Visual Media Coding and Transmission

Wireless Channel Models 281 y > 504 x TTI block xxxx Filler TTI –Transmission Time Interval bits Figure 8.30 Example of block segmentation at the channel encoder Eight tail bits with binary value 0 are added to the end of the code block before encoding, and initial values of the shift register are set to 0s when the encoding is started. The generator polynomials used in the encoding as given in [38] are: Rate 1/2 convolutional coder: G0 ¼ 1 þ D2 þ D3 þ D4 þ D8 ð8:4Þ G1 ¼ 1 þ D þ D2 þ D3 þ D5 þ D7 þ D8 Rate 1/3 convolutional coder: G0 ¼ 1 þ D2 þ D3 þ D5 þ D6 þ D7 þ D8 ð8:5Þ G1 ¼ 1 þ D þ D3 þ D4 þ D7 þ D8 G2 ¼ 1 þ D þ D2 þ D5 þ D8 Note that the UTRA uses two different sets of generator polynomials to achieve two different convolutional code rates. If Ki denotes the number of bits in the ith code block before encoding then the number of bits after encoding, Yi, is: Yi ¼ 2Ki þ 16 with 1/2 rate coding. Yi ¼ 3Ki þ 24 with 1/3 rate coding. Turbo Coding Turbo codes employ two or more error control codes, which are arranged in such a way as to enhance the coding gain. They have been demonstrated to closely approach the Shannon capacity limit on both AWGN and Rayleigh fading channels. Traditionally, two parallel or serial concatenated recursive convolutional codes are used in the encoder implementation. A bit interleaver is used in between the encoders. Generated parity bitstreams from two encoders are finally multiplexed to produce the output turbo-coded bitstream. Turbo decoding is carried out iteratively. The whole process results in a code that has powerful error-correction properties. The defined turbo coder for use in UMTS is a parallel-concatenated convolutional code with two eight-state constituent encoders and one turbo-code internal interleaver. The coding rate of the turbo coder is 1/3. Figure 8.31 shows the configuration of the turbo coder.

282 Visual Media Coding and Transmission 1st constituent encoder xk DDD zk xk Input Input 2nd constituent encoder Output DDD Turbo code z’k internal interleaver x’k Output x’k Figure 8.31 Structure of rate 1/3 turbo coder [38]. Reproduced, with permission, from “3rd Generation Partnership Project; technical specification group radio access network; multiplexing and channel coding (FDD) (release 4)”, 3GPP TS 25.212 V4.6.0 (2002 09). Ó2002 3GPP. Ó1998 3GPP. Reproduced by permission of Ó European Telecommunications Standards Institute 2008. Further use, modification, redistribution is strictly prohibited. ETSI standards are available from http://pda.etsi.org/pda/ The transfer function of the eight-state constituent code is defined as:  g1 ðDÞ GðDÞ ¼ 1; g0 ðDÞ ð8:6Þ where: g0ðDÞ ¼ 1 þ D2 þ D3 ð8:7Þ g1ðDÞ ¼ 1 þ D þ D3 The initial values of the shift registers are set to 0s at the start of the encoding. Output from the turbo code is read as X1, Z1, Z01, and so on. Termination of the turbo coder defined in UMTS is performed in a different way to conventional turbo code termination, which uses 0 incoming bits to generate the trellis termination bits. Here, the shift register feedbacks after all information bits have been encoded are used to generate the termination bits. To terminate the first constituent encoder, switch A in Figure 8.31 is set to the lower position while the second constituent encoder is disabled. Likewise, the second constituent encoder is terminated by setting switch B in Figure 8.31 to the lower position while the first constituent encoder is disabled. The turbo code internal interleaver arranges incoming bits into a matrix. If the number of incoming bits is less than the number of bits that the matrix could contain, padding bits are used. Then intra-row and inter-row permutations are performed according to the algorithm given in [38]. Pruning is performed at the output, so the output block size is guaranteed to be equal to

Wireless Channel Models 283 the input block size. If Ki denotes the number of input bits in a code block, the number of turbo code output bits Yi, is Yi ¼ 3Ki þ 12 for 1/3 code rate. The minimum block size and the maximum block size for turbo coding are defined as 40 bits and 5114 bits respectively. Data sizes below 40 bits can be coded with turbo codes; however, in such a case, dummy bits are used to fill the 40 bit minimum-size interleaver. If the incoming block size exceeds the maximum size then segmentation is performed before channel coding. 8.3.2.2 Rate Matching Rate matching is used to match the incoming data bits to available bits on the radio frame. Rate matching is achieved either by bit puncturing or by repetition. If the amount of incoming data is larger than the number of bits that can be accommodated in a single frame then bit puncturing is performed. Otherwise, bit repetition is performed. In the case of transport channel multi- plexing, rate matching should take into account the number of bits arriving in other transport channels. The rate matching algorithm depends on the channel coding applied. The corresponding rate matching algorithms for convolutional and turbo coding are defined in [38]. In the simulation under discussion, rate matching is only performed for the signaling bearer. As signaling data is protected using convolutional codes, the rate matching algorithm is implemented only for convolutional coding. 8.3.2.3 Interleaving In UTRA, data interleaving is performed in two steps: first and second interleaving. These are also known as inter-frame interleaving and intra-frame interleaving, respectively. The first interleaving is a block interleaver with inter-column permutations (inter-frame permutation) and is used when the delay budget allows more than 10 ms of interleaving. In other words, the specified transmission time interval (TTI), which indicates how often data comes from higher layers to the physical layer, is larger than 10 ms. The TTI is directly related to the interleaving period and can take values of 10, 20, 40 or 80 ms. Table 8.14 shows the inter-column permutation patterns for first interleaving. Each column contains data bits for 10 ms duration. The second or intra-frame interleaving performs data interleaving within a 10 ms radio frame. This is also a block interleaver with inter-column permutations applied. Incoming data Table 8.14 Inter columns permutation patterns for first interleaving [38]. Reproduced, with permission, from “3rd Generation Partnership Project; technical specification group radio access network; multi plexing and channel coding (FDD) (release 4)”, 3GPP TS 25.213 V4.3.0. (2002 06). Ó2002 3GPP. Ó1998 3GPP. Reproduced by permission of Ó European Telecommunications Standards Institute 2008. Further use, modification, redistribution is strictly prohibited. ETSI standards are available from http:// pda.etsi.org/pda/ TTI Number of columns Inter column permutation patterns 10 ms 1 <0 > 20 ms 2 <0,1 > 40 ms 4 <0,2,1,3 > 80 ms 8 <0,4,2,6,1,5,3,7 >

284 Visual Media Coding and Transmission Table 8.15 Inter columns permutation patterns for second interleaving [38]. Reproduced, with permission, from “3rd Generation Partnership Project; technical specification group radio access network; multiplexing and channel coding (FDD) (release 4)”, 3GPP TS 25.213 V4.3.0. (2002 06). Ó2002 3GPP. Ó1998 3GPP. Reproduced by permission of Ó European Telecommunications Standards Institute 2008. Further use, modification, redistribution is strictly prohibited. ETSI standards are available from http://pda.etsi.org/pda/ Number of columns Inter column permutation patterns 30 <0,20,10,5,15,25,3,13,23,8,18,28,1,11,21, 6,16,26,4,14,24,19,9,29,12,2,7,22,27,17 > bits are input into a matrix with n rows and 30 columns, row by row with a starting position of column 0 and row 0. The number of rows is the minimum integer n, which satisfies the condition: Total number of bits in radio block n  30 If the total number of bits in the radio block is less than that, it is necessary to fill the whole matrix, and bit padding is performed. The inter-column permutation for the matrix is performed based on the pattern shown in Table 8.15. Output is read out from the matrix column by column, and finally pruning is performed to remove padding bits that were added to the input of the matrix before the inter-column permutation. 8.3.2.4 Spreading and Scrambling The spreading in the downlink is based on the channelization codes and is used to preserve the orthogonality among different downlink physical channels within one cell (or sector of a cell), and to spread the data to the chip rate, which is 3.84 Mcps. In UTRA, spreading is based on the orthogonal variable spreading factor (OVSF) technique. The OVSF code tree is illustrated in Figure 8.32. Typically only one OVSF code tree is used per cell sector in the base station (or node B). The common channels and dedicated channels share the same code tree resources. The codes are normally picked from the code tree; however, there are certain restrictions as to which of the codes can be used for a downlink transmission. A physical channel can only use a certain code from the tree if no other physical channel is using a code that is on an underlying branch. Neither can a smaller spreading factor code on the path to the root of the tree be used. This is because even though all codes from the same level are orthogonal to each other, two codes from different levels are orthogonal to each other only if one of them is not the mother code of the other. The radio network controller in the network manages the downlink orthogonal codes within each base station. The spreading factor on the downlink may vary from 4 to 512 (an integer power of 2), depending on the data rate of the channel. Table 8.16 summarizes the channel bit rates, data rates, and spreading factors for downlink dedicated physical channels. In addition to spreading, a scrambling operation is performed in the transmitter. This is used to separate base stations (cell sectors) from one another. As the chip rate is already achieved

Wireless Channel Models 285 (c, c) c (c, -c) C4,1 = (1,1,1,1) C4,1= (1, 1, 1, 1, 1, 1, 1, 1) C2,1= (1,1) C4,1= (1, 1, 1, 1,-1,-1,-1,-1) C1,1= (1) C4,2 = (1,1,-1,-1) C4,1= (1, 1,-1,-1, 1, 1,-1,-1) C4,1= (1, 1,-1,-1,-1,-1, 1, 1) C4,3 = (1,-1,1,-1) C4,1= (1,-1, 1,-1, 1,-1, 1,-1) C4,1= (1,-1, 1,-1,-1, 1,-1, 1) C2,2= (1,-1) C4,4 = (1,-1,-1,1) C4,1= (1,-1,-1, 1, 1,-1,-1, 1) C4,1= (1,-1,-1, 1,-1, 1, 1,-1) SF4 SF8 SF16 SF32 Figure 8.32 Example of OVSF code tree used for downlink with spreading, the symbol rate is not affected by scrambling. The downlink scrambling uses the Gold codes [39]. The number of primary scrambling codes is limited to 512, simplifying the cell search procedure. The secondary scrambling codes are used in the case of beam-steering and adaptive antenna techniques [40]. 8.3.2.5 Modulation Quadrature phase shift keying (QPSK) modulation is applied on time-multiplexed control and data streams on the downlink. Each pair of consecutive symbols is serial-to-parallel converted and mapped on to I and Q branches. The symbols on I and Q branches are then spread to the chip rate by the same real-valued channelization code. The spread signal is then scrambled by a cell- specific complex-valued scrambling code. Table 8.16 Downlink dedicated channel bit rates Spreading factor Channel bit rate (kbps) User data rate with 1/2 rate coding (approx.) 4 1920 936 kbps 8 960 456 kbps 16 480 215 kbps 32 240 105 kbps 64 120 45 kbps 128 60 20 24 kbps 256 30 6 12 kbps 512 15 1 3 kbps

