Portal de Periódicos da CAPES

PSNR Enhancement in Image Streaming over Cognitive Radio Sensor Networks

2017; Electronics and Telecommunications Research Institute; Volume: 39; Issue: 5 Linguagem: Inglês

10.4218/etrij.17.0116.0887

2233-7326

Mahdi Bahaghighat, Seyed Ahmad Motamedi,

Advanced Wireless Network Optimization

ETRI JournalVolume 39, Issue 5 p. 683-694 ArticleFree Access PSNR Enhancement in Image Streaming over Cognitive Radio Sensor Networks Mahdi Bahaghighat, Mahdi Bahaghighat m.bahaghighat@aut.ac.ir Search for more papers by this authorSeyed Ahmad Motamedi, Corresponding Author Seyed Ahmad Motamedi motamedi@aut.ac.ir Search for more papers by this author Mahdi Bahaghighat, Mahdi Bahaghighat m.bahaghighat@aut.ac.ir Search for more papers by this authorSeyed Ahmad Motamedi, Corresponding Author Seyed Ahmad Motamedi motamedi@aut.ac.ir Search for more papers by this author First published: 11 October 2017 https://doi.org/10.4218/etrij.17.0116.0887Citations: 7 Seyed Ahmad Motamedi (corresponding author, motamedi@aut.ac.ir) and Mahdi Bahaghighat (m.bahaghighat@aut.ac.ir) are with the Wireless Sensor Laboratory, Electrical Engineering Department, Amirkabir University of Technology, Tehran, Iran. AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Share a linkShare onFacebookTwitterLinkedInRedditWechat Abstract Several studies have focused on multimedia transmission over wireless sensor networks (WSNs). In this paper, we propose a comprehensive and robust model to transmit images over cognitive radio WSNs (CRWSNs). We estimate the spectrum sensing frequency and evaluate its impact on the peak signal-to-noise ratio (PSNR). To enhance the PSNR, we attempt to maximize the number of pixels delivered to the receiver. To increase the probability of successful image transmission within the maximum allowed time, we minimize the average number of packets remaining in the send buffer. We use both single- and multi-channel transmissions by focusing on critical transmission events, namely hand-off (HO), No-HO, and timeout events. We deploy our advanced updating method, the dynamic parameter updating procedure, to guarantee the dynamic adaptation of model parameters to the events. In addition, we introduce our ranking method, named minimum remaining packet best channel selection, to enable us to rank and select the best channel to improve the system performance. Finally, we show the capability of our proposed image scrambling and filtering approach to achieve noticeable PSNR improvement. 1 Introduction In recent decade, a lot of researches have been conducted on wireless sensor networks (WSNs). Currently, this concept covers different aspects of communication network technologies and applications. Similar to other wireless networks, multimedia transmission over WSNs (WMSNs) is a challenging issue 1. Wireless multimedia applications require significant bandwidth and often have to satisfy relatively tight delay constraints. In addition, delay-sensitive data transmission, such as image transmission, conflicts with the limited available WSN resources. Although much work has been done in the field of WMSNs, to date, this problem has remained unsolved 2. The main obstacle that hinders the development of WMSNs is the band-limited nature of traditional WSNs. Generally, radio spectrum is a scarce resource. The limited availability of bandwidth is considered one of the major bottlenecks that hinder the development of high-quality multimedia wireless services 3. The concept of the cognitive radio sensor network (CRSN or CRWSN) was first introduced by 4. However, up to the present there have been few reported studies on image transmission over light and low end devices in CRSNs. Multimedia and delay-sensitive data applications in CRSNs require efficient real-time communication and dynamic spectrum access (DSA) capabilities 5. Compared with traditional WSNs, intelligent DSA in CRSNs not only improves spectral utilization, but also enhances the quality of service (QoS) in multimedia transmission systems. Delay-sensitive and multimedia communication in CRSNs was introduced by 5, but their model was customized for only 500 kV substations. A similar idea was also proposed by 6. On the other hand, optimal spectrum sensing is a key enabling technology for cognitive radio networks. The main objective of spectrum sensing is to provide more spectrum-access opportunities to cognitive radio users without interfering with the operations of licensed network users 7. Spectrum sensing has the potential to significantly impact the QoS 8 in both primary and secondary networks. A large inter-time spacing between two successive spectrum sensing (SS) runs may lead to collision, and consequently retransmission, while a small inter-time spacing can increase the overhead of systems in terms of time and power consumption. Realizing the lifetime maximization of