SMART ANTENNAS IN UMTS LCR-TDD

Implementation of RSS Beamforming

Arjang Hessamian-Alinejad, Achim Seebens, Tobias Scholand, Admir Burnic,

Guido H. Bruck and Peter Jung

Lehrstuhl für KommunikationsTechnik, Universität Duisburg-Essen, 47048 Duisburg, Germany

Keywords: Beamforming, DSP (Digital Signal Processor), LCR-TDD (Low Chip-Rate Time Division Duplex), Regular

Spatial Sampling (RSS) Smart Antennas, UMTS (Universal Mobile Telecommunications System).

Abstract: In this communication a novel beamforming scheme is presented, which is based on regular spatial

sampling (RSS) of signals. RSS beamforming allows low-complexity algorithms for smart antennas, while

at the same time improving the performance of those systems. An implementation concept is presented,

which shows the feasibility of implementing RSS beamforming on today’s readily available hardware

platforms.

1 INTRODUCTION

In the LCR-TDD (Low Chip-Rate Time Division

Duplex) mode of UMTS (Universal Mobile

Telecommunications System), the application of

smart antennas is anticipated (Holma et al., 2002). A

smart antenna is the combination of an array antenna

with an appropriate signal processing algorithm,

which is applied to the signals coming from or going

to the antenna elements of the array antenna (Lehne

et al., 1999), (Correia,2001). Smart antenna

concepts range from simple switching between

directive antenna elements of an array antenna to

highly-sophisticated adaptive beamforming

techniques, which aim to maximize the signal-to-

interference-and-noise ratio (SINR) for the user of

interest during reception, while in the transmit

direction reducing the interference to other users as

far as possible (Lehne et al., 1999). Beamforming

can be viewed as spatial filtering by adjusting the

beam pattern of the array antenna through the

application of beamforming weights to the signals of

the antenna elements.

Here, we will focus on receiver beamforming. The

majority of beamforming concepts apply array

antennas with

a

K

identical wired antenna elements,

making the analysis and design of the array antenna

very simple. In order to avoid large unwanted side

lobes the antenna spacing must not be larger than

2

λ

, where

λ

is the wavelength of the radio signals

in mind. However, placing the antenna elements too

close increases the coupling between the antenna

elements and broadens the minimum achievable

beam width. Hence, in most arrays the antenna

element spacing is

2

λ

.

First the general concepts of conventional and the

novel Regular Spatial Sampling (RSS) beamforming

shall be considered. In figure 1, a scenario

consisting of three user signals impinging from the

Directions Of Arrival (DOA)

60

β

=°, 120

β

=°, and

240

β

=

°

, onto an array antenna is shown. The array

antenna is a UCA consisting of

a

4K = antenna

elements with an antenna element spacing of

2

λ

.

The conventional beamformer expresses a single

beam for each impinging wave, which are shown in

figure 1 in blue, magenta and red for the signals

impinging from

60

β

=

° , 120

β

=°, and 240

β

=

° ,

respectively. Looking at the blue beam steered in the

direction

60

β

=

° this beam not only receives the

user of interest, but also due to its side lobes the

other two impinging signals. These are considered

as interference in the conventional beamformer.

Likewise it is for the other two impinging signals.

The conventional beamformer can be viewed as a

spatial matched filter as it maximizes the signal-to-

noise ratio (SNR) in the receiver. However, since

the beamformer steers the beam into the directions

of the impinging signals, these directions have to be

estimated. Appropriate DOA algorithms like e.g.

177

Hessamian-Alinejad A., Seebens A., Scholand T., Burnic A., H. Bruck G. and Jung P. (2006).

SMART ANTENNAS IN UMTS LCR-TDD - Implementation of RSS Beamforming.

In Proceedings of the International Conference on Wireless Information Networks and Systems, pages 177-184

Copyright

c

SciTePress

ESPRIT (Paulraj et al., 1986) can become

computationally very complex.

The expression of beams into the DOAs of the

impinging signals can be seen as sampling the space.

