throughput enhancement in

Upload: gauravyadav0074017

Post on 08-Aug-2018

213 views

Category:

Documents


0 download

TRANSCRIPT

  • 8/22/2019 Throughput Enhancement In

    1/14

    1

    THROUGHPUT ENHANCEMENT IN

    WCDMA USING THE GENERALIZEDRAKE RECEIVER

    Jay Kumar Sundararajan

    Vaibhav Maheshwari

    R. David Koilpillai

    Department of Electrical Engineering

    Indian Institute of Technology, MadrasChennai, India 600036

    Abstract

    In a multipath fading channel, when there are many interfering CDMA signals, the

    multipath causes loss of orthogonality between different signals leading to intracell interfer-

    ence. The Generalized RAKE receiver, proposed recently in the literature, is based on the

    concept that this interference is not white but colored. Therefore, unlike the conventional

    RAKE receiver, GRAKE tries to match to the channel as well as whiten the interference.

    The GRAKE gives a typical improvement of 1-3 dB in SNR for a moderate increase in

    complexity.

    In this paper, we quantify the benefits of using the GRAKE in place of the RAKE

    in a WCDMA downlink under realistic conditions as specified in the 3GPP standards. The

    comparison is made both in the presence and absence of channel coding. After performing

    channel coding, the block error rate and hence the system throughput is evaluated for both

    receivers in a 144 kbps test case. It is demonstrated that the GRAKE provides significant

    improvement in BER performance, which translates to higher data throughput, as well as

    increased capacity.

    Currently at MIT, USA; E-mail: [email protected] Currently at ETH, Zurich; E-mail: [email protected] Currently at IIT Madras; E-mail: [email protected]

  • 8/22/2019 Throughput Enhancement In

    2/14

    2

    Index Terms

    Wideband CDMA, DS-CDMA downlink receivers, Generalized RAKE Receiver, Col-

    oration of interference

    I. INTRODUCTION

    The RAKE receiver is equivalent to a matched filter receiver that matches to the

    multipath channel. Therefore it is the optimal solution in a situation where the only

    impairments to the transmitted signal are the multipath fading channel and receiver

    white noise.

    However, in realistic situations, one has to account for time-dispersive propagation

    environments. In general, channels are expected to appear more dispersive in third

    generation cellular systems than in second generation cellular systems, as the signal

    bandwidth in 3G systems is much higher. In particular, with the use of multi-code

    transmission and low spreading factors, the multipath channel causes loss of orthogo-

    nality between the signals of different CDMA signals. So this discussion becomes even

    more important in third generation systems which use CDMA and their evolutions such

    as HSDPA, cdma2000 and 1X-EV.

    In CDMA systems, the signals of all users are added up and sent over the same

    frequency band at the same time. So all of them pass through the same radio channel.

    There will be inter-symbol and multiple access interference caused by delayed replicas

    of the transmitted signal as well as the presence of many users in the system. Besides,

    there are signals from neighboring base stations which also produce interference. The

    interference gets distorted in the frequency domain due to the time dispersion and hence

    cannot be treated as white noise. The problem with RAKE receiver is that it assumes

    that such interference can be included under white noise. Actually, the interference

    is colored because of multipath as well as pulse shaping. Thus, in the presence of

    interfering signals, the RAKE receiver is not the best approach.

  • 8/22/2019 Throughput Enhancement In

    3/14

    3

    The Generalized RAKE (GRAKE) receiver [1] is a new receiver that accounts for

    these discrepancies in the RAKE. In the GRAKE, this idea is generalized in two ways:

    The fingers can be placed in locations other than the channel taps

    The weights associated with the fingers need not be conjugates of the channel taps

    I I . INTERPRETATIONS OF THE GRAKE

    An important property, specific to DS-CDMA systems, is that on the downlink, signals

    within the same cell are transmitted using orthogonal waveforms so that they will not

    interfere with one another. However, if the channel has multipath propagation, this

    orthogonality property no longer holds. Thus intracell interference can be mitigated by

    equalizing the dispersive channel.

    Intercell interference does not have this orthogonality property. So, the approach of

    equalizing the channel will not work here. Since the signal from the interfering base

    station goes through a different multipath channel as compared to the signal from the

    desired base station, the differences in spectral coloration must be exploited. Since the

    multipath channel gives multiple images of the signal, one image of the interference is

    used to cancel another image.

