Pilot Patterns for the Primary Link in a MIMO-OFDM Two-Tier Network

Transcription

1 Pilot Patterns for the Primary Link in a MIMO-OFDM Two-Tier Network by Sara Al-Kokhon A thesis submitted in conformity with the requirements for the degree of Master of Applied Science Electrical and Computer Engineering University of Toronto Copyright by Sara Al-Kokhon 2017

2 Pilot Patterns for the Primary Link in a MIMO-OFDM Two-Tier Network Sara Al-Kokhon Master of Applied Science Electrical and Computer Engineering University of Toronto 2017 Abstract To meet the exponentially growing demand for high data rates in wireless mobile networks, high capacity cells are required. One way of increasing the cell s capacity is by the use of MIMO- OFDM two-tiered networks. These networks can also be used to efficiently connect IoT devices, of which many are expected to be stationary and deployed within a small area, to the internet. As primary links could create a bottleneck in the system, we focus on increasing the capacity of these links through achieving a more accurate channel estimate without adding overhead to the systemwhen compared to the 3GPP LTE system. We achieve this by proposing a new Reference Signal (RS) structure that focuses on reducing the effect of the AWGN on the different pilot symbol positions without reducing the amount of power or bandwidth available for data transmission, and without adding complexities to the system. We show the effectiveness of the proposed design and the impact of AWGN at pilot symbol positions on the link s capacity through mathematical analysis and simulation results. ii

3 Acknowledgements I would like to thank my supervisor Prof. Elvino Sousa for his support and guidance, and for the valuable knowledge he imparted to me. I would also like to thank him for giving me the opportunity to learn, immerse in research, and get this work done. I would also like to thank my committee members: Prof. Raviraj Adve, Prof. Dimitrios Hatzinakos and Prof. Natalie Enright Jerger for their time and valuable feedback; and all UofT professors who I took courses with for their knowledge and inspiration. Finally, I would like to thank my family for their love, motivation, encouragement, and constant support; my lab colleagues for the valuable discussions; and my friends for their continued support and encouragement. iii

4 TABLE OF CONTENTS ACKNOWLEDGEMENTS... iii LIST OF TABLES... vi LIST OF FIGURES... viii LIST OF ACRONYMS... xi 1. INTRODUCTION... 1 LITERATURE REVIEW... 4 CONTRIBUTION... 8 ORGANIZATION LTE DOWNLINK PHYSICAL CHANNELS AND PHYSICAL SIGNALS INTRODUCTION LTE DL TRANSMISSION LTE DL PHYSICAL CHANNELS LTE DL PHYSICAL SIGNALS SUMMARY LTE DL REFERENCE SIGNAL STRUCTURE INTRODUCTION COMMON RSS UE-DEDICATED RSS MBSFN-REGION DEDICATED RS SUMMARY PROPOSED REFERENCE SIGNAL STRUCTURE INTRODUCTION BASIC SL-RS STRUCTURE SUMMARY MATHEMATICAL ANALYSIS INTRODUCTION iv

5 5.1 MIMO SYSTEM MODEL CHANNEL ESTIMATION ANALYSIS POST-EQUALIZATION SINR SIMULATION RESULTS INTRODUCTION SIMULATION PARAMETERS SIMULATION RESULTS CONCLUSION CONCLUSION FUTURE WORK APPENDIX A- MATLAB CODE FLOW CHART APPENDIX B-WIM CDL POWER-DELAY PROFILE (PDP) APPENDIX C- ADDITIONAL SIMULATION RESULTS REFERENCES v

6 List of Tables TABLE 2-1: LTE TRANSMISSION BANDWIDTH CONFIGURATIONS TABLE 2-2: PHYSICAL RESOURCE BLOCK PARAMETERS TABLE 2-3: ANTENNA PORT MAPPING FOR DIFFERENT LTE RS TYPES TABLE 3-1: RS CATEGORIES, TYPES AND SEQUENCE INITIALIZATION PARAMETERS TABLE 3-2: COHERENCE BANDWIDTH SUPPORTED BY CSR APS TABLE 3-3: CSR PILOT OVERHEAD TABLE 3-4: CHANNEL PARAMETERS SUPPORTED BY AP TABLE 3-5: AP COMBINATIONS USED FOR MULTI-LAYER BEAMFORMING TABLE 3-6: CHANNEL PARAMETERS SUPPORTED BY AP7 AND AP8 RS STRUCTURES TABLE 3-7: LENGTH-2 OCC CODE USED FOR DUAL-LAYER & 4-LAYER BEAMFORMING TABLE 3-8: LENGTH-4 OCC CODE USED FOR 8-LAYER BEAMFORMING TABLE 3-9: LTE RS STRUCTURE PARAMETERS TABLE 4-1: N P OFDM FOR SISO SL-RS STRUCTURE TABLE 4-2: N P OFDM FOR 2X2, 4X4, AND 8X8 MIMO SL-RS STRUCTURES TABLE 6-1: WIM PARAMETERS TABLE 6-2: SIMULATION PARAMETERS TABLE 6-3: AVERAGE GAIN ACHIEVED BY SL-RSS (N P OFDM =2) FOR DIFFERENT SNR VALUES WHEN T c = 1 ms TABLE 6-4: AVERAGE GAIN ACHIEVED BY SL-RSS (N P OFDM =4) WHEN T c = 2 ms TABLE 6-5: AVERAGE GAIN ACHIEVED BY SL-RSS WHEN USING N P OFDM =2 FOR SNR 20 db &T c = 2 ms TABLE 6-6: AVERAGE GAIN ACHIEVED BY SL-RSS (N P OFDM =6) WHEN Tc = 3 ms TABLE 6-7: AVERAGE GAIN ACHIEVED BY SL-RSS WHEN USING N P OFDM =4 FOR SNR 20 db & T c = 3 ms TABLE 6-8: AVERAGE GAIN ACHIEVED BY SL-RSS (N P OFDM =8) WHEN T c = 4 ms TABLE 6-9: AVERAGE GAIN ACHIEVED BY SL-RSS WHEN USING N P OFDM =4 FOR SNR 20 db & T c = 4 ms TABLE 6-10: AVERAGE GAIN ACHIEVED BY SL-RSS (N P OFDM =10) WHEN T c = 5 ms vi

7 TABLE 6-11: AVERAGE GAIN ACHIEVED BY SL-RSS WHEN USING N P OFDM =4 FOR SNR 20 db & T c = 5 ms TABLE 6-12: AVERAGE GAIN ACHIEVED BY SL-RSS (N P OFDM =12) WHEN T c = 6 ms TABLE 6-13: AVERAGE GAIN ACHIEVED BY SL-RSS WHEN USING N P OFDM =4 FOR SNR 20 db & T c = 6 ms TABLE 6-14: AVERAGE GAIN ACHIEVED BY SL-RSS (N P OFDM =12) WHEN T c = 7 ms TABLE 6-15: AVERAGE GAIN ACHIEVED BY SL-RSS WHEN USING N P OFDM =4 FOR SNR 20 db & T c = 7 ms TABLE 6-16: N P OFDM FOR 2X2 MIMO TABLE 6-17: AVERAGE GAIN ACHIEVED BY 2X2 MIMO SL-RSS WHEN T c = 1 ms TABLE 6-18: AVERAGE GAIN ACHIEVED BY 2X2 MIMO SL-RSS WHEN T c = 2 ms TABLE 6-19: AVERAGE GAIN ACHIEVED BY 2X2 MIMO SL-RSS WHEN T c = 3 ms TABLE 6-20: N P OFDM FOR 4X4 MIMO TABLE 6-21: AVERAGE GAIN ACHIEVED BY 4X4 MIMO SL-RSS WHEN T c = 1ms TABLE 6-22: AVERAGE GAIN ACHIEVED BY 4X4 MIMO SL-RSS WHEN T c = 2 ms TABLE 6-23: AVERAGE GAIN ACHIEVED BY 4X4 MIMO SL-RSS WHEN T c = 3 ms TABLE 6-24: N P OFDM FOR 8X8 MIMO TABLE 6-25: AVERAGE GAIN ACHIEVED BY 8X8 MIMO SL-RSS WHEN T c = 1 ms TABLE 6-26: AVERAGE GAIN ACHIEVED BY 8X8 MIMO SL-RSS WHEN T c = 2 ms TABLE 6-27: GAIN ACHIEVED BY SISO SL-RS STRUCTURE FOR DIFFERENT T c COMPARED TO LTE CSR AND UE-SPECIFIC RSS TABLE 6-28: GAIN ACHIEVED BY 2X2 MIMO SL-RS STRUCTURE FOR DIFFERENT T c COMPARED TO LTE CSR AND UE-SPECIFIC RSS TABLE 6-29: GAIN ACHIEVED BY 4X4 MIMO SL-RS STRUCTURE FOR DIFFERENT T c COMPARED TO LTE CSR AND UE-SPECIFIC RSS TABLE 6-30: GAIN ACHIEVED BY 8X8 MIMO SL-RS STRUCTURE FOR DIFFERENT T c COMPARED TO LTE CSR AND UE-SPECIFIC RSS vii

8 List of Figures FIGURE 1-1: TWO-TIER CELL... 1 FIGURE 1-2: LTE CSR PATTERN FOR TWO ANTENNA PORTS... 5 FIGURE 1-3: RS STRUCTURE PROPOSED IN [15]... 5 FIGURE 2-1: TRANSMITTER SIDE OFDM OPERATION FIGURE 2-2: RECEIVER SIDE OFDM OPERATION FIGURE 2-3: IFFT/FFT OFDM SYSTEM IMPLEMENTATION FIGURE 2-4: RESOURCE BLOCK STRUCTURE FIGURE 2-5: LTE FRAME STRUCTURE TYPE FIGURE 2-6: LOCATION OF PSS AND SSS IN A FDD RADIO FRAME WITH NORMAL CP FIGURE 3-1: CSR REFERENCE SIGNAL STRUCTURE OF AP0-AP3 WHEN NORMAL CP IS USED FIGURE 3-2: CSI-RS STRUCTURE FOR NORMAL CP FIGURE 3-3: UE-SPECIFIC RS STRUCTURE FOR AP5 WHEN NORMAL CP IS USED FIGURE 3-4: UE-SPECIFIC RS STRUCTURE FOR AP7 & AP8 WHEN NORMAL CP AND DUAL-LAYER BEAMFORMING ARE USED FIGURE 3-5: UE-SPECIFIC RS STRUCTURE FOR AP7 AP10 WHEN NORMAL CP AND 4-LAYER BEAMFORMING ARE USED FIGURE 3-6: UE-SPECIFIC RS STRUCTURE FOR AP7 AP14 WHEN NORMAL CP AND 8-LAYER BEAMFORMING ARE USED FIGURE 3-7: MBSFN-RS STRUCTURE SHARED TRANSMISSION MODE ( F=15 KHZ & EXTENDED CP). 41 FIGURE 3-8: MBSFN-RS STRUCTURE - DEDICATED TRANSMISSION MODE ( F=7.5 KHZ & EXTENDED CP) FIGURE 4-1: BASIC SL-RS STRUCTURE WITH ONE AP USED FOR CONTROL CHANNEL TRANSMISSION FIGURE 4-2: SL-RS STRUCTURE FOR SINGLE-LAYER TRANSMISSION FIGURE 4-3: SL-RS STRUCTURE FOR 8N X 8N MIMO (N=4) - RS STRUCTURE FOR GROUP 1 TRANSMIT ANTENNAS FIGURE 4-4: SL-RS STRUCTURE FOR 8N X 8N MIMO (N=4) - RS STRUCTURE FOR GROUP 2 TRANSMIT ANTENNAS FIGURE 4-5: SL-RS STRUCTURE FOR 8N X 8N MIMO (N=4) - RS STRUCTURE FOR GROUP 3 TRANSMIT ANTENNAS viii