a node is an important challenge in WSNs, and is directly related to power consumption 9. The use of orthogonal frequency-division multiplexing (OFDM) based traffic allocation in the subchannels of spectrum pools is an appropriate method for increasing the system reliability 10, but there have been few reported works that focus on multichannel image transmission over CRSNs. Reference 11 demonstrates a new model that exploits the impact of an SS frequency and packet-loading scheme for multimedia transmission over cognitive radio networks. Although their approach was the first work that derived the benefits from the SS frequency, their model suffers from a nonadaptive structure during multimedia transmission, high complexity, and large processing time, which is not suitable for resource-limited sensor nodes in WSNs. In this paper, we model an effective approach to achieving image transmission over CRSNs, focusing on the image quality at the receiver. We developed an adaptive structure model based on SS frequency optimization. In our model, we focus on optimizing the SS frequency in order to minimize the number of packets remaining in the send buffer. The parameters of our model are dynamically adapted to channel situations, and our updating process, called the dynamic parameter updating procedure (DPUP) is applied depending on whether hand-off (HO) events occur or not. In addition, we apply our proposed channel-ranking procedure, minimum remaining packet best channel selection (MRBC) to increase the probability of successful transmission of the image within the maximum tolerable delay time. In order to improve the performance of the transmission system, we develop our work according to multichannel packet scheduling, and then we incorporate some image processing based approaches to increase PSNR as much as possible; so, we end up our proposed PSNR enhancement method by scrambling the image in the transmitter and applying average filtering enhancement in the receiver (MC-SAFE), which significantly affects the final PSNR. 2 Proposed Model In this section, we present our proposed model for image transmission over CRSNs. 2.1 Primary User Activity Model In CRNs, the primary user activity significantly impacts the network performance, and it is difficult to obtain a good estimate of this activity in SS. According to most related works, we assume that primary user (PU) activity can be modeled as exponentially distributed inter-arrival times. In this applicable model, the traffic regime of the PUs can be considered as a two-state birth-death process (BDP), with a death rate of α and a birth rate of β. An ON (Busy) state represents the period used by PUs, while an OFF (Idle) state represents the unused period 12, 13. Based on the assumption of independent arrival of the PUs, each transition follows the Poisson arrival process. Thus, the lengths of ON and OFF periods are exponentially distributed, and PU activity can be modeled as exponentially distributed inter-arrival times. 2.2 Secondary User Activity Model The secondary user (SU) activity in CRSNs can be restricted by the PU's arrival. When a PU reoccupies the channel that would be "Idle" in the last SU's sensing period, the SU must vacate the current channel to prevent additional degradation of the QoS in the PU's network. In this case, the SU should try to switch to an alternate channel J in the available channel list (J = 1, 2, 3, … , S0; where S0 is the number of available channels in the spectrum pool). As mentioned before, the PU arrival process in an arbitrary channel i is modeled as a Poisson process with the parameter (that is, the arrival rate); therefore, the PU inter-arrival time τi will be an exponential distribution, with the mean arrival time given as 2, 5. 2.3 Adaptive Image Transmission Through Single Channel For image transmission, it is assumed that the image data stream can be fitted into N packets. Therefore, N indicates the total number of picture packets that are available in the send buffer before transmission. In the beginning phase of the image transmission, an active SU makes its decision among S available licensed channels in the spectrum pool (our new channel selection process will be explained at the end of this section). After the selection of an available idle channel, the SU turns on its RF link and begins to transmit the packet streams over the air medium. Generally, wireless multimedia transmission systems require significant bandwidth and often have to satisfy relatively tight delay constraints. In related works 3, 11, Dmax is defined as the maximum tolerable time in which an image data stream, including N packets, should be sent by the transmitter. From this definition, the reception of any packet after Dmax indicates that some packets were lost during transmission, and the PSNR of the image at the receiver may be highly degraded. Here, we define