    The GRAKE receiver utilizes these propeties of interference and uses a model for the

    coloration of the interference to give a better downlink performance. It aims to achieve

    a trade-off between two goals:

    To equalize the multipath channel to restore orthogonality of the intracell interfering

    signals

    To use one image of the interference signal to cancel another image, thereby

    suppressing intercell interference

    Another interpretation of the GRAKE is that it is a whitening matched filter. This is

    essentially an application of the matched filter theory for colored noise [7]. The structure

    is similar to that of the conventional RAKE receiver. Assuming that the finger locations

  • 8/22/2019 Throughput Enhancement In

    4/14

    4

    have been chosen, the weights are decided using a maximum likelihood formulation. For

    this, the interference at the output of the fingers is modeled as colored noise. If J is the

    number of fingers chosen, then a J x J noise correlation matrix R is formed where the

    (i, j)th entry is the correlation between the noise terms on the ith and jth fingers. The

    ML formulation results in the decision variable:z = cHR1x where c is the vector of

    channel coefficients, x is the vector of finger outputs and R is the noise correlation matrix

    described above. There is a trade-off between matching to the channel and whitening

    with an aim to maximise the overall signal-to-interference-plus-noise ratio (SINR). It

    can be shown that the R matrix can be factorized into one term corresponding to the

    pre-whitening filter and another corresponding to the matching operation [2].

    III. THE MATHEMATICAL FORMULATION

    The notation followed here is the same as that used in [1]

    A. The system model

    The transmitted signal for user k is

    xk(t) =Ek

    i=

    sk(i)ak,i(t iT) (1)

    where Ek is the average symbol energy, T is the symbol duration, sk(i) is the ith symbol,

    ak,i(t) is the user symbol-period-dependent spreading waveform which is expressed as:

    ak,i(t) =

    1N

    N1

    j=0

    ck,i(j)p(tjTc) (2)

    The data symbol is assumed to have unit energy.

    The transmitted signal passes through an L-tap multipath channel whose baseband

    equivalent impulse response is given by:

    g() =L1l=0

    gl( l) (3)

  • 8/22/2019 Throughput Enhancement In

    5/14

    5

    Assuming there are K users totally, the received signal is expressed as

    r(t) =K1j=0

    L1l=0

    glxk(t l) + n(t) (4)

    The GRAKE receiver structure consists of a bank of J GRAKE fingers, each corre-

    lating to a different delay of the received signal. Let the delay of the jth finger be dj .

    For a given symbol, say symbol 0 of user 0, the correlator output (after despreading)

    corresponding to this delay is:

    y(dj) =

    r(t)a0,0(t dj)dt (5)

    These correlator outputs are combined through a weighted summation to produce a

    decision statistic:

    z=J

    j=1

    wjy(dj) = wHy (6)

    where w = [w1, w2, , wJ]T is the vector of combining coefficients and

    y = [y(d1), y(d2), , y(dJ)]T is the vector of correlator outputs.

    B. Calculating the combining weights

    If the correlator output vector can be expressed as :

    y = hs0(0) + u (7)

    then the combining weights are calculated using the following equation which is derived

    using the ML formulation:

    w = Ru

    1h (8)

    where Ru = E[uuH] is the noise correlation matrix of vector u (which is assumed to

    be a complex-valued Gaussian noise vector with zero-mean).

    The computation of the h vector and the noise correlation matrix is as described

  • 8/22/2019 Throughput Enhancement In

    6/14

    6

    in [1]. The correlation in the noise terms is a result of several factors inter-symbol

    interference, intracell interference and white noise. Each of these contribute to the noise

    correlation matrix.

    Once Ru has been found, it is inverted and multiplied with h vector, as in eqn. (8)

    to get the GRAKE weights. These are then used as the combining coefficients during

    maximal ratio combining.

    IV. SIMULATION SETUP

    A. Transmitter

    1) Physical Channels incorporated: The following physical channels have been in-

    corporated in the simulation as mentioned in [3]:

    1) DPCH: Downlink Physical Channel

    2) P-CPICH: Primary Common Pilot Channel

    3) PICH: Paging Indicator Channel

    4) P-CCPCH: Primary Common Control Physical Channel

    5) SCH: Synchronization Channel

    6) OCNS:Orthogonal Channel Noise Source

    The spreading factors and format details of these channels have been specified in [4].

    The OCNS is used to simulate the effect of the DPCHs of other users within the same

    cell.

    2) Power Allocation for the various channels: The relative powers of the various

    physical channels has been fixed according to the Table C.3 in specification [3]. The

    DPCH power is assumed to be -12 dB with respect to the total power. The total power

    for all the OCNS signals put together is calculated by subtracting the power of all the

    other channels from the total transmitted power. Note that the SCH and P-CCPCH are

    time-multiplexed. Therefore, while subtracting the powers of the other channels, the

    power of SCH and P-CCPCH are not subtracted separately, but only once commonly.