9 FIGURE 4-6: SL-RS STRUCTURE FOR 8N X 8N MIMO (N=4) - RS STRUCTURE FOR GROUP 4 TRANSMIT ANTENNAS FIGURE 4-7: SL-RS STRUCTURE FOR 8N X 8N MIMO, WHERE N> FIGURE 5-1: LINEAR INTERPOLATION/EXTRAPOLATION FIGURE 6-1: SISO CE-MSE FOR T c = 1 ms FIGURE 6-2: SISO EFFECTIVE THROUGHPUT FOR T c = 1 ms FIGURE 6-3: SISO CE-MSE FOR T c = 2 ms FIGURE 6-4: SISO EFFECTIVE THROUGHPUT FOR T c = 2 ms FIGURE 6-5: SISO CE-MSE FOR T c = 3 ms FIGURE 6-6: SISO EFFECTIVE THROUGHPUT FOR T c = 3 ms FIGURE 6-7: SISO CE-MSE FOR T c = 4 ms FIGURE 6-8: SISO EFFECTIVE THROUGHPUT FOR T c = 4 ms FIGURE 6-9: SISO CE-MSE FOR T c = 5 ms FIGURE 6-10: SISO EFFECTIVE THROUGHPUT FOR T c = 5 ms FIGURE 6-11: SISO CE-MSE FOR T c = 6 ms FIGURE 6-12: SISO EFFECTIVE THROUGHPUT FOR T c = 6 ms FIGURE 6-13: SISO CE-MSE FOR T c = 7 ms FIGURE 6-14: SISO EFFECTIVE THROUGHPUT FOR T C = 7 MS FIGURE 6-15: 2X2 MIMO CE-MSE FOR T C = 1 MS FIGURE 6-16: 2X2 MIMO EFFECTIVE THROUGHPUT FOR T c = 1 ms FIGURE 6-17: 2X2 MIMO CE-MSE FOR T c = 2 ms FIGURE 6-18: 2X2 MIMO EFFECTIVE THROUGHPUT FOR T c = 2 ms FIGURE 6-19: 2X2 MIMO CE-MSE FOR T c = 3 ms FIGURE 6-20: 2X2 MIMO EFFECTIVE THROUGHPUT FOR T c = 3 ms FIGURE 6-21: 4X4 MIMO CE-MSE FOR T c = 1 ms FIGURE 6-22: 4X4 MIMO EFFECTIVE THROUGHPUT FOR T c = 1 ms FIGURE 6-23: 4X4 MIMO CE-MSE FOR T c = 2 ms FIGURE 6-24: 4X4 MIMO EFFECTIVE THROUGHPUT FOR T c = 2 ms FIGURE 6-25: 4X4 MIMO CE-MSE FOR T c = 3 ms FIGURE 6-26: 4X4 MIMO EFFECTIVE THROUGHPUT FOR T c = 3 ms FIGURE 6-27: 8X8 MIMO CE-MSE FOR T c = 1 ms FIGURE 6-28: 8X8 MIMO EFFECTIVE THROUGHPUT FOR T c = 1 ms FIGURE 6-29: 8X8 MIMO CE-MSE FOR T c = 2 ms ix

10 FIGURE 6-30: 8X8 MIMO EFFECTIVE THROUGHPUT FOR T c = 2 ms LIST OF APPENDIX FIGURES FIGURE A-1: MATLAB CODE FLOW CHART FOR TRANSMITTER AND CHANNEL PARTS FIGURE A-2: MATLAB CODE FLOW CHART FOR RECEIVER PART FIGURE B-1: PDF FOR B1 SCENARIO OF THE WIM II-CDL CHANNEL MODEL FIGURE C-1: SISO CE-MSE ACHIEVED BY DIFFERENT N P OFDM FOR T c = 2 ms FIGURE C-2: SISO EFFECTIVE THROUGHPUT ACHIEVED BY DIFFERENT N P OFDM FOR T c = 2 ms FIGURE C-3: SISO CE-MSE ACHIEVED BY DIFFERENT N P OFDM FOR T c = 3 ms FIGURE C-4: SISO EFFECTIVE THROUGHPUT ACHIEVED BY DIFFERENT N P OFDM FOR T c = 3 ms FIGURE C-5: SISO CE-MSE ACHIEVED BY DIFFERENT N P OFDM FOR T c = 4 ms FIGURE C-6: SISO EFFECTIVE THROUGHPUT ACHIEVED BY DIFFERENT N P OFDM FOR T c = 4 ms FIGURE C-7: SISO CE-MSE ACHIEVED BY DIFFERENT N P OFDM FOR T c = 5 ms FIGURE C-8: SISO EFFECTIVE THROUGHPUT ACHIEVED BY DIFFERENT N P OFDM FOR T c = 5 ms FIGURE C-9: SISO CE-MSE ACHIEVED BY DIFFERENT N P OFDM FOR T c = 6 ms FIGURE C-10: SISO EFFECTIVE THROUGHPUT ACHIEVED BY DIFFERENT N P OFDM FOR T c = 6 ms FIGURE C-11: SISO EFFECTIVE THROUGHPUT ACHIEVED BY DIFFERENT N P OFDM FOR T c = 7 ms x

11 List of Acronyms AP: Antenna Port CDMA: Code Division Multiple-Access CE: Channel Estimation CP: Cyclic Prefix CSI: Channel State Information CSR: Cell-Specific Reference Signal DL: Downlink FDMA: Frequency Division Multiple-Access LTE: Long Term Evolution MBMSFN: Multicast Broadcast Single Frequency Network MIMO: Multiple Input Multiple Output OFDM: Orthogonal Frequency Division Multiplexing PDSCH: Physical Downlink Shared Channel PMCH: Physical Multi-Cast Channel PSLCH: Physical Static Link Channel RB: Resource Block RS: Reference Signal SF: Sub-Frame SL-RS: Static Link Reference Signal TDMA: Time Division Multiple-Access UE: User Equipment xi

12 Introduction To meet the exponentially growing demand for high data rates in wireless mobile networks, high capacity cells are required. One way of increasing the cell s capacity is by the use of two-tiered networks. These networks can also be used to efficiently connect IoT devices, of which many are expected to be stationary and deployed within a small area [6], to the internet. In a two-tier network, devices- not limited to cellular mobile terminals- can be connected to the cell s basestation through high capability transceivers, which we refer to as secondary nodes. These secondary nodes break the connectivity between the connected device (terminal station) and the base station (primary node) down into two links: primary link, i.e. cellular link between primary node and secondary node, and a secondary link, i.e. communication link between the secondary node and the terminal station, see Figure 1-1 below. The secondary link is not necessary a cellular link, it can be of any different type. Figure 1-1: Two-Tier Cell 1

13 As a large number of devices are to be supported by the secondary node, of which many demand high data rates, high capacity primary links are desired. In this thesis, we focus on optimizing the performance of the primary links, and assume that the links are well behaved. Wellbehaved links are links in which the channel changes slowly, or in which channel variations are predictable. For the purpose of this thesis, we assume channel remains static for at least 1 ms. One way of achieving high capacity links is by the use of MIMO-OFDM technology. In OFDM, the frequency domain channel is divided into a number of overlapping narrow-band orthogonal sub-channels (subcarriers) on which modulated symbols are transmitted. This technique leads to high spectral efficiency, robustness against frequency selective fading, and simplified channel equalization. Moreover, the use of cyclic prefix in OFDM makes it immune to Inter-Symbol Interference (ISI). In addition to OFDM, many standards such as 3GPP LTE/4G, IEEE , IEEE , DVB-T etc., make use of the MIMO technique, i.e. the use of multiple antennas at the transmitter and the receiver sides [11], to further increase the OFDM channel (link s) capacity. To fully exploit the advantages of these technologies, an accurate channel estimate is required to recover the transmitted signal at the receiver side. In OFDM based systems, one practical way of estimating the channel is by the use of pilot signals (reference signals). These signals are known at the receiver side, and are transmitted over predefined time/frequency radio resources; thus, adding overhead to the system. This amount of overhead scales up with the number of antennas used for MIMO transmission, as more channels need to be estimated [18]. In [18], it is shown that to obtain a meaningful channel estimate H C NxM, the number of channel measurements made should be at least as large as the number of unknowns, which is equal to NxM - the product of the number of antennas used at the transmitter side and the receiver side, respectively. Another important factor impacting the link s capacity through channel estimation is the position of the pilot signals across the time-frequency domains, i.e. the pilot pattern design. To capture the channel variations in both the time and the frequency domains, the pilot symbols are usually scattered across the time-frequency grid [16]. One way of positioning the pilot signal is by defining the pilot s spacing in both the time and the frequency domains with respect to the 2

14 channel s maximum delay-doppler spread (τ max & f m ), respectively [17]. However, the channel s behavior characterized by the coherence time and the coherence bandwidth is not the only factor effecting the pilot placements. In [21], it was found that in the absence of noise, the frequency spacing between pilot symbols in the frequency domain across one OFDM symbol has no impact on the channel estimation accuracy, i.e. pilot symbols can be placed anywhere on the grid and still achieve the same result. On the contrary, in the presence of noise, pilot placement becomes critical with equi-spaced and equi-powered pilots shown to be optimal in terms of channel estimation MSE [21], hence, making up the regular pilot pattern. In the literature, many researchers attempted to find the optimal spacing between the pilot symbols in both the frequency and time domains by optimizing different cost functions. In this thesis, we take another approach in designing the pilot patterns. By reviewing the channel estimation process in which the channel is first estimated at the different pilot positions and then interpolated to obtain the full channel response, we note that two types of error result from this process. One error is what we call the pilot Channel Estimation (CE) error (channel estimation error at pilot positions), and the other error is the channel interpolation error (channel estimation error at the remaining positions). Adapting the pilot spacing to the varying channel conditions and/or using different interpolation methods could result in a lower channel interpolation error, however, it has no effect on the pilot CE error. On the other hand, reducing the pilot CE error results in a lower channel interpolation error, as more accurate channel estimates are used for interpolation. In addition, in [14] it is shown that pilot spacing adjustments do not yield any performance enhancements for low SNR scenarios. The problem of enhancing the channel estimation accuracy is usually thought of in terms of power optimization or channel estimation algorithm design. However, the two approaches could add complexities to the systems. In this thesis, we tackle this problem by proposing new reference signal structures that achieves lower pilot CE error as well as lower channel interpolation error without adding overhead or complexities to the LTE system - unlike the proposed patterns in the literature, which efforts to reduce the channel interpolation error. For low SNR scenarios, the proposed pattern make use of the stationarity of the channel to enhance the channel estimation at the pilot positions. Moreover, for high SNR scenarios, we propose pilot patterns that make use of the stationarity of the link to support more than the 8 spatially multiplexed layers- the maximum 3