286 Visual Media Coding and Transmission cos(ωt) Re{S} Complexed Split real & Pulse valued chip- shaping sequence from {S} imaginary Im{S} spreading Pulse operations parts shaping –sin(ωt) Figure 8.33 Downlink modulation [39]. Reproduced, with permission, from “3rd Generation Partner ship Project; technical specification group radio access network; spreading and modulation (FDD) (release 4)”, 3GPP TS 25.213 V4.3.0. (2002 06). Ó2002 3GPP. Ó1998 3GPP. Reproduced by permission of Ó European Telecommunications Standards Institute 2008. Further use, modification, redistribution is strictly prohibited. ETSI standards are available from http://pda.etsi.org/pda/ Figure 8.33 shows the spreading and modulation procedure for a downlink physical channel. A square-root raised cosine filter with a roll-off factor of 0.22 is employed for pulse shaping, and the pulsed shaped signal is subsequently up-converted and transmitted. 8.3.2.6 Physical Channel Mapping The frame structure for a downlink dedicated physical channel is shown in Figure 8.34. Each radio frame has 15 equal-length slots. The slot length is 2560 chips. As shown in Figure 8.34, the DPCCH and DPDCH are time-multiplexed within the same slot [41]. Each slot consists of pilot symbols, transmit power control (TPC) bits, transport format combination indicator (TFCI) bits, and bearer data. The number of information bits transmitted in a single slot depends on the source data rates, the channel coding used, the spreading factor, and the channel symbol rate. The exact number of bits in the downlink DPCH fields is given in [41] and is summarized in Table 8.17. 8.3.2.7 Propagation Model The channel model used in the simulator is the multipath propagation model specified by IMT2000 in [36]. This model takes into account that the mobile radio environment is Slot 0 Slot 1 One radio frame, 10 ms Slot 13 Slot 14 Slot i One time slot, 2560 Data 1 TPC TFCI Data 2 Pilot DPCCH DPDCH DPCCH DPDCH Figure 8.34 Frame structure for downlink DPCH

Wireless Channel Models 287 Table 8.17 DPDCH and DPCCH fields (3GPP TS 25.211). Reproduced, with permission, from “3rd Generation Partnership Project; technical specification group radio access network; physical channels and mapping of transport channel on to physical channel (FDD) (release 4)”, 3GPP TS 25.211 V4.6.0. (2002 09). Ó2002 3GPP. Ó1998 3GPP. Reproduced by permission of Ó European Telecommunications Standards Institute 2008. Further use, modification, redistribution is strictly prohibited. ETSI standards are available from http://pda.etsi.org/pda/ Spreading factor DPDCH (bits/slot) DPCCH (bits/slot) Ndata1 Ndata2 NTPC NTFCI NPilot 512 0 2 2 2 4 256 2 6 2 2 8 128 6 22 2 2 8 64 12 48 4 8 8 32 28 112 4 8 8 16 56 232 8 8 16 8 120 488 8 8 16 4 248 1000 8 8 16 dispersive, with several reflectors and scatterers. For this reason, the transmitted signal may reach the receiver via a number of distinct paths, each having different delays and amplitudes. The multipath fast fading is modeled by the superposition of multiple single-faded paths with different arrival times and different average powers for specified power-delay profiles in [36]. Each path is characterized by Rayleigh distribution (first-order statistic) and classic Doppler spectrum (second-order statistic). Figure 8.35 shows a block diagram of a four-path frequency selective fading channel. UTRAN defines three different multipath power-delay profiles for use in different propaga- tion environments. There are indoor office environments, outdoor-to-indoor and pedestrian environments, and vehicular environments. All of these models are implemented in the simulator and the tapped-delay line parameters for the vehicular environment are shown in Table 8.18. Mobile channel impulse response is updated 100 times for every coherence time interval. Power0 INPUT Rayleigh fade Power1 Rayleigh Delay1 fade Power2 Delay2 Delay3 Rayleigh Power3 fade ∑ Rayleigh fade Figure 8.35 Four path frequency selective fading channel

288 Visual Media Coding and Transmission Table 8.18 Vehicular A test environment [36]. Reproduced, with permission, from “Universal Mobile Telecommunications System (UMTS); selection procedures for the choice of radio transmission technologies of the UMTS (UMTS 30.03 version 3.2.0)”, TR 101 112 V3.2.0 (1998 04) Ó1998 3GPP. Ó1998 3GPP. Reproduced by permission of Ó European Telecommunications Standards Institute 2008. Further use, modification, redistribution is strictly prohibited. ETSI standards are available from http:// pda.etsi.org/pda/ Taps Delay (nsec) Power (dB) Doppler Spectrum 10 0 Classic 2 310 1.0 Classic 3 710 9.0 Classic 4 1090 10.0 Classic 5 1730 15.0 Classic 6 2510 20.0 Classic After the multipath channel shown in Figure 8.29, white Gaussian noise is added to simulate the effect of overall interference in the system, including thermal noise and inter-cell interference. 8.3.2.8 Rake Receiver The rake receiver is a coherent receiver that attempts to collect the signal energy from all received signal paths that carry the same information. The rake receiver therefore can significantly reduce the fading caused by these multiple paths. The operation of the rake receiver follows three main operating principles (see Figure 8.36). The first is the identification of the time-delay positions at which significant energy arrives, and the time alignment of rake fingers for combining. The second is the tracking of fast-changing phase and amplitude values originating from the fast-fading process within each correlation receiver, and the removal of these from the incoming data. Finally, the demodulated and phase- adjusted symbols are combined across all active fingers and passed to the decoder for further processing. The combination can be processed using three different methods: Input Signal Correlator Phase Delay rotator equalizer Code Channel Combiner generators estimator Timing (multipath tracking) Finger 1 Finger 2 Finger 3 Figure 8.36 Block diagram of a rake receiver

Wireless Channel Models 289 . Equal gain combining (EGC), where the output from each finger is simply summed with equal gain. . Maximal ratio combining (MRC), where finger outputs are scaled by a gain proportional to the square root of the SNR of each finger before combining [42]. In this case, the finger output with the highest power dominates in the combination. . Selective combining (SC), where not all finger outputs are considered in the combination, but some fingers are selected for combining according to the received power on each. Two types of rake receiver have been developed for the downlink: 1. Ideal rake receiver. 2. Quasi-ideal rake receiver. The following subsections provide details of the designs of these rake receivers. Ideal Rake Receiver The block diagram of an ideal rake receiver is shown in Figure 8.37. Here, perfect channel estimation and perfect finger time alignment is assumed. This is implemented by storing all the fast-fading channel coefficients as a complex vector, where the vector length equals the number of frequency selective fading paths. This vector is then fed from the channel directly to the ideal receiver. At the receiver, the coefficients for each path are first separated and then applied to each rake finger, after being time-aligned in accordance with the delay (from channel delay profile) in each reflected path. Of the three rake finger combination methods, EGC is selected for the ideal receiver as it gives the optimal performance in the presence of ideal channel estimation and perfect time alignment. Quasi-ideal Rake Receiver The quasi-ideal rake receiver resembles a practical rake receiver in terms of implementation. However, as depicted in Figure 8.38, ideal finger search for the rake receiver is assumed. That is, each finger in the receiver is assumed to have perfect synchronization with the corre- sponding path in the channel. First the received data is time-aligned according to the channel Input Finger time Complex alignment correlator Scrambling Spreading Complex From Output code code conjugate To Finger2 Equal To FingerN Finger2 Gain Channel From coefficient Combiner FingerN Figure 8.37 Ideal rake receiver

290 Visual Media Coding and Transmission Input Finger time De-scrambling/ Delay alignment De-spreading Matched Complex FIR Complex From Output filter filter conjugate Finger2 To Finger2 Maximum To FingerN Pilot bits Ratio From Combiner FingerN Figure 8.38 Qausi ideal rake receiver delay profile. Then the data on each finger undergoes a complex correlator process to remove the scrambling and spreading codes. As in an actual receiver implementation, pilot bits are used to estimate the momentary channel state for a particular finger. Channel estimation is achieved through the use of a matched filter, which is employed only during the period in which the pilot bits are being received. The pilot bits are sent in every transmit time slot; therefore, the maximum effective channel updating interval is equivalent to half a slot. Output from the matched filter is further refined by using a complex FIR filter. Here, the weighted multislot averaging (WMSA) technique proposed in [43] is employed to reduce the noise variance, and also to track fast channel variation between consecutive channel estimates. Two different sets of weighting for the WSMA filter are used for low vehicular speeds and high vehicular speeds respectively. This is because the limiting factor in channel estimation errors at low vehicular speed is the channel noise rather than the channel variation. Therefore, the noise averaging effect is more desirable at low vehicle speed. By comparison, at high vehicle speed channel variation becomes the limiting factor, hence weighting based on interpolation should be considered. The WMSA technique requires a delay of an integer number of time slots for the channel processing. The time-varying channel effect is removed from the de-scrambled and de-spread signal before it is sent to the signal combiner. MRC is used for the rake finger combination as it gives better performance. Intersymbol interference due to the multipath is implicit in the resulting output. 8.3.2.9 Channel Decoding In the implementation, a soft-decision Viterbi algorithm is used for the decoding of the convolutional codes. Turbo decoding is based on the standard LogMap algorithm (which is provided in SPW), which tries to minimize the bit error ratio rather than the block error ratio [42]. Eight iterations are performed. 8.3.3 Model Verification for Forward Link The theoretical formula for the BER probability with an order L MRC diversity combiner is given in [44], and is stated as:

Wireless Channel Models 291 1 XL  r gk  2 1 1 þg Pb ¼ k¼1 pk À ð8:8Þ k where Pb is the bit error probability, L denotes the number of diversity path, and gk is the mean Eb/h for kth diversity path. pk is given by: pk ¼ YL gk ð8:9Þ i¼1 gk Àgi i„k If the rake receiver is assumed to behave as an order L MRC diversity combiner, Equation (8.9) gives the lower bound of the BER performance. Other test conditions assumed in Equation (8.9) are: . Perfect channel estimation. . No intersymbol interference presence. . Each propagation path has a Rayleigh envelope. Using Equation (8.9), the theoretical lower bound of the performance for the power delay profile that is specified in the case 3 outdoor performance measurement test environment in annex B [30] is calculated. It is depicted in Figure 8.39. Here, a mean SNR value for each Figure 8.39 Performance of uncoded channel