  • 8/22/2019 Throughput Enhancement In

    7/14

    7

    Fig. 1. Operations performed in throughput simulation

    3) Channel Coding: The throughput simulation is performed for the test case given

    in Appendix A.3.3 in [3]. This is a 144 kbps. The slot format structure and all other

    parameters are given in [4]. Each block of data bits is transmitted over 2 radio frames.

    The raw data bits are subjected to several operations before being placed in the frame.

    These operations have been shown in Figure 1. The full details of this sequence of steps

    are available in the 3GPP specification [5].

    4) Convolutional Coding: The block of 2904 bits, that results after 8 zeros have

    been appended, is sent through a rate 1/3 convolutional encoder with a memory of 8

  • 8/22/2019 Throughput Enhancement In

    8/14

    8

    bits. The initial state of the encoder is the all-zero state. The resulting bit sequence has

    a length of 8712.

    5) Modulation and Spreading: The bits of all the physical channels are modulated

    using Quadrature Phase Shift Keying (QPSK).

    The complex symbol is then spread using the channelization code that has been

    assigned to the particular channel resulting in complex-valued chips whose number

    depends on the spreading factor. The chips of all the channels are added and then

    filtered through a pulse shaping SRRC filter with roll-off factor chosen to be 0.22 as

    per the WCDMA specifications [6].

    The chip sequence, after spreading, is multiplied with a scrambling code. The scram-

    bling sequence varies from symbol to symbol.

    B. Channel Profile used

    The following is the channel profile that we have used in the simulations (Table 1).

    It is based on an example in [1]. The multipath fading channel has been normalized in

    such a manner that the sum of squares of the magnitude of the various taps is unity

    on the average. This is achieved by dividing each tap by the square root of the sum

    of squares of the nominal channel tap magnitudes. Note that the chip period in the

    WCDMA system is 260.4 nanoseconds.

    Delay Gain

    (in nanoseconds) (in dB)

    0 0260.4 -1.5

    520.8 -3.0

    780.3 -4.5

    TABLE I

    CHI P-SPACED CHANNEL

  • 8/22/2019 Throughput Enhancement In

    9/14

    9

    V. RECEIVER

    A. Channel Estimation

    The received signal is correlated with a known version of the pilot channel (P-CPICH)

    to get the estimates of the channel. A chunk of the pilot channel of length equal to

    5120 chips is used to do the correlation. The chunk is chosen such that its center

    point corresponds to the time instant at which the channel impulse response has to be

    found. Exact knowledge of the channel tap delays is assumed. Now the pilot chunk

    is positioned at these delays and the correlation with the received signal is computed

    at these delays. Thus, the channel coefficients are found. The estimates of the channel

    taps are divided by the norm of the pilot chunk used for the correlation in order to

    normalize them.

    It is assumed that the channel remains constant for half a slot (1280 chips). At the

    Doppler frequency that has been used in our simulation, this is a reasonable assumption.

    After half a slot, the channel estimates are updated. For this, a new pilot chunk is selected

    from the known pilot signal, with its center point at the new time instant corresponding

    to the next half-slot. The correlation is then performed as described in the previous

    paragraph.

    B. Finger Placement

    The location of fingers need not coincide with the actual channel taps in the case of

    GRAKE receiver. However, there is no closed form expression to find out exactly where

    to place them. It has been observed that the performance of the GRAKE is very sensitive

    to the finger locations. The methods suggested in [1] involve a brute force search among

    all possible potential combinations of finger delays in order to find the one which gives

    maximum SNR. However, this method has a very high complexity. The method used

    in our simulation is described here. Symmetrical finger placement algorithm [2]: First

    of all, GRAKE fingers are placed at the actual channel tap locations. The algorithm

  • 8/22/2019 Throughput Enhancement In

    10/14

    10

    tries to cancel the interference on a channel tap by placing extra GRAKE fingers in

    symmetrical places around that path. This is done on the channel taps in decreasing

    order of their energy. It is ensured that none of the new fingers is too close to already

    existing fingers, so that the diversity is utilized best. The total number of GRAKE

    fingers is limited to twice the number of RAKE fingers as it is found that using more

    fingers doesnt enhance the performance further significantly [1]. This is a sub-optimal

    algorithm but is found to work well.

    For each finger, the received signal is tapped at the correct position corresponding

    to the finger delay. This segment of the signal is passed through a receive filter which

    is matched to the SRRC filter used in the transmitter. The output of the filter is then

    downsampled to one sample per chip. We assume that the ideal sampling point is known.

    The estimated chip sequence is then subject to descrambling and despreading and then

    maximal ratio combining is performed.

    Subsequently, Viterbi decoding is performed for decoding the convolutional coding.

