MIMO for Mobile Satellite Digital. Broadcasting: From Theory to Practice

Transcription

1 MIMO for Mobile Satellite Digital 1 Broadcasting: From Theory to Practice Aaron Byman, Ari Hulkkonen, Pantelis-Daniel Arapoglou, Massimo Bertinelli, Riccardo De Gaudenzi Abstract This paper presents the detailed design and the key system performance results of a comprehensive laboratory demonstrator (test-bed) for a hybrid satellite/terrestrial S-band mobile digital broadcasting system. The physical layer is based on an enhanced version of the Digital Video Broadcasting Satellite to Hand-held (DVB-SH) standard exploiting dual polarization Multiple-Input Multiple Output (MIMO) technology. This complete digital MIMO demonstrator, the first of its kind, allows in-depth verification and optimization of the MIMO techniques applied to satellite broadcasting networks. Moreover, this demonstrator allows complementing and confirming the theoretical or simulation-based findings published so far. It is shown that dual polarization MIMO diversity is able to provide remarkable gains in terms of satellite/terrestrial transmit power reduction and/or capacity increase compared to more conventional non MIMO solutions. It is also demonstrated that the adoption of a relatively simple spatial multiplexing MIMO technique represents the best way to grasp these gains. The paper provides an extensive set of laboratory measurement results for existing stochastic satellite and hybrid MIMO channels as well as results based on S-band satellite measured dual polarization time series recently collected during a campaign sponsored by the European Space Agency (ESA). Results obtained using MIMO techniques are also compared to a dual and single polarization Single Input Single Output (SISO) DVB-SH benchmark system as well to computer simulation results. Keywords (MIMO, Dual Polarization, Mobile Satellite Digital Broadcasting, MIMO test-bed, LMS.) A. Byman and A. Hulkkonen are with Elektrobit Ltd, Tutkijantie 8 FIN Oulu, Finland first.last@elektrobit.com ( ). P.-D. Arapoglou, M. Bertinelli and R. De Gaudenzi are with the European Space Agency-ESTEC, Keplerlaan 1, 2200 AG, Noordwijk ZH, The Netherlands. pantelis-daniel.arapoglou@esa.int, Massimo.Bertinelli@esa.int, rdegaude@xrsun0.estec.esa.nl.

2 2 I. INTRODUCTION The commercial success of Sirius-XM [1] S-band digital mobile satellite broadcasting services in the United States is opening up new possibilities for satellite digital broadcasting worldwide. While the wide coverage of satellite broadcasting systems is very attractive, their power and bandwidth limitation is making the design of high capacity and quality of service networks very challenging. The satellite radio frequency (RF) power limitations are mainly related to the platform constraints in terms of available Direct Current (DC) power, power dissipation and RF power handling. As a consequence, the achievable link margins are typically not able to cope with deep shadowing or non-line-of-sight (NLOS) conditions which are experienced in representative mobile (vehicular) applications. Those conditions are therefore counteracted through powerful physical or link layer forward error correcting (FEC) schemes combined with time interleaving spanning up to 5-10 seconds [2]. For urban areas, subject to frequent and persistent NLOS link situations, the satellite coverage is typically complemented by terrestrial gap fillers termed Complementary Ground Component (CGC). A state-of-the-art standard for hybrid satellite/terrestrial digital broadcasting is represented by the ETSI DVB-SH [3] comprising an SH-A and SH-B air interface based on Orthogonal Frequency Division Multiplexing (OFDM) and Time Division Multiplexing (TDM), respectively. Details on the DVB-SH capabilities, application scenarios, typical system configurations and performance assessment results can be found in the standard guideline document [4] and in the open literature [2], [5]. State-of-the-art DVB-SH does not explicitly support dual polarization MIMO as in the more recent terrestrially driven ETSI DVB Next Generation Hand-held (NGH) standard [6]. Therefore, in the following, we will focus on comparing the DVB-SH performance to an enhanced DVB-SH physical layer capable of supporting dual polarization MIMO in a hybrid satellite/terrestrial mobile digital broadcasting network. Typically, the available mobile satellite digital broadcasting spectrum is limited to a few MHz. This constrained band needs to support the whole satellite coverage as well as the terrestrial gap fillers when they are not transmitting at the same frequency the satellite is operating. This is the case in the Sirius-XM system. To achieve robust reception in the harsh land-mobile satellite (LMS) channel conditions, the physical layer design requires the adoption of low-rate FEC scheme combined with extensive time interleaving and robust modulation formats. These

3 3 physical layer design drivers are limiting the achievable spectral efficiency. That is why MIMO technology, that has received a great deal of interest in the last decades for increasing the spectral efficiency in Third Generation Partnership (3GPP) and Long Term Evolution (LTE) as well in Wireless Local Area Networks (WLAN) standards [7], has attracted also the attention of mobile hybrid broadcasting standards. Nevertheless, the L/S-band LMS channel has very different characteristics compared to its terrestrial counterpart as line-of-sight (LOS) operation is necessary due to the satellite RF power limitations previously mentioned. This limits the amount of signal multipath diversity and, in turn, the MIMO gain potential [8]. Deployment of dual transmit satellite antennas with enough spacing to obtain sizeable signal decorrelation is unfeasible considering the required large antenna reflector size and the lack of a rich scattering environment. References [9], [10], [11], [12], [13], [14] report recent investigations on more practical ways to exploit MIMO technology for enhancing mobile satellite broadcasting power and/or spectral efficiency. Reference [9] and [10] describe why MIMO, in the polarization domain, is the most realistic option for mobile satellite broadcasting. References [11] and [12] present simulation results in a DVB-SH framework of various MIMO schemes for TDM and OFDM, respectively. Finally, [13] investigates by simulation the impact of interleaver length, cross-polarization isolation, imperfect channel estimation and non-linear amplification on MIMO performance and [14] translates the MIMO benefits from the single link level to the multibeam system level. In addition to the MIMO satellite specific literature, owing to DVB-NGH, there have been significantly efforts towards hybrid satellite/terrestrial MIMO for mobile digital broadcasting (see, for example, [15] and the references therein). The most practical MIMO solution was found to be the exploitation of dual-polarized satellite antenna, i.e. reusing the same frequency in the two right/left hand circular polarizations (RHCP/LHCP). Applying MIMO over a RHCP/LHCP antenna allows the satellite to reuse the same antenna reflector for generating both polarizations. Also, the user terminal can exploit the existing antenna s horizontal and vertical elements to simultaneously generate left and right polarized signal components. In the satellite, two distinct RF chains will be present (one per polarization). In this work, apart from the wide use of MIMO LMS channel models [16], we also exploit the dual-polarization S-band MIMO LMS channel characterization completed through an extensive ESA-funded campaign [17], [18]. Reference [16] describes a statistical MIMO LMS channel