292 Visual Media Coding and Transmission individual path is calculated from the global Eb/No by simply multiplying it by the fraction of power carried by each path (given in power delay profile). The number of rake fingers equals the number of propagation paths, which is four in this case. Figure 8.39 shows the performance in terms of raw BER (uncoded) for varying spreading factors in the above-described test environment. A single active connection is considered. The dashed lines show the performance obtained by Olmos and Ruiz in [44] for similar test conditions. Figure 8.39 clearly illustrates the close match of results obtained from the described model to those given in [44]. As the spreading factor reduces, the performance deviates from that of the theoretical bound, due to the presence of intersymbol interference. For non-ideal channel estimation, raw BER performance (see Figure 8.40) deviates considerably from the ideal channel estimation performance. Performance degradation is about 3 4 dB for operation at lower Eb/No, and increases gradually as Eb/No increases. It should be emphasized here that the channel-coding algorithm can correct almost all of the channel error occurrences if the raw BER values are less than 10 2. Therefore, the region that is interesting for multimedia applications is limited to the top-left-hand corner in Figure 8.40. 8.3.3.1 Model Performance Validation Reference performance figures for the downlink dedicated physical channels are given in [30]. These allow for the setting of reference transmitter and receiver performance figures for nominal error ratios, sensitivity levels, interference levels, and different propagation conditions. Reference measurement channel configurations are specified in Annex A [30], Figure 8.40 Comparison of ideal and non ideal channel estimate performance

Wireless Channel Models 293 Table 8.19 BLER performance requirement [30]. Reproduced, with per mission, from “3rd Generation Partnership Project; technical specification group radio access network; user equipment (UE) radio transmission and reception (FDD) (release 4)”, 3GPP TS 25.101 V410.0 (2002 03). Ó2004 3GPP. Ó1998 3GPP. Reproduced by permission of Ó European Telecommu nications Standards Institute 2008. Further use, modification, redistribution is strictly prohibited. ETSI standards are available from http://pda.etsi.org/pda/ Test number DPCH EC/Ior BLER 1 2 11.8 dB 10À2 8.1 dB 10À1 3 7.4 dB 10À2 6.8 dB 10À3 4 9.0 dB 10À1 8.5 dB 10À2 8.0 dB 10À3 5.9 dB 10À1 5.1 dB 10À2 4.4 dB 10À3 while the reference propagation conditions are specified in Annex B [30]. A mechanism to simulate the interference from other users and control channels in the downlink (named orthogonal channel noise simulator (OCNS)) on the dedicated channel is shown in Annex C [30]. The performance requirements are given in terms of block error ratio (BLER) for different multipath propagation conditions, data rates (hence spreading factors), and channel coding schemes. For example, Table 8.19 lists the required upper bound of BLER for the reference parameter setting shown in Table 8.20. The power-delay profile of the multipath fading propagation condition used in the reference test is given in Table 8.21. Physical channel parameters, transport channel parameters, channel coding, and channel mapping for the 64 kbps reference test channel are depicted in Figure 8.41. As in a typical operating scenario, two transport channels, the data channel and the signaling channel, are multiplexed and mapped on to the same physical channel. Table 8.20 Reference parameter setting [30]. Reproduced, with permission, from “3rd Generation Partnership Project; technical specification group radio access network; user equipment (UE) radio transmission and reception (FDD) (release 4)”, 3GPP TS 25.101 V410.0 (2002 03). Ó2004 3GPP. Ó1998 3GPP. Reproduced by permission of Ó European Telecommunications Standards Institute 2008. Further use, modification, redistribution is strictly prohibited. ETSI standards are available from http://pda.etsi. org/pda/ Parameter Unit Test 1 Test 2 Test 3 Test 4 Ior/Ioc dB 3 3 3 6 Ioc dBm/3.84 MHz 60 60 60 60 Information data rate kbps 12.2 64 144 384

294 Visual Media Coding and Transmission Table 8.21 Power delay profile for case 3 test environment [30]. Reproduced, with permission, from “3rd Generation Partnership Project; technical specifi cation group radio access network; user equipment (UE) radio transmission and reception (FDD) (release 4)”, 3GPP TS 25.101 V410.0 (2002 03). Ó2004 3GPP. Ó1998 3GPP. Reproduced by permission of Ó European Telecommu nications Standards Institute 2008. Further use, modification, redistribution is strictly prohibited. ETSI standards are available from http://pda.etsi.org/pda/ Relative delay ns Average power dB Fading 00 Classical Doppler 260 3 Classical Doppler 521 6 Classical Doppler 781 9 Classical Doppler 8.3.3.2 Calculation of Eb/No and DPCH EC/Ior Reference test settings are given in terms of the ratio of energy per chip to the total transmit power spectral density of the node B antenna connector. The relationship between Eb/No and the setting of the variance s of the AWGN source and the conversion of DPCH Ec/Ior to Eb/No is given in Equations (8.10) and (8.11) respectively. The derivations of these equations are given in Appendix A. DTCH DCCH Information data 1280 CRC16 Information data 100 CRC12 CRC detection 1280 CRC detection 100 112 Tail8 Turbo code R=1/3 1296 3888 Termination 12 Tail bit discard Viterbi decoding R=1/3 360 Rate matching 4014 Rate matching 372 1st interleaving 4014 1st interleaving 372 #1 2007 #2 2007 #1 2007 #2 2007 #1 93 #2 93 #3 93 #4 93 Radio Frame Segmentation 2007 93 2007 93 2007 93 2007 93 2100 2nd interleaving 2100 2100 2100 slot segmentation 14 0 1 140140 140 • • • • 01 14 0 1 14 0 1 14 140 140 140 • • • • 140 140140 • • • • 140 140 140 • • • • 120ksps DPCH 0 1 •••• 14 0 1 • • • • 14 0 1 • • • • 14 0 1 • • • • 14 (including TFCI bits) Radio frame FN=4N Radio frame FN=4N+1 Radio frame FN=4N+2 Radio frame FN=4N+3 Figure 8.41 Channel coding of DL reference measurement channel (64 kbps) [30]. Reproduced, with permission, from “3rd Generation Partnership Project; technical specification group radio access network; user equipment (UE) radio transmission and reception (FDD) (release 4)”, 3GPP TS 25.101 V4.10.0. (2002 03). Ó2004 3GPP. Ó1998 3GPP. Reproduced by permission of Ó European Telecommunications Standards Institute 2008. Further use, modification, redistribution is strictly prohibited. ETSI standards are available from http://pda.etsi.org/pda/

Wireless Channel Models 295 Table 8.22 Performance validation for convolutional code use Propagation DPCH Ec/Ior at BLER ¼ 1% environment 3GPP upper bound CCSR model AWGN 16.6 dB 18.5 dB Case 1 15.0 dB 16.0 dB Case 2 12.5 dB Case 3 7.7 dB 14.6 dB 11.8 dB Eb ¼ RC Á ch os ð8:10Þ No 2 Á Rb Á s2 Eb DPCH Ec Á RC Á ch os No Rb ¼ Ior ð8:11Þ Ioc ^Ior where RC and Rb are the chip rate and the channel bit rate respectively, and ch os denotes the channel over sampling factor. Equation (8.11) is equivalent to the equation proposed by Ericsson in [28]. 8.3.3.3 UMTS DL Model Verification for Convolutional Code Use Reference interference performance figures [30] for convolutional code with 12.2 kbps data were compared to the results obtained using the designed simulation model, which is referred to as the CCSR model. A comparison is given in Table 8.22. The reference results are given in terms of the upper bound of the average downlink power, which is needed to achieve the specified block error ratio value. The results listed in Table 8.22 show that the CCSR model performance is within the performance requirement limits in all propagation conditions. The performance requirements specified in 3GPP are limited to a single value, which is insufficient to test the model performance over a range of propagation conditions. Therefore, performance tests were carried out for a range of DPCH Ec/Ior for different reference propagation environments, and the results are compared to those obtained by Ericsson [28] and NTT DoCoMo [29]. These results are shown in Figure 8.42. Figure 8.42 clearly illustrates the close performance of the CCSR model to the results given in the above references. BLER curves for case 1 and case 3 are virtually identical to those given in the above references. In the case 2 propagation environment, the CCSR model outperforms the other two reference figures. The reason for this may be the use of an EGC rake receiver in the CCSR model. Case 2 represents an imaginary radio environment, which consists of three paths with large relative delays and equal average power. Use of EGC in this environment combines energy from all three paths with equal gain, resulting in maximum power output, and shows optimal performance. The path combiner structures used in the reference models are unknown. Performance over AWGN environment shows slight variation at low-quality channels. Howev- er, the performance gets closer to that of the reference figure when the channel quality gets better.

296 Visual Media Coding and Transmission Figure 8.42 Comparison of reference performance for convolutional code use: (a) 12.2 kbps measure ment channel over AWGN channel; (b) 12.2 kbps measurement channel over case 1 environment; (c) 12.2 kbps measurement channel over case 2 environment; (d) 12.2 kbps measurement channel over case 3 environment 8.3.3.4 UMTS DL Model Verification for Turbo Code Use [30] also presents the upper bounds for performance of turbo codes under different propagation conditions. The results generated with the CCSR model at a BLER of 10% and 1% were compared to the above reference figures, as listed in Table 8.23 for the 64 kbps test channel and in Table 8.24 for the 144 kbps test channel. The results listed in the tables clearly show that under the propagation conditions described in [30], the performance of the CCSR model at a resulting BLER of 10% and 1% is within the required maximum limits. As for convolutional codes, performance for turbo codes are evaluated and compared to performance figures obtained by Ericsson [28] and NTT DoC- oMo [29]. The performance traces under different conditions are shown in Figure 8.43 for 64 kbps and in Figure 8.44 for 144 kbps. Dashed-dotted lines denote the results of Ericsson, while dashed lines with star marks show the results of NTT DoCoMo.

Wireless Channel Models 297 Table 8.23 Performance validation for 64 kbps channel Propagation environment DPCH/Ior at BLER ¼ 10% 3GPP upper bound CCSR model AWGN 13.1 dB 15.2 Case 1 13.9 dB 15.0 Case 2 10.5 Case 3 6.4 dB 10.9 8.1 dB AWGN DPCH/Ior at BLER ¼ 1% 14.95 Case 1 12.8 dB 10.7 Case 2 10.0 dB 7.5 Case 3 2.7 dB 10.1 7.4 dB The result for case 1 is very close to the results given in the above references. The result over AWGN channel shows closer performance to Ericsson figures. As with the convolutional code, the CCSR model outperforms the other two models in the operation over case 2 propagation environment. However, for the 144 kbps channel over case 3, the CCSR model results do not closely follow the reference traces. Moreover, even the reference traces do not show close performance in this environment. There are several reasons for this behavior. First, case 3 resembles a typical outdoor vehicular environment. Mobile speed is set to 120 kmph in this condition. Therefore the effect of time-varying multipath channel conditions and intercell interference are more evident in this condition, resulting in a variation in performance. Second, the implementation of the intercell interference could be different in each model. In the CCSR model, intercell interference is evaluated mathematically and is mapped on to the variance of the Gaussian noise source. Third, the decoding algorithm used for turbo iterative decoding in the CCSR model is the LogMap algorithm, while the reference models use the MaxLogMap algorithm. Even though these two algorithms show similar performance for AWGN channels, when applied over multipath propagation conditions their performance depends on other conditions, such as the length of the turbo-internal interleaver, input block length, and input signal amplitude [42]. Table 8.24 Performance validation for 144 kbps channel Propagation environment DPCH/Ior at BLER ¼ 10% AWGN 3GPP upper bound CCSR model Case 1 Case 2 9.9 dB 11.86 dB Case 3 10.6 dB 12.5 dB 13.0 dB AWGN 8.1 dB 12.3 dB Case 1 9.0 dB Case 2 DPCH/Ior at BLER ¼ 1% 11.72 dB Case 3 9.8 dB 7.6 dB 6.8 dB 9.5 dB 5.1 dB 10.75 dB 8.5 dB

