Effect of Satellite System Impairments on a Multilevel Coding System for Satellite Broadcasting

Transcription

1 Effect of Satellite System Impairments on a Multilevel Coding System for Satellite Broadcasting Aharon Vargas 1, Cédric Keip 1, Wolfgang H. Gerstacker 2, and Marco Breiling 1 1 Fraunhofer Institute for Integrated Circuits (IIS), Am Wolfsmantel 33, D Erlangen, Germany, {aharon.vargas,cedric.keip,marco.breiling}@iis.fraunhofer.de 2 Chair of Mobile Communications, Universität Erlangen-Nürnberg, Cauerstraße 7, D Erlangen, Germany, gersta@lnt.de Abstract In this paper, we evaluate a multilevel coding (MLC) scheme with multistage decoding (MSD) designed for satellite broadcasting communications. The impact of three different satellite system impairments on the decoding performance is analyzed. First, the influence of errors introduced by the channel estimation is discussed, assuming a typical data aided (DA) channel estimator with different pilot lengths. Second, the impact of the residual phase noise present after the phase recovery is investigated using a model based on a normal distribution. Finally, the degradation introduced by the non linearities of the satellite power amplifiers is also analyzed. The impact of these effects is investigated via the mutual information. Besides, bit error rate (BER) simulations are performed for each impairment effect. The considered MLC scheme is compared to a classical bit interleaved coded modulation (BICM) scheme, showing that the MLC scheme provides different grades of robustness for each level. I. INTRODUCTION The growing interest in broadcasting a huge amount of data, e.g. high definition TV content, has motivated the deployment of satellite broadcasting systems, where high rate information streams can be distributed to a large region. Recently, appropriate standards have been released, e.g. the ETSI Satellite Digital Radio (ESDR) standard [1], [2] and the Digital Video Broadcasting Satellite to Handhelds (DVB SH) standard [3], which support services with high data rates of several Mbit/s. The information carried by each service, e.g. audio, video, news or traffic information, can have different levels of importance which necessitates a configurable content dependent quality of service (QoS). The ESDR standard provides different protection levels by splitting the data in different pipes, which are protected using one of the different code rates available in the transmitter and the forward error correction (FEC) block. For example, the 3GPP2 Turbo code [4] supports code rates ranging from 1/5 to 6/7. On the other hand, the services can also be unequally protected by using hierarchical modulation in combination with the concept of multilevel codes (MLC) [5], as originally proposed in [6]. An MLC scheme suitable for satellite broadcasting communications has been proposed in [7] and further investigated in [8], where a typical bit interleaved coded modulation (BICM) [9] scheme and an MLC scheme are compared. The MLC approach presents more flexibility when setting different levels of QoS, and the highly protected levels can operate at lower signal to noise ratios (SNRs) than the BICM approach. This is especially interesting for satellite communication systems due to the low link budgets available for such systems. In [7] and [8], a simple additive white Gaussian noise (AWGN) channel or a Land Mobile Satellite (LMS) channel model have been assumed, respectively. However, the impairments associated to satellite systems have not been taken into account so far. In this paper, we focus on three typical impairments present in satellite systems and their impact on an MLC scheme with multistage decoding (MSD). The first effect is related to the errors introduced by the channel estimation unit, which leads to a degradation when demapping the different levels of the MLC scheme. We employ a data aided SNR estimation algorithm, which is typically used in satellite systems due to its high performance for low SNRs. Second, we analyze the residual phase noise which remains after the phase recovery. The phase noise can be accurately described by a normal distribution for the system simulations. The third impairment effect is related to the non linearity of the satellite high power amplifiers (HPAs). Due to the long distances between the satellite and the user terminals, the power amplifiers located in the satellite emit a very high power, which leads to non linearities in the output signal. We model this effect by a complex valued attenuation factor. Finally, we compare the MLC scheme and a typical BICM scheme proposed by the ESDR and DVB SH standards under the different impairments. The remainder of this paper is structured as follows. Section II introduces the considered system concept including encoder and decoder. The above mentioned impairment effects present in a satellite communication system are described in Section III. In Section IV, the proposed MLC system is compared to a typical BICM system for the impairment effects of Section III. Finally, some conclusions are drawn in Section V. II. SYSTEM MODEL MLC has been proposed for satellite broadcasting systems in [6] and [7], where only the effect of an AWGN channel has been analyzed. A more accurate channel description, based on