4 4 model based on a Markov-chain approach for modeling the high dynamic range encountered in the various propagation environments. References [17], [18] describe an S-band dual polarization measurement campaign in Erlangen, Germany, over a variety of propagation environments. Despite the modest MIMO LMS channel diversity mentioned, the MIMO simulation results reported in [11]- [13] indicate that sizeable gains of 3-4 db in terms of satellite RF power for a given throughput are possible in intermediate tree shadowed (ITS) LMS conditions. This truly remarkable gain can be achieved exploiting the relatively simple Spatial Multiplexing (SM) MIMO scheme. It is remarked that more elaborate full rate-full diversity space time codes do not offer any additional performance advantage. This paper describes a comprehensive DVB-SH MIMO-enabled laboratory real-time demonstrator (test-bed), developed by a team of companies and research institutions with ESA Advanced Research in Telecommunication Systems (ARTES) R&D funding [19]. The test-bed is capable of faithfully demonstrating the real potential of dual-polarization MIMO in an hybrid satellite/terrestrial mobile digital broadcasting system. The demonstrator described in this paper represents a key complement to the standardization and analytical/simulation work cited hitherto. Moreover, it confirms through laboratory experiments the great MIMO potential in a realistic satellite system environment. Note that the end-to-end system implemented in hardware cannot be approximated by analytical means. Simulations, although useful in a first phase, are too time consuming to perform extensive system analysis and parameters optimization. Hence, reliable performance results can only be obtained through the test-bed demonstrator. Although a plethora of terrestrial MIMO test-beds have been reported in the literature (see e.g. [20]), the one presented in this paper demonstrates, for the first time, the practical feasibility of dual-polarization satellite and hybrid satellite/terrestrial MIMO. The rest of the paper is organized as follows: following this introduction, in Section II a multibeam and a single-beam satellite system reference scenario for the demonstrator is presented and key system parameters are derived. Section III illustrates the MIMO demonstrator key design elements. Section IV reports the main demonstrator experimental performance results, and compares them to simulation findings, when possible. Finally, in Section V the conclusions are drawn.

5 5 II. REFERENCE SYSTEM SCENARIO The objective of the system scenario definition is to provide a reference satellite system that renders the evaluation of the MIMO performance in a DVB-SH framework realistic, particularly for the hybrid satellite/terrestrial system tests. The proposed system targets typical S-band satellite systems and CGC with DVB-SH and MIMO on the forward link whose mission is to offer a realistic mobile broadcasting communication system that could be operational in the mid-term timeframe. A. System Features The selected reference mission aims at the provision of digital mobile broadcasting over Europe by means of a geostationary satellite complemented by a CGC to cover urban areas. The typical applications envisaged are audio/video broadcasting and software updates for mobile platforms. In principle, the broadcasting mission can be complemented by interactive return link capability for messaging services. This can be accomplished by exploiting the complementary S-band return link ETSI S-band Mobile Interactive Multimedia (S-MIM) standard [21]. In the following, we will focus on the forward link broadcasting mission. The discussion related to the possible reverse link services is extensively covered in [21]. Figure 1 shows the high-level architecture of the S-band hybrid satellite/terrestrial digital mobile broadcasting network. The digital content is uploaded to the satellite through a Ku-band feeder uplink ground station. Then, the received feeder link multiplex carriers (one per beam) are on-board transparently frequency converted, amplified and retransmitted at S-band. The S-band signal is broadcasted to the ground users through one or more satellite beams depending on the specific market constraints. In Europe, linguistic beams are preferably adopted to customize the digital content to the specific language adopted in the covered regions. This multi-beam approach allows to better focus the satellite power and the reuse of frequency among the beams when there is enough isolation (typically 10 db or more). A single wide area beam (single-beam) approach can be employed when the digital content does not need to be differentiated among service beams. This is the case of Sirius XM covering the continental United States territory. The satellite also feeds the CGC, which is typically deployed in densely populated urban areas, at Ku-band. The CGC repeaters are typically converting the Ku-band downlink CGC feeder link signal into an S-band DVB-SH terrestrial signal. In the Single Frequency Network (SFN)