298 Visual Media Coding and Transmission Figure 8.43 Comparison of reference performances for 64 kbps channel: (a) 64 kbps measurement channel over AWGN environment; (b) 64 kbps measurement channel over case 1 environment; (c) 64 kbps measurement channel over case 2 environment; (d) 64 kbps measurement channel over case 3 environment 8.3.4 UMTS Physical Link Layer Simulator From a user point of view, services are considered end-to-end; that means from one terminal equipment to another terminal equipment. An end-to-end service may have a certain quality of service, which is provided for the user by the different networks. In UMTS, it is the UMTS bearer service that provides the requested QoS, through the use of different QoS classes, as defined in [45]. At the physical layer, these QoS attributes are assigned a radio access bearer (RAB) with specific physical-layer parameters in order to guarantee quality of service over the air interface. RABs are normally accompanied by signaling radio bearers (SRBs). Typical parameter sets for reference RABs, signaling RBs, and important combinations of the two

Wireless Channel Models 299 Figure 8.44 Comparison of reference performances for 144 kbps channel: (a) 144 kbps measurement channel over AWGN environment; (b) 144 kbps measurement channel over case 1 environment; (c) 144 kbps measurement channel over case 2 environment; (d) 144 kbps measurement channel over case 3 environment (downlink, FDD) are presented in [37]. In the simulation, 3.4 kbps SRB, which is specified in [37], is used for the dedicated control channel (DCCH). Transport channel parameters for the 3.4 kbps SRB are summarized in Table 8.25. Careful examination of parameter sets for RABs and SRBs, which are specified in [37], shows that the minimum possible rate-matching ratios for RABs and SRBs vary with the physical-layer spreading factor being used. This is because when a higher spreading factor is used it adds transmission channel protection to the transmitted data in addition to the channel protection provided by the channel coding. Therefore, the channel bit error rate reduces with increase in spreading factor, and it is possible to allow higher puncturing in these scenarios without loss of performance. Table 8.26 shows the variation of calculated minimum rate- matching ratios (maximum puncturing ratio) with a spreading factor for FDD downlink control channels.

300 Visual Media Coding and Transmission Table 8.25 Transport channel parameter for 3.4 kbps signaling radio bearer [37]. Reproduced, with permission, from “3rd Generation Partnership Project; technical specification group terminals; common test environments for user equipment (UE) conformance testing (release 4)”, 3GPP TS 34.108 V4.7.0 (2003 06). Ó2003 3GPP. Ó1998 3GPP. Reproduced by permission of Ó European Telecommunications Standards Institute 2008. Further use, modification, redistribution is strictly prohibited. ETSI standards are available from http://pda.etsi.org/pda/ RLC Logical Channel Type DCCH DCCH DCCH DCCH RLC Mode UM AM AM AM Payload Size, bit 136 128 128 128 Max Data Rate, bps 3400 3200 3200 3200 AMD/UMD PDU 8 16 16 16 Header, bit MAC MAC Header, bit 4 Layer 1 MAC Multiplexing 4 logical channel TrCH Type multiplexing TB Sizes, bit DCH TTI, ms 148 Coding Type 40 CRC, bit CC 1/3 Max Number of bits/ 16 TTI before Rate 516 Matching In the simulation, rate-matching attributes for an SRB are set according to the minimum rate-matching ratios shown in Table 8.26 for different physical channels. The actual information data rate is a function of spreading factor, rate-matching ratio, type of channel coding, channel coding rate, number of CRC bits, and transport block size. Table 8.27 is a list of all the parameters that are user-definable, either by modifying the parameters of hierarchical models, by changing the building blocks that constitute the model, or by using different schematics. 8.3.4.1 BLER/BER Performance of Simulator Simulations were carried out for different radio bearer configuration settings. For higher spreading factor realizations, the simulation period was set to 60 s duration. However, for Table 8.26 Minimum rate matching ratios for SRB SF Minimum rate matching ratio 128 0.690 64 0.73 32 0.99 16 4 1.0

Wireless Channel Models 301 Table 8.27 UTRAN simulator parameters Parameters Settings CRC attachment 24, 16, 12, 8 or 0 Channel coding scheme supported No coding, 1/2 rate convolutional coding, 1/3 rate convo lutional coding, 1/3 rate turbo coding Interleaving 1st interleaving: block interleaver with interframe permutation Rate matching 2nd interleaving: block interleaver with inter columns per mutation (Permutation patterns are specified in [38]) TrCH multiplexing The algorithm (for convolutional rate matching) as specified Transport format detection in 3GPP TS 25.212. Rate matching ratio (repeat or punc Spreading factor turing ratio) is user definable Transmission time interval Experiments were conducted for two transport channels Pilot bit patterns TFCI based detection Interference/noise characteristics 512, 256, 128, 64, 32, 16, 8, 4 10 ms, 20 ms, 40 ms, 80 ms Fading characteristics As specified in 3GPP TS 25.211 User defined values are converted to the variance of AWGN Multipath characteristics source at receiver Mobile terminal velocity Rayleigh fading mobile channel impulse response is updated Chip rate 100 times for every coherence time interval Carrier frequency Vehicular, pedestrian Antenna characteristics User definable. Constant for the simulation run Receiver characteristics 3.84 Mcps 2000 MHz Transmission diversity characteristic 0 dB gain for both transmitter and receiver antenna Channel decoding Rake receiver with maximum ratio combining, equal gain combining, or selective combining. Number of rake fingers is Performance measures user definable Simulation length Closed loop fast power control [46] Soft decision Viterbi convolutional decoder Standard LogMap turbo decoder Number of turbo iterations is user definable Bit error patterns and block error patterns 3000 6000 radio frames, equivalent to 30 60 s duration spreading factor 8, the simulation period was limited to 30 s. This is to compensate for the higher processing time requirement seen at high data rates. 3000 6000 radio frames (10 ms) accommodate about 15 000 20 000 RLC blocks in a generated bit error sequence, and that is sufficiently long enough to obtain a meaningful BLER average. Furthermore, experimental results show that the selected simulation duration is sufficient to capture the bursty nature of the wireless channel and its effect on the perceptual quality of received video. Effect of Spreading Factor For the purpose of a performance comparison of the effect of spreading-factor variation, experiments were conducted for various spreading-factor allocations. The other physical channel parameters are set to their nominal values, which are shown in Table 8.28. The

302 Visual Media Coding and Transmission Table 8.28 Nominal parameter settings Spreading Factor 32 Transmission Time Interval 20 ms CRC Attachment 16 bits Channel Coding 1/2 CC, 1/3 CC, 1/3 TC Mobile Speed 3 kmph, 50 kmph Rate Matching Ratio 1.0 Operating Environment Vehicular A, pedestrian B Table 8.29 Channel throughput characteristics Turbo coding (1/3 rate) Spreading factor Convolutional coding (1/2 Convolutional coding (1/3 rate) rate) RLC payload Rate (kbps) RLC payload Rate (kbps) RLC payload Rate (kbps) 128 45 15.75 49 9.8 45 9.0 32 320 96.0 320 64.0 320 64.0 16 320 192.0 320 128.0 320 128.0 8 640 416.0 640 288.0 640 288.0 4 640 896.0 640 608.0 640 608.0 calculated possible information data rates are based on the specified SRBs for a given composite transport channel (consisting of one signaling channel and one dedicated data channel) for FDD downlink channels and are presented in Table 8.29. Table 8.29 also lists the RLC payload setting used in different bearer configurations. Figure 8.45 shows the BER performance for the transmission of uncoded data (raw BER/ channel BER) over the vehicular A propagation environment. It clearly illustrates the error- flow characteristics experienced due to the intersymbol interference in multipath channels. The effect of intersymbol interference increases with a reduction in spreading factor. However, the error-flow characteristic is not very pronounced in terms of coded BER performance, except for very low spreading factor allocations (see Figure 8.46). This is due to the effect of the channel coding algorithm, which tends to correct most of the errors if the channel bit error ratio is lower than 10 2. Figure 8.46(a) shows the performance of convolutional code, while the performance of turbo code is shown in Figure 8.46(b). The effect of spreading factor variation on the performance of turbo codes is similar to that of convolutional codes. However, the performance for spreading factor 8 is closer to that of other spreading factors than to the convolutional coding case. This is mainly due to the behavior of turbo codes. It is known that the higher the input block size of the turbo code, the better the performance. A high-bit-rate (with low spreading factor) service can accumulate more bits in a TTI than a low-bit-rate service. The better performance of the turbo code seen with large input block sizes compensates for the reduced robustness against interference provided by low spreading factor realizations. In Figure 8.46(a) and (b), the performance for 128 spreading factor is worse than that for 16 and 32 spreading factors. A possible reason is the poor performance of the interleavers (the first interleaver in convolutional

Wireless Channel Models 303 100 Performance over vehA 10−1 128sf 32sf 16sf 8sf ber 10−2 10−3 10−4 −10 0 10 20 30 40 −20 Eb/No (dB) Figure 8.45 Performance of uncoded channel over vehicular A environment coding and the first and turbo internal interleaver in turbo coding) in the presence of smaller input block sizes. Effect of Channel Coding Figure 8.47 illustrates the effect of channel coding schemes on the block error ratio and bit error ratio performances. The vehicular A channel environment is considered the test environment, while the spreading factor is set to 32. As expected, turbo coding shows better performance than the other channel coding schemes, while the 1/3 rate convolutional code outperforms the 1/2 rate convolution code. It must be emphasized that the plots shown are the BLER/BER performances vs Eb/No. If the BLER/BER performances are viewed vs transmitted power, Figure 8.46 Spreading factor effect for vehicular A environment: (a) 1/3 rate convolutional code; (b) 1/3 rate turbo code

304 Visual Media Coding and Transmission 100 TC1/3 100 TC1/3 10−1 CC1/3 10−1 CC1/3 10−2 CC1/2 10−2 CC1/2 10−3 10−3 10−4BER 10−4 10−5 BLER 10−6 2 4 6 8 10 12 10−50 2 4 6 8 10 12 0 Eb/No (dB) Eb/No (dB) (a) (b) Figure 8.47 Effect of channel coding scheme: (a) BER; (b) BLER performance significant improvements will be visible for the 1/3 rate coding scheme compared to the 1/2 rate coding scheme. This is because the transmit power is directly proportional to the source bit rate. As 1/2 rate coding supports higher source rates, the corresponding curve will be shifted more to the right than others. Furthermore, the convolutional code and the turbo code show closer BLER performance (shown in Figure 8.47(b)) than the BER performance shown in Figure 8.47 (a). This is due to the properties of the implemented LogMap algorithm at the turbo decoder, which is optimized to minimize the number of bit errors rather than the block error ratio [42]. Effect of Channel Environment Experiments were conducted to investigate the BLER performance for the pedestrian B channel environment. The mobile speed is set to 3 kmph. Results for 1/3 rate convolutional code with different spreading factors are shown in Figure 8.48. As is evident from the figure, the Figure 8.48 1/3 rate convolutional coding performance for the pedestrian B environment

