Tsoulos George (ред.) MIMO System Technology for Wireless Communications

Подождите немного. Документ загружается.

234 MIMO System Technology for Wireless Communications

vector is a difﬁcult non-convex optimization problem. All these multiuser

MIMO schemes assume perfect CSI at the transmitter. Under partial CSI as

in UTRA FDD, the performance of multiuser MIMO is still an open problem.

8.5.2 Proposed MIMO Algorithms within 3GPP Standardization

The standardization of MIMO for HSDPA is still ongoing, and many algo-

rithms have been proposed by different standardization bodies within 3GPP.

In the following, we brieﬂy summarize the algorithms proposed so far; their

acronyms are presented in Table 8.4. All concepts adjust the modulation and

coding of independent substreams using CQI feedback, and D-BLAST-type

processing is not applied. In addition, some schemes employ space–time

coding and four transmit and two receive antennas to gain diversity and

improve link-level performance (DSTTD-SGRC and RC-MPD). Alternatively,

link performance is improved by linear precoding in the transmitter (SS CL

MIMO, PU

RC, S-PARC, D-TxAA).

The ﬁrst two proposals, PARC and DSTTD-SGRC, are well documented,

but documentation on the other six algorithms is sparse. Therefore, we will

introduce them only brieﬂy. A more detailed description and performance

analysis can be found in [10] and numerous standardization contributions

showing performance results for the algorithms of Table 8.4. However, there

is no widespread agreement concerning the ranking of the candidate algo-

rithms, and therefore, we refrain from comparing them here.

All proposals invariably increase the overhead of CSI feedback when com-

pared to HSDPA in Release 5. Multiple data streams require multiple CQIs,

and additional transmit antennas require additional CSI when compared to

2 × 1 MISO conﬁguration supported in UTRA FDD Release 5. However,

reducing the overhead due to feedback signaling has been identiﬁed as one

of the goals in enhanced HSDPA [13]. Uplink interference due to feedback

signaling may cause problems, especially when the mobile is located on the

cell edge.

TABLE 8.4

Proposed MIMO Algorithms within

3GPP Standardization

MIMO Scheme

Space–Time

Coding

Linear

Precoding

PARC

DSTTD-SGRC ⻬

SS CL MIMO ⻬

RC-MPD ⻬

PU2 RC ⻬

TPRC for CD-SIG

S-PARC ⻬

D-TxAA ⻬

4190_book.fm Page 234 Tuesday, February 21, 2006 9:14 AM

Multiuser MIMO for UTRA FDD 235

Per-Antenna Rate Control (PARC). Figure 8.7 shows a block diagram of

PARC architecture for two transmit antennas. The HS-DSCH data packet is

demultiplexed into two low-rate substreams. Both substreams are turbo

encoded, interleaved, and mapped to either QPSK or 16 QAM symbols. Code

rates and symbol mappings between the substreams may vary, and therefore

the number of information bits assigned to the streams can be different.

Furthermore, symbols in each spatial substream are divided into K subchan-

nels, where K is the maximum number of HS-PDSCHs deﬁned by the UE

capability. After spreading these subchannels by different orthogonal spread-

ing codes (denoted by OSC 1-OSC K in Figure 8.7), the subchannels are

summed up and modulated by a scrambling code and are ﬁnally transmitted

from their designated antennas. Thus, PARC includes multiplexing in spatial

and code domains.

The base station selects data rates and corresponding modulation, coding,

and channelization codes for different antennas based on antenna-speciﬁc

CQI feedback. For this purpose, UE estimates SINR for all antennas, maps

SINR values to CQI values, and sends this information to the base station

through a feedback channel. The number of possible transport combinations

is large, and CQI must be quantized to avoid excessive feedback overhead.

In order to guarantee the performance of PARC at poor channel conditions,

the lower end of the CQI values should correspond to selective transmit

diversity (STD) transmission.

To emphasize the importance of STD feature for the PARC performance,

probabilities for dual stream and STD transmission for 2 × 2 PARC in a

macrocell environment are presented in Figure 8.8. Probabilities are pre-

sented both in ITU Pedestrian and Vehicular A channels. It should be noted

that the probability for the single stream transmission is as high as 75 to

80%. Moreover, Vehicular A has higher dual stream transmission probability

as it provides more multipath diversity than Pedestrian A.