6 6 Fig. 1. S-band hybrid satellite/terrestrial mobile digital broadcasting system architecture. case, the CGC operates at the same frequency as the satellite component. Instead, for Multi- Frequency Network (MFN), the CGC operates in a different frequency band. This distinction leads to different MIMO configurations. Figure 2 shows a hypothetical system with 8 linguistic beams of European coverage achievable with a large S-band antenna reflector and a three colour frequency reuse scheme (see Table I). Each beam is using both LHCP and RHCP. The latter is also true for the single-beam system. Typical parameters for the reference space segment concerning both a multi-beam and a single-beam system inspired by [4] are summarized in Table I. For the single beam case a higher power per beam can be transmitted assuming both satellite architectures can fit in a standard 12.5 kw platform. The feeder uplink is in the 14 GHz band. The 15 MHz beam band is used for one SH-A or SH-B carrier occupying 5 MHz/polarization of the 15 MHz available (3 colour scheme, see Figure 3). Each beam can simultaneously transmit over both LCHP/RCHP, thus the effective bandwidth used for each one is 10 MHz. The resulting user link frequency reuse scheme is represented in Figure 3. The ground segment parameters are very much dependent on the CGC deployment area and scenario as well as operator s choice in terms of transmitter size. For small sites an RF power of W per sector will be sufficient to cover a cell size of 3-6 km [2]. A more powerful transmitter will allow coverage over larger areas [4].

7 7 Fig. 2. S-band regional (linguistic) service areas. TABLE I SPACE SEGMENT MAIN PARAMETERS. Multi-beam Single-beam Feeder link band Ku ( GHz) User link band S (2.2 GHz) Bandwidth/beam 2 5 MHz Number of beam colours 3 1 Overall user bandwidth 15 MHz 5 MHz Number of linguistic beams 8 1 Polarization LHCP/RHCP Satellite EIRP/beam 63 dbw 70 dbw Inter-beam C/I 14 db NA B. MIMO Configurations The MIMO configurations implemented in the test-bed are depicted in Table II and more extensively discussed in [22]. For comparison fairness, in all cases the total RF power of the satellite and terrestrial components is the same compared to single polarization reference system. As mentioned before, the simulation findings [11]- [13], [22] confirmed that for both the satellite only and the hybrid dual polarization MIMO configurations SM yields the most promising results.

8 8 Fig. 3. User link frequency plan. For the satellite only configuration, also the 2 SISO (single input single output) technique 1 [23] represents an intermediate step between SISO and MIMO. For the satellite/terrestrial hybrid configurations there are distinct system limitations and channel characteristics that limit the selection of MIMO techniques implemented in the test-bed. The most important one is that true hybrid coverage, where both components are available, is limited in terms of coverage area, hence reception should be possible in the absence of any of the two components. Therefore, when dual polarization is available at both components, in addition to the SM technique also the Alamouti (ALA) Space-Frequency Block Coding (SFBC) has been retained in the hardware. The ALA MIMO scheme is particularly suited for single stream satellite and CGC combination. Also in general, MIMO can also be applied to the case of dual satellite systems exploiting spatial diversity to enhance the system availability. However, this configuration cannot be realized in the case of SFN as it is not possible to synchronize the signals from the two satellites except for a small part of the coverage area. Concerning MFN, this dual satellite MIMO configuration was not pursued in the test-bed as it requires two satellites in simultaneous view which is considered an expensive option. C. Satellite Link Budgets We now introduce some reference system link budget to provide typical ranges of signal-tonoise ratio (SNR) at which the demodulator is going to operate. Table III reports hypothetical yet realistic examples of system link budgets for portable and vehicular type of users for the multibeam and the single-beam system inspired by [4]. The user terminals are supposed to include 1 According to this technique, the two polarizations are treated as independent streams of data without any special processing either at the transmitter or the receiver.

9 9 TABLE II NETWORK CONFIGURATIONS: SP STANDS FOR SINGLE POLARIZATION, DP FOR DUAL POLARIZATION. Configuration Satellite Terrestrial DVB-SH CGC Benchmark MIMO ID Links Links/Polarizations mode configuration scheme scheme 0-a 1 polarization Not active SH-A SFN SP-satellite SISO (OFDM) SP-receiver 0-b 1 polarization Not active SH-B MFN SP-satellite SISO (TDM) SP-receiver 1-a 2 polarizations Not active SH-A SFN SP-satellite 2 2 Satellite only (RCHP/LCHP) (OFDM) DP-receiver SM DP - SFN 1-b 2 polarizations Not active SH-B MFN SP-satellite 2 2 Satellite only (RCHP/LCHP) (TDM) DP-receiver SM DP - MFN 2-a 2 polarizations 2 polarizations SH-A SFN SP-satellite 4 2 Hybrid (RCHP/LCHP) H/V (OFDM) DP-receiver SM DP - SFN 2-b 2 polarizations 2 polarizations SH-B MFN SP-satellite 4 4 Hybrid (RCHP/LCHP) H/V (TDM) DP-receiver SM DP - MFN 3 1 polarization 1 polarization SH-a SFN SP-satellite 2 1 Hybrid (RCHP or LCHP) H or V (OFDM) SP-receiver ALA SP - SFN antenna diversity, due to which an array gain is accounted for in the link budget, but no extra diversity gain. The signal power (C), the interference power (I), the SNR or the signal-to-noise plus interference power ratio (SNIR) all refer to unfaded LOS conditions. The adopted space and ground segment parameters are very much in line with the values suggested in the DVB-SH Implementation Guidelines (IG) [4] and representative of current space segment capabilities. In Section IV it it is shown that the MIMO gains are dependent on the SNIR the system is operating under. III. MIMO HARDWARE DEMONSTRATOR DESIGN Figure 4 presents a high level overview of the MIMO hardware demonstrator test-bed architecture. The main blocks of the demonstrator are: the Control and Monitoring unit, the Transmitter

