13_Low Complexity Motion Estimation for H.264 AVC Based Depth Encoding in Free Viewpoint Video
- 格式:pdf
- 大小:1.92 MB
- 文档页数:15
第 22卷第 7期2023年 7月Vol.22 No.7Jul.2023软件导刊Software Guide基于提前终止策略改进的运动估计算法朱鑫磊,汪伟(上海理工大学光电信息与计算机工程学院,上海 200093)摘要:针对HM-16.14中TZSearch标准算法存在的计算复杂度高、耗时相对较长等问题,提出一种基于提前终止策略的改进TZSearch算法。
首先,根据编码产生的率失真代价对编码单元、变换单元和预测单元的深度进行划分,有效避免了额外的划分深度;然后,在TZSearch初始网格搜索过程中,采用钻石搜索和六边形搜索两种搜索方式,根据运动矢量分布位置选择一种更为有效的方式,精确找出最佳匹配点;最后,使用OARP栅格搜索和精细搜索完成运动估计。
由实验结果可知,该方法与标准算法相比,平均降低了60%以上的TZSearch运动估计耗时,且基本不影响视频质量。
关键词:TZSearch算法;提前终止策略;栅格搜索;精细搜索;运动估计DOI:10.11907/rjdk.221887开放科学(资源服务)标识码(OSID):中图分类号:TP391.1 文献标识码:A文章编号:1672-7800(2023)007-0051-08A Modified Motion Estimation Algorithm Based on Early Termination StrategyZHU Xinlei, WANG Wei(School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology,Shanghai 200093, China)Abstract:Considering the high computational complexity and relatively long time consumption of the TZSearch standard algorithm within HM-16.14, an improved TZSearch algorithm based on early termination strategy is proposed to improve the efficiency of video coding. Firstly,the depth sorting of the coding unit, transform unit and prediction unit is calculated according to the performance of rate distortion, which can effectively decrease additional division depths. Secondly, two search methods, i.e. diamond search and hexagonal search, are employed within the initial grid search step of TZSearch in order to precisely find the best matching point according to the motion vector distribution. Finally,OARP raster search and fine search are used to acquire the motion estimation results. Compared with the standard algorithm, experimental re‐sults show that the proposed method reduces more than 60% motion estimation time consumption on average, yet keeps the similar video quali‐ty .Key Words:TZSearch algorithm; early termination strategy; raster search; fine search; motion estimation0 引言随着视频技术的快速发展,依靠视频传递信息变得越来越普及,这使得视频流数据在互联网传输中的占比越来越大。
《Broadcasting, IEEE Transactions on》期刊第3页200条数据https:///academic-journal-foreign_broadcasting-ieee-transactions_info_128_1/1.《A New Blind SLM Scheme With Low Decoding Complexity for OFDM Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html2.《Implementation and Co-Simulation of Hybrid Pilot-Aided Channel Estimation With Decision Feedback Equalizer for OFDM Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html3.《A Depth-Aware Character Generator for 3DTV》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html4.《Novel End-to-End Quality of Service Provisioning Algorithms for Multimedia Services in Virtualization-Based Future Internet》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html5.《FPGA Design and Performance Evaluation of a Pulse-Based Echo Canceller for DVB-T/H》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html6.《On the Provisioning of Mobile Digital Terrestrial TV Services to Vehicles With DVB-T》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html7.《Reception Quality Prediction in a Single Frequency Network for the DTMB Standard》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html8.《Signal-to-Noise Ratio Estimation Algorithm for Advanced DVB-RCS Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html9.《Augmented Data Transmission Based on Low Density Parity Check Code for the ATSC Terrestrial DTV System》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html10.《Adaptive Digital Predistortion for Wideband High Crest Factor Applications Based on the WACP Optimization Objective: A Conceptual Overview》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html11.《Initial-Estimation-Based Adaptive Carrier Recovery Scheme for DVB-S2 System》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html12.《Quantifying Subjective Quality Evaluations for Mobile Video Watching in a Semi-Living Lab Context》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html13.《Perceived 3D TV Transmission Quality Assessment: Multi-Laboratory Results Using Absolute Category Rating on Quality of Experience Scale》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html14.《Efficient Pilot Patterns and Channel Estimations for MIMO-OFDM Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html15.《Coding Distortion Elimination of Virtual View Synthesis for 3D Video System: Theoretical Analyses and Implementation》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html16.《An Efficient Nonlinear Companding Transform for Reducing PAPR of OFDM Signals》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html17.《Saliency Inspired Full-Reference Quality Metrics for Packet-Loss-Impaired Video》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html18.《Improved CIR-Based Receiver Design for DVB-T2 System in Large Delay Spread Channels: Synchronization and Equalization》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html19.《High Power Amplifier Pre-Distorter Based on Neural-Fuzzy Systems for OFDM Signals》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html20.《Performance Analysis of Inter-Layer Prediction in Scalable Video Coding Extension of H.264/AVC》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html21.《Study of Rating Scales for Subjective Quality Assessment of High-Definition Video》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html22.《Planning Factors for Digital Local Broadcasting in the 26 MHz Band》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html23.《Peak-to-Average Power Ratio Reduction of OFDM Signals Using PTS Scheme With Low Computational Complexity》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html24.《Feedback Cancellation for T-DMB Repeaters Based on Frequency-Domain Channel Estimation》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html25.《Efficient Multi-Reference Frame Selection Algorithm for Hierarchical B Pictures in Multiview Video Coding》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html26.《Performance Comparisons and Improvements of Channel Coding Techniques for Digital Satellite Broadcasting to Mobile Users》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html27.《Burst-Aware Dynamic Rate Control for H.264/AVC Video Streaming》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html28.《Helicopter-Based Digital Electronic News Gathering (H-DENG) System: Case Study and System Solution》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html29.《Transmit Diversity for TDS-OFDM Broadcasting System Over Doubly Selective Fading Channels》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html30.《Interference Cancellation Techniques for Digital On-Channel Repeaters in T-DMB System》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html31.《Field Measurements of EM Radiation From In-House Power Line Telecommunications (PLT) Devices》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html32.《A Novel Scheme of Joint Channel and Phase Noise Compensation for Chinese DTMB System》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html33.《Path Loss Prediction for Mobile Digital TV Propagation Under Viaduct》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html34.《Efficient Motion Vector Interpolation for Error Concealment of H.264/AVC》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html35.《3D-TV Content Creation: Automatic 2D-to-3D Video Conversion》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html36.《A Novel Rate Control Technique for Multiview Video Plus Depth Based 3D Video Coding》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html37.《The Effect of Crosstalk on the Perceived Depth From Disparity and Monocular Occlusions》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html38.《Semi-Automatic 2D-to-3D Conversion Using Disparity Propagation》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html39.《Display-Independent 3D-TV Production and Delivery Using the Layered Depth Video Format》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html40.《3DTV Roll-Out Scenarios: A DVB-T2 Approach》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html41.《PAPR Reduction Using Low Complexity PTS to Construct of OFDM Signals Without Side Information》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html42.《Quality-Oriented Multiple-Source Multimedia Delivery Over Heterogeneous Wireless Networks》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html43.