2 Fig. 1. Multilevel coding system, composed of a multilevel encoder, a channel and a multistage decoder. an LMS channel model [10], is considered in [8] to obtain more realistic performance results for MLC in a satellite broadcasting system. In this paper, we focus on different impairments associated to satellite communications: channel estimation errors, phase noise, and non linearity of the satellite HPAs. In order to analyze these effects, we consider the system model shown in Fig. 1 based on the DVB SH implementation guidelines [11] and the DVB S2 standard [12], where the error correction scheme corresponds to MLC using MSD at the receiver side, in contrast to [11], [12]. The codewords C i, i = 1,2,...,k, are generated in parallel by encoding a set of k data services with information words {D 1,...,D k }, and mapped onto symbols by reading k bits c i, one from each codeword, in parallel. The codeword bits c i are mapped onto a unique constellation point to generate a channel symbol, determined by the block labeling (BL) partitioning, cf. also [13]. BL partitioning is based on an iterative procedure, where the constellation is divided into subsets grouped as clusters well separated from each other. A channel interleaver assures that the channel effects of consecutive channel symbols at the decoder input are well separated over the time. Considering an AWGN channel, the received signal is given by y = e jzp a x+n, (1) where x is the transmit signal, e jzp with random z p represents the phase noise, a is a complex valued attenuation factor introduced by the HPA which is extracted for each symbol from an appropriate model described in Section III.C, and n corresponds to a complex valued white Gaussian noise process with double sided power spectral density N 0 in equivalent complex baseband. At the receiver, the phase recovery block uses typically a phase locked loop (PLL) [14] to compensate for the phase noise introduced by the channel. The output symbols are de interleaved and passed on to the channel estimation block, which provides a reliable estimate of the status of the channel (noise variance) to the demapper. The de interleaved symbols ẙ are demapped resulting in a posteriori probabilities (APPs) of the codeword symbols c i based on the received signal. The APPs are represented as conditional log likelihood ratios (LLRs), expressed as L(c i ẙ) = ln P (c i = 1 ẙ) P (c i = 0 ẙ), (2) which are fed to the FEC block which decodes the codewords in a recursive MSD process. The levels are decoded sequentially using the decoding results of the previous levels as a priori information. The FEC block is a hard output Turbo decoder [15] which outputs the estimated data sequence ˆD i based on the LLRs. Subsequently, ˆDi is re encoded and used as a priori information for the demapping of the next levels. III. IMPAIRMENT EFFECTS IN A SATELLITE BROADCASTING SYSTEM A. Channel Estimation Errors The demappper block requires information about the quality of the incoming channel symbols ẙ in order to calculate the LLRs properly. An inaccurate channel estimation could give high importance to LLRs with wrong sign, producing errors when decoding the codewords. The channel estimation block measures the channel quality by estimating the variance of the noise σ 2 n. Data aided SNR estimation algorithms are based on known pilot symbols and provide more accuracy than the non data aided estimators [16]. In satellite communications with low link budgets, a data aided channel estimator is preferred. The DVB S2 standard [12] proposes the signal to noise ratio estimator (SNORE) algorithm, originally devised by NASA for

3 Fig. 2. Phase noise profile before and after the phase recovery block. deep space missions and further analyzed in [16]. Defining the average power of the pilots and the average power of the received pilots as P s and P p, respectively, the SNORE estimator delivers an estimate N SNR ˆ P s = = P p P s 1 N 2 1 N ẙ p p=1 N ẙ p 2 N 2 (3) 1 N ẙ p p=1 p=1 for the SNR, where N represents the number of transmitted pilots involved in the SNR estimation. Note that only the received symbols ẙ p which correspond to the pilots are used in the SNR estimation. Without loss of generality, we assume that all the symbol pilots are equal to one. The estimated value SNR ˆ is used in the demapping of a block of M data symbols, until the next N pilots are received. Obviously, the SNR estimation error increases as the length of the pilots N decreases. B. Phase Noise The phase noise present in a communication system can be represented by a term e jzp, which is multiplied by the transmitted signal x, and accounts for the total phase noise of frequency sources in the system. The phase noise can be described in the frequency domain providing single side band (SSB) phase noise values at different frequency offsets with respect to the carrier [17]. At the receiver, the phase recovery module implemented as a PLL corrects the phase deviations of the received signal by means of low pass filtering. Therefore, a PLL can follow and correct phase changes at low