10 10 TABLE III EXAMPLE LINK BUDGET FOR THE SATELLITE COMPONENT. Multi-beam system Single-beam system Parameter Portable Vehicular Portable Vehicular Unit Uplink SNIR db Satellite EIRP/beam dbw Satellite C/I db Free space loss db Atmospheric losses db User terminal G/T db/k Polarization losses db Antenna diversity gain db Signal noise bandwidth MHz SNR downlink db SNIR downlink db Total SNIR db Units (TXU) emulating the satellite gateway and the terrestrial CGC transmitter, the Radio Channel Emulator, and the Receiver Unit (RXU) emulating the User Receive Terminal (UT). The complete MIMO demonstrator has been integrated in one rack as shown in Figure 5. In the following, a very brief description of the main blocks and design aspects is provided. Due to space limitations, the description will be limited to the MIMO aspects of the test-bed and will ignore the elements related to the implementation of the DVB-SH chipset or DVB-SH specific design aspects; for which the interested reader is referred to [26] and [4], respectively. A. Satellite & Terrestrial Transmitter Units The transmitter units are built-up with both COTS (Component Off The Shelf) and custom circuit boards. The RF and clock circuitry is custom designed electronics and the digital processing is done with the ML605 FPGA board from Xilinx [27]. The Digital-to-Analog conversion is handled on a DAC5682z EVM from Texas Instruments [28]. The transmitter units also contain standard ATX PC systems to host the unit firmware. The block diagram of the signal processing chain employed at each of the two TXU is presented in Figure 6. The physical layer of the

11 11 Fig. 4. High level block diagram of MIMO hardware demonstrator for hybrid satellite/terrestrial mobile digital broadcasting. MIMO demonstrator TXU consists of six major building blocks (in order of processing): the TX Data Source, the COMMON TX, the OFDM or TDM PHY, the Sample Rate Conversion, and the Power Amplifier Model. The Power Amplifier Model is not part of the normal transmitter baseband signal processing chain. Rather, it is used to apply a model of a realistic satellite non-linear High Power Amplifier (HPA) model to the transmit signal. The TX Data Source supplies the transmitter chain with the broadcast payload data received from the control software via a FIFO buffer implemented in the physical synchronous dynamic random access memory (SDRAM). Once the transmitter is enabled, the TX Data Source supplies the chain with packets upon request from the downstream transmitter. The broadcast payload data processing begins in the COMMON TX processing block, which builds SH-frames and performs the forward error correction (FEC) encoding. An important function is the convolutional time interleaver which spreads the payload data in time with programmable interleaver depths up to roughly 10 seconds. Thus, the interleaver memory is implemented in the dedicated off-chip SDRAM. The encoded and interleaved SH-frames are then forwarded to either the OFDM or the TDM PHY blocks. The output of the PHY block is a sample stream at the specified baseband sampling rate. The baseband sampling rate must be matched to the DAC sampling rate (fixed at 160 MHz) and

12 12 Fig. 5. Test-bed rack integration. digitally up-converted to a configurable intermediate frequency (50 MHz or 70 MHz). This functionality is performed in the Sample Rate Conversion block. 1) OFDM PHY: As shown in Figure 6 the OFDM PHY block takes up to two parallel bit streams, groups them into interleaved groups that fit into OFDM symbols and maps them to a configurable constellation (QPSK, 16QAM). The complex constellation symbols are MIMO encoded with a configurable encoding scheme (SISO, 2 SISO, SM, Alamouti). The antenna streams are then inserted into a frame structure in the Framing block. The Framing block inserts TPS (Transmission Parameter Signalling) symbols and reference pilot sequences for both antennas. Finally the frequency domain symbol stream is transformed to the time-domain in the OFDM FE block.

13 13 Fig. 6. Transmitter signal processing chain. MIMO Encoder: The MIMO Encoder block takes one or two input symbols streams and encodes these with the selected MIMO modulation scheme. The MIMO modulation schemes are SISO (single stream input unmodified to output), 2 SISO, SM (both dual stream input unmodified to output) and Alamouti (single stream duplicated and encoded in the subcarrier domain with the Alamouti SFBC). The MIMO Encoder block, contains an input FIFO where the input data streams are buffered. Then, the MIMO encoding is done based on the selected mode to produce two spatial streams. The block output is always two antenna streams. The spatial streams are modified by a vector-matrix precoding multiplier that allows to switch the spatially stream to different antennas, as well as nulling an antenna if necessary. In the case of hybrid scenarios where one of the antenna streams comes from one transmit device (e.g. satellite) and the other stream from a second transmit device (e.g. CGC) we can use the precoding multiplier to switch off the unused antennas of the two transmitting devices. Pilot Generation: The PRBS (pseudo random binary sequence) generator handles the reference pilot generation and generates the first bit of the TPS packet for two antennas. This

14 14 requires modifications to the DVB-SH standard, where only a single antenna stream is specified and does not specify the pilot patterns for multi-stream transmission. Thus, the TXU for the OFDM mode slightly diverges from the standard in order to allocate pilot sequences that allow MIMO channel estimation at the receiver. The solution selected for the TXU was to use locally orthogonal sequences that are transmitted on the same pilot sub-carriers for all antennas. Further the pilot sequence for antenna 1 is identical to the DVB-SH pilot sequence. This method requires very minimal changes to the DVB-SH standard. The pilot sequence used for channel estimation are based on the optimal MIMO pilots described in [25] and are formed such that they are locally orthogonal. This is achieved by applying an phase shift to all the original pilot symbols to form the locally orthogonal pilot sequence sent from the second antenna. 2) TDM PHY: The TDM PHY block (see Figure 6) modulates the bit streams with a time domain multiplexing scheme. The block takes up to two parallel bit streams and maps them to a configurable constellation (QPSK, 8PSK, 16APSK). The complex constellation symbols are MIMO encoded with a configurable encoding scheme (SISO, 2 SISO, SM, Alamouti). The antenna streams are then inserted into a frame structure in the Framing block, which inserts reference pilot sequences and scrambles the frames with a complex scrambling sequence. Finally the TDM symbol stream is pulse shaped and the symbol rate is matched to the OFDM symbol sampling rate. The TDM to OFDM sample rate ratio depends on the OFDM guard interval and the TDM roll-off factor [4]. The Sample Rate Conversion block expects inputs sampled at the OFDM sampling rates. Therefore, the TDM sample rate must be adjusted to match. As for MIMO channel estimation, in the TDM, unlike OFDM, the existing DVB-SH options based on quasi-orthogonal scrambling sequences separating the two streams proved to be adequate to support this functionality 2 [13]. B. Radio Channel Emulator The radio channel emulator is a commercially available emulator, namely the Propsim F8, a simplified block diagram of which is presented in Figure 7. For the purpose of the test-bed, the internal PC controls the MIMO fading channel emulator engine and allows remote control of the channel emulation via a LAN interface. A detailed description of the complex MIMO 2 Note that the whole physical layer (PL) slot including data and pilot symbols is multiplied with the scrambling sequence.