Double STTD with Subgroup Rate Control (DSTTD-SGRC). This scheme

assumes that M

= 2M

. Transmit antennas are divided into M

two-antenna

subgroups that apply STTD and AMC. Both antennas within the subgroup

FIGURE 8.7

Transmitter structure for PARC.

Coding

interleaving

mapping

MCS 1

HS-DSCH

data stream

MCS 2

OSC K

OSC 2

OSC 1

Demux

Coding

interleaving

mapping

4190_book.fm Page 235 Tuesday, February 21, 2006 9:14 AM

236 MIMO System Technology for Wireless Communications

use the same modulation and coding scheme (MCS), but the data rates of

different groups can be adjusted independently or jointly by selecting a

suitable MCS. In WCDMA, the maximum number of transmit antennas is

expected to be four, and thus, at maximum, two independent data streams

can be transmitted.

DSTTD-SGRC is a natural extension to the conventional STTD supported

by UTRA FDD Release’99. While conventional STTD employs two transmit

antennas and a single data stream, DSTTD-SGRC doubles the number of

transmit antenna and data streams, provided that UE is equipped with at

least two antennas. Figure 8.9 shows the structure of the DSTTD-SGRC with

four transmit antennas. The incoming HS-DSCH data are divided into two

substreams. Selected MCS and the number of orthogonal spreading codes

(OSC 1-OSC K) deﬁne the number of information bits allocated to each

substream. For both substreams, information bits are coded, interleaved, and

modulated according to the MCS scheme. After STTD encoding, the two

substreams are split into K parallel streams corresponding to K spreading

codes. In the last stage, streams are combined, scrambled, and transmitted.

Rate-Control Multipaths Diversity (RC-MPD). Here each data stream is trans-

mitted from at least two antennas, and the number of data streams is equal

to the number of transmit antennas. A pair of data streams that share the

same two antennas apply the same MCS. The basic idea is to transmit a

second copy of the signal after one chip delay by using the STTD encoding.

Hence, in the case of two transmit antennas, two data streams and corre-

sponding symbols s

and s

, the transmitted signal consists of symbols s

and

FIGURE 8.8

Stream distribution for 2–2 PARC in ITU Pedestrian A and Vehicular A channels. Round robin

scheduler is used and 70% of the total transmit power in the base station is allocated to

HS-DSCH.

1 & 2 1 2

VehA at 30 km/h

1 & 2 1 2

PedA at 3 km/h

4190_book.fm Page 236 Tuesday, February 21, 2006 9:14 AM

Multiuser MIMO for UTRA FDD 237

at time T and symbols and at time T + T

, where T

is the chip interval.

The aim of the method is to achieve multipath diversity that is orthogonal-

ized through STTD encoding.

Single Stream Closed-Loop MIMO (SS CL MIMO). This is a four-antenna

extension of the two-antenna closed-loop Mode 1 that is supported by UTRA

FDD Release’99. The scheme supports only a single data stream but requires

more CSI than Mode 1 due to additional transmit antennas.

Per-User Unitary Rate Control (PU

RC). This algorithm is based on the

singular value decomposition of the MIMO channels. Precoding in the trans-

mitter is based on the unitary matrix that is a combination of the selected

unitary basis vector from all UEs. The aim is to utilize multiuser diversity

on top of MIMO transmission.

Tx Power Ratio Control for Code Domain Successive Interference Cancellation

(TPRC for CD-SIC). Here the receiver applies code-domain SIC to suppress

the impact of code domain interference in addition to space–time interfer-

ence. System performance is further boosted by employing a code-domain

transmit power ratio control that requires additional feedback signaling.

Power control allocates different powers to different spreading codes accord-

ing to the order of successive interference cancellation in code domain.

Selective Per-Antenna Rate Control (S-PARC). The aim is to improve the per-

formance of conventional PARC by combining PARC with antenna selection.

Double Transmit Antenna Array (D-TxAA). Composite data stream is split

into two substreams and each substream is transmitted from two antennas

by applying either of the closed-loop methods in UTRA FDD Release’99.

Hence, the total number of transmit antennas is four.

8.5.3 Practical Considerations

The high speed downlink packet access in Release 5 contains several enhance-

ments to UTRA FDD downlink compared to earlier releases as discussed in

Section 8.1. MIMO concepts presented above show one possibility to further

FIGURE 8.9

Transmitter structure for DSTTD-SGRC.

