Measurement Based Capacity Scavenging via Whitespace Modeling ...
Measurement Based Capacity Scavenging via Whitespace Modeling in Wireless Networks Anthony Plummer Jr., Mahmoud Taghizadeh and Subir Biswas Department of Electrical and Computer Engineering Michigan State University, USA email: {plumme23, taghizad, sbiswas}@msu.edu
Abstract—Dynamic Spectrum Access can enable secondary network users to access unused spectrum, or whitespace, which is found between the transmissions of primary users in a wireless network. The main design objectives for secondary user access strategy are to be able to ¨scavenge¨ spatio-temporally fragmented whitespace opportunities while limiting the amount of interference caused to the primary users. In this paper, we propose a novel secondary user access strategy which is based on measurement and modeling of the whitespace as perceived by the secondary network users. A secondary user continually monitors its surrounding whitespace, models it, and then attempts to access the available spectrum holes so that the effective secondary throughput is maximized while the resulting interference to the primary users is limited to a pre-defined bound. We first develop analytical expressions for the secondary throughput and primary interference, and then perform ns2 based simulation experiments to validate the effectiveness of the proposed access strategy, and evaluate its performance numerically using the developed expressions. Index Terms—Dynamic Capacity Scavenging, Cognitive Networks, Secondary User Access Strategy, Whitespace Modeling.
I. I NTRODUCTION
L
ICENSED primary users have been able to freely utilize the available spectrum due to static spectrum assignments by government organizations such as the FCC in the United States. With the continual increase in usage of wireless devices this static assignment can lead to underutilized spectrum. Unused spectrum reduces the efficiency of the finite spectrum resources. Research into the dynamic spectrum access paradigm has provided the opportunity for secondary users (SU) with a cognitive radio to access underutilized spectrum between primary users (PU) transmissions. Allowing an SU to utilize the idle spectrum would increase the efficiency of the spectrum usage and stem the effects of increasing wireless devices and decreasing static spectrum availability. It has been shown that SU access of underutilized spectrum is feasible through studies of the spectrum usage of PUs and the policies and protocols that govern them [1], [2]. An SU may experience different PU networks or allocations depending on the purpose and application of the network. For example, a PU network could be TV bands where the initial task of the SU is to discover the frequencies that are currently not used by the PUs (television stations). Then the SU would access those frequencies that are not interfering with
1 This work was partially supported by a grant from the National Science Foundation (CMMI-0800103)
the PUs [1], [3]. This represents SU interacting with PUs in scenarios when the PUs’ idle periods or whitespaces last for long durations, typically of the order of hours. An example of the other extreme of whitespaces lasting for much shorter durations (of the order of milliseconds) would be when the SUs attempt opportunistic spectrum access in a PU network operating packet-based MAC protocols. The SUs, in these scenarios will attempt transmissions in between PUs’ packet transmissions [2]. We term this opportunistic access during those ultra-short and non-deterministic whitespaces as bandwidth scavenging by the secondary users. In other words, the SUs scavenge capacity left over by the PUs. In this paper, we focus on the latter of the two examples by proposing a secondary user channel access strategy for efficient bandwidth scavenging. Using an access strategy for the SU is important to satisfy two major goals. First is to minimize the amount of interference that is experienced by the PUs when the SUs are accessing available whitespaces. The second goal is to maximize the secondary user throughput using the available whitespace, without compromising the first goal. The steps adopted in designing an access strategy are to first discover and measure the whitespaces available in the PU transmissions. Second is to model the whitespace over time, which is based on the SU view of the whitespace periods between PU transmissions. Finally, to develop an access strategy based on the model to allow SU to access the whitespace. The contributions of this paper are as follows. First, we develop a practical whitespace measurement scheme that can be used by each secondary user to assess the instantaneous spectrum holes due to the primary user behavior in its immediate neighborhood. Second, a model of the whitespace as measured by an SU is developed for statistically interpreting the available whitespace and identifying effective transmission opportunities. This modeling mechanism is independent of the MAC and routing layer PU protocols and the PU traffic characteristics. Third, a comprehensive channel access strategy based on the previously mentioned whitespace model is developed to facilitate capacity scavenging by the SUs. In this access strategy, the SU throughput for a given whitespace is attempted to be maximized in the presence of a pre-defined bound on the interference that an SU can cause on the PUs in its neighborhood. Finally, we experimentally demonstrate that the proposed access strategy is able to accomplish the targeted goals of high SU throughput and low interference on primary users for multiple traffic characteristics.
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
II. R ELATED W ORK Approaches to the design of secondary network access strategies have taken many forms. The authors in [2], use the assumption that the PU network is based on an 802.11 MAC. Their physical layer measurements of the PU transmissions are modeled as a semi-Markov process based on the transition states of DATA, SIFS, and ACK of the 802.11 syntax. In our work we do not assume that the PU MAC layer is known to the SU. The access strategy we propose takes a completely measurement based approach which allows the SU to operate and be flexible in varying PU environments and protocols. Game theoretic designs which use game theory to provide a framework for spectrum usage between PUs and SUs have been introduced in [4], [5]. In this work, the PUs cooperate with the SUs through pricing strategies and subscription fees to facilitate spectrum sharing. These approaches require the PUs to have knowledge of the SUs presence, and to assist them to access the spectrum. In our work, there is no cooperation requirement for the PUs. Thus our strategy will not require any changes to the PUs behavior. The idea here is to provide an access mechanism for the SUs for scavenging PUs’ leftover capacity, without the PUs being aware of such scavenging. In [6], [7], the proposed access strategies rely on specific types of PU networks. Protocols are developed in [6] for a slotted PU network where the SUs are synchronized with the PU’s slots. In this case, if a PU does not use a slot then the SU can transmit after a small duration of time in the same slot. The PUs in [7] are assumed to use a specific combination of time and frequency multiplexing to prevent inter-PU access collisions. In this paper, we formulate the PU network more generally by not assuming any specific PU MAC layer such as the slotted ones in [6]. The target here is to develop an SU access mechanism which can deal with unslotted as well as stochastic PU MAC layers such as 802.11, CSMA and ALOHA protocols. This assumption of a general PU access protocol broadens the applicability of the SU access approach proposed in this paper. The authors in [8], [9], do not restrict PUs to a slotted MAC protocol although their access strategy requires a slotted SU network. During the beginning of each slot the SU senses the channel and if the channel is free then the SU transmits. A greedy version also uses probabilistic analysis to decide whether or not to transmit in a given slot. This framework requires time synchronization across the SUs. In our approach, SUs use a stochastic mechanism and are not required to maintain time synchronization. III. BANDWIDTH S CAVENGING : C ONCEPT S UMMARY From an SU’s standpoint, once a whitespace is identified, the next step is to send a packet in that whitespace, only if the estimated chance of completing the transmission before the whitespace ends is high. The quality of whitespace access, that is capacity scavenging in this context, is determined by the resulting Effective SU Throughput (EST) which should be maximized, and the Primary User Interference (PUI), which should be minimized for a given whitespace profile. Under the present model of non-cooperating primary users, the SU whitespace access strategy depends solely on the
Figure 1: Different whitespace profiles and their CDFs
