An SPW Implementation of Wideband DS-CDMA Forward Link Physical Layer Murali Krishnan, John Lundell, Siavash M. Alamouti [email protected]; [email protected]; [email protected]

Abstract This paper describes an implementation of the forward link physical layer of the wideband DS-CDMA as described in the preliminary ETSI W-CDMA specification for UMTS [1]. The channel definitions, the frame structure, and the transmission formats for the forward link channels are presented, and the implementation in SPW is described in detail. The description of forward error correction coding and decoding is excluded. However, the implementation of the pilot-aided Rake receiver is explained in great detail. 1. Introduction Cadence Design Systems, Inc. first implemented the WCDMA system in Signal Processing WorkSystem (SPW1) as specified by the ETSI SMG2 Wideband DSCDMA Concept Group (Alpha). The specifications for the European UMTS and the Japanese ARIB W-CDMA systems are currently being harmonized by the Third Generation Partnership Program (3GPP). Although these specifications have not been finalized, we have built the appropriate W-CDMA libraries and have been updating them and building additional libraries as the standards evolve. The first demonstration system in this series is the mobile station fast cell search algorithm[2]. Other models include a forward link (downlink) model with a six finger Rake receiver, and a reverse link (uplink) model with interference cancellation. In this paper, we present the forward link of the UTRA FDD Mode, the second in a series of W-CDMA demonstration systems developed by Cadence. We do not discuss forward error correction (FEC) coding and decoding on the forward traffic channels. The paper is primarily focused on the symbol transmission structure and the implementation of the Rake receiver. However, 1. Signal Processing Work System and SPW are registered trademarks of Cadence Design Systems, Inc.

models of the forward and reverse links including FEC coding are now available. The forward link channel definitions, the frame structure and the transmission formats are described in sufficient detail in Section 2 to Section 4. In Section 5, the various short spreading codes and long scrambling codes used in the downlink traffic channels are explained. The SPW implementation of the forward link UTRA FDD Mode is presented in Section . 2. The Forward Link Channel Parameters The fundamental chip rate is 4.096 Mchips/s. The nominal minimum channel separation to accommodate the 4.096 Mchips/s is 5 MHz. The specified pulse shaping roll off factor is 0.22. This results in an effective minimum bandwidth of 4.096x(1+.22) which is approximately 5 MHz. The W-CDMA carriers are specified on a grid with a resolution of 200 KHz. The typical separation between the carriers is 5 MHz, but it may vary depending on the deployment scenario. The transmissions from all the logical channels within a 5 MHz channel are code division multiplexed and transmitted over the same band. In this paper, we discuss 2 downlink channels: • Dedicated Physical Data Channels (DPDCH) • Dedicated Physical Control Channels (DPCCH) Table 1 shows the bit rates, symbol rates and the spreading factors for the DPDCH/DPCCH channels. The raw data rates of these channels on the uplink is the same as the downlink. The channels are spread by a short spreading code with a spreading factor that may vary from 4 to 256 depending on the data rate of the channel. The channels are also scrambled by a long scrambling code specific to the base station. The symbol rate is simply the chip rate divided by the spreading factor. For instance, if the spreading factor is 16, then the symbol rate is 4.096 Mchips/s divided by 16, or 256 ks/s. The bit rate is then 256x2=512 kb/s.

Version 0.4, February 15, 1999

An SPW Implementation of Wideband DS-CDMA Forward Link Physical Layer

channel

bearer bit rate kb/s

symbol rate ks/s

spreading factor

32

16

256

64

32

128

128

64

64

256

128

32

512

256

16

1024

512

8

2048

1024

4

DPDCH/ DPCCH

Table 1 The bit rates, symbol rates and spreading factors for DPDCH/DPCCH channels.

3. Forward Link Frame Structure The frame structure for the forward link DPCCH/DPDCH channel is shown in Figure 1. A super frame is made up of 72, 10 ms frames. Each frame has 16 slots which are 625 µs (or 2560 chips). As shown in Figure 1, the DPCCH and DPDCH are time multiplexed within the same slot.

