+2)th Slot

hil ( n − 2 ) hil ( n − 1) hil ( n ) hil ( n + 1) hil ( n + 2 ) hil ( n + 3)

α ( −2 ) α ( −1) α ( 0) α (1) α ( 2) α ( 3)

hl ( n )

24
 Weighted-Multislot Averaging
First, instantaneous channel estimation is performed by using
the Np pilot symbols belonging to the nth slot:
N p −1
1
hil ( n ) =
Np
∑ r ( n, m )
m=0
l

The channel estimate to be used for the n-th slot data is obtain
by using 2K consecutive channel estimates:
K
hl = ∑
i =− K +1
α ( i ) ⋅ hil ( n + i )
where α(i)≤1 is the weighting factor.
By properly choosing weighting factors, α(i), accurate
channel estimation is possible, particularly in slow fading
environments.

25
 Alpha Tracker
Y(n)=(1-α)·X(n-1)+α·Y(n-1)
D 1-α Y(n)
X(n) X(n-1) Y(n-1)
α D

The alpha tracker can be thought of as an IIR low-pass filter.

The difference equation describing the alpha track is:

y ( n ) = (1 − α ) ⋅ x ( n − 1) + α ⋅ y ( n − 1)
The filter can be described in the z domain:

Y ( z ) (1 − α )z −1
H (z ) = =
X ( z ) 1 − αz −1
26
 Alpha Tracker

Substituting z with ejω, we can obtain the frequency response:

1−α
H (e jω
)=
(1 + α − 2α cos ω )
1
2 2

sinω
∠H e ( )
jω
= tan−1

α − cosω

The cut-off frequency ωc of the IIR filter is given by:

− α 2
+ 4α −1
ωc = cos−1

2α
27
 Other Channel Estimation Techniques
There are a variety of channel estimation techniques that can
be applied:
Adaptive Filter Techniques
Simon Haykin, “Adaptive Filter Theory,” 3rd Edition, Prentice-
Hall, 1996.
Wiener Filter
B. Lindoff, C. Ostberg, and H. Eriksson, “Channel Estimation for
the WCDMA Systems, Performance and Robustness Analysis from
a Terminal Perspective,” IEEE Vehicular Technology Conference,
Vol. 2, 1999.
Forward Linear Prediction
S. Cacopardi, F. Gatti, and G. Reali, “Channel estimation using
linear prediction for wireless indoor communications,” IEEE ICC
1995.
28
Interpolation

29
 Newton’s Divided-Difference Formula
Assume values of function, f(x), are known at suitably
specialized points, x0, x1, …, xn, which need not be equally
spaced or taken in consecutive order.
Newton’s divided-difference formula:
f ( x ) = f ( x0 ) + ( x − x0 ) f ( x0 , x1 ) + ( x − x0 )( x − x1 ) f ( x0 , x1 , x2 ) + "
+ ( x − x0 )( x − x1 )" ( x − xn −1 ) f ( x0 , x1 ," , xn )
+ ( x − x0 )( x − x1 )" ( x − xn ) f ( x, x0 , x1 ," , xn )
≡ pn ( x ) + rn +1 ( x )
pn ( x ) = f ( x0 ) + ( x − x0 ) f ( x0 , x1 ) + ( x − x0 )( x − x1 ) f ( x0 , x1 , x2 ) + "
+ ( x − x0 )( x − x1 )" ( x − xn −1 ) f ( x0 , x1 ," , xn )
rn +1 ( x ) = ( x − x0 )( x − x1 )" ( x − xn ) f ( x, x0 , x1 ," , xn ) ≡ error term
30
 Lagrange’s Interpolation Formula
Lagrange’s interpolation formula is a polynomial of degree n
(or less) since each term on the right hand side is a polynomial
of degree n.
When x=xi, i=0,1,…n, every fraction except the (i+1)th
vanishes because of the factor (x-xi).