statistical profile of the whitespace, which in turn, is determined by the topology, traffic and routing characteristics of the primary user network. Depending on these factors, the available whitespace can have widely varying characteristics. For example, the neighborhood of an SU may experience a large number of whitespace slots but each whitespace lasts only for a short duration (see profile-1 in Figure 1.). Conversely, there can be only a few whitespaces, but each with very long durations, as shown in profile-2 in Figure 1 The cumulative distribution functions (CDF) of the two whitespace profiles are also shown in the bottom part of the figure. These two different whitespace profiles indicate that even when the average white space length are the same, for a given target Primary User Interference (PUI), profile-1 will be able to support less number of fixed length SU packet transmissions due to a higher level of capacity fragmentation during the ends of the whitespaces. Therefore, the Effective SU Throughput (EST) for profile-1 will be less than that for profile-2, for the same level of target PUI. With non-deterministic primary traffic patterns, an SU is not able to deterministically predict when a detected whitespace slot will end. As a result, the SU access strategy cannot be deterministic either. The best an SU can perform is to access a given whitespace based on a previously observed statistical whitespace profile. An aggressive access may result in higher Effective SU Throughput (EST), but can also increase Primary User Interference (PUI) which happens when an SU’s transmission (the last packet transmission in a whitespace) does not end before the completion of the whitespace in question. A conservative access strategy may produce a reverse effect, meaning a low PUI, but at the expense of an EST that is lower than what could be achieved for a given whitespace profile. Therefore, the objective for an SU is to be able to make a transmission decision based on previously observed whitespace statistics, such that by prudently transmitting packets, it is able to maximize the achievable EST and minimize the PUI. The primary contribution of this paper will be to develop such decision mechanisms. Note that traditionally, interference on another user of a network is defined as a user attempting to transmit in the middle of another user’s transmission. In the framework assumed in this paper, however, when an SU initially accesses a whitespace it is assumed that the PU will not be transmitting at that time; otherwise it is not considered a whitespace by the SU in the first place. Therefore the traditional definition of PUI does not hold in this case. In this paradigm, PUI can
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
be defined as when an SU transmits during a whitespace, and the time needed to successfully complete the transmission is more than how long the current whitespace will last purely from the PU traffic standpoint. This modified definition takes into account the fact that whitespaces are of fixed duration, and that once it ends the PUs’ transmissions will be affected in that the PUs will be prevented from transmitting. The key assumptions made in this paper are: 1) the ability of continuous channel sensing [2] by the secondary users, 2) no inter-secondary user channel access conflicts [10], 3) the ability of secondary users to differentiate between the PU and SU transmissions through physical layer waveform detection [11], 4) the ability of an SU to detect when a PU wants to transmit, and to vacate a whitespace [12], and finally, 5) fixed secondary packet size. IV. W HITESPACE M EASUREMENT The available whitespace can be measured by a secondary user by detecting the received signal strength (RSSI) at a given channel frequency. Durations when the measured RSSI are below a pre-defined threshold are considered as whitespaces. Such whitespaces are a function of the physical locations, topology, traffic profile, and the MAC and routing protocols used by the primary users. Interpretation of such measured whitespace depends on the channel sensing periodicity, which is considered to be very high (perfect sensing) for all results presented in this paper. The proposed method of capturing whitespace and modeling could be extended to work in environments where sensing is done with finite periodicity. Figure 2 shows a simple network of primary users and two secondary users which are within up to the interference range of the primary users. In Figure 3, we show the RSSI observed by the SU1 as a result of a data flow from PU1 to PU5 . These results are collected from an 802.11 simulation run in ns2 network simulator. The traffic flow consists of UDP data at 10 packets per second with a packet duration of 2.43 ms. Figure 3 shows the RSSI trace, as observed by SU1 , as a result of the packet flows from PU1 to PU5 . Observe that since SU1 is physically closest to the primary user PU3, the transmissions from that primary user appears with highest signal strength compared to the other PUs whose signal strength monotonically reduces with increasing distances. Also, since the PUs in this experiment use 802.11 as the MAC protocol, each packet transmission creates an RTS-CTS-DATA-ACK cycle, resulting in white spaces of multiple durations representing different 802.11 inter frame spacing (i.e. SIFS, DIFS, etc.). Three types of whitespaces can be observed for this example 802.11 primary traffic. As pointed in Figure 3, the first is the control whitespace caused by the short inter frame spacing (SIFS) periods between the RTS, CTS, DATA, and ACK transmissions. The second whitespace type, termed as forwarding whitespace, represents the duration between a PU receiving data and forwarding data. The last whitespace type is the inter-packet whitespace, representing the inter packet duration which can vary greatly because it is based on the rate of packet transmission by the PUs. V. W HITESPACE S TATISTIICS The measured whitespace can be expressed [2] by a probability density function (PDF) w(t), in which the continuous
Figure 2: Example primary topology with secondary locations
Figure 3: RSSI for primary traffic as viewed by SU1 in Figure 2
Figure 4: PDF and CDF of a whitespace trace
random variable t represents the duration of individual white spaces. This function can capture the impacts of all primary user properties including physical location, network topology, traffic characteristics, and MAC/Routing protocols. In other words, once the w(t) for a measured whitespace is computed, the SUs can rely solely on this function for optimal channel access without having to worry about the individual PU properties. Figure 4 shows the PDF w(t) and the corresponding Cumulative Distribution Function (CDF) W(t) of a whitespace trace that resulted from a PU1 -to-PU5 data stream (see Figure 2) with an average rate of 20 packets per second, and the interpacket interval uniformly distributed between ±50% of the average rate. This whitespace is observed by SU1 in Figure 2. For clarity, the PDF graph is split into two sub-graphs, showing the ranges of durations where the probability is nonzero (between 0-1 ms and 9.5 to 62 ms). The probability is zero for all other whitespace durations. A number of key observations can be made from the whitespace statistics shown in Figure 4. First, the detected whitespaces from the measured RSSI trace vary a great deal in terms of their durations. Second, for this particular trace, created by 802.11 primary traffic, a vast majority of the whitespaces (more than 95%) last for less than 1 ms. The rest of the 5% is uniformly distributed between 10 ms and 60 ms. The initial peak in the PDF indicates the control and the forwarding whitespaces as defined in Section IV. The larger whitespaces are mainly from the inter-packet intervals which
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
Algorithm 1: MBAS Scheme if (channel free) then while (channel remains free) do determine , the duration spent in the current whitespace; if ( ≤ μ) then continue; else ∞ if ( w(t) dt ≥ 1 − I) then send packet; +S
else break; end end end
have the 10 ms to 60 ms spread created by the jitter in the packet generation by PU1 in Figure 2. These observations indicate that an efficient access strategy for the secondary users will have to avoid transmitting during any small whitespaces, but access the longer duration whitespaces. Without having the knowledge of the actual duration of a whitespace at the beginning of the whitespace, an SU needs to rely on statistical measures in order to maximize the EST while keeping the PUI within tolerable bounds. The access strategy in the next section attempts to accomplish these goals. VI. C HANNEL ACCESS S TRATEGY A. Access Logic The access strategy comprises of three main steps. First, when an SU intends to make a packet transmission it waits till the next white space is detected. Second, at any given point within a detected whitespace, an SU will transmit only when the estimated probability of interference, computed based on the measured whitespace PDF w(t), is lower than a pre-defined threshold. The third component requires the SU to wait for a minimum wait-threshold duration μ even before it estimates the interference probability. This minimum wait-threshold duration μ is required for considering the initial peak of the whitespace PDF function as shown in Figure 4. This peak indicates that for 802.11 style primary traffic, due to the inter-frame spacing at the MAC layer, over almost 95% of the detected whitespaces are too small to fit the secondary packets. As a result, if an SU transmits at the beginning of each whitespace then a vast majority of them will end up in primary interference. However, deferring the transmission attempts for a suitably chosen waitthreshold duration μ can potentially reduce such Primary User Interference (PUI) without scarifying the Effective Secondary Throughput (EST). After the initial wait-threshold duration μ, if a packet is transmitted at time since the beginning of a whitespace, the probability of interference can be computed using the quantities and the measured whitespace PDF w(t). If the computed PUI is less than a pre-set bound I then the SU makes a packet transmission. After the packet is transmitted, if the SU has more packets to send, it re-computes the interference with a newly measured value of , and the process continues until the probability of interference becomes greater than the chosen bound, I. The channel access strategy is summarized in the form of a pseudo code in Algorithm 1. Note that at the point of time when the remaining whitespace reduces to a level so that the expected interference is