Figure 2 shows the forward link DPCCH/DPDCH transmit processing for one channel in a 10 ms frame. The process shown in the figure is repeated every 10 ms1. A DPCCH/DPDCH channel carries 2n bits of information in every 10 ms frame. The actual number of bits depends on the data rate and hence the spreading factor. Please note that some of the bits are to be used as pilot symbols, so the pilot insertion process is included in this step. The 2n bits are mapped onto n QPSK complex symbols. The QPSK symbols (in the ith channel) are first spread using a short spreading code Ci of length L to produce 40960 chips. In other words nxL=40960. The resulting chips are then combined with the chips from all the other channels which require scrambling2 within the same cell or sector and are scrambled by a long code Cscramb of length 40960. The resulting chips are then combined with those that do not require scrambling and are passed through a square root raised cosine filter (SRRC) with roll-off factor α=0.22, are RF converted and finally transmitted through the air. DPDCH & DPCCH

40960

QPSK 2n

mapper

n Cscramb

Ci

super frame= 720 ms Frame 1

Frame 2

Frame i

Frame 72

IF/RF

Slot 2

Slot i

Slot 16

other channels

slot = 625 µsec pilot TPC

DPCCH

TFCI

combiner

combiner 40960

frame = 10 ms Slot 1

SRRC Filter

40960

40960 scrambled channels

Figure 2 The transmit processing for DPCCH/DPDCH channels in a 10 ms frame.

Data

DPDCH

Figure 1 The frame structure definitions for W-CDMA forward link.

The DPCCH channel consists of pilot symbols, transmit power-control (TPC) bits and transport format combination indicator (TFCI) bits. The number of symbols (or bits) in each of these fields has yet not been specified. The DPDCH channels contain bearer data. The number of bits transmitted in a single slot depends on the data rate of the channel. 4. The Forward Link Traffic Channel Transmission Structure (Forward DPDCH/DPCCH)

The short spreading codes are orthogonal sequences whose elements are either 1 or -1, and so are the elements of the non-orthogonal long scrambling codes. The mathematical description of the forward link transmit processing is given in this section. You may skip over this section if not interested. In fact, the mathematical representation uses matrices whereas most implementations may involve only vectors. Nevertheless, you may use these descriptions to gain insight to the

1. Figure 2 describes the base station baseband processing for a single traffic channel. The transmission for multiple channels may be combined differently based on the preferred implementation of the base station transmitter. 2. A given base station may serve multiple sectors within the cell, and each sector may be assigned a different scrambling code. Note, however, that not all channels are scrambled. For instance, the synchronization channels are transmitted without any scrambling.

Version 0.4, February 15, 1999

2 of 10

An SPW Implementation of Wideband DS-CDMA Forward Link Physical Layer

transmit processing. Let the QPSK symbol sequence of length n be denoted by the row vector di. The sequence is first spread by the short spreading code Cch,i of length L which may be denoted by the row vector Cch,i. The long scrambling code may be described by an n by L matrix Cscramb. The spread and scrambled sequence may then be represented by: T

Si = d i C ch, i • C scramb

(1)

where T denotes matrix transpose operation and ● denotes element-wise matrix multiply. The actual transmission is done serially starting with the first element in the first row, followed by the second element in the first row, and ending with the last element in the last row. When the transmissions from all the code channels are combined, the resulting CDMA signal may be described by: T

S = ∑ d C ch, j • C scramb

(2)

j

j

The short spreading code vectors have the following useful properties: 1 T --- CiCi = 1 L T

CiC j = 0

∀i (3) ∀i ≠ j

where L is the length of the spreading vector (spreading factor). The long scrambling code has the following property: C scramb • C scramb = U

transmissions of the various traffic and control channels within a given cell, and the non-orthogonal long scrambling codes are used to reduce inter-cell interference amongst the channels in neighbouring cells. At the time this paper was published, some parameters of the codes were not yet specified. The codes described in this paper are those implemented in the SPW example model. The codes are parameterized in the model and may hence be easily modified. The short spreading codes are used to ensure orthogonality between the forward link channels with different spreading factors and rates and are hence called Orthogonal Variable Spreading Factor (OVSF) codes. The OVSF codes are generated using the code tree shown in Figure 3. The construction of these codes is very similar to Hadamard codes. Two branches emanate from each code in the code tree. The top emanating branch is simply the code from the mother branch repeated twice and the bottom branch is that same code followed by its negation. Not all the OVSF codes are mutually orthogonal. A given code can be used in a cell if and only if there are no other codes used on the path from that given code, to the root of the tree, or any code belonging to the sub-tree generated from that specific code. For instance, if C4,1 is used, then C2,1 and C1,1 may not be used in the same cell as they are in the path to the root. Also, C8,1 and C8,2 and all other codes derived from C4,1 cannot be used in the same cell. SF=1

