Turbo Coded - QAM Modem One versatile solution for all your needs

Transcription

1 Turbo Coded - QAM Modem One versatile solution for all your needs Sorin Adrian Barbulescu adrian.barbulescu@unisa.edu.au Institute for Telecommunications Research University of South Australia Mawson Lakes, South Australia Abstract Traditionally satellite systems were considered power limited. However the latest trends show that this is changing dramatically and in the very near future the system bottleneck will be bandwidth availability. With the discovery of turbo codes, the operating point for a communication system has decreased by 3 to 4 db. Thus the thermal noise becomes the dominant source of perturbation reducing the system sensitivity to other distortions and interferers. This characteristic is exploited by specific strategies used in turbo coded systems in conjunction with higher order constellations. Today there is at least one commercial satellite system in operation using this technique [1]. This paper describes a single versatile coding solution that can be used with different modulation schemes and optimised for various system requirements e.g., available power/bandwidth, delay, bit error rate (BER), etc. Compared with the best commercial modems on the market, the performance is improved by 2 db in most cases across a wide range of modulation schemes and data rates at BERs less than The impact the low operating BER has on applications such as Internet services over a satellite link is also discussed. 1. Introduction The Inmarsat Aeronautical High Speed Data service (Aero-H) described in [1] is the first commercial satellite service exploiting turbo codes in conjunction with 16QAM. This service operates at 64 kbit/s in an aeronautical fading channel (C/M 10 to 15 db, fading bandwidth of 20 to 100 Hz, differential delays of 10 to 15 microseconds) [1]. The most important conclusion from [1] is that a careful optimisation of transmitter design parameters, including power amplifier back-off, modulator shaping filter and signal constellation allowed the optimisation of power amplifier efficiency while limiting receiver performance loss and side-lobe regrowth, thereby attaining optimal power efficiency while minimising interference effects on adjacent and co-channels. A 3 rd generation mobile communications system via satellite is already under development, as described in [2]. The system targets hand held telephony, data and Internet access to laptop and palm computers with data rates up to 432 kbit/s. The target satellite payload power is 9 kw with 200 spot beams. Average power for systems such as Agrani, Thuraya and ACeS has already increased to 12 kw. Given that designers plan for large flexible solar arrays generating 50 to 60 kw with solar cell efficiencies above 30% [3], satellite systems will soon become bandwidth limited instead of power limited. It becomes necessary to design the satellite transmitter and receiver such that it can easily adapt its modulation and coding architecture to whatever link margin is available, thus achieving maximum efficiency from the user point of view. Section 2 describes briefly the measured performance of the Turbo Codec QPSK Modem developed recently at ITR. Section 3 describes the main issues related to the development of the new Turbo Coded QAM Modem. This represents a natural extension of the previous work and aims to provide a versatile solution to the needs of a power/bandwidth efficient market segment. 2. Turbo Codec QPSK Modem The Institute for Telecommunications Research (ITR) at the University of South Australia has developed its own turbo coding technology to provide significant performance improvements for modern communications systems. The proof-of-concept Turbo Codec QPSK Modem, shown in Figure 1, uses serially concatenated convolutional codes (SCCC) for either rate ½ or ¾ FEC coding. The project was developed under an R&D contract awarded by INTELSAT and executed in collaboration with DSpace Pty. Ltd., and OKI Techno Centre (Singapore) Pty Ltd. The system provides more than 3 db improvement in performance over the INTELSAT IESS-309 (Rev. 7) specifications.