As can be seen from the above discussion this is an

irregular spatial sampling, since the beams are

directed into the DOAs of the impinging waves,

which may be any directions. From the processing

of time domain signals it is known that processing of

irregularly sampled signals is often more complex

than the processing of regularly sampled signals.

This is the main idea of the novel RSS beamforming

technique, which represents a regular sampling of

the space (Seebens et al., 2004), (Scholand et al.,

2004).

Figure 1: Conventional beamforming.

Figure 2: Regular spatial sampling (RSS) beamforming.

Figure 3: Burst structure for the LCR-TDD mode of the

UMTS system (3GPP TS 25.221 V5.2.0, 2002).

Figure 4: Block diagram for the Node B transmitter.

Figure 2 shows the concept of RSS beamforming for

the same scenario as above for the conventional

beamforming. Contrary to conventional

beamforming the number of expressed beams does

not depend on the number of users. Instead a fixed

number of beams is expressed, which is suitable to

sample the space. For the circular array with

a

4K

=

antenna elements like the one for the conventional

beamformer there are four independent beams

possible, shown in figure 2 in blue, magenta, red and

violet for the directions

0° , 90° , 180° , and 270° ,

respectively. Because of the beam width and the side

lobes of the beams each beam naturally receives

signals from more than one direction. These are not

considered as interference as in the conventional

beamforming concept, but contribute to the overall

received energy due to maximum ratio combining

schemes. Since the RSS beamformer directs its

beams into predetermined directions it can totally

renounce on DOA estimation techniques. This

allows a low-complexity beamformer with very

good performance.

In this communication the performance of the RSS

beamforming concept is investigated with respect to

WINSYS 2006 - INTERNATIONAL CONFERENCE ON WIRELESS INFORMATION NETWORKS AND SYSTEMS

178

the LCR-TDD mode of UMTS. Especially,

simulation results are shown with a promising

performance. An implementation concept for a

demonstrator is presented, which considers RSS

beamforming with four antenna elements at a Node

B of LCR-TDD of UMTS, demonstrating the

feasibility of implementing RSS beamforming using

today’s conventional hardware.

Figure 5: Block diagram for the Node B receiver.

This communication is organized as follows. In

Sect. 2, the system model for the channel estimation

and the data detection will be presented. Sect. 3 will

provide a novel channel estimation and an

appropriate data detection algorithm, which are both

based on the RSS beamforming concept. In Sect. 4

an implementation concept for a demonstrator will

be presented. The performance of the RSS

beamforming for the LCR-TDD mode of UMTS

will be investigated in terms of simulations in

Sect. 5. Both floating-point as well as fixed-point

simulation results will be presented. Finally, in

Sect. 6 a conclusion of this communication will be

given.

In what follows, the matrix-vector notation is used.

Matrices are denoted as upper case characters in

bold face italics, vectors are lower case characters in

bold face italics. Furthermore, complex-valued

variables are underlined.

()

T

⋅

and

()

H

⋅ denote matrix

and vector transposition and complex conjugate

transposition of a matrix or a vector.

2 SYSTEM MODEL

In this section, the appropriate system models for

channel estimation and data detection are presented.

These models assume the LCR-TDD mode of

UMTS, where each burst consists of two data parts,

data 1 and data 2, separated by a midamble as shown

in figure 3. The two data parts hold the user data and

consist of 352 chips each. The midamble transmitted

in each burst is used for channel estimation and has

a length of 144 chips. To mitigate different signal

run times within a cell between near and far users

each burst is appended by a guard period with a

length of 16 chips. The total block length including

the guard period is thus 864 chips.