no longer less than the bound I, the secondary user abandons transmission attempts and starts channel sensing for detecting the next whitespace. B. Primary User Interference (PUI) PUI for an access strategy is defined as the probability that an SU will cause interference (as defined in Section III) to PUs during a white space. According to the access strategy described in Algorithm 1, after a whitespace starts, a secondary user waits for a period μ before it attempts transmission, and continues to transmit packets until the expected PUI reaches a value that is larger than the pre-defined bound I. Let us define a term Jmax , which denotes the maximum number of packets that the secondary user can send before the expected interference reaches bound I. There are three access scenarios that can arise. First, a whitespace duration is less than or equal to μ, in which case the whitespace is not accessed by the SU, and therefore, there is no interference to the PUs. Second, a whitespace duration is in the range between μ and μ+Jmax S+(Jmax -1)γ, where S is the secondary packet duration and γ is the inter-packet duration for the secondary traffic. In this case, there will be PUI since the last packet will last at least till the end of the mark of the whitespace. The third scenario corresponds to when a whitespace duration is greater than μ+Jmax S+(Jmax 1)γ. No PUI will be caused in this case because the SU will vacate or exit the whitespace after Jmax transmissions. Thus, for a given whitespace profile w(t), the PUI can be defined by the probability that the whitespace duration is in the range between μ and μ+Jmax S+(Jmax -1)γ. Therefore, an SU accessing a whitespace characterized by PDF w(t) will create primary user interface (PUI): μ+Jmax S+(J max −1)γ
PUI =
w(t) dt
(1)
μ
According the access strategy proposed in Algorithm 1, the number of packets the SU is allowed to transmit in a given whitespace, Jmax , is determined by finding the maximum j that satisfies the inequality: ∞ 1−
w(t) dt ≤ I
(2)
μ+(j−1)(S+γ)
The left side of the above Equation 2 represents the PUI when the SU sends Jmax packets in a given whitespace. Therefore, the inequality in (2) gives the Jmax that will reduce the interference to the pre-defined bound I. By plugging in this value of Jmax in Equation 1, we can find the overall PUI caused by our proposed access strategy. C. Effective Secondary Throughput (EST) The Effective Secondary user Throughput (EST) is defined as the number of packets successfully transmitted per whitespace. To model EST we first define Sj , the probability of sending exactly j packets in a whitespace. This corresponds to
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
the event in which the jth packet from an SU in a whitespace has interfered with the PUs, i.e. the whitespace had ended during the transmission of the jth packet. The quantity Sj can be written as: μ+jS+(j−1)γ
Sj =
w(t) dt
(3)
μ+(j−1)(S+γ)
For a whitespace that lasts between μ and μ+(Jmax -1)(S+γ) the number of packets sent by an SU is j. For whitespace duration between μ+Jmax S+(Jmax -1)γ and ∞, the number of packets sent by an SU is Jmax , since the secondary vacates the whitespace after sending Jmax packets. Therefore EST, the expected number of secondary packets sent for a given whitespace can be expressed as: EST =
Jmax −1
∞
jSj +
j=1
Jmax Sj
Figure 5: The interaction between µ and the PDF w(t)
(4)
j=Jmax
Where j is the number of transmitted SU packets, W is the number of whitespaces per unit time, and Sj is the probability that an SU will transmit j packets in a whitespace. D. Dimensioning Wait-threshold μ The selection of wait-threshold μ is critical since it impacts both PUI and EST as shown in Equations 1 and 4. A large μ can reduce primary user interference by preventing the SUs from transmitting during very short whitespaces (see Figure 5). But a large μ can also bring down the Effective Secondary Throughput (EST), since some portions of large whitespaces will be lost due to this conservative wait period. These two observations are consistent with the dependencies of PUI and EST on μ, as captured in Equations 1 and 4. The goal is to optimally pre-dimension the parameter μ, based on the measured PDF function w(t), in order to strike a desirable balance between the PUI and EST. According to the access strategy described in Algorithm 1, the wait-threshold period determines the PUI caused by the first SU packet during a whitespace. For a wait-threshold tx , such PUI caused by the first SU packet can be written as: t x +S
f (tx ) =
w(t)dt
(5)
tx
Now, the minimum and the maximum values of tx , are zero and 2S, where S is the secondary packet duration. The maximum wait-threshold is 2S, since waiting for a duration that is greater than or equal to 2S would mean missing out sufficient whitespace that could have been used by the secondary user to send at least one packet. Therefore, for a given whitespace w(t), the optimal waitthreshold μ can be chosen as the smallest tx , over the range 0 to 2S, such that the quantity f(tx ) in Equation 5 is minimized. It should be noted that there may multiple values of tx that cause minimal first-packet interference, but by selecting the smallest tx , the EST can be maximized.
Figure 6: Evaluation of the impacts of µ and interference bound I
To summarize, after a secondary user profiles its observed whitespace as the PDF w(t), it first chooses an optimal waitthreshold μ that minimizes the first-packet interference while maximizing the EST. It then computes the maximum number of possible transmissions per whitespace (Jmax ) for a given PUI bound I from Equation 2. The SU then executes the access mechanism in Algorithm 1 with the computed parameters μ and Jmax . The resulting PUI and EST can then be computed using Equations 1 and 4. VII. ACCESS S TRATEGY E VALUATION Performance of the proposed access strategy has been evaluated both numerically and experimentally (using ns2 simulator) in terms of its Primary user Interface (PUI) and the Effective Secondary Throughput (EST). The network simulator ns2 is used to create a network topology as shown in Figure 2. A multi-hop primary traffic flow from node PU1 to PU5 is created. As described in Section IV, the primary users use 802.11 as the MAC layer protocol, and the resulting whitespace profile as observed by the secondary user SU1 is shown in Figure 3. All secondary traffic in this Section corresponds to those from nodes SU1 to SU2 . A. Impacts of wait-threhold μ and interference bound I The wait-threshold μ determines the duration an SU waits before transmitting in a whitespace, and the tolerable interference bound I determines the quantity Jmax , the number of packets an SU sends before vacating a whitespace. The simulation experimental results in Figure 6 correspond to a PU traffic flow with an average transmission rate of 20
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
Figure 7: Impacts of Secondary Packet Size
Figure 8: Effects of primary user traffic load
B. Impacts of Secondary User Packet Size packets per second (pps) and a 20% uniformly distributed inter-packet transmission variation. The SU packet duration S was chosen to be 1.215 ms, which is half the used primary packet duration. The results are produced using ns2 simulation. Observe that for all μ, a higher interference bound I provide a higher secondary throughput, although at the expense of increased primary interference. For the most liberal case I = 1.0, indicating that an SU never vacates a whitespace before it interferes with a PU, the EST is the highest since the SU in this case is able to exploit the entire whitespace. As expected, as the interference bound I is made more conservative (smaller), both PUI and the EST values are brought down. For μ = 0 ms, representing the case when the SU transmits immediately at the start of a whitespace, the amount of PUI is very high and the EST is low for all interference bounds. This is because as shown in Figure 4, most of the whitespaces are of very short duration for 802.11 primary traffic. As μ increases, the PUI first decreases abruptly and then relatively moderately, but in a monotonic manner. With increasing μ, the EST first shows an increasing trend up to a threshold value (0.715 ms in Figure 6), beyond which it either falls slightly or remains steady. This threshold μ provides the best EST while minimizing the PUI. For a given w(t), This can be computed from Equation 5 in Section VI-D. EST is measured in number of secondary packets successfully sent per whitespace. With μ set to zero, the SU attempts packet transmission in all whitespaces. However, such transmissions succeed only for a small percentage of the whitespaces, since the w(t) in Figure 4 indicates that a large number of whitespaces are smaller than the chosen secondary packet duration of 1.215 ms. That is why the EST is very low for zero μ. With increasing μ, small whitespaces are avoided, causing an increase in the EST. Once μ reaches the threshold value, defined by Equation 5, the EST reaches its peak. The values of μ, larger than that threshold, cause the access strategy to be over conservative, thus missing out useful whitespace that could contribute to successful secondary packet transmissions. This explains the downward trends in the EST values beyond the μ threshold.