15 15 Fig. 7. Propsim F8 with 4 RF channels. fading channel emulator engine is beyond the scope of this document and the interested reader is referred to the Propsim F8 brochure [29]. The Propsim F8 is a file-based channel emulator, i.e. the channel model that is to be emulated is generated off-line and stored in a file. These channel time series may be either the result of a field measurement or generated via a stochastic channel model. This allows both the playback of measured channel or synthesized channel models. To support the hybrid MIMO configurations, up to 4 2 on a single frequency or two times 2 2 on multiple frequency bands are required, as illustrated in Figure 8. Both cases are supported by Propsim F8 both with internally or externally generated RF local oscillators. In MFN mode the satellite transmit signal is applied to inputs #1 and #2 and the terrestrial transmit signal is applied to inputs #3 and #4. The outputs #1 and #2 are connected to the satellite receiver RF chain and outputs #3 and #4 are applied to the terrestrial receiver RF chain. In SFN, the four channel emulator inputs are from a dual-antenna satellite and dual-antenna terrestrial gap filler. Both MFN and SFN scenarios can be reduced to the corresponding single-antenna cases by disabling the second antenna sub-channels. To support the MFN operating scenario in both UHF and S-band which corresponds to an RF frequency range between 300 MHz to 4000 MHz, two internal local oscillators are employed. C. Receiver Unit The RXU, capable of demodulating up to two parallel data streams is implemented on three Xilinx Virtex-6 LX240T FPGA circuits. It is built-up with both COTS and custom circuit boards.

16 16 Fig. 8. Channel emulator MIMO setups. The RF and clock circuitry is custom designed electronics and the digital processing is COTS. The digital processing is implemented with the ML605 FPGA board from Xilinx [27]. Analogto-Digital conversion is implemented with a ADS62P48EVM from Texas Instruments [30]. The receiver unit also contains a standard ATX PC system to host the unit firmware. As illustrated in Figure 9, the receiver contains two parallel radio chains. The first chain is the primary radio that is used in all usage scenarios including DVB SH-A and SH-B with SFN and MFN. The RF tuner of the primary radio is tuned to the first center frequency. The second chain is the secondary radio and is used in MFN scenarios where the two signals are on separate frequency bands. The RF tuner of the secondary radio is then tuned to the second of the two center frequencies. The primary and secondary radio FPGAs (FPGA 1 and FPGA 2) contain the physical layer processing required to demodulate a MIMO signal to form receive bit Log- Likelihood Ratio (LLRs) for up to two parallel spatial streams. The physical layer processing is configurable to be either TDM or OFDM. In an MFN scenario the receive signal streams from the two bands are combined with complementary code combining. This combining takes place in the third FPGA (FPGA 3 COMMON), which contains the FEC decoding and MPEG-TS packet stream demodulation in addition to the MFN combining. A fourth FPGA is connected

17 17 Fig. 9. MIMO hybrid receiver architecture. to FPGA-3 to offload some of the processing if required and for securing the possibility of a future expansion. The functional receive chain partitioning is illustrated in the block diagram in Figure 10. The content of FPGA1 and FPGA2 is identical containing the signal processing required to demodulate 1 or 2 spatial streams received on 1 or 2 antennas. The PHY FPGAs (FPGA1 and FPGA2) contain processing for both TDM and OFDM, however only one of the waveforms will be active at any given time. When receiving in OFDM mode, the dedicated TDM blocks are disabled and vice versa. The output of FPGA1 and FPGA2 is one or two parallel streams of received bit LLRs. These streams are routed to FPGA3 where the COMMON processing is performed. The COMMON processing deinterleaves the LLR streams, combines the streams in the case of code combining, and decodes the Turbo encoded LLR streams. The decoded FEC words are stream and mode de-adapted to form the receive MPEG-TS payload streams. Each FPGA is connected to the serving PC via a Peripheral Component Interconnect Express (PCIe) interface. This connection, in addition to the off-chip SDRAM allows data tracing from various points along the chain. The physical layer receiver of the MIMO demonstrator consists of five major building blocks (in order of processing): the Sample Rate Conversion, OFDM or TDM PHY, COMMON RX, and the RX Data Sink. In fact, the receiver contains two instances (FPGA1 and FPGA2) of the Sample Rate Conversion and OFDM/TDM PHY. In the case of