Wireless Channel Models 305 resulting performance over the pedestrian B channel is much lower than that over the vehicular A channel environment when operating without fast power control. This is due to the slow channel variation associated with low mobile speeds. A larger number of consecutive information blocks can experience a long weaker channel condition during the transmission. This reduces the performance of block-based de-interleaving and channel decoding algo- rithms, resulting in a high block error ratio. On the other hand, a faster Doppler effect results in alternating weak and strong channel conditions of short durations at high vehicular speeds. This effect behaves as a time-domain transmit diversity technique and enhances the performance of block-based interleaving and channel coding algorithms. 8.3.4.2 Eb/No to Eb/Io and C/I Conversion The BER performance of UMTS-FDD systems depends on many factors, such as mean bit energy of the useful signal, thermal noise, and interference. Interference can be divided into three main parts, as intersymbol interference, intra-cell interference, and inter-cell interference. Due to the multiple receptions, the signal is received with significant delay spread in a multipath propagation environment. This causes the intersymbol interference. Orthogonal codes are used to separate users in the downlink. Without any multipath propagation, these codes can be considered as perfectly orthogonal to each other. However, in a multipath propagation environment, orthogo- nality among spreading codes deviates from perfection, due to the presence of delay spread of the received signal. Therefore, the mobile terminal sees part of the signal, which is transmitted to other users, as interference power, and it is labeled intra-cell interference. The interference power seen among users in neighboring cells is quantified as the inter-cell interference. BER performance is commonly written as a function of the global Eb/h, where the definition is given as: Eb ¼ No þ c þ ð1 Eb Á Io þ hISI ð8:12Þ h À rÞ Eb \"  1  1  1   # h Eb Eb Eb Eb No c Io hISI 1 ¼ þ þ ð1 À rÞ Á þ ð8:13Þ where Eb is the received energy per bit of the useful signal, No is the power density representing the system-generated thermal noise, h is the global noise power spectrum density, x is the inter- cell interference power spectral density, r is the orthogonality factor (OF), Io is the intra-cell interference power spectral density, and hISI is the power spectral density of the intersymbol interference of the received signal. These factors depend on: . The operating environment. . The number of active users per cell. . The used spreading factors in the code tree. . The cell site configurations. . Presence of the diversity techniques.

306 Visual Media Coding and Transmission . The mobile user locations. . The type of radio bearer. . The voice activity factor. The loss of orthogonality between simultaneously-transmitted signals on a WCDMA downlink is quantified by the OF. The lower the value of the OF, the smaller the interference; an OF of 1 corresponds to the perfect orthogonal case, while an OF near 0 indicates considerable downlink interference. The introduction of the orthogonality factor in modeling intra-cell interference allows the employment of the Gaussian hypothesis. It is employed where the equivalent Gaussian noise with power spectral density is equal to (1 À r) times the received intra-cell interference power, and is simply added at the receiver input. The statistics of the OF are normally derived from measurement data gathered through extensive field trial campaigns. In the designed UTRA downlink simulator, the OFs, which are derived based on the gathered channel data and are presented in [47], are used to simulate the intra-cell interference power. The inter-cell interference can also be modeled with the Gaussian hypothesis. The inter-cell interference power spectral density and the intra-cell interference power spectral density can be explicitly obtained through system-level simulations or analytical calculations based on simplifying assumptions and cell configuration [31]. However, intersymbol interference can only be obtained by chip-level simulation and does not depend on other factors, apart from used spreading factor, propagation condition, and mobile speed. Therefore, it is sufficient to obtain the Eb/h performance for one single connection (x ¼ 0, Io ¼ 0) of each of all the possible bit rates or spreading factors by chip-level simulation. Then the Eb/Io performances can easily be derived from Eb/h using Equation (8.14). No ( (1 À r)ÁIo is assumed, and the intersymbol interference is implicit in the simulation. Eb ¼ ð1 À rÞ Á Eb ð8:14Þ Io h Equation (8.15) shows the relationship between average energy per bit and average received signal power, S. S ¼ Eb Á R ð8:15Þ where R denotes data bit rate. Therefore: SIRc ¼ R  Eb ð8:16Þ c SIRI ¼ R  Eb ¼ ð1 À rÞ Á R Á Eb ð8:17Þ Io No SIRTotal ¼ ½SIRc 1 þ SIRI 1Š 1 ð8:18Þ

Wireless Channel Models 307 Table 8.30 Orthogonality factor variation for different cellular environments [47] Environment Code orthogonality factor Urban small Mean Std Urban large Rural large 0.571 0.159 0.514 0.212 0.626 0.190 [T]where SIRx,SIRI, and SIRTotal denote signal-to-inter-cell interference ratio, signal-to-intra- cell ratio, and signal-to-interference ratio respectively. Assume No ¼ 0.0002, x ¼ 0.005, and vehicular A propagation environment. From Table 8.30, the orthogonality factor is 0.514. The Eb/Io for Eb/h values shown in Figure 8.46(a) are calculated from Equation (8.14) and are shown in Figure 8.49. 8.3.5 Performance Enhancement Techniques 8.3.5.1 Fast Power Control for Downlink As can be seen in Figure 8.48, data transmission over the pedestrian B slow-speed propagation environment shows worse performance than over the high-speed propagation environment. This is mainly because the error-correcting methods are based on the interleaving and block- based methods, which do not work effectively in the presence of long-duration weak channel conditions caused by Rayleigh fading at low mobile speeds. This condition (weak long radio link) can be improved by the application of a fast power control algorithm. A closed-loop fast Figure 8.49 BLER performance: solid line shows BLER vs. Eb/No; dashed line shows BLER vs. Eb/h

308 Visual Media Coding and Transmission From Transmit Wireless Rake To channel transmitter power receiver decoder Feedback adjustment delay Transmit power Received decision power making unit Figure 8.50 Fast power control for UTRA FDD downlink power control algorithm has been designed and is incorporated in the simulator. The implementation and the resulting performance improvement are described below. 8.3.5.2 Algorithm Implementation A block diagram representation of the implemented power control algorithm is depicted in Figure 8.50. According to the measured received pilot power at the receiver, the UE generates appropriate transport power control (TPC) commands (whether to adjust transmit power up or down) to control the network transmit power and sends them in the TPC field of the uplink dedicated physical control channel (DPCCH). The TPC command decision is made by comparing the average received pilot power (averaged over an integer number of slots to mitigate the effects of varying interference and noise) to the pilot power threshold, which is predefined by the UTRAN based on the outer-loop power control [48]. Upon receiving the TPC commands, UTRAN adjusts its downlink DPCCH/DPDCH power according to Equa- tion (8.19). PðkÞ ¼ Pðk À 1Þ Æ DTPC ð8:19Þ where P(k) denotes the downlink transmit power in the kth slot, and DTPC is the power control step size. Æ is decided from the uplink TPC command. This algorithm is executed at a rate of 1500 times per second for each mobile connection. Settings used in the implementation are listed in Table 8.31. Note: the aggregated power control step is defined as the required total changes in a code channel in response to ten multiple consecutive power control commands. Table 8.31 Power control parameter settings [48] Power Control Step Size 0.5 Æ 0.25 dB Aggregated Power Control Step Change 4 6 dB Power Averaging Window Size, n 4 Feedback Delay 3 slots Algorithm Execution Frequency 1.5 kHz

Wireless Channel Models 309 2.5 1500 2000 2500 3000 2 1.5 1 0.5 0 1000 2.5 2 1.5 1 0.5 0 1500 2000 2500 3000 1000 Figure 8.51 Characteristic of the fast power control algorithm The control algorithm adjusts the power of the DPCCH and DPDCH; however, the relative power difference between the two is not changed. Figure 8.51 shows how a downlink closed-loop power control algorithm works on a fading channel at low vehicular speed. Node B transmit power varies inversely proportional to the received pilot power. This closely resembles the time-varying channel at low mobile speeds. Transmit power cut-off values are defined by the maximum and minimum power limits set by node B. Receive power at the receiver shows very little residual fading. Figure 8.52 illustrates the performance of the power control algorithm for data transmis- sion over the vehicular A propagation environment with a spreading factor 16 and 1/3 rate convolutional coding. The experiment was carried out at three different mobile speeds settings, namely 3, 50 and 120 kmph. The performance improvement by power control is evident at low speed, while at high mobile speed the improvement is largely insignificant. This is because a transmission diversity gain is provided by the highly time-varying channel at high mobile speed. 8.3.6 UMTS Radio Interface Data Flow Model The designed physical link layer simulator alone provides a necessary experimental platform to examine the effects of the radio link upon the data transmitted through the physical channel. However, in order to investigate the effect of channel bit errors upon the end-application, the application performance must be validated in an environment as close as possible to that of the real world. Therefore, not only the effect of the physical link layer but also the effect of UMTS protocol layer operation on multimedia performance should be investigated. A UMTS data flow model was designed in Microsoft Visual C þþ to emulate the protocol layer behavior.

310 Visual Media Coding and Transmission 100 10−1 120km/h 50km/h 3km/h FPC−120km/h FPC−50km/h FPC−3km/h bler 10−2 10−3 10−4 12 0 2 4 6 8 10 Eb/No (dB) Figure 8.52 Fast power control algorithm performance The design criteria follow a modular design strategy. Each of the protocol layers was implemented separately and protocol interaction is performed through the specified interfaces. This allows individual protocol-layer optimization, or improvement and testing of novel performance-enhancement algorithms in the presence of a complete system. Here, only the protocol-layer effect on multimedia performance is considered. The protocol layers imple- mented include the application layer, transport layer, PDCP layer, RLC/MAC layer, and layer 1. The block diagram of the data flow model is shown in Figure 8.53. The effect of protocol headers on application performance was emulated by allocating dummy headers. The application consists of a full error-resilience-enabled MPEG-4 video source. In addition to the employed error-resilience techniques, the TM5 rate-control algorithm is used in order to achieve a smoother output bit rate. An adaptive intra refresh algorithm is also implemented to stop temporal error propagation and to achieve a smoother output bit rate. The output source bit rate is set according to the guaranteed bit rate, which is a user-defined QoS parameter. For packet-switched connections, encoded video frames are forwarded to the transport layer at regular intervals, as defined by the video frame rate. Each video frame is encapsulated into an independent RTP/UDP/IP packet [22] for forwarding down to the PDCP layer. The PDCP exists mainly to adapt transport-layer packet to the radio environment by compressing headers with negotiable algorithms [49]. The current version of the data flow model implements the resulting compressed header sizes, but not the actual header- compression algorithms. More extensive and detailed examination of different header- compression algorithms and their effects on multimedia performance can be found in [50]. For circuit-switched connections, the output from the video encoder is directly forwarded to the RLC/MAC layer. At the RLC layer, forwarded information data is further segmented into RLC blocks. The size of an RLC block is defined by the transport block (TB) size, which is an implementation- dependent parameter. Optimal setting of TB size should include many factors, such as application type, source traffic statistics, allowable frame delay jitter, and RLC buffer size.