Figure 7 shows the changes in primary interference and secondary throughput as a result of varying secondary packet durations. These results correspond to a wait-threshold of 0.715 ms, with back-to-back secondary packets (i.e. γ is set to zero). Observe that with increasing secondary packet duration (i.e. size) the EST slightly increases. This is because the last transmitted secondary packet (Jmax th packet) manages to use more amount of whitespace with larger packet durations. But it also means that in more occasions, the SU interferes with the primary users since with larger packets, it has higher chances of not finishing the Jmax th packet transmission before the white space ends. This explains why the PUI also goes up, although slightly, with increasing packet size. Figure 7 presents both experimentally measured (via ns2 simulation) and theoretical values (using Equations 1 through 4), which are marked as EXP and THY respectively. It is evident that the experimental results match very well with the theoretical results. Furthermore, the figure shows that with a more conservative I, the EST is lower than with a larger I. Also, the results for all I values indicate that the access strategy is indeed able to contain the primary user interference within the pre-defined interference bound I, for all experimented secondary packet sizes. C. Effects of Primary User Traffic Load For the next two experiments we use the optimal value of μ = 0.715 ms which was calculated using Equation 5, and was experimentally validated using results presented in Figure 6. The secondary packet duration is 1.215 ms which represents half of the PU packet duration. Figure 8 demonstrates the throughput and interference variation with different primary user packet rates, spanning across 1 pps to 33 pps. For all the primary packet rates, a 20% uniformly distributed inter-packet transmission variation was introduced. As expected, at low PU traffic rates, the secondary throughput is high due to lot of available whitespaces. With increasing PU packet rates, however, the EST comes down quite drastically. This reduction is due to a left-shift in the whitespace
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
Figure 9: Impacts of primary traffic rate variability
PDF w(t), and the subsequent reduction in the maximum number of transmitted secondary packets per whitespace (Jmax ), as computed in Equation 2. It is worth noting that the primary interference (PUI) is quite low and, more importantly, relatively insensitive to the primary packet rates. This actually shows that with varying primary packet rates, by appropriately computing Jmax , the secondary user is able to vacate the whitespace, thus keeping the PUI insulated from the primary packet rates. This demonstrates the flexibility of the proposed access strategy. D. The Variability in Primary User Packet Transmissions The results in Figure 9 report primary user interference and secondary throughput with different amounts of variability in the inter-packet spacing for the primary users. All graphs in Figure 9 corresponds to an average data rate of 10 packets per second, with a packet spread factor ranging from 0 to 1. A spread factor of 0 indicates that the inter-packet interval for all primary packets is strictly 100 ms. A 0.2 spread factor means a 20% (of the average interval of 100 ms) uniformly distributed inter-packet interval variation. The secondary packet duration, and the primary packet durations were chosen to be 1.215 ms, and 2.43 ms respectively. For small spread factors, since the whitespace PDF is highly modal (a narrow probability peak in the PDF), the maximum transmittable packet count Jmax , as defined in Equation 2, can be customized for that narrow pick. Thus the available whitespace is very well utilized by the SU with high efficiency, leading to high EST. As the spread factor increases, wider and shorter peaks form in the PDF, leading to Jmax computation that needs to be conservative due to the presence of shorter whitespaces. As a result, the SU may often evacuate a whitespace much earlier than needed, leading to lower secondary throughputs. This explains as to why the EST reduces with larger spread factors, for all interference bound I. Due to the same reason, the PUI reduces with higher spread factor. It reduces rather rapidly initially, but then somewhat settles down to a smaller value, indicating prudent Jmax computation by the access mechanism in Algorithm 1.
of the whitespace as perceived by the secondary network users. A secondary user continually monitors its surrounding whitespace, models it, and then attempts to access the available spectrum holes so that the effective secondary throughput is maximized while the resulting interference to the primary users is kept limited to a pre-defined bound. The strategy is general in that it does not reply on the information about the primary network protocols. Once the surrounding whitespace of a secondary user is modeled by the user, it simply executes the proposed access strategy without having to rely on primary cooperation or any information about specific primary protocols. Through ns2 simulation experiments and developed analytical expressions it has been demonstrated that with 802.11 MAC traffic, the access strategy is able to provide reasonable amount of secondary throughput while bounding the primary interference to pre-set values. Our ongoing work on this topic includes, 1) relaxing the assumptions as listed in Section III, 2) using more elaborate and larger network topologies, and 3) using a wider range of primary MAC protocols including CSMA, ALOHA and other stochastic protocols to establish the general applicability of the proposed concept. R EFERENCES [1] V. R. Petty, R. Rajbanshi, D. Datla, W. Frederick, D. Daniel, et.al, ¨Feasibility of Dynamic Spectrum Access in Underutilized Television Bands,¨ in New Frontiers in Dynamic Spectrum Access Networks. DySPAN 2007., Apr. 2007. [2] S. Geirhofer, L. Tong, and B. M. Sadler, ¨Dynamic Spectrum Access in WLAN Channels: Empirical Model and Its Stochastic Analysis,¨ in ACM Workshop on Technology and Policy for Accessing Spectrum (TAPAS), August 2006. [3] Y. Yuan, P. Bahl, R. Chandra, P. A. Chou, J. I. Ferrell, et al., ¨KNOWS: Cognitive Radio Networks Over White Spaces,¨ in New Frontiers in Dynamic Spectrum Access Networks. DySPAN 2007., Apr. 2007. ¨ [4] I. Stanojev, O. Simeone, Y. Bar-Ness, and T. Yu, Spectrum Leasing via Distributed Cooperation in Cognitive Radio,¨ın IEEE International Conference on Communications. ICC ’08. , May 2008. [5] A. O. Ercan, L. Jiwoong, S. Pollin, and J. M. Rabaey, ¨A Revenue Enhancing Stackelberg Game for Owners in Opportunistic Spectrum Access,¨ in New Frontiers in Dynamic Spectrum Access Networks, (DySPAN 2008), Oct. 2008. [6] C. Yunxia, Z. Qing, and A. Swami, ¨Joint Design and Separation Principle for Opportunistic Spectrum Access in the Presence of Sensing Errors,¨ IEEE Transactions on Information Theory, vol. 54, no. 5, May 2008. [7] E. Jung and L. Xin, ¨Opportunistic Spectrum Access in Heterogeneous User Environments,¨ in New Frontiers in Dynamic Spectrum Access Networks. DySPAN 2008., Oct. 2008. [8] Q. Zhao, S. Geirhofer, L. Tong, and B. M. Sadler, ¨Optimal Dynamic Spectrum Access via Periodic Channel Sensing,¨ in Proc. IEEE Wireless Communications and Networking Conference (WCNC), March 2007. [9] S. Geirhofer, L. Tong, and B. M. Sadler, ¨Cognitive Medium Access: Constraining Interference Based on Experimental Models,¨ IEEE Jrnl on Selected Areas in Communications, vol. 26, no. 1, Jan. 2007. [10] M. Wellens, J. Riihijarvi, M. Gordziel, and P. Mahonen, ¨Evaluation of Cooperative Spectrum Sensing Based on Large Scale Measurements,¨ in New Frontiers in Dynamic Spectrum Access Networks. DySPAN 2008., Oct. 2008. [11] D. Cabric, S. M. Mishra, and R. W. Brodersen, ¨Implementation Issues in Spectrum Sensing for Cognitive Radios,¨ in Conference on Signals, Systems and Computers, Nov. 2004. [12] L. Xin and D. Zhi, ¨ESCAPE: A Channel Evacuation Protocol for Spectrum-Agile Networks,¨ in New Frontiers in Dynamic Spectrum Access Networks. DySPAN 2007., Apr. 2007.
