OpenBTS: a step forward in the cognitive direction - Semantic Scholar

OpenBTS: a step forward in the cognitive direction Pasquale Pace and Valeria Loscr´ı Department of Electronics, Computer and Systems Sciences - DEIS University of Calabria Arcavacata di Rende - ITALY Email: {ppace,vloscri}@deis.unical.it

Abstract—The paper proposes the integration of new cognitive capabilities within the well known OpenBTS architecture in order to make the system able to react in a smart way to the changes of the radio channel. In particular, the proposed spectrum sensing strategy allows the OpenBTS system to be aware of other active transmissions by forcing to choose a new radio channel, within the GSM frequency band, when a licensed primary user has to transmit on a busy channel. The implemented scheme, representing a solid step forward in the cognitive direction, has been validated throughout a detailed testbed pointing out strengths and limitations in realistic communication environments. Index Terms—Dynamic Spectrum Access, Cognitive Radio, Spectrum Sensing, Software Defined Radio, OpenBTS, GSM.

I. I NTRODUCTION In recent years we have witnessed the emergence and affirmation of new wireless communication technologies for data transmission in a short range such as Wi-Fi and Bluetooth even if the research community has investigated few limitations mainly related to the interference due to the transmission in an unlicensed frequency band and the lack of a wise mechanism to avoid collisions or concurrent transmissions; in addition, these technologies need a new hardware specifically designed according to different communication standard that are quickly changing becoming obsolete very soon. On the contrary, if you only need a robust voice communication, GSM technology is still the easiest solution since it is the most successful standard in the world covering over 90% of the globe’s population; moreover, the mobile equipment for the transmission is quite common and extremely cheap. Since the GSM transmission is performed on licensed frequency band, the only drawback of this technology is the connection cost to be payed to Mobile Network Operators (MNOs) that own the transmission license and the entire network architecture implemented with very expensive hardware components. However, recent development in signal processing and Software Defined Radio (SDR) solutions [1] have made the GSM architecture much more economical to be implemented substituting most of the back end hardware with real-time software applications. In particular, thanks to the development of GNU radio [2] and Universal Software Radio Peripheral (USRP) [3], an effective SDR solution consisting of an emR ulation of GSM architecture called OpenBTS° [4], has been successfully implemented and tested in order to provide realtime voice connectivity for rural and underdeveloped countries

at a very low costs [10] [11]. Of course, the use of this solution is absolutely prohibited in dense areas where the transmission on licensed frequency bands is permitted only to MNOs as already explained; nevertheless, in the next future such transmission may be performed thanks to the implementation of new communication paradigms mostly based on Dynamic Spectrum Access (DSA) techniques [5] [6]. In more detail, the DSA paradigm will allow the unlicensed secondary user (SU) to access and utilize a spectrum hole for a period of time and move to another spectrum hole whenever the licensed primary user (PU) appears. Starting from this intuition and taking into account the modern Cognitive Radio (CR) paradigm [7], we propose to extend the potential of the OpenBTS system architecture by implementing a new software module to provide specific cognitive features including spectrum sensing and spectrum mobility; in this way, the Cognitive OpenBTS system, periodically performs spectrum sensing to detect the presence of licensed users signals before a spectrum portion is accessed by unlicensed users as to avoid harmful interference. The spectrum sensing procedure, based on energy detection techniques, is frequently repeated in order to avoid a decrease of QoS guaranteed to the PUs also making the SUs transmission completely transparent and harmless to the PUs full paying for the real-time voice service. We demonstrated the feasibility of our cognitive strategy throughout a detailed testbed pointing out the right working conditions of the whole system in terms of interference limits tolerated by primary users; however, we would like to emphasize that in order to make our solution fully sealable and attractive, the Spectrum Management Authority and the industry have to work in close cooperation to realize the goal of Dynamic Spectrum Access by adapting the RF spectrum regulations and making the use of RF spectrum more flexible. The rest of the paper is organized as follows. In Section II, we present a review of works based on the OpenBTS system pointing out the main application fields of this open softwarehardware architecture. In section III, we briefly describe the OpenBTS system in order to provide the reader with all the basic knowledge to understand the system operation whilst the new features of the cognitive module, to be integrated in the whole system, are presented in section IV. Performance evaluation and numerical results coming from a realistic testbed are presented in Section V. Finally, Section VI concludes the paper by proposing future research directions on this exciting