frequencies, but not at high frequencies. The range of frequencies that can be tracked by the PLL depends on the frequency of the pilots and the transmission rate [14]. Fig. 2 shows the phase noise profile (or power density spectrum function), obtained from the DVB S2 standard user guide [18]. The dashed line represents the profile without Fig. 3. Pdf of the residual phase error for the phase profile of Fig. 2. Fig. 4. AM AM and AM PM characteristic of a commercial TWTA. any processing, while the solid line is the phase noise profile after the phase recovery block, assuming that this block tracks frequency offsets below 1 khz. We focus on the residual phase noise profile after the phase recovery because we are interesting in the effect of the residual phase noise in the demapping and decoding block. The residual phase noise can be simulated by applying the inverse fast Fourier transform (IFFT) to the noise profile to obtain a noise shaping filter response, but this operation increases the simulation time of the whole system. Instead, we directly model the phase deviations at the output of the phase recovery block in order to reduce the simulation time. A selected symbol of the constellation is transmitted applying only phase noise, and the pdf of the angle deviation errors in relation to the transmitted symbol is calculated. Fig. 3 shows the pdf of the residual phase error after the phase recovery block for the phase profile of Fig. 2. Additionally, a normal distribution with a standard deviation of σ p = 3.4 is also represented with a dashed line. Note that the distribution of the residual phase error can be approximated well by a normal distribution, as stated in [19], where the standard deviation of the normal distribution is proportional to the true standard

4 deviation of the residual phase noise. This approximation is used in Section IV for the simulations. C. Non Linearity of Power Transmitters Satellite transponders have to be designed in order to maximize the radiated power. Therefore, high signal levels are usually employed at the input of the HPA, which forces the HPA to work close to the saturation point producing distortion in the transmitted signal due to non linearities. The effects of non linearities can be reduced by applying an input back off (IBO), which reduces the input power of the power amplifier, moving the operation point below the non linear segment of the amplifier characteristic. However, the IBO reduces also the output level, which decreases the efficiency of the amplifier. This effect arises in traveling wave tube amplifiers (TWTAs) and solid state power amplifiers (SSPAs), which are commonly used in satellite transponders [20]. The behavior of these HPAs is usually described in terms of the amplitude modulation to amplitude modulation (AM AM) and the amplitude modulation to phase modulation (AM PM) relationships. Fig. 4 shows the AM AM and AM PM characteristics given in [18] for a TWTA, which are used to adjust the output amplitude of each symbol by applying a factor a (see Fig. 1). Therefore, this factor a is a function of the instantaneous signal power and depends on the input amplitude of each symbol. This model is used in Section IV for the simulations, combined with pulse shaping to limit the effective bandwidth of the transmission. For pulse shaping, a raised cosine filter with a roll off factor of 0.15 has been employed. IV. SIMULATION RESULTS The system model shown in Fig. 1 is employed in all the simulations, where the FEC block is a fixed point 3GPP2 Turbo decoder [4] for a block size of bits and a code rate of 2/5, which uses the max log MAP algorithm described in [21] for the MAP decoding block. A 32 ary amplitude phase shift keying (32 APSK) is used as modulation format, and a BL partitioning is applied which results in five different levels of protection, k = 5. The effects of the impairments described in Section III are analyzed separately, measuring the bit error rate (BER) at the output of the FEC block. Besides, for each level the mutual information (MI) I [13] between the output LLRs l of the demapper of the level and the original input data d to the mapper of the level is calculated as I = d {0,1} + p(l,d) log ( p(l,d) p 1 (l)p 2 (d) ) dl, (4) where p(l, d) denotes the joint probability density function (pdf) of l and d, and p 1 (l) and p 2 (d) are the corresponding marginal pdfs. Note that the impairment effects described in Section III cannot be influenced by the designer. Thus, they can be seen as part of the channel. Therefore, the general MI expression (4) can be used in all the simulations, considering the impairment effects and the AWGN channel as an entire block which distorts the transmitted signal. For the MLC Fig. 5. Mutual information of a classical BICM scheme and the proposed MLC scheme for different pilot lengths N. Fig. 6. BER of a classical BICM scheme and the considered MLC scheme for a pilot length of N = 2 and for perfect SNR knowledge. scheme, the MI for each level is calculated independently. For the BICM scheme, the MI of all bits of each signal point is calculated and the results are averaged. Via the MI approach, the performance of the considered