VIII. C ONCLUSIONS In this paper, we have proposed a novel secondary user access strategy which is based on measurement and modeling 978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
Abstract—Dynamic Spectrum Access can enable secondary network users to access unused spectrum, or whitespace, which is found between the transmissions of primary users in a wireless network. The main design objectives for secondary user access strategy are to be able to ¨scavenge¨ spatio-temporally fragmented whitespace opportunities while limiting the amount of interference caused to the primary users. In this paper, we propose a novel secondary user access strategy which is based on measurement and modeling of the whitespace as perceived by the secondary network users. A secondary user continually monitors its surrounding whitespace, models it, and then attempts to access the available spectrum holes so that the effective secondary throughput is maximized while the resulting interference to the primary users is limited to a pre-defined bound. We first develop analytical expressions for the secondary throughput and primary interference, and then perform ns2 based simulation experiments to validate the effectiveness of the proposed access strategy, and evaluate its performance numerically using the developed expressions. Index Terms—Dynamic Capacity Scavenging, Cognitive Networks, Secondary User Access Strategy, Whitespace Modeling.
I. I NTRODUCTION
L
ICENSED primary users have been able to freely utilize the available spectrum due to static spectrum assignments by government organizations such as the FCC in the United States. With the continual increase in usage of wireless devices this static assignment can lead to underutilized spectrum. Unused spectrum reduces the efficiency of the finite spectrum resources. Research into the dynamic spectrum access paradigm has provided the opportunity for secondary users (SU) with a cognitive radio to access underutilized spectrum between primary users (PU) transmissions. Allowing an SU to utilize the idle spectrum would increase the efficiency of the spectrum usage and stem the effects of increasing wireless devices and decreasing static spectrum availability. It has been shown that SU access of underutilized spectrum is feasible through studies of the spectrum usage of PUs and the policies and protocols that govern them [1], [2]. An SU may experience different PU networks or allocations depending on the purpose and application of the network. For example, a PU network could be TV bands where the initial task of the SU is to discover the frequencies that are currently not used by the PUs (television stations). Then the SU would access those frequencies that are not interfering with
1 This work was partially supported by a grant from the National Science Foundation (CMMI-0800103)
the PUs [1], [3]. This represents SU interacting with PUs in scenarios when the PUs’ idle periods or whitespaces last for long durations, typically of the order of hours. An example of the other extreme of whitespaces lasting for much shorter durations (of the order of milliseconds) would be when the SUs attempt opportunistic spectrum access in a PU network operating packet-based MAC protocols. The SUs, in these scenarios will attempt transmissions in between PUs’ packet transmissions [2]. We term this opportunistic access during those ultra-short and non-deterministic whitespaces as bandwidth scavenging by the secondary users. In other words, the SUs scavenge capacity left over by the PUs. In this paper, we focus on the latter of the two examples by proposing a secondary user channel access strategy for efficient bandwidth scavenging. Using an access strategy for the SU is important to satisfy two major goals. First is to minimize the amount of interference that is experienced by the PUs when the SUs are accessing available whitespaces. The second goal is to maximize the secondary user throughput using the available whitespace, without compromising the first goal. The steps adopted in designing an access strategy are to first discover and measure the whitespaces available in the PU transmissions. Second is to model the whitespace over time, which is based on the SU view of the whitespace periods between PU transmissions. Finally, to develop an access strategy based on the model to allow SU to access the whitespace. The contributions of this paper are as follows. First, we develop a practical whitespace measurement scheme that can be used by each secondary user to assess the instantaneous spectrum holes due to the primary user behavior in its immediate neighborhood. Second, a model of the whitespace as measured by an SU is developed for statistically interpreting the available whitespace and identifying effective transmission opportunities. This modeling mechanism is independent of the MAC and routing layer PU protocols and the PU traffic characteristics. Third, a comprehensive channel access strategy based on the previously mentioned whitespace model is developed to facilitate capacity scavenging by the SUs. In this access strategy, the SU throughput for a given whitespace is attempted to be maximized in the presence of a pre-defined bound on the interference that an SU can cause on the PUs in its neighborhood. Finally, we experimentally demonstrate that the proposed access strategy is able to accomplish the targeted goals of high SU throughput and low interference on primary users for multiple traffic characteristics.
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
II. R ELATED W ORK Approaches to the design of secondary network access strategies have taken many forms. The authors in [2], use the assumption that the PU network is based on an 802.11 MAC. Their physical layer measurements of the PU transmissions are modeled as a semi-Markov process based on the transition states of DATA, SIFS, and ACK of the 802.11 syntax. In our work we do not assume that the PU MAC layer is known to the SU. The access strategy we propose takes a completely measurement based approach which allows the SU to operate and be flexible in varying PU environments and protocols. Game theoretic designs which use game theory to provide a framework for spectrum usage between PUs and SUs have been introduced in [4], [5]. In this work, the PUs cooperate with the SUs through pricing strategies and subscription fees to facilitate spectrum sharing. These approaches require the PUs to have knowledge of the SUs presence, and to assist them to access the spectrum. In our work, there is no cooperation requirement for the PUs. Thus our strategy will not require any changes to the PUs behavior. The idea here is to provide an access mechanism for the SUs for scavenging PUs’ leftover capacity, without the PUs being aware of such scavenging. In [6], [7], the proposed access strategies rely on specific types of PU networks. Protocols are developed in [6] for a slotted PU network where the SUs are synchronized with the PU’s slots. In this case, if a PU does not use a slot then the SU can transmit after a small duration of time in the same slot. The PUs in [7] are assumed to use a specific combination of time and frequency multiplexing to prevent inter-PU access collisions. In this paper, we formulate the PU network more generally by not assuming any specific PU MAC layer such as the slotted ones in [6]. The target here is to develop an SU access mechanism which can deal with unslotted as well as stochastic PU MAC layers such as 802.11, CSMA and ALOHA protocols. This assumption of a general PU access protocol broadens the applicability of the SU access approach proposed in this paper. The authors in [8], [9], do not restrict PUs to a slotted MAC protocol although their access strategy requires a slotted SU network. During the beginning of each slot the SU senses the channel and if the channel is free then the SU transmits. A greedy version also uses probabilistic analysis to decide whether or not to transmit in a given slot. This framework requires time synchronization across the SUs. In our approach, SUs use a stochastic mechanism and are not required to maintain time synchronization. III. BANDWIDTH S CAVENGING : C ONCEPT S UMMARY From an SU’s standpoint, once a whitespace is identified, the next step is to send a packet in that whitespace, only if the estimated chance of completing the transmission before the whitespace ends is high. The quality of whitespace access, that is capacity scavenging in this context, is determined by the resulting Effective SU Throughput (EST) which should be maximized, and the Primary User Interference (PUI), which should be minimized for a given whitespace profile. Under the present model of non-cooperating primary users, the SU whitespace access strategy depends solely on the
Figure 1: Different whitespace profiles and their CDFs
