communications by introducing agility in frequency, time and space as well as .... On both SSCN and SUN, external Atheros WiFi adapters are used for SU ...
Cognitive Agile Networking Testbed Hanwen Cao, Christoph K¨onig, Andreas Wilzeck, M.-D. P´erez Guirao Institute of Communications Technology, Leibniz Universit¨at Hannover Appelstr. 9A, 30167 Hannover, Germany Email:{hanwen.cao, christoph.koenig, andreas.wilzeck, dolores.perez}@ikt.uni-hannover.de Abstract— Cognitive radio is a well motivated concept for improving the inefficient spectrum utilization and apparent scarcity that result from traditional spectral regulations. So far, most efforts in the cognitive radio research field focus on techniques for the enhancement of the amount of available spectrum under the stringent constraint of causing no harmful interference to legacy users. But communication reliability also can substantially benefit from the cognitive radio approach by means of agile utilization of transmission opportunities in frequency, time and space. In this paper, we present our Cognitive Agile Networking (CAN) framework that aims at achieving robust low-latency communications by introducing agility in frequency, time and space as well as learning and reasoning for optimizing network operations. In order to respect practical issues of real-world applications, we are building a COgnitive Agile Spectrum Testbed (COAST) for experiments and validations. An initial version of this testbed is introduced here and its spectral agility is demonstrated. Index Terms— Cognitive Radio, Spectral Agility, Testbed, Spectrum Sensing
I. I NTRODUCTION The idea of cognitive radio was originally motivated by the apparent spectrum scarcity caused by inefficient command-and-control regulations, and, on the other side, by the technical feasibility of improving spectrum utilization by allowing Secondary Users (SU) to share the available spectrum with incumbent Primary Users (PU) on a non-interfering basis. During the last ten years, many efforts have been made to solve spectrum scarcity in order to stimulate new wireless broadband applications and markets. For instance, we have witnessed the advent of the first cognitive radio standard IEEE 802.22 [1] along with the FCC’s opening of white space spectrum in VHF and UHF band [2]. But cognitive radio involves much more than a set of techniques for improving spectrum utilization. In essence, a cognitive radio is “an intelligent wireless communication system that is aware of its environment and uses the methodology of understanding-by-building to learn from the environment and adapt to statistical variations in the input stimuli” [3]. In [3], the author also pointed out that another important objective of cognitive radio is to establish “highly reliable communication whenever and wherever needed”. With the intelligent cognition cycle
composed of observation, orientation, planning, learning, decision and action [4], cognitive radio is capable of discovering transmission opportunities in the varying wireless environment and optimizing their use to smartly improve communication efficiency and reliability. In this paper, we propose the concept of Cognitive Agile Networking (CAN) where agility in frequency, time and space is emphasized in order to achieve robust, lowlatency communications for application scenarios in which reliability is of most concern, e.g. industrial, emergency, and public safety communications. With the hope of meaningfully contributing to realworld applications, we are building the COgnitive Agile Spectrum Testbed (COAST). This testbed will allow us to develop, test and evaluate algorithms, protocols, and networking applications assisted by environmental cognition and multi-dimensional radio agility. The rest of the paper is organized as follows: in Section II the CAN concept along with suitable application scenarios are illustrated. Section III introduces the architecture and the spectrum sensing algorithm of the initial COAST platform. Demonstration and measurements of spectral agility with COAST are presented in Section IV. Finally, Section V concludes the paper. II. C OGNITIVE AGILE N ETWORKING A. Application Scenarios In the field of industrial, emergency, and public safety communication, reliability and latency issues are of most concern. For industrial wireless communications the data-rate is not the focused topic. Instead, reliability, low latency, and the ability to connect hundreds of different devices (e.g. sensors, actors, control units) with a high Quality of Service (QoS) guarantee are major requirements [5]. During extreme situations or emergency cases, traditional wireless systems may cease to work because of the damaged infrastructure or a overloaded spectrum. Thus, for such cases new wireless technologies are required to guarantee efficient communications [6]. Cognitive radio is a promising approach for providing reliable and low-latency communications in the application scenarios described above thanks to its fourfold
Secondary Sensing & Control Node (SSCN)
Cognitive Engine
Wispy DBx Sensor
Primary User Node with USRP (PUN)
Internal WiFi Adapter
Internal WiFi Adapter
External WiFi Adapter
Channel Controller
User Applicatio ns
Secondary User Nodes (SUN)
Fig. 1.
