Convergence of Hardware and Software in Platforms for Radio Technologies

Transcription

1 NETWORK-CENTRIC MILITARY COMMUNICATIONS Convergence of Hardware and Software in Platforms for Radio Technologies Jerker Björkqvist, A bo Akademi University Seppo Virtanen, University of Turku ABSTRACT Technologies used in wireless broadband systems are typically data-driven and require a very high processing speed. In this article we discuss technological issues that could be utilized in designing converged hardware/software platforms for future radio technologies in both civilian and military communications. We focus on current and future radio applications as well as hardware and software technologies, and discuss issues in integrating these technologies into converged single-processor radio platforms that can switch dynamically from one radio standard to another with intelligent application software, and that can also be taken advantage of as parts of complex systems and networks-on-chips. INTRODUCTION The current trend in telecommunications is to provide wireless broadband data services for both static and mobile users. Once the data service is available, the data itself can be used for a multitude of different applications, of which Digital Video Broadcasting-Terrestrial (DVB-T) and Wireless Local Area Networks (WLAN) are currently the best known. Novel systems and services are constantly developed for mobile users; for example, DVB-H (DVB for handheld devices) enhances an existing system for mobile use. There are quite a few activities in the area of using WLANs for mobile real-time usage, such as Voice over Internet Protocol (VoIP) calls, video broadcast, or a combination of both (e.g., video conferencing or peer-to-peer video calls). A very significant characteristic of this development is the emergence of free, fully functional software utilities for establishing such services between the interested parties. Broadband data-delivery systems are used for carrying standard Internet Protocol (IP) data, in which case often the term IPDC, or IP data casting, is used. Another trend is to provide devices that are lightweight, mobile, power effective, and capable of handling multiple radio interfaces, for example, multiband-gsm, 3G, DVB-H, WLAN, Infrared, and Bluetooth in the same device. At the same time, these handheld devices are supposed to handle increasing streams of data and are required to do more processing, as the complexity and number of the modulators, codecs, and protocol layers increase. This immediately raises the question of hardware complexity: does each radio interface require its own chip or IC and related software, or would it be possible to use a single programmable processor that could be configured by its application software on-the-fly to switch from the radio interface defined by one standard to an interface defined by another one? The traditional solution of using generalpurpose programmable microprocessors with accompanying application-specific chips (application-specific integrated circuits, or ASICs) is neither an efficient nor a flexible enough solution to handle the increasing demands. SoC and NoC devices face the same problems if the intellectual property blocks incorporated into the devices are fixed or minimally programmable/configurable. The concepts of software-defined radio (SDR) and open wireless architecture (OWA) are recent attempts to provide an answer to this problem. With SDR the goal is to design systems that use their application software for reconfiguring the hardware from one radio interface or application to another without modifying the hardware or software components in the system. Ideally, there would be software modules for different radio applications running on a softwareconfigurable hardware platform. OWA extends this idea further by attempting to provide a generic (even standardized) and flexible hardware platform specification, or application programming interface (API), for which operators and vendors could provide software modules for novel radio applications. In this article we focus on discussing technological issues and potential solutions that could be utilized in designing converged hardware/software radio platforms for future technologies both in commercial and in military communications. We focus on current and future radio applications (with an emphasis on OFDM-based applications due to OFDM s robust non-line-of-sight radio capabilities) as well as hardware and software technologies, and we discuss the issues in integrating these tech /06/$ IEEE IEEE Communications Magazine November 2006

2 nologies into converged single-processor radio platforms that can be programmed to switch on-the-fly from one radio standard or application to another. TECHNOLOGY Data in FEC encoder Digital domain Modulation D/A RF upconvert In this section we discuss technologies in the key areas related to designing converged hardware/software radio platforms for future civilian and in military communications. We focus on radio technologies, encryption technologies, and hardware/software platform technologies. Transmitter Receiver Data out FEC decoder Demodulation D/A Figure 1. General block diagram of digital radio systems. RADIO TECHNOLOGY Radio technology has been dependent on hardware implementations of most radio systems for a long time. The most basic modulation for transmitting speech has been amplitude modulation, where the amplitude of the instantaneous air displacement is added to the amplitude of a carrier signal, which in turn is transmitted using electromagnetic waves. This is the system of the AM radio still used mainly for civilian long-distance radio