15 number of layers supported by the current LTE-Advanced standard- without the use of extra overhead. Literature review The authors in [15] investigated the impact of pilot patterns (for SISO and 2x2 MIMO cases) on the performance of the LTE system by proposing new pilot patterns and comparing their performance in terms of BER with the 3GPP LTE conventional patterns (LTE CSR antenna port 0 and antenna port 1 [3]). The LTE CSR pilot pattern has a diamond-shaped structure, which can be decomposed into two rectangular patterns with pilots equi-spaced in both the frequency and the time domains- with D f subcarrier spacing in the frequency domain and D t OFDM symbol spacing in the time domain. The two patterns are separated by D f /2 and D t /2 in the frequency domain and the time domain, respectively (see Figure 1-2). For antenna port 1 and antenna port 2, LTE uses FDM across the two antenna ports, i.e. both antenna ports use the same pilot pattern with a frequency shift of 1 subcarrier applied to antenna s 2 pattern. In the proposed pilot pattern designs, the authors used a diamond-shaped pattern with the same amount of overhead as the LTE pilot structure, however, they used unequal pilot spacing in the frequency domain, non-identical rectangular patterns (different pilot frequency spacing per pattern) across one antenna port, and different pilot patterns across the different antenna ports (Figure 1-3 shows one of the proposed designs). The simulation results showed that the performance of the proposed patterns for the 2x2 MIMO underperformed the conventional LTE pilot structure for block fading channels. In [14], the authors also investigated the effect of varying the pilot frequency and time spacing of the LTE CSR SISO pilot pattern on the BER of the LTE system under slow-fading channels. They found that varying the time spacing within one slot did not yield any performance enhancements, which is expected for block-fading channels. Moreover, it was found that for low SNR values frequency domain adjustments had no impact on the system performance. The results also verifies that equispacing pilot is optimal in case of high SNR values. 4

16 Figure 1-2: LTE CSR pattern for two antenna ports [15] Figure 1-3: RS structure proposed in [15] 5

17 In [36], the authors proposed a pilot pattern for SISO OFDM systems that minimizes the system s BER. The performance of the proposed pattern was compared to the conventional DVB- T pilot pattern. In the proposed design, the authors grouped the pilot symbols in the frequency domain into clusters consisting of two neighboring pilots each. They used equal spacing between the clusters in the frequency domain; however, to keep the same amount of overhead as the DVB- T pilot pattern, the authors doubled the spacing between the clusters. The proposed pattern achieved better performance only when the channel s SNR was low. At high SNR values, the larger pilot spacing yielded a less accurate channel estimate, which in turn resulted in a worse BER. This shows the tradeoff between fighting noise and combating fading [36]. The authors further recommended switching between the two patterns depending on the channel s SNR to utilize the benefits of both patterns. Constraints on pilot pattern designs for SISO OFDM systems, such as number of pilots and pilot s spacing were discussed in [21], [17] and [35]. In [21], it was found that the minimum number of pilot symbols (transmitted per one OFDM symbol) that is required for exact channel estimation in the absence of noise is equal to the maximum channel length (number of channel taps (L)). To estimate the channel s impulse response, IFFT was applied to the channel s frequency response estimates obtained at the different pilot positions, i.e. IFFT was used for interpolation. Another finding is that, in the absence of noise, the position of the pilot tones across the frequency domain does not affect the estimation accuracy; however, in the case of AWGN, pilot positioning becomes critical. The set of equally spaced pilots was found to be optimal in terms of minimizing the channel s impulse response MSE. Moreover, the authors compared the performance of two pilot structures (block type and comb type pilot structures) for time-invariant and time varying channels. In the block type structure, pilots are transmitted across all frequency resources once every pilot period (number of OFDM symbols), where in the comb type structure, pilots are transmitted across a set of frequency resources in every OFDM symbol, i.e. transmitted in all OFDM symbols, the optimal set was assumed. For time invariant channels, both structures when using the same amount of overhead yielded the same MSE, however, for time-varying channels the comb type showed to better track the variations of the channel, hence resulting in a lower channel MSE. From simulations, one can note that block type pilot structure performed slightly better in case of time invariant channels. An overview of the different pilot structures along with their preferred usage scenarios, gains and constraints are summarized in [35]. 6

18 Another approach for positioning pilots across the time-frequency grid, is by defining the pilot s spacing in both the frequency and the time domains with respect to the channel s maximum delay-doppler spread (τ max & f m ), respectively. To capture the channel s variation in both the frequency and the time domains, in [17] it is suggested that at least one pilot symbol should be transmitted every 1/(τ max ) Hz in the frequency domain (F.D) and every 1/(f m ) seconds in the time domain (T.D). However, in [35] a higher sampling rate, i.e. transmitting one pilot every 1/(4. τ max ) Hz in the F.D and every 1/(4. f m ) seconds in the T.D, is suggested for achieving a more accurate channel estimate. On the other hand, increasing the sampling rate increases the pilot density (pilot overhead). In OFDM systems, the pilot density is usually designed to accommodate devices in extreme conditions, i.e. with a maximum delay-doppler spread that exceeds the design limit, however the amount of these devices is usually rare leading to spectrum efficiency loss. To reduce the amount of pilot overhead, the authors in [17] proposed an adaptive pilot pattern in which the pilot density is adjusted with respect to the channel s conditions. The pattern initially uses a sampling rate slightly higher than the Nyquist rate for both domains (T.D & F.D). In case of high channel frequency/time variations, extra pilots referred to as data pilots are transmitted. In this paper, MMSE estimation was considered for obtaining the channel s frequency response using pilot symbol observations. The addition of extra pilots in [17] does not only add overhead, but also add complexity to the channel estimation problem, as the insertion of extra pilots in the F.D and T.D breaks the regular structure of the pilot pattern. As a solution, the authors in [17] further proposed a reducedcomplexity MMSE estimator, which uses the data pilot observations (extra pilots) and the estimated channel frequency response of some of the neighboring pilot positions, instead of using all pilot symbol observations. However, the reduced complexity resulted in loss of performance (lower channel estimation accuracy). In [30], the authors proposed a method for finding the optimal spacing between adjacent pilot symbols in the frequency domain for time-invariant channels, in addition to finding the optimal power distribution between pilot and data symbols. The authors used an approximation of the system s capacity that takes channel estimation error into account as the cost function for their optimization problem. They first derived the post equalization SINR expression under imperfect channel knowledge, after which they decomposed the expression into two functions; a power 7

19 allocation function that depends on the noise power, data symbol power, channel frequency autocorrelation, and pilot frequency spacing, and a function that depends on the actual channel frequency response matrix. Furthermore, the authors defined the system s capacity using the postequalization SINR, and approximated the expression for high and low SNR values to eliminate the dependency of their cost function (system s capacity) on the actual channel matrix. After which a combination of an optimal frequency spacing and a power offset parameter, which maximizes the cost function under two equality constraints- on the total data transmission bandwidth and the total transmit power- is found for a given SNR and channel frequency autocorrelation. For low SNR values, it was found that the number of pilot symbols can be decreased, but their power have to be significantly increased compared to the power of the data symbols [30]. However, for high SNR, smaller pilot spacing with smaller power offset is required to achieve optimal performance. In [23], the authors proposed an adaptive pilot pattern, in which the pilot s time spacing of the LTE release 8 Cell-Specific RS pattern is varied with respect to the channel s coherence time. The performance of the proposed scheme was compared to the regular LTE pilot pattern. For slowly varying channels, the proposed scheme used less pilots in the time domain, and hence achieved higher throughput. However, in terms of BER, the proposed scheme did not provide any performance enhancements compared to the regular LTE pattern. Contribution In this thesis, we address the following. Study and analyze the different LTE/LTE-Advanced pilot structure designs. Propose new pilot patterns for well-behaved primary links, which in low SNR scenarios focus on enhancing the channel estimation accuracy, and in high SNR scenarios support more than 8 spatially multiplexed layers. For SISO and MIMO (up to 8 spatially multiplexed layers), the amount of overhead used in the proposed pilot pattern designs is upper bounded by the LTE s pilot overhead. For pilot patterns supporting more than 8-spatially multiplexed layers, the amount of overhead is upper bounded by the LTE-Advanced 8x8 MIMO pilot overhead. 8

20 For low SNR scenarios, the proposed patterns enhance the link capacity by achieving a more accurate channel estimate, and by using less amount of overhead compared to the conventional LTE pilot patterns when the channel permits, i.e. when the channel is static for more than a threshold value. For high SNR values, the proposed designs achieve capacity enhancements by supporting more than 8 spatially multiplexed layers using the same amount of overhead as the LTE- Advanced 8-layer pattern. This is in addition to using less overhead compared to the LTE/LTE-Advanced patterns (in case of single layer and multi-layer - up to 8 layerstransmissions) when the channel permits, i.e. when the channel is static for more than a threshold value. Effectiveness of the proposed design is validated through mathematical analysis and simulation results. A comparison of the simulated channel estimation MSE and effective-throughput between the proposed patterns and the LTE/LTE-Advanced patterns is made to show the performance enhancements achieved by the proposed patterns. Organization The remainder of this thesis is organized as follows: In chapter 2, we present an overview of the different LTE/LTE-Advanced downlink physical channels and physical signals. In chapter 3, we present and analyze the different LTE/LTE-Advanced reference signal structures. In chapter 4, we present the proposed pilot structure designs with the mathematical analysis presented in chapter 5 and the simulation results presented in chapter 6. Finally, we conclude our thesis, and give directions for future work in the conclusions and future work section, respectively. 9

21 LTE Downlink Physical Channels and Physical Signals 2.0 Introduction In this chapter, an overview of the LTE Downlink (DL) transmission, along with the different LTE DL physical channels and signals is presented. 2.1 LTE DL Transmission In LTE, Orthogonal Frequency Division Multiple Access (OFDMA) is used for Downlink (DL) transmission. OFDMA is a multiuser version of the OFDM transmission scheme. In OFDM, the available spectrum is divided into a number of narrowband parallel (orthogonal) channels referred to as subcarriers - on which modulated data symbols are transmitted at a reduced signaling rate. As shown in Figure 2-1, each subcarrier is modulated by a data symbol, after which the modulated subcarriers are added to form an OFDM signal. The resulting OFDM signal experiences a high Peak-to-Average Power Ratio (PAPR) making it not suitable for uplink transmissions. OFDM also make use of a Cyclic Prefix (CP), which is formed by copying a portion of the end of an OFDM symbol and adding it to the beginning of the symbol to avoid Inter-OFDM Symbol Interference (ISI) caused by the multipath nature of the channel. The cyclic prefix length is generally chosen to accommodate the maximum delay spread of the wireless channel [20]. 10

22 Figure 2-1: Transmitter Side OFDM Operation 1 [20] Figure 2-2: Receiver Side OFDM Operation 2 [20] At the receiver side (Figure 2-2), the CP part of the OFDM symbol is first removed, after which the inner product of the remaining part of the signal and each of the different subcarriers is taken to obtain the received symbol on each of the different subcarriers. Note that OFDM operations are usually implemented using an IFFT, and FFT operations, respectively, as shown in Figure CP addition is not shown in Figure CP removal is not shown in Figure

23 Figure 2-3: IFFT/FFT OFDM system implementation [20] To reduce the inter-subcarrier interference (ICI) effect caused by the channel s Doppler spread, the subcarrier spacing should be greater than the maximum Doppler spread that the system is designed to support. The maximum Doppler spread (f d ) is a function of the terminal s speed (v) and the channel s carrier frequency (f c ), i.e. f d = v f c c. On the other hand, to avoid ISI, the subcarrier spacing should not be too large. This is because the OFDM symbol duration (T u ) is inversely proportional to the subcarrier spacing ( f), i.e. T u = 1, and the length of the CP should be at least as large as the channel s maximum delay spread to avoid ISI. At the same time, for the system to be efficient, the ratio of the length of CP to symbol duration should not be too large. This shows the tradeoff between the subcarrier spacing and the ratio of the CP length to OFDM symbol duration that should be made when designing OFDM systems. In LTE, two subcarrier spacing values are defined, 15 KHz and 7.5 KHz. The 15 KHz spacing is used by all Physical Downlink channels except the Physical Multicast Channel (PMCH) when used in dedicated mode, i.e. when used on a dedicated carrier. The PMCH transmission modes are further explained below (section 2.2). As the PMCH is used for transmission of Enhanced Multimedia Broadcast Multicast Services (embms) via Single Frequency Network 12 Δf