《Efficient PAPR Reduction in OFDM Systems Based on a Companding Technique With Trapezium Distribution》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html44.《Objective Video Quality Assessment Methods: A Classification, Review, and Performance Comparison》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html45.《Pixel Interlacing Based Video Transmission for Low-Complexity Intra-Frame Error Concealment》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html46.《Fountain Codes With PAPR Constraint for Multicast Communications》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html47.《RF Watermark Backward Compatibility Tests for the ATSC Terrestrial DTV Receivers》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html48.《61st Annual IEEE Broadcast Symposium — Save the Date》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html49.《Evaluation of Asymmetric Stereo Video Coding and Rate Scaling for Adaptive 3D Video Streaming》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html50.《Stereoscopic Perceptual Video Coding Based on Just-Noticeable-Distortion Profile》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html51.《A Depth Information Based Fast Mode Decision Algorithm for Color Plus Depth-Map 3D Videos》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html52.《3D-TV Production From Conventional Cameras for Sports Broadcast》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html53.《A Digital Blind Watermarking for Depth-Image-Based Rendering 3D Images》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html54.《Object-Based 2D-to-3D Video Conversion for Effective Stereoscopic Content Generation in 3D-TV Applications》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html55.《3D-TV Content Storage and Transmission》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html56.《New Depth Coding Techniques With Utilization of Corresponding Video》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html57.《3DTV Broadcasting and Distribution Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html58.《Boundary Artifact Reduction in View Synthesis of 3D Video: From Perspective of Texture-Depth Alignment》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html59.《Stereoscopic 3D-TV: Visual Comfort》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html60.《A Novel Inpainting-Based Layered Depth Video for 3DTV》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html61.《3D-TV R&D Activities in Europe》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html62.《A Directional-View and Sound System Using a Tracking Method》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html63.《Joint Maximum Likelihood Estimation of Carrier and Sampling Frequency Offsets for OFDM Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html64.《Perceptual Issues in Stereoscopic Signal Processing》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html65.《Performance Evaluation of Multimedia Content Distribution Over Multi-Homed Wireless Networks》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html66.《The Relationship Among Video Quality, Screen Resolution, and Bit Rate》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html67.《Corrections to “Efficient Motion Vector Interpolation f or Error Concealment of H.264/AVC”》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html68.《PAPR Reduction of OFDM Signals by PTS With Grouping and Recursive Phase Weighting Methods》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html69.《Improve the Performance of LDPC Coded QAM by Selective Bit Mapping in Terrestrial Broadcasting System》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html70.《The Importance of Visual Attention in Improving the 3D-TV Viewing Experience: Overview and New Perspectives》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html71.《Co-Channel Analog Television Interference in the TDS-OFDM-Based DTTB System: Consequences and Solutions》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html72.《Prediction and Transmission Optimization of Video Guaranteeing a Bounded Zapping-Delay in DVB-H》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html73.《IBC2011 Experience the Future》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html74.《61st Annual IEEE Broadcast Symposium — Save the Date》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html75.《A 2$times$2 MIMO DVB-T2 System: Design, New Channel Estimation Scheme and Measurements With Polarization Diversity》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html76.《A Pilot Symbol Pattern Enabling Data Recovery Without Side Information in PTS-Based OFDM Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html77.《Efficient Incremental Raptor Decoding Over BEC for 3GPP MBMS and DVB IP-Datacast Services》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html78.《Spatio-Temporally Consistent Novel View Synthesis Algorithm From Video-Plus-Depth Sequences for Autostereoscopic Displays》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html79.《IBC2011 Experience the Future》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html80.《Evaluation of Stereoscopic Images: Beyond 2D Quality》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html81.《An Evaluation of Parameterized Gradient Based Routing With QoE Monitoring for Multiple IPTV Providers》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html82.《Three-Dimensional Displays: A Review and Applications Analysis》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html83.《Guest Editorial Special Issue on 3D-TV Horizon: Contents, Systems, and Visual Perception》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html84.《LIVE: An Integrated Production and Feedback System for Intelligent and Interactive TV Broadcasting》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html85.《$MS^{2}$ : A New Real-Time Multi-Source Mobile-Streaming Architecture》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html86.《A Reverse-Order Scheduling Scheme for Broadcasting Continuous Multimedia Data Over a Single Channel》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html87.《An Automatic Recommendation Scheme of TV Program Contents for (IP)TV Personalization》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html88.《Irregular Mapping and its Application in Bit-Interleaved LDPC Coded Modulation With Iterative Demapping and Decoding》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html89.《Time Diversity in Mobile DVB-T2 Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html90.《An Audio Quality Evaluation of Commercial Digital Radio Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html91.《Spectral Geometry Image: Image Based 3D Models for Digital Broadcasting Applications》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html92.《Novel Approach to Support Multimedia Services Over DTMB System》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html93.《Collaborative Content Fetching Using MAC Layer Multicast in Wireless Mobile Networks》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html94.《Energy-Efficient Transmission for Protection of Incumbent Users》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html95.《Normalization Factor of Hierarchically Modulated Symbols for Advanced T-DMB System》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html96.《Pooling-Based Intra Prediction Mode Coding for Mobile Video Applications》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html97.《A Suboptimal Tone Reservation Algorithm Based on Cross-Entropy Method for PAPR Reduction in OFDM Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html98.《A Measurement Method of the DTMB Modulator》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html99.《Interference Elimination for Chinese DTMB System With Transmit Diversity》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html100.《61st Annual IEEE Broadcast Symposium》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html101.《IBC2011 Experience the Future》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html102.《A Low-Complexity SLM Scheme Using Additive Mapping Sequences for PAPR Reduction of OFDM Signals》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html103.《Illumination-Sensitive Background Modeling Approach for Accurate Moving Object Detection》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html104.