SF=2

1, 1, 1, 1

(4) 1, 1

where U is the unity (all ones) matrix of the same dimension as Cscramb. If all short spreading codes (Cj) are orthogonal, then the original data for channel i may be fully recovered. For instance, it may be shown using properties in Equation 3, and Equation 4 that:

SF=4

c2,1

c4,1 1, 1, -1, -1

c4,2

1

c1,1

1, -1, 1, -1

c4,3 1, -1

T

d i = ( S • C scramb C ch, i )

T

c2,2

(5)

SF=8 c8,1 c8,2 c8,3 c8,4 c8,5 c8,6

1, -1, -1, 1

c8,7

c4,4

5. The Spreading and Scrambling Codes

c8,8 Figure 3 The code tree for the OVSF codes.

There are two major types of codes specified for the WCDMA forward link channels: orthogonal short spreading codes and non-orthogonal long scrambling codes. The orthogonal short spreading codes are used to separate the

Depending on the required data rate, the length of the short spreading code (spreading factor) applied on a given DPCCH/DPDCH forward link channel may vary from 4 to 256 corresponding, respectively, to data rates of 2 Mb/s

3 of 10

Version 0.4, February 15, 1999

An SPW Implementation of Wideband DS-CDMA Forward Link Physical Layer

and 32 kb/s. The spreading factor must a power of two so the possible spread factors are; 4,8,16,32,64,128,256. In order to reduce the interference from the channels in adjacent cells, every base station uses a long scrambling code. The long scrambling codes are from the well-known family of Gold codes[3] built by the product of two maximum linear PN sequences. These codes have good cross-correlation properties and are widely used in commercial spread spectrum systems. In general, using two m-stage maximal sequence shift registers, 2m+1 Gold sequences of period 2m-1 may be constructed. The Gold codes specified for scrambling of DPDCH/ DPCCH denoted by Cscramb are generated using the product of two 18-stage (m=18) shift registers and have the following generator polynomial: g( x ) = g1( x )g2( x )

(6)

where 18

7

18

10

g1 ( x ) = x + x + 1 7

(7)

5

g2 ( x ) = x + x + x + x + 1

Figure 4 shows the feed-back shift register description of the long scrambling code.

Figure 4 The shift register implementation of the Gold code used as the long scrambling code for DPDCH/DPCCH.

It is possible to generate 2m+1=262,145 sequences of period 2m-1=262,143 using this implementation. However, only 512 sequences of length 40960 chips are specified for use in the W-CDMA forward link channels. Furthermore, the 512 sequences are divided into 16 sets of 32 sequences. Each base station is normally assigned one sequence in one of the sets. However, there may be circumstance where the use of multiple scrambling codes within one cell may be beneficial. The initial value of the top registers for all the codes is the all ones sequence. The

initial value of the bottom registers determines which of the 512 sequences is generated. The initial value for the code in the jth sequence of the ith code group is simply the binary representation of 16i+j (i varies from 0 to 15 and j from 0 to 31). For instance, the 8th code in the 5th code group would have the initial value of 88 which is 000000000001011000. 6. The SPW Model for the W-CDMA Forward Link The W-CDMA forward link system has been implemented in SPW 4.1. The model includes all the components discussed in this paper. In this section, we will discuss the various blocks that make up the model. The system model for the W-CDMA forward link is forward_link.system. The parameters in the model are as follows: • Transmit Power for Data Symbols This parameter is set to 0.5 Watts. This is the power of each transmitted QPSK symbol. • Transmit Power for Pilot Symbols This is the power of the pilot symbols set at a default value of 1.0 Watts which is 3 dB higher than the power of the data symbols. • Spread Factor This is the spreading factor which may vary between 4 and 256. The default value in the SPW model is 16, corresponding to a symbol rate of 256 ks/s. • Number of Symbols per Slot This parameter is calculated1 based on the spreading factor. As explained in Section 3, there are 2560 chips in each slot. The number of symbols within a slot is therefore 2560 divided by the spreading factor. For a spreading factor of 16 this results to 160. The number of symbols per slot may vary between 10 (for spreading factor 256) and 640 for spreading factor of 4. • Number of Pilot Symbols per Slot This parameter is yet not specified. We have given this parameter a default value of 8 for a spreading factor of 16. However, this value may not be reasonable for higher spreading factors. For instance, for a spreading factor of 256, there are only 10 symbols per slot. It would be very inefficient to designate 8 of these symbols as pilots. • Number of Slots for Channel Estimation