statistical profile of the whitespace, which in turn, is determined by the topology, traffic and routing characteristics of the primary user network. Depending on these factors, the available whitespace can have widely varying characteristics. For example, the neighborhood of an SU may experience a large number of whitespace slots but each whitespace lasts only for a short duration (see profile-1 in Figure 1.). Conversely, there can be only a few whitespaces, but each with very long durations, as shown in profile-2 in Figure 1 The cumulative distribution functions (CDF) of the two whitespace profiles are also shown in the bottom part of the figure. These two different whitespace profiles indicate that even when the average white space length are the same, for a given target Primary User Interference (PUI), profile-1 will be able to support less number of fixed length SU packet transmissions due to a higher level of capacity fragmentation during the ends of the whitespaces. Therefore, the Effective SU Throughput (EST) for profile-1 will be less than that for profile-2, for the same level of target PUI. With non-deterministic primary traffic patterns, an SU is not able to deterministically predict when a detected whitespace slot will end. As a result, the SU access strategy cannot be deterministic either. The best an SU can perform is to access a given whitespace based on a previously observed statistical whitespace profile. An aggressive access may result in higher Effective SU Throughput (EST), but can also increase Primary User Interference (PUI) which happens when an SU’s transmission (the last packet transmission in a whitespace) does not end before the completion of the whitespace in question. A conservative access strategy may produce a reverse effect, meaning a low PUI, but at the expense of an EST that is lower than what could be achieved for a given whitespace profile. Therefore, the objective for an SU is to be able to make a transmission decision based on previously observed whitespace statistics, such that by prudently transmitting packets, it is able to maximize the achievable EST and minimize the PUI. The primary contribution of this paper will be to develop such decision mechanisms. Note that traditionally, interference on another user of a network is defined as a user attempting to transmit in the middle of another user’s transmission. In the framework assumed in this paper, however, when an SU initially accesses a whitespace it is assumed that the PU will not be transmitting at that time; otherwise it is not considered a whitespace by the SU in the first place. Therefore the traditional definition of PUI does not hold in this case. In this paradigm, PUI can
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
be defined as when an SU transmits during a whitespace, and the time needed to successfully complete the transmission is more than how long the current whitespace will last purely from the PU traffic standpoint. This modified definition takes into account the fact that whitespaces are of fixed duration, and that once it ends the PUs’ transmissions will be affected in that the PUs will be prevented from transmitting. The key assumptions made in this paper are: 1) the ability of continuous channel sensing [2] by the secondary users, 2) no inter-secondary user channel access conflicts [10], 3) the ability of secondary users to differentiate between the PU and SU transmissions through physical layer waveform detection [11], 4) the ability of an SU to detect when a PU wants to transmit, and to vacate a whitespace [12], and finally, 5) fixed secondary packet size. IV. W HITESPACE M EASUREMENT The available whitespace can be measured by a secondary user by detecting the received signal strength (RSSI) at a given channel frequency. Durations when the measured RSSI are below a pre-defined threshold are considered as whitespaces. Such whitespaces are a function of the physical locations, topology, traffic profile, and the MAC and routing protocols used by the primary users. Interpretation of such measured whitespace depends on the channel sensing periodicity, which is considered to be very high (perfect sensing) for all results presented in this paper. The proposed method of capturing whitespace and modeling could be extended to work in environments where sensing is done with finite periodicity. Figure 2 shows a simple network of primary users and two secondary users which are within up to the interference range of the primary users. In Figure 3, we show the RSSI observed by the SU1 as a result of a data flow from PU1 to PU5 . These results are collected from an 802.11 simulation run in ns2 network simulator. The traffic flow consists of UDP data at 10 packets per second with a packet duration of 2.43 ms. Figure 3 shows the RSSI trace, as observed by SU1 , as a result of the packet flows from PU1 to PU5 . Observe that since SU1 is physically closest to the primary user PU3, the transmissions from that primary user appears with highest signal strength compared to the other PUs whose signal strength monotonically reduces with increasing distances. Also, since the PUs in this experiment use 802.11 as the MAC protocol, each packet transmission creates an RTS-CTS-DATA-ACK cycle, resulting in white spaces of multiple durations representing different 802.11 inter frame spacing (i.e. SIFS, DIFS, etc.). Three types of whitespaces can be observed for this example 802.11 primary traffic. As pointed in Figure 3, the first is the control whitespace caused by the short inter frame spacing (SIFS) periods between the RTS, CTS, DATA, and ACK transmissions. The second whitespace type, termed as forwarding whitespace, represents the duration between a PU receiving data and forwarding data. The last whitespace type is the inter-packet whitespace, representing the inter packet duration which can vary greatly because it is based on the rate of packet transmission by the PUs. V. W HITESPACE S TATISTIICS The measured whitespace can be expressed [2] by a probability density function (PDF) w(t), in which the continuous
Figure 2: Example primary topology with secondary locations
Figure 3: RSSI for primary traffic as viewed by SU1 in Figure 2
Figure 4: PDF and CDF of a whitespace trace
random variable t represents the duration of individual white spaces. This function can capture the impacts of all primary user properties including physical location, network topology, traffic characteristics, and MAC/Routing protocols. In other words, once the w(t) for a measured whitespace is computed, the SUs can rely solely on this function for optimal channel access without having to worry about the individual PU properties. Figure 4 shows the PDF w(t) and the corresponding Cumulative Distribution Function (CDF) W(t) of a whitespace trace that resulted from a PU1 -to-PU5 data stream (see Figure 2) with an average rate of 20 packets per second, and the interpacket interval uniformly distributed between ±50% of the average rate. This whitespace is observed by SU1 in Figure 2. For clarity, the PDF graph is split into two sub-graphs, showing the ranges of durations where the probability is nonzero (between 0-1 ms and 9.5 to 62 ms). The probability is zero for all other whitespace durations. A number of key observations can be made from the whitespace statistics shown in Figure 4. First, the detected whitespaces from the measured RSSI trace vary a great deal in terms of their durations. Second, for this particular trace, created by 802.11 primary traffic, a vast majority of the whitespaces (more than 95%) last for less than 1 ms. The rest of the 5% is uniformly distributed between 10 ms and 60 ms. The initial peak in the PDF indicates the control and the forwarding whitespaces as defined in Section IV. The larger whitespaces are mainly from the inter-packet intervals which
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
Algorithm 1: MBAS Scheme if (channel free) then while (channel remains free) do determine , the duration spent in the current whitespace; if ( ≤ μ) then continue; else ∞ if ( w(t) dt ≥ 1 − I) then send packet; +S
else break; end end end
have the 10 ms to 60 ms spread created by the jitter in the packet generation by PU1 in Figure 2. These observations indicate that an efficient access strategy for the secondary users will have to avoid transmitting during any small whitespaces, but access the longer duration whitespaces. Without having the knowledge of the actual duration of a whitespace at the beginning of the whitespace, an SU needs to rely on statistical measures in order to maximize the EST while keeping the PUI within tolerable bounds. The access strategy in the next section attempts to accomplish these goals. VI. C HANNEL ACCESS S TRATEGY A. Access Logic The access strategy comprises of three main steps. First, when an SU intends to make a packet transmission it waits till the next white space is detected. Second, at any given point within a detected whitespace, an SU will transmit only when the estimated probability of interference, computed based on the measured whitespace PDF w(t), is lower than a pre-defined threshold. The third component requires the SU to wait for a minimum wait-threshold duration μ even before it estimates the interference probability. This minimum wait-threshold duration μ is required for considering the initial peak of the whitespace PDF function as shown in Figure 4. This peak indicates that for 802.11 style primary traffic, due to the inter-frame spacing at the MAC layer, over almost 95% of the detected whitespaces are too small to fit the secondary packets. As a result, if an SU transmits at the beginning of each whitespace then a vast majority of them will end up in primary interference. However, deferring the transmission attempts for a suitably chosen waitthreshold duration μ can potentially reduce such Primary User Interference (PUI) without scarifying the Effective Secondary Throughput (EST). After the initial wait-threshold duration μ, if a packet is transmitted at time since the beginning of a whitespace, the probability of interference can be computed using the quantities and the measured whitespace PDF w(t). If the computed PUI is less than a pre-set bound I then the SU makes a packet transmission. After the packet is transmitted, if the SU has more packets to send, it re-computes the interference with a newly measured value of , and the process continues until the probability of interference becomes greater than the chosen bound, I. The channel access strategy is summarized in the form of a pseudo code in Algorithm 1. Note that at the point of time when the remaining whitespace reduces to a level so that the expected interference is