System architecture of the initial version of COAST
approaches: (i) exploitation of multi-dimensional communication opportunities; (ii) anti-jamming by using redundant/backup channels intelligently; (iii) minimizing interference among coexisting heterogenous systems via adaptive techniques; (iv) integration of context-aware information in the decision making process. B. Overview of the CAN Research Framework Here we outline the scope of the Cognitive Agile Networking (CAN) research framework. CAN aims at introducing agility in the frequency, time and spatial domains in order to enhance network connectivity with reliability and latency guarantees. In the frequency domain, spectral opportunities are exploited on a signal classification basis since the identification of wireless signal characteristics (modulation, power, protocol, etc.) can lead to more exhaustive understanding of the radio environment. In the extreme case, when no purely vacant frequency bands are available, the exploitation in the time domain allows CAN to use temporal transmission opportunities, e.g. idle TDMA slots or even random intervals between bursty signals. Although this is technically challenging, some pioneering work has been done in [7]. Another important feature of CAN is the inclusion of learning, reasoning and adaptation techniques for optimizing and maintaining reliability in a fast changing radio environment, e.g. learning and prediction of certain incumbent signal’s temporal or spectral behaviors can facilitate the discovery of transmission opportunities for SUs. III. CO GNITIVE AGILE S PECTRUM T ESTBED (COAST) The validation of the techniques described in section II-B requires the use of a testbed for experimentations in RF environments that respects the practical issues in realworld applications. In this section, the design of the initial version of our COAST testbed as well as its classificationbased spectrum sensing algorithm are introduced. A. Testbed Architecture COAST targets an application scenario in which a SU network and a PU with dynamic spectral behavior
(frequency hopping) share the same frequency band, which is divided into a number of non-overlapping channels. If the SU network and the PU attempt to transmit in the same channel, interference will occur. To avoid interference, one of the secondary devices acts as a central sensing and control node. It keeps track of the spectrum usage in the band and order a channel switch for the SU network if the presence of the PU is detected. The channel switch command is sent on a fixed, out-of-band control channel. The testbed architecture is depicted in Fig. 1 and consists of three types of nodes: (i) a Secondary Sensing and Controlling Node (SSCN) that acts as the SU network coordinator, (ii) a Secondary User Node (SUN) and (iii) a Primary User Node (PUN) that generates the dynamic primary signal. The computation cores of all these nodes are HP 2140 netbooks. On SSCN, spectrum sensing is performed by the USB-interfaced spectrum analyzer WiSpy DBx [8]. With this device, energy detection within a single frequency band is possible at a time. It is planned to replace it with a more powerful sensing device capable of performing advanced detection methods. This will result in a more accurate sensing and enhance the decision making process. On both SSCN and SUN, external Atheros WiFi adapters are used for SU ad-hoc communication in the 5 GHz band. Besides, the internal WiFi adapters of the netbooks are used for the control channel in the 2.4 GHz band. Except for spectrum sensing, SSCN has the full functionality of SUN. On PUN, the software-defined radio front-end Universal Software Radio Peripheral (USRP) [9] is used to transmit the hopping PU signal in the 5 GHz band. All the software programs for SSCN, SUN and PUN are running under Ubuntu Linux v9.04 operating system. A modified version of Spectools [10] is used to drive the WiSpy DBx device, in which sensing parameters (e.g. scanning frequency range, bandwidth, resolution) can be configured with a XML file. The Atheros WiFi card is driven by the ath5k driver [11], which enables fast channel switching. The spectrum sensing algorithm and channel control program are implemented in C programming language. On PUN, USRP is driven by GNU Radio [12]. A photo of the testbed is shown in Fig. 2. The cognition cycle of the secondary network is performed as follows: • SSCN order SUNs to start transmissions in one of the channels identified as idle. • When SSCN senses the presence of the PU on the currently occupied data channel, it randomly selects an idle channel and broadcasts a switch command over the control channel to all SUNs. • On the SUN, a channel control program listens for incoming switch commands and changes the frequency of the external adapter to the indicated channel.