《Coordinating Allocation of Resources for Multiple Virtual IPTV Providers to Maximize Revenue》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html105.《Inter-Sequence Error Concealment Techniques for Multi-Broadcast TV Reception》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html106.《Performance Evaluation of the DVB-RCT Standard for Interactive Services》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html107.《An Efficient Predistorter Design for Compensating Nonlinear Memory High Power Amplifiers》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html108.《Accurate BER Analysis of OFDM Systems Over Static Frequency-Selective Multipath Fading Channels》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html109.《A Frame-Related Approach for Performance Improvement of MPE-FEC in DVB-H》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html110.《Balanced Multiple Description Coding for 3D DCT Video》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html111.《Performance Validation of the DVB-SH Standard for Satellite/Terrestrial Hybrid Mobile Broadcasting Networks》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html112.《An Improved Tone Reservation Scheme With Fast Convergence for PAPR Reduction in OFDM Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html113.《Metaheuristic Procedure to Optimize Transmission Delays in DVB-T Single Frequency Networks》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html114.《Adaptive Resource Allocation for MIMO-OFDM Based Wireless Multicast Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html115.《An Analytical Approach for Performance Evaluation of Hybrid (Broadcast/Mobile) Networks》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html116.《Cost-Aware Wireless Data Broadcasting》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html117.《Subspace-Based Semi-Blind Channel Estimation in Uplink OFDMA Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html118.《Performance of the Consumer ATSC-DTV Receivers in the Presence of Single or Double Interference on Adjacent/Taboo Channels》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html119.《A Cooperative Cellular and Broadcast Conditional Access System for Pay-TV Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html120.《A Narrow-Angle Directional Microphone With Suppressed Rear Sensitivity》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html121.《Peak-to-Average Power Ratio Reduction in OFDM Systems Using All-Pass Filters》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html122.《Development of Advanced Terrestrial DMB System》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html123.《HDTV Subjective Quality of H.264 vs. MPEG-2, With and Without Packet Loss》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html124.《Estimation of RF Electromagnetic Levels Around TV Broadcast Antennas Using Fuzzy Logic》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html125.《Statistical Multiplexing for Digital Audio Broadcasting Applications》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html126.《A Composite PN-Correlation Based Synchronizer for TDS-OFDM Receiver》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html127.《Application of Quantum-Inspired Evolutionary Algorithm to Reduce PAPRof an OFDM Signal Using Partial Transmit Sequences Technique》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html128.《Improved Decoding Algorithm of Bit-Interleaved Coded Modulation for LDPC Code》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html129.《Precoding for PAPR Reduction of OFDM Signals With Minimum Error Probability》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html130.《Network Design and Field Application of ATSC Distributed Translators》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html131.《On the Channel and Signal Cross Correlation of Downlink and Uplink Mobile UHF DTV Channels With Antenna Diversity》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html132.《Performance Evaluation of TV Over Broadband Wireless Access Networks》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html133.《IBC2010 Experience the State-of-the-Art》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html134.《Peak-to-Average Power Ratio Reduction of OFDM Signals With Nonlinear Companding Scheme》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html135.《Motion-Compensated Frame Rate Up-Conversion—Part I: Fast Multi-Frame Motion Estimation》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html136.《Comments on Equation (4) in “Single Frequency Networks in DTV”》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html137.《Motion-Compensated Frame Rate Up-Conversion—Part II: New Algorithms for Frame Interpolation》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html138.《A Novel Equalization Scheme for ZP-OFDM System Over Deep Fading Channels》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html139.《A Synchronization Design for UWB-Based Wireless Multimedia Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html140.《Frequency Domain Decision Feedback Equalization for Uplink SC-FDMA》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html141.《A 2 2 MIMO DVB-T2 System: Design, New Channel Estimation Scheme and Measurements With Polarization Diversity》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html142.《Impact of the Receive Antenna Arrays on Spatio-Temporal Availability in Satellite-to-Indoor Broadcasting》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html143.《Reducing Channel Zapping Time in IPTV Based on User's Channel Selection Behaviors》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html144.《On the Methodology for Calculating SFN Gain in Digital Broadcast Systems》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html145.《Statistical Multiplexing of Upstream Transmissions in DOCSIS Cable Networks》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html146.《Bit-Rate Allocation for Broadcasting of Scalable Video Over Wireless Networks》原文链接:https:///academic-journal-foreign_broadcasting-ieee-transactions_thesis/020*********.html147.《Full-Reference Video Quality Metric for Fully Scalable and Mobile SVC Content》。
基于导频受限短突发通信的高精度快速频偏估计袁静珍【摘要】针对导频符号辅助调制(PSAM)的导频受限短突发通信载波同步,提出了一种高精度快速的频偏估计算法.其基本原理是:利用自相关算子得到具有较大估计范围和较低信噪比门限的联合自相关算法,再利用互相关算子获得具有高精度的简化互相关算法.理论分析和仿真结果表明,提出的算法能够消除大频偏,而且还具有较低的复杂度和良好的解调性能.【期刊名称】《电讯技术》【年(卷),期】2018(058)010【总页数】5页(P1201-1205)【关键词】短突发通信;载波同步;导频符号辅助调制;频偏估计【作者】袁静珍【作者单位】韩山师范学院物理与电子工程学院,广东潮州521041【正文语种】中文【中图分类】TN911.231 引言近年来,短突发通信已经广泛应用于卫星遥感、深空通信等前沿领域,同时还将应用到第五代(5G)移动通信中[1-2]。
在这些通信领域中,通信双方的相对移动会产生多普勒效应,大多普勒频移会造成同步接收机无法实现相干解调,从而导致通信质量的急剧下降。
为了对抗大载波频偏,传统的估计算法可以分为数据辅助、非数据辅助两大类[3-4],其中,非数据辅助这一类估计算法的信噪比门限和复杂度较数据辅助估计算法高,因此,在短突发通信中,普遍采用基于已知的数据符号的数据辅助估计算法。
用于第二代数字视频广播(Digital Video Broadcasting-Second Generation,DVB-S2)的数据帧结构将90个已知的数据符号作为帧头,再以1 440个数据符号附加36个导频符号为单元周期地构成DVB-S2数据帧结构。
文献[5-6]提出了一种基于导频符号辅助调制(Pilot-Symbol-Assisted-Modulation,PSAM)的数据帧结构,即将一定长度的导频符号分成两部分,含有若干个连续符号的部分作为帧头,细分成单个离散符号的部分插至帧中和帧尾。
MeGUI参数设置1 Main图1 Main选项卡1.1 Encoding ModeABR:一次成形。
码率的分配来自即时验算。
通常不应该使用这个模式。
ConstQuantizer:固定量化值输出。
每一类帧采用相同的量化值来量化,使得全片的质量大体相近。
这里指定的应该是P帧的qr,I和B帧的qr由第2页的比例计算来得到。
CQ模式适用与追求质量而不计全片码率和文件大小的片子使用。
2pass - 1st/2nd:手动进行2pass压制。
1st时可以用turbo以加快压制速度。
在1st pass时MeGUI不会输出任何的视频信息,也不会生成一个空的文件。
Automated 2pass:自动2pass压制。
一次设置就能生成2个任务,加快了处理速度。
3pass:高级功能,供高级用户使用。
Const Quality:恒定画质,对于体积没有要求的压制。
Quality中的数字越大表明质量越差文件越小,越小表明质量越好文件越大,0代表无损。
1.2 TuningsTunings下拉菜单中有多种片源预置参数可供选择,一般可以直接选择Default默认,然后在后面的选项卡中再进行具体的设置)。
1.3 A VC ProfilesHigh Profiles:主要用于电脑播放和高清视频的制作。
Main Profiles:主要用于移动设备视频的制作,如PSP等设备。
Baseline Profiles:主要用于低清视频的制作。
1.4A VC Level移动视频选择Level 3.0即可。
电脑视频可选择Level 4.1以上的选项。
默认值:Unrestricted/Autoguess建议值:Unrestricted/Autoguess2 Frame-Type2.1 H.264 FeaturesDeblocking:开启环路滤波,除马赛克。
默认值:勾选(当不勾选该项时该参数显示为:--no-deblock)建议值:勾选Deblocking Strength:--deblock 0:0:设置环路滤波的AlphaC的参数,范围-6-6。