Coding

interleaving

mapping

STTD

Demux

OSC 1

OSC K

OSC 1

OSC K

MCS 1

HS-DSCH

data stream

MCS 2

Demux

Coding

interleaving

mapping





4190_book.fm Page 237 Tuesday, February 21, 2006 9:14 AM

238 MIMO System Technology for Wireless Communications

improve the performance of HSDPA. Naturally, the improvement due to

MIMO has to be substantial when compared to the complexity of the imple-

mentation. We recall the most important requirements that should be taken

into account when evaluating different candidate MIMO techniques. In the

following, each paragraph starts with a direct quotation from [10].

MIMO proposals shall be comprehensive to include techniques for one, two, and

four antennas at both the base station and UE. Deploying multiple antennas in

the user equipment or base stations to support MIMO techniques is not

straightforward due to concerns of cost, implementation complexity, and

visual impact. This is especially true in the present mobile terminals, where

basic products with large production volumes may have, at most, two antennas.

Present macro base stations typically employ two or four antennas, and it is

expected that two-antenna base stations will dominate in number in the near

future.

For each proposal, the transmission techniques for the range of data rates from

low to high SIR shall be evaluated. This requirement is due to the fact that the

gain from information MIMO greatly depends on the SINR. For example, in

macrocell environments the operating SINR in HSDPA is less than 10 dB

most of the time. Therefore, practical performance differences between var-

ious diversity and information MIMO techniques are not great because the

differences become only signiﬁcant at high SNR region.

Operation of the MIMO technique should be speciﬁed under a range of realistic

conditions. The requirement sparked the development of a new MIMO chan-

nel model for standardization purposes [47]. The model is a geometry-based

stochastic model, which replaced the earlier stochastic channel model [48,49].

Link-level simulations with the spatial channel model are only used for

calibration purposes, and system-level simulations are required to verify the

performance of MIMO proposals. Moreover, implementation un-idealities

should be taken into account when modeling realistic operation environments.

The MIMO technique shall have no signiﬁcant negative impact on features avail-

able in earlier releases. This requirement makes the design of competitive four-

antenna MIMO algorithms difﬁcult, because the present speciﬁcation con-

tains only two primary common pilot channels. To support four-antenna

MIMO, a straightforward solution would be to deﬁne two additional primary

common pilot channels. However, since the total transmit power in the base

station cannot increase — otherwise the base station would generate more

interference to the network — the transmit power per antenna should be

halved. Legacy UEs can receive only two common pilot channels so that, in

case of a four-antenna cell, they would be able to gather only half of the

pilot power when compared to a two-antenna cell. This would seriously

limit the coverage and service availability within the four-antenna cell.

MIMO techniques should demonstrate signiﬁcant incremental gain over the best

performing systems supported in the current release with reasonable complexity.

Theoretical capacity curves suggest that information MIMO may give

remarkable gains when compared to diversity MIMO. It has been observed,

4190_book.fm Page 238 Tuesday, February 21, 2006 9:14 AM

Multiuser MIMO for UTRA FDD 239

however, that with two transmit and receive antennas, practical gains from

information MIMO can be small [50]. Increasing the number of antennas

increases the gain of information MIMO, but at the same time, the imple-

mentation complexity grows rapidly, and backward compatibility, such as

the pilot design problem, becomes an issue.

The focus shall be on strengthening the UTRA system as a reliable and cost-

effective access technique in urban and suburban areas. This means that the goal

is to increase the number of users, and/or to increase their coverage com-

pared to earlier releases. In other words, the improvement of the service

availability as compared to earlier releases shall be used as a primary evaluation

criterion. The increase in maximum data rate per cell is also of interest. In

WCDMA downlink, capacity and coverage are highly interdependent. Mul-

tiple intracell users, which share the same transmit power resource, interfere

with each other. In general, the system can support more users at low power

than at high power, and the availability of services in a cell depends on

required transmit powers. This requirement suggests that instead of increas-

ing data rates with parallel multiplexing, multiple transceiver antennas

should be used for diversity or beamforming to improve the quality of the

received signal.

8.5.4 Multiuser Beamforming

In this section we present a concept that fulﬁlls the requirements cited in the

previous section without additional feedback overhead or receiver complex-

ity when compared to WCDMA Release 5. This can be accomplished by

utilizing the existing closed-loop modes and CQI feedback in UTRA FDD

together with physical layer scheduling.