1. In your version of the model, the parameter may have been fixed. Please ensure that you change the parameter when the spreading factor is changed. If you have any problems modifying the parameter please email any of the authors for assistance.

Version 0.4, February 15, 1999

4 of 10

An SPW Implementation of Wideband DS-CDMA Forward Link Physical Layer

These are the number of slots used to estimate the phase and the amplitude of the channel. The appropriate number of slots used for channel estimation depends on the doppler rate and the signal to interference and noise ratio. Increasing the number of slots, enhances the estimate. However, the resulting delay may degrade the performance. • Channel Estimation Weights File The channel estimation is based on a weighted average of Nc slots where Nc is the parameter Number of Slots for Channel Estimation. The weight file therefore contains Nc weights. • Number of Slots for Pilot Power Estimation These are the number of slots used to estimate the power of the pilot symbols received on each finger. If the received power is below a given threshold then the corresponding finger is not used for combining. This threshold is given by the parameter Fraction of Maximum Power Used for Combining. • Fraction of Maximum Power Used for Combining This parameter is the power threshold for the selection of a given finger for combining. This figure is the percentage of the maximum received power of the pilot symbols on all the fingers. The default value of the this parameter is 10%. Any finger whose power is more than 10% of the power of the pilots on the best finger is selected for combining. • Channel Oversampling Factor This is the number of samples per chip used in the square root raised cosine (SRRC) channel filter. • Raised Cosine Filter File This is SRRC with a roll-off factor of 0.22. This block runs faster than the standard SPW SRRC block as it is not a generalized block. The blocks in the system model forward_link.system are the main blocks for the SPW implementation of the FMA forward link model. The processing of the blocks is done in a serial fashion from the left-most block (QPSK MOD AND SPREADING) to the right (QPSK DEMOD). The right most set of blocks (BIT ERROR RATE AND RECEIVED CONSTELLATION DISPLAY) are only used for the interactive simulations (ISL) display and do not play a rule in the forward link system functionality. 6.1

QPSK MOD and SPREADING

From the system model click on the TRAFFIC CHANNEL block. Inside the TRAFFIC CHANNEL 5 of 10

block, there are the building blocks for the generation of forward link transmission formats for DPCCH/DPDCH. The transmissions on these two channels are time multiplexed as shown in Figure 1. The data on DPDCH is generated using a binary random generator followed by binary to integer conversion and a QPSK mapper. The pilot insertion is done using a VEC JOIN VEC COMPLEX block. The pilot symbols are generated using a CONSTANT VECTOR block. In our model, we do not insert transport format combination indicator (TFCI) and transmit power-control (TPC) bits as we do not employ power control or rate adaptation in the model. Nevertheless, these functions may be incorporated into the existing model. The resulting vector sequence of length Ns (is then spread by a factor of L, where Ns is the number of symbols within one slot set by the parameter Number of Symbols per Slot and L is the spreading factor set by the parameter Spread Factor. Recall that the chip rate is to remain constant. Therefore, the product LxNs must remain constant at 2560. For the default values, Ns=160, and L=16. The parameter Ns in the model has been fixed to 160. To be compliant to the specification, it may be better to introduce a parameter Number of Chips per Slot which may be fixed to 2560. Then the parameter Number of Symbols per Slot Ns may be calculated as Number of Chips per Slot divided by Spread Factor. The spread sequence is then multiplied by the long scrambling code described before. The OVSF CODE block is a custom C code that generates OVSF codes of arbitrary length. The Short Code Class parameter in the block determines the length of the code. If the parameter is set to n, then a code of length 2n is generated. The parameter Short Code Number determines which code within the set is selected. This may be better understood using Figure 3. The Class i and Number n indicates code Ci,j where j= 2n. The DOWNLINK LONG CODE block generates the Gold Code of length 40960 shown in Figure 4 using the product of two PN sequence generators. 6.2

The Transmit Filter

The Transmit Filter is a Square Root Raised Cosine Filter (SRRC) with roll-off factor 0.22 which is constructed using a complex interpolator with an oversampling rate set by the parameter Channel Oversampling Factor and the filter file is defined using the parameter Raised Cosine Filter File.