专利名称:Low complexity cost function for sub-pixelmotion estimation发明人:Richard John Allen申请号:US13867788申请日:20130422公开号:US09667960B1公开日:20170530专利内容由知识产权出版社提供专利附图:摘要:Methods and circuitry are provided for finding an optimum vector for motion-compensated prediction in video coding. Processing circuitry is operable to define a block in an image, the defined block having a plurality of pixels. A first set of residuals iscalculated using a first motion vector. Each residual of the first set of residuals corresponds to a respective pixel within the plurality of pixels. Additionally, a first plurality of absolute differences is calculated. Each absolute difference in the first plurality of absolute differences is between a pair of residuals in the first set of residuals. The first plurality of absolute differences is summed, and is compared to sums of absolute differences of residuals calculated using other motion vectors. The motion vector corresponding to the lowest sum may be identified as an optimum motion vector.申请人:Altera Corporation地址:San Jose CA US国籍:US代理人:Michael H. Lyons更多信息请下载全文后查看。
Research on Super-resolution Reconstruction Algorithm Based on L1 normA Thesis Submitted to Chongqing Universityin Partial Fulfillment of the Requirement for theMaster’s Degree of EngineeringByWang BinSupervised by Prof. He Zhong ShiSpecialty: Computer Software and TheoryCollege of Computer Science of ChongqingUniversity, Chongqing, ChinaApril, 201摘要在道路病害检测中,由于摄像机多处于野外环境,通常摄像设备较为简陋,获取的道路图片分辨率较低。
同时由于硬件成本较高,且成像系统本身的技术瓶颈,使得在很多领域高分辨率图像通常难以获取。
图像超分辨率重建技术的基本原理是通过软件手段从多幅具有非冗余互补信息的低分辨率图像重构出包含更多信息的高分辨率图像的过程。
超分辨率重建技术不需要在硬件上改变成像系统便能提高图像的空间分辨率,这样能极大降低高分辨率图像的获取成本。
超分辨率重建技术广泛应用的同时也存在着许多亟待解决的问题,如超分辨率重建过程是病态的,解不唯一的。
另外超分辨率重建的全过程涉及多个环节,如图像配准中的运动估计、插值放大、能量函数选择、算法求解、去模糊,如何保证高效的实现重建过程需要各个环节的优化协调。
本文主要研究内容及成果如下:①提出基于高斯金字塔的小十字形搜索算法:针对图像配准中求取运动位移估计计算量大的缺点,采用基于高斯金字塔思想的运动估计算法,通过使用高斯金字塔将图像分层,先在金字塔上层搜索运动偏移量,然后以此偏移量为初始偏移量,再在金字塔下层进行搜索,由于金字塔上层搜索范围远小于原图,因此能大大提高搜索速度。
中国科学技术大学硕士学位论文第三章新一代视频编码标准——H264/AVC得多,从而实现更高效的数据压缩。
量化是一种有损压缩过程,实际上就是将变换的系数除以相应的量化步长后四舍五入取整数。
tt.264/AVC采用的一个重要变换方法是类似DCT变换的4X4整数变换“”:”…,是一种只需加法和移位运算即可实现的低复杂度的精确变换。
同时,H.264/AVC将变换过程的系数缩放运算并入量化过程,并将缩放系数与量化步长联合通过查表实现,既简化了核心变换运算,又避免了量化过程的除法运算。
3.3.1变换的类型H.264/Avc采用各种优化技术提高了预测的效果,但同时对预测误差的编码提出更高的要求,希望预测误差编码无漂移(Drift—Free)。
另外,为了适应更广泛的应用环境,H.264/AVC希望变换只需16一bit运算。
而4X4整数变换很好地满足了以上要求,是H.264/AVC最重要、晟基本的一种变换。
一当宏块采用Intra_16X16预测方式时,由于图象相对平滑,交换后大部分能量集中在Dc系数上。
为了进一步减小Dc系数之间的相关性,将16个(图3.3.1中编号O一15)4X4块的Dc系数构成一个4X4的DC系数块(图3.3.1中编号一1),再进行Hadamard变换处理,以进一步减,j,DC系数之间的相关性。
对于色度分量也作类似处理,得N2个2×25jDC系数块(图3.3.1中编号16—17),然后再进行Hadamard变换。
采用这种分级变换结构进一步提高了压缩效率。
图3.3.144块的扫描顺序以及Dc块的构成”1H.264/AVC共支持3种变换方式,即针对4X4差值信号块的整数变换、针对4X4。
440C-CR30COMLOCK 00 01 02 03 04 05 06 07 08 0 10 1112 13 14 15 16 17 18 1 20 21+DC -DC 24 24Note: Configuration is what we tested,not the max capacityMicro800 PTO Motion PopularConfiguration Drawing• Optimized for compatibility with MicroLogix™ , SLC™, and Micro800 controllers.• Features include unicode language switching, alarm messages and history, and basic recipe capability• User Interface through Web browser•Panelview Component Configuration Publication CCSIMP-QR002A-EN-P• Supports communication via RS-232 (DH-485), RS-232 (DF1), RS-485 and Ethernet• Two USB ports for transferring files or updating firmware , Supports SD memory cardsPanelview Component Capacity• 4.3, 5.7 ,10.7 inch Display Sizes • Monochrome Transmissive FSTN Options/Color Transmissive TFT, Analog Touch• Automatic motor recognition and gain settings • Cost-effective motion control solution for smaller, low-axis count applications• Flexible command interfaces including digital I/O, analog, pulse train and Modbus-RTU• Configurable with Connected Component Workbench SoftwareKinetix 3 Configuration• Input volt range of 170…264V AC single and three-phase• Continuous power output of 0.05 W…1.5 kW• Modbus-RTU serial communications for changing drive parameters during runtimeKinetix 3 Capacity• CCAT contains Building blocks for indexing up to 3 axisKinetix 3 PerformanceExternal Power Supply• Embedded I/O 28 In, 20 Out,Base Analog I/O channels via plug in or expansion. Maximum Digital I/O :132, Max Expansion I/O Modules:4, 6 High Speed Counters, 3 PTOMicro850 Configuration• Designed for larger standalone machine applications that require more I/O or higher performance analog I/O• Number of I/O points embedded in the base 24,48 points (2085 expansion I/O’s)Micro850 PerformanceMicro 850 Overview 2080-LC50-48xxx Embedded communication USB, Serial, Ethernet Basic I/O size 48 points Max number of I/OUp to 132Max number of Expansion I/O 4Motion axis (Only available in model with transistor output)Up to 3 axis HSCUp to 6 embeddedTo Plant Wide Ethernet network To Computer for ProgrammingStratix 2000PanelView ComponentMicro850Kinetix 3External Encoder Differential / Open collector typeStratix 2000 Unmanaged Switch - 5 Copper PortsPanelview Component Performance 3• Compact DIN rail mount safety relay• 22 embedded Safety I/O • Support for 16 standardI/O via plug-ins440C-CR30 Safety Relay Micro850 Capacity• LC50-48xxx has 10Ksteps of instructions and 20KB of data memory and supports 5 plug in I/O modules & 4 expansion modules.• Supports Ladder Programming,Function Block Diagram, Structured TextC400Micro850 with2080-SERIALISOL (slot 1) & 2080-MOT-HSC (slot 2-4) plug-in modules.• Rotary Motors • TL-Series • Linear Motors• LDC-Series Iron Core • LCL-Series IronlessServo motorsSafety Diagnostics to Micro850 and to PVcSafety I/OConnectionsPulse Train Outputs (24V)High Speed feedback signals from Kinetix 3 Servo drive 5V Differential3High Speed feedback signals from External Encoder . (Up to 5V/24V, Up to 100kHz open collector 250kHz differential)Kinetix 3Kinetix 3Protocol RS-232Micro850: Modbus RTU Master 440C-CR30: SlaveDaisy ChainRS-485 to three Kinetix 3About this ConfigurationThis Micro 850 Controller based low cost system demonstrates the power and scalability of the Component class on an Ethernet I/P based network. This system utilizes standard Ethernet I/P technology to connect to plant wide network. This configuration also highlights Micro850 Capabilities in Connecting Safety relay, Motion control & HMl .A key advantage of this architecture is the ability to use Connected Components Workbench a common integrated environment for programming, configuration, commissioning, and motion tools for the Kinetix 3 ,Panelview Component , and Micro 800 Controller family of products.About the ProductsMicro 850 Controllers• The Micro850 controller is equipped with the same form factor, plug-in support, instruction/data size and motion capabilities as the 24-pt and 48-pt Micro830 controllers• Support up to four Micro850 Expansion I/O modules, Up to a maximum of 132 I/O points (with 48-pt model)• Designed for larger standalone machine applications that require more I/O or higher performance analog I/O than supported by Micro830.• Structured Text, Ladder Diagram and Function Block editors that support • Symbolic addressing• Connected Components Workbench software is used for PLC programming, HMI, drives and safety relay configurationMicro 850 Plugin Module2080-MOT-HSC• The new Allen-Bradley Micro800 Motion Feedback Axis plug-in enhances high speed counter function to Micro800 controllers.• With input frequency to 250 kHz, it supports both 5V differential line driver input and 24VDC input with 1 physical digital output and 15 virtual digital outputs.• Its programmable limit switch function makes it easy for user to turn ON and OFF output based on the counter value.• Through its motion feedback axis mode, virtual axis mode and touch probe function, the new plug-in addition provides an easy and intuitive way for users to obtain motion positioning information..440C-CR30 Safety Relay• Programmable Safety Relay designed for flexibility in a simple safety system• Predefined Safety Functions provide an easy to use system that is programed and configured using CCW software• Compact DIN rail mount designed with 22 embedded Safety I/O and support for 16 standard I/O via plug-ins in a 110mm wide package • Built in communications for programming and HMI diagnostic displayPanelView Component• A full line of displays ranging from 2” to 10”,• Communicate to MicroLogix, Micro800 and SLC controllers via serial (RS232 or RS422/RS485) networks on all terminals, and Ethernet on the C400, C600 and C1000 displays.