the channel capacity Ri as the total number of packets that can be transmitted during Dmax. It has been reported in some previous works 2, 3 that some major metrics in SS models, such as the probability of correct spectrum detection (PCD), the probability of false alarm (PFA), and the probability of miss detection (PMD), can play key roles in the performance of CRNs. In the same way, as reported by 11, we prefer to evaluate the impacts of PFA; therefore, we assume that PCD = 1 and PMD = 0. As previously mentioned, the impacts of the SS frequency and packet-loading scheme on multimedia transmission over cognitive radio networks have been investigated by 11. They introduced their new and creative model to perform the SS procedure using a periodic pattern. Further, they defined parameter f as the number of packets that are ready to be sent before running the next SS. Inter-sensing time , represents the actual time between two successive SS iterations in the channel "i". Using this notation, fi is the number of packets that should be transmitted over channel "i" before running the spectrum sensing once more, and are the packet time and sensing time duration, respectively. Here, increasing "f" means that the number of SS iterations decreases, and as a result, the SS time overhead will experience a noticeable reduction. This reduced time overhead may increase the probability of achieving successful image transmission within the time range [0, Dmax]. On the other side, by increasing f without any constraint, the probability of collision between PUs and SU in the re-occupied channel will significantly increase. This phenomenon requires intelligent trade-offs between the collision probability and time overhead. This tradeoff can be handled by optimizing parameter "f" according to an appropriate cost function. In (1): (1) where NRP is defined as the number of packets remaining in the send buffer after the th round of continuous transmission of packets in the channel "i" 11. At this time, the SS shows that due to the new arrival of some PUs, the channel "i" is no longer available; therefore, the sensor node should switch to a new idle channel (such as j) to fulfil the transmission task. Based on the PU's activity model mentioned earlier, the probability P{Xi ≥ x} can be written as follows 11: (2) where . In (3), the outcome of the last equation is presented: (3) Now, we can estimate the mathematical expectation of the remaining packets in the sending buffer E(NRP) 11: (4) Clearly, the limit of E(NRP) as λi approaches infinity will be equal to N: (5) This implicitly indicates that if PU's channels are in the saturation state, there are always N packets in the transmission buffer waiting to be sent. Therefore, in this study, we consider the unsaturated channel condition to investigate the results of our model. Similarly, in our work, E(NRP) is defined as the cost function, and according to (6), the target is to accurately estimate fi* in order to minimize the cost of the system. (6) Without loss of generality and to enable a feasible solution, fi should be between the upper and lower bounds, as indicated below: (7) To estimate fi*, we extract the following results from the equation : (8) Figure 1 demonstrates the impact of the maximum tolerable delay, Dmax, and fi on E(NRP). As shown in the figure, Dmax is increased from 0.1 s to 3 s in four independent scenarios. In the first scenario, Dmax = 0.1 s, and the lowest fi is selected as the optimum point, while the highest number of packets (more than 600 packets) remain in the send buffer, compared to less than 275 packets in the last scenario (Dmax = 3) with the highest value of fi, for the first round of transmission. Figure 1Open in figure viewerPowerPoint Impact of maximum tolerable delay on mathematical expectation of remaining packets. Reference 11 constructed a model based on the optimization of E(NRP) at the beginning of transmission on the available channel "i," and before HO of the current channel to the alternate channel "j" due to the arrival of the PU. However, in our work, we consider the critical situation after each HO to update the parameters in our system based on the new channel access scenario. Consequently, we applied our proposed procedure with dynamic updating of important parameters (DPUP), as indicated by Procedure 1. Figure 2 depicts our proposed system model and its dynamic updating approach depending on two different states: HO and no HO (NHO). Figure 2Open in figure viewerPowerPoint Proposed Methods for Image Transmission over CRSN. First, consider the condition where there is no need for an HO after a new SS. This implies that we have a successful transmission of "f" packets in round K; and both parameters should therefore be updated to extract from: (9) Second, in the situation where the HO is an obligation owing to the new arrival of PUs, it is interpreted