A Real Time Cognitive Radio Testbed for Physical and Network level Experiments

Transcription

1 A Real Time Cognitive Radio Testbed for Physical and Network level Experiments Shridhar Mubaraq Mishra, Danijela Cabric, Chen Chang, Daniel Willkomm, Barbara van Schewick, Adam Wolisz and Robert W. Brodersen School of Electrical Engineering and Computer Science University of California, Berkeley, California Telecommunication Networks Group, Dept. of Electrical Engineering, Technical University of Berlin, Berlin, Germany Abstract Cognitive Radios have been advanced as a technology for the opportunistic use of under-utilized spectrum. However, Primary users of the spectrum have raised concerns with regards to interference from Cognitive Radios. On the other hand, a variety of techniques have been proposed for reliable sensing and non-interfering use of the spectrum which have yet to be validated in an actual system. In this paper we present a testbed that will allow us to experiment with sensing algorithms and to demonstrate a working prototype of an indoor cognitive radio network. The testbed is based on the BEE2, a multi-fpga emulation engine which is capable of connecting to 18 radio front-ends. The testbed will be used to experiment with various baseband sensing algorithms and cooperative sensing schemes. I. INTRODUCTION It is commonly believed that there is a spectrum scarcity at frequencies that can be economically used for wireless communications. This concern has arisen from the intense competition for use of spectra at frequencies below 3 GHz. As seen in Figure 1, the Federal Communications Commission s (FCC) frequency allocation chart indicates multiple allocations over all of the frequency bands, which reinforces the scarcity mind set. On the other hand, actual measurements taken at the BWRC (see spectrogram in Figure 2) indicate low utilization especially in the 3-6 MHz bands. This view is supported by recent studies by the FCC s Spectrum Policy Task Force (SPTF) which reported vast temporal and geographic variations in the usage of allocated spectrum with utilization ranging from 15% to 85% [1]. Proposals to encourage efficient use of the spectrum have focused on introducing secondary users into frequency bands already allocated to Primary users (a Primary user of a frequency band has been allocated the band by the FCC). Here we list three specific proposals to accomplish this: Negotiated Spectrum Sharing: Under this regime, secondary users of the spectrum can negotiate usage of the spectrum from the Primary user. Specifically, Primary users may need to introduce a beaconing scheme to signal intent to reclaim the channel. Negotiated spectrum sharing requires economic incentives to persuade Primary users to modify Fig. 1. FCC spectrum allocation chart their equipment for beaconing their availability. Receiver Interference Announcement: This regime is derived from the interference temperature proposal by the FCC [1] and assumes that there is no power constraints on secondary transmitters. Primary receivers measure interference and announce when the received interference becomes unacceptable. Acceptable interference is a function of receiver quality. A poor receiver will signal interference early, a behavior which run counter to the aim of introducing efficient use of the spectrum. Opportunistic Spectrum Sharing via Cognitive Radios: Secondary users must sense the presence of a Primary user and use the spectrum only if the Primary user is not detected. Secondary radios that can sense the spectrum are called Cognitive Radios (CR radios). The FCC has issued a Notice of Proposed Rule Making [2] in which it has advanced Cognitive Radio technology as a candidate to implement opportunistic spectrum sharing. Cognitive Radios offer the possibility of efficiently reusing the spectrum without modification to Primary user equipment.