• Features include unicode language switching, alarm messages and history, and basic recipe capability• PanelView Component software (DesignStation) is an offline programming software offering better user experience with significant DesignTime performance improvementKinetix 3 Servo DrivesSingle-axis solution for low-complexity motion applications, with or without a PLC • Digital I/O, analog, preset velocity, and pulse-train command interfaces• Performs indexing on up to 64 points through serial communication or over digital I/O • 170…264V AC, (200V-class) single-phase or three-phase1783 Stratix 2000™ Unmanaged Switches• Multiple port count and fiber options are available • Copper ports are 10/100 Mbps, full- or half-duplex• Fiber ports are 100 Mbps, full-duplex, with an LC fiber optic connector • Ideal Ethernet switch for small, isolated networks • Auto negotiates speed and duplex settings • Operates on 20V AC or 24V DC power • Automatic cable cross-over detectionFor More Information and HelpFor more information contact your local distributor or Rockwell Automation sales representative.• • Publication Library • My Support• A – Z Product DirectoryBill of MaterialsQty Catalog #DescriptionSystem: Controller Hardware 12080-LC50-48QBB Micro 850 Controller12080- SERIALISOL RS232/485 isolated serial port Plug in Module 32080-MOT-HSC Micro800 Motion Feedback Axis plug-in ModuleHMI Hardware12711C T4T PanelView C400, touch terminalSafety Relay Module1440C-CR30-22BBB Compact DIN rail mount safety relay designed with 22 embedded Safety I/O and support for 16 standard I/O via plug-ins 11761-CBL-HM02Communication cableMotion Hardware22071-AP1Kinetix 3 , 2.38Amps 2TL-A130P-BJ32AA Servo Motor22090-CFBM6DD-CCAA03Cable,Feedback,M-Circ.Plastic.D-DB15,INCR CC ,Std,3M 22090-CPWM6DF-16AA03Cable,Power,M-Circ.Plastic.D-Flying Lead,16AWG ,Std,3M12090-CCMDSDS-48AA01Serial Communication cable between K3 drive & 2080-SERIALISOL plug-in 1Serial Communication cable between K3 drivesNote: Catalog numbers consist of characters, each of which identifies a specific version or option for that component. Reference Publication GMC-SG001-EN-E Kinetix Motion Control Selection Guide for additional informationSystem: Communication Hardware11783-US05T Stratix 2000 Unmanaged Switch - 5 Copper Ports 41585J-M8TBJM-1M Shielded Ethernet CableConfiguration Tool Required19328-SO001-EN-CConnected Components Workbench software。
doi:10.3969/j.issn.1003-3114.2022.02.017引用格式:施育鑫,鲁信金,孙艺夫,等.MIMO通信模型下的相关矩阵组低复杂度设计[J].无线电通信技术,2022,48(2):327-335.[SHIYuxin,LUXinjin,SUNYifu,etal.Low⁃complexityDesignofCorrelationMatrixGroupunderMIMOCommunicationModel[J].RadioCommunicationsTechnology,2022,48(2):327-335.]MIMO通信模型下的相关矩阵组低复杂度设计施育鑫1,鲁信金2,孙艺夫2,雷㊀菁2,李玉生1(1.国防科技大学第六十三研究所,江苏南京210000;2.国防科技大学电子科学学院,湖南长沙410000)摘㊀要:矩阵组常用于无线通信中的数据表示㊂在多输入多输出(Multiple⁃InputMultiple⁃Output,MIMO)通信模型中,基站利用信道数据设计适应信道的最小均方误差(MinimumMeanSquareError,MMSE)均衡器接收矩阵组,以低复杂度地处理来自多用户的上行数据㊂首先分析了矩阵数据的关联性,通过时谱图确定矩阵组在时间维所具有的强相关性;其次采用插值算法进行低复杂度的矩阵估计,并提出最大插值比搜索算法计算各类插值算法的性能及其复杂度;接着利用一种改进的Strassen矩阵求逆算法来降低MMSE求逆过程的复杂度㊂相比传统的接收矩阵组,显著降低了计算复杂度㊂关键词:MIMO通信模型;奇异值分解;最小均方误差;计算复杂度中图分类号:TN929.5㊀㊀㊀文献标志码:A㊀㊀㊀开放科学(资源服务)标识码(OSID):文章编号:1003-3114(2022)02-0327-09Low⁃complexityDesignofCorrelationMatrixGroupunderMIMOCommunicationModelSHIYuxin1,LUXinjin2,SUNYifu2,LEIJing2,LIYusheng1(1.63rdResearchInstitute,NationalUniversityofDefenseTechnology,Nanjing210000,China;2.SchoolofElectronicScience,NationalUniversityofDefenseTechnology,Changsha410000,China)Abstract:Matrixgroupsareoftenusedtorepresentdatainwirelesscommunication.IntheMIMO(Multiple⁃InputMultiple⁃Output)communicationmodel,thebasestationoftenusesthechanneldatatodesignachannel⁃adaptedMMSE(MinimumMeanSquareError)equalizerreceivingmatrix,inordertoprocesstheuplinkdatafrommultipleuserswithlowcomplexity.First,thecorrelationofmatrixdataisanalyzed,andthestrongcorrelationofthematrixgroupinthetimedimensionisdeterminedthroughtime⁃spectrogram.Secondly,aninterpolationalgorithmisusedforlow⁃complexitymatrixestimation,andamaximuminterpolationratiosearchalgorithmisproposedtocalculatetheperformanceandcomplexityofvariousinterpolationalgorithms.ThenanimprovedStrassenmatrixinversional⁃gorithmisusedtoreducethecomplexityoftheMMSEinversionprocess.Comparedwiththetraditionalreceivingmatrix,thecomputa⁃tionalcomplexityissignificantlyreduced.Keywords:MIMOcommunicationmodel;singularvaluedecomposition(SVD);MMSE;computationalcomplexity收稿日期:2021-12-16论文来源:基于2021年 华为杯 第十八届中国研究生数学建模竞赛A题建模改写,参赛团队获得竞赛一等奖(获奖率约1.1%)及华为专项奖(共10项)㊂0㊀引言矩阵常常被用于表示无线通信㊁图像视频处理㊁计算机视觉㊁相控阵雷达的数据表示㊂随着用户需求的不断增加,数据规模㊁通信阵列的持续扩大,矩阵的大小和维度也随之快速增加,这给矩阵的数据存储㊁算法计算带来了很大的困难㊂矩阵的关联性是指矩阵数据在某些维度上的相关特性,例如视频中时间相邻的帧具有很强的矩阵关联性㊂因此,充分挖掘矩阵间关联性,以实现低复杂度的计算具有十分重要的价值和意义㊂1㊀问题描述对于所给定复数矩阵H=Hj,k{},Hj,kɪMˑN,j=1,2, ,J;k=1,2, ,K㊂其中,矩阵之间以及同一矩阵的元素之间有一定的相关性,包括:相同j下标㊁不同k下标的矩阵间存在一定的关联,即Hj,1,Hj,2,Hj,3, ,Hj,K{}间存在关联;且矩阵的各个元素间h(j,k)m,n{},m=1,2, ,M;n=1,2, ,N也存在关联,矩阵Hj,kɪMˑN,j=1,2, ,J;k=1,2, ,K可表示为:Hj,k=h(j,k)1,1h(j,k)1,2h(j,k)1,3h(j,k)1,Nh(j,k)2,1h(j,k)2,2h(j,k)2,3 h(j,k)2,N︙︙︙︙h(j,k)M,1h(j,k)M,2h(j,k)M,3 h(j,k)M,Néëêêêêêùûúúúúú㊂(1)此外定义矩阵组H=Hj,k{}的一组数学运算,其中间结果V=Vj,k{}由式(2)给出:Vj,k=svd(Hj,k)Hj,k=Uj,kSj,kV Hj,kVj,k=V Hj,k(:,1:L)ìîíïïïï,(2)式中,j=1,2, ,J;k=1,2, ,K,svd(㊃)为矩阵奇异值分解(SingularValueDecomposition,SVD)中求解右奇异向量的过程;Vj,k是由Hj,k的前L个右奇异向量构成的矩阵,维度为NˑL㊂为得到最终输出结果W=Wj,k{},先将不同j下标㊁相同k下标的Vj,k进行横向的拼接,得到维度NˑLJ的Vk=[V1,k, Vj,k, VJ,k],然后根据式(3)获取Wk:Wk=Vk(VHkVk+σ2I)-1,(3)式中,σ2为固定常数;Wk维度同Vk;I为单位矩阵,维度为LJˑLJ㊂最后将各Wk按如式(4)进行拆解:Wk=[W1,k, Wj,k, WJ,k],(4)式中,Wj,k为Wk中顺序排列的子矩阵,维度为NˑL㊂为了降低计算和储存的复杂度,分析相关矩阵组的关联性,通过建模对输出结果进行估计,建模过程可用式(5)表示:W^=f(H),(5)式中,W^即为对输出结果W的建模估计㊂该建模过程可拆分为如式(6)的两个步骤㊂V^=f1(H)W^=f2(V^){,(6)式中,f1(㊃)表示从输入矩阵组H到中间结果V的建模过程,V^表示中间结果V的建模估计,f2(㊃)表示从中间结果V到最终结果W的建模过程,W^表示最终结果W的建模估计㊂定义W的建模估计精度为:ρl,j,k(W)=W^Hl,j,kWl,j,k2W^l,j,k2 Wl,j,k 2,l=1,2, ,L,(7)式中, ㊃ 2表示矢量的欧几里得范数(即2范数,对于列矢量a, a 2=aHa),Wl,j,k表示Wj,k的第l列㊂上式中,W^Hl,j,kWl,j,k为复数标量,此处取其欧几里得范数即获取其模值㊂为描述方便,额外定义W的最低建模精度为ρmin(W):ρmin(W)minlɪ{1,2, ,L}jɪ{1,2, ,J}kɪ{1,2, ,K}ρl,j,k(W),(8)式中,minl,j,k(㊃)表示在l,j,k三个维度上取最小值㊂另外,中间结果V的建模估计精度ρl,j,k(W)的定义及最低建模精度ρmin(W)的定义与此相同㊂计算复杂度定义为由矩阵组H计算得到结果矩阵组W所需要的总计算复杂度㊂复数矩阵运算可拆解为基本的复数运算,而基本的复数运算又可进一步拆解为基本的实数运算㊂例如,复数乘法(a+bj)(c+dj)=(ac-bd)+(ad+bc)j的复杂度为4次实数乘法和2次实数加(减)法㊂实数基本运算的复杂度按照表1计算㊂表1㊀实数基本运算的计算复杂度Tab.1㊀Computationalcomplexityofbasicoperationsonrealnumbers运算类型计算复杂度加(减)法1乘法3倒数25平方根25自然指数25自然对数25正弦25余弦25其他1002021研究生数学建模A题提供的数据集附件(Data1 Data6)给出的详细数据,包括输入矩阵组㊁标准中间矩阵组和标准输出矩阵组的数据及其维度,其中M=4,N=64,L=2,J=4,K=384,σ2=0.01,数据为十进制格式㊂根据所给数据Data1 Data6中的M=4,N=64,J=4,可对应通信模型中共有J=4个用户,每个用户的发射天线数为M=4,基站的接收天线数为N=64㊂L=2表示取信道衰落程度最低的2个信道向量,即对矩阵进行压缩㊂K表示信道探测时隙数㊂σ2表示信道中高斯白噪声的噪声方差㊂基于给定的所有矩阵数据,本文通过分析数据间的关联性,解决相关矩阵组的低复杂度计算问题,即以减少计算复杂度为目标进行模型优化㊂设计相应的近似分析模型W^=f(H),在满足ρmin(V)ȡρth=0.99的情况下,使根据表格计算的总计算复杂度最低㊂2 通信模型建立建立基于矩阵的多输入多输出(Multiple⁃InputMultiple⁃Output,MIMO)通信模型[1]如图1所示,J个用户发送信息,信号经过信道H到达基站,基站有N根接收天线㊂图1㊀矩阵关系的通信模型建立示意图Fig.1㊀Schematicdiagramofestablishingcommunicationmodelofmatrixrelationship如图2所示,对于某个用户,拥有M个天线,各个天线均可与基站天线进行通信㊂其通信信道矩阵Hj,k=Uj,kSj,kVHj,k,通过SVD分解,将信道矩阵分解成方向酉矩阵和信道随机衰落矩阵的乘积,其中,Sj,k为随机矩阵,代表波束随机衰落主信道矩阵,Uj,k和V Hj,k分别是用户和基站特征向量矩阵的相关矩阵㊂由于VHj,k与基站和用户位置相关且各个节点的位置相对固定,可以取信道衰落程度最低的L个信道向量压缩V Hj,k矩阵,即Vj,k=VHj,k(:,1:L)㊂以上模型建立与式(2)一致㊂图2㊀单个用户和基站的通信示意图Fig.2㊀Schematicdiagramofcommunicationbetweenasingleuserandabasestation此时,可使用压缩矩阵Vj,k来表示H矩阵㊂进一步,引入最小均方误差(MinimumMeanSquareError,MMSE)的概念来解释题设条件㊂如图3所示的信道模型中,信号x经过信道V=Vj,k,由于白噪声n的影响,接收信号y可表示为:y=Vx+n㊂(9)图3㊀信号传输模型(求解MMSE流程)Fig.3㊀Signaltransmissionmodel(processofsolvingMMSE)MMSE的目的是找到一个矩阵W=Wj,k{},使得Wy更加接近x㊂得到x =Wy与原始发送信号x的差值为:e=x -x=Wy-x㊂(10)此时的MMSE为:MMSE=EeHe{}㊂(11)假设接收到的数据y和误差e是不相关的,即Ee㊃yH{}=0㊂(12)将式(10)代入式(12)可得:E(Wy-x)㊃yH{}=0㊂(13)将式(13)左边进一步展开可得:㊀㊀E(Wy-x)㊃yH{}=EWyyH{}-ExyH{}=WEyyH{}-ExyH{}㊂(14)由式(13)和式(14)可得:W=ExyH{}EyyH{}-1㊂(15)接下来对EyyH{}和ExyH{}进行处理,首先对于EyyH{},将其进一步展开:㊀㊀EyyH{}=E(Vx+n)(Vx+n)H{}=EVxxHVH+VxnH+nxHVH+nnH{}㊂(16)此处假设输入信号x和噪声n不相关,则nxH与nxH值为0,可得:㊀㊀㊀EyyH{}=EVxxHVH+nnH{}=VExxH{}VH+EnnH{}=V(P㊃I)VH+σ2㊃I,(17)式中,P为发送信号x的能量,σ2为噪声n的方差㊂其次对于ExyH{},展开如下:㊀㊀㊀ExyH{}=Ex(Vx+n)H{}=ExxHVH+xnH{}=ExxHVH{}=ExxH{}VH=(P㊃I)VH㊂(18)得到EyyH{}和ExyH{}后,将其代入式(15)可得到W的表达式:㊀㊀W=ExyH{}EyyH{}-1=(P㊃I)VH(V(P㊃I)VH+σ2㊃I)-1=P㊃VH(PVVH+σ2㊃I)-1=VH(VVH+㊃I)-1㊂(19)当发送信号x的能量P为1时,则可得:W=VH(VVH+σ2㊃I)-1㊂(20)综合上述分析,MIMO模型中利用信道数据计算信道的MMSE均衡器接收矩阵的复杂度主要来源于SVD分解与式(20)中的矩阵求逆㊂3㊀相关矩阵组的低复杂度计算由前文可知,H㊁V和W之间的关系可以由图4表示㊂图4㊀H㊁V和W的关系示意图Fig.4㊀RelationshipofH,VandW3.1㊀利用相关矩阵组的关联性降低计算复杂度利用相关矩阵组的关联性以降低计算复杂度,其具体分析及操作如下㊂3.1.1㊀矩阵数据的关联性对于信道系数复数矩阵H=Hj,k{},其中Hj,kɪMˑN,j=1,2, ,J;k=1,2, ,K㊂因此,该矩阵是一个MˑNˑJˑK维的信道矩阵,其中M表示接收天线的数量,N表示发射天线的数量,J表示用户数量,K表示时隙个数㊂由前文建立的通信模型,对H矩阵内在的关联性进行分析㊂首先,Hj,k内部的关系可表示为不同天线构建出的不同信道之间的相关性㊂在一般高斯白噪声信道条件下,天线阵列之间固定的距离和入射角关系将带来一定的规律,但数据集中未能发现Hj,k内部可靠的相关特性㊂这可能是由于天线之间的方向性㊁距离之间的差异较大,使得信道在空间上的相关特性不再明显㊂考虑不同k时隙,同一用户j的信道系数情况,即MN个信道的时间相关性㊂图5给出了MN个信道在时隙k=1,2,3情况下的幅度响应和相位响应㊂可以看出,在不同k下的幅度和角度的变化程度不大,这可以理解为在相关时间内,信道的变化很小,这进一步验证通信建模的可行性㊂(a)不同信道系数的幅度响应(b)不同信道系数的相位响应图5㊀同j不同k的H块之间的幅度和相位响应关系Fig.5㊀AmplitudeandphaseresponserelationshipbetweenblocksHwiththesamejanddifferentk基于上述两层分析,得出Hj,k块在时间维上的相关性㊂而在时间维上利用SVD与MMSE求W矩阵是独立的,无法直接用于算法简化,为此,本文利用时间相关性,并运用插值算法直接估计W矩阵㊂具体的,由于同j不同k的块在时间维上的相关性,在经过函数W^=f(H)后,具有相关性的输入H,与得到的W之间将保持相关性㊂因此,可以利用同j不同k的W的相关性,通过插值算法获取某些k值上的W矩阵㊂这将直接减少SVD与MMSE求逆过程的计算数量㊂为了便于理解,将L维与J维(用户数)进行合并,因此W矩阵可以改写为三维矩阵㊂将矩阵按照K维展开,不同k下标的矩阵可以由二维平面示意,其插值过程如图6所示㊂图6㊀W矩阵的插值示意图Fig.6㊀SchematicdiagramofinterpolationofWmatrix3.1.2矩阵数据W的插值算法对于矩阵数据W的插值算法,采用linear插值法㊁spline插值法与Pchip插值法进行建模插值[2-4]㊂此处,引入 最大插值比 作为评估方法来评价插值性能,参数寻优的过程可以表示为:㊀㊀Rate=max1R{}s.t.