as a collision event during the last transmission of packets; thus, about packets may be lost, and they should be retransmitted in the new channel j. Therefore, only will be updated to , and will remain the same as . In this case, we can obtain from (10): (10) In addition to updating parameters dynamically to best adapt to the based on channels and timing conditions, we add our proposed rule to select the optimized channel from the spectrum after running the SS. This task can be effectively carried out by selecting the best channel that minimizes . We call this process MRBC, and it is formulated as shown: (11) 2.4 Packet Loading Through Multichannel Communication (MCPL) In this section, we extend our proposed approach to multichannel transmission. In this case, an active SU can stream an N-packet input image into all of the S available channels simultaneously; thus, the sending SU fragments N packets into S fragments: such that Nj refers to the number of packets that should be sent through the channel "j." Therefore, it follows that: (12) As a direct extension of the single-channel case, the remaining packets of all channels can be defined as the summation of the remaining packets of each channel 11: (13) Now, we focus on the minimum of the E(NRP) function in order to achieve the optimum solution of this new problem. By evaluating the equation, it is clear that E(NRP) is the function of and variables, and it depends on the following parameters: ; therefore, the optimization problem has 2S variables and can be formulated as shown below: (14) To reduce the high complexity of this problem, 11 assumed that all channels have an equal value of "f" (), and that the channel rates are the same for all channels (). Consequently, the optimization problem is converted to a packet-loading problem and SS frequency optimization based on just (S + 1) variables : (15) (16) To solve the last problem, 11 proposed two algorithms, namely the Hughes-Hartogs and discrete particle-swarm optimization (DPSO) algorithms. For a fixed SS frequency (f), the Hughes-Hartogs algorithm can find the optimal packet-loading results (); then, the DPSO algorithm is used to find the optimal particle in each iteration. After several iterations, we can obtain the optimal packet-loading results and SS frequency. The main problems of the two above-mentioned methods are their high complexity and large processing times, in particular DPSO. Actually, in CRSNs, there are many hardware restrictions, and practical nodes are usually embedded using only small CPUs and limited memories. Accordingly, the methods with lower processing times are preferred. As a result, we discard the high cost DPSO algorithm and propose our method, which uses the following procedure: (17) By getting index "I" which minimizes , it is easy to estimate and f* just by using following equations: and f* = fI. Hence (S + 1) variables are estimated. It should be noted that our numerical analysis of the E(NRP) functional behavior in terms of the f variation, especially the soft variations of f around f*, shows that this function will experience a very smooth change around f*. This important feature explicitly indicates that the high-accuracy estimation of f* does not benefit the proposed transmission system. Moreover, its approximate calculation can reduce the total processing time. For example, applying the simple replacement of f(i+1) = f(i) + 2 instead of f(i+1) = f(i) + 1 (in Step 2 of our method) will reduce the overall processing time by 50%. 2.5 Image Scrambling and Filtering-Based PSNR Enhancement in Multichannel Communication (MC-SAFE) In previous sections, we presented our enhanced methods for image transmission over CRSNs. The main parameter that is considered in the model is Dmax or the maximum tolerable delay. However, in spite of our attempts to minimize the number of packets that remain in the send buffer, depending on the PUs' traffic, it is possible to have a timeout event without ending the image stream. In this case, the transmitter node prevents the transmission of the rest of the information of the input image. Consequently, on the other end of the channel, the receiver receives only a part of the image. This condition is depicted in Fig. 3. The input image is a 512 × 512 RGB image that includes three color channels: red, green, and blue. The black pixels in Fig. 3(f) are related to the packet loss. Actually, because of a timeout event, around 12% of all pixels remain in the send buffer. The sender node arranges the pixels from the first row of subchannel R to the last row of subchannel B as a vector (payload array) in the transmit buffer, after which it proceeds to send this vector. Chronologically, the last pixels of the B channel will be the last pixels that should be transmitted in the allowed time range: [0, Dmax]. In the case where a timeout event has occurred, the receiver