First the system model for the channel estimation is

described, which is based on the transmitted

midamble. For each user

,1kk K= a user specific

midamble denoted by the

(

)

11LW+−× vector

() () () ()

(

)

T

12 1

kkk k

LW

mm m

+−

=m is transmitted, where

L

is the midamble length and W denotes the

number of paths. According to the RSS

beamforming concept (Seebens et al., 2004),

(Scholand et al., 2004) for user

k there are

b

N

predetermined directions, into which beams are

steered. The array may receive midamble signals by

all beams. The channel impulse response associated

with each beam is termed Directional Channel

Impulse Response (DCIR) and is denoted by the

1W

×

vector

() () () ()

(

)

ddd d

T

,,, ,

dd,1d,2 d,

kk kk kk kk

W

hh h=h for

the user

,1kk K= …

and for the direction

dd b

,1kk N= …

. Arranging these user and direction

specific DCIR for all users in a new vector one

obtains the direction specific DCIR vector given by

the

1KW

×

vector

() () () ()

(

)

dd d d

T

1, T 2 , T , T

dd d d

kkk Kk

hh h=h for direction

d

k .

The arrangement of the direction specific DCIR

vectors associated with all directions in a new vector

?SMART ANTENNAS IN UMTS LCR-TDD - Implementation of RSS Beamforming

179

leads to the

b

1KN W × vector

() () ()

(

)

b

T

1T 2T T

dd d d

N

hh h=h , which is termed the

combined DCIR vector.

With the

K

WKW× phase rotation matrix for

antenna element

a

k and direction

d

k

()

()

()

()

ad

ad

ad

ad

1, ,

s

2, ,

,

s

s

,,

s

kk

W

kk

kk

W

Kk k

W

b

b

b

⎛⎞

⎜⎟

⎜⎟

=

⎜⎟

⎜⎟

⎜⎟

⎝⎠

I0 0

0I0

A

00 I

, (1)

which contains phase factors associated with

antenna element

a

k and direction

d

k , we define the

phase rotation matrix for all antenna elements and

all directions by the

ab

K

KW KNW

×

matrix

() () ()

() () ()

() () ()

b

b

aa ab

1,1 1, 2 1,

ss s

2,1 2,2 2,

ss s

s

,1 , 2 ,

ss s

N

N

KK KN

⎛⎞

⎜⎟

⎜⎟

=

⎜⎟

⎜⎟

⎜⎟

⎝⎠

AA A

AA A

A

AA A

. (2)

With the

L

KW× matrix of the user specific

midamble training sequences

G

(Steiner and Jung,

1994), the

a

1KK W × received midamble vector

m

e

can be expressed by

(

)

a

s

mdm

K

=⊗ +eIGAhn, (3)

where the

a

1KK W × vector

m

n models additive

noise associated with the received midamble and

⊗

is the Kronecker product. The system equation in (3)

is the basis for the channel estimation.

For derivation of the system model for data

detection we set out from the data parts. In what

follows the

1N

×

data vector

() () () ()

()

T

12

kkk k

N

dd d=d

with a total of

N

data

symbols represents either the first or the second data

part of the burst for each user

k , 1kK

=

.

According to (Seebens et al., 2005) the spreading of

the data symbols by an Orthogonal Variable

Spreading Factor (OVSF) code and the transmission

over the directional channel can be expressed by the

(

)

b

1NNQW N+−×

directional system matrix (Lu,

2002)

() () ()

dd

,1

kkk

kK==AHC . (4)

The

(

)

b

1

N

NQ W NQ+−× matrix

()

d

k

H describes the

transmission over the directional channel and the

NQ N×

matrix

()

k

C

performs the spreading of the

data symbols by applying the OVSF code.

The arrangement of the individual data vectors

()

k

d

of all users in a new vector yields the

1KN × overall

data vector

() () ()

(

)

T

1T 2T 2T

=dd d d . According

to (Seebens et al., 2005) the system equation for

detection of the data transmitted in one of the data

parts of a specific burst is given by

(

)

sd

1

dd

NQ W

+−

=

⊗+eBI Adn. (5)

The

(

)

a

11KNQW

+

−×

vector

d

e in (5) represents

the received data vector and the

(

)

a

11KNQW+−×

vector

d

n is additive noise associated with the

received data. The

(

)

b

1

K

NNQW KN+−× matrix

d

A

is the overall directional system matrix and is

composed of directional system matrices for

individual users

()

d

k

A , 1kK

=

(Seebens et al.,

2005). The

ab

K

KN

×

combined steering matrix

s

B

is composed of steering matrices for the individual

users

()

s

k

B , 1kK

=

(Seebens et al., 2005). The

bases of the

ab

K

N

×

steering matrices

()

s

k

B for user

k are the

a

1K

×

steering vectors

()

d

,

s

kk

b for the

predetermined beams (Godara, 1997). The system

equation (5) is used to derive the estimation matrix

of the Spatio-Temporal Zero-Forcing Block Linear

Equalizer (ST-ZF-BLE) (Seebens et al., 2005).