Version 0.4, February 15, 1999

An SPW Implementation of Wideband DS-CDMA Forward Link Physical Layer

6.3

The Channel Model

The channel model used in the SPW system is that specified by IMT2000[4]. The model takes into account both slow and frequency selective fast fading. The slow fading phenomenon is modeled by log-normal shadow fading. Frequency selective fast fading is modeled by the superposition of multiple single faded paths (flat fades) with different arrival times and different average powers. The arrival time and the average power for each path is assumed to be fixed and is determined by the channel impulse response. Each path has a Rayleigh distribution with the power spectrum as suggested by Jakes. The IMT2000 has specified three basic environments: indoor, pedestrian and vehicular. These models have been built using a generalized SPW block. The parameters in the generalized model may be changed to simulate all specified environments. For the purpose of illustration, we discuss only one of the environments: the Vehicular Model A. The remaining environments may be created by changing the parameters in the model. The Vehicular Model A block, shown in Figure 5, can be found under the following logical grouping name: >Blocks>IMT-2000 Channels>Vehicular Model A.

functions for the Rayleigh channel. Test systems may be found under the logical grouping >IMT-2000 Tests. The IMT-2000 Vehicular channel model consists of two main parts: slow time varying path loss and faster time varying channel impulse response. The main window, and the various parameters in the model are shown in Figure 5 6.3.1 Path Loss The path loss is made up of four components: • free space loss, Lfs, • diffraction loss from rooftop to the street, Lrts, • diffraction loss from multiple screening past rows of buildings, Lmsd, and • lognormal shadow fading, Llsf. Loss = Lfs + Lrts + Lmsd + Llsf The free space loss is given by: c L fs = –20 ⋅ log  ------------------------------ 2 2 ⋅ π ⋅ D ⋅ f 

(8)

where D is the transmitter-receiver separation in meters, f is the carrier frequency in Hz and c is the speed of light in meters per second. Lrts is given by 1  2  c ⋅  --1- – ----------------  ϑ 2 ⋅ π + ϑ  Lrts = –10 ⋅ log -------------------------------------------2 2⋅π ⋅r⋅ f

(9)

where r and ϑ are given by 2

2

r = hm + x ,

hm ϑ = atan -----x

(10)

and hm is the difference between the mean building and mobile heights and x is the distance between the mobile and the diffraction edges. Lmsd is given by b Lmsd = –20 ⋅ log  ----  D Figure 5 The SPW block for IMT-2000 Vehicular Model A channel.

The model includes static path loss, lognormal shadow fading, and Rayleigh fading. Verification consists of matching the autocorrelation function for the Lognormal shadow fading and matching the level crossing rates and cumulative distribution

(11)

where b is the average separation between rows of buildings. The lognormal shadow fading is given by: Llsf = σdev ⋅ X ln

(12)

where Xln is a zero mean Gaussian random variable with autocorrelation function:

Version 0.4, February 15, 1999

6 of 10

An SPW Implementation of Wideband DS-CDMA Forward Link Physical Layer

6.3.3 ln 2 R( ∆x ) = exp  – ∆x ⋅ ----------  d cor

(13) Parameter

where dcor is the decorrelation distance, typically 20m. The ∆x can be replaced by v ⋅ t where v is the mobile velocity yielding: ln 2 R( t ) = exp  – v ⋅ t ⋅ ----------  d cor

(14)

The random variable Xln is generated by passing white Gaussian noise through a shaping filter with frequency response of FFT ( R( t ) )

(15)

The standard deviation ( σdev ) is set as a parameter, usually 8 dB. The model ensures that the combined path loss is never less than the free space loss (Lfs). 6.3.2 Channel Impulse Response The channel impulse response for the Vehicular Model A channel is shown in Table 2 Delay (nsec)

Average Power (dB)

0

0

310

–1.0

710

–9.0

1730

–15.0

2510

–20.0

1090

–10.0

Parameters List

Carrier frequency

Carrier frequency in GHz.