2 The main features are: ❹ Information data rates (kbit/s): 64, 128, 256, 512, 1024 and 2048 ❹ Bit error rate : < ❹ E b /N 0 for rate ½: < 2.0 db ❹ E b /N 0 for rate ¾: < 3.5 db ❹ 2048 kbit/s: < 30 ms The modem can be configured via a Graphical User Interface (GUI) from a laptop. This feature makes monitoring and controlling tasks very user friendly. The terrestrial interfaces are: RS 422, RS 232, NRZ and 70 MHz IF for the modulated signal. It is housed in a single 19 x 3U unit with a universal power supply. Figure 1 Turbo Codec QPSK Modem The motherboard contains all the modulator and demodulator functions together with the microcontroller. The turbo codec is designed as a daughter board for future upgrades. It is based on a parallel implementation with up to sixteen blocks being MAP decoded simultaneously. Specific delay reduction techniques have been used as described in [4]. The hardware has been successfully tested over an INTELSAT VI simulator hemi transponder at INTELSAT headquarters in Washington, US. Live satellite tests were performed in Adelaide, Australia, over the Optus B3 satellite. 3. Turbo Coded-QAM Modem ITR has recently installed a very small aperture terminal (VSAT) network that operates in conjunction with the MEOSAT satellite. This platform enables ITR to conduct live tests over a satellite channel in order to verify and fine tune the performance of turbo coding schemes designed for satellite applications. It is also a useful tool to study non-linearities and other distortions introduced in a satellite link. At the core of this network is the Turbo Coded-QAM Modem. This is a pre-engineering model that can be easily integrated in any existing VSAT network in order to increase the bandwidth/power efficiency of the system. However, greater advantages can be achieved by using the built-in network adaptor module. Turbo codes combined with higher order modulations are not just an upgrade of convolutional coded modulated schemes. There is a misconception that 16QAM, for example, will behave similarly when traditional convolutional/reed-solomon codes are replaced with turbo codes. It is well known that QAM signals exhibit a higher sensitivity to phase noise and amplifier non-linearity than QPSK signals. This nonlinearity can be seen in the AM/PM and AM/AM transfer characteristics of typical travelling wave tube amplifiers (TWTA). This is why these amplifiers should be operated with about 7 db output back-off from the saturation point [5]. The specific issues for turbo coded QAM can be summarised as follows: (a) The low operating point of turbo coding schemes makes the system less sensitive to Carrier to Interferer (C/I) ratio due to adjacent channel interference (ACI) or co-channel interference (CCI). The reason for this is that the thermal noise is the dominant source of system degradation. This can be seen in Figure 2 of [6], showing that the loss versus SNR for various C/I ratios is significantly reduced due to the lower operating point of the turbo coded system. (b) Higher order modulation systems usually require a larger back-off than constant amplitude modulation schemes in order to maintain the operating point of a High Power Amplifier (HPA) in the linear region to avoid saturation. Due to the specific mapping of the information bits, a turbo coded system can sustain a higher level of distortion applied to the 16QAM constellation than a conventional convolutional code. As shown in Figure 1 of [1], the 16QAM constellation can sustain significant distortion with the outer points becoming almost circular, without any penalty in performance. This is translated directly in a more efficient use of the HPA, with smaller back-off than for the traditional 16QAM systems based on convolutional codes. This is an extremely important consideration, which is specific to the ability of turbo codes to operate in a very low signal-to-noise ratio environment. 3.1 Modem / Codec Performance This new Turbo Coded-QAM Modem will handle various modulation schemes, e.g., QPSK, 8PSK, 16QAM and 64QAM, using the same SCCC scheme and one or two transmitter/receivers.