3 CHANNEL ESTIMATION AND

DATA DETECTION USING

THE RSS BEAMFORMING

Setting out from the system equation (3) the

maximum likelihood (ML) estimate of the

b

1KN W

×

combined DCIR vector

d

ˆ

h can be expressed as

follows

(

)

()

()

m m

1

HH1 HH1

sss

d m

ˆ

aa a

KK K

−

−−

⎡⎤

=⊗ ⊗ ⊗

⎣⎦

nn

hAIGRIGAAIGRe

. (6)

The

aa

K

KW KKW

×

covariance matrix of the noise

m

n

R

associated with the received midamble, owing

to the spatio-temporal equalizer, consists of two

components and can be written as

ma

a

DOA m th,m0

DOA m

0

K

K

N

N

=⊗+ ⊗

⎡⎤

≈+ ⊗

⎣⎦

n

RR R I R

RIR

. (7)

The first Kronecker product in (7) takes the spatial

component into account, where the

aa

K

K× matrix

DOA

R is the spatial covariance matrix and the

K

WKW

×

matrix

m

R

is the temporal covariance

matrix of the interferers associated with the received

midamble, thus representing the spectral form of the

interfering signals. The second Kronecker product

WINSYS 2006 - INTERNATIONAL CONFERENCE ON WIRELESS INFORMATION NETWORKS AND SYSTEMS

180

denotes the thermal noise with the

K

WKW× matrix

th,m

R

being the normalized temporal covariance

matrix of the colored noise and

0

N

being the

spectral noise density. The approximation in (7) is

owed to the assumption that

m

R

is approximately

identical to

th,m

R

.

Using (7) and writing the received midambles in a

a

K

WK× matrix as

() () ()

(

)

a

12

m

mm m

K

=Eee e

, (8)

we can rewrite (6) as

()

(

)

(

)

aa

1

1

11

11

HH * *T* *

m DOA s s DOA sm0 0

d

ˆ

vec

KK

NN

−

−

−−

−−

⎧⎫

⎛⎞

⎡⎤⎡⎤

≈++

⎨⎬

⎜⎟

⎣⎦⎣⎦

⎝⎠

⎩⎭

hGRGGRERIBBRIB

. (9)

In this equation the operator

{

}

vec ⋅ stacks the

columns of the matrix.

To determine the data detector we set out from the

system equation (5). The ML estimate of the

1KN

×

data vector

ˆ

d is then given by (Seebens et al., 2005)

()

()

()

d d

1

HH 1 HH 1

ds s d ds

11 1

d

ˆ

NQ W NQ W NQ W

−

−−

+− +− +−

⎡⎤

=⊗ ⊗ ⊗

⎣⎦

nn

dAB I RBI AAB I Re

. (10)

In accordance to (7) the

(

)

(

)

aa

11KNQW KNQW+−× +− covariance matrix

of the noise

d

n

R

associated with the received data

can be approximated as

d

a

DOA d

0 K

N

⎡⎤

≈+ ⊗

⎣⎦

n

R

RIR

, (11)

where the

()

(

)

11NQ W NQ W+−× +− matrix

d

R

is

the temporal covariance matrix of the interferers

associated with the received data. A comparison of

(7) with (11) yields, that the covariance matrices

m

n

R

and

d

n

R

have the same properties. Note,

however, that these matrices have different

dimensions, because the midamble and the data parts

have different dimensions.

