Pseudo-Analog
Yüklə 1,04 Mb.
|
|
P +σn
= E( Gλ2 Y − X)2 (5) from video data is utilized as the prior knowledge in the video = , λ 1+γ · where E( ) represents the expectation. Comparing Eqn. (5) with Eqn. (2), one can conclude that the analog method can also achieve the same optimal performance as the digital method. Pseudo-Analog Wireless Transmission Technology Conventional digital wireless video transmission systems often divide a video into several groups of pictures (GOPs) via standard encoder e.g., JPEG 2000 and H.264, and adopt predictive source coding [14-19]. Intra- and inter-frame corre- lations of the video enable high compression efficiency. The digital communication system can select a suitable channel modulation/coding scheme (MCS) to overcome the channel interference. However, the selected MCS may not guarantee a predetermined packet loss ratio (PLR) due to the fluctuations of the channel quality. The cliff effect will occur under deep channel fading. Especially in the broadcast scenario, the receivers have different channel qualities. Therefore, the transmitter should select a MCS according to the worst channel quality to guarantee the correct demodulation of all receivers. Consequently, the existing digital wireless video transmission system is ineffective in the broadcast scenario. The pseudo- analog transmission technology has been proposed to improve the effectiveness of the video broadcast [22-28]. In the pseudo- analog video broadcast system, the transmitter does not need to know the receivers’ channel qualities because the receivers can demodulate the video adaptively according to their own channel qualities. The first pseudo-analog video transmission system named SoftCast is proposed in [22]. At the transmitter, SoftCast first divides the original video into multiple GOPs. Next, SoftCast performs three dimensional discrete cosine transform (3D-DCT) for each GOP to remove the spatial and temporal redundancy. The transformed coefficients are divided into reconstruction. However, the quality of reconstructed video does not increase linearly with the increase of signal-to-noise ratio (SNR) due to mutual interference. Huang et al. propose a knowledge-enhanced wireless video transmission system named KMV-Cast [25-27] which could further eliminates the mutual interference. Liu et al. add the multiple-input multiple- output (MIMO) technique into SoftCast and point out that important coefficients should be transmitted on channels with large gain [28]. HARDWARE IMPLEMENTATION DIFFICULTIES. In this section, we will describe two hardware implemen- tation difficulties in the designed system including the data format modification and the non-linear distortion. Firstly, we will explain the difference between the conventional digital system and the designed system in terms of the coding and modulation scheme. Then, we will introduce the causes of non-linear distortion in the designed system and the counter- measures we take to reduce such non-linear distortion. Data Format Modification In the conventional digital system, the physical layer re- ceives packets of bits from the application layer. These bits are first error protected by forward error correction (FEC) coding and interleaving, and then modulated into complex samples which are transmitted over the channel [29]. Fig. 2 shows two examples of the coding and modulation scheme in the conventional OFDM system and the designed OFDM system, respectively. In Fig. 2(a), the conventional OFDM system first adopts the FEC coding to increase the redundancy and then adopts the 16 quadrature amplitude modulation (16-QAM) to map each four data bits into a complex I/Q signal. Different from the conventional digital system, the physical layer of the designed system receives packets of real-value samples from the application layer. Specifically, these real-value samples are power-adjusted using the joint source-channel coding for error protection. The designed physical layer bypasses the FEC coding and interleaving modules, and directly maps each two real-value samples into a complex I/Q signal, as shown in Fig. 2(b). In order to integrate such design into the standard commu- nication system, we take advantage of the fact that the existing OFDM system separates the channel estimation and tracking from data transmission [30]. Therefore, we can change the way that the transmitted data is encoded and modulated without affecting the performance of OFDM. Specifically, the existing OFDM system divides spectrum into many independent sub- carriers, some of which are used for channel tracking (the specific signals transmitted on these subcarriers are called pilots) while others are used for data transmission. Since the designed pseudo-analog video transmission system does not modify the pilots, the original OFDM functions including synchronization, carrier frequency offset estimation, channel estimation, and