field. II. R ELATED WORKS In the last few years different works have been conducted with the aim of investigating the potential of open software and hardware architectures. The feasible implementation of a simple and standard GSM Base Transceiver System (BTS) by using Universal Software Radio Peripheral, which is a multipurpose motherboard for Software Defined Radio (SDR), and a Personal Computer (PC), has been described and validated in [8] whilst in [9] the authors implemented a base station using state-of-the-art software and hardware components, namely GNU Radio, OpenBTS and USRP software and hardware platforms to demonstrate the coexistence of two heterogeneous wireless systems, GSM and CSMA, in the same GSM-900 band. In addition, more recently, the research community interest has been moved to implement complex real testbed such as those proposed in [10] and [11] mainly focused on providing network connectivity and coverage extension in low population densities and low income rural areas of the developing world, where big telecoms often defer from deploying expensive infrastructure. This goal has been reached by using OpenBTS-based GSM microcells to implement a low-cost and low power rural network architecture that extends data and voice connectivity from the closest city/town to nearby rural regions. Since these large testbeds follow the open software and hardware approach, the authors argued that the proposed architectures are simple and easy to deploy, yet robust also requiring no modification to GSM handsets. Despite the cited works, our contribution aims at implementing cognitive capabilities within the OpenBTS system architecture making it smarter and more reactive to the radio channel variation. Moreover, the proposal aims to offer a fair QoS to the active connections without causing interference to those users for whom the transmission has to be always guaranteed with the maximum quality because they pay a specific fee to the mobile network operators (MNOs) for transmitting in a licensed band. III. O PEN BTS SYSTEM ARCHITECTURE In this section we describe the general OpenBTS system architecture and the role of the specific subsystems making a differentiation between software and hardware components. Figure 1 summarizes the whole system architecture. A. OpenBTS OpenBTS is a set of open source software modules that allow to create an own GSM network by replacing standard GSM network infrastructure from the BTS upwards. Calls generated by standard GSM mobile terminals, are handled by an Asterisk PBX server instead of a MSC and mobile users, despite of calling each other, can also send SMS messages. Such an OpenBTS based GSM network consists of a dedicated USRP hardware component connected to a USB port of a computer equipped with Linux operating system, Asterisk,

Fig. 1.

OpenBTS hardware and software system architecture.

GNU Radio and OpenBTS. The general structure of the network is presented in the figure 1. The OpenBTS module is written in the C++ programming language and it is released as free software under the AGPL license. The mobile station connects first to the USRP working as a standard GSM cellular BTS. This is the only hardware module which enabling radio communications. The transmitted data are later elaborated throughout the GNU Radio software in order to be correctly processed by the OpenBTS software. Those three components (i.e. USRP, GNU Radio and OpenBTS) create together the standard “Um” interface. Finally the Asterisk server is used to implement user management and call forwarding functionalities fulfilling the functions of HLR and AuC. B. GNU Radio GNU Radio is a free & open-source software development toolkit to implement software radio capabilities which allow processing of the high frequency transmitter or receiver signals. It is used mainly as the USRP controlling program which prepares data for signal processing. The signal processing operation is later done by custom modules. GNU Radio offers an easily reconfigurable radio system allowing its users to create different devices without the need to buy several expensive radios. C. Asterisk Asterisk is a call center system consisting of an IP PBX with integrated VoIP gateway. It is used for all call control functions as well as parts of mobility management tasks. It utilizes subscriber IMSIs (International Mobile Subscriber Identity) as Session Initiation Protocol (SIP) user names and presents each GSM handset to Asterisk as a SIP client. This software component makes the entire OpenBTS project even more cost-effective. IV. C OGNITIVE MODULE FOR O PEN BTS The aim of this work is the design of an application that cooperates with Open-BTS to realize a GSM Base Transceiver

Station arranging communications channels in a cognitive fashion. In order to make this possible, it is necessary the implementation of a spectrum sensing procedure that scans the surrounding and detects available frequencies where it would be possible to start a communication. To the follow we give some details about the spectrum sensing concept and some techniques known in literature.

value. The last step allows the evaluation of the average power related to the specific bandwidth dimension in order to obtain the PSD value expressed in W/Hz. The algorithm is repeated for each GSM channel and, as a result, we can argue that if the computed PSD value is greater than a certain threshold value likely on this specific GSM channel there is, for sure, a communication in progress.

A. Spectrum sensing capability Spectrum Sensing is a very important feature for technologies that allow opportunistic access to few portions of the radio spectrum not used by licensed users. In this context, a cognitive radio device operates as a “smart” radio apparatus able to make a frequencies scanning in order to tune on channels not “occupied” by licensed users (spectrum holes). Moreover, cognitive devices have to be able to make free frequency channels when a licensed user needs to transmit on those specific channels. This opportunistic and extremely dynamic spectrum access could allow better usage of transmission resources and occupied bandwidth; thus, a fundamental feature of cognitive networks is the correct and rapid detection of the spectrum “holes” representing the portion of the spectrum not used in a certain time. In Fig. 2 we summarize the most important spectrum sensing techniques [12],[13].