18 18 Fig. 10. Receiver signal processing chain. MFN operation, the two PHYs are demodulating signals from different RF carrier frequencies. Combination of the two radio signal is then done in the COMMON RX block (FPGA3). As is evident from Figure 10, there are a number of blocks in the TDM and OFDM chain that perform roughly the same computation. In particular the MIMO Detector and SNIR estimation are blocks that are similar enough for TDM and OFDM so that they can be shared. The FPGA will be configured at run-time to be in either TDM or OFDM mode. Taking this into account the block diagram of the PHY FPGAs is shown in Figure 11. The ADC sample stream is converted to a baseband data rate that is a function of the bandwidth option selected in the Sample Rate Conversion block. This is a flexible decimator that converts the input rate from a constant 160 MHz to the proper sample rate depending on the bandwidth selected. In TDM mode, the symbol timing is recovered in the Timing Recover block using a Gardner Timing Error Detector [33] and interpolator. Once the symbol timing is found and corrected

19 19 Fig. 11. FPGA1 and FPGA2 top level block diagram. the synchronization procedure searches for the PL-Slot boundaries in the PL-Slot SYNC block, after which channel estimation can be performed on the pilot fields of the PL-Slots (CH-Est in Figure 10). The channel estimator exploits the pilot fields of the PL-Slot to form a sample set of the noise. This noise sample set is forwarded to the SNIR-Est block where the noise variance is computed and tracked. The noise-variance, the received data symbol vectors and the MIMO channel estimate matrices are input to the MIMO Detector where the detection is performed to produce soft-bit LLRs of one or two parallel spatial streams similar to [31]. From here the TDM PHY completes with a start of frame (SOF) Sync block where the start of the SH-Frame is found. The block then forwards complete SH-Frames to the COMMON processor for detection. In OFDM mode, the synchronization is handled in a multi-step process. In the PreFFT-Sync block the OFDM symbol timing and fractional carrier frequency error are estimated and tracked. Once the acquisition of these parameters is obtained the Fast Fourier Transform (FFT) block converts the time-domain OFDM symbols to the frequency domain. The FFT block utilizes the FFT Engine where the actual Fourier transform is performed. The PostFFT-Sync block estimates the integer-sub-carrier frequency offset and synchronizes to the scattered pilot pattern. Once these two are found, the channel estimation of the receive symbols can be started. The CH-Est block utilizes the FFT Engine in computing the channel estimate matrix for each sub-carrier. The CH- Est block also exploits the scattered pilot symbols to form a sample set of the noise present

20 20 on the signal. This noise sample set is sent to the SNIR-Est block where the noise-variance is computed and tracked. The noise-variance, the received data symbol vectors, and the channel estimate matrices are input to the MIMO Detector where the detection is performed to produce soft-bit LLRs of one or two parallel spatial streams. From here the blocks of data are symbol deinterleaved in the Symbol De-Interleave block before forwarding to the COMMON processor for detection. The COMMON processing deinterleaves the LLR streams, combines the streams in the case of code combining, and decodes the Turbo encoded LLR streams. The decoded FEC words are stream and mode de-adapted to form the receive MPEG-TS payload streams. MIMO Detector: The MIMO Detector block (Figure 12) is shared between the TDM and OFDM PHY processing chains and essentially removes the radio channel effects as well as the MIMO encoding that is present in the input symbol vector. The supported MIMO schemes are: SISO (1 and 2 ), SM, and Alamouti. The block always produces two soft-bit LLR streams and depending on the spatial rate (1xSISO, Alamouti = 1 stream, 2 SISO, SM = 2 stream) either one or both are considered valid. The first block in the chain is the Alamouti decoder, which block decodes the Alamouti SFBC when the Alamouti MIMO mode is enabled. In all other modes the block simply passes the input directly to the output. The second block, namely the Input Format, buffers the incoming receive symbols, channel matrices, and SNR estimates. The block then feeds the Symbol LLR block with the received data and combined constellation candidates. The block contains ROM storage of the candidate vectors. The Symbol LLR block computes the symbol LLRs for each of the constellation candidates, which corresponds to maximum likelihood MIMO decoding [31]. The same block is used for all MIMO modes. The modes are distinguished by their candidate lists. In the worst case (16QAM/16APSK SM) the candidate list is of length 256. The Bit LLR block computes the receive bit LLRs based on the symbol LLRs. In the worst case the Bit LLR computes 8 bits in parallel. All modes and modulations use the same chain. Finally the soft-bit LLRs are written out of the block in serial fashion. The number of soft-bits written per iteration depends on the modulation order. IV. TEST-BED EMULATION RESULTS The MIMO test-bed for mobile satellite/terrestrial broadcasting has been run to perform an extensive emulation and system test campaign to evaluate a number of performance aspects of the

21 21 Fig. 12. MIMO detector block diagram. dual polarization DVB-SH system described in Section II. The key findings out of a huge volume of measurements are summarized in the following paragraphs. The control and monitoring station equipment has been developed by Elektrobit to perform test-bed automation tasks, to monitor the platform in real-time and to perform post-processing of the results. For the purposes of the TABLE IV LIOLIS-CTTC MIMO LMS CHANNEL MODEL PARAMETERS. Parameter Value Used Carrier frequency S-band 2.2 GHz Vehicle speed 60 km/h Polarization RHCP/LHCP Propagation environment ITS Elevation angle 40 Small-scale correlation Polarization correlation coef. Tx/Rx 0.5 Large-scale correlation Correlation between LOS components XPD parameters Antenna polarization discrimination 15 db Environment cross-polarization 5.5 db 2-state first-order Markov model Good > Good 0.6 Bad > Bad Good > Bad 0.4 Bad > Good Markov state update interval 0.18 s (3 m movement) Loo distribution triplet As in Tables III and IV of [24] Doppler spectrum Low pass filter