phase tracking will not be affected [31]. Non-linear Distortion In the pseudo-analog video transmission system, the dis- tribution of the input real-value samples directly determines the spectrum power distribution. However, the hardware can not support unbounded signals in practical due to the limited range of D/A converters. As a consequence, the hardware will truncate signals with high amplitudes, thus introducing the non-linear distortion [32,33]. In order to avoid such non- linear distortion, the transmitted packets are required to be normalized by the application layer, ensuring that the average of the input signals in each packet is equal to 0. Actually, the average does not carry too much information but greatly increases the total transmission power. Besides the normaliza- tion, whitening should also be performed on the transmitted signals, which is similar to pseudo-random scrambling and bit stream interleaving in the conventional digital system. Two whitening methods are adopted in the existing literature of pseudo-analog wireless video transmission system including Hadamard transform [22] and unitary transform [25]. Whiten- ing can bring at least two benefits to the designed physical layer. Firstly, it ensures that the power spectrum distribution of the transmitted signals is flat so that the transmitted signals can shield the frequency selective fading. Secondly, it reduces the PAPR of AWGN, thus protecting the transmitted signals to suffer from less error caused by the noise. TRANSCEIVER FRAMEWORK In this section, we will introduce the detailed implemen- tations of the designed transceiver framework. Firstly, we will give the overview of the designed transceiver. Then, we will introduce the function of each module contained in the transmitter, channel and receiver, respectively. Overview of the Designed Transceiver The designed GNU Radio-based pseudo-analog wireless video transmission transceiver is an open-source real-time signal processing framework, which can support all packet sizes and MCSs (including dense modulation). Fig. 3 shows the detailed implementation block diagram of the transceiver in GNU Radio Companion [34] which is a graphical user interface that can be used to simplify the creation and con- figuration of flow charts and help researchers understand the data processing process more clearly. Although Fig. 3 does not show the detailed functions of each module, it provides the concept of data flow in GNU Radio. Specifically, the signal processing modules are represented by boxes, and data streams are depicted by arrows connecting signal processing modules. Arrows of different colors represent different data formats, e.g., red for bytes, blue for complexes, and orange for floats. For the ease of understanding, we also provide a simplified transceiver diagram in Fig. 4. Transmitter Compared with the receiver, the transmitter is relatively simpler to implement. As shown in Fig. 4, the input normalized and power-adjusted video data are whitened first. Then, these video data is mapped into complex signals and filled into the frame according to the general frame format as shown in Fig. 5. Specifically, the short training sequence is used for frame detection while the long training sequence is used to determine the start position of the fast Fourier transform (FFT) window. The pilots are used for channel estimation. In the designed system, the payload data represents the modulated complex signals. At the transmitter, we mainly accomplish the following tasks: Group DCT coefficients into packets with uniform size and assign a tag containing the length information for each packet. Generate a header (including the short training sequence and the long training sequence) for each packet which can be used for coarse frequency offset tuning at the receiver. Map the packet header to the complex constellation points according to the selected modulation scheme, such as e.g., BPSK, QPSK, 16-QAM, or 64-QAM, and map the payload data to the complex constellation points according to the intensive modulation scheme. Interleave the modulated payload data with pilots. Convert the transmitted signals from frequency domain into time domain using FFT. Add cyclic prefixes to the time-domain signals to deal with inter-symbol interference. The main modules of the transmitter are shown in Fig. 3 and their functions are as follows: Stream to tagged stream: Add tags to the data stream at fixed intervals, which can help the subsequent signal processing modules align data. Packet header generator: Generate a packet header (in- cluding the short training sequence and the long training sequence) for the tagged stream which can be used for frame detection and channel estimation at the receiver. Chunks to symbols: Map the bit stream to complex signals according to the selected modulation scheme. Tagged stream mux: Merge multiple tagged streams.