ρmin(W)minlɪ{1,2, ,L}jɪ{1,2, ,J}kɪ{1,2, ,K}ρl,j,k(W)>0.99,(21)式中,R表示每隔R个点进行一次插值运算㊂因此,式(21)表示插值结果满足最小建模精度的约束下,使得插值数量越多的寻优目标㊂因此,为满足题设对于拟合后W矩阵对最小建模精度的要求,需选择合适的插值方法并计算最大插值比,为此进一步提出了最大插值比搜索方法,以评估在不同信道条件数据集下可用的插值参数,最大插值比搜索算法的详细过程在算法1中给出,其基本思想为通过给定的初始插值比Rate,和选定的插值类型,不断迭代和逼近给定插值类型下的满足要求的最大插值比㊂当取得的插值比越大时,意味着W矩阵的更多部分可以通过在K维度上的相关性插值得到,不需要通过SVD与MMSE求逆过程,这将大大减少计算的复杂度㊂此外,Data集最大插值比的计算过程可以理解为适应信道的训练过程㊂在后续过程中,在外部信道条件未剧烈改变的情况下,不需要再次执行,因此最大插值比搜索可以在线下执行,不会影响算法的复杂度㊂算法1所提出的最大插值比搜索算法输入:训练数据集Data,包含通过MMSE计算的标准W矩阵;初始化:初始插值比Rate=1/R,插值间隔R的初始值可以设定为R=2㊂R的中止值可以设定为Rmax=200选定的插值类型:Linear,Spline,Pchip㊂执行过程:ForR<Rmax=200㊀ForiN=1:N㊀㊀ForiLJ=1:LJ1.根据选定的插值比Rate,确定插值点所在的横坐标序列x,其中x=[1,R,2R, ,pR]T,pRɤK=3842.合并J个用户的W矩阵,使其降维为NˑLJˑK3.将W矩阵的第三维度中与x重合的部分置零,新建为W^,在Matlab中可采用setdiff函数㊂置零部分准备后续进行插值填充4.执行插值操作W^(iN,iLJ,1:K)=interp1(z,y,1:K,插值类型)其中interp1表示插值函数,具体使用方法可参考Matlab中对应函数5.评估插值结果ρl,j,k(W)=W^Hl,j,kWl,j,k 2W^l,j,k 2 Wl,j,k 2,l=1,2, ,L若ρmin(W) minlɪ{1,2, ,L}jɪ{1,2, ,J}kɪ{1,2, ,K}ρl,j,k(W)>0.99,则跳出循环(break)㊀㊀Endif㊀EndifEndif输出:插值类型,最大插值比Rate表2给出了3种常见插值方法在6个Data集的最大插值比㊂由于Data1 Data6来自不同的信道条件,因此同一插值方法在插值过程中计算出的最大插值比有较大区别㊂例如Data3数据集的最大插值比显著较小,这可以理解为信道的时变性强或受到干扰噪声较大,插值算法在此时难以满足要求,需要降低最大插值比㊂进一步,横向对比3种插值方法,可见在多数的数据集下,Linear插值的最大插值比最小,性能最差,这是由于简单的Linear插值精度较低㊂Spline插值与Pchip插值的最大插值比性能相近,Spline插值在Data1与Data3上表现比Pchip插值较好㊂从原理上分析,可以理解为当基础函数振荡时,Spline比Pchip更好地捕获点之间的移动,后者会在局部极值附近急剧扁平化,这在该场景下带来了更好的插值性能[5-6]㊂表2㊀3种插值方法在6个Data集的最大插值比Tab.2㊀Maximuminterpolationratioofthethreeinterpolationmethodsin6datasets最大插值比Linear插值Spline插值Pchip插值Data11/201/151/20Data21/301/301/30Data31/1301/1111/118Data41/31/21/2Data51/91/91/9Data61/171/101/10复数矩阵运算可拆解为基本的复数运算,而基本的复数运算又可进一步拆解为基本的实数运算,实数基本运算的复杂度按照表1计算㊂对于Linear插值法,根据上述描述,可得需要加减法6次㊁乘法2次㊁倒数1次,且对于复数,幅度和相位要分别插值,总复杂度是实数插值的两倍㊂然而,特殊的是这里插值点的横坐标是均匀的,因此计算的复杂度可以大大简化,仅需要3次加法㊁2次乘法和1次倒数,因此总复杂度为68㊂每个插值时刻k,共需要NJL次插值,因此需要复杂度68NJL㊂本文的参数取值为N=64,J=4,L=2,因此计算复杂度为17408㊂对于Spline插值法,其计算步骤为:求三次函数的系数,然后将插值点横坐标代入三次函数,计算又需要6次加法㊁6次乘法㊂且对于复数,幅度和相位要分别插值,总复杂度是实数插值的2倍,因此总复杂度为124,每个插值时刻k,计算复杂度124NJL㊂取本文参数,计算复杂度为31744㊂对于Pchip插值,由于其性能不如Spline且计算复杂度相似,因此不在此处进行考虑㊂3.2㊀降低矩阵求逆的计算复杂度对于求解逆矩阵Vk(VHkVk+σ2I)-1过程中的计算复杂度,当使用高斯消元法时[7],求解维度LJˑLJ的矩阵的逆矩阵的复杂度近似为O((LJ)3);当矩阵求逆过程中使用的矩阵乘法使用文献[8]中的Strassenᶄs方法时,其提出的矩阵相乘公式将常规的矩阵相乘的运算量减少很多,可以将上述复杂度降低到O((LJ)2.807)㊂为此,进一步研究矩阵求逆降低复杂度算法,明显看出Vk(VHkVk+σ2I)-1为Hermite正定阵[9],采用了一种改进的Strassen矩阵求逆算法[10],该算法结合Strassen矩阵求逆的高效性以及Hermite正定阵的共轭对称性特点,使得算法运算量小,结构也简化许多㊂首先,对于矩阵分块直接求逆,假设一个N阶(这里的N被重新定义)的方阵A,分块如下:A=[a11]nˑn[a12]nˑm[a21]mˑn[a22]mˑnæèçöø÷NˑN㊂(22)设A的逆矩阵分块如下:A=[c11]nˑn[c12]nˑm[c21]mˑn[c22]mˑnæèçöø÷NˑN,(23)则根据矩阵分块求逆的原理有:c11=(a11-a12ˑa121ˑa21)-1c12=-c11ˑa12ˑa-122c21=-a-122ˑa21ˑc11ìîíïïï,(24)式中,对于NˑN阶矩阵,需要的矩阵相乘的运算量级为N3㊂对于n>1,m>1,直接分块求逆算法在具体实现中需要利用递归实现,具体的运算量按照复数乘加次数统计,对于上述的直接求逆算法,以乘加次数统计理论运算量,式(24)各部分运算量:c11的运算量为4m2n+4mn2-mn,c12和c21的运算量相等,均为4m2n+4mn2-6mn,同理,c22的运算量为4m2n+4mn2-m2,为此,矩阵A利用一次分块求逆的总的运算量为:㊀㊀㊀T(1)(N)=T(m)+T(n)+16m2n+12mn2-13mn-m2+4m3,(25)式中,T(1)(N)表示利用一次矩阵分块求逆算法计算矩阵求逆的总计算量,T(m)和T(n)分别表示对m和n阶复矩阵求逆所需的运算量㊂由式(24)可知,经过一次分块求逆之后的运算量依然很高,即T(1)(N) O[(max(m,n))3],同样可知,式(24)中的(a11-a12ˑa-122ˑa21)-1和a-122可以继续作为需要求逆的复矩阵,利用式(22) (24)可以继续分块求逆,所需运算量即式(25)中T(m)和T(n)部分㊂采用改进的Strassen矩阵求逆算法,结合式(22) (23),Strassen算法应用到求逆运算有如下公式:R1=a-111R2=a21ˑR1R3=R1ˑa12R4=a21ˑR3R5=R4-a22R6=R-15c12=R3ˑR6c21=R6ˑR2R7=R3ˑc21c11=R1-R7c22=-R6ìîíïïïïïïïïïïïïïïïï㊀㊀㊂(26)式中,对于NˑN阶矩阵,需要的矩阵相乘的运算量级为Nlog62=N2.585,相比于矩阵分块直接求逆,运算量随着矩阵维数将有显著降低㊂按照前面相同的运算量统计方法,根据式(25),矩阵利用一次分块求逆的总运算量为:㊀㊀㊀㊀T(1)=T(m)+T(N)+13m2n+11mn2-4mn+3m2㊂(27)对于n>1,m>1,Strassen矩阵求逆算法也是利用递归实现的,但因为Strassen算法减少了矩阵复乘次数,所以相比直接分块的常规算法运算量有明显降低㊂又由于a11,a22,a-122均为Hermite矩阵,且aH12=a21,代入到式(24)中得R3=RH2,c21=cH12,根据Her⁃mite矩阵的共轭对称性,式(26)可进一步改写为:R1=a-111R2=a21ˑR1R3=RH2R4=a21ˑR3R5=R4-a22R6=R-15c12=R3ˑR6c21=cH12R7=RH2ˑc21c11=R1-R7c22=-R6ìîíïïïïïïïïïïïïïïïï,(28)式中,对于NˑN阶矩阵,需要的矩阵相乘的运算量级为N2㊂为了便于比较新求逆算法的性能改善,按照前面相同的运算量统计方法,根据式(28),矩阵A利用一次分块求逆的总的运算量为:T(1)(N)=T(m)+T(N)+8m2n+8mn2+mn,(29)式中,T(1)(N)表示利用改进算法计算一次矩阵求逆的运算量,T(m)和T(n)部分表示对m和n阶复矩阵求逆所需的运算量㊂由式(28)可知,经过一次分块求逆之后的运算量T(1)(N) O[(max(m,n))3]㊂同样可以得出,式(28)中的R-15可以继续作为需要求逆的复矩阵,利用式(28)可以继续分块求逆,所需运算量即式(29)中T(m)和T(n)部分㊂和矩阵直接分块求逆算法相比,新的求逆算法虽然增加了一些加减运算,但复乘次数降低㊂对于维数较高的矩阵,其中有大量的复矩阵运算,复乘消耗的运算量将远大于加减法,而这个运算量随着矩阵维数增加将有显著增大,因此新算法对于复乘次数的降低将显著改善求逆运算的实时性能㊂和常规Strassen矩阵求逆算法相比,改进的算法由于利用了求逆矩阵的特点,即对Hermite矩阵进行求逆运算,所以在运算量和算法复杂度上都有明显的降低㊂常规算法计算一次矩阵求逆的计算复杂度如表3所示,改进算法计算一次矩阵求逆的计算复杂度如表4所示㊂表3㊀常规算法计算一次矩阵求逆的计算复杂度Tab.3㊀Computationalcomplexityoftheconventionalalgorithmtocalculatetheinversionofamatrix单项乘法次数加法次数其他R1T(n)求逆R24mn24mn2-2mnR34mn24mn2-2mnR44m2n4m2n-2m2R52m2R6T(m)求逆R74mn24mn2-2n2c112n2c214m2n4m2n-2mnc124m2n4m2n-2mnc22m2复杂度合计9mn(4m+4n-1)+3m2+T(n)+T(m)表4㊀改进算法计算一次矩阵求逆的计算复杂度Tab.4㊀Improvedalgorithmtocalculatethecomputationalcomplexityofamatrixinversion单项乘法次数加法次数其他R1T(n)求逆R24mn24mn2-2mnR3R44m2n4m2n-2m2R52m2R6T(m)求逆R74mn24mn2-2n2c112n2c21c124m2n4m2n-2mnc22m2复杂度合计8mn4m+4n-12()+3m2+T(n)+T(m)在本文中,当矩阵维度为8(J=4)时,改进算法总共复杂度为4776,常规算法总共5255,复杂度降低10.03%㊂进一步,图7给出了不同用户数量J时,改进的Strassen算法与常规Strassen算法的复杂度比较㊂可以看出,随着用户数量增加,矩阵求逆时的维度增加,采用改进的Strassen算法对复杂度的降低更加明显㊂图7㊀在不同用户个数J下,改进的Strassen算法与常规Strassen算法的复杂度比较Fig.7㊀UnderdifferentnumberofusersJ,thecomplexitycomparisonoftheimprovedStrassenalgorithmandtheconventionalStrassenalgorithm3.3㊀所提算法对最小建模精度的影响利用改进的Strassen矩阵求逆算法求得的W^Hl,j,k与原来的矩阵求逆算法求得的Wl,j,k进行建模估计精度计算,对于所给的数据集Data1 Data6,仿真不同数据集的最小建模精度,发现各数据集最小建模精度均为1,如图8所示㊂可见所提改进Strassen矩阵求逆算法不会对建模精度带来影响,这是因为该算法采用了分块迭代方法在变换的过程中不会带来计算误差㊂图8㊀采用改进的矩阵求逆算法后对各数据集最小建模精度的影响Fig.8㊀Influenceoftheimprovedmatrixinversionalgorithmontheminimummodelingaccuracyofeachdataset3.4㊀综合复杂度分析本节分别对插值算法和改进的矩阵求逆的综合复杂度进行分析㊂利用相关矩阵组的关联度降低计算复杂度,即通过Spline插值操作降低SVD的总复杂度㊂其中,所需的乘法次数(5MN2-MN)㊁加法次数(3MN2-MN)㊁除法次数(0.5N(N-1)+2MN)以及平方根次数(MN2)㊂最终的复杂度合计为Nite(43MN2+752N(N-1)+148MN)㊂另外,通过改进的矩阵求逆,即采用改进的Strassenᶄs矩阵求逆进一步降低矩阵求逆Vk(VHkVk+σ2I)-1的复杂度,得出了改进后矩阵求逆算法后所需的乘法次数(4ML2+4N2LJ)㊁加法次数(4ML2+4N2LJ-2ML-2N2)以及求逆复杂度(4776)㊂最终的复杂度合计为2N2(8LJ-1)+2ML(8L-1)+4776㊂对于不采用插值算法的情况,每个插值时刻k,由H矩阵到W矩阵,需要进行J(J=4)次SVD和1次MMSE均衡(主要复杂度在于求逆)的计算㊂其中4次SVD需要3574400次计算㊂MMSE均衡需要521112次计算,因此共需要计算复杂度C1=4095512㊂采用Spline插值时,每个插值时刻的复杂度为31744㊂可以计算出每次插值的复杂度收益为C2=3574400+521112-31744=4063768㊂因此采用插值的最终复杂度收益为:ΔC=(C1-C2)ˑKˑRate㊂(30)假设Rate=1/3时,ΔC=524225536㊂可见,插值对于计算复杂度的降低比较明显㊂同样,计算复杂度可以降低为:ΔRC=C1(1-Rate)+C2ˑRateC1㊂(31)当Rate=1/3时,计算复杂度降低为原来的66.93%㊂当Rate=1/10时,计算复杂度降低为原来的90.08%㊂综上,当采用改进的Strassen矩阵求逆算法时,复杂度降低了10.03%㊂进一步采用插值算法后,计算复杂度能够在上述的基础上再降低9.92%(插值比为1/10)㊁33.07%(插值比为1/3)㊂4㊀结论本论文主要解决MIMO场景下的相关矩阵组的低复杂度计算问题,首先利用H矩阵在时间相关性推导了W矩阵的相关性,通过对已有W矩阵的相关性直接插值出部分缺失的W㊂这使得在接收H矩阵时,在求取部分W矩阵后通过相关性重建完整的W矩阵;也避免了一部分H矩阵的存储以及这部分H矩阵计算SVD与求逆获得W矩阵的过程㊂相比SVD与求逆的复杂度,插值的复杂度要低得多㊂此外,采用了一种改进的Strassen矩阵求逆算法来降低求逆过程的复杂度㊂该算法结合了Strassen矩阵求逆的高效性以及Hermite正定阵的共轭对称性特点,结构更简化㊂参考文献[1]㊀RUSEKF,PERSSOND,LAUBK,etal.ScalingUpMIMO:OpportunitiesandChallengeswithVeryLargeArrays[J].IEEESignalProcessingMagazine,2013(30)1:40-60.[2]㊀蔡锁章,杨明,雷英杰.数值计算方法[M].2版.北京:国防工业出版社,2016.[3]㊀许小勇,钟太勇.三次样条插值函数的构造与Matlab实现[J].兵工自动化,2006(11):76-78.[4]㊀陈帅,岳迎春,徐巍,等.小波时间序列对非平稳信号中突变点的辨识与处理[J].测绘科学,2013,38(5):11-12.[5]㊀DEBOORC.APracticalGuidetoSplines[J].AppliedMathematicalSciencesNewYorkSpringer,1978,27(149):157-157.[6]㊀FRITSCHFN,CARLSONRE.MonotonePiecewiseCubicInterpolation[J].SiamJournalonNumericalAnalysis,1980,17(2):238-246.[7]㊀颜志升,郑昱.基于高斯消元的自适应信号处理的实现方法:CN111427014A[P].2020-07-17.[8]㊀STRASSENV.GaussianEliminationisNotOptimal[J].NumerischeMathematik,1969,13(4):354-356.[9]㊀杨忠鹏,林志兴.关于Hermitian矩阵的特征的注记[J].大学数学,2003,19(5):52-53.[10]李瑞,李晓明,董晔.STAP中的协方差矩阵求逆快速算法研究[J].计算机仿真,2011,28(2):25-28.作者简介:㊀㊀施育鑫㊀国防科技大学第六十三研究所博士研究生㊂主要研究方向:通信抗干扰㊁OFDM㊂㊀㊀鲁信金㊀国防科技大学电子科学学院博士研究生㊂主要研究方向:信息论㊁索引调制㊁polar码㊁物理层安全㊁无线通信技术等㊂在各类期刊和会议论文集上发表论文多篇㊂㊀㊀孙艺夫㊀国防科技大学电子科学学院博士研究生㊂主要研究方向:可重构信息超表面㊁通信抗干扰㊁物理层安全㊂㊀㊀雷㊀菁㊀国防科技大学电子科学学院教授,博士生导师㊂主要研究方向:信息论㊁LDPC码㊁物理层安全㊁polar码㊁无线通信等㊂㊀㊀李玉生㊀国防科技大学第六十三研究所正高级工程师,硕士生导师㊂主要研究方向:通信抗干扰㊂。
第十一届全国研究生数学建模竞赛C题无线通信中的快时变信道建模一、背景介绍1.基本模型宽带移动通信传输正在改变着人们的生活,更为快速和准确的传递信息是其基本需求。
据预测,到2020年,数以千亿的“物”,包括汽车、计量表、医疗设备和家电等都将连入移动通信网络,人们的移动数字生活也将更加美好。
由于移动通信网络连接环境复杂多变,对实现高速宽带数据传递提出了更高的要求和挑战。
例如,高速铁路和高速公路的开通和应用,使未来移动通信系统面临高速移动环境,而在高速移动环境下,无线通信信道会发生快速变化,若不能适应这种变化,通信系统性能将会受到严重影响,极大降低信息传输的速度和质量。
分析现有通信模型的不足,建立新的数学模型,对提升信道容量、增加信息传输速率和降低误码率会有很好的促进作用。