22 22 test-bed, the stochastic MIMO LMS channel model described in [16] was employed as a starting point. The Liolis-CTTC model is a statistical dual polarization MIMO channel model of an ITS environment, the parameterizations of which can be found in Table IV. The model assumes a fixed UT speed of 50 km/h. To increase the realism of the results obtained, dual polarization channel samples measured during the ESA MIMOSA real world satellite campaign [17], [18] were stored and played back in the test-bed for a varying UT speed. The MIMOSA channel samples are extracted from channel sounding measurements performed in an ITS and a MIX (of suburban, tree-shadowed and open) environment. The MIMOSA field experiment campaign was carried out at Erlangen and Lake Constance, in Germany. The campaign exploited two circularly polarized overlapping beams from the Solaris S-band payload embarked on the Eutelsat 10A geostationary satellite. The two beams are in opposite circular polarization thus allowing the simultaneous transmission of two orthogonal polarizations. The elevation at which the satellite was seen was ranging from 33 to 35 degrees. A snapshot of the ITS measurements is given in Figure 13. In this figure it is apparent some degree of uncorrelation among the four LMS channel H i,j coefficients, essential to achieve the MIMO gain [17], [18]. As concluded by the measurement campaign, the co-polar to co-polar correlation is very high as expected. Amid the cross-polarized channels, a lower correlation is witnessed, but effects like Angle-of-Arrival (AoA) dependent XPD of the antennas combined with partly correlated deep fading influence the characteristics of the correlation. Concerning the co-polar to cross-polar correlation, for urban environments a change of polarization is observed with significant probability. For fades higher than 10 db the cross polarized channel is partly stronger than the co-polarized one. For suburban and tree shadowing this effect is observed with much lower probability (suburban) or is non existing (tree shadowed). In the remainder, the three satellite channels are referred to as Liolis-CTTC, ITS and MIX. As explained in Sect. I, in all MIMO configurations in Table II there are no additional antennas, only single antennas with dual polarization capability. The corresponding terrestrial MIMO channel model adopted in the test-bed is the generic WINNER II channel model [32], and particularly the IMT-A (International Mobile Telecommunications) Suburban Macro variant, which follows a geometry-based stochastic channel modelling approach. Concerning the performance metrics used, the test-bed keeps counters for the BER/FER (Bit/Frame Error Rate), but more importantly includes an automated algorithm for calculating the error second ratio ESR5(20), which is

23 23 Fig. 13. Snapshot of MIMO dual polarization channel dynamics: measured ITS time series. fulfilled in a time interval of 20 seconds if there is at most one second in error [5]. The result is a second order statistical performance metric better suited in assessing the MPEG video quality. In DVB-SH, the operational point is defined by the SNR value for which the ESR5(20) criterion is fulfilled for 90% of the time. For the test-bed results reported hereinafter, all references to SNR will imply the fulfillment of this criterion unless explicitly stated. It should be noted that in order to achieve reliable ESR5(20) estimates each measurement point has been obtained after collecting 2-4 hours of real time data. When not mentioned otherwise, the SH-A (OFDM) results have been obtained over a 5 MHz bandwidth, 2K FFT size and Guard Interval of 1/4, whereas the SH-B (TDM) results have been obtained for a 5 MHz channel bandwidth and a 0.25 roll-off. Both modes employ a Uniform Long interleaver of 10 seconds. It is worth highlighting that for all hardware emulations, the implementation losses with reference to the software simulation results [22] are kept very reasonable, that is within 1.5 db and typically less than 1 db.

24 24 A. Satellite Only Performance In this subsection we will report on the satellite only SH-B 3 emulations in the form of spectral efficiency 4 vs. LOS SNR curves for SISO, 2 SISO and MIMO (SM). This corresponds to Configuration 1-b in Table II. For fairness of comparison, the total SISO transmit RF power has been split among the two polarizations in 2 SISO and MIMO. We start in Figure 14 with the Liolis-CTTC model and proceed in Figure 15 and Figure 16 by presenting the results for the MIX and ITS measurements, respectively. The first conclusion from all three figures is that MIMO is always better than SISO and 2 SISO, with the improvement becoming higher as the SNR increases. This result is in line with previous analytical and simulation findings [11], [22]. There are two ways of interpreting this improvement: either by fixing the spectral efficiency and quantifying the power saving from employing MIMO or by assuming an operational SNR and then evaluating the improvement in data rate. For example, with SISO technology based on DVB-SH, a typical system configuration would be one based on QPSK 1/2. At that level of spectral efficiency 1 bit/symbol, MIMO yields at least 3 db of valuable on-board power savings independently of which of the three channels we look at. Vice-versa, looking at the link budgets in Table III, MIMO will achieve, for example, 12.2 % better data rate compared to SISO in the ITS channel of a multi-beam system for a vehicular terminal. In the single-beam system, where the power can be pushed to higher levels, the corresponding gains are more impressive and amount to 108% in the MIX channel and 100% in the ITS channel. Although 2 SISO is better than SISO in Figures 14 and 15, this trend is reversed in Figure 16 due to the higher impact of the cross-polarized channels. Note that the same ESR5(20) will in general correspond to a different FER, a fact that explains the differences in slope between the SISO, 2 SISO and MIMO curves. Table V presents the very good match between computer simulation and hardware test-bed results for a few configurations of the satellite only scenario, which is of special importance to this test-bed. This confirms the earlier general statement that the difference is kept within the 1 db range. In reality, the difference is even negative in some lines of Table V due to the slightly different algorithm used for computing the ESR5(20) metric. 3 Actually, when the equivalent satellite radiated isotropic power is the same, there is very little difference in performance for the satellite channel which is flat between SH-A (OFDM) and SH-B (TDM). Minor difference (e.g. 0.3 db) comes from the different framing between the two options. 4 As previously mentioned the results are provided for a 90% ESR5(20) fulfillment.