Pilot Data Conventional digital system Time Frequency ...+19, -9, +40, +3, -30, -27... Designed transcei Fig. 2 The coding and modulation scheme in the conventional digital system and the designed system. Pilot Frequency
Data Time
Fig. 3 The detailed implementation block diagram of the transceiver. Fig. 4 The simplified diagram of the transceiver. Fig. 5 The frame format. time
Schmidl & Cox OFDM synch [35]: Use pseudo-random preambles to detect the beginning of a frame and produce fine estimation of carrier frequency offset. Frequency modulator: Convert time-domain signals to OFDM carrier allocator: Generate OFDM symbols ac- cording to the frame format. IFFT: Transform OFDM symbols from frequency domain into time domain. OFDM cyclic prefixer: Add cyclic prefixes and perform pulse shaping on OFDM symbols. Channel One of GNU Radio-based transceiver is that it can provide seamless switch between simulations and experiments. The simulations can be implemented by circulating the transmitted signals back to the receiver without connecting to real radio devices, e.g., USRP. GNU Radio supports researchers to select the simulation environment by providing propagation models such as phase noise, clock drift, AWGN, Rayleigh, and multipath fading. In Fig. 3, we only show the implementation block diagram for simulations. In fact, researchers can switch from simulations to experiments by replacing the channel mode module with the UHD module. The main modules and functions of each module are as follows: Tag Gate: Control whether the tags are transmitted. Channel model: A basic channel model simulator which can be used to help evaluate, design and test various signals, waveforms, and algorithms. Throttle: Ensure that the average rate does not exceed the allowed maximum number of samples per second. Receiver The design of the receiver is more complicated than that of the transmitter. At the receiver, the decoding of the received signals is mainly divided into two steps: 1) Decode the packet header. The short training sequence and the long training sequence contained in the packet header are used for frame detection and channel estimation, respectively (the channel estimation result will be used for the decoding of payload data); 2)Decode the payload data. After removing the packet headers from the received signals, the remaining payload data frequency-domain signals. Delay: Delay the signals by a certain amount of sampling time. Multiply: Multiply all the input signals. Header payload demux: Demultiplex packet data from burst transmission. When the length of the received packet has not been determined, the packet header is passed to other blocks for demodulation, and then the payload data is demodulated using the information pro- vided by the packet header. OFDM channel estimation: Estimate the channel condi- tion and coarse frequency offset of OFDM with the help of the packet header. OFDM frame equalizer: Equalize the tagged OFDM frames. OFDM serializer: Discard the pilots and extract the payload data. SIMULATIONS AND EXPERIMENTS In this section, we will analyze the performance of the designed GNU Radio-based pseudo-analog wireless video transmission system. Firstly, we will give the parameter set- tings of the hardware platform. Then, we will compare the simulation and experiment results from the perspectives of both the objective evaluation metric and the subjective visual quality. Parameter Settings The wireless pseudo-analog video transmission system de- signed in this paper is shown in Fig. 6. The object in the red dashed box represents the Ettus Research X310, which is a universal radio device equipped with two antennas (as shown in the orange dashed box) supporting the full-duplex operation from 120 MHz to 6 MHz in frequency. Ettus Research X310 also has two daughter boards, multiple high-speed interface options (e.g., PCIe, dual 10 GigE, and dual 1 GigE), and a large programmable Kintex-7 FPGA. The open-source soft- ware architecture of Ettus Research X310 also supports cross- platform UHD drivers which make it compatible with many development frameworks. In this paper, we run the designed transceiver on a PC (as shown in the green dashed box) which has an Intel i5 CPU with 16 GB RAM. The operating system is Ubuntu 16. 04 with Linux 4. 14. 