no longer less than the bound I, the secondary user abandons transmission attempts and starts channel sensing for detecting the next whitespace. B. Primary User Interference (PUI) PUI for an access strategy is defined as the probability that an SU will cause interference (as defined in Section III) to PUs during a white space. According to the access strategy described in Algorithm 1, after a whitespace starts, a secondary user waits for a period μ before it attempts transmission, and continues to transmit packets until the expected PUI reaches a value that is larger than the pre-defined bound I. Let us define a term Jmax , which denotes the maximum number of packets that the secondary user can send before the expected interference reaches bound I. There are three access scenarios that can arise. First, a whitespace duration is less than or equal to μ, in which case the whitespace is not accessed by the SU, and therefore, there is no interference to the PUs. Second, a whitespace duration is in the range between μ and μ+Jmax S+(Jmax -1)γ, where S is the secondary packet duration and γ is the inter-packet duration for the secondary traffic. In this case, there will be PUI since the last packet will last at least till the end of the mark of the whitespace. The third scenario corresponds to when a whitespace duration is greater than μ+Jmax S+(Jmax 1)γ. No PUI will be caused in this case because the SU will vacate or exit the whitespace after Jmax transmissions. Thus, for a given whitespace profile w(t), the PUI can be defined by the probability that the whitespace duration is in the range between μ and μ+Jmax S+(Jmax -1)γ. Therefore, an SU accessing a whitespace characterized by PDF w(t) will create primary user interface (PUI): μ+Jmax S+(J max −1)γ
PUI =
w(t) dt
(1)
μ
According the access strategy proposed in Algorithm 1, the number of packets the SU is allowed to transmit in a given whitespace, Jmax , is determined by finding the maximum j that satisfies the inequality: ∞ 1−
w(t) dt ≤ I
(2)
μ+(j−1)(S+γ)
The left side of the above Equation 2 represents the PUI when the SU sends Jmax packets in a given whitespace. Therefore, the inequality in (2) gives the Jmax that will reduce the interference to the pre-defined bound I. By plugging in this value of Jmax in Equation 1, we can find the overall PUI caused by our proposed access strategy. C. Effective Secondary Throughput (EST) The Effective Secondary user Throughput (EST) is defined as the number of packets successfully transmitted per whitespace. To model EST we first define Sj , the probability of sending exactly j packets in a whitespace. This corresponds to
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
the event in which the jth packet from an SU in a whitespace has interfered with the PUs, i.e. the whitespace had ended during the transmission of the jth packet. The quantity Sj can be written as: μ+jS+(j−1)γ
Sj =
w(t) dt
(3)
μ+(j−1)(S+γ)
For a whitespace that lasts between μ and μ+(Jmax -1)(S+γ) the number of packets sent by an SU is j. For whitespace duration between μ+Jmax S+(Jmax -1)γ and ∞, the number of packets sent by an SU is Jmax , since the secondary vacates the whitespace after sending Jmax packets. Therefore EST, the expected number of secondary packets sent for a given whitespace can be expressed as: EST =
Jmax −1
∞
jSj +
j=1
Jmax Sj
Figure 5: The interaction between µ and the PDF w(t)
(4)
j=Jmax
Where j is the number of transmitted SU packets, W is the number of whitespaces per unit time, and Sj is the probability that an SU will transmit j packets in a whitespace. D. Dimensioning Wait-threshold μ The selection of wait-threshold μ is critical since it impacts both PUI and EST as shown in Equations 1 and 4. A large μ can reduce primary user interference by preventing the SUs from transmitting during very short whitespaces (see Figure 5). But a large μ can also bring down the Effective Secondary Throughput (EST), since some portions of large whitespaces will be lost due to this conservative wait period. These two observations are consistent with the dependencies of PUI and EST on μ, as captured in Equations 1 and 4. The goal is to optimally pre-dimension the parameter μ, based on the measured PDF function w(t), in order to strike a desirable balance between the PUI and EST. According to the access strategy described in Algorithm 1, the wait-threshold period determines the PUI caused by the first SU packet during a whitespace. For a wait-threshold tx , such PUI caused by the first SU packet can be written as: t x +S
f (tx ) =
w(t)dt
(5)
tx
Now, the minimum and the maximum values of tx , are zero and 2S, where S is the secondary packet duration. The maximum wait-threshold is 2S, since waiting for a duration that is greater than or equal to 2S would mean missing out sufficient whitespace that could have been used by the secondary user to send at least one packet. Therefore, for a given whitespace w(t), the optimal waitthreshold μ can be chosen as the smallest tx , over the range 0 to 2S, such that the quantity f(tx ) in Equation 5 is minimized. It should be noted that there may multiple values of tx that cause minimal first-packet interference, but by selecting the smallest tx , the EST can be maximized.
Figure 6: Evaluation of the impacts of µ and interference bound I
To summarize, after a secondary user profiles its observed whitespace as the PDF w(t), it first chooses an optimal waitthreshold μ that minimizes the first-packet interference while maximizing the EST. It then computes the maximum number of possible transmissions per whitespace (Jmax ) for a given PUI bound I from Equation 2. The SU then executes the access mechanism in Algorithm 1 with the computed parameters μ and Jmax . The resulting PUI and EST can then be computed using Equations 1 and 4. VII. ACCESS S TRATEGY E VALUATION Performance of the proposed access strategy has been evaluated both numerically and experimentally (using ns2 simulator) in terms of its Primary user Interface (PUI) and the Effective Secondary Throughput (EST). The network simulator ns2 is used to create a network topology as shown in Figure 2. A multi-hop primary traffic flow from node PU1 to PU5 is created. As described in Section IV, the primary users use 802.11 as the MAC layer protocol, and the resulting whitespace profile as observed by the secondary user SU1 is shown in Figure 3. All secondary traffic in this Section corresponds to those from nodes SU1 to SU2 . A. Impacts of wait-threhold μ and interference bound I The wait-threshold μ determines the duration an SU waits before transmitting in a whitespace, and the tolerable interference bound I determines the quantity Jmax , the number of packets an SU sends before vacating a whitespace. The simulation experimental results in Figure 6 correspond to a PU traffic flow with an average transmission rate of 20
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
Figure 7: Impacts of Secondary Packet Size
Figure 8: Effects of primary user traffic load
B. Impacts of Secondary User Packet Size packets per second (pps) and a 20% uniformly distributed inter-packet transmission variation. The SU packet duration S was chosen to be 1.215 ms, which is half the used primary packet duration. The results are produced using ns2 simulation. Observe that for all μ, a higher interference bound I provide a higher secondary throughput, although at the expense of increased primary interference. For the most liberal case I = 1.0, indicating that an SU never vacates a whitespace before it interferes with a PU, the EST is the highest since the SU in this case is able to exploit the entire whitespace. As expected, as the interference bound I is made more conservative (smaller), both PUI and the EST values are brought down. For μ = 0 ms, representing the case when the SU transmits immediately at the start of a whitespace, the amount of PUI is very high and the EST is low for all interference bounds. This is because as shown in Figure 4, most of the whitespaces are of very short duration for 802.11 primary traffic. As μ increases, the PUI first decreases abruptly and then relatively moderately, but in a monotonic manner. With increasing μ, the EST first shows an increasing trend up to a threshold value (0.715 ms in Figure 6), beyond which it either falls slightly or remains steady. This threshold μ provides the best EST while minimizing the PUI. For a given w(t), This can be computed from Equation 5 in Section VI-D. EST is measured in number of secondary packets successfully sent per whitespace. With μ set to zero, the SU attempts packet transmission in all whitespaces. However, such transmissions succeed only for a small percentage of the whitespaces, since the w(t) in Figure 4 indicates that a large number of whitespaces are smaller than the chosen secondary packet duration of 1.215 ms. That is why the EST is very low for zero μ. With increasing μ, small whitespaces are avoided, causing an increase in the EST. Once μ reaches the threshold value, defined by Equation 5, the EST reaches its peak. The values of μ, larger than that threshold, cause the access strategy to be over conservative, thus missing out useful whitespace that could contribute to successful secondary packet transmissions. This explains the downward trends in the EST values beyond the μ threshold.