transmitters. However, issues like quality of sound, spectrum utilization efficiency, energy used per amount of transmitted information and digital transmissions have lead to a number of new techniques for transmitting information. Frequency modulation (FM) improves the quality of analog transmissions and has been used for a number of radio communication systems, such as civilian FM radio and VHF radio systems. Currently, radio systems are converging to digital systems, where a digitized data stream is sent over a radio link. This digital data stream might correspond to voice audio, video signals, data, and so forth. The data can also be encrypted using any encryption system, hence providing an eavesdrop-safe communications channel. The GSM mobile phone system efficiently uses many of these techniques to provide safe mobile communication. Not only are the data sent over the network being digitalized, but also the actual radio device modulating the data for transmission over the air. An increasing portion of the receivers is being designed using pure digital technology, using either application-specific integrated circuits (ASICs), digital signal processors (DSPs), field-programmable gate arrays (FPGAs), or a combination of these and general-purpose processors. In these solutions, only the radio frequency parts work using analog technology: the intermediate frequency (IF) is fed into a fast analog-to-digital converter (ADC), and the rest is handled digitally. Figure 1 shows a functional block diagram of a digital radio system. OFDM Orthogonal frequency division multiplexing (OFDM) is becoming the chosen modulation technique for wireless communications. OFDM can provide large data rates with sufficient robustness to radio-channel impairments. The attraction of OFDM is mainly due to how the system handles the multipath interference at the receiver. In OFDM, a large number of subcarriers are spread in the available bandwidth. The carriers are orthogonal, which means that the subcarrier separation is theoretically minimal, giving a high spectral utilization. OFDM efficiently solves two problems in multipath radio communications: frequency-selective fading, meaning that some parts of the used bandwidth are faded; and intersymbol interference (ISI), meaning that two consecutive symbols overlap in time. The former problem is solved by the division into subcarriers, as these correspond to narrowband channels, where as the intersymbol interference is reduced by the low symbol rate. In practice, a guard interval is also inserted between symbols in order to further decrease the effects of ISI. Additionally, OFDM systems are combined with forward error correcting (FEC) RF downconvert IF A/D FFT Channel estimate/ correct Symbol demapper Soft decision slicer Deinterleaver Viterbi decoder RS decode Descramble Figure 2. Functional blocks of a WiMAX radio receiver. IEEE Communications Magazine November

3 DVB-T DAB 1 WiMax 2 WLAN 3 UWB 4 Bandwidth 6, 7, 8 MHz 1.5 MHz MHz 16.6 MHz 528 MHz Center frequency 700 MHz 200 or 1400 MHz 2 6 GHz 5 GHz 3 10 GHz No. of carriers 2048, Modulation BPSK, QPSK, QAM16, QAM64 DQPSK BPSK, QPSK, QAM16, QAM64 BPSK, QPSK, QAM16, QAM64 QPSK FEC Convolutional + RS Convolutional Convolutional + RS Convolutional Convolutional Throughput 5 31 Mb/s 1.6 Mb/s Mb/s 1 11 Mb/s Mb/s Bit/s/Hz Range > 100 km 100 km 5 km 200 m 2 10 m 1 Eureka IEEE IEEE a 4 IEEE a proposal by Intel Table 1. Parameters of some common OFDM radio systems. techniques. In FEC codecs, the original data is in the transmitter end and is fed into an encoder, which adds redundancy to the signal according to mathematical rules. This redundancy can be used in the receiver for detecting and correcting errors that might have been introduced during transmission. The FEC is also augmented by data interleaving, which in turn spreads out errors so that the FEC codecs are better used. As errors often occur as bursts of errors, but the error-correcting codes can only handle a limited number of errors at a time, the interleaver makes sure that bursts of errors are spread out. Figure 2 shows the functional blocks of a WiMAX radio receiver. The same blocks can be identified in most OFDM-based radio systems. OFDM technology is today used in the European Digital Video Broadcasting (DVB-T/H) standard, Digital Audio Broadcasting (DAB), Asynchronous Digital Subscriber Lines (ADSL), Wireless Local Area Networks (WLANs), Ulta Wide Band (UWB), WiMAX, and so on. Cellular 4G systems are clearly driven towards OFDM technology, on the cost of Code Division Multiple Access (CDMA) systems used in 3G radio communications. In Table 1 the main parameters of some OFDM systems are shown. ENCRYPTION TECHNOLOGY At the same time as the world is getting smaller, thanks to the integration of networks and provision of wireless access to almost any resource at any time, the need for guaranteed secure data transmissions is also increasing. Even if much of the information moving in the networks is public, there is always information that for various reasons must be made unavailable to unauthorized actors on the network. The traditional wired networks could basically be made totally secure by ensuring that nonauthorized actors never have physical access to the network. However, this is seldom the case, and in a wireless environment, basically anyone near enough can eavesdrop on the communication. Hence, methods are needed for making it impossible, or very hard, for eavesdroppers to actually use the data being sent over the communications link. This is the basic encryption and decryption problem. Person A wants to send a message to person B, making sure that potential listener C cannot make