0. The specific details of the hardware are shown in Table 1. We compare the simulation and experiment results from perspectives of both the objective evaluation metric, e.g., peak signal-to-noise ratio (PSNR) [37] and the subjective visual quality. The PSNR is a standard objective measurement of video/image quality, which can defined as a function of the MSE: Fig. 6 The designed GNU Radio-based platform. TABLE I The parameter settings. Componets Types Operating System Ubuntu 16.04 with Linux 4. 14. 0 GNU Radio Version 3.7.10 UHD Version 3.15.0 SDR Ettus Research X310 FPGA Kintex-7 Daughter boards UBX-160 Interface PCIe/dual 10 GigE/dual 1 GigE Supported Frequency 120 MHz-6 GHz Maximum Bandwidth 20MHz Performance Analysis of Simulation and Experiment Re- sults × In this paper, we use the standard video sequence “Fore- man” to verify the effectiveness of the designed system [36]. “Foreman” has 300 frames with a resolution of 176 144. Firstly, we test the experiment performance of the designed transceiver at the frequency of 2.4GHz with a channel band- width of 20 MHz. Since the USRP radios operate in the same frequency band as 802.11 WLANs, there is unavoidable interference. To reduce the impact of interference, we repeat each experiment ten times. Secondly, we test the simulation performance of designed transceiver over AWGN channel, and set the channel SNR to the same as that of the experiment (please note that the channel SNR can be calculated by the pilots). In addition, we also implement the simulation of video transmission in the conventional digital system under the same channel SNR. Specifically, we generate MPEG4 streams using the H.264/AVC codec provided by the FFmpeg software and the X264 codec library in the conventional digital system. To ensure that they take the same channel bandwidth, the MPEG4 streams are encoded into bit streams at 1/3 code rate and mapped into complex signals using 16QAM. 255 PSNR = 20log( √MSE ) (6) where MSE represents the mean squared error between all pixels of the reconstructed video and the original version. The simulation and experiment results are shown in Figs. 7-8. For simplicity, we select the 2nd frame (see Fig. 7) and the 180th frame (see Fig. 8) of “Foreman” to analyse the performance of the designed transceiver. In Figs. 7-8, we present four images from left to right including 1) the original frame, 2) the reconstructed frame using the designed transceiver by experiment, 3) the reconstructed frame using the designed transceiver by simulation, 4) the reconstructed frame using the digital system by simulation. From Figs. 7-8, one can see that the designed system can achieve the similar performance in the simulation and experiment in terms of subjective visual quality. Since the thermal noise caused by the actual radio device may affect the quality of the reconstructed frame, the PSNR performance of the simulation results are slightly higher than that of the experiment results using the designed transceiver. One can also conclude from Figs. 7-8 that pseudo-analog video trans- mission system can obtain better performance compared with the digital video transmission system. Although the frames reconstructed by the pseudo-analog system are a little vague compared with the original frames, one can also receive the most information contained in the frames. However, in the same channel condition, the reconstructed frames by the digital system are prone to distortion which are not suitable for watch- ing. This is because the digital system adopts the inter-frame compression and motion compensation technologies, resulting in a high correlation between frames. The demodulation error of a single frame may result in the loss of several-second video clip. CONCLUSIONS In this paper, a GNU Radio-based pseudo-analog video transmission transceiver has been proposed which can provide seamless switch between simulations and experiments. Firstly, we have described the fact that pseudo-analog transmission system can also achieve the same optimal performance as digital system from the perspective of rate-distortion theory. Then, we have analyzed the two difficulties in the practical implementation process including data format modification and non-linear distortion. Next, we have provided the detailed flow chart of the proposed transceiver and introduced the detailed function of each module. Finally, we compare the simulation and experiment results in terms of PSNR and the subjective visual quality. The testing results have shown that the designed transceiver can achieve the