UTRA FDD two-antenna closed-loop modes [51] aim to maximize the

received SNR by signaling transmit weights from UE to BS. Due to the

limited overhead of the feedback channel, the transmit weights must be

quantized. Mathematical formulation of the problem for ﬁnding the optimal

transmit weight w

is given by

where and in the presence of ﬂat

fading consists of complex impulse responses corresponding to

the ﬁrst and the second channel. It can be assumed without loss of generality

that w

is real. Therefore, the solution can be characterized by a single

complex coefﬁcient and

Find wWhw hw

 :=



max ,

Ww w= = :|| ||= {(,) , },ww ww C

12 12

1and

h =,()hh

wze

(,) argmax [ ] [ )zzhzehz

KKU

=  +: ,, ,10102

¯¯

4190_book.fm Page 239 Tuesday, February 21, 2006 9:14 AM

240 MIMO System Technology for Wireless Communications

With UTRA FDD closed-loop Mode 1 and Mode 2, the feedback weights

are quantized to follow a time-varying QPSK and 16-QAM signal constella-

tion, respectively.

Assume now that the base station transmits to two users simultaneously

employing either of the present closed-loop modes, and let the channeliza-

tion and scrambling codes be the same for both users. Such assumptions are

reasonable in HSDPA, where the shortage of channelization codes puts strict

limits on the cell throughput. After despreading, the received signals of the

two users become

where r

is the received signal of user i and is the impulse

response vector between the transmitter and the user i, w

is the transmit

weight vector requested by user i, s

is the transmitted symbol, and n

is the

noise term.

Mutual interference between simultaneously scheduled users that use the

same channelization code can be suppressed by selecting users i and j in

such a way that their transmit weights are orthogonal, i.e., . Now

the expectation L

i,i

for the desired power of user i, (i {1,2}) is given by

(8.23)

and the expected undesired interference powers between the users are

Moreover, assuming that the channels corresponding to separate users are

uncorrelated, we ﬁnd that

(8.24)

When the base station transmits with a ﬁxed power, as in HSDPA, the

expectation of Equation 8.25 vanishes due to orthogonal transmit weights.

Computation of SIR can be carried out provided that the expectations of

Equation 8.23 and Equation 8.24 can be evaluated. This can be done by

rssn

11111221

22112222

=++

hw hw

iii

hh=,

()

†

= 0

ii i i i,

== :

{}

EEhw hw w W

max

12 1 2

== :

{}

EEhw hw w Wmin ,

=== :

{}

EEhw hw w W

min

EE()

†

hw hw

11 12 1121 11



= ww h

²²



+.ww h

12 22 21

4190_book.fm Page 240 Tuesday, February 21, 2006 9:14 AM

Multiuser MIMO for UTRA FDD 241

following the method of [52], and in the case of Rayleigh fading, the resulting

SIR values corresponding to Mode 1 and Mode 2 are given by

Performance of the proposed multiuser beamforming (MUB) technique is

simulated with a quasi-static system level simulator assuming UTRA FDD

closed-loop Mode 1. Average throughput per sector is used as a performance

measure and 2 × 2 PARC provides a reference case. The simulation param-

eters are listed in Table 8.5. The primary user is chosen by using a round

robin (RR) scheduler, which does not take into account CSI. The secondary

user with the same spreading code and orthogonal transmit weight vector

is chosen by either RR or max SINR scheduler. If there is no user having

orthogonal transmit weights with the primary user, full HS-DSCH power is

allocated to the primary user. Otherwise, HS-DSCH power is evenly divided

between the primary and the secondary user. Thus, the proposed scheduling

is a simple extension of the conventional RR providing a fair share of trans-

mission resources to primary users in a RR fashion while additional

resources are given to secondary users depending on the orthogonality of

transmit weights.