Fig. 2.

Fig. 3.

Power Spectral Density computation.

B. The Cognitive Algorithm The different steps within the proposed cognitive strategy to improve the awareness and responsiveness of the OpenBTS system are shown in Fig. 4.

Spectrum Sensing Techniques.

In particular, the transmitter detection approach supports the capability to determine if a signal from a primary transmitter is locally present in a certain spectrum. Transmitter detection methods currently used are: • Matched-Filter Detection; • Energy Detection; • Cyclostationary Feature Detection. In this work we will focus on Energy Detection techniques implementing a specific algorithm for the Power Spectral Density (PSD) evaluation on the GSM channels as shown in figure 3. When a generic next generation cognitive user has not enough information regarding signals transmitted by a primary user, the best method to verify the presence of a communication is to measure the PSD of the signal in the considered frequency bandwidth. In particular, we executed the spectrum analysis in the frequency domain throughout the discrete fourier transformation (DFT) by acquiring a vector of 256 complex samples in order to compute the average power

Fig. 4.

Flowchart of the Cognitive OpenBTS algorithm.

Going into the details, we can see that at the application starting phase, the system first makes the spectrum sensing procedure to scan the GSM frequency channels, then it takes the decision whether to perform the channel change. Each channel is identified through a specific ARFCN (Absolute Radio Frequency Channel Number) value in a range of 125 channels corresponding to the standard GSM frequencies from 890M Hz to 915M Hz for downlink and, from 935M Hz to 960M Hz for the uplink, with a channel dimension of 200KHz. More details about this can be found in [14]. During

the spectrum sensing phase, the Power Spectral Density (PSD) on the downlink channels is measured. We decided to implement the spectrum sensing feature on the downlink channels because each BTS periodically sends signal messages even if there are not any connected users. In fact, in this latter condition the uplink spectrum sensing procedure would not reveal any occupied frequency because there was no communication. Once the PSD is detected for each channel, the system build a list of 125 channels sorted in ascending, thus obtaining the optimum channel with less noise in the first position of the list. OpenBTS can now handle the communication on this optimal channel by automatically setting the GSM.ARF CN parameter into the specific OpenBTS configuration file. Once OpenBTS started, the spectrum sensing procedure is performed again after a certain time δ (i.e., 2 minutes in our testbed) in order to verify that the chosen ARFCN is not “occupied” by a newly arrived PU. After that, the system evaluates the quality offered to the active connections, if any, by making decisions about what to do according to the two following conditions: •

•

There are active calls: the system checks the quality of the current communications by verifying the FER (Frame Error Rate) associated to traffic and signal channels. Whether all the communications have too low quality, the system infers that the used ARFCN is too noisy due to an external transmission; thus becomes necessary to execute a switch procedure consisting in “abandoning” the current channel in order to tune on a new free frequency detected during the previously spectrum sensing phase. Whether there is at least a single communication with a poor quality, lower than a threshold value (i.e., FER > 15% in our testbed), the system makes the channel switch looking for a new channel; There are no active calls: the system checks the position of the used ARFCN channel inside the list obtained throughout the previous spectrum sensing phase. If the channel has moved to a location far from the head of the list, it means that there is a lot of interference mostly due to the presence of a new PU, thus it is necessary to choose a new channel less disturbed by taking the first in the list otherwise the system does not make any frequency change keeping active the same channel.

We would like to point out that “Spectrum Sensing” and “Quality Check” procedures are executed periodically. Unfortunately, the mandatory switch channel process causes the drop of the active communications due to the OpenBTS reconfiguration that cannot be handled “on the fly” by using the current release of OpenBTS. V. T ESTBED AND R ESULTS In this section we show the testbed setup and the performance analysis of the implemented cognitive architecture. Moreover, the overall behavior and the system reactiveness in the switching phase of the ARFCN channel will be tested in different working conditions. In particular, we will make