similar performance in the simulation and experiment which shows the effectiveness of the designed transceiver. In addition, we have also compared Fig. 7 Simulation and experiment results (Foreman #2). From left to right: (a) the original frame; (b) the reconstructed frame using the designed transceiver by experiment, PSNR = 28.71dB; (c) the reconstructed frame using the designed transceiver by simulation, PSNR = 28.95dB; (d) the reconstructed frame using digital transmission system by simulation, PSNR = 23.52dB. Fig. 8 Simulation and experiment results (Foreman #180). From left to right: (a) the original frame; (b) the reconstructed frame using the designed transceiver by experiment, PSNR = 27.14dB; (c) the reconstructed frame using the designed transceiver by simulation, PSNR = 27.36dB; (d) the reconstructed frame using digital transmission system by simulation, PSNR = 21.11dB. the simulation and experiment results obtained by the designed transceiver with the simulation results obtained by the digital system. The simulation results have indicated that the pseudo- analog system can achieve over 6.25dB PSNR gains than the digital system under the same channel condition. ACKNOWLEDGEMENT This work was supported in part by the National Natural Science Foundation of China under Grants No. U1733114 and No. 61631017, in part by the Fundamental Research Funds for the Central Universities, in part by Shanghai Rising- Star Program under Grant No. 19QA1409100, and in part by Shanghai Science and Technology Innovation Action Plan under Grant No. 19DZ1201100. REFERENCES B. Mao, F. Tang, Z. M. Fadlullah, and N. Kato, “An Intelligent Route Computation Approach Based on Real-Time Deep Learning Strategy for Software Defined Communication Systems,” IEEE Transactions on Emerging Topics in Computing, Early Access, pp. 1-12, Feb. 2019. Z. Lan, K. Mizutani, G. Villardi, and H. Harada, “Design and Imple- mentation of a Wi-Fi Prototype System in TVWS based on IEEE 802.11 a/f, in Proc. IEEE WCNC, Jul. 2013, pp. 750-755. B. Bloessl, F. Klingler, F. Missbrenner, and C. Sommer, “A Systematic Study on the Impact of Noise and OFDM Interference on IEEE 802.11p,” in Proc. IEEE VNC, Nov. 2017, pp. 1-4. Z. Geng, X. Wei, H. Liu, R. Xu, and K. Zheng, “Performance Analysis and Comparison of GPP-based SDR Systems, in Proc. IEEE MAPE, Oct. 2017, pp. 124-129. S. Navya, P. K. Yadav, and G. Ramesh, “Design and Implementation of OFDM Transmitter in WARP v3 Board,” in Proc. ICSCN, Mar. 2015, pp. 2-4. L. Lin, S. Chen, and T. Zhang, “Performance Evaluation for OFDM Link Based on LabVIEW and USRP,” in Proc. ICIM, Sept. 2014, pp. 897-901. B. Mao, F. Tang, Z. M. Fadlullah, N. Kato, O. Akashi, T. Inoue, and K. Mizutani, “A Novel Non-Supervised Deep-Learning-Based Network Traffic Control Method for Software Defined Wireless Networks,” IEEE Wireless Communications, vol. 25, no. 4, pp. 74-81, Aug. 2018. K. Tan, L. He, J. Zhang, Y. Zhang, and J. Fang, “Sora: High-performance Software Radio using General-purpose Multi-core Processors,” Commu- nications of the ACM, vol. 54, no. 1, pp. 99-107, Jan. 2011. R. Zitouni, S. Ataman, and L. George, “RF Measurements of the RFX 900 and RFX 2400 Daughter Boards with the USRP N210 Driven by the GNU Radio Software,” in Proc. CDCKD, Oct. 2013, pp. 490-494. N. B. Truong, Y. J. Suh, and C. Yu, “Latency Analysis in GNU Radio/USRP-Based Software Radio Platforms,” in Proc. MILCOM, Feb. 2014, pp. 305-310. X. Liu, and X. Zhang, “NOMA-based Resource Allocation for Cluster- based Cognitive Industrial Internet of Things,” IEEE Transactions on Industrial Informatics, vol. 16, no. 8, pp. 5379-5388, Aug. 2020 Y. Gao, and E. Tuncel, “Separate Source-Channel Coding for Transmit- ting Correlated Gaussian Sources Over Degraded Broadcast Channels,” IEEE Transactions on Information Theory, vol. 59, no. 6, pp. 3619-3634, Feb. 2013. T. Goblick, “Theoretical Limitations on the Transmission of Data from Analog Sources,” IEEE Transactions on Information Theory, vol. 11, no. 4, pp. 558-567, Oct. 1965. N. Cvijetic, S. G. Wilson, and R. Zarubica, “Performance Evaluation of a Novel Converged Architecture for Digital-Video Transmission Over Optical Wireless Channels,” Journal of Lightwave Technology, vol. 25, no. 11, pp. 3366-3373, Nov. 2007. Z. Chen, P. V. Pahalawatta, A. W. Tourapis, and D. Wu, “Improved Es- timation of Transmission Distortion for Error-Resilient Video Coding,” Yüklə 1,04 Mb. Dostları ilə paylaş:
|