Writing the received data symbols in a

()

a

1

N

QW K+−× matrix as

() () ()

(

)

a

12

d

dd d

K

=Eee e , (12)

one can rewrite (10) by using(11) as (Seebens et al.,

2005)

{

}

a a

1

1

1

11

HH**

DOA d d d d DOA s

00

ˆ

vec

KK

NN

−

−

−

−−

⎡⎤

⎡⎤

⎡⎤

≈+⊗ +

⎣⎦

⎣⎦

⎢⎥

⎣⎦

dAR I RAA RER I B

. (13)

In (13) the beamforming can be identified as being

carried out by the matrix product

a

1

**

DOA s

0 K

N

−

⎡⎤

+

⎣⎦

R

IB. Since the matrix

s

B

is

composed of the steering vectors

()

d

,

s

kk

b , the optimal

beamforming weights can be written by

a

1K

×

vectors (Seebens et al., 2005)

() ()

dd

a

1

,* ,

DOA

0db

opt s

,1,1

kk kk

K

N

kKk N

−

⎡⎤

=+ = =

⎣⎦

wR Ib

. (14)

4 IMPLEMENTATION CONCEPT

To allow the investigation of the performance of

RSS beamforming in a real environment, an

implementation concept for a demonstrator is

presented, which considers RSS beamforming with

four antenna elements at a Node B of LCR-TDD of

UMTS. For demonstration purposes only simple

speech transmission is considered. The User

Equipment (UE) is assumed to have a single antenna

element, hence modeling a standard LCR-TDD

transceiver. This communication concentrates on the

implementation of the Node B, since this is

considered to be interesting for RSS beamforming in

first place. In order to determine the required

functions to implement in the Node B, block

diagrams for the Node B transmitter and receiver are

established.

The block diagram for the Node B transmitter is

given in figure 4. The speech data coming from the

PCI interface are first processed in a standard way as

described in (3GPP TS 25.222 V5.2.0, 2002),

(3GPP TS 25.223 V5.1.0, 2002). The subframes

generated by the subframe assembler are multiplied

with the beamforming weights

a

k

w ,

a

14k

=

,

which have been generated by the receiver for the

previous burst. These weights are used to generate

the desired radiation pattern in transmit direction. In

order to overcome inhomogeneities in the transmit

paths the signals are also subject to calibration by

applying calibration weights

a

r,k

c prior to RRC

filtering. Then the signals are passed to the RF

(Radio Frequency) frontend and radiated by the

antenna elements.

The block diagram of the Node B receiver is shown

in figure 5. The signals coming from the RF

frontends are calibrated by appropriate calibration

weights

a

r,k

c ,

a

14k

=

after RRC filtering. The

calibration weights are determined from the

previous burst and updated with each burst. The

calibrated signals are used for channel estimation

and data detection. For channel estimation the Joint

Channel and DOA Estimation (JCDE) block is

responsible. This block determines the DCIR and

provides it to the Joint Spatial Detector (JSD). With

the help of the DCIR the JSD detects the transmitted

data symbols and determines also the beamforming

weights. These are transferred to the transmitter. The

detected symbols are used to generate Log

?SMART ANTENNAS IN UMTS LCR-TDD - Implementation of RSS Beamforming

181

Likelihood Ratios (LLR), which are used for the

further signal processing according to (3GPP TS

25.222 V5.2.0, 2002), (3GPP TS 25.223 V5.1.0,

2002).

For the digital baseband signal processing Harrier

boards from MangoDSP are considered for the Node

B as well as for the UE. Each Harrier board provides

15 TI C6416 DSPs (Digital Signal Processor), five

Altera Stratix FPGAs (Field Programmable Gate

Array), four MangoLink connectors, and a PCI

interface. The 15 DSPs are arranged in five clusters

with three DSPs each. Each DSP cluster is

associated with an FPGA. The FPGAs may be used

for digital signal processing, which requires simple

operations at high sample rates. The FPGAs can also

be programmed to connect the DSPs with each

other. Furthermore, the FPGAs are required for

operating the MangoLink connectors. There is also a

hierarchical system of PCI busses. These allow the

DSPs to communicate with each other and with any

device attached to the external PCI interface.