Mobile velocity

Mobile velocity in km/hr.

Sampling frequency

Number of samples per second the model processes.

Propagation distance

Separation between transmitter and receiver in meters.

Mean building row separation

Average distance between rows of buildings in meters.

Base/mobile height difference

Difference between the mean building and mobile heights in meters.

Diffraction distance

Distance between the mobile and the diffraction edges.

Path loss standard deviation

Standard deviation of the lognormal shadow fading path loss in dB.

Decorrelation length

Decorrelation distance of the lognormal shadow fading path loss in meters.

Delay 1 to Delay 6

Delay in ηsecs for the different channel paths.

Power 1 to Power 6

Average relative power of the different channel paths.

Inputs in

Complex baseband equivalent input signal.

fd

Doppler frequency. If connected, overrides the mobile velocity parameter.

Outputs out

Table 2 Channel impulse response for Vehicular Model A

All paths are complex Gaussian random variables with power spectra of 1 P( f ) = ------------------------------f 2 π ⋅ 1 –  ------  f d

(16)

where fd is the doppler frequency. This is the classic Rayleigh fading spectrum.

7 of 10

Description

Complex baseband equivalent output signal.

6.3.4 Additive White Gaussian Noise (AWGN) The last element is the AWGN source which simulates the effect of overall interference in the system including intercell interference and thermal noise. Since W-CDMA systems may have a unity reuse factor, the dominating source of interference is inter-cell interference. The level of inter-cell interference is often denoted by the term carrier-to-interference-ratio (C/I). In CDMA Systems this value may be 0 dB or lower. In other words, the energy of the desired signal may be lower than the aggregate energy of the interferers in the neighbouring cells.

Version 0.4, February 15, 1999

An SPW Implementation of Wideband DS-CDMA Forward Link Physical Layer

6.4

The RECEIVE FILTER

The received signal (the output of the channel) is passed through the receive filter which is a SRRC filter with identical parameters as that in the transmit filter described in Section 6.2. 6.5

The COMPLEX VARIABLE DELAY

The next element is the COMPLEX VARIABLE DELAY which is used to set the reference time of the Rake fingers. In other words, the delay brings all the fingers into time alignment for combining. The delay values are set to align all the fingers to the start of the next slot. There are 2560*8=20480 samples per slot. The appropriate delay for the first path with zero delay is therefore 20480 samples minus the total filter delay which happens to be 128 samples. The first delay value is thus 20480128=20352. The second delay value corresponds to the second path which according to Table 2 is delayed by 310 ηs which is (310x10-9)x(8x4.096x106) samples which is approximately 10 samples. The second delay value is therefore 20352-10=20342, and so on. 6.6

The WCDMA RAKE RECEIVER

The nominal chip rate in W-CDMA is 4.096 Mchips/s for a 5MHZ channel. This results in a transmit signal that occupies much higher bandwidth than narrowband CDMA systems, and hence greater immunity to frequency selective fading. As in second generation systems, a Rake receiver is used to demodulate the wideband spread spectrum signals. However, given the performance requirements, the number of Rake fingers1 needed for required diversity improvement is greater for 3G WCDMA. In our demonstration models, we use a six finger Rake receiver, with finger selection capability for combining, wherein fingers processing weaker multipath signals are not selected for combining. As shown in Figure 1, on the Forward link, timemultiplexed pilot and data symbols are used. Each 10 ms frame is divided into 16 slots, and known pilot symbols are inserted at the beginning of each slot. The pilot symbols serve as reference symbols for channel amplitude and phase estimation and Rake finger power estimation for finger selection. Channel estimation is performed by simply averaging over the duration of the pilot symbols. 1. By definition, a finger denotes a resolvable path for demodulation of the received signal. If there are N fingers, N reflections of the transmitted signal arriving at the receiver with different delays may be detected and combined.