use of the message. The traditional solution is to apply an encryption scheme to the message, so that an encryption key is needed to encrypt the original message to a cipher text, which only can be decrypted using a decryption key. If the encryption key and decryption key are the same, the scheme is called a symmetric encryption system. If they are different, it is called an asymmetric encryption system. Two fundamental problems remain in designing encryption systems; How to make the encryption system safe enough, and how to distribute the necessary keys. Discrete mathematics has provided tools for construction and analysis of encryption systems. Currently, however, the length of the encryption key is the dominant parameter: the longer the key, the safer the system is. The drawback with long encryption keys is that the computational power required for encryption and decryption increases with key length. This provides motivation for exploring hardware-based encryption and decryption solutions. There have been various implementations of hardware-based solutions, but the solutions are often stiff and constructed for certain decryption systems. Here it would be advantageous to have a reconfigurable encryption system. The second problem, exchanging keys, is often solved by adapting additionally a Public Key Infrastructure (PKI) encryption system. In this system, there are always two encryption keys involved, the public key and the secret, or private, key. They are used to form an asymmetric encryption system, where a message encrypted 54 IEEE Communications Magazine November 2006

4 with the public key can only be decrypted with the private key and vice versa. Now the advantage is that the public key can freely be distributed to anyone interested in communicating with the owner of the corresponding private key. When a message is encrypted with the public key, only the owner of the corresponding private key can decrypt the message. By sending messages encrypted with the other party s public key, two persons can securely communicate with each other. The key length of public-key systems is often considerably longer than that of symmetric keys, because getting access to someone s private key, for instance, by exhaustive key search, invalidates the security of the system. The PKI-encryption system can also be used to solve another problem: How can I be sure that I am communicating with the right person? This can be solved by sending a challenge from A to B. The challenge is normally a random message. If person B encrypts the challenge with his private key, decrypting this response with B s public key should unveil the original challenge. Failing to generate the appropriate response means that we are not engaged in a discussion with the right person. A PKI system using long keys efficiently solves both of the abovementioned problems with encryption systems. Including hardware support for PKI infrastructure using long key lengths would facilitate a very secure way of communicating over wireless systems. SDR Software-defined radio (SDR) is the term for software and hardware technologies that enable reconfigurable radio systems. SDR-enabled devices can be dynamically programmed to reconfigure the characteristics of the equipment. This means that many systems can use a common hardware platform that is configured using software to function in a specific way. Provided there is a well-defined platform architecture, it is easy to implement new radio systems in a robust way. A well-defined platform should provide both the functional blocks that the radio system requires (as a reconfigurable on-chip communication system) and enable data transports between the functional blocks. Current OFDM systems use functional blocks of certain types. The FEC codes commonly used are convolutional coding (commonly using Viterbi decoders) and block coding (using Reed Solomon codecs). The actual OFDM modulation is performed using fast Fourier transforms (FFTs) and Inverse FFTs. The interleaving mechanisms mainly require some memory and logic. The differences in the coding schemes of current radio technologies are such that hardware implementations are achievable using parametrizable functional blocks instead of dedicated blocks for each scheme. Similarly, encryption systems use functional blocks of certain types, which by parametrization can be applied to different standards. Hence, the technologies used for providing the two essential parts of a radio communication system, an efficient data stream (OFDM) and securing the data stream (encryption), are both good components for the flexible radio platform. PLATFORM TECHNOLOGY In designing small handheld communication devices for multiple radio technologies, the traditional solution of using general-purpose programmable microprocessors (CPUs) with accompanying application-domain-specific ASICs is neither an efficient nor a flexible enough solution to handle the increasing demands. The lack of flexibility in such systems is mostly due to the often minimal programmability of ASICs: they are designed for a specific application, and are either minimally or not at all capable of performing other similar applications. In addition to the lack of flexibility in such systems, devices consisting of general-purpose processors and accompanying ASICs may require excess power (depending on the CPU used) or may cause unnecessary enlargement of overall device size due to the area occupied on the circuit board. One suggested solution to this is to use FPGAs that make true hardware reconfigurability possible. Ideally, with FPGAs it may be possible to store multiple hardware configurations in the device memory or