f ( x) =
( x − x1 )( x − x2 )" ( x − xn )
f ( x0 )
( x0 − x1 )( x0 − x2 )" ( x0 − xn )
+
( x − x0 )( x − x2 )" ( x − xn )
f ( x0 ) + "
( x1 − x0 )( x1 − x2 )" ( x1 − xn )
+
( x − x0 )( x − x1 )" ( x − xn −1 )
f ( xn )
( xn − x0 )( xn − x1 )" ( xn − xn−1 )
31
 Forward Gregory-Newton Interpolation
Formula
When the points x0, x1, …, xn on which Newton’s divided-
difference formula is based are regularly spaced with tabular
interval h, it is generally more convenient to express Newton’s
divided-difference formula in terms of ordinary differences.
Let x=x0+rh and xk=x0+kh, we have x-xk=h(r-k) k=0,1,…,n.
Therefore, (x-x0)(x-x1)…(x-xj)=hj+1r(r-1)…(r-j).
Using: ∆ j +1 f 0 ∆f k = f k +1 − f k
f ( x0 , x1 ,..., x j +1 ) =
( j + 1)!h j +1
∆ n f k = ∆ n −1 f k +1 − ∆ n −1 f k
We have, from Newton’s divided difference formula, Forward
Gregory-Newton Interpolation formula:
r ( r − 1) 2 r ( r − 1)( r − 2 ) 3
f ( x ) ≡ f ( x0 + rh ) = f 0 + r ∆f 0 + ∆ f0 + ∆ f 0 +"
2! 3!
32
 Backward Gregory-Newton Interpolation
Formula
Forward Gregory-Newton Interpolation formula is especially
adapted to interpolation for smaller value of x.
Let x=x0+rh and xk=x0-kh, we have x-xk=h(r+k) k=0,1,…,n.
Therefore, (x-x0)(x-x1)…(x-xj)=hj+1r(r+1)…(r+j).
j +1
∆
( ) ( j + 1)!h
Using: f x , x ,..., x f − j −1
0 1 = j +1 j +1

We have, from Newton’s divided difference formula,

Backward Gregory-Newton Interpolation formula:

r ( r + 1) 2 r ( r + 1)( r + 2 ) 3
f ( x ) ≡ f ( x0 + rh ) = f 0 + r ∆f −1 + ∆ f −2 + ∆ f −3 +"
2! 3!

33
 Forward Newton-Gauss Interpolation
Formula
Let x0=x0, x1=x0+h, x2=x0-h, x3=x0+2h, x4=x0-2h, …
j +1
Using: ∆
f ( x0 , x1 ,..., x j +1 ) =
f − j −1
( j + 1)!h j +1
We have, from Newton’s divided difference formula, Forward
Newton-Gauss Interpolation formula:
f ( x ) ≡ f ( x0 + rh )
r ( r − 1) 2 r ( r − 1)( r + 1) 3 r ( r − 1)( r + 1)( r − 2 ) 4
= f 0 + r ∆f 0 + ∆ f −1 + ∆ f −1 + ∆ f −2 + ...
2! 3! 4!
r ( r − 1) 2 ( r + 1) r ( r − 1) 3 ( r + 1) r ( r − 1)( r − 2 ) 4
= f 0 + rδ f1/ 2 + δ f0 + δ f1/ 2 + δ f 0 + ...
2! 3! 4!

where δ f k +1/ 2 ≡ f k +1 − f k .
34
 Backward Newton-Gauss Interpolation
Formula
Let x0=x0, x1=x0-h, x2=x0+h, x3=x0-2h, x4=x0+2h, …
We have, from Newton’s divided difference formula,
Backward Newton-Gauss Interpolation formula:
f ( x ) ≡ f ( x0 + rh )

= f 0 + rδ f −1/ 2 +
( r + 1) r δ 2 f ( r + 1) r ( r − 1) δ 3 f ( r + 2 )( r + 1) r ( r − 1) δ 4 f
0 + −1/ 2 + 0 + ...
2! 3! 4!
We can also have Stirling’s interpolation formula:

r δ f1/ 2 + δ f −1/ 2 r 2 2
f ( x ) ≡ f ( x0 + rh ) = f 0 + ⋅ + δ f0
1! 2 2!
r ( r 2 − 1) δ 3 f1/ 2 + δ 3 f −1/ 2 r 2 ( r 2 − 1) 4
+ ⋅ + δ f 0 + ...
3! 2 4!
35
Fast Channel Estimation

36
 Fast Channel Estimation
In order for the power control to be effective, the TPC symbol
must be decoded with minimum possible delay to adjust the
transmit power so that the overall power control delay should
be one time slot.
Similar comment also applies for the SIR measurement since
SIR must be calculated and compared to a set point to
determine the TPC bits.
The only difference between the fast channel estimation and
the ordinary (slow) channel estimation is whether interpolation
is used or not.
As a result, fast channel estimation is obtained by averaging
the Np pilot symbols within a time slot.
Np
1
hl ( k ) = ∑
N p n =1
hil ( n, k )

37