Figure 6 shows the board diagram of the Node B.

The heart of the Node B are two Harrier boards, one

for the transmitter and one for the receiver,

respectively. Each of these Harrier boards is

connected with the two RF boards, which provide

the RF frontend for two antenna elements each.

There is an additional RF board which is used to

operate a single calibration antenna element, which

is required to calibrate the receive paths of the four

RF frontends. The board named “Local Oscillator”

is used to provide a system wide stable clock which

is required for the exact timing.

Figure 6: Board diagram for the Node B.

Figure 6 shows the board diagram of the Node B.

The heart of the Node B are two Harrier boards, one

for the transmitter and one for the receiver,

respectively. Each of these Harrier boards is

connected with the two RF boards, which provide

the RF frontend for two antenna elements each.

There is an additional RF board which is used to

operate a single calibration antenna element, which

is required to calibrate the receive paths of the four

RF frontends. The board named “Local Oscillator”

is used to provide a system wide stable clock which

is required for the exact timing.

The total system is implemented in a 19” rack which

also has a PCI backplane.Besides the two Harrier

boards a Single Board Computer (SBC) and a sound

card are also attached to the PCI backplane. The

SBC is used to control the whole setup and also for

the multirate adaptive speech codec, while the sound

card is used for speech input and output. However,

in order to allow undisturbed real-time

communications the Harrier boards are also

connected by MangoLinks.

Figure 7: Comparison of the fixed-point implementation

with the floating-point implementation.

In order to distribute the identified functions on the

DSPs the complexity of each functional block is

estimated in terms of Mega Instructions Per Second

(MIPS). Implementation guidelines are to implement

as many functions on DSPs as possible in order to

be as flexible as possible. Hence, only rather simple

functions running at high sample rates are

implemented on the FPGAs. Moreover, as few as

possible DSPs shall be used to keep the overall

complexity moderate. Furthermore, the total

estimated load of the DSPs shall be below the

theoretical maximum of the DSPs keeping a

relatively large margin. This analysis reveals that the

Node B transmitter can be realized largely on a

single DSP. Only the weighting with the calibration

and beamforming weights and the RRC filtering is

performed in the FPGA. The total load of the DSP is

then 8.9526 MIPS (Seebens et al., 2005).

The Node B receiver is much more complex than the

WINSYS 2006 - INTERNATIONAL CONFERENCE ON WIRELESS INFORMATION NETWORKS AND SYSTEMS

182

Node B transmitter and requires three DSPs and one

FPGA for implementation (Seebens et al., 2005).

The FPGA is needed for the RRC filtering and the

application of the calibration weights on the

received data. Note, that in the Node B receiver, the

application of the beamforming weights is

performed in the JSD (Seebens et al., 2005). DSP 1

is used for the channel estimation, power estimation,

and calculation of the calibration weights. The total

load of this DSP with the mentioned functions is

estimated to be 61,008 MIPS (Seebens et al., 2005).

Since the JSD is the most complex function of the

receiver, it requires a DSP more or less on its own.

Only minor functions like parallel-to-serial

multiplexing, SIR (Signal-to-Interference Ratio)

estimation and LLR generation are also located on

this DSP. The load of DSP 2 is estimated to be

1583.2884 MIPS (Seebens et al., 2005). The

remaining functions according to (3GPP TS 25.222

V5.2.0, 2002), (3GPP TS 25.223 V5.1.0, 2002) are

implemented in DSP 3 (Seebens et al., 2005). The

total load of this DSP is estimated to be 51.6948

MIPS (Seebens et al., 2005). However, this load can

be further reduced by using the internal Viterbi

Coprocessor of the TI C6416 DSP.

5 SIMULATION RESULTS

In (Seebens et al., 2004), (Scholand et al., 2004)

Matched Filter Bounds (MFB) were simulated to

investigate the performance of the RSS beamformer

in the uplink transmission direction. MFBs provide a

lower limit for the achievable Bit Error Ratio (BER),

that one can expect from an algorithm. The

respective MFBs were obtained by considering a

spatial mobile channel as described in (Scholand et

al., 2004), (

Blanz and Jung, 1998). In (Seebens et al.,

2004), (Scholand et al., 2004) it has been shown,

that the RSS beamformer outperforms the

conventional beamformer by comparing the MFBs.