2 Inability to determine channel to a Primary user: Knowing the presence of a Primary transmitter, still leaves open the problem of knowing the location of the corresponding receivers. This is especially true in the broadcast case where the receivers are passive. In such a case, the Cognitive Radio must be certain that it is far enough so that its transmission cannot interfere with any receiver at the edge of the Grade B contour of the TV reception region [3]. Even if the position of the Primary receivers was known, the channel to the Primary receivers cannot be determined since there is no feedback path from the Primary receiver to the secondary transmitter. Fig. 2. Spectrum use of 0-2GHz frequencies over 10mins A. Challenges in implementing Opportunistic Spectrum Sharing In order to use spectrum in an opportunistic manner a Cognitive Radio must be able to demonstrate usage with no or minimal interference to the Primary user. This task is rendered difficult due to the following physical constraints: Difficulty in sensing the spectrum reliably: If a Cognitive Radio does not see energy in a particular band can it assume that the Primary user is not present? Answering this question is difficult since a secondary user may suffer severe multipath and/or shadowing. For example, in the TV regime, a TV receiver may be elevated (top of the roof) and hence may have better reception as compared to the Cognitive Radio which may be at the ground level. Furthermore, the Cognitive Radio may be inside a building while the TV antenna is outdoors. In this case, the Cognitive Radio will experience additional building penetration loss. To account for losses from multipath, shadowing and building penetration, the secondary user must be 20-30dB more sensitive than the TV receiver. To get a better understanding of the problem, consider this: a typical Digital TV receiver must be able to decode a signal level of at least -83dBm without significant errors [3]. The typical TV signal is 6MHz wide. The thermal noise in this band is -106dBm. Hence a Cognitive Radio which is 30dB better has to detect a signal level of -113dBm, which is below the noise floor. A very low SNR signal can be sensed reliably, provided enough samples are used for detection as demonstrated in [4]. However if there is ambiguity associated with receiver noise, [5] have proved the presence of a minimum SNR value (called SNR wall) below which robust detection of the primary is not possible. For a receiver noise uncertainty of 1dB the lowest detectable SNR value is -6dB. Transitory nature of Primary users: Primary users that use the spectrum intermittently (some examples would be Public Safety and other packet radio networks in the licensed band), impose limitations on the sensing time of the secondary. For packet radio networks, the Cognitive Radios must detect the presence of a packet and back-off. B. Wealth of techniques to aid Opportunistic use The previous discussion paints a bleak picture for Opportunistic use of the spectrum. However, a wealth of isolated techniques have been proposed to enhance sensing capabilities. Enhanced detection using a pilot signal: Results in [4] have demonstrated the fact that the presence of a pilot can greatly enhance detection in the very low SNR regime. Cooperative Sensing: In [6] cooperation has been showed to greatly reduce the probability of interference to a TV receiver. Presence of a network of radios provides diversity and helps overcome destructive multipath at a single radio. Overcoming shadowing is not as easy, since shadowing demonstrates distance dependent correlation [7]. Cyclostationary Feature Detection: In [6], cyclostationary detectors have been shown to perform better than energy detectors even at -20dB SNR for a 4-FSK modulated continuous phase signal. Cyclostationary analysis can be used to detect features like the number of signals, their modulation types, symbol rates and presence of interferers. High Sensitivity of Secondary users: Current day CMOS technology allows very sensitive radios. The AR5004X and AR5004G WiFi chipsets from Atheros have a receiver sensitivity of -105dB [8]. Wide availability of Geo-locationing devices in wireless handhelds: Current day cell phones come with built-in GPS receivers. Such geo-locationing devices should enable receivers in determining their positions and thus incorporating prior information about transmitters in the area into their detection process.

3 While these techniques offers optimism, their performance in a real system has yet to be demonstrated. C. Concerns of existing Primary users FCC proposal for negotiated use of the TV bands has witnessed many rebuttal comments which highlight concerns of Primary users [9] [10] [11] [12]. While the offered comments are specific to the TV scenario, we have extracted concerns that apply to most situations: Challenges in updating databases of Cognitive Radios. Especially in the TV case it is virtually impossible to have prior information about 10,000 protected contours. Multiple Primary users that coexist on the same bands may occupy different sized frequency bands. For example, analog TV uses a 6MHz wide spectrum. However, when the same spectrum is used by FM Wireless Microphones, only 250kHz spectrum is used. It is difficult to design a system which can scan variable sized frequency bands. While receiver sensitivity of existing Primary receivers may be