24 (SFN), and as the signal received from multiple cells would need to appear as a single cell transmission, a long CP is needed to cover the time difference between the cells. To accommodate the different transmission requirements, LTE defines two CP types normal CP, and extended CP. Moreover, to keep the ratio of the CP length to OFDM symbol length reasonable in case of embms transmissions, the PMCH can use a smaller subcarrier spacing of 7.5 KHz, when defined on a dedicated subcarrier. The values of the different CP types are shown in Table 2-2 below. The number of subcarriers per OFDM symbol is a function of the frequency subcarrier spacing, and the transmission bandwidth. In LTE, six different transmission bandwidths ranging from 1.4 MHz to 20 MHz are supported. These are shown in Table 2-1. Each transmission bandwidth is made up of a number of Resource Blocks (RB) with a bandwidth of 180 KHz each. In the frequency domain, a Resource Block (RB) is composed of 12 subcarriers when a 15 KHz spacing is used, and 24 subcarriers when a 7.5 KHz spacing is used. In the time domain, a RB is composed of one time slot (0.5 ms duration), which consists of 3, 6 or 7 OFDM symbols depending on the frequency subcarrier spacing and the type of the cyclic prefix (CP) used. One subcarrier per OFDM symbol is known as a Resource Element (RE). The different RB configurations are shown in Table 2-1 below; a RB representation when normal CP and a 15 KHz subcarrier spacing are used is shown in Figure 2-4 below. In this thesis, normal CP and 15 KHz subcarrier spacing are considered. Table 2-1: LTE Transmission Bandwidth Configurations [2] Transmission Bandwidth (MHz) Number of Resource Blocks (N RB )

25 Table 2-2: Physical resource block parameters [1] Configuration Number of Subcarrier per RB RB N sc Number of OFDM symbols per time slot N DL symb Normal cyclic prefix 5.2 µsec for first symbol 4.7 µsec for the rest f 15 khz 12 7 Extended cyclic prefix 16.7 µsec for all symbols f 15 khz 6 f 7.5 khz 24 3 In LTE, a minimum of two physical resource blocks that are transmitted over two consecutive time slots are assigned to each UE. This pair of resource blocks can either be composed of the same physical resource blocks (i.e. same frequency components) or of different physical resource blocks, hence applying frequency hopping between the different resource blocks over the two consecutive time slots. These two configurations are defined by the localized and distributed virtual physical resource blocks, respectively [1]. The virtual resource block is of the same size as the physical resource block, assigned in pairs and its type defines the way its resource block-pair is formed (i.e. defines the mapping between the virtual RBs and the physical RBs) [1]. Note that a localized virtual resource block pair is equivalent to a physical resource block-pair. 14

26 One downlink slot Tslot N DL symb OFDM symbols k N DL RB N RB sc 1 Resource block N N DL symb RB sc resource elements subcarrier s RB sc DL N RB N subcarrier s RB N sc Resource element ( k, l) DL l 0 l N symb 1 k 0 Figure 2-4: Resource block structure [1] LTE Radio Frame Structure In LTE, transmissions are organized into radio frames. A radio frame is 10 ms long, and it is composed of ten 1 ms subframes. Each subframe consists of two slots (0.5 ms each) with a number of resource blocks per each slot. The way these subframes are used, e.g. for uplink transmission, downlink transmission, etc., is defined by the frame structure type. In 3GPP TS [1], three different frame structure types are defined, i.e. Type 1, Type 2, and Type 3. In early LTE releases (prior to release 13) only frame structure Type 1 and Type 2 were defined. These two frame structures are defined for transmissions within the LTE Licensed band - 15

27 for FDD and TDD transmission modes, respectively, whereas frame structure Type 3, which was first introduced in release 13, is defined for Licensed-Assisted Access (LAA), i.e. aggregated LTE transmissions in licensed and unlicensed bands. In this thesis, only licensed band transmission is considered. For LTE licensed band transmissions, i.e. transmissions in the licensed spectrum, two different transmission modes are supported: Frequency Division Duplex (FDD) and Time Division Duplex (TDD). In FDD, a paired spectrum (two separate frequency bands) is used for uplink and downlink transmissions, i.e. one for uplink and one for downlink, where in TDD, a single band is shared between both uplink and downlink transmissions, i.e. uplink and downlink transmissions use the same frequency but are time multiplexed. LTE frame structure type 1 is used for LTE- FDD, and LTE frame structure type 2 is used for LTE-TDD. In LTE-FDD, two 10 ms radio frames of structure type 1 are used, one for uplink and one for downlink. In frame structure type 1, all subframes within a radio frame are used for either uplink or downlink transmissions. Frame structure type 1 is shown in Figure 2-5 below. On the other hand, for LTE-TDD, a one 10 ms radio frame of type 2 is shared between uplink and downlink transmissions. For a type 2 radio frame, subframes within one radio frame are configured as downlink, uplink and special subframes. Frame structure type 2 along with its subframe configuration can be found in [4]. In this thesis, LTE-FDD is considered. One slot, T slot = 15360T s = 0.5 ms One radio frame, T f = T s = 10 ms #0 #1 #2 #3 #18 #19 One subframe Figure 2-5: LTE Frame Structure Type 1 [4] 16

28 2.2 LTE DL Physical Channels Downlink physical channels are used to transport data over predefine resource elements. In LTE, the DL physical channels can be categorized into two groups: Control Channels Data Channels layers. : transport control data that is generated at the physical layer. : transport user-plane and control-plane data that originate at higher The control data that is transported by the physical control channels is used to support transmissions on the physical data channels. The different LTE physical data, and physical control channels are listed and described briefly in the following subsections below. For more information, please refer to [1] and [12] Control Channels Control channels are transmitted in a predefined region, which occupies the first N OFDM symbols of every subframe; N {1, 2, 3, 4} for non-mbms DL subframes, and N {1, 2} for MBMS DL subframes. Following are the three different control channels defined in LTE release 8 and above. Physical Control Format Indicator Channel, PCFICH: carries the Physical Control Format Indicator (PCFI) that identifies the number of OFDM symbols used for PDCCH; transmitted in the first OFDM symbol of each DL subframe. Physical Hybrid ARQ Indicator Channel, PHICH: carries the Hybrid ARQ Indicator that identifies uplink transmission acknowledgments - acknowledgements corresponding to the M th subframe s uplink transmissions are transmitted in the (M + 4) th subframe. The PHICH is transmitted in the first one or three OFDM symbols of each DL subframe depending on the number of uplink transmission acknowledgements to be transmitted. Physical Downlink Control Channel, PDCCH: carries downlink control information (DCI) that supports PDSCH transmissions of UE data, system information and paging messages [12]. The DCI includes DL resource allocations, modulation order, coding rate, uplink scheduling grants, power control commands and other information that is intended for one specific user or a group of users. The PDCCH is transmitted on the remaining 17

29 available resources within the control region (i.e. control region s resource elements that are not used by PCFICH and PHICH). In addition to the channels listed above, two special types of the PDCCH, the Enhanced Physical Downlink Control Channel (EPDCCH), and the Machine Type Communication (MTC) Physical Control Channel (MPDCCH) were defined in release 11 and release 13, respectively, to support the additional features supported by release 11 and above Data Channels These channels are transmitted in the remaining subframe s OFDM symbols, i.e. symbols that are not occupied by the control channels. Following are the three different Data channels defined in LTE release 8 and above. Physical Broadcast Channel, PBCH: used to transport system information Master Information Block (MIB) - that is essential for initial cell access and network configuration. The MIB carries the downlink system bandwidth, PHICH size and system frame number, and is transmitted periodically, with a period of 40 ms (i.e. 4 radio frames). In the frequency domain, the MIB is transmitted over the central six RBs, which is equal to the minimum bandwidth supported by LTE. This is to allow the UE to allocate the PBCH regardless of the actual system bandwidth. To allow the UE to efficiently decode the PBCH, forward error coding is applied, in addition to time diversity. That is, after applying FEC coding, the coded information is divided into four equal-sized individually self-decodable units, with each unit transmitted in one of the four consecutive radio frames. This allows the UE to perform soft combining if failed to decode the MIB from the first time. In the time domain, the PBCH is transmitted in the first four OFDM symbols of the second slot of the first subframe of each of the four consecutive frames. Physical Downlink Shared Channel, PDSCH: used to transport all user data, paging messages, and system information that is not broadcasted on the PBCH [5]; transmitted on all available resource elements (i.e. elements that are not used by PBCH or physical signals described below) in the data channel region. Physical Multicast Channel, PMCH: used to transport Multimedia Broadcast and Multicast Services (MBMS) data. The PMCH can be configured in two modes: Dedicated 18

30 mode, and Shared mode. In the dedicated mode, the PMCH is transmitted on a dedicated carrier (separate carrier), whereas in the shared mode, the PMCH is transmitted on the same carrier as the PBCH and PDSCH, but on dedicated subframes (MBMS subframes). The PMCH uses a 7.5 KHz, and a 15 KHz subcarrier spacing when transmitted in dedicated mode, and shared mode, respectively. 2.3 LTE DL Physical Signals Physical signals are signals that are generated at the physical layer, multiplexed with both the physical data channels and the physical control channels, and known at the receiver side. They are generated at both the transmitter and the receiver sides. At the receiver side, they are used to obtain different system values, such as cell ID, cyclic prefix type, frequency/time shifts, estimated channel value, etc. This is usually achieved by comparing the received physical signals with physical signals generated at the receiver side. In LTE, two different DL physical signal types are defined: Synchronization Signals (SS), and Reference Signals (RS). The SS enable the UE to perform cell synchronization, and to determine the cell ID, cyclic prefix, and duplex mode. The RS enable the UE to estimate the channel, which is required for Channel State Information (CSI) measurements and data demodulation, in addition to performing Radio Resource Measurements (RRM) required for cell selection/reselection and handover purposes. Moreover, a special RS type, the Positioning Reference Signal (PRS) enables the UE to allocate its position. A brief description of the different LTE DL physical signals is provided in the subsections below. For more information, please refer to [1], [12] and [25] Synchronization signals These signals are the first signals that the UE tries to detect when trying to camp on the cell. They are multiplexed with the PDSCH data and are used for cell synchronization, and for acquiring the cell ID, cyclic prefix, and duplex mode. In LTE, two downlink synchronization signals are defined: Primary Synchronization Signal (PSS) and Secondary Synchronization Signal (SSS). These signals are transmitted in the first slot of the 1 st and 6 th subframes of every radio frame (i.e. twice 19