25 25 Spectral Efficiency [bits/symbol] Liolis CTTC SISO 2xSISO MIMO (SM) LOS SNR [db] Fig. 14. Satellite only performance based on the Liolis-CTTC channel model. Spectral Efficiency [bits/symbol] MIX 2xSISO MIMO (SM) SISO LOS SNR [db] Fig. 15. Satellite only performance based on the MIX measured channel.

26 26 Spectral Efficiency [bits/symbol] ITS 2xSISO MIMO SISO LOS SNR [db] Fig. 16. Satellite only performance based on the ITS measured channel. TABLE V COMPARISON BETWEEN COMPUTER SIMULATION RESULTS AND HARDWARE TEST-BED. Configuration Physical Channel SNR 90% ESR5(20) SNR 90% ESR5(20) Difference Layer (Computer Simulation) (Hardware Test-Bed) [db] 1-b QPSK 1/2 MIX b QPSK 1/4 ITS b QPSK 1/4 MIX b QPSK 1/2 MIMOSA ITS a QPSK 1/3 MIX a QPSK 1/3 Liolis CTTC a QPSK 1/3 Liolis CTTC b QPSK 1/2 MIX, phase noise B. Hybrid SFN/MFN Performance The next set of results will refer to the comparison between MIMO and SISO in a hybrid configuration, where both satellite and terrestrial signal components are received by the UT. In this context, SISO refers to each component contributing a single stream to the total received

27 27 power, which is equivalent to the existing DVB-SH standard unmodified, and MIMO refers to a dual polarization stream being transmitted from each component, where each transmitter repeats the same data employing SM. Figure 17 and Figure 18 depict this comparison for the hybrid SFN (hence OFDM) configuration (2-a in Table II) for the MIX and ITS environments, respectively. A ratio of satellite over terrestrial received power equal to 37/100 has been used assuming a particular terrestrial link budget and repeater to UT link distance. As before, the SNR reported in the x-axis of the figures refers to LOS SNR (also for the terrestrial link). At the level of 1 bit/symbol, adding a second stream (polarization) to the SISO hybrid system yields impressive in MIX (4.0 db) and more moderate in ITS (1.5 db) power savings. These performance gains increase together with the operational SNR. This is in accord with the DVB-SH IG computer simulation findings reported in Annex A of the [4]. However, it is worth noting that for the ITS environment, MIMO seems beneficial only after crossing a certain spectral efficiency level of around 0.7 bits/symbol. This is in line with typical theoretical MIMO findings, where the MIMO multiplexing gain appears at higher SNR levels [7]. Next in Table VI, we evaluate how the different satellite and terrestrial components contribute to various MIMO system configurations. Specifically, we explore a satellite only reception, a terrestrial only reception in two frequency bands (UHF and S), a hybrid SFN (configuration 2-a) and a hybrid MFN (configuration 2-b) reception, all set up for providing equal spectral efficiency. The channel used to obtain the results in Table VI is the Liolis-CTTC one. What is worth highlighting is the remarkable improvement, up to 10 db, when there is coverage from both components, instead of only one of them, which can be attributed to more power being received, but also to the increased diversity. Furthermore, hybrid MFN is achieving better performance than hybrid SFN. However, the resources it exploits are double: MFN implies a 2 2 MIMO scheme both in UHF and in S bands. C. Sensitivity Analysis After presenting the general performance trends and significant improvements coming from the use of MIMO in dual polarization DVB-SH, the focus of this subsection will be on the impact of particular physical layer aspects on the MIMO performance. Including these important practical effects in the results increases the confidence of the performance obtained, which has been the main motivation for developing the MIMO test-bed.

28 28 Fig. 17. Hybrid SFN performance based on the MIX measured channel. TABLE VI CONTRIBUTION OF SATELLITE AND TERRESTRIAL COMPONENTS IN VARIOUS MIMO SYSTEM CONFIGURATIONS. Configuration Transmission Spectral 90% Technique Efficiency ESR5(20) [bits/symbol] level [db] SAT ONLY SM QPSK 1/3 4/ TERR 700MHz SM QPSK 1/3 4/ TERR MHz SM QPSK 1/3 4/ MFN Hybrid SM QPSK 1/3 4/ SFN Hybrid SM QPSK 1/3 4/3 6.4 Although all the results presented in the previous subsections correspond to a uniform long interleaver of 10 seconds, the longer time interleaving leads to longer zapping times degrading the viewer s video experience. In some systems, this might be seen as a commercial limitation. Then, MIMO should keep its high performance also over shorter interleaving depths. Table VII

29 29 Fig. 18. Hybrid SFN performance based on the ITS measured channel. compares 10 seconds interleaving with 5 and 1 seconds in the case of satellite only and hybrid SFN reception (configuration 3 in Table II). In the former case, the MIMO SM scheme is employed, whereas in the latter the Alamouti SFBC scheme. Interestingly, the results reveal a very small impact on the performance when the interleaving length is reduced from 10 to 5 seconds. This experimental testbed finding is in line with the analysis reported in [13]. The performance degradation becomes higher when the smallest interleaver of 1 second is used, but still only around 2.2 db for SM and 3.5 db for Alamouti. This is in sharp contrast with the SISO performance which depends critically on the size of the interleaver and substantially degrades for shorter interleaver lengths. Indicatively, moving from 10 to 5 seconds of interleaving length costs Config. 1-b SISO QPSK 1/2 about 5 db in ITS [2]. Hence, it is possible to draw the conclusion that MIMO preserves the performances even at reduced interleaving depth, which is another appealing added value of MIMO versus SISO. An aspect of mobile satellite channels that is not usually part of terrestrial MIMO investigations is the HPA non-linear behavior on