a transmission on an already busy ARFCN channel implementing a specific GNU Radio script in order to simulate the behavior of a PU transmitting in a licensed frequency band; this transmission will be detected by the cognitive module installed on the OpenBTS forcing the system to find a new channel throughout the spectrum sensing capability. In this way, a specific handover procedure, towards a free ARFCN channel, will be implemented if the interference produced by the transmission of the PU is detected to be greater than a certain threshold. A. Hardware and Software components The implemented testbed consists of 2 Personal Computers (PC) and 4 USRP equipped with specific daughterboards and software modules as explained in the following: • 1 PC equipped with OpenBTS version 2.6 and the proposed Cognitive module; • 1 PC dedicated to generate the noise signal in order to mimic the behavior of a generic PU; at the same time this PC will be used as a radio spectrum analyzer thanks to a specific software module supported by GNU Radio 3.4.2; • 1 USRP, working as a standard BTS of the GSM network, equipped with 2 Flex 900 daugtherboard for the transmission on the downlink and uplink directions; • 1 USRP equipped with a Flex 900 daugtherboard to perform the spectrum sensing functions on the GSM channels; • 1 USRP equipped with a Flex 900 daugtherboard to emulate the PU transmission in the GSM frequency band on a specific ARFCN channel; • 1 USRP equipped with a Flex 900 daugtherboard working as a spectrum analyzer to verify the system operation during different working phases. The figure 5 shows the 4 USRP performing different tasks during the testbed. In particular, the USRP working as a standard BTS and the USRP used for implementing the spectrum sensing capabilities (fig. 5.a) have been connected to the usb ports of the first PC on which both OpenBTS and the new cognitive module, have been installed and configured. The URSP used to mimic the behavior of a PU transmitting in the licensed GSM frequency band and the USRP working as a spectrum analyzer (fig. 5.b) have been connected on the second PC on which two different GNU Radio applications, are executed at the same time. B. Testing the cognitive module To validate the new cognitive capabilities of the OpenBTS system, we activated the cognitive module by executing a specific script aims to automatically find the best ARFCN channel on which establish the GSM service provided by OpenBTS. This new sensing capability is performed according to the energy detection strategy described in section IV.A, the figure 6.a shows the power spectral density (PSD) of all the 125 GSM channels detected by the sensing procedure performed before to start the OpenBTS service. As we can

Fig. 6. PSD of the GSM channels: a) Before to start the OpenBTS service, b) During the PU transmission.

Fig. 5. Cognitive OpenBTS testbed: a) Spectrum sensing & OpenBTS, b) Spectrum analyzer & Primary User.

see, since the GSM channel number 7 presents the lowest PSD value, the sensing module suggests to choose this channel to perform the OpenBTS service in order to guarantee a good QoS for the users handled by OpenBTS without annoying any PUs. The selected ARFCN channel will be used by the OpenBTS system until the cognitive application will not detect an increase of the PSD value on that specific channel, representing the transmission of a potential PU; otherwise, the cognitive module will decide to perform frequency hopping selecting a free channel with the lowest PSD value. To validate the proposed strategy, we transmitted an audio signal consisting of an audio wav file modulated on the GSM channel number 7 (fig. 7) and representing a generic PU; in particular the figure 7.c shows the GNU Radio implementation of this specific signal whilst the output of the spectrum analyzer before and after the transmission is shown in figures 7.a and 7.b respectively. The sensing procedure is performed by the cognitive module every 2 minutes in order to detect the presence of a PU or any change in the radio channel. After this time, the output of the sensing procedure in terms of PSD values on the GSM channels is shown in figure 6.b. As we can see, the PSD value around the channel number 7 is now considerably higher than the previous case; thus, the system has correctly detected the PU transmission and the new sensing phase also selecting the channel number 114 as the more suitable for the OpenBTS system. We would like to remark that the choice of a new ARFCN channel for the OpenBTS system causes the drop of all active

Fig. 7. Spectrum analyzer on GSM channel 7: a) Before noise signal, b) After noise signal, c) GRC script of the noise signal.

calls because the whole system needs to be restarted. For this reason we decided to investigate the conditions under which the OpenBTS system reboot is mandatory in order to guarantee an adequate QoS to the active connections also preserving the PU’s transmission. For this study, we considered the QoS of each phone call in terms of frame error rate (FER) because this index is already implemented in the OpenBTS architecture and it can be easily computed for each couple of mobile devices involved in a GSM communication. In our testbed we decided to fix the quality FER threshold (QT h ), above which the grade of service is too low, equal to 15%. We conducted few ad hoc tests to practically validate this choice experiencing a very low speech quality when the FER value of a GSM call is over this threshold making the conversation really annoying. According to this remarks, we decided to reboot the OpenBTS system setting a new ARFCN channel only if the FER experienced by at least one of the active calls is above the quality threshold QT h .