download new hardware configurations when necessary. In this way, a new configuration could be brought into use whenever the system needs to switch from one application or wireless networking technology to another (e.g., from WLAN to 3G or vice versa). The clear benefit of this technology is that overall device size could be very small, since the application or technology-specific circuitry (separate ASICs for each technology) would be replaced ideally by just one reconfigurable circuit. However, the on-the-fly reconfiguration times of FPGAs, as of today, are not fast enough to facilitate a roaming user switching from one radio network to another. Also, depending on the target hardware technology and FPGA device, the achievable processing speed of the FPGA implementation may be considerably lower than for ASICs designed for the same application of technology. According to our studies (see, e.g., [1]), a complex transport triggered platform could be implemented as an ASIC running at over 200 MHz, while the FPGA implementation would reach clock speeds of less than 50 MHz for the same application using a 150 MHz FPGA device due to the data transport and interconnectivity characteristics of the implemented transport triggered platform. Whether resorting to ASICs or FPGAs as cocircuitry, the requirements of the target wireless technology are still likely to require the CPU to run at a very high clock speed, thus consuming more power and draining the power source quickly. Taking this and the aforementioned limitations of ASIC and FPGA technologies into account, in our opinion the research emphasis should be on exploring the concept of application-domain-specific hardware (processor) platforms. In such platforms the executional units are optimized for a particular family of operations in the target application domain, for example, wireless networking. The executional units should be designed so that each unit supports multiple similar operations of the domain and the functionality chosen for execution is deter- Provided there is a well-defined platform architecture, it is easy to implement new radio systems in a robust way. A well-defined platform should provide both the functional blocks that the radio system requires and enable data transports between the functional blocks. IEEE Communications Magazine November

5 Program memory Generic registers Input and output socket connections Interconnection network controller Interconnection network User data memory ummu Input FU Output FU dmmu Protocol data memory Figure 3. A functional view of the TACO architecture. Data packets and host interface mined by the application software. While programmable in their own domains, due to the application-domain optimization of such hardware platforms, they will not perform well in other application domains (for example, a hardware platform designed for processing network data packets would not be efficient in calculating DSP algorithms). However, it is also possible to design a multidomain-optimized hardware platform, as demonstrated in [2], with a platform designed for both the protocol processing and the DSP domains. Still, the important question that must be raised here is how to determine and group the similar operations of the target application domain to be implemented into executional units. In our view, this requires development of application analysis methods, such as those presented in [3], for example. With such methods, both the application domain and a given target application can be analyzed to determine the operations that should be grouped and implemented into executional units. However, to accomplish this, the basic underlying hardware solution must first be specified. As an example, the recently developed transport triggered architecture (TTA)-based [4, 5] TACO hardware platform [6] serves as a basis for developing application-domain specific processors. In TTAbased processors, operations are triggered by programmed data transports. This is contrary to the traditional approach, in which programmed operations cause data transports to occur. A TTA-based processor is composed of functional units (FUs) that communicate via an interconnection network of buses. The FUs are connected to the buses through modules called sockets. In the TACO hardware platform each FU performs a group of related processing tasks or operations of the target application domain. The initial focus has been on developing FUs for protocol processing [6], but multidomain optimization (running DSP and protocol processing in parallel) [2] and software-defined radio [7] have also been targeted recently. A functional view of the TACO hardware platform is given in Fig. 3. CONVERGENCE OF PLATFORM ARCHITECTURES AND RADIO TECHNOLOGIES In the previous sections we have presented the technological background for recent OFDMbased radio systems, an introduction to encryption methods, the general trends of trying to provide reconfigurable radio platform, and an emerging TTA-based hardware platform for implementing domain-specific applications. Looking at the requirements for the technologies providing radio communication and the possibilities that a TTA-based hardware platform provides, a clear match can be seen. Radio technologies need certain types of operations, such as FFT, IFFT, convolutional codecs, encryption ciphers, Reed Solomon codecs, modulators, demodulators, and so forth. In TTA-based platforms the devices are constructed of functional units that are interconnected using a number of buses. It would be quite logical to build a set of functional units supporting the typical operations of recent radio technologies. If these FUs are well defined and