    This is done blockwise. The block is tested using the CRC. If the block fails the CRC

    test, a block error is recorded. In this manner the block error rate (BLER) can be found.

    The throughput is then computed for the given BLER value using the following formula.

    Throughput = (1 BLER) 100% (9)

    VI. RESULTS

    Channel profile : Chip-spaced channel

    Test case : 144 kbps test case

    Finger placement strategy : Symmetrical finger placement

    For the throughput simulations, the same simulation model is used except that channel

    coding blocks are included in the model. Rate 1/3 convolutional coding is performed

    followed by puncturing as described in Section IV-A.3.

  • 8/22/2019 Throughput Enhancement In

    11/14

    11

    Fig. 2. RAKE vs GRAKE: Raw Bit Error Rate

    The first plot (Figure 2) shows the raw BER between the point after the channel

    coding and before the decoding. The second plot (Figure 3) compares the information-

    bit error rate. In this plot, the SNR values correspond to the SNR per information bit

    with rate R = 2904/8464 coding. This is related to the actual SNR per channel bit in

    the following manner.

    SNR per information bit = SNR per channel bit k

    n (10)

    where k/n is the effective rate of the code after puncturing. In this simulation, this

    corresponds to a 4.646 dB difference between the SNR per channel bit and the SNR

    per information bit. The third plot (Figure 4), gives the comparison of the throughput

    obtained by using the RAKE and the GRAKE. This figure shows that the RAKE reaches

    an error floor due to which, the throughput cannot go higher than about 88% even for

  • 8/22/2019 Throughput Enhancement In

    12/14

    12

    Fig. 3. RAKE vs GRAKE: Information Bit Error Rate

    high SNR. However, the GRAKE enhances the throughput at low SNR values and is

    able to reach close to full throughput at high SNR values.

    VII. CONCLUSIONS

    We have demonstrated that the generalized RAKE receiver gives a significant perfor-

    mance gain over the RAKE receiver in terms of raw bit-error rate, information bit-error

    rate as well as throughput of the system. For an information bit-error rate of 1%, the

    RAKE requires 8 dB SNR per bit, whereas the GRAKE gives 1% error rate at 6 dB

    itself. Thus there is a gain of 2 dB. This gain translates into a lower block error rate

    for the GRAKE case. In turn, there is an enhancement of throughput.

    For a throughput of 80 %, the GRAKE gives a gain of about 5.5 dB. Moreover, the

    throughput of RAKE saturates at about 88% whereas the GRAKE gives a throughput

  • 8/22/2019 Throughput Enhancement In

    13/14

    13

    Fig. 4. Throughput of the system

    of almost 100 % for high SNR. This is because, even when the white noise is almost

    absent, the RAKE is till affected by interference and therefore, the RAKE performance

    reaches an error floor. This effect can be seen from the raw-bit-error rate plot of the

    144 kbps test case plots.

    At the same time, it is to be noted that the performance gains are sensitive to the

    accuracy of channel estimation as well as the finger placement strategy used. Future

    work can include a detailed study of these effects. Further investigation is necessary on

    what kind of finger placement strategies work best and whether there is a simple way

    to find the optimal location of the fingers. These aspects are critical in order to obtain

    the promised enhancement in performance of the GRAKE over the coventional RAKE

    receiver.

  • 8/22/2019 Throughput Enhancement In

    14/14

    VIII. ACKNOWLEDGEMENTS

    We would like to thank Dr. Gregory Bottomley of Ericsson Research, USA, for useful

    discussions related to this work.

    REFERENCES

    [1] Bottomley, G.E.; Ottosson, T., Wang, Y.-P.E. A generalized RAKE receiver for interference suppression, IEEE

    J. Selected Areas Comm. 18, 8 (August 2000), 1536-1545.

    [2] Kutz G and Amir Chass, On the Performance of a Practical Downlink CDMA Generalized RAKE Receiver,

    IEEE 56th Vehicular Technology Conference, VTC Fall 2002.[3] UE radio transmission and reception (FDD). 3GPP TS 25.101, V5.5.0 (2002-12).

    [4] Physical channels and mapping of transport channels onto physical channels (FDD). 3GPP TS 25.211, V 3.3.0

    (2002-06).

    [5] Multiplexing and channel coding (FDD). 3GPP TS 25.212, V3.3.0 (2002-06).

    [6] Spreading and modulation (FDD). 3GPP TS 25.213, V3.3.0 (2002-06).

    [7] S. M. Kay, Fundamentals of statistical signal processing. Volume 2: detection theory, Prentice Hall, 1998.

    [8] Jay Kumar Sundararajan, Throughput Enhancement in WCDMA using The Generalized Rake Receiver, B.Tech.

    project thesis, IIT Madras, 2003.

    14