scheme can be evaluated over a wide range of SNRs in a short simulation time, while the BER is mainly relevant in the SNR region where the FEC starts to decode the codewords correctly. Fig. 5 shows the MI versuse s /N 0 (E s : average received energy per symbol) of a typical BICM scheme and the proposed MLC scheme for different pilot lengths N, assuming a data block length of M = 100 symbols per each sequence of pilots. Therefore, N = 2 and N = 5 correspond to a pilot overhead of 2% and 5%, respectively. The value N + corresponds to the ideal case of perfect knowledge of the channel. Phase noise and non linearity of the HPA are not considered, i.e., a = 1 and z p = 0. Note that only the best and worst levels of the MLC scheme are shown. The rest of the levels performs between these two levels. The loss introduced by the channel

5 Fig. 7. Mutual information of a classical BICM scheme and the proposed MLC scheme for different phase noise distributions. Fig. 8. BER of a classical BICM scheme and the considered MLC scheme for a phase noise distribution with σ p = 5. estimator is approximately constant within the simulated SNR range (-4 to 16 db) and it corresponds to about 1.5 db for the case of N = 2. Increasing the pilot length to N = 5 reduces the loss by about 0.5 db. The BICM scheme presents the same behavior. The BER corresponding to the same scenario is shown in Fig. 6, where the performance for perfect knowledge of the channel is also shown (dashed line). The operation points extracted from Fig. 5 for a code rate of 2/5 do not match exactly with the waterfall region of the BER curves. This loss corresponds to the fixed point implementation of the 3GPP2 Turbo decoder and the considered suboptimal max log MAP algorithm, which introduce a loss of about 0.5 db. Obviously, this effect arises for all the simulations. The pilot length is fixed to N = 2 for the simulations of Fig. 6, resulting in a constant loss of about 1.5 db for the MLC and BICM schemes. According to Fig. 5, if a code rate of 2/5 is used for the BICM scheme, the received codewords should be decodable for SNRs higher than 5 db for the case of N +. However, Fig. 6 shows that the waterfall region of the BICM scheme is located at 8 db. This loss occurs typically for a BICM scheme, where no iterations between the demapper and the FEC block are done which implies that the statistical dependencies between the five bits of each symbol of the 32 APSK modulation are not taken into account. Note that this effect becomes more critical for higher modulation orders. The approach described in Section III.B is employed to simulate the residual phase noise after the phase recovery block. Different phase noise profiles can be modeled by adjusting the variance σ 2 p of a normal distribution. Fig. 7 shows the MI of BICM and MLC schemes for different levels of phase noise. Other impairment effects are not considered. The first level of the MLC scheme shows almost no loss for the considered phase noise levels. Only for high SNRs and a phase noise with σ p = 5, a loss of around 1 db can be observed. The same effect can be noticed for the BICM scheme. However, the worst level of the MLC scheme suffers from a higher degradation (2 db), which is due to the proximity of the constellation points, where small phase deviations lead to a considerable performance degradation. This loss does not arise for low SNRs, because here the phase noise is below the AWGN. Fig. 8 shows the BER of the BICM and the MLC scheme in comparison, where the standard deviation of the phase noise is fixed to σ p = 5. For the selected code rate of 2/5, the first level of the MLC scheme exhibits no degradation, and the last level and the BICM scheme show a degradation below 0.5 db. Therefore, if the phase recovery block is implemented properly, the remaining residual phase noise does not have a big impact in the FEC block. Fig. 9 shows a comparison in terms of MI between the BICM scheme and the considered MLC scheme for an HPA with IBO = 3 db, which represents a compromise between output signal power and linearity of the HPA (see Fig. 4). Higher values of IBO move the operation point to the left, reducing the effect of the non linearities but also decreasing the output signal power. The first level of the MLC scheme shows almost no loss for low and medium SNR values, while the loss increases dramatically for high SNRs to more than 2 db for an SNR of 16 db. This effect is even much stronger for the BICM scheme, where the loss introduced for high SNRs might be unacceptable for many communication systems. The same problem arises in the fifth level of the MLC scheme. In this case, a different partitioning for the considered modulation format could help to mitigate the impact of the non linearities of the HPA. Note that the BL partitioning is based on the distance between the signal points, and it does not consider the distortion introduced by non linear components. Fig. 10 shows the corresponding BER for the considered schemes for a code rate of 2/5, which reflects the degradation of the BICM scheme and the worst level of the MLC scheme. V. CONCLUSIONS In this paper, an MLC scheme with MSD suitable for satellite communications has been evaluated. The effect of different