Figure 7 shows the changes in primary interference and secondary throughput as a result of varying secondary packet durations. These results correspond to a wait-threshold of 0.715 ms, with back-to-back secondary packets (i.e. γ is set to zero). Observe that with increasing secondary packet duration (i.e. size) the EST slightly increases. This is because the last transmitted secondary packet (Jmax th packet) manages to use more amount of whitespace with larger packet durations. But it also means that in more occasions, the SU interferes with the primary users since with larger packets, it has higher chances of not finishing the Jmax th packet transmission before the white space ends. This explains why the PUI also goes up, although slightly, with increasing packet size. Figure 7 presents both experimentally measured (via ns2 simulation) and theoretical values (using Equations 1 through 4), which are marked as EXP and THY respectively. It is evident that the experimental results match very well with the theoretical results. Furthermore, the figure shows that with a more conservative I, the EST is lower than with a larger I. Also, the results for all I values indicate that the access strategy is indeed able to contain the primary user interference within the pre-defined interference bound I, for all experimented secondary packet sizes. C. Effects of Primary User Traffic Load For the next two experiments we use the optimal value of μ = 0.715 ms which was calculated using Equation 5, and was experimentally validated using results presented in Figure 6. The secondary packet duration is 1.215 ms which represents half of the PU packet duration. Figure 8 demonstrates the throughput and interference variation with different primary user packet rates, spanning across 1 pps to 33 pps. For all the primary packet rates, a 20% uniformly distributed inter-packet transmission variation was introduced. As expected, at low PU traffic rates, the secondary throughput is high due to lot of available whitespaces. With increasing PU packet rates, however, the EST comes down quite drastically. This reduction is due to a left-shift in the whitespace
978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
Figure 9: Impacts of primary traffic rate variability
PDF w(t), and the subsequent reduction in the maximum number of transmitted secondary packets per whitespace (Jmax ), as computed in Equation 2. It is worth noting that the primary interference (PUI) is quite low and, more importantly, relatively insensitive to the primary packet rates. This actually shows that with varying primary packet rates, by appropriately computing Jmax , the secondary user is able to vacate the whitespace, thus keeping the PUI insulated from the primary packet rates. This demonstrates the flexibility of the proposed access strategy. D. The Variability in Primary User Packet Transmissions The results in Figure 9 report primary user interference and secondary throughput with different amounts of variability in the inter-packet spacing for the primary users. All graphs in Figure 9 corresponds to an average data rate of 10 packets per second, with a packet spread factor ranging from 0 to 1. A spread factor of 0 indicates that the inter-packet interval for all primary packets is strictly 100 ms. A 0.2 spread factor means a 20% (of the average interval of 100 ms) uniformly distributed inter-packet interval variation. The secondary packet duration, and the primary packet durations were chosen to be 1.215 ms, and 2.43 ms respectively. For small spread factors, since the whitespace PDF is highly modal (a narrow probability peak in the PDF), the maximum transmittable packet count Jmax , as defined in Equation 2, can be customized for that narrow pick. Thus the available whitespace is very well utilized by the SU with high efficiency, leading to high EST. As the spread factor increases, wider and shorter peaks form in the PDF, leading to Jmax computation that needs to be conservative due to the presence of shorter whitespaces. As a result, the SU may often evacuate a whitespace much earlier than needed, leading to lower secondary throughputs. This explains as to why the EST reduces with larger spread factors, for all interference bound I. Due to the same reason, the PUI reduces with higher spread factor. It reduces rather rapidly initially, but then somewhat settles down to a smaller value, indicating prudent Jmax computation by the access mechanism in Algorithm 1.
of the whitespace as perceived by the secondary network users. A secondary user continually monitors its surrounding whitespace, models it, and then attempts to access the available spectrum holes so that the effective secondary throughput is maximized while the resulting interference to the primary users is kept limited to a pre-defined bound. The strategy is general in that it does not reply on the information about the primary network protocols. Once the surrounding whitespace of a secondary user is modeled by the user, it simply executes the proposed access strategy without having to rely on primary cooperation or any information about specific primary protocols. Through ns2 simulation experiments and developed analytical expressions it has been demonstrated that with 802.11 MAC traffic, the access strategy is able to provide reasonable amount of secondary throughput while bounding the primary interference to pre-set values. Our ongoing work on this topic includes, 1) relaxing the assumptions as listed in Section III, 2) using more elaborate and larger network topologies, and 3) using a wider range of primary MAC protocols including CSMA, ALOHA and other stochastic protocols to establish the general applicability of the proposed concept. R EFERENCES [1] V. R. Petty, R. Rajbanshi, D. Datla, W. Frederick, D. Daniel, et.al, ¨Feasibility of Dynamic Spectrum Access in Underutilized Television Bands,¨ in New Frontiers in Dynamic Spectrum Access Networks. DySPAN 2007., Apr. 2007. [2] S. Geirhofer, L. Tong, and B. M. Sadler, ¨Dynamic Spectrum Access in WLAN Channels: Empirical Model and Its Stochastic Analysis,¨ in ACM Workshop on Technology and Policy for Accessing Spectrum (TAPAS), August 2006. [3] Y. Yuan, P. Bahl, R. Chandra, P. A. Chou, J. I. Ferrell, et al., ¨KNOWS: Cognitive Radio Networks Over White Spaces,¨ in New Frontiers in Dynamic Spectrum Access Networks. DySPAN 2007., Apr. 2007. ¨ [4] I. Stanojev, O. Simeone, Y. Bar-Ness, and T. Yu, Spectrum Leasing via Distributed Cooperation in Cognitive Radio,¨ın IEEE International Conference on Communications. ICC ’08. , May 2008. [5] A. O. Ercan, L. Jiwoong, S. Pollin, and J. M. Rabaey, ¨A Revenue Enhancing Stackelberg Game for Owners in Opportunistic Spectrum Access,¨ in New Frontiers in Dynamic Spectrum Access Networks, (DySPAN 2008), Oct. 2008. [6] C. Yunxia, Z. Qing, and A. Swami, ¨Joint Design and Separation Principle for Opportunistic Spectrum Access in the Presence of Sensing Errors,¨ IEEE Transactions on Information Theory, vol. 54, no. 5, May 2008. [7] E. Jung and L. Xin, ¨Opportunistic Spectrum Access in Heterogeneous User Environments,¨ in New Frontiers in Dynamic Spectrum Access Networks. DySPAN 2008., Oct. 2008. [8] Q. Zhao, S. Geirhofer, L. Tong, and B. M. Sadler, ¨Optimal Dynamic Spectrum Access via Periodic Channel Sensing,¨ in Proc. IEEE Wireless Communications and Networking Conference (WCNC), March 2007. [9] S. Geirhofer, L. Tong, and B. M. Sadler, ¨Cognitive Medium Access: Constraining Interference Based on Experimental Models,¨ IEEE Jrnl on Selected Areas in Communications, vol. 26, no. 1, Jan. 2007. [10] M. Wellens, J. Riihijarvi, M. Gordziel, and P. Mahonen, ¨Evaluation of Cooperative Spectrum Sensing Based on Large Scale Measurements,¨ in New Frontiers in Dynamic Spectrum Access Networks. DySPAN 2008., Oct. 2008. [11] D. Cabric, S. M. Mishra, and R. W. Brodersen, ¨Implementation Issues in Spectrum Sensing for Cognitive Radios,¨ in Conference on Signals, Systems and Computers, Nov. 2004. [12] L. Xin and D. Zhi, ¨ESCAPE: A Channel Evacuation Protocol for Spectrum-Agile Networks,¨ in New Frontiers in Dynamic Spectrum Access Networks. DySPAN 2007., Apr. 2007.
VIII. C ONCLUSIONS In this paper, we have proposed a novel secondary user access strategy which is based on measurement and modeling 978-1-4244-4148-8/09/$25.00 ©2009 This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the IEEE "GLOBECOM" 2009 proceedings.