The average sector throughput values in Pedestrian A channel are pre-

sented in Figure 8.10. When the secondary user is scheduled by using RR

scheduler (legend “MUB, RR”), the gain in throughput over single-user

transmission is approximately 40%, whereas for max SINR scheduling (leg-

end “MUB, max SINR”) the gain is 70%. Multiuser beamforming with RR

scheduler provides almost the same average sector throughput as 2 × 2

PARC, whereas with max SINR scheduler MUB provides 15% better average

sector throughput than PARC. We emphasize that the MUB schemes employ

only one receive antenna in UE. It has been shown that the performance of

the MUB can be further improved by slightly modifying the calculation

of CQI in user equipment [53]. However, here the simulations assume that

TABLE 8.5

Simulation Assumptions

Site-to-site distance 2.8 km

Pathloss model 3GPP UMTS

Number of sectors Three/cell

Power delay proﬁle ITU Pedestrian A, ITU Vehicular A

Number of HS-DSCH codes 10

MCS set {QPSK (1/2), QPSK (3/4), 16QAM (1/2),16QAM (3/4)}

Receiver Time-domain LMMSE equalizer

Power of HS-DSCH 70% of the total BS power

Power of overhead channels 30% of the total BS power

Feedback delay 1 slot

Feedback error rate 4%

SIR

dB for Mode

===

765 1

13 5 BB for Mode 2

4190_book.fm Page 241 Tuesday, February 21, 2006 9:14 AM

242 MIMO System Technology for Wireless Communications

UEs are perfectly backward compatible to earlier UTRA FDD releases and

CQI calculation is not altered.

Cumulative distribution functions (CDF) for user throughputs are pre-

sented in Figure 8.11. For MUB with RR scheduler, the shape of the distri-

bution is maintained but the distribution is shifted to the right when

compared to the single-user distribution. The shape of the distribution of

MUB with max SINR scheduling is less steep than the CDF of MUB with

FIGURE 8.10

Average sector throughputs in ITU Pedestrian A channel with 3 km/h mobile speed.

FIGURE 8.11

User throughput CDFs, ITU Pedestrian A channel with 3 km/h mobile speed.

1000

2000

3000

4000

5000

6000

Scheme

roughput

(

kbps/sector)

Avg sector t

roug

put in ITU Pe

A at 3

CL Mode 1

MUB, RR

MUB, max SINR

2 × 2 PARC

0 100 200 300 400 500 600 700 800 900 1000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

roughput(kbps)

Cumu

ative pro

ity

CDF for throughput in ITU pedA at 3 km/h

CL Mode 1

MUB, RR

MUB, max SINR

2 × 2 PARC

4190_book.fm Page 242 Tuesday, February 21, 2006 9:14 AM

Multiuser MIMO for UTRA FDD 243

RR scheduling. Multiuser beamforming with max SINR scheduling increases

the probability of serving good users when compared to MUB with RR,

which explains the higher rates. On the other hand, users with a low SINR

are only scheduled as primary users in MUB with max SNR scheduling,

whereas MUB with RR schedules weak users as primary and secondary

users alike. This explains the shape of the CDF at low data rates. When

comparing the average user throughput CDFs of MUB against the PARC,

we note that MUB with RR provides best average user bit rates for users in

poor channel conditions, whereas PARC provides better average bit rates for

users in good channel conditions than MUB with RR scheduler. The max

SINR scheduler provides clearly the best average bit rates, but on the other

hand, it also provides the smallest average bit rates for the users in poor

channel conditions.

Average sector throughputs in ITU Vehicular A channel at 30 km/h are

presented in Figure 8.12, and the corresponding CDFs for average user

throughputs are shown in Figure 8.13. Comparing the results in Pedestrian

A and in Vehicular A channels reveals the following differences: First, the

average sector throughputs in Vehicular A channel are smaller than in Pedes-

trian A channel for all simulated schemes. For the closed-loop Mode 1 and

the MUB schemes the reduced throughputs are explained by reduced beam-

forming gains. For PARC the performance reduction is due to the interference

between the streams, which increases in multipath channels. Second, in Vehic-

ular A channel MUB with RR scheduler outperforms PARC, while the oppo-

site is true in Pedestrian A channel. This is due to the increase in interference

between users and substreams, the increment being larger for PARC than for

MUB. This is based on the fact that MUB minimizes the interference between

FIGURE 8.12

Average sector throughputs in ITU Vehicular A channel with 30 km/h mobile speed.

1000

2000

3000

4000

5000

6000

Scheme

roughput (kbps/sector)

Avg sector t

roug

put in ITU Ve

A at 30

CL Mode 1

MUB, RR

MUB, max SINR

2 × 2 PARC

4190_book.fm Page 243 Tuesday, February 21, 2006 9:14 AM