Additionally, a weighted sum of the averaged channel estimates for multiple slots is used as the final estimate for one slot. Depending on the type of fading (fast or slow, depending on mobile velocity), the number of slots used for channel estimation as well as the weighting coefficients may be varied. The composite received signal at the mobile station consists of the sum of multiple reflections of the transmitted signal with different delays, amplitudes and phase shifts. In the forward link demonstration model, only steady state Rake processing is modeled i.e. the receiver assumes knowledge of the multipath delays, and the optimum sampling instants of the RC matched filter output for the various Rake fingers are delay adjusted to correspond to the beginning of the second time slot. At the base station transmitter, the I and Q channels are spread using the same OVSF channelization code and the same long scrambling code. Hence, the Rake demodulator, to begin with, consists of a despreading unit, that despreads the I and Q channels separately using the product of long scrambling and short spreading codes. Correlation is then performed over one symbol duration, which corresponds to different number of chips for different traffic bit rates. This results in an estimate of the transmitted symbol corrupted by multiplicative fading. To compensate for the amplitude and phase shift introduced in the fading channel, estimates of the channel coefficients are made by correlating the despread estimates obtained above with the known pilot symbols at pilot symbol positions. Additionally, as described earlier, a final estimate is obtained by computing the weighted average of estimates obtained for multiple slots. This final estimate is multiplied out from the data stream after appropriate delay to account for the delay in the channel estimation process. This process constitutes the maximal ratio weighting of Rake fingers before they are combined. Not all Rake fingers contain usable information for data demodulation. Depending on the state of the fading channel, certain fingers, at times, may be processing heavily faded signals that result in very noisy channel and symbol estimates. Unreliable information of this nature may be discarded by not selecting certain Rake fingers for combining. This decision is made from time to time, and hence the Rake fingers selected for combining change over time. In the forward link demonstration model, Rake finger selection is based on the estimated power of the pilot symbols detected within each finger. Only those fingers which contain pilot symbol power within a certain tolerance of the maximum received pilot symbol power

Version 0.4, February 15, 1999

8 of 10

An SPW Implementation of Wideband DS-CDMA Forward Link Physical Layer

are selected for combining. A QPSK slicer/demodulator is then used to make estimates of the transmitted bits. Now, having described the fundamentals concepts of the Rake receiver, we may describe in more detail the implementation of the Rake receiver in the SPW forward_link.system block. The output of the COMPLEX VARIABLE DELAY block is down-sampled by a factor of 8 before entering the Rake receiver. The Rake receiver, therefore, operates on chips. You may now click into the W-CDMA RAKE RECEIVER BLOCK also shown in Figure 6. In the top left of the block, there is a VECTOR TO SCALAR block which splits the vector into 6 parallel branches where each branch is fed into one finger. In the bottom left, there are two operations: spreading code and scrambling code generation and reference pilot symbol generation. The short spreading code and long scrambling code are those explained in Section 6.1. It is assumed that the mobile receiver has synchronized and acquired the long scrambling code through the cell search algorithm[2] and has negotiated the short spreading code with the base station. The pilot symbols are known at the receiver. The reference pilot symbols are generated and fed into each Rake finger after the appropriate (one slot) delay.

Figure 6 The SPW block for the W-CDMA Rake receiver.

Now you may click into any one of the 6 Rake fingers. The rake fingers are identical. The detail of a Rake finger is shown in Figure 7.

9 of 10

Figure 7 The SPW block for the W-CDMA Rake finger.

The input sequence is first despread using in-phase and quadrature despreading by the sequence generated in the mother block. The despreading is done using SIMPLE CORRELATE AND DUMP blocks which despreads and accumulates the incoming sequence and resets when the RESET input goes high. The despreading function is done without multiplication as the despreading code takes on values of either 0 or 1 hence simplifying the despreading function. The despread sequence is delayed by Nc/2 slots where Nc is the number of slots used for channel estimation set by the parameter Number of Slots for Channel Estimation. The default delay value shown should be 480 which is Nc=6 divided by 2 times Ns where Ns is the Number of Symbols per Slot, which equals to 480. In earlier versions of the system the value is mistakenly set at 640, (Nc/2+1)xNs. That should be corrected. This value ensures that the channel estimate is applied in the middle of the estimation period which is the beginning of the third slot for the default value of 6. The channel estimation is done using the blocks on the top right corner. The despread symbols are correlated with the known pilot sequence generated in the mother block. This operation generates the channel estimates. The COMPLEX DOWN SAMPLE BLOCK ensures that only the first Np samples of the correlation process are collected where Np is set by the parameter Number of Pilot Symbols per Slot. The output of the COMPLEX DOWN SAMPLE BLOCK is fed into a weighted averaging filter whose characteristic is set defined by a filter file. Click on the COMPLEX FIR FILTER BLOCK. The file is defined in the main parameter Impulse response