6 satellite system impairments, and the worst level suffers from almost the same degradation than the BICM approach. Further analysis could be performed for different labelings of the MLC scheme, especially for the case of non linearities, where the BL partitioning does not show the desirable behavior of decreasing protection to distortion for increasing level index of the MLC scheme. Another interesting topic of further work could be to analyze the combination of predistorters and BL partitioning. Fig. 9. Mutual information of a classical BICM scheme and the proposed MLC scheme for an HPA with IBO = 3 db. Fig. 10. BER of a classical BICM scheme and the considered MLC scheme for an HPA with IBO = 3 db. satellite system impairments encountered in practical applications has been analyzed. A typical DA channel estimator has been considered with pilot sequences of different length, and its impact on the decoding block has been analyzed. A constant loss of about 1 db has been observed for the simulated SNR range for typical pilot lengths. The residual phase noise at the input of the decoding block has been also taken into account, where a simplified model for simulation of the phase noise has been presented. It has been shown that the residual phase noise has almost no impact on the performance of the FEC block. Finally, the non linearities introduced by the satellite power amplifier have been evaluated using a typical power amplifier model employed in satellite communications. A high performance degradation has been observed, especially for high SNRs. Besides, the considered MLC scheme has been compared with a typical BICM approach proposed by the satellite communication standards [1], [12]. According to this, the MLC scheme exhibits different grades of robustness for each level, where the first level is almost not affected by the REFERENCES [1] Satellite Earth Stations and Systems (SES); Satellite Digital Radio (SDR) Systems; Outer Physical Layer of the Radio Interface, ETSI TS [2] Satellite Earth Stations and Systems (SES); Satellite Digital Radio (SDR) Systems; Inner Physical Layer of the Radio Interface, ETSI TS [3] Digital Video Broadcasting (DVB); Framing Structure, Channel Coding and Modulation for Satellite Services to Handheld Devices (SH) below 3 GHz, ETSI EN [4] Physical layer standard for CDMA2000 spread spectrum systems, Third Generation Partnership Project 2 (3GPP2), Feb [5] H. Imai and S. Hirakawa, A new multilevel coding method using error correcting codes, IEEE Transactions on Information Theory, vol. 23, pp , May [6] D. Schill, D.-F. Yuan, and J. Huber, Efficient hierarchical broadcasting using multilevel codes, in Proceedings of Information Theory and Networking Workshop, [7] A. Vargas, M. Breiling, and W. Gerstacker, Design and Evaluation of a Multilevel Decoder for Satellite Communications, in Proceedings of the International Conference on Communications (ICC 09), Dresden, Germany, June [8] A. Vargas, W. Gerstacker, M. Breiling, and A. Heuberger, Multilevel Codes for Satellite Broadcasting under LMS Channels, in Proceedings of the Vehicular Technology Conference (VTC Fall 10), Ottawa, Canada, Sept [9] E. Zehavi, 8-PSK trellis codes on Rayleigh channel, in Proceedings of Military Communications Conference (MILCOM 89), Oct. 1989, pp [10] E. Lutz, D. Cygan, M. Dippold, F. Dolainsky, and W. Papke, The land mobile satellite communication channel recording, statistics, and channel model, IEEE Transactions on Vehicular Technology, vol. 40, no. 2, pp , May [11] DVB SH Implementation Guidelines, ETSI A120. [12] Digital Video Broadcasting (DVB); Second generation framing structure, channel coding and modulation systems for broadcasting, interactive services, news gathering and other broad-band satellite applications, ETSI EN [13] U. Wachsmann, R. Fischer, and J. Huber, Multilevel Codes: Theoretical Concepts and Practical Design Rules, IEEE Transactions on Information Theory, vol. 45, pp , July [14] R. E. Best, Phase Locked Loops. McGraw Hill, [15] C. Berrou, A. Glavieux, and P. Thitimajshima, Near Shannon Limit Error-Correcting Coding and Decoding: Turbo-Codes (1), in Proceedings of the International Conference on Communications (ICC 93), May 1993, pp [16] D. Pauluzzi and N. Beaulieu, A Comparison of SNR Estimation Techniques for the AWGN Channel, IEEE Transactions on Communications, vol. 48, no. 10, pp , Oct [17] A. Demir, A. Mehrotra, and J. Roychowdhury, Phase Noise in Oscillators: a Unifying Theory and Numerical Methods for Characterization, IEEE Transactions on Circuits and Systems, vol. 47, no. 5, pp , May [18] Digital Video Broadcasting (DVB); User guidelines for the second generation system for broadcasting, interactive services, news gathering and other broadband satellite applications, ETSI TR [19] P. B. M. C. Jeruchim and K. S. Shanmugan, Simulation of Communication Systems. Kluwer Academic/Plenum Publishers, [20] R. Strauss, Reliability of SSPA s and TWTA s, IEEE Transactions on Electron Devices, vol. 41, no. 4, pp , Apr [21] J. Vogt and A. Finger, Improving the max-log-map Turbo Decoder, Electronics Letters, vol. 36, no. 23, pp , Nov