poor, users may boost gain by employing high gain antennas. As opposed to this, Cognitive Radios are are limited by a fixed antenna. Secondary users may be unable to distinguish between Primary and secondary users of the spectrum. Certain primary users are more susceptible to receiver interference than others. For example, all problems in digital TV appear as a blue screen squelch which is difficult to diagnose. It may not be possible to control Cognitive Radios when they are in the field. A Cognitive Radio may get hacked. There are Primary users which receive extremely low SNR signals (for example, -180dB for radio astronomy). These users must be protected from out-of-band emissions from other bands. Primary users do not have the incentive to put out a control signal. Listen-before-transmit Cognitive Radios may be able to reliably detect presence of a Primary but will be unable to detect the reappearance of a Primary user once secondary transmission has started. What is needed are Listenwhile-transmit radios. While some of these concerns are best addressed by policy (for example, how does one ensure that radios in the field are always complaint?), some of these concerns are technical and should be verified in an actual system. D. Lack of metrics to evaluate interference to Primary users While there is acceptance in the community that Cognitive Radios will introduce a certain level of interference, the level of acceptable interference has yet to be identified (the FCC has made an attempt at this by defining the notion of interference temperature per frequency band). What is required is a set of tests to prove that a Cognitive Radio can reliably detect a Primary user under different circumstances and also vacate a frequency band if a primary user reappears. The above discussion stresses the need for a controlled testbed where the Cognitive Radio idea can be verified against Primary user concerns. In this paper, we propose such a setup based on the Berkeley Emulation Engine 2 (BEE2) platform to experiment with various sensing techniques and develop a set of tests which will allow us to measure the sensing performance of these techniques. Section II discusses the basic architecture and implementation of the testbed. Section III explains the setup using BEE2 and the proposed set of experiments. Following that, Section IV goes into metrics to evaluate performance of various channel sensing and channel use schemes. Finally, conclusions are offered in Section V. II. TESTBED ARCHITECTURE We identified the following list of features for a testbed for Cognitive Radios: Ability to support multiple radios which can serve as Primary or secondary users. Ability for PHY/MAC layer adaptation and fast information exchange between multiple radios for sensing and cooperation. Ability to perform rapid prototyping in order to experiment with different sensing algorithms. Figure 3 shows an abstract diagram of the emulation platform. To implement multiple radios, the emulation platform must provide plenty of parallelism and mechanisms to connect to multiple frontends. Further more, the latency to exchange information between the various radios should be small. Fig. 3. Emulation platform for Cognitive Radios These requirements are met by the Berkeley Emulation Engine (BEE2), which is a generic, multi-purpose, FPGA based, emulation platform for computationally intensive applications. Each BEE2 can connect to 18 frontend boards via multigigabit interfaces. The case for FPGAs, over DSPs and Microprocessors, has been argued in [13]. FPGAs offer rapid reconfigurability, exhibit rapidly increasing computational power per unit area and demonstrate the best computational performance per unit power consumed for key computational modules [13]. A. The BEE2 board The BEE2 consists of 5 Vertex-2 Pro 70 FPGAs. Each FPGA embeds a PowerPC 405 core which minimizes the

4 latency between the microprocessor and reconfigurable logic. These 5 FPGAs form a single Compute Module. Each FPGA can be connected to 4 GBytes of memory with a raw memory throughput of 12.8Gps. Four FPGAs are used for computation and one for control as shown in Figure 4. Adjacent FPGAs are connected via onboard low-voltage 40Gbps (LVC-MOS) parallel interfaces. All computation FPGAs are connected to the control FPGA via 20Gbps links. These high bandwidth, low latency links allow the the five FPGA to form a virtual FPGA of five times the capacity. is available on the control FPGA. The Power PC of the control FPGA can run Linux and a full IP protocol stack. The board also contains USB and JTAG interfaces along with provision for a flash card. The 100 Base-T interface allow remote management and control. B. Modular Front-end system The Front-end system has been designed in a modular fashion. The Analog/baseband board contains the filters, ADC/DAC chips and a Xilinx Vertex-II Pro FPGA. Digital-toanalog conversion is performed by a 14-bit DAC running up to 128MHz, while analog-to-digital conversion is performed by a 12-bit ADC running up to 64MHz. The FPGA performs data processing and control, and supports 4 optical 1.25 Gb/s links for transmitting and receiving data to/from BEE2. The optical link provides good analog signal isolation from digital noise sources and allows the frontend to be moved up to a third of a mile from BEE2 for wide range wireless experimentation. A separate RF modem module connects to the