Version 0.4, February 15, 1999

An SPW Implementation of Wideband DS-CDMA Forward Link Physical Layer

signal file whose default value is “wcdma_lib/alpha”. Highlight the parameter, press the right button and select Edit Text File. The file should be opened. This file may be edited. However, the default weights are 0.3, 0.8, 0.1, 0.1, 0.8, 0.3. The number of weights is equal to the parameter Nc. So if Nc is changed, the files must be modified accordingly. The weights have been selected to give more weight to the estimate in the middle and less weight to the edges of the estimation period. The channel estimate is then upsampled by 160; i.e., Ns. This will provide an estimate for every symbol in the slot. The incoming delayed signal is then multiplied by the complex conjugate of the channel estimate. The sequence is then fed into a COMPLEX SCALAR TO VECTOR BLOCK of length Ns, and subsequently the pilots and the data signals are split apart into two vectors of length 152; i.e, (Ns-Np) and 8; i.e., (Np). The pilot power is then estimated using a magnitude squared followed by a summer and is sent to the output (pilot-pwr port). The channel compensated symbols are also sent to the output (out port). This process is performed in parallel for all the fingers. Now, go up one level to the RAKE RECEIVER block. The pilot power estimate out of each finger is fed into a SCALAR TO VECTOR to block followed by a vector element-wise summer of length Nsp, where Nsp is set by the parameter Number of Slots for Pilot Power Estimation. The default value of Nsp is 5. In other words, the powers of the pilots in 5 slots is summed to improve the power estimate. The power estimates out of the six fingers are then turned merged into a single vector and fed into the first RAKE FINGER SELECT which has six outputs o1 to o5. The output is binary value (0 or 1). The binary values are upsampled by a factor (Ns-Np)xNsp whose default value is 152*5=760. This provides a binary value for each data symbol within the slot. The value is one if the power estimate is greater than ktPmax where kt is set by the parameter Fraction of Maximum Power Used for Combining and Pmax is the power of the pilot symbols on the best finger. The default value of kt is 10%. The data symbol sequence from each finger and the binary finger selection information are then fed into the second RAKE FINGER SELECT block. Click on the second RAKE FINGER SELECT block. There are 6 complex switches. The switches are fed with the data symbol sequence and controlled by the binary finger selection information. The output of these switches are then fed into summers. This would ensure that only the data symbol sequence corresponding to the selected fingers will be combined.

Now go two levels up to the W-CDMA RAKE RECEIVER. There are many blocks on the top right side which pertain to the interactive simulations (ISL) capabilities. We will not discuss the functions of these blocks as they are not integral to the functionality of the models. 6.7

The QPSK DEMOD

The Rake combined data symbols are finally fed into a QPSK SLICER which selects the most likely QPSK symbol and outputs the corresponding number (between 0 and 3). The number is fed into an INTEGER TO BITS block. The final output corresponds to an estimate of the transmitted binary sequence. The remaining blocks not explained in this paper pertain to the interactive simulations. References [1] “Concept Group Alpha-Wideband Direct Sequence CDMA (W-CDMA) Evaluation Document (3.0), Part 1: System Description, Performance Evaluation”, ETSI TDOC SMG2 359/ 97. [2] S. Alamouti, M. Krishnan,“An Implementation of the Cell Search Algorithm for Wideband DS-CDMA Using SPW”, Cadence Design Systems Publication, Version 0.5, Sept. 5, 1998. [3] Robert Gold,“Optimal Binary Sequences for Spread Spectrum Multiplexing”, IEEE Transactions on Information Theory, Vol. IT-13, no. 4, pp. 619-621, 1967. [4] “Guidelines for Evaluation of Radio Transmission Technologies for IMT-2000/FPLMTS”, ITU Document 8/29-E Task Group 8/1.

Version 0.4, February 15, 1999

10 of 10