3 Eb/No [db] ITR EFData TPC 0 QPSK Rate 1/2 QPSK Rate 3/4 16QAM Rate 1/2 16QAM Rate 3/4 Figure 2 Turbo Coded QAM Modem performance for BER = The modem will operate at data rates in the 20 Mbit/s range, in single or multichannel decoding mode. Delay sensitive applications will be serviced for specified maximum programmable delay. For data transfer applications a maximum coding gain mode can be defined by increasing the interleaver size and the number of iterations. The design is scalable such that if 8 iterations are set for 20 Mbit/s, 16 iterations can be achieved at 10 Mbit/s and so on, up to 1000 iterations at 64 kbit/s data rate. The slight change in name from turbo codec to turbo coded signifies the feedback loop from the decoder to the demodulator, which is included in the current architecture. This option allows testing of turbo equalisation algorithms as well as improved modem synchronisation strategy using the soft decoded outputs. Simulation results for the 16QAM configuration are shown in Figure 2. They outperform the COMTECH EFData SDM 8000 performance specifications by more than 2.5 db in an AWGN channel. Turbo product codes (TPC) were used in the CDM 600 modem, only for QPSK. The rate achieves the lowest specified BER of 10-8 at 3.0 db (typical value). Extrapolating this curve based on 0.2 db decrease in BER per order of magnitude, the performance could be achieved at 3.4 db. This is 1.4 db more than required by the ITR solution. The rate ¾ TPC achieves the lowest specified BER of 10-8 at 4.0 db (typical value). Extrapolating this curve based on 0.3 db decrease in BER per order of magnitude, the performance could be achieved at 4.6 db. It is clear from Figure 2 that the ITR s SCCC solution achieves again more than 1 db coding gain than the TPC solution. 3.2 Sliding Window Decoding Algorithm A sliding window (SW) algorithm was introduced in [7] and later on in [8]. Since no tail can be appended for each window due to a significant reduction in throughput, a buffer space was defined where the unreliable decoded data is simply ignored in the current window and left to be decoded again with a higher reliability in the following window. The solution described in [9] ensures that the penalty in using a sliding window decoding algorithm is less than 0.05 db for no buffer, see Figure 3, and negligible for 4 or 8 bit buffer size. The advantages of using the SW are reduced complexity / memory requirements and higher decoding speeds in field programmable gate arrays. The INTELSAT IESS-308 standard specifies that a 72 MHz C-Band transponder must be allocated in the case of two 45 Mbit/s Rate ¾ FEC QPSK/IDR carriers. Recent tests performed between the Teleglobe earth station at Lake Cowicham, Canada and the Telstra earth station in Sydney, Australia proved that three 45 Mbit/s 16QAM carriers can be sent through a 72 MHz INTELSAT transponder [10]. Thus the throughput could be increased from 90 Mbit/s to 135 Mbit/s. Initial simulations have shown that using the Turbo Coded QAM Modem the throughput can be increased up to 180 Mbit/s at a target BER of

4 E E E-02 BER 1.0E E E-05 SW400B0 SW400B4 SW400B8 continuous 1.0E E E-08 Eb/No [db] Figure 3 Performance of the Sliding Window algorithm 3.3 Internet over satellite It is well known that to include satellite links as part of the overall Internet requires breaking the network (spoofing). This is necessary in order to interface the Transmission Control Protocol (TCP) with a unit which solves critical issues like: slow start, large windows, delayed ACKs, fast recovery, fast retransmit, etc. This enhancer unit or interworking unit is usually a proprietary solution for a specific system, e.g., Astrolink, Loral Cyberstar, Skybridge, Spaceway, Freedom IP, isky, Teledesic, etc [11]. All the above schemes use convolutional codes and deal only with protocol issues designed to solve just one part of the problem. The optimum solution resides in finding the best system configuration that suits the particular application required by the user. The Turbo Coded QAM Modem interface with the VSAT network will be used for developing strategies where the source coding, channel coding, modulation, and possibly network parameters are jointly determined to yield the best end-to-end system performance. This joint optimisation is transparent to the user, e.g., the system could be optimised for speed when Internet browsing, whereas downloading a file would reconfigure the system for maximum coding gain. 3.4 Multiple Tx/Rx System The current Turbo Coded QAM Modem architecture provides for two independent transmitters and two independent receivers simultaneously. The purpose of this design is two fold: ❹ Test the performance of various diversity techniques and joint decoding for different jamming scenarios, ❹ Investigate the performance of space-time coding to increase throughput in a fading channel. There are many applications in which diversity combining techniques are used to improve system performance. The iterative decoding techniques can be used against fading channels or as protection against jamming strategies. The reason for improved performance is the interleaver embedded in the encoder, which acts, at the receiver end, as a fade decorrelator. With turbo code like systems reaching channel capacity, the only way forward is to move the capacity. This is possible using multiple antennas, the capacity growing linearly with the number of transmit antennas (it is assumed that the number of receiving antenna is at least equal with the number of transmit antennas) [12].