Wireless Channel Models 311 Rate Control Application Layer: Application Layer: M Video frame M Video frame Transport Layer: Transport Layer: PDCP Layer: PDCP Layer: Video Packetization RLC/MAC Layer: RLC/MAC Layer: Header Layer 1: Layer 1: compression/ No.compression (for PS) Bit error injection from physical link layer simulator Figure 8.53 UTRAN data flow model RLC block header size depends on the selected RLC mode for the transmission. Transparent mode adds no header as it transmits higher-layer payload units transparently. Unacknowledge mode and acknowledgement mode add 8 bit and 16 bit headers to each RLC block, respectively [34]. Apart from the segmentation and addition of a header field, other RLC layer functions such as error detection and retransmission of erroneous data are not im- plemented in the current version of the model as the main use of the model is to investigate the performance of conversational-type multimedia applications. The MAC layer can be either dedicated or shared. A dedicated mode is responsible for handling dedicated channels allocated to a UE in connected mode, while shared mode takes responsibility for handling shared channels. If channel multiplexing is performed at the MAC layer then a 4 bit MAC header is added to each RLC block [35]. Layer 1 attaches a CRC to forwarded RLC/MAC blocks. According to the specified TTI length, higher-layer blocks are combined to form the TTI blocks and store them in a buffer for further processing before transmitting over the air interface. The number of higher-layer PDUs to be encapsulated within the TTI block depends on the selected channel coding scheme, the spreading factor, and the rate-matching ratio. In a practical system, the selection of channel coding scheme, TTI length, and CRC size is normally performed by the radio resource management algorithm according to the end user quality requirement, application type, operating environment, system load, and so on. For experimental sake, these parameters are to be user-definable in the designed data flow model. An error-prone radio channel environment is emulated by applying generated bit errors from the physical link layer simulator to the information data at layer 1.

312 Visual Media Coding and Transmission The receiver side is emulated by reversing the described processing. Layer 1 segments the TTI blocks received over the simulated air interface into RLC/MAC blocks. After detaching CRC bits, RLC/MAC blocks are passed on to the RLC/MAC layer. At the RLC/MAC layer the received data is reassembled into PDCP data units for packet-switched connections. If IP/UDP/RTP headers are found to be corrupted, data encapsulated within the packet is dropped at the network layer. Finally, the received source data is displayed using an MPEG-4 decoder. This layered implementation of the UMTS protocol architecture allows the investigation of the effects of physical layer-generated bit errors upon different fields of the payload data units at each protocol layer. In other words, the data flow model can be used to map channel errors on to different PDU fields and to optimize protocol performance for the given application. 8.3.7 Real-time UTRAN Emulator The above-described UMTS data flow model is integrated with the physical link layer model to form the UTRAN emulator. The emulator software suite provides a graphical user interface for connection setup, radio bearer configuration, and performance monitoring. The emulator model considers the emulated system to be a black box, whose input output behavior intends to reproduce the real system without requiring knowledge of the internal structure and processes. It has also been designed for accurate operation in real time with moderate implementation complexity. The emulator was implemented in Visual C þþ , as it provides a comprehensive graphical user interface design environment. Figure 8.54 depicts the block diagram of the designed emulator architecture. It consists of three main parts, namely content server, UMTS emulator, and mobile client. An “MPEG-4 file transmitter” is used as the content server. It selects the corresponding video sequence, which is encoded according to the requested source bit rate, frame rate, and other error-resilience parameters, and transmits the video to the UMTS radio link emulator. At the emulator the received source data passes through the UMTS data flow model and the simulated physical link layer, and is finally transmitted to the mobile client for display. Here, a PC-based MPEG-4 decoder is used to emulate the display capabilities of the mobile terminal. The UMTS configuration options dialog box (Figure 8.55) is designed for interactive radio bearer configuration for a particular connection. The QoS parameter page shows the user- requested quality of service parameters, such as type of service, traffic class, data rates, residual bit error ratio, and transfer delay. In addition, operator control parameters, connection type, PDCP connection type, and the number of multiplexed transport channels are shown. The transport channel parameter page for the data channel shows the transport channel- related network parameter settings (Figure 8.56). Logical channel type, RLC mode, MAC channel type, MAC multiplexing, layer 1 parameters, TTI, channel coding scheme, and CRC are user-definable emulator parameters, while TB size and rate matching ratio are calculated and displayed from the other input parameter values. If the number of multiplexed transport channels is set to 1 then the transport channel parameter page for the control channel is disabled. Otherwise it shows the transport channel parameters that are related to the control channel. The appropriate spreading factor for transmission is calculated based on the requested QoS parameters and is displayed on the physical/radio channel parameter page (Figure 8.57). Radio channel-related settings (carrier frequency, channel environment, mobile speed) and receiver

Wireless Channel Models 313 Figure 8.54 UMTS emulator architecture characteristics (number of rake fingers, rake combining, diversity techniques, power control) are selected on the physical/radio parameter page. Figure 8.58 illustrates the user interfaces of the designed emulator. In addition to the radio bearer configuration parameter pages described so far, the emulator also displays the instantaneous performance in terms of Eb/No, Eb/Io, C/I, and BER. Furthermore, it allows interactive manipulation of the number of users in the cell (hence co-channel interference), and monitoring of the video performance in a more realistic operating environment. 8.3.8 Conclusion The design and evaluation of a UMTS-FDD simulator for the forward link has been described. The CCSR model gives performances that satisfy the requirements shown in 3GPP perfor- mance figures. Furthermore, the performance of the CCSR model closely follows the performance traces published by different terminal manufactures on most test configurations. However, some performance variation was visible for operation over the case 2 propagation environment and 144 kbps reference channel over the case 3 test environment. As mentioned earlier, there are several things that could contribute to this. The most likely is the different implementation strategies followed in the receiver design and in channel decoding. Another

314 Visual Media Coding and Transmission Figure 8.55 QoS parameter option page Figure 8.56 Transport channel parameter option page

Wireless Channel Models 315 Figure 8.57 Physical/radio channel parameter option page possible contributor is the different simulation techniques used for propagation modeling and interference modeling. The differences seen in the performance of turbo codes are greater than in the performance of convolutional codes, where the coding/decoding technology is fairly stable and consolidated. In addition, the performance of the LogMap algorithm implementa- tion (provided in the SPW package) is highly sensitive to the amplitude of the decoder input signal. The input amplitude setting was based on the conducted experimental results and may cause slight performance degradation. Even though the CCSR model matches the reference performance traces, or comes very close to them under the particular bearer configurations tested, the bit error sequences generated by the CCSR model should be considered, to a certain extent, as worst-case performance figures. The designed quasi-ideal rake receiver employs non-ideal channel estimation based on weighted multislot averaging (WMSA) techniques. The settings of the WSMA filter parameters might not be considered optimal in varying simulation environments. Also, the employment of advanced power control techniques can result in improved performance over the less complex fast power control algorithm implemented. In fact, the CCSR model shows about 3 dB performance loss over the results published by Olmos [51] using a non-ideal rake receiver and fast power control. The intention of the UMTS physical layer simulation model was to facilitate investigation of multimedia performance over the UMTS air interface. Although the performance test results were presented as block error ratio values, the output of the physical link simulator produces bit error sequences to characterize the actual physical link layer. After integrating with the UMTS

316 Visual Media Coding and Transmission Figure 8.58 Real time UTRAN emulator protocol data flow model, the physical link layer-generated bit error patterns can be used in audiovisual transmission experiments. In addition to the high degree of correlation shown between the performance of the CCSR model and the quoted reference figures, generated bit error patterns are suitable for employment in a radio bearer optimization for multimedia communication, as they exhibit relative differences between various network parameter and interference settings despite the type of receiver architecture implemented. 8.4 WiMAX IEEE 802.16e Modeling 8.4.1 Introduction This section features a discussion of the WiMAX IEEE 802.16e system. First, the function of basic components in the WiMAX model is discussed. Second, the simulation results facilitate a study of the performance of WiMAX for different mobile speeds, channel models, and channel coding schemes. Furthermore, this section outlines the developed WIMAX software for trace generation, the error pattern collection method, and the format of the error pattern.

Wireless Channel Models 317 In Subsection 8.4.2, the WiMAX physical layer system is outlined. Subsection 8.4.3 presents the results of WIMAX physical layer performance with different settings and different channel models. Finally, Subsection 8.4.4 discusses the error pattern generation format and parameters. Basic assumptions for the error pattern generation are also presented. 8.4.2 WIMAX System Description The IEEE 802.16e-2005 standard [52] provides specification of an air interface for fixed, nomadic, and mobile broadband wireless access systems with superior throughput perfor- mance. It enables non-line-of-sight reception, and can also cope with high mobility of the receiving station. The IEEE 802.16e extension enables nomadic capabilities for laptops and other mobile devices, allowing users to benefit from metro area portability of an xDSL-like service. The standard allows the physical layer to be scalable in bandwidth ranging from 1.25 to 20 MHz with fixed subcarrier spacing, while at the same time providing advanced features such as adaptive modulation and coding (AMC), advance antenna systems (AAS), coverage enhancing safety channel, convolutional turbo coding (CTC), and low-density parity check (LDPC) code. This rich set of IEEE802.16e features allows the equipment manufacturer to pick and mix the features to provide superior throughput performance, while at the same time allowing the system to be tailored to cater for restrictions in certain countries. The WIMAX forum is currently standardizing system profiles, which encompass subsets of IEEE 802.16e features considering country restriction, and at the same time allowing interoperability between equipment from different companies. This subsection describes the implemented WIMAX baseband simulation model. The implemented features consider a broadcasting deployment scenario. The system specifications are outlined, and an overview of every component in the baseband system model is given. 8.4.2.1 System Specifications and Baseband System Model Figure 8.59 shows the block diagram of the physical layer of the IEEE.802.16e standard. The block diagram specifies the processing of data streams. The components of the system are: Encoding and Modulation Data from MAC Randomiz- FEC Encoder Bit Sub- Subchannel MIMO Tx Chain and Adaptation ation interleaving channel Allocation and Processing IFFT Mapper Layers Boosting Downlink/Uplink Demodulation and Decoding Data to MAC De FEC Decoder Bit Subchannel Channel MIMO Rx Chain and Adaptation Randomization Deinterle- Demapper Estimation Processing FFT and Adapter Layers aving Figure 8.59 Physical layer of the IEEE 802.16e standard