在通信系统中,发送端通过信道传输信号到接收端,在传输过程中,不可避免地要引入干扰噪声。
接收端对包含噪声的信号进行合理解码,得到正确的信息,完成信息传输过程,原理用图1表示。
图1 通信基本模型示意图通信过程的数学模型可以表示为:WXHY+⋅=(1)从式(1)可以看出,在已知接收端信号Y的情况下,要得知发送端的信号X,还需要知道信道变量H和噪声W的统计特征。
W可视为加性高斯白噪声AWGN(Additive White Gaussian Noise),因此问题的关键就是对H规律的探索。
在无线信道中,发送和接收之间通常存在多于一条的信号传播路径。
多径的存在是因为发射机和接收机之间建筑物和其他物体的反射、绕射、散射等引起的,其传播特征如图2所示。
图2 无线信道传播特征图中LOS(line of sight)是信号直接到达的传播路径。
可以看出,由于环境的复杂性,信号传播途径也复杂多变,需要对其进行简化和抽象,建立描述、估计信道传播的数学模型。
发端信号传输信道H噪声WYX收端信号当信号在无线信道传播时,多径反射和衰减的变化将使信号经历随机波动。
无线多径传输系统的时间离散形式的数学表达式为[1]:∑-=-=+-=101,...,0],[][][][L l l K n n w l n x n h n y (2)式中L 为信道的多径数,K 为传输信号的长度,)(n w 可视为AWGN ,[]l h n 就是信道参数。
Low Complexity Mode Decision and Motion Estimation for H.264/A VC Based Depth Maps Encoding in Free Viewpoint VideoGianluca Cernigliaro,Fernando Jaureguizar,Juli´a n Cabrera,and Narciso Garc´ıaAbstract—Within free viewpoint video,the3-D reconstruction of the scene is created from a high number of viewpoints.Every viewpoint is represented by a traditional sequence,called texture, and its associated depth information.This is known as a View plus Depth environment.In this paper,a novel low complexity mode decision and motion estimation algorithm for the H.264/A VC based encoding of depth sequences is proposed.Given that a tex-ture sequence and its associated depth represent the same scene from the same point of view,they should have similar motion characteristics.The complexity reduction of the depth encoding, in the proposed algorithm,is obtained by taking advantage of the texture motion information that has been previously processed by a traditional H.264/A VC encoder.The characteristics of depth and texture sequences are analyzed,focusing on similarities and differences that are properly managed to design an algorithm able to detect when the motion information of the texture might be usefully exploited in the encoding of the corresponding depth sequence.The proposed method is able to achieve the same objective quality,measured by means of the PSNR and the VQM, than a traditional H.264/A VC encoder with a reduction of up to 58%of the computational burden.Index Terms—3-D video(3DV),depth maps,free viewpoint video,H.264/A VC,mode decision,motion estimation,multiview plus depth.I.IntroductionF REE VIEWPOINT video(FVV)and3-D video(3DV)[1],[2]represent the next generation of video paradigms whose goal is the involvement of the observer,thanks to the 3-D scene reconstruction,or the depth perception without stereoscopic glasses or other additional devices[3].There are several targetfields for3DV and FVV covering a large number of areas like cinema,home theaters,and video conferences. In all these areas the immersion is provided by representing the scene complying with the depth and by displaying the 3-D world showing a high number of viewpoints to freely change the perspective and to give to the spectator a real-world experience[4].Immersion performance depends on the Manuscript received February3,2012;revised June12,2012;accepted August18,2012.Date of publication October9,2012;date of current version May1,2013.This work was supported in part by the Ministerio de Econom´ia y Competitividad of the Spanish Government under Project TEC2010-20412(Enhanced3DTV).This paper was recommended by Associate Editor P.Salama.The authors are with the Grupo de Tratamiento de Imagenes,Universidad Polit´e cnica de Madrid,Madrid28040,Spain(e-mail:glc@gti.ssr.upm.es; fjn@gti.ssr.upm.es;julian.cabrera@gti.ssr.upm.es;narciso@gti.ssr.upm.es). Color versions of one or more of thefigures in this paper are available online at .Digital Object Identifier10.1109/TCSVT.2012.2223632number of viewpoints that,when high,improves the3-D sensation.However,when increasing the number of cameras to capture the scene,a considerable amount of information must be transmitted or recorded[5].To overcome this disadvantage the number of viewpoints captured should be limited,bringing a data decrease,but also a worsening of the3-D effect. Researchers found the solution to the issue offinding the best tradeoff between the number of viewpoints and the perception of the reality by generating synthetic views located where cameras are not present[6].The virtual view generation needs to be set in a well-defined environment composed of one or more texture sequences and by their corresponding depth sequences.This setting is called the View plus Depth (V+D)environment[7]where the depth information is usually represented as a gray scale video sequence that describes the position of the objects in the scene[8].The virtual sequence is generated by an interpolation of the warped reference textures to the location of the virtual one,using the depth sequences to locate the objects in the3-D space.One of the synthetic view generation methods is called depth-image-based rendering(DIBR)[9].The work presented in this paper focuses on the compression stage of the V+D sequences. Taking into account their new features,the V+D encod-ing should be suitably assessed to obtain improvements in terms of performance or complexity reduction.Focusing on the complexity,due to the presence of the depth,the time needed to encode every view is doubled.Given that a depth sequence represents the same scene captured by the traditional texture camera,some aspects of these two videos could be similar,for example,the motion of the objects.This motion redundancy could be exploited to reduce the computational complexity of the depth encoding.The related work section of this paper(Section II)explains how some methods exploit the analogies between texture and depth and use the texture motion information to reduce the complexity of the mode decision(MD)and the motion estimation(ME)of the depth encoding.However,it may be useful to state the distinctions between the depth video and the associated texture.The depth sequence only provides information about how the objects are far from the camera;therefore,although the objects in the texture and in the depth are the same,a surface parallel to the camera will be represented with the same depth gray level, losing the information related to the texture patterns.Hence some depth characteristics are definitely different from the1051-8215/$31.00c 2012IEEEcorresponding textures and motion information sharing could generate quality losses[10].This paper has a double purpose.Thefirst one is to obtain an analysis of the V+D sequences,focusing on the video characteristics strictly related to the compression and which have a direct impact on the motion evaluation.The target of this paper is to know when the sharing of motion information between texture and depth is able to ensure a good rate distortion(RD)performance[11],or when the video features are different enough to produce a worsening.This paper is centered on the video properties that affect directly the MD and ME stages such as temporal and spatial variances.The second purpose is the introduction of a novel low complexity MD and ME algorithm,based on H.264/A VC[12],which, to the best of the authors’knowledge,is thefirst one able to reduce the depth encoding computational burden obtaining the best quality achievable with a traditional H.264/A VC encoder. The main novelty is the approach used to detect when the texture information is useful to obtain an acceptable prediction for the corresponding depth.The observations made in the analysis are exploited to make the algorithm able to detect when the reuse of the texture motion information over the depth is a reliable solution and when this option is not opti-mum.The target quality of the method is the RD performance obtained by a traditional H.264/A VC encoder.Therefore,the aim is a complexity reduction without losses in terms of quality.The depth compression quality is usually evaluated through objective metrics,e.g.the well-known peak-signal-to-noise-ratio(PSNR),applied directly on the depth data or on the synthetic views.However,due to the rendering algorithms,the synthetic views could be affected by errors or artifacts that are not correctly handled by a fully objective metric,so it might be important to know how is perceived the degradation of the synthetic views,due to the depth compression.The application of subjective experiments requires a huge amount of tests and people involved in the video evaluation,so the use of an objective metric is considered more viable.In this paper