baseband board. The current RF modem module is capable of up/down converting 20MHz RF bandwidth at 2.4 GHz. The RF frequency is fully programmable in the entire 80MHz ISM band. A block diagram of a single RF modem is shown in Figure 6, while Figure 7 shows the RF and baseband boards. Fig. 4. BEE2 Compute Module These FPGAs can connect to the external world using serial Multi-Gigabit (MGT) interfaces. Four MGTs are channel bonded to form a physical into a physical Infiniband 4X (IB4X) electrical connector, to form a 10 Gps full duplex interface. There are a total of 18 IB4X connectors per board. The Infiniband connectors allow the BEE2 Compute module to connect to an Infiniband switch which enables multiple BEE2 Compute models to communicate and exchange data. Figure 5 shows a picture of the BEE2 board. Fig. 6. RF Modem Module and Analog/baseband board Fig. 5. BEE2 board Each BEE2 board supports one 100 Base-T Ethernet which Fig. 7. Front-end boards

5 Scalability is achieved through parallel RF modem modules being provided with a common RF reference and clock signals. Two configurations are supported by this architecture: All front-ends operate at the same radio frequency (The radios need to operate in Time Division Duplex (TDD) mode in a single 20MHz band) Groups of 4 or more antennas operate in different bands (The radios operate in Frequency Division Duplex (FDD) mode and occupy the entire 80MHz band) C. BEE2 Programming model using Simulink The BEE2 can be programmed using Matlab/Simulink from Mathworks coupled with the Xilinx system generator. The tool chain is augmented with BWRC developed automation tools for mapping high level block diagrams and state machine specifications to FPGA configurations. A set of parameterized library blocks have been developed for communications, control operators, memory interfaces and I/O modules. III. COGNITIVE RADIO SETUP Since each BEE2 Compute board allows connection to 18 Front-ends, we can split the 18 interfaces between Primary and Secondary users. This will enable us to construct scenarios with multiple Primary users exhibiting different channel use patterns. Primary user traffic pattern can be controlled via the BEE2. Performance of energy and cyclostationary feature detectors can be characterized as a function of input SNR, sensing time, and modulation types. The on-board BEE2 implementation of various cooperation schemes will allow us real-time experimentation, even in dynamic Primary user traffic patterns. In addition, the optical links from BEE2 to front-end boards that reach 1/3 mile, facilitate experimentation in different shadowing and multipath environments. For the distributed detection of Primary users, protocols for the exchange of control information are necessary. Since a CR system does not provide a priori communication, a dedicated control channel must be used to exchange control information. The protocols used to implement these control channels are an integral part of the testbed. A. First experimental setup For our first experimental demonstration we chose the unlicensed 2.4GHz band in indoor environments. The 2.4GHzISM band is suitable for several reasons: 1) It is an unlicensed spectrum so the cognitive radio operation in this band is not a subject to an agreement with licensed users. Furthermore, it is considered as a very crowded spectrum with many unlicensed devices that are not able to intelligently control and avoid mutual interference. 2) Commercially available WLAN devices for 2.4GHz band, such as IEEE b/g cards within laptops, are quite programmable and allow user to control their transmission parameters. Therefore, they can be used for primary user emulation in a controlled fashion as well as secondary user transmitters. 3) All hardware and software support for 2.4GHz bands is already developed within BWRC to support cognitive radio experiments. Our BEE2 infrastructure supports multiple connections of laptop cards and 2.4GHz frontends that can be combined as a cognitive radio system capable of sensing and transmission. Furthermore, our 2.4GHz are configurable to sense whole 80MHz of spectrum instantaneously while commercial devices can sense only single 20MHz channel. 4) We believe that the performance of sensing algorithms for indoor 2.4GHz experiments, if reported as function of input SNR, can be further extended to other frequency bands. Figure 8 illustrates the setup that combines two primary users and a cognitive radio network connected to BEE2. Note that each cognitive radio is composed of a laptop computer with b/g radio card used for cognitive radio transmission and 2.4GHz 80MHz wide front-end for sensing. Ability to transmit standard compliant b/g waveforms on the secondary links and coordinate control of transmission times, will allow us easy experimentation of protocols for medium access control. Information between the sensing radios and the transmission laptops is exchanged via the standard Ethernet interface which serves as the control channel in the first implementation. Fig. 8. Primary/Secondary user setup using the BEE2 IV. METRICS FOR COGNITIVE RADIOS We need a set of metrics to report the performance of Cognitive Radios under various conditions. There are two set of metrics for evaluating Cognitive Radios: metrics to measure the interference caused to a Primary user and metrics to evaluate the hardware costs of implementing a Cognitive Radio.