318 Visual Media Coding and Transmission . randomizer . FEC encoder: convolutional turbo code (CTC), repetition code, etc. . bit interleaver . data and pilot modulation . subchannel allocation: FUSC, PUSC, etc. . MIMO processing: space time coding (STC), spatial multiplexing (SM), etc. . FFT/IFFT: 2048, 1024, 512, 256 points. Channel coding procedures include randomization, FEC encoding, bit interleaving, and repetition coding. When repetition codes are used, allocation for the transmission will always include an even number of adjacent subchannels. The basic block will pass the regular coding chain where the first subchannel sets the randomization seed. The data will follow the coding chain up to the QAM (Quadrature Amplitude Modulation) mapping. The data outputted from the QAM mapper will be loaded on to the block of pre-allocated subchannels for transmis- sion. The subchannel allocation follows one of the subcarrier permutation schemes, for example FUSC or PUSC. After that, multiple-antenna signal processing is applied if available in the system, and finally the data is passed to the OFDM transceiver (IFFT block) for transmission. A subset of the features of IEEE 802.16e is implemented according to the broadcasting scenario. The functionality and requirements of each block are specified below. The C þþ software implementation of the baseband model uses the IT þþ communication signal processing library [53]. Data Randomizer Data randomization is performed on data transmitted on the downlink and uplink. The randomization is initialized on each FEC block (using the first subchannel offset) and the OFDMA symbol offset on which that block is mapped. Symbol offset, for both UL and DL, is counted from the start of the frame, where the DL preamble is count 0. If the amount of data to transmit does not exactly fit the amount of data allocated, padding of 0 Â FF (“1” only) is added to the end of the transmission block, up to the amount of data allocated. Each data byte to be transmitted enters sequentially into the randomizer, MSB first. Preambles are not randomized. The seed value is used to calculate the randomization bits, which are combined in an XOR operation with the serialized bitstream of each FEC block, as shown in Figure 8.60. The randomizer sequence is applied only to information bits. Convolutional Turbo Coding (CTC) Figure 8.61 shows the convolutional turbo coding (CTC) encoder. Incoming bits are fed alternately into A and B. During the first encoding operation, A and B are connected to position 1 of the constituent encoder to generate parity C1. In the second step, the interleaved A and B bits are connected to position 2 of the constituent encoder to generate parity C2. A, B, C1 and C2 together are the mother codeword that can be punctured to obtain different code rates for transmission. The constituent encoder uses duo-binary circular recursive systematic convolu- tional (CRSC) code, where the encoding operation is performed on a pair of bits, hence “duo- binary”. The code is circular in such a way that the ending state matches the starting state. The polynomials defining constituent encoder connections are described in hexadecimal and binary symbol notations as follows:

Wireless Channel Models 319 Initalization 0 1 1 0 1 1 1 0 0 0 1 0 1 01 Sequence MSB LSB 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Data In Data Out Figure 8.60 Data randomizer . For the feedback branch: 0 Â B; equivalently: 1 þ D þ D3. . For the Y parity bit: 0 Â D; equivalently: 1 þ D2 þ D3. . For the W parity bit: 0 Â 9; equivalently: 1 þ D3. The number of input bits depends on the number of slots allocated by the MAC layer to the user for transmission. Table 8.32 shows the number of bits per data slot. Concatenation of a number of data slots is performed in order to make larger blocks for coding whenever possible, with the limitation of not exceeding the largest block defined in Table 8.32 for a given applied modulation and coding rate. A larger coding block improves the coding performance, but introduces higher decoding and computational complexity. To decode the above duo-binary CTC codes, MaxLogMap [54] has been adopted. Figure 8.61 Convolutional turbo coding (CTC)

320 Visual Media Coding and Transmission Table 8.32 Encoding slots concatenation for different rates in CTC Modulation Number of bits Maximum number and rate per data slot of concatenated slots QPSK 1/2 48 10 QPSK 3/4 72 6 16 QAM 1/2 96 5 16 QAM 3/4 144 3 64 QAM 1/2 144 3 64 QAM 2/3 192 2 64 QAM 3/4 216 2 64 QAM 5/6 240 2 Interleaver The interleaving operation ensures that adjacent coded bits are mapped onto nonadjacent OFDM subcarriers, and at the same time maps coded bits alternately on to less and more significant bits of the modulation constellation. The intention is to increase the robustness and avoid long runs of low-reliability bits. The operation is divided into two permutation steps. The first permutation equation is: i ¼ ðNCBPS=16Þðkmod16Þ þ bk=16c k ¼ 0; . . . ; NCBPS À 1 ð8:20Þ where i is the index after the first permutation of index k. k is the index of coded bits before first permutation. NCBPS is the encoded block size. The second permutation mapping index i to j is: j ¼ s  bi=sc þ ði þ NCBPS À b16  i=NCBPScÞmod s i ¼ 0; . . . ; NCBPS À 1 ð8:21Þ where s ¼ max(NBPSC/2,1), with NBPSC being the number of bits per modulation symbol, for example two bits for QPSK. Note that the interleaver is not used for CTC. Modulation Permutation definition Figure 8.62 shows the PRBS used to produce a random sequence, wk, that will be used for subcarrier randomization (after QAM mapping) and pilot modulation in the following. The polynomial for the PRBS generator will be X11 þ X9 þ 1. The initialization vector for the memory will follow the steps in [52]. LSB MSB 1 2 3 4 5 6 7 8 9 10 11 wk Figure 8.62 PRBS for pilot modulation

Wireless Channel Models 321 Figure 8.63 QPSK, 16 QAM and 64 QAM constellations point Data modulation OFDM subcarriers are modulated using QPSK, 16-QAM, and 64-QAM constellations. The encoded and interleaved serial input data is divided into groups of NBPSC, i.e. 2, 4, or 6 bits, which will then be converted into a complex number, I þ Q Â j, representing either a QPSK, 16-QAM, or 64-QAM constellation point. The mappings for QPSK, 16-QAM and 64-QAM are shown in Figure 8.63. Finally, the resulting complex value is normalized by multiplying it by the normalization factor, KMOD, specified in Table 8.33. The constellation-mapped data is subsequently modulated on to the allocated data subcarriers, and each subcarrier is multiplied by the factor 2 Â (1/2 À wk) according to the subcarrier index k. wk is derived using the method described above for permutation definition. Pilot modulation In the downlink, the pilot is transmitted with a boosting of 2.5 dB over the average nonboosted power of each data tone. The pilot subcarriers are modulated with sequence wk, defined earlier using the following equation:  Refckg ¼ 8 1 À wk 3 2 Imfckg ¼ 0 ð8:22Þ In the downlink, for PUSC, FUSC, AMC, and optional FUSC permutations, all pilots (of the segment, in the case of PUSC) will be modulated, whether or not all the subchannels are Table 8.33 Modulation dependent normalization factor Modulation KMOD p QPSK 16 QAM 1=p2 64 QAM 1=p10 1= 42

322 Visual Media Coding and Transmission allocated in the DL-MAP. For AMC permutation in AAS zone, the BS is not required to modulate the pilots that belong to bins not allocated in the DL-MAP, or allocated as gaps (UIUC ¼ 13). Subchannel Allocation There are many types of subchannel allocation mechanism, grouped according to whether the transmission is uplink or downlink. For the downlink, the two main subchannel allocation methods are: . Partial usage of subchannels (PUSC), where some of the subchannels are allocated to the transmitter. . Full usage of subchannels (FUSC), where all of the subchannels are allocated to the transmitter. FUSC employs full channel diversity by distributing data subcarriers to subchannels using a permutation mechanism designed to minimize interference between cells. It is somewhat similar to the idea of the classical frequency hopping technique. For PUSC, subchannels are divided and assigned to three segments, which can be allocated to sectors of the same cell. As with FUSC, a permutation mechanism is applied to allocate subcarriers to subchannels to harvest the interference averaging and fast-fading averaging effects. See [52] for details of the permutation mechanism. Figure 8.64 shows the PUSC data frame, consisting of L subchannels across the time interval. Data can be transmitted over the subchannels depicted in the figure. In PUSC mode, a data slot is composed of one subchannel and two OFDMA symbols. The data region is the allocated area for user data transmission. The mapping of encoded data blocks onto the subchannels is depicted in Figure 8.64. The mapping follows the order in Figure 8.64 (vertical direction first, then proceed to the next two OFDMA time symbols). Figure 8.64 Example of mapping encoded data blocks to subchannels in downlink PUSC mode [52]

Wireless Channel Models 323 Table 8.34 FUSC FFT parameters 2048 345 System parameter Description 1703 1536 FFT size 128 512 1024 166 No. of guard subcarriers 22 86 173 No. of used subcarriers 106 426 851 No. of data subcarriers 96 384 768 No. of pilot subcarriers 9 42 83 IFFT/FFT The IFFT block transforms the data from frequency to time domain, using inverse fast Fourier transforms at the transmitter while the FFT performs the reverse operation at the receiver. A highlight of the IEEE 802.16e standard is the idea of scalable OFDMA (SOFDMA), where the FFT size can be adjusted while fixing the subcarrier frequency spacing at a particular value. This is advantageous for supporting a wide range of bandwidths in order to flexibly address the need for various spectrum allocations, ranging from 1.25 Mhz to 20 MHz. The relevant FFT parameters for FUSC and PUSC schemes are shown in Tables 8.34 and 8.35, respectively. 8.4.3 Physical Layer Simulation Results and Analysis This subsection describes the physical layer performance of the WIMAX system. The simulation parameters used are the WIMAX system parameters, which can be found in Table 8.36. Other simulation parameters and assumptions are: . ITU vehicular A and vehicular B channel models [55]. . A spectral mask of À5 dBc/Hz flat up to 10 kHz, and then reducing 20 dB/dec up to À120. . Perfect channel knowledge. Unless otherwise specified, the parameters above are assumed for the simulation results presented below. 8.4.3.1 Performance of WIMAX at Different Mobile Speeds Figure 8.65 shows the performance of PUSC schemes for mobile speeds of 60 and 100 kmph in the ITU vehicular A channel. The FEC blocks sizes are indicated in the graph legends, in the Table 8.35 PUSC FFT parameters Description System parameter 512 1024 FFT size 128 91 183 2048 No. of guard subcarriers 43 421 841 367 No. of used subcarriers 85 360 720 1681 No. of data subcarriers 72 60 120 1440 No. of pilot subcarriers 12 240