Fig. 2.
Photo of the initial version of COAST
Fig. 3. Spectrum sensing with the combination of PSD shape and temporal features(continuous or bursty)
B. Spectrum Sensing The WiSpy DBx spectrum analyzer tunes to a number of carriers with a fixed interval separated in a certain frequency range and measures the power of the filtered signal in order to get the power spectrum. In our algorithm, the scanned frequency band is divided into several continuous and non-overlapping channels with 20 MHz bandwidth. Assuming there are N channels, each of which contains M samples of the signal power in dBm. In total, we have M N power samples within the scanned band: {pji } i ∈ {0, 1, . . . , M N − 1}
(1)
where j is the index of the scanning period for the whole band. Firstly, the minimal power level among all the M N samples is found: pjmin = min({pji }) i ∈ {0, 1, . . . , M N − 1}
(2)
Then, the detection value for each channel is calculated: gnj = max({pji }) − pjmin n ∈ {0, 1, . . . , N − 1} i ∈ {nM, nM + 1, . . . , n(M + 1) − 1}
(3)
Where n is the index of the channels. If gnj is larger than a threshold value ξ, channel n is consider to be occupied. The selection of ξ depends on the requirement of false alarm and detection probability. Lower threshold value results in higher detection probability but also higher false alarm probability. In IEEE 802.22, spectrum sensing is performed in quiet periods [1] when all nodes temporally cease transmission synchronously. In order to maintain high spectrum utilization and avoid possible link outage, the interval between two successive quiet periods should be as long as possible. However, longer intervals make the system less agile to react to the appearance of PU signals, in other words, causes longer time of mutual interference.
Moreover, synchronization of different nodes’ quiet periods requires additional signalling overhead which further degrades spectrum utilization. In our testbed design where spectrum agility is specially emphasized, quiet periods are bypassed and sensing is performed independently of SU transmission. The question arising is how to identify the just appeared PU signal when it is mixed up with the SU signal. We have developed a method to differentiate continuous PU from bursty SU signals by exploring both their spectral and temporal characteristics. In our demonstration we have assumed bursty IEEE 802.11a SU signals and continuous PU signals. The assumption for the last is based on the fact that digital/analog TV broadcasting and wireless microphones concerned most in TV white space nowadays use continuous transmission. In the frequency domain, PU and SU signals can be differentiated based on the shape of their Power Spectral Density (PSD). IEEE 802.11a signal is wide-band OFDM modulated with flat-top PSD, while continuous PU signal is narrow-band single carrier modulated with peak-top PSD. If the SSCN continuously observes a peak-top shaped PSD over a given number of scanning periods, it can assume that the PU is using the channel. IV. D EMONSTRATION AND M EASUREMENT The configuration of the testbed for demonstration is listed in Table I. SSCN senses the whole band continuously for occupancies of the channels using the previously introduced algorithm. As in our demonstration scenario, the signal received are very strong, the threshold value ξ is set to 20 dB by which the false alarm probability is negligibly small. The waterfall view in Fig. 4 shows the SU and PU’s behaviors in both time and frequency domains. A. PU Detection Delay An important aspect for the performance and the protection of PU is the detection time of the PU signal by the
TABLE I PARAMETERS OF THE D EMONSTRATION Parameter Spectral Band Primary User Signal Secondary User Signal Power Samples Common Control Channel Detection threshold ξ
Value 5.49 - 5.59 GHz, 5 Channels QPSK, 1M baud/s, 10 sec/hop IEEE 802.11a 20 samples/channel, 1 MHz interval 2.412 MHz 20 dB
Fig. 4. Waterfall view of the spectrum showing spectral agility behaviors
SSCN. This time depends on the required sensing accuracy that is reflected in the probabilities of detection and false alarm. We figured out that stable detection of PU signal can be achieved via continuous observation of matched PSD shape in four successive scanning periods. The measured average detection time is around 500 ms. The measurement is performed as follows. The PUN and SSCN are directly connected via Ethernet and every time the PUN switches the channel, a notification is sent to SSCN where the arrival time is recorded with SystemTaptool [13]. The appearance time of the PU signal as detected by the SSCN is also recorded and hence the detection delay can be calculated. The impact of the Ethernet delay of 0.2 ms is negligible in comparison to the mean detection time of 500 ms. B. Channel Switching Delay A further performance aspect is the time a SU device need to switch the channel. Using the ath5k driver, channel switching actions are performed on the fly swiftly without restarting the linux network interface and hence without a temporary break down of the protocol stack. The measurement of the channel switching time is also performed with the help of the SystemTap-tool. It is taken from the difference between the time at which the change frequency command is started from the kernel at the ath5k-driver and the time at the return from this call. The channel switch costs about 50 ms.