31 every radio frame), and are located in the central 6 Resource blocks of the last and 2nd last OFDM symbols as shown in Figure 2-6 below. Figure 2-6: Location of PSS and SSS in a FDD radio frame with normal CP [12] The PSS is the first signal that the UE tries to acquire for cell synchronization. The PSS sequence is derived from the frequency domain Zadoff Chu (ZC) sequences of length 63 [12]. Once the PSS sequence is detected, OFDM symbol boundary, slot boundary, and subframe boundary synchronization is achieved. In addition, the physical layer identity (cell sector of values 0 to 2) is obtained. The SSS enables the UE to acquire radio frame boundary synchronization, and hence declare cell synchronization. Unlike the PSS transmission in which the same sequence is transmitted repeatedly, the SSS transmission involves the transmission of two different SSS sequence versions per radio frame to allow UEs to determine the radio frame boundary. Furthermore, the SSS acquisition allows the UE to obtain the physical cell identity, the cell s CP length and duplex mode. The SSS sequence is generated from 3 maximum-length base sequences (M-Sequences). For more information regarding PSS and SSS sequence generation, please refer to [12- Chapter 4]. After synchronization, the UE attempts to demodulate the PBCH using the channel estimate it obtains from detecting the Cell-Specific RSs (CSR) described below. 20

32 2.3.2 Reference signals In LTE, different reference signals are defined for different purposes, such as, measuring Channel State Information (CSI), demodulating data, locating the UE terminal, etc. These signals are multiplexed with the different Physical Data and Physical Control channels, and have a predefined structure. The RS structure is defined by the reference symbols positions across one RB. In LTE, different LTE RS types have different RS structures, and each RS structure defines an antenna port, which is mapped to one or more physical antennas. Table 2-3 shows the different RS types, along with the corresponding antenna ports. The different RS types defined in LTE release 8 and above are briefly described below, with the corresponding structures presented in Chapter 3. Table 2-3: Antenna port mapping for different LTE RS types. RS Type Antenna Port LTE release Cell Specific (CSR) MBSFN RS 4 8 UE-Specific (Single layer Beamforming) 5 8 Positioning 6 9 UE-Specific (Dual layer Beamforming) UE-Specific (Eight layer Beamforming) Channel State Information (CSI-RS) 7,

33 Cell-Specific RS (CSR) These are the first reference signals that the UE attempts to detect after acquiring synchronization, i.e. detecting the PSS and SSS, after which coherent demodulation of the PBCH is performed. These reference signals are transmitted in all DL subframes and across all RBs, embedded in all physical channels except the PMCH, and are used for estimating the DL channel; hence enabling coherent detection and Channel State Information Measurements (CSI). They are also used for obtaining Radio Resource Management (RRM) measurements. The CSR signals are used for data demodulation of all physical control channels, the PBCH, and the PDSCH (only when single antenna, multiple antenna transmit diversity and spatial multiplexing transmissions are used). In addition, they are used to obtain CSI measurements, such as Channel Rank indicator, Precoding matrix indicator and Channel quality indicator, which are fed back to the enodeb for scheduling and link adaptation purposes. Moreover, CSR signals are used for Radio Resource Management (RRM) measurements, such as RSRP and RSRQ, which are required for cell selection/reselection, and handover purposes. The CSR signals are defined for both normal and extended CP, and support transmissions of up to 4 spatial multiplexed layers. The CSR signals define antenna ports 0, 1, 2 and 3, and were first introduced in LTE release Multicast Broadcast Single Frequency Network (MBSFN) RS The CSR are embedded in all physical channels except the PMCH. The PMCH uses the MBSFN RSs for demodulating the MBMS data. These signals are transmitted in MBSFN subframes, and are embedded in the PMCH only. Since the transmission of embms via Single Frequency Network (SFN) requires the use of an extended CP, the MBSFN RSs are defined for extended CP only. These signals are associated with antenna port 4, and were first defined in LTE release UE-Specific RS The UE-Specific RSs are embedded in the UE s RBs, only when beamforming is used. These signals are transmitted on the PDSCH, precoded using the same beamforming weights as the user s data, and are inserted in the corresponding user s RBs only; hence, the name UE-Specific. Since 22

34 these signals are precoded in the same way as the user s data, they enable the UE to estimate the effective channel, i.e. the channel along with the precoding matrix, and hence demodulate the transmitted data without the need of knowing the employed precoding matrix. As these signals are not transmitted across the whole transmission bandwidth, and as they are precoded in the same way as the user s data, they cannot be used for CSI and/or RRM measurements. Therefore, CSR and CSI-RS are transmitted along with the UE-Specific RSs to enable CSI and RRM measurements. In this thesis, we further classify the UE-Specific RSs into two RS types/categories depending on the number of beamforming layers that they can support. The two categories are: Single-Layer UE-Specific RSs, and Multi-Layer UE-Specific RSs. Single-Layer UE-Specific RS: These reference signals are used for demodulating the PDSCH data when single layer beamforming is used (i.e. Transmission Mode (TM) 7). Similar to CSR, the UE-specific RSs are defined for normal CP and extended CP. The Single layer UE-Specific RSs are associated with antenna port 5 and were first defined in LTE rel-8. In addition to the UE-Specific RSs, CSR signals are transmitted to enable CSI measurements. Multi-layer UE-Specific RS: These RSs were defined to support multi-layer beamforming. In release 9, two new UE-specific RS structures associated with antenna port 7 and antenna port 8 were defined to support dual layer beamforming (TM 8). Furthermore, release10 defines another six UE-specific RS structures to support transmissions of up to 8 layers, i.e. 8x8 MIMO (TM 9). The rel10 UE-Specific RSs are associated with antenna ports In addition to these demodulation signals, to estimate the channel for CSI measurements, CSR signals and CSI-RSs are transmitted when TM 8 and TM 9 are used, respectively Channel State Information RS (CSI-RS) The CSI-RSs were first introduced in release 10 to support multi-layer beamforming transmissions of more than 2 layers. These signals are used for CSI-measurement, which need to be fed back to the base-station to perform beamforming. For LTE releases prior to release 10, the CSRs are used for CSI-measurements. However, since CSR can only support transmissions of up to 4 spatial multiplexed layers, the CSI-RSs were introduced to support 8 layer beamforming, i.e. 8x8 MIMO 23

35 (TM 9). These signals are associated with antenna ports 15-22, defined for normal and extended CP, and are transmitted periodically across the whole transmission bandwidth in subframes where PBCH, PSS and SSS are not transmitted. These signals are embedded in the PDSCH only. For CSI-RS subframe configuration, please see [4] Positioning RS (PRS) The PRS signals were introduced in LTE release 9 for estimating the UE location within a network of enodebs (base-stations). The UE uses these signals for measuring the relative timing difference between the serving cell s PRS time of arrival and the neighboring cells PRS time of arrival, i.e. the Reference Signal Time Difference (RSTD), and feeds back the different measured RSTD corresponding to different Serving-neighbor cells pair to its serving cell for UE location calculations. However, the CSR signals can be used for measuring the relative timing difference between the serving cell and neighboring cells, but for better hearability (hearing ability) of the neighboring cells the PRSs were introduced. The PRS signals are multiplexed with both PDSCH and PMCH data, transmitted in pre-defined subframes over pre-defined resource blocks, defined for both normal and extended CP, and are associated with antenna port 6. These signals are not transmitted in subframes where PBCH, PSS and SSS are not transmitted. For more information, please refer to [9] and [19]. 2.4 Summary In LTE, transmissions are arranged in subframes (1 ms each). Within one subframe, different physical data channels, physical control channels, and physical signals are transmitted. To decode the physical data and the physical control channels, detection of physical signals is required. Physical Synchronization signals are the first signals that the UE tries to detect during cell camping, after which cell synchronization is achieved and basic cell information is obtained. After detecting these signals, the UE detects the cell s reference signals- CSR, from which an estimate of the downlink channel is obtained. This channel estimate is further used for coherent demodulation of the physical data and physical control channels. This highlights the importance of physical signal transmissions within the LTE system. 24

36 LTE DL Reference Signal Structure 3.0 Introduction In this chapter, we present the reference signal structure associated with each of the different LTE DL RS types. As mentioned earlier, the different LTE DL RSs are associated with different antenna ports, and are introduced in different releases to support different transmission modes. Moreover, the transmission of these RSs can either be intended for all users within a cell, a specific user within a cell, or a group of users within the area of multiple cells. Therefore, the different LTE DL RSs can be classified into 3 categories 3 : Common, UE-dedicated, and MBSFN area dedicated, respectively, with the Common type including CSR signal, CSI-RS and PSR signal; the UEdedicated type including UE-Specific RSs (i.e. Single-Layer UE-Specific RS and Multi-layer UE- Specific RS mentioned in Chapter 2); and the MBSFN-region dedicated including MBSFN RS. The RS sequence for all three different RS categories is generated using a 31-length gold sequence that is initialized differently for each category. For the Common type, the sequence generator is initialized using the physical cell identity, the slot number, and the CP type, whereas for the UE-dedicated type, the generator is initialized using the Physical cell ID, the slot number, and the UE-identity (in case of Single-layer UE-Specific RS sequence generation only). Moreover, for the MBSFN-region dedicated, the generator is initialized using the slot number, and the MBSFN area ID. The different RS categories along with the corresponding RS types and sequence initialization parameters are summarized in Table 3-1. The different RS sequences are QPSK modulated 4, precoded in the same way as the UE s data (only in the case of UE-dedicated type), and mapped to a set of pre-located REs defined over one PRB. The set of pre-located REs defines the RS structure. Each RS type is associated with one or more predefined structures, and each structure is associated with an Antenna Port (AP) mapped to one or more physical antennas. 3 Note that this type of RS classification, in addition to the Single-layer and multi-layer UE-Specific terms are only used for clarification purposes and are not 3GPP classifications or terms. 4 QPSK modulation is used due to its noise robustness, and low PAPR compared to higher order modulation schemes. 25

37 Table 3-1: RS Categories, types and sequence initialization parameters RS Category RS Types Sequence Initialization Parameters Common CSR CSI-RS PRS physical cell identity; slot number; CP type UE-dedicated Single layer UE-Specific RS Multi-layer UE-Specific RS physical cell identity; slot number; UE-identity physical cell identity; slot number MBSFN-region dedicated MBSFN RS slot number; MBSFN area ID The number of antenna ports associated with one RS type depends on the number of transmission layers that the RS type supports, i.e. if one type supports transmissions of up to 4 layers, such as the CSR, 4 different antenna ports will be associated with that RS type as shown in Table 2-3 (Chapter 2). APs of the same RS type are mapped to different physical antennas. For example, AP0-AP3 are transmitted on 4 different antennas to support 4x4 MIMO. However, APs corresponding to different RS types are sometimes mapped to the same physical antennas, such as AP0 and AP5 are mapped to the same physical antenna to support CSI/RRM measurements and data demodulation when single layer beamforming is utilized. In this case, the RS positions across the two corresponding RS structures should not overlap. This is further reflected in the RS structures described in the subsections below. 3.1 Common RSs Common RSs are known signals transmitted to all users within a cell. These signals are transmitted in predefine RBs and predefined subframes; and do not undergo precoding, i.e. they are only mapped to their predefined RE set after modulation. 26