TABLE I FER VALUES [%]

EXPERIENCED BY DIFFERENT CALLS WITH AND WITHOUT EXTERNAL INTERFERENCES .

Call number

Traffic channel

Without PU

PU Low power 29dBm

PU Medium power 34dBm

PU High power 39dBm

1

TCH 1 TCH 2

5,87E-24 2,13E-16

0,0047 0,0270

6,93 7,59

30,24 24,69

2

TCH 1 TCH 2

1,07E-22 4,3E-23

0,0063 0,0057

0,82 1,90

42,01 17,27

3

TCH 1 TCH 2

6,3E-46 3,65E-09

3,02E-05 0,143

2,21 5,87

8,14 41,04

Table I shows the results obtained by activating 3 simultaneous calls on the OpenBTS system while a PU transmits on the same channel with different power levels compliant to the GSM standard (i.e., 29dBm, 34dBm and 39dBm [14]). As we can see, the presence of a PU transmitting with a low or medium power level does not require any ARFCN change because the quality level offered to the voice calls is always respected. On the contrary, if the system detects a low quality for the active calls due to the high power transmission of the PU, it will look for a new ARFCN by using its own cognitive capabilities. VI. C ONCLUSION This paper has outlined the future cognitive direction of the next generation wireless system extending the potential of the OpenBTS system architecture by integrating a smart software module able to use temporary free spectrum portions without annoying any primary users. The proposed cognitive extension has been validated throughout a detailed testbed pointing out strengths, such as system awareness and reactiveness to radio channel changes, and limitations mainly related to the impossibility of dynamically changing the transmission channel (rebooting OpenBTS) without dropping in progress calls. Future research directions will take into account the implementation of a totally transparent handover mechanism between two USRP representing two different OpenBTS systems in order to avoid the active calls dropping. ACKNOWLEDGMENT This work has been carried out under the framework of STEM-Net, PRIN-National Italian Project #H21J11000050001, financed by the Italian Ministry of University and Research. R EFERENCES [1] J. Mitola, “The software radio architecture, IEEE Communications Magazine, 06 August 2002. vol 33 issue 5, pp. 2638. [2] GNU Radio. http://gnuradio.org. - Accessed on March. 2, 2012. [3] Ettus Research LLC. http://www.ettus.com - Accessed on March. 2, 2012. [4] OpenBTS project official site. http://openbts.sourceforge.net - Accessed on March. 2, 2012. [5] I. F. Akyildiz, W.-Y. Lee, M. C. Vuran, and S. Mohanty “Next Generation Dynamic Spectrum Access Cognitive Radio Wireless Networks: A Survey, Computer Networks. 2006. vol. 50, pp. 2127 2159.

[6] Y. Zhang “Dynamic Spectrum Access in Cognitive Radio Wireless Networks,” IEEE International Conference on Communications, ICC ’08, Beijing, China, 19-23 May 2008, pp. 4927-4932. [7] S. Haykin “Cognitive radio: Brain-empowered wireless communications, IEEE Journal on Selected Areas in Communications, February 2005, vol. 25, pp. 201-220. [8] E. Natalizio, V. Loscr´ı, G. Aloi, N. Paoli and N. Barbaro, “The practical experience of implementing a GSM BTS through Open Software/Hardware,” Proceedings of ISABEL 2010, Rome, November 8-10, 2010. [9] S. Liao and L. Bao, “Implementing a Base Station Using the SDR Platform for Coexistence of Heterogeneous Wireless Systems,” Demo in SDR Forum Technical Conference and Product Exposition (SDR), Washington, DC, 2009. [10] A.Anand, V. Pejovic, E. M. Belding and D. L. Johnson, “VillageCell: Cost Effective Cellular Connectivity in Rural Areas”, To appear in ICTD’12, Atlanta, Georgia, March 2012. [11] A. Dhananjay, M. Tierney, J. Li and L. Subramanian, “WiRE: a new rural connectivity paradigm,” Proceedings of the ACM SIGCOMM 2011, pp. 462-463. [12] T. Yucek and H. Arslan, “A survey of spectrum sensing algorithms for cognitive radio applications,” IEEE Communications Surveys & Tutorials, Volume: 11 Issue:1. First Quarter 2009. [13] D.D. Ariananda, M.K. Lakshmanan and H. Nikoo, “A survey on spectrum sensing techniques for Cognitive Radio,” Second International Workshop on Cognitive Radio and Advanced Spectrum Management, May 2009. CogART 2009. [14] 3GPP TS 05.05 version 8.20.0 Release 1999. [On-Line]. Available: http://www.3gpp.org/specs/numbering.htm