324 Visual Media Coding and Transmission Table 8.36 System parameters of the WIMAX platform System parameter Description Duplexing TDD Multiple access OFDMA Subcarrier permutation PUSC Carrier frequency 2.3 GHz Channel bandwidth 8.75 MHz FFT 1024 Subcarrier spacing 9.765 625 kHz Symbol duration, TS 102.4 ms Cyclic prefix, TG 12.8 ms OFDM duration 115.2 ms TDD frame length 5 ms No. of symbols in a frame 42 DL/UL ratio 27/15 TTG/RTG 121.2/40.4 ms range of 384 480 bits, depending on the modulation and coding setting. It can be seen that WIMAX is fairly robust to Doppler spread and the performance loss is minimal. Figure 8.66 further shows the BER performance at a mobile speed of 150 kmph. Comparing Figure 8.65 and Figure 8.66, the performance loss is still small, even when the 64-QAM 1/2 is used. Figure 8.67 further shows the result for ITU vehicular A BER and PER at 100 kmph using FEC blocks of 144 and 192 bits. When compared to Figure 8.65, there is a small performance loss. This is due to small block size being used; for the class of CTC codes, larger block sizes will have a better BER/PER performance. Nevertheless, the performance is still robust to Doppler spread, even at 100 kmph. 8.4.3.2 Performance of WIMAX with Different Channel Models Figure 8.68 shows the BER and PER performance of ITU vehicular A and vehicular B channel at 60 kmph using FEC blocks of length 384 480 bits as defined in the graphs, depending on the modulation and coding setting. It can be seen that ITU vehicular B beyond 16-QAM 3/4 model has large performance degradation compared to the ITU vehicular A channel. This is due to the larger channel echo of the ITU vehicular B model than the guard interval of the defined system parameters in Table 8.36. While the defined system parameters target smaller cell size and low antenna, the figure shows that the system is still usable (and robust for modes below 16-QAM 1/ 2) at the larger cell size of the normal GSM/3G system. Note that this problem can be solved easily, by using a larger guard interval at the cost of reduced bit rate if a larger cell size is desired. 8.4.4 Error Pattern Files Generation 8.4.4.1 Data Flow over the Physical Layer Figure 8.69 shows the physical layer time division duplex (TDD) frame of the WIMAX system. It contains downlink (DL) and uplink (UL) subframes with various regions for performing protocol functions and data transmissions. Basically, the DL data burst #x region is the

Wireless Channel Models 325 Figure 8.65 BER and PER performance of PUSC scheme at mobile speeds 60 kmph and 100 kmph Figure 8.66 BER performance of PUSC scheme at mobile speed 150 km/h

326 Visual Media Coding and Transmission Figure 8.67 BER and PER performance of the PUSC scheme at 100 kmph Figure 8.68 BER and PER performance with ITU vehicular A and vehicular B channel

Wireless Channel Models 327 Figure 8.69 WIMAX physical layer TDD frame. Reproduced by Permission of Ó2004, 2007 IEEE allocated data region where a user can transmit data. As shown in Figure 8.69, a data burst generated by users is placed at the correct DL data burst, depending on the MAC scheduler allocation decision. The allocation decision is transmitted in the DL-MAP section of the DL subframe. 8.4.4.2 Pregenerated Trace Format (Portions reprinted, with permission, from C.H. Lewis, S.T. Worrall, A.M. Kondoz, “Hybrid WIMAX and DVB-H emulator for scalable multiple descriptions video coding testing”, International Symposium on Consumer Electronics (ISCE 2007), June 20 23, 2007, Dallas, Texas, USA. Ó2007 IEEE.) The developed WIMAX simulator has been compared to the literature for validation. It has been used to generate error pattern files. The format of the trace matches the TDD frame format. The system parameters are shown in Table 8.36. As PUSCs have been used, one data slot is equal to one subchannel and two time symbols. Thus, the DL subframe is a matrix of 30 Â 13 data slots, excluding the preamble symbol. In order to reduce data storage requirements, the error pattern is saved in the form of a data- slot error pattern instead of a bit error pattern. The data-slot error pattern is obtained by comparing all the data bits within an original data slot to the transmitted data slot. If there is any bit error within the data slot, it is declared as an error. Note that we have not specifically assumed any IP packet size or data burst size within a physical layer frame. Also, no MAC layer packet encapsulation procedures have been performed. This decision was made to allow flexibility on the choice of packet size and the data throughput during video transmission simulation. Data slots can be aggregated to different MAC frame sizes for different packet sizes, and different burst sizes for different data throughput. This also allows the study of efficient packetization schemes for video transmission. Maximum FEC code block size has been assumed for the trace generation. The maximum code block size is MCS mode-dependent, as defined in the IEEE 802.16e-2005 standard.

328 Visual Media Coding and Transmission Table 8.37 Parameters used for trace generation Parameter Values Length of trace 15 s Permutation PUSC Channel coding CTC Terminal speed 60, 120 kmph Test environment ITU vehicular A MCS mode QPSK 1/2, QPSK 3/4, 16 QAM 1/2, 16 QAM 3/4, 64 QAM 1/2 SNR range 0 30 dB, which will have 5 7 data points for each MCS mode For error pattern generation, the system parameters in Table 8.36 and the simulation parameters in Table 8.37 are used. Nevertheless, error patterns for other scenarios may be generated easily with the WIMAX simulator. For each MCS mode, 5 7 data points, each representing different SNR/BER levels, are generated. In the generated error pattern files, symbol 1 refers to a data slot error while symbol 0 means that there is no error. 8.5 Conclusions Wireless communication technologies have experienced rapid growth and commercial success during the last decade. Driven by the powerful vision of being able to communicate from anywhere at any time with any type of data, the integration of multimedia and mobile technologies is currently underway. Third- and beyond-third-generation communication systems will support a wide range of communication services for mobile users from any geographical location, in a variety of formats, audiovisual services, and applications. This document has described the design of channel simulators for GPRS/EGPRS and UMTS radio access networks. The simulators can be used to investigate resource-allocation and quality- enhancement methods for the transmission of audiovisual services over heterogeneous radio access networks. The design and implementation of a physical link layer simulation model of the GPRS and EGPRS packet data channels was presented. The model was integrated with a GPRS radio interface data flow model to simulate the GERAN radio access networks. The design of the UMTS radio access network simulator was presented. A UMTS-FDD forward-link physical layer simulation model was designed using the Signal Processing WorkSystem (SPW) software simulation tools. The model includes all the radio configurations, channel coding/ decoding, modulation parameters, transmission modeling, and their corresponding data rates for a dedicated physical channel according to the UMTS specifications. The physical link simulator generated bit error patterns, which correspond to various radio bearer configurations. These error patterns were integrated with the UMTS protocol data flow model, designed in Visual C þþ . This integration was implemented in a real-time emulator so as to allow for interactive monitoring of the effects of network parameter settings upon the received multimedia quality. The developed simulators match the reference performance traces, or the performances are within the operational margins specified by the relevant standardization bodies. However, the bit error sequences generated by the simulators should be considered, to a certain extent, as

Wireless Channel Models 329 worst-case performance figures. The receivers implemented in the simulators are customer- designed and are based on less-complex non-ideal channel estimation. The parameter settings for these channel estimators and receivers might not be considered as optimal compared to the commercially-available advanced receiver architectures. The performances of developed simulators may further be improved with the use of advanced receiver architectures and receiver/transmitter diversity techniques. The intention of the physical-layer simulation models was to facilitate performance investigation of audiovisual applications over the heterogeneous wireless interface. The outputs of the physical link simulators produce bit error sequences, which reflects the characteristics of the actual physical link layer. These bit error sequences are integrated with the higher-protocol-layer data flow models to emulate the effect of end-to-end communication systems. The emulated systems allow us to investigate the end-to-end quality of audiovisual applications over a number of wireless communication systems. In addition to the high degree of correlation shown between the performance of the developed simulators and the quoted reference figures, generated bit error patterns are nevertheless suitable for employment in a radio bearer optimization for audiovisual communication, as they exhibit relative differences between various network parameter and interference settings, despite the type of receiver architecture implemented. This chapter also described the WIMAX simulation models developed. An overview of the WIMAX baseband model and its specification was provided. The implemented functions are a subset of the IEEE 802.16e standard, as not all the features are relevant or necessary for the purpose of the study. The physical link-level performance of WIMAX was presented. It can be seen that WIMAX is fairly robust to high vehicular speeds (Doppler spread) from the simulations carried out for 150 kmph. There is some performance degradation for existing system parameters when the popular GSM/3G, large cell size, ITU vehicular B channel is considered for 16-QAM 3/ 4 mode and above. This is because the defined parameters are targeting smaller cell sizes (smaller delay spread) than the ITU vehicular B channel. Nevertheless, the system is still robust for 16-QAM 1/2 mode and below. A comparison of different coding schemes, convolutional coding (CC), convolutional turbo coding (CTC), and low-density parity check (LDPC) was made. It was shown that the gain of LDPC over CTC is marginal (less than 1dB), and only when the block size is large. In an OFDMA system, where the allocated resource unit can be small, there is no huge advantage in an LDPC scheme. The implemented trace-generation software and the format of the error pattern were also described. The generated trace considers QPSK 1/2, QPSK 3/4, 16-QAM 1/2, 16-QAM 3/4, and 64-QAM 1/2 for the ITU vehicular A channel. Nevertheless, other parameters can be considered easily with the simulator. 8.6 Appendix: Eb/No and DPCH Ec/Io Calculation Eb ¼ total signal power Rb 2s2 No ¼ RC Á ch os where RC and Rb are chip rate and channel bit rate, respectively. ch os denotes the channel- over-sampling factor. Setting total signal power to 1, Eb/No becomes:

330 Visual Media Coding and Transmission Eb ¼ RC Á ch os No 2Á Rb Á s2 where ^Ior is the received power spectral density of the downlink as measured at the UE antenna connector, Ioc is the power spectral density of a band-limited white noise source as measured at the UE antenna connector, Ior is the total transmit power spectral density of the downlink at the node B antenna connector, DPCH Ec is the average energy per chip for DPCH, and transmit signal power. DPCH EC ¼ chip rate Assume no path loss. Eb ¼ transmit signal power bit rate ¼ DPCH factor Á ^Ior Á chip rate bit rate total noise power ¼ power on OCNS þ AGWN ¼ ðOCNS factor * ^Ior þ IocÞ Á chip rate Code orthoganality is assumed. total noise power ¼ No Á chip rate Á ch os DPCH factor Á ^Ior Á chip rate2 Á ch os Eb ¼ bit rate Á ðOCNS factor Á ^Ior þ IocÞ Á chip rate No Eb ¼ DPCH factor Á ^Ior Á chip rate Á ch os No bit rate Á ðOCNS factor Á ^Ior þ IocÞ Eb ¼ DPCH Ec Á chip rate Á ch os No Ior bit rate Á ðOCNS factor þ Ioc=^IorÞ as: Ioc=^Ior ) OCNS factor DPCH Ec Á chip rate Á ch os Ior Eb ¼ bit rate No Ioc ^Ior References [1] M. Nilsson, “Third generation radio access standards,” Ericsson Review, No. 3, 1999. [2] T. Nilsson, “Toward third generation mobile multimedia communication,” Ericsson Review, No. 3, 1999. [3] H. Granbohm and J. Wikulund, “GPRS: general packet radio service,” Ericsson Review, No. 2, 1999. [4] P. Stuckmann, The GSM Evolution: Mobile Packet Data Services, John Wiley & Sons, Ltd., 2003. [5] B. Sarikaya, “Packet mode in wireless networks: overview of transition to third generation,” IEEE Commu nications Magazine, Vol. 38, No. 9, pp. 164 172, Sep. 2000.