6 A. Interference to Primary users Among this set of metrics we are interested in the fraction of time that the CR interferes with the Primary user. In particular, we are interested in the amount of time it takes for the CR to detect the presence of the Primary user. We are also interested in measuring the accuracy of this detection as the environment is varied: no shadowing to large shadowing, no multi-path to large multi-path. These scenarios translate into difference in received signal strengths at various Cognitive Radios. Furthermore, we are interested in the time it takes the CR to cease transmission once a Primary user reappears. Using our setup, we are also interested in the case when a Primary user occupies different bandwidths for transmission with respect to the Secondary user (Primary user uses a single 20MHz channel while the Primary user uses a larger bandwidth and visa versa). We would also like to evaluate the tradeoff between Interference to Primary users and the throughput/delay of the CR system. We are interested in the sustained throughput that the CR system can support as well as delay incurred per packet as a function of the interference caused to the Primary user. Delay and throughput are also functions of the system load. B. Hardware Costs It is important to normalize any CR performance and interference metric with respect to the hardware area, power dissipation and per-unit cost. For power dissipation, the critical design parameters will be the power required for the ADC which is a function of the number of bits, speed and bandwidth. Similarly we expect different amounts of power to be consumed for varying levels in resolving the Primary signal in the presence of interference (more power is required as linearity is increased). ACKNOWLEDGMENT The authors would like to thank the BEE2 team for their valuable comments. REFERENCES [1] Spectrum policy task force report. Technical Report , Federal Communications Commision, Nov [2] Ness, Furchtgott-Roth, and Tristani. Promoting efficient use of spectrum through elimination of barriers to the development of secondary markets. Notice of Proposed Rulemaking , Federal Communications Commision, [3] Office of Engineering and Technology (OET). Longley-rice methodology for evaluating tv coverage and interference. OET Bulletin 69, Federal Communications Commision, Feb [4] A. Sahai, N. Hoven, and R. Tandra. Some fundamental limits on cognitive radio. In Allerton Conference on Communication, Control, and Computing, [5] R. Tandra and A. Sahai. Fundamental limits on detection in low snr under noise uncertainty. In Proc. of the WirelessCom 05 Symposium on Signal Processing, [6] D. Cabric, S. M. Mishra, and R. W. Brodersen. Implementation issues in spectrum sensing for cognitive radios. In Asilomar Conference on Signals, Systems, and Computers, [7] M. Gudmundson. Correlation model for shadow fading in mobile radio systems. Electronic Letters, 27(23): , [8] [9] retrieve.cgi?native or pdf=pdf&id document= [10] retrieve.cgi?native or pdf=pdf&id document= [11] retrieve.cgi?native or pdf=pdf&id document= [12] retrieve.cgi?native or pdf=pdf&id document= [13] Chen Chang, John Wawrzynek, and Robert W. Brodersen. Bee2: A high-end reconfigurable computing system. IEEE Design and Test of Computers, 22(2): , V. CONCLUSIONS In this paper we have presented a testbed for experimenting with Cognitive Radios at the Physical and Network layer. The motivation for a testbed to evaluate Cognitive Radios in a controlled environment stems from studying various spectrum sensing schemes and analyzing the concerns of existing Primary users. This testbed allows us to emulate Primary as well as secondary users and enables us to evaluate the performance of various spectrum sensing schemes. The 2.4GHz spectrum was chosen for initial experimentation due to the availability of off-the-shelf transmission equipment and the ability to emulate Primary users in a controlled manner. These 2.4GHz radios are connected to the Berkeley Emulation Engine 2 (BEE2) which is a multi FPGA emulation platform. Multiple Cognitive and Primary Radios can be implemented on the BEE2. Furthermore, the Cognitive Radios can exchange information in a timely manner since the BEE2 FPGAs are connected via high bandwidth low latency links. BEE2 enables us to implement a control channel and protocols for cooperation among multiple Cognitive Radios.