A comparison between the MFBs in (Seebens et al.,

2004), (Jung, 2004) and link-level simulation results

for a microcellular channel model is given in

(Seebens et al., 2005). For the link-level simulations,

the baseband signal processing according to (3GPP

TS 25.222 V5.2.0, 2002), (3GPP TS 25.223 V5.1.0,

2002) was implemented in the UE transmitter. In the

Node B receiver the functional blocks for the

baseband signal processing were implemented as

given in figure 5. Especially, the channel estimation

and the data detection were performed by using (9)

and (13) with the noise covariance matrices assumed

to be the identity matrices. In (Seebens et al., 2005)

it was shown, that the implementation loss for a

UCA with

a

4K

=

antenna elements at a BER of

3

10

−

is about 1.4 dB. This degradation is due to

imperfect channel estimation and data detection. It

should be noted, that the implementation loss of 1.4

dB is a promising result, since the MFBs represent

the best possible performance one can expect.

For the link-level simulations in (Seebens et al.,

2005) a floating point implementation of the

simulator was considered. According to the selected

hardware all functional blocks on the DSPs and

FPGAs of the demonstrator have to run in a fixed-

point implementation. For further validation, a

fixed-point implementation of the simulator is used

to find limitations and degradations of the fixed

point implementation. The functional blocks realized

in fixed point can then be transferred directly to the

demonstrator hardware.

Figure 7 depicts a comparison of the fixed-point

implementation with the floating-point

implementation of the simulator for

a

1, 4K = antenna

elements. For the simulations the microcellular

channel model in (Scholand et al., 2004), (Blanz and

Jung, 1998) was considered. In accordance to the

simulations in (Seebens et al., 2005) the uplink

transmission was considered, where a single user

was placed randomly in the environment for each

simulation run. For a given signal to noise ratio

(

)

10 b

0

10log EN

a total of 1000 simulation runs were

considered. At each simulation run 100 bursts as

shown in figure 3 were transmitted. With QPSK

modulation and a spreading factor of eight a total of

17.6 Mio. bits were transmitted for each signal-to-

noise value. For the fixed-point implementation a 16

bit quantization was used. This means that the input

and the output of each arithmetic operation are

fixed-point values with a word length of 16 bit.

However, intermediate results may require more bits

and have to be scaled appropriately.

In figure 7 the abbreviations “fl. p.” and “fix. p.”

stand for floating point and fixed point, respectively,

distinguished by solid and dashed lines. The violet

and the blue curves represent the BER for

a

1K

=

and

a

4K

=

antenna elements, respectively. From

figure 7 it can be observed, that the performance of

the fixed-point implementation is slightly worse than

that of the floating-point implementation for

a

1K

=

and

a

4K

=

antenna elements, respectively. For a

UCA with

a

4K

=

antenna elements the performance

degradation of the fixed-point implementation with

respect to the floating-point implementation at a

?SMART ANTENNAS IN UMTS LCR-TDD - Implementation of RSS Beamforming

183

BER of

2

10

−

is only about 0.2 dB.

6 CONCLUSIONS

In this communication an overview of a novel low-

cost beamforming concept based on regularly spatial

sampling of signals is presented. The performance

of the RSS beamforming is investigated by means of

simulations with respect to the LCR-TDD mode of

UMTS. It has been demonstrated, that the RSS

beamformer shows promising performance in terms

of bit error ratios. Moreover, a robust fixed-point

implementation of the RSS beamforming is

presented. For further validation of the performance

of the RSS beamformer, an implementation concept

for a demonstrator is developed in order to show the

feasibility of implementing RSS beamforming on

today’s readily available hardware platforms. It has

been shown, that the total digital baseband signal

processing in the Node B receiver can be

implemented on three standard DSPs.