fills the empty rows and columns with zero values or equivalently black Pixels, which is called zero-padding. Figure 3Open in figure viewerPowerPoint Black pixels show the lost packets due to a timeout event during transmission: (a) original (red), (b) original (green), (c) original (blue), (d) received (red), (e) received (green), and (f) received (blue). In order to compare the performance of the image transmission system, we use the most important metrics to evaluate the image quality, that is, the PSNR. The PSNR definition is presented in (17) 2, 14: (18)where MSE is the mean square error. According to these formulations, the direct approach to improve the PSNR is to reduce the MSE between the transmitted and received images 14, 15. In time-out situations, when there are black pixels with zero values in the last rows of the blue subimage, the MSE will increase. In addition, in this case, there is no way of enhancing the image quality based on typical image-processing methods, such as 3 × 3 average filters and smoothing filters (because such filters have no gain in the black parts related to the pixel loss; thus, no actual enhancement will be achieved). To address this challenging issue, we propose the MC-SAFE method, which suggests the distribution of the black pixels throughout all of the three subimages instead of in just the last part of the blue image. In fact, to change the spatial position of the pixels, we first used a reversible scrambling function g(x, y) to rearrange them in the RGB image ISender and generate a new image ; then, we converted three-two-dimensional (2D) matrices into a vector VSender according to the following proposed procedure: The transmitter node sends VSender and the receiver will receive V*Sender. In a perfect transmission scenario, V*Sender = VSender, but in practice, owing to packet loss, some of the pixel positions in V*Sender may be empty. In this case, the empty pixel positions are then zero padded. Now, the receiver should reverse the procedure proposed for transmission by converting to , and then applying the inverse function to obtain . This is the best time to apply our image- processing enhancement method based on the 3 × 3 average filter to estimate the value of pixels using their eight neighbors located in the 3 × 3 mask around the central pixesl. The MC-SAFE approach not only can markedly enhance the PSNR of the image, but also it can improve the maximum tolerable delay effectively. This improvement will be discussed in more detail in the next section. 3 Simulation and Results As previously presented, our work focuses on image transmission over CRSNs. In order to investigate the performance of our proposed algorithm, we used "lena.bmp," which is an eight-bit RGB image with a size 512 × 512. We take advantage of the main performance metric in image transmission, that is, the PSNR, to evaluate the performance of the algorithm in different conditions. In our model, we defined N and f as optimizing parameters to minimize our cost function E(NRP). According to (4) and (13), important parameters such as can affect this nonlinear function; thus, we considered these parameters in our simulations using Matlab software, and we investigated appropriate values based on related literatures 2, 11, 16. Table 1 shows these parameters along with their descriptions. Table 1. Parameters defined in the model Row Parameter Description 1 N Number of packets in input image 2 R i Channel capacity 3 f i Spectrum sensing frequency 4 λ i Arrival rate of pus 5 T s Spectrum sensing duration time 6 D max Maximum tolerable delay 7 S Number of available channels in pool 8 P fa Probability of false alarm First, we evaluate the proposed DPUP and its impact on the system functionality. We consider the situation: , and we assume that the PUs' traffic is sufficient low to enable the continuous transmission of all of the input image Lena on just one channel. By applying the proposed dynamic procedure that was previously mentioned, both Dmax(K+1) and N(K+1) are updated to provide a dynamic estimation of for each SS iteration (SSI) K. In Fig. 4, the optimal frequency f* is estimated to be around 89 at the beginning phase of the transmission for the first SSI (SSI = 1); then, it falls sharply to 29 for SSI = 2. Next, f* shows a soft decline trend down to 10 from SSI = 3 to SSI = 29. Finally, the new value is raised up to 25 for SSI = 41 at the end point of transmission. This "U" shaped variation in the f* pattern is well adapted to time-variant channel scenarios. To elaborate, in the initial phase, our proposed algorithm tries to send as many packets as possible, around 89, for SSI = 1. The SU's arrival time is the random time point between two successive PU intervals, and will take μi = (λi)–1 seconds on average. If this time point is near the left side of the idle time window, the selection of a higher f* will be