38 In LTE, there are three different RS types that can be classified within this category: CSR, CSI-RS, and PRS. The CSR is transmitted in all subframes across the whole bandwidth, the CSI is transmitted periodically across the whole system bandwidth, and the PRS is transmitted in predefined subframes across predefined PRBs. Since the CSR signals are transmitted in all DL subframes across the whole bandwidth, the RE sets for the CSI-RS and the PRS should not intersect with the RE set corresponding to the CSR signals. The different RE sets corresponding to CSR and CSI-RS types are further explored in the following subsections below. Since PRSs are used for UE-location estimation and are not intended for channel estimation, the PRS structure is not presented in this thesis CSR Structure The CSR signals are used for demodulating both physical control and physical data channels, in addition to performing CSI and RRM measurements. These signals are defined for transmission on up to 4 antenna ports (AP0-AP3), hence supporting transmissions of up to 4 spatial multiplexed layers. Figure 3-1 shows the CSR signal structure of antenna port 0, 1, 2 and 3. For all antenna ports, the CSR signals are multiplexed in both the frequency domain and the time domain, hence utilizing a hybrid FDM/TDM approach. This hybrid FDM/TDM approach enables UE battery power saving by enabling the UE to decode the control channel that is transmitted at the beginning of each subframe in the first 2-3 OFDM symbols, and to go to microsleep mode (switch off its receiver) in case no transmission is scheduled for the UE within that subframe [20, Chapter 9]. It also enables power sharing between RS signals and data transmissions [20]. The CSR signals have a diamond shaped pattern with the RSs equi-spaced in both the frequency domain and the time domain, and are orthogonal across the different antenna ports, i.e. the RE set corresponding to one antenna port is nulled on the other for avoiding interference on pilot subcarriers, and hence enabling multilayer transmission. Diamond shaped patterns, in addition to orthogonal RS transmissions are shown to be optimal- in terms of achieving the Channel Estimation (CE) MMSE- for time-varying channels [16, 18]. 27

39 As can be seen in Figure 3-1, the CSR signal structure can be decomposed into two identical rectangular patterns (dark-colored and light-colored patterns) with RSs equi-spaced in both the frequency domain and the time domain. For both patterns, the frequency domain spacing (D f ) is equal to 6 subcarriers for all antenna ports, while the time domain spacing (D t ) is equal to 7 OFDM symbols for Antenna Port (AP) 0 and AP 1, and 14 OFDM symbols for AP 2 and AP 3. Moreover, for all antenna ports, the two patterns are separated by D f in the frequency domain 2 and by D t in the time domain [29]. 2 Figure 3-1: CSR Reference signal structure of AP0-AP3 when normal CP is used 5 5 Resource element nulling is performed only when MIMO transmission is employed. 28

40 In the frequency domain, the CSR can estimate channels with coherence bandwidths of at least 90 KHz, or 45 KHz depending on the channel s coherence time and the antenna port used. For AP 0 and AP 1, the CSR can estimate channels with coherence bandwidths of at least 90 KHz ( τ d μsec ) if the channel s coherence time is less than or equal to 4 OFDM symbols ( ms), and channels with coherence bandwidths of at least 45 KHz (τ d μsec) otherwise. Similarly, for antenna port 2 and antenna port 3, the value of the supported coherence bandwidth is reduced to 45 KHz when the channel s coherence time (T c ) is greater than 8 OFDM symbols ( 0.57 ms). Therefore, for slowly varying channels, the CSR signals can better estimate frequency selective channels with smaller coherence bandwidths, i.e. 45 KHz compared to 90 KHz, or equivalently larger delay spreads, i.e μsec compared to μsec. The different channel coherence bandwidths supported by the different CSR APs are summarized in Table 3-2 below. Table 3-2: Coherence bandwidth supported by CSR APs CSR Antenna Port (AP) Channel s Coherence Time (T c ) Supported Channel parameters Minimum Maximum delay Coherence spread (τ d ) bandwidth (B c ) AP 0, AP1 AP 2, AP3 4 OFDM symbols 90 KHz μs > 4 OFDM symbols 45 KHz μs 8 OFDM symbols 90 KHz μs > 8 OFDM symbols 45 KHz μs 29

41 In the time domain, the CSR signals can estimate channels with coherence time of at least 4 OFDM symbols ( 0.29 ms) and 8 OFDM symbols ( 0.57 ms) when AP0/AP1 and AP2/AP3 are used, respectively. Therefore, for transmissions of 4 spatial multiplexed layers, a good channel estimate is achieved when the channel has a coherence time that is equal to or greater than 8 OFDM symbols, explaining the reason why transmission on 4 antenna ports would require lower user mobility. From the RS frequency and time spacing values mentioned above, it can be deduced that 8 REs/PRB 6 are used for pilot transmission on antenna port 0 and antenna port 1, whereas only 4 REs/PRB are used for pilot transmission on antenna port 2 and antenna port 3. Furthermore, since orthogonal RS transmission is used across the different APs, the total amount of overhead per AP is increased when MIMO transmission is used. Table 3-3 shows the total amount of overhead per AP when different numbers of APs are used. It can be noted that there is a linear increase of overhead across AP0 and AP1 when two APs are used. As this linear increase might result in nulling throughput gains achieved by using more spatial layers and/or obtaining a more accurate channel estimate, a smaller amount of pilots is used across AP2 and AP3. Table 3-3: CSR pilot overhead Total amount of overhead Number of APs AP/APs used Amount (no. of REs/PRB/AP) Percent % 1 AP {AP0, AP1} {AP0, AP1, AP2, AP3} Localized virtual resource-block assignments are assumed in this thesis, i.e. 1 PRB consists of a pair of contiguous RBs across the Time Domain (T.D). 1 PRB equals 14 OFDM symbols in the T.D and 12 subcarriers in the F.D. 30

42 3.1.2 CSI-RS Structure The CSI-RS were introduced in release 10 to support spatial multiplexing of up to 8 layers, i.e. to support TM 9 (8x8 MIMO). These RSs are used for CSI measurements required for link adaptation, and are multiplexed with the PDSCH only, i.e. inserted only in the PDSCH region. As shown in Figure 3-2, these signals are neither transmitted in the sub-frame s control region (first 1-4 OFDM symbols), nor inserted in CSR-OFDM symbols (OFDM symbols containing CSR signals). Furthermore, these signals are transmitted periodically - once every 5, 10, 20, 40, or 80 subframes - in subframes where PBCH, synchronization signals, system information or paging signals are not transmitted. In other words, the CSI-RSs are transmitted in predefined subframes once every CSI-RS period. The CSI-RSs are inserted in all PRBs within their transmission subframe, however, only one RE per PRB per AP (<1 % overhead) is used for their transmission when more than one AP is used, and two REs per PRB when a single AP is used ( 1.2 % overhead). This is to minimize the effect of the CSI-RSs on rel8 and rel9 terminals - the CSI-RS puncture rel8 and rel9 terminals data as they are unaware of their use, and to avoid populating the PRB with pilot overhead since CSR and DMR are transmitted in the same PRB as CSI-RS. The CSI-RS signals are defined for transmissions on single, dual, four and eight APs-AP15, AP15-AP16, AP15-AP18, AP15-AP22, respectively. As can be seen in Figure 3-2, depending on the number of antenna ports used per transmission, the CSI-RS can have different configurations. For transmissions using AP15 and/or {AP15, AP16}, the CSI-RS can have 20 different locations (C0-C19) that can be used by 20 different neighboring cells. On the other hand, only 10 and 5 different configurations- (C0-C9) and (C0-C4) - are available when 4 APs and 8 APs are used, respectively. 31

43 Figure 3-2: CSI-RS structure for normal CP (created with reference to Table in [4]) The CSI-RSs use a hybrid CDM/FDM across the different antenna ports. CDM is used across one pair of APs, and FDM is used between the different AP pairs. For a pair of APs, the CSI-RS are spread across the time domain - over two consecutive OFDM symbols - using a length 2 orthogonal code, and are transmitted over the same frequency/time resources. The rows of the following 2x2 Walsh matrix make up the spreading codes for one pair of AP. [ ] {(1 1) code for AP i (1 1) code for AP i+1 Therefore, unlike CSR, the CSI-RS transmitted on one antenna port are not nulled on the other, and are also sparser in the frequency and time domains, hence making them more suitable for channels with large coherence bandwidth (B c 180 KHz, or equivalently τ d 5.56 µsec), and large coherence time (T c 5 ms). 32

44 3.2 UE-Dedicated RSs The UE-Dedicated RSs are used for demodulating the user s data when beamforming is used. They are precoded in the same way as the UE s data, and transmitted in UE-dedicated RBs, hence allowing the UE to demodulate its data without the need of knowing the precoding matrix used. More specifically, the precoded RSs are used to estimate the effective channel, i.e. the channel along with the applied precoding weights. Since UE-dedicated RSs are only used for data demodulation, CSR and/or CSI-RS are transmitted along with these signals- within the same PRBto allow for CSI and RRM measurements. As mentioned earlier, there are two different UE-dedicated RSs defined in LTE: Single- Layer UE-Specific RS, and Multi-Layer UE-Specific RS. The Single-Layer UE-Specific RS is used to support single layer beamforming, whereas Multi-Layer UE-Specific RS is used to support up to 8-layer beamforming. These signals are defined on different antenna ports, and were introduced in different LTE releases. The structure of each of the different UE-dedicated RSs is presented in the subsections below Single-Layer UE-Specific RS structure The Singe-Layer UE-Specific RSs were first introduced in Release 8 to support single layerbeamforming (i.e. TM 7). These signals are defined on AP 5, and are accompanied with CSR signals to allow for CSI measurements. Hence, they are not transmitted on CSR signal positions. They are also not inserted in the sub-frame s control channel region as they are only used for UE data demodulation. As can be seen in Figure 3-3, the Single-Layer UE-Specific RS has a lattice structure similar to the CSR structure, but with a pattern s frequency spacing and time spacing (D f & D t ) equal to 4 subcarriers and 3 OFDM symbols, respectively. In the frequency domain, these signals can estimate channels with coherence bandwidths of at least 60 KHz (or equivalently, τ d μsec) if the channel s coherence time (T c ) is less than or equal to 3 OFDM symbols ( ms). For channel s with larger coherence time, i.e. T c > ms, the Single-layer UE-Specific RS can estimate channels with coherence bandwidth of at least 30 KHz (or τ d μsec). In 33

45 the time domain, these signals can support channels with coherence time of 3 or more OFDM symbols. The different channel parameters supported by AP 5 are summarized in Table 3-4. Figure 3-3: UE-Specific RS Structure for AP5 when normal CP is used In terms of overhead, AP 5 uses 12 REs/PRB for RS transmission resulting in a pilot overhead of 7.14 %. However, since the CSR signal needs to be transmitted in the same PRB as the Singe-Layer UE-Specific RS, the total amount of overhead/prb increases to %. Table 3-4: Channel parameters supported by AP5 Antenna Port (AP) Channel s Coherence Time (T c ) Supported Channel parameters Minimum Maximum delay Coherence spread (τ d ) bandwidth (B c ) AP 5 3 OFDM symbols 60 KHz μs > 3 OFDM symbols 30 KHz μs 34

46 3.2.2 Multi-Layer UE-Specific RS Structure These signals are an extension to release 8 UE-Specific RSs. They are defined to support multilayer beamforming. In release 9, AP 7 and AP 8 were defined to support dual-layer beamforming (TM 8), whereas in release 10, six more antenna ports (AP 9-AP 14) were defined to support up to 8-layer beamforming (TM 9). That is, AP7 and AP8 are used to support 4-layer and 8-layer beamforming, in addition to dual-layer beamforming. Table 3-5 shows the AP combinations used to support two, four and eight layer beamforming. Table 3-5: AP combinations used for multi-layer beamforming Number of transmission layers Antenna ports Supported by 2 [7,8] Rel9 & above 4 [7,8,9,10] Rel10 & above 8 [7,8,9,10,11,12,13,14] Rel10 & above For AP7 and AP8, the same RS pattern is used across both APs. Within this pattern, 12 REs are used for RS transmission. As shown in Figure 3-4, these RSs are multiplexed in both the frequency domain and the time domain. In the frequency domain, the RS symbols are transmitted on 3-subcarriers that are 5-subcarriers apart. In the time domain, the RSs are spread across multiple OFDM symbols (2-4 OFDM symbols) using Orthogonal Cover Codes (OCC), i.e. CDM is used across the two APs to differentiate between their RS transmissions. The length of the OCC used depends on the number of transmission layers that the AP supports. For dual-layer or 4-layer beamforming, a length 2-OCC is used; however, for 8-layer beamforming a length 4-OCC is used across the ports. Since the RSs are spread across the time domain, the OCC length affects the time-spacing between the RS symbols, and hence affects the channel s coherence time that the AP can support. However, in the frequency domain, the RS subcarrier spacing is not affected by the OCC length. 35