5 The Turbo Coded QAM Modem will be used as a test bed to implement a two antenna transmitter / receiver for higher order modulation schemes. Further code design will aim to ensure that the additional diversity advantage obtained using multiple tx/rx antennas is fully exploited. 4. Conclusions This paper described technical issues related to a new platform based on the Turbo Coded QAM Modem. While specific performance parameters, e.g., BER, were shown to outperform other technologies, the main advantage resides in the flexibility of the hardware to switch between different data rates, coding rates and modulation schemes. The potential of joint optimisation of different layers in the communication system is also possible. The flexibility of a serially concatenated convolutional code that can be packed in any block size is a definite advantage over fixed length codes like low density parity check codes or turbo product codes. This is crucial for operation in both continuous and burst mode of variable length, e.g., demand assigned multiple access mode. The user is interested only in the application layer. The system must be smart enough to reconfigure itself in order to achieve maximum benefit for the application the user is running. For example, a delay insensitive data file transfer can use a 64 kbit interleaver that gives much higher coding gain than a 400 bit interleaver that could be used for an ATM cell size application. The performance of the SCCC scheme described here is with more than 1 db more power efficient than the nearest competitor. However, the true advantage of a system operating at a BER less than is not measured in db but in network throughput. For successful TCP transfer, a BER lower than 10-8 is required otherwise packets are lost and the TCP layer assumes that congestion is at fault, thus, slowing the throughput. This is why the SCCC scheme was specifically designed to work efficiently at this low BER, close to that of a terrestrial network. The performance of the solution for the physical layer was proven by the proof-of-concept hardware built at ITR, University of South Australia. It was shown that it can achieve a BER less than at E b /N 0 less than 2.0 db for a rate ½ code. A more advanced, pre-engineering version, was also described from the point of view of its performance and potential applications. The Turbo Coded-QAM Modem is more than an optimised hardware solution. It integrates the communication system from the tip of your fingers all the way down. 5. References [1] E. Trachtman and T. Hart, Research Elements Leading to the development of Inmarsat s new mobile multimedia Service, Sixth international mobile satellite conference, Ottawa, Canada, June [2] M. Franchini, Wideband multimedia over versatile satellites Astrium solutions, IEE Conference Broadband satellites: the critical success factors, London, UK, pp.24/1-24/8, Oct [3] J. N. Pelton and A. U. Mac Rae, Key trends in the field of global satellite communications, Space Communications 16 (2000), pp.71-84, [4] S. A. Barbulescu and J. Buetefuer, Practical Delay Reduction Techniques in Serial and Parallel Concatenated Schemes (Turbo Codes), The 22 nd Symposium on Information Theory and its Applications (SITA 99), Niigata, Japan, pp , Nov [5] H. Stewart and R. Cannon, Advanced Modulation Techniques for Digital Satellite Modems, PTC 96, pp , [6] M. Rice, P. Gray and S. A. Barbulescu, Coding and Modulation Techniques for High Speed Data Services by Satellite, ICICS 97, Singapore, Sep [7] S. A. Barbulescu, Iterative decoding of turbo codes and other concatenated codes, University of South Australia, PhD Dissertation, Aug [8] S. Benedetto, G. Montorsi, D. Divsalar and F. Pollara, Soft-output decoding algorithms in iterative decoding of turbo codes, TDA Progress Report , Jet Propulsion Laboratory, Pasadena, California, pp.63-87, Fe [9] S. A. Barbulescu, On Sliding Window and Interleaver Design, submitted to Electronics Letters, July [10] Contribution of the Signatories of Australia, Canada & USA, Report on testing of 16QAM services, BG/T Inf E, Aug [11] D. Roddy, Satellite Communications, McGraw- Hill TELECOM Engineering, [12] V. Tarokh, N. Seshadri and A. R. Calderbank, Space-Time Codes for High Data Rate Wireless Communication: Performance Criterion and Code Construction, IEEE Transactions on Information Theory, Vol.44, No.2, pp , March 1998.