V. C ONCLUSION AND F UTURE W ORK This paper presents our vision of Cognitive Agile Networking (CAN) in which agile utilization of multidimensional transmission opportunities with adaptive and intelligent techniques is emphasized for achieving robust and low-latency communication. The spectral agility is demonstrated with our initial version of COgnitive Agile Spectrum Testbed (COAST), which is developed using standard easy-to-use components. Future work will focus on sensing techniques that can discover opportunities agilely and accurately (e.g. feature detection, cyclostationary detection, distributed sensing), especially short temporal and spacial opportunities. Additionally, adaptive and intelligent techniques to further enhance network connectivity in the fast-changing radio environment will be investigated. Furthermore, as the centralized structure is a weak point of the system, we are working on a distributed coordination approach. For testbed development and experiment, more sophisticated hardware components like FPGA or DSP will be employed, with which PHY-MAC algorithms can be fully controlled or defined. DSP and FPGA hardware design work with low and deterministic latencies and is thus suitable for temporal agile designs and efficient data processing. Possible candidates are USRP2 [9], WARP [14] and Lyrtech SFF SDR [15]. R EFERENCES [1] C. Stevenson et al., “IEEE 802.22: The first cognitive radio wireless regional area network standard,” IEEE Communication Magazine, vol. 47, no. 1, pp. 130–238, Jan. 2009. [2] FCC Adopted Rules For Unlicensed Use of Television White Spaces. http://hraunfoss.fcc.gov/edocs public/attachmatch/FCC-08 -260A1.doc. [3] S. Haykin, “Cognitive radio: Brain-empowered wireless communications,” IEEE Journal on Selected Areas in Communications, vol. 23, no. 2, pp. 201–205, Feb. 2005. [4] J. Mitola III., “Cognitive radio: an integrated agent architecture for software defined radio,” Ph.D. dissertation, Royal Institute of Technology (KTH), Stockholm, Sweden, June 2000. [5] T. Kaiser et al., “Cognitive radio & networks in the perspective of industrial wireless communications (invited),” in 2009 2nd International Workshop on Cognitive Radio & Advanced Spectrum Management (CogART 2009), May 2009. [6] A. Gorcin et al., “Public safety and emergency case communications: Opportunities from the aspect of cognitive radio,” in 3rd IEEE Symposium on New Frontiers in Dynamic Spectrum Access Networks (DySPAN 2008), Oct. 2008, pp. 1–10. [7] L. Biard et al., “A hardware demonstrator of a cognitive radio system using temporal opportunities,” in 4th International Conference on Cognitive Radio Oriented Wireless Networks and Communications (CROWNCOM 2009), June 2009. [8] MetaGeek LLC. http://www.metageek.net/. [9] Ettus Research LLC. http://www.ettus.com/. [10] Spectools. http://www.kismetwireless.net/spectools/. [11] Linux Wireless ath5k driver. http://www.linuxwireless.org/. [12] GNU Radio. http://www.gnuradio.org/. [13] SystemTap. http://sourceware.org/systemtap/. [14] Wireless open-Access Research Platform(WARP). http://warp.rice.edu/. [15] Lyrtech Small Form Factor SDR. http://www.lyrtech.com/.