47 When AP7and AP8 are used for dual-layer/4-layer beamforming, the RS symbols are spread across two consecutive OFDM symbols, making the total amount of RS symbols equal to 6 symbols per PRB. These RS symbols are transmitted over 3 subcarriers in the frequency domain, and two OFDM symbol pairs in the time domain. The RS subcarriers are 5-subcarrier apart, and the OFDM symbol pairs are 7-OFDM symbols apart. This frequency and time spacing allow AP7 and AP8 to support channels with a minimum coherence bandwidth of 75 KHz (or maximum delay spread of μs), and a minimum coherence time of 0.5 ms, respectively. On the other hand, when the APs are used for 8-layer beamforming, the coherence time that the APs can support is increased to 1 ms, as the RSs are spread across all available RS-OFDM symbols using a length 4- OCC codes. However, the supported coherence bandwidth remains unchanged. The channel parameter values supported by AP 7 and AP8 are summarized in Table 3-6. Table 3-6: Channel Parameters supported by AP7 and AP8 RS structures Supported Channel parameters Antenna Port (AP) Transmission layers OCC code length Minimum Coherence bandwidth Maximum Delay Spread Minimum Coherence time (B c ) (τ d ) (T c ) AP 7/AP8 2/ KHz μs 0.5 ms KHz μs 1 ms For AP9 and AP10, these APs are used along with AP 7 and AP8 to support 4-layer and 8- layer beamforming. As shown in Figure 3-5, these APs have the same RS structure as AP 7 and AP8, but with a one subcarrier frequency shift, i.e. FDM is applied across the two AP pairs - {AP7,AP8} and {AP9, AP10}. Moreover, the AP pair {AP9, AP10} uses the same OCC as {AP7, AP8}. That is, a length 2-OCC and a length-4 OCC is used across the ports to support 4-layer and 36

48 8-layer beamforming, respectively. Therefore, AP9 and AP10 support the same channel parameters as AP7 and AP8- when used for 4-layer and 8-layer beamforming. For the remaining APs, i.e. AP11-AP14, these APs are used along with AP7-AP9 to support 8-layer beamforming, and can be divided into two AP pairs, with one pair, i.e. {AP11,AP13}, using the same RS pattern as {AP7, AP8}, and the other pair (i.e. {AP12,AP14}) using the same pattern as {AP9, AP10}. Therefore, AP7-AP14 can be divided into 2-AP groups: {AP7, AP8, AP11, AP13}, and {AP9, AP10, AP12, AP14} with each group having the same RS structure. Furthermore, each group uses a set of length 4-OCCs. These OCCs used for 8-layer beamforming are shown in Table 3-8. Since AP11-AP14 use the same RS structure as AP7-AP10, they support the same channel parameters as AP7-AP10 when used for 8-layer beamforming. Since multi-layer UE-Specific RSs are only used for data demodulation, CSR and/or CSI- RS are transmitted with the UE-Specific RSs (inserted within the UE s PRB) for CSI measurements. For dual-layer beamforming, CSR AP0 and AP1 are used for CSI-measurement. Whereas for 4-layer and 8-layer beamforming, CSI-RSs are needed for CSI-measurements, and hence are transmitted across the different antenna ports. In the case of the 4-layer beamforming, the CSI-RS are used instead of the CSR signals to have less amount of overhead per PRB. In terms of overhead, for dual-layer beamforming 28 REs/PRB (16.7 %) are used for UE- Specific RS and CSR transmissions, whereas for the 4-layer beamforming and the 8-layer beamforming, 28 REs/PRB (16.7 %) and 32 REs/PRB (19.0 %) are used, respectively, for UE- Specific RS and CSI-RS transmissions (excluding CSR). 37

49 Figure 3-4: UE-Specific RS Structure for AP7 & AP8 when normal CP and dual-layer beamforming are used Figure 3-5: UE-Specific RS Structure for AP7 AP10 when normal CP and 4-layer beamforming are used 38

50 Table 3-7: Length-2 OCC code used for dual-layer & 4-layer beamforming OCC length OCC code AP Port 2 [+1 +1] AP7, AP9 [+1 1] AP8, AP10 Figure 3-6: UE-Specific RS Structure for AP7 AP14 when normal CP and 8-layer beamforming are used. 39

51 Table 3-8: Length-4 OCC code used for 8-layer beamforming OCC OCC code for Group 1 OCC code for Group2 length Group1 APs APs Group2 APs APs [ ] AP7 [ ] AP9 4 [ ] AP8 [ ] AP10 [ ] AP11 [ ] AP12 [ ] AP13 [ ] AP MBSFN-region dedicated RS Only one type of MSFN-region dedicated RS is defined in LTE, i.e. MBSFN RS. The MBSFN RS is defined in LTE release 8, and its RS structure is presented in the following subsection below MBSFN RS Structure The MBSFN RSs are used to support simulcast transmission of MBMS services, i.e. simultaneous transmission of identical data streams from multiple time synchronized cells using the same RF carrier [19]. As mentioned earlier, these signals are transmitted in the PMCH using antenna port 4, and are used for MBMS data demodulation. As the PMCH has two transmission modes: shared, i.e. PMCH uses the same carrier as the PDSCH with 15 KHz subcarrier spacing, and dedicated, i.e. PMCH transmitted on a dedicated carrier with a 7.5 KHz subcarrier spacing, two different MBSFN RS patterns are defined- one for each transmission mode. In addition, as PMCH only supports extended CP, MBSFN RSs are defined for extended CP only. Figure 3-7 and Figure 3-8 show the MBSFN-RS structure for shared and dedicated mode, respectively. Note that, for extended cyclic prefix, the number of OFDM symbols per PRB is less than 14 OFDM symbols. In case of shared transmission mode, i.e. 15 KHz subcarrier spacing, there are 12 OFDM symbols/prb. Whereas in the case of the dedicated mode, one PRB contains 40

52 6 OFDM symbols in the time domain, this is due to the use of smaller subcarrier spacing (7.5 KHz), in addition to the use of extended CP. For both modes, the MBSFN-RSs use a lattice structure similar to the CSR signals structure, but with a different RS frequency and time spacing. Similar to CSR, this structure can be decomposed into two patterns with RSs equi-spaced in both the frequency domain and the time domain. The RS spacing within one pattern is 30 KHz in the frequency domain (D f = 2 subcarriers in case of shared mode, and D f = 4 subcarriers in case of dedicated mode), and around 0.63 ms in the time domain (D t = 8 OFDM symbols in case of shared mode, and D t = 4 OFDM symbols in case of dedicated mode). The two patterns are further separated by D f in the frequency domain 2 and by D t in the time domain. 2 Therefore, both structures can support channels with a minimum coherence bandwidth of 15 KHz (or equivalently a maximum delay spread of 66.6 µsec), and a minimum coherence time of 0.33 ms. In terms of overhead, 12.5% of the REs per PRB are used by MBSFN-RSs. Figure 3-7: MBSFN-RS Structure Shared transmission mode ( f=15 KHz & extended CP) 41

53 Figure 3-8: MBSFN-RS Structure - dedicated transmission mode ( f=7.5 KHz & 3.4 Summary extended CP) Table 3-9 summarizes the characteristics of the different RS structures used to support transmissions of 1, 2, 4 and 8 spatial multiplexed layers. From the table below, it can be noted that to support higher order MIMO, channels with a large coherence time are required. It can also be noted that the maximum coherence time assumed in the design of the LTE RS structures is 1ms, which could be less than the coherence time experience by well-behaved primary links. Therefore, many of the benefits that well-behaved links provide are not exploited within the LTE RS structure designs. In the next chapter, we propose new RS structures that better utilize the channel characteristics of well-behaved links. 42

54 Table 3-9: LTE RS structure parameters Channel s parameters supported by DMR RS Overhead (OH) Tx mode Demodulation RS (DMR) Tx layers CSI- Measurement (CSI-Meas.) RS Min. T c (ms) Min. B c (KHz) DMR REs/ % PRB Total CSI-Meas. RS RS OH REs/ % % PRB = CSR > = CSR 2 CSR > = CSR > = Unicast 1 CSR > UE-Specific RS 2 CSR CSI-RS CSI-RS Broadcast MBSFN-RS (shared) = > MBSFN-RS (dedicated) = > , 8 it is assumed that CSR signals are transmitted on one AP only when 4-layer and 8-layer UE-Specific RSs are used. 43

55 Proposed Reference Signal Structure In this chapter, we present the proposed RS structures for well-behaved MIMO-OFDM links. The performance of the proposed structure is evaluated through mathematical analysis provided in Chapter 5, and simulation results presented in Chapter Introduction In this thesis we propose a new Reference Signal (RS) structure for well-behaved links- links in which the channel remains stationary for a long period of time, or in which channel variations are predictable. We assume the channel is stationary for at least 1 subframe, i.e. the channel has a coherence time (T c ) of at least 1 ms. Similar to the Physical Multi-Cast Channel (PMCH), we suggest the use of a new physical downlink channel for data transmission over the aforementioned links, we name this channel Physical Static Link Channel (PSLCH) and the proposed RSs as Static Link RSs (SL-RS). The PSLCH can be transmitted along with PDSCH using TDMA, i.e. using shared transmission mode, or on a dedicated carrier, i.e. using dedicated transmission mode. For both modes, SL-RSs are used across the PSLCH region; however, different RS types are used across the control channel region. The basic SL-RS structure for both modes when one Antenna Aport (AP) is used for control channel transmission is shown in Figure 4-1 below. It is assumed that the first two OFDM symbols of each subframe are used for control channel transmissions, i.e. 12 OFDM symbols per PRB are available for PSLCH transmission. For the shared transmission mode, LTE Cell-Specific Reference (CSR) signals are transmitted in the control region to allow all User Equipment (UEs) within the cell (stationary & non-stationary stations) to decode the control channels. Whereas for the dedicated transmission mode, SL-RSs are transmitted in both the control region and the PSLCH region, this is to allow users to further enhance their channel estimate using the control region RSs. For both transmission modes, the control region s reference signals are used for CSI in addition to control channel demodulation, and are transmitted in all downlink subframes across the entire system bandwidth, 44

56 whereas the data region s SL-RSs are used for PSLCH demodulation and are transmitted periodically depending on the number of subframes at which the terminal is assumed to be stationary. Since the PSLCH SL-RS periodicity depends on the terminal s stationarity and this could vary from one terminal to another, these RSs can be assumed to be UE-dedicated. (a) Shared mode (b) Dedicated mode Figure 4-1: Basic SL-RS Structure with one AP used for control channel transmission- (a) Shared mode, (b) Dedicated mode 45