ACKNOWLEDGEMENTS

The authors wish to thank SAMSUNG Electronics,

Global Standards and Strategy (GSR) for their

generous support. Finally, they gratefully

acknowledge the support of their colleagues.

REFERENCES

Holma, H.; Toskala, A.: WCDMA for UMTS. Second

Edition, Chichester: John Wiley & Sons, 2002.

Lehne, P.H.; Pettersen, M.: An Overview of Smart

Antenna Technology for Mobile Communications

Systems.

IEEE Communication Surveys, vol. 2 (1999),

pp. 2-13.

Correia, L.M.:

Wireless Flexible Personalised

Communications

. Chichester: John Wiley & Sons

Ltd., 2001.

Paulraj, A.; Roy, R.; Kailath, T.: ESPRIT – a subspace

rotation approach to signal parameter estimation.

Proceedings of the IEEE. vol. 74 (1986), pp. 1044 –

1045.

Seebens, A.; Hessamian-Alinejad, A.; Kim, S.H.;

Scholand, T.; Jung, P.: Low-Cost Exploitation of

Spatial Diversity – Regular Spatial Sampling (RSS)

Beamforming.

COoperation européenne dans le

domaine de la recherche Scientifique et Technique

(COST) 273

, 11th Meeting of the Management

Committee, Duisburg, Germany, 2004.

Scholand, T.; Kim, S. H.; Hessamian-Alinejad, A.;

Seebens, A.; Jung, P.: Regular spatial sampling (RSS)

beamforming for the low-cost exploitation of spatial

diversity.

Proceeedings of the Third IASTED

International Conference on Communications,

Internet and Information Technology (CIIT 2004)

, St.

Thomas, US Virgin Islands, 2004, pp. 437 – 442.

3GPP TS 25.221 V5.2.0 (2002-09):

Physical Channels

and Mapping of Transport Channels onto Physical

Channels (TDD)

, Release 5, 2002.

Steiner, B.; Jung, P.: Optimum and suboptimum channel

estimation for the uplink of CDMA mobile radio

systems with joint detection.

European Transactions

on Telecommunications (ETT)

, vol. 5 (1994), pp. 39-

50.

Seebens, A.; Ann, J.; Hessamian-Alinejad, A.; Burnic, A.;

Lee, E.; Scholand, T.; Jung, P.: Performance of RSS

Beamforming for TD-SCDMA Transceibvers.

Proceedings of the Third IASTED International

Conference on Communications and Computer

Networks (CCN 2005)

, Marina del Rey, CA, USA,

2005.

Lu, Y.:

Contributions to the application of adaptive

antennas and CDMA code pooling in the TD-CDMA

downlink

. Ph.D. Dissertation, Universität

Kaiserslautern, (2002), http://kluedo.ub.uni-

kl.de/volltexte/2002/1290/pdf/dissertation_7.pdf.

Godara, L. C.: Application of Antenna Arrays to Mobile

Communications, Part II: Beam-Forming and

Direction-of-Arrival Considerations.

Proceedings of

the IEEE

, vol. 85 (1997), pp. 1195 – 1245.

3GPP TS 25.222 V5.2.0:

Technical Specification;

Multiplexing and channel coding (TDD)

. Release 5,

2002.

3GPP TS 25.223 V5.1.0:

Technical Specification;

Spreading and modulation (TDD)

. Release 5, 2002.

Seebens, A.; Hessamian-Alinejad, A.; Burnic, A.; Jeong,

K.Y.; Scholand, T.; Lee, H.W.; Jung, P.:

Implementation Concept of a LCR-TDD Node B with

RSS Beamforming Capabilities.

Proceedings of the

Third IASTED International Conference on

Communications and Computer Networks (CCN

2005)

, Marina del Rey, CA, USA, 2005.

Blanz, J. J.; Jung, P.: A flexibly configurable spatial

model for mobile radio channels.

IEEE Transactions

on Communications

, vol. 46 (1998), pp. 367 – 371.

WINSYS 2006 - INTERNATIONAL CONFERENCE ON WIRELESS INFORMATION NETWORKS AND SYSTEMS

184