parameterized, they could provide the classes of operations needed. Hence, by building one flexible hardware platform and applying a software-based configuration to the platform, a specific radio system can be implemented. Moreover, depending on the capabilities of the platform, it could simultaneously provide two or more radio systems. This seems very attractive compared to the alternative of providing one radio module for each radio system implemented in a specific device. Currently, one concern in the mobile phone business is the energy consumption of the devices. The traditional way of building systems, using general-purpose processors connected to specific hardware modules, is leading to systems with poor energy efficiency. The need for increasing the number of radio technologies in the same devices leads to a situation where several radio modules implementing the required radio systems are attached to the device. As a result, both energy and silicon area are unnecessarily used in the device, and many of the implemented functionalities are redundant in one way or another. Using a strategy of providing one configurable platform with the needed FUs would decrease the redundancy and make better use of the silicon attached to the device. By optimizing the FUs and communication paths, energy and area can be further saved. This kind of approach is depicted in Fig. 4. At this stage, the reasoning is mostly theoretical, as both SDR and TTA are emerging as means of providing efficient devices to radio systems. However, initial studies have already been performed to validate the use of TTA in implementing a DVB-T system [7]. The work is 56 IEEE Communications Magazine November 2006

6 focused on designing and implementing functional units that could be generalized for a variety of radio applications and systems such as those described earlier in this article. The initial focus was to build a DVB-T receiver on our TTA platform. The designed and implemented functional units were for demapping, depuncturing, and bit, symbol, and convolutional interleaving. Synthesis of these units was successfully done, targeting over 200 MHz on a 0.18 µm CMOS process. At this clock speed, the throughput of the platform for the DVB application was clearly more than sufficient, since our first tests suggest that the TACO TTA platform requires a clock speed of less than 30 MHz for the DVB receiver application. Another conclusion reached in [7] was that once the basic functional units were defined, it was easy to incorporate support for other, originally unplanned, radio systems into the platform as well. CONCLUSION The emerging variety of radio systems is a challenge for radio-device developers, as development of specific radio modules requires time. The trend is to adopt SDR methodologies, where a general hardware platform is configured to provide the required system and functionality. Even when targeting a broad bouquet of radio systems, several basic operations on which most of them rely (whether parameterizable or not) can most certainly be found. In this article the emphasis has been on operations derived from OFDM-based technologies. These common basic operations can be classified into a limited number of classes, which again can be specified as functional units in a TTA-based platform design. This yields a system which provides the flexibility, energy efficiency, and reusability that will be crucial when developing the radio devices of the future. REFERENCES [1] S. Virtanen et al., Highly Automated FPGA Synthesis of Application-Specific Processors, Proc. 15th Int l. Conf. on Field Programmable Logic and Applications (FPL2005), Tampere, Finland, [2] J. Paakkulainen, S. Virtanen, and J. Isoaho, Tuning a Protocol Processor Architecture Towards DSP Operations, LNCS 3553 (SAMOS V), Berlin: Springer-Verlag, 2005, pp [3] D. Truscan, J. M. Fernandes, and J. Lilius, Tool Support for DFD-UML Model-based Transformations, Proc. Int l. Conf. Engineering of Computer-Based Systems (ECBS 04), Brno, Czech Rep., [4] H. Corporaal, Microprocessor Architectures from VLIW to TTA, Chichester, England: Wiley, DVB-T WiMAX DAB... FU1 FFT Platform defined adaptation rules FU2 RS Figure 4. Using a TTA platform methodology for developing radio systems. [5] D. Tabak and G. J. Lipovski, MOVE Architecture in Digital Controllers, IEEE Trans. Comput., vol. 29, no. 2, 1980, pp [6] S. Virtanen et al., A System-Level Framework for Designing and Evaluating Protocol Processor Architectures, Int l. J. Embedded Syst., vol. 1, nos. 1 2, 2006, pp [7] M. I. Anwar and S. Virtanen, Mapping the DVB Physical Layer onto SDR-Enabled Protocol Processor Hardware, Proc. 23rd IEEE NORCHIP Conf., Oulu, Finland, BIOGRAPHIES Software/configuration Platform FU3 Conv. JERKER BJÖRKQVIST received an M.Sc. in process control in 1996 and a D.Sc. (Tech.) in process design in 2002 from Åbo Akademi University, Finland. Since 2002, he has been an assistant professor at the Faculty of Technology, Department of Information Technologies. His research interests include digital television (DVB-H), error-correcting coding, and optimization. SEPPO VIRTANEN (seppo.virtanen@utu.fi) received a B.Sc. in applied physics and an M.Sc. in electronics and information technology in 1998, and a D.Sc. (Tech.) in communication systems in 2004, from the University of Turku, Finland. Since 2005 he has been a senior lecturer in the Department of Information Technology of the University of Turku. His research interests include hardware/software codesign, domain-specific sysftem-level design, communication systems, and communication protocols. FUn Radio system Software Hardware IEEE Communications Magazine November