57 4.1 Basic SL-RS Structure Since SL-RSs are used for enabling coherent demodulation of the PSLCH, their basic structure is designed to allow a more accurate channel estimate to be achieved without adding extra costs to the system in terms of overhead and/or demodulation delay when compared to the LTE system. As shown in Figure 4-2, the basic SL-RS structure is defined by three parameters: D f (Pilot frequency Spacing), T p (Pilot time periodicity) and N POFDM (Pilot OFDM symbols). The reason for having each of the different parameters is further explained below. In Pilot-assisted channel estimation methods, channel estimates usually suffer from two different types of errors: pilot channel estimation error (channel estimation error at pilot positions), and channel interpolation error (channel estimation error at the remaining RE positions). Pilot channel estimation error results from the AWGN effect of the channel; hence, it is not a function of the pilot symbol position. In addition, this error propagates through the interpolation process resulting in a greater interpolation error. For these two reasons, the SL-RS structure focuses on reducing the pilot channel estimation error. This is achieved by transmitting the RSs over N POFDM consecutive OFDM symbols, and using the average pilot channel estimate across each frequency resource element for interpolation in case of SISO transmission, and by using orthogonal spreading in case of MIMO transmission, i.e. RSs are spread over the N POFDM symbols using orthogonal cover codes. To avoid using more overhead compared to the LTE RS structures, the SL-RSs are transmitted periodically once every T p subframes, where T p equals the user s channel coherence time (T c ). In LTE, the channel is estimated every 1 ms; hence, to not add any extra data demodulation delays, N POFDM is upper bounded by the number of OFDM symbols available for PSLCH transmission, which is equal to 12 symbols in this thesis. Since equi-spaced RSs are shown to be optimal in terms of achieving channel estimation MMSE [18], SL-RSs are equi-spaced in the frequency domain with a frequency spacing (D f ) of 4 subcarriers, which is equal to the minimum subcarrier spacing used by the LTE unicast RS structures. D f is set to equal 4 subcarriers to allow for a minimum of 2 N POFDM symbols to be used without encountering any extra overhead. This is also equal to the minimum subcarrier spacing 46

58 used by the LTE unicast RS structures. Since D f is pre-set to 4 subcarriers, given T p, N POFDM for the different transmission modes can be set as explained in the following subsections below SL-RS for Single-Layer Transmission (SISO) For SISO transmission, the SL-RS uses the basic structure as shown in Figure 4-2 below. To find the maximum N POFDM that can be used without adding extra overhead to the system, we upper bound the total amount of SL-RSs by the minimum amount of overhead used by the LTE RS structures. In LTE, CSR (AP 0) and UE-Specific RS (AP 5) are used for single-layer transmission. Since AP 0 uses less overhead than AP 5, we upper bound the total amount of SL-RSs by the number of CSR (AP 0) signals transmitted in the PSLCH region, which is equal to 6 REs/PRB/SF. Figure 4-2: SL-RS Structure for Single-Layer Transmission For a given T c, we calculate N POFDM as follows: We denote the total amount of overhead used for RS transmission during T c subframes by 1 AP0 OH for the LTE CSR signal structure, and by SLRS OH for the proposed SISO SL-RS structure. AP0 OH = 6 x T c (4.1) 47

59 1 SLRS OH = 3 x N POFDM (4.2) 1 We upper bound SLRS OH by AP0 OH to get: N POFDM 2 x T c (4.3) Since N POFDM cannot be larger than 12 OFDM symbols, N POFDM is upper bounded by min{2t c, 12}, i.e. N POFDM min{2t c, 12}, for T c 1 SF. To achieve maximum channel estimation enhancements, we set N POFDM to its upper limit as follows: N POFDM = min{2t c, 12}, for T c 1 SF (4.4) From simulation results, it was found that for high SNR values (SNR 20 db), using N POFDM = 2 OFDM symbols for T c =2 SFs, and N POFDM = 4 for T c >2 SFs yield better results compared to using a larger N POFDM, as it combines gains achieved from using less overhead and obtaining a more accurate channel estimate. Therefore, for high SNR values, we set N POFDM to equal min{t c ( 2 T c + 1), 4 } for T c 1 SF. Table 4.1 and Equation 4.5 show the value of N POFDM for high and low SNR cases. N POFDM = { min{2t c, 12} ; SNR < 20 db min{t c ( 2 T c + 1), 4 }; SNR 20 db ; for T c 1 SF (4.5) From Table 4-1 below, one can note that the proposed structure uses less overhead compared to AP0 when T c 2 SFs in case of high SNR, and when T c > 6 SFs in case of low SNR. 48

60 Table 4-1: N POFDM for SISO SL-RS Structure N POFDM (OFDM symbols) T c (SFs) Low/moderate SNR SNR < 20 db High SNR SNR 20 db 1 T c < 3 2T c 2 3 T c 6 2 T c 4 T c > SL-RS for Multi-Layer transmission (2x2, 4x4, 8x8 MIMO) For spatial multiplexing of up to 8-layers, i.e. 2x2, 4x4, and 8x8 MIMO, the SL-RS uses Code Division Multiplexing (CDM) across the different antenna ports (APs). The same RS structure proposed for SISO is used across the different APs, i.e. at each AP, RSs are equi-spaced with a 4- subcarrier spacing in the frequency domain and transmitted over consecutive N POFDM OFDM symbols once every T p in the time domain. However, to differentiate between the APs transmissions, RSs at different APs are spread over the N POFDM OFDM symbols using orthogonal cover codes (OCC). Each AP uses one row of an N POFDM x N POFDM Hadamard matrix, where N POFDM N l and N l equals the number of spatial layers, i.e. 2, 4 or 8. In this thesis, we set N l = N t. Similar to the SISO case, N POFDM is upper bounded by min { min (OH LTE) T c, 12}, for T 3 c 1 SF, where min (OH LTE ) represents the minimum amount of overhead used by LTE multi-layer RSs per PRB within the PSLCH region. To achieve maximum channel estimation enhancements without using extra overhead, we set N POFDM to equal its upper limit. Furthermore, since N POFDM represents the size of a Hadamard matrix, we set N POFDM for the different transmission layer configuration (N POFDM ) N lxn l as follows. 49

61 (N POFDM ) N lxn l = max{min {4 min (OH LTE) T c /4, 12}, N 3 l } ; for T c 1 SF (4.6) We now derive N POFDM for the different MIMO cases. In the following we assume a maximum of two APs are used for control channel transmissions. For 2x2 MIMO, both LTE CSR (AP 0 and AP 1) and multi-layer UE-Specific RSs (AP 7 and AP 8) use 12 REs/PRB within PSLCH region. Therefore, min (OH LTE ) = 12 and N POFDM for 2x2 MIMO (N POFDM ) 2x2 is set as follows. (N POFDM ) 2x2 = max{min{4t c, 12}, 2}; for T c 1 SF (4.7) For 4x4 MIMO, the LTE CSR (AP 0-AP 4) uses an overhead of 16 REs/PRB/PSLCH across (AP0 and AP1) and 24 REs/PRB/PSLCH across (AP3 and AP4), while multilayer UE-Specific RSs (AP7-AP 10) use an overhead of 24 REs/PRB/PSLCH across all APs. Therefore, min (OH LTE ) = 16 and (N POFDM ) 4x4 is set as follows. (N POFDM ) 4x4 = max{min {4 16 T c 3 /4, 12}, 4}; for T c 1 SF (4.8) For 8x8 MIMO, only UE-Specific RSs (AP7-AP 14) are supported in LTE. These RSs use 24 REs/PRB/PSLCH. Therefore, min (OH LTE ) = 24 and (N POFDM ) 8x8 is set as follows. (N POFDM ) 8x8 = max{min{8t c, 12}, 8}; for T c 1 SF (4.9) The different N POFDM values for 2x2, 4x4 and 8x8 MIMO transmission are summarized in Table 4-2 below. It can be noted that in all MIMO cases, the proposed structure uses a longer code length compared to LTE multi-layer UE-Specific RSs. By using a longer code-length, the effect of the AWGN on pilot channel estimates is reduced, and hence a more accurate channel estimate is achieved. Since the proposed structure is upper-bounded by the minimum amount of overhead used by LTE MIMO RSs, the achieved channel estimation accuracy enhancements result in a higher link capacity. This is further shown in Chapter 5 and Chapter 6. 50

62 Table 4-2: N POFDM for 2x2, 4x4, and 8x8 MIMO SL-RS Structures Transmission mode N POFDM (OFDM symbols) T c = x 2 T c = 2 8 T c 3 12 T c = x 4 8 x 8 T c = 2 8 T c 3 12 T c = 1 8 T c SL-RS for Multi-Layer transmission (8n x 8n MIMO,n 2) As shown in the Table 4-2, the basic SL-RS structure can support up to 8-spatial multiplexed layers for T c = 1. For T c 2, the proposed RS structure increases the link capacity in two ways: Uses a longer OCC code length across the basic SL-RS structure to achieve a better channel estimate. Extends the basic structure proposed for 8x8 MIMO to supports 8xn spatial multiplexed layers, where n 2, for the same amount of overhead used by the LTE 8-layer UE- Specific RSs. To support 8n x 8n, where n 2, we propose a structure in which the structure proposed for 8x8 MIMO (when T c =1 ms) is used across each group of 8 transmit antennas, i.e. across the n antenna groups. Since the different antenna groups use the same OCC set, Frequency Division 51

63 Multiplexing (FDM) is applied across the groups to maintain orthogonality between RSs across the different APs. As shown in Figure 4-3, a maximum of 4 eight-transmit antenna groups (n=4) can have their RSs transmitted within 1 subframe, i.e. a maximum of 32 spatial multiplexed layers can be supported without adding any extra demodulation delays to the system. To support more than 32 transmit antennas, the receiver requires to wait for at least 2 ms (2 subframes) to receive all RSs, and hence demodulate the signal. The RS structure extension for n>4 is shown in Figure 4-7. For n>4, the RS structure used within subframe 1 is repeated across the different subframes. The following matrix makes up the set of orthogonal codes used by the different APs within one transmit antenna group. Each AP uses one row of the matrix C 8x8 = [ ] 52

64 Figure 4-3: SL-RS structure for 8n x 8n MIMO (n=4) - RS structure for group 1 transmit antennas Figure 4-4: SL-RS structure for 8n x 8n MIMO (n=4) - RS structure for group 2 transmit antennas 53

65 Figure 4-5: SL-RS structure for 8n x 8n MIMO (n=4) - RS structure for group 3 transmit antennas Figure 4-6: SL-RS structure for 8n x 8n MIMO (n=4) - RS structure for group 4 transmit antennas 54

66 4.2 Summary Figure 4-7: SL-RS Structure for 8n x 8n MIMO, where n>4 In this Chapter, we propose the use of a new physical downlink channel for data transmission across well-behaved primary links, which we assume is stationary for at least 1 ms. We name this Channel as PSLCH and RSs transmitted across this channel as SL- RSs. We further propose a RS structure for the SL-RSs. The proposed RS structure utilizes the stationarity of the channel to achieve a better channel estimate, and to support more than 8-spatiall multiplexed layers without adding extra overhead or demodulation delays to the system. The basic SL-RS structure, which is used for single-layer and multi-layer transmissions of up to eight layers is defined by three parameters: D f, T p and N POFDM. For multi-layer transmission of up to eight layers, CDM is used across the different APs for orthogonal RS transmission. To support n groups of eight-layers, where n 2, the SL-RS uses CDM within one antenna group, and FDM across the groups to maintain RS orthogonality between the different APs. A maximum of 32 antennas can be supported without introducing extra demodulation delays to the system. 55