the best action, as performed by our algorithm. On the other hand, if it is far from the left side, this implicitly means that we have a few channels available time; therefore, the HO (and consequently retransmission) will be unavoidable in the next SS for both low and high values of f*. As a result, a larger value may be a better choice provided that the SU arrival time is sufficient to ensure no collision for the first SS. Figure 4Open in figure viewerPowerPoint Variation of optimal frequency based on dynamic updating of parameters. After SSI = 1, if the channel is still available, our algorithm will update the parameters and try to reduce the f* gradually from 29 to 10 in order to decrease the collision probability. For SSI = 29, about 25% of the total packets are still in the transmitting queue; thus, f* should be increased to complete the transmission within the short remaining time. At the final attempt, for SSI = 41, the nodes empty the buffers by sending all of the remaining packets. Now, we analyze the behavior of our proposed methods in terms of channel decision and ranking. We set the parameters to the following values: , and we then apply three different scenarios: In scenario (1), the arrival rates of the PU's channel increase from 300 (arrivals per second) in channel 1 to 3,000 in channel 10. In this case, channels 1 and 10 are clearly the best and worse channels, respectively (from the CR user perspective). Figure 5 shows the channel's histogram for transmission of ''Lena.bmp." In this figure, the vertical and horizontal axes represent the number of iterations of each channel in the image-transmission process and the channel index, respectively. The results of Fig. 5(b) clearly show that our proposed method effectively selects the best channel (channel 1 with 14 iterations), the second best channel (channel 2 with 12 iterations), and the third best channel (channel 3 with 6 iterations). Our channel decision and ranking algorithm avoids channels with high arrival rates to reduce the probability of collisions in the system. Conversely, in scenario (2), the arrival rates for channels 1 to 10 vary from 3,000 to 300. The results show the same rule indicated in scenario (1) for channels 10, 9, and 8. Here, the worse channels, such as 1, 2, and 3, have no share in data transmission in the systems (Fig. 5(c)). In the last scenario, we assume that all channels have the same value of arrival rate, that is, about 300. In this case, our method tries to select all 10 available channels, and balances the transmission load in the spectrum band (please refer to Fig. 5(a)). The results of these three scenarios verify that our proposed method rigidly tries to involve just the channels that have the appropriate conditions in order to reduce the probability of retransmission (because retransmission has a high cost on the actual time of transmission). Actually, providing that there is a retransmission, the probability of successful transmission of an image within the time range [0, Dmax] will decrease. Figure 5Open in figure viewerPowerPoint Channel histograms for different PU arrival rates: (a) equal rate, (b) incremental rate, and (c) deceremental rate. In addition, we performed our evaluation of the proposed method by setting the model parameters as follows: Figure 6 shows the original color image (Lena) at the sender with three sub images in R, G, and B channels. The results of our model when transmitting an image over CRSNs are depicted in Figs. 6(e), (f), (g), and (h). In order to compare the subchannel degradation, the error energy of each channel is calculated separately, and the outcomes strongly indicate that in this case, there is a zero energy error for all color channels at the receiver sensor node. This is confirmed by Figs. 6(i), (j), (k), and (l), for the final RGB color in the receiver sensor node. The results of Fig. 6 show that in this example, we do not have any loss; thus, we obtain PSNR = Inf. Figure 6Open in figure viewerPowerPoint Error Energy Comparison: original image with its channels at the transmitter node (a) red, (b) green, (c) blue, (d) color, received with its sub channels at the received node (e) red, (f) green, (g) blue, (h) color, and error Energy Images for all images (i) red, (j) green, (k) blue, (l) color. To continue our analysis, we present Fig. 7. Figure 7(a) shows the value of "f" in terms of the SSI, while Fig. 7(b) shows the channel number indicator (CNI). From the figure, the total numbers of SS and HO iterations are 24 and 8 respectively. As a consequence, our algorithm tries to assign a lower f when an HO event forces the sensor node to switch to the channel with a higher arrival rate. For example, after SSI = 3, the transmitter node performs an HO from CNI = 1 (with λ1 = 200) to CNI = 2 (with λ2 = 400), and owing to this change, f decreases from 24 to 16. This behavior of our alg