we propose to assess the depth encoding quality evaluation on the synthesized view by means of the video quality metric(VQM)[13],given its good correlation with the quality perceived bya human observer.The paper is structured as follows.In Section II the work related to the depth encoding is reviewed,highlighting the novelties of the proposed method.Section III shows the study carried out comparing the characteristics of texture and depth sequences.The architecture of the proposed Low Complexity MD and ME algorithm is explained in Section IV.Section V illustrates the complexity reduction obtained.Results on RD performance and on complexity reduction are presented in Section VI.Finally,Section VII describes the conclusions about the work.II.Related WorkAs explained above,V+D sequences are composed by texture and depth information.Researchers started exploring new approaches in order to take advantage of the newcharac-Fig.1.Encoding time comparison between texture and depth for a set offive V+D sequences(Beergarden,Book Arrival,Cafe,Kendo,and Newspaper). teristics of the V+D sequences or to exploit appropriately the additional information available in this environment.The cur-rent state-of-the-art related to the depth encoding optimization is reviewed in this section.The depth video,due to these specific characteristics,should be differently handled with respect to the texture one,in order to adapt the compression to this specific content.For example,in depth maps,patterns inside objects are lost and edges are the more relevant information because they define the boundaries of the objects in the3-D scene,so the quality of their reconstruction affects the virtual view rendering.The typical block divisions used in standard video coding could not provide optimized performances,because large homogenous areas could be divided into small blocks and,to the contrary, objects boundaries could be encoded in the same block.Mor-van et al.[14]proposed a platelet-based depth map encoder that geometrically adapts the area of the blocks according to the boundaries,refining the prediction in the depth areas that are more relevant in view synthesis.Apart from the differences between texture and depth,the similarities can also be used to optimize the V+D encoding. When a texture video has an associated depth sequence,it is possible to take advantage of the similar characteristics between texture and depth to reduce the complexity of the texture encoding process.So,in the authors’previous work [15],[16],the depth information has been used to guide the block division of the MD stage,reducing the number of candidate modes and,as a consequence,reducing the texture encoding complexity.Another target of researchers has been depth encoding optimization.The depth information is used to represent,with other magnitude,the same scene captured by the correspond-ing texture cameras;therefore,some similarities in terms of motion can be found.Daribo et al.[17]considered the correlation between the motion of the texture and of the depth and they proposed a motion vectors(MVs)sharing algorithm, where the ME is done by a joint motion estimator able to generate common MVs for both texture and depth.Through this method it is possible to improve the RD performance saving bit-stream because,for the encoding of two videos (texture and depth),only the MVs needed to encode one video are used.Seo et al.[18]empirically traced how depth mode selection is different from texture mode selection,finding that,CERNIGLIARO et al.:LOW COMPLEXITY MODE DECISION AND MOTION ESTIMATION771in depth maps,direct modes are used more often than in textures.Due to this difference,a MV sharing between texture and depth is not always reliable,so Seo et al.designed an algorithm able to select the best candidate for depth between intramodes,intermodes,and the mode used in the texture encoding.When the texture motion information provides a good prediction,the bit saving is obtained by sending only one bit to indicate the reuse of the texture MVs.Hewage et al.[19]exploited the correlation between texture and depth se-quences to design a concealment algorithm able improve the performance in case of error in the transmission of the depth information.Another issue is the complexity reduction of the depth encoding process.Focusing on it,time needed for the encoding of the depth sequences has been evaluated and compared to the corresponding textures.It is possible to observe in Fig.1 a comparison between the average encoding time needed to process texture and depth of one view of a set of V+D sequences:the time needed is comparable.Hence for the encoding of V+D sequences,due to the presence of a second video associated to every texture,the doubling of time,or of resources,is required for the encoding.With the complexity reduction purpose,Oh et al.[10]intro-duced the idea of exploiting the similarities,in motion,with the corresponding textures.By extrapolating texture motion,a good approximation of the depth motionfield can be obtained. To know how the motion data of the depth is similar to the texture one,Oh et al.observed the MVs selected by a traditional H.264/A VC encoder to compress both texture and depth.In this analysis,the difference in terms of magnitude between the MVs has been measured:when such difference was low,it was possible to approximate the depth motion with the texture motion information.The results of this analysis were highly dependent on the sequences characteristics,so it was not possible to know a priori when to trust in the texture motionfield.The solution found consisted of taking advantage of the texture modes and MVs,previously selected,to generate the candidates as mode and MVS for the depth.The algorithm selects the best ones by RD optimization.Other researchers considered other features of depth maps. Chiang and Chen[20]exploited depth edges to reduce the complexity of the MD stage.Edges are detected with a Sobel operator and they are used to guide the macro-block(MB) segmentation according to H.264/A VC modes.The suggested number of candidates is less than the modes allowed in H.264/A VC so,although in some sequence RD performance gets worse,the complexity always decreases.Also,the work of Zhu et al.[21]focused on the features of depth motionfields.In this case,depth maps are divided into subregions according to edges or moving objects and,depend-ing on the characteristics of every region,different approaches are used to evaluate the mode selection.A reduced number of modes is considered in every region so,as in the previous cases,the complexity of the MD stage is reduced.The Zhu et al.algorithm obtains a considerable complexity reduction (up to80%),but the quality is highly dependent on the edge detector refinement that,when reduces the complexity,causes quality losses.To the best of the authors’knowledge,these represent the most relevant works related to the depth map coding where the depth features are analyzed and where the correlation between depth and texture is exploited.Some of them have the aim to improve the RD performance,while others have the computational burden reduction purpose.When the goal is the depth encoding computational burden reduction,the RD performance is worsened;therefore,until now,the complexity reduction is payed with a reduced quality of the compressed signal.The algorithm proposed in this paper has the goal of the complexity reduction of the MD and ME stages in depth maps encoding by exploiting the previously evaluated texture motion information with two main novelties.Thefirst novelty is the modus operandi used to obtain an a priori analysis of the features that have a direct impact on MD and ME.In the authors’previous study[22],an analysis of the different features between textures and depth has been carried out.Such analysis was based on a study of the MD and ME stages of the depth encoding process,observing the differences between the best modes used for the encoding of the depth maps and of the texture.This analysis has been useful for the design of afirst low complexity depth maps encoder but it was not definitely conclusive because the complexity has been reduced according to the features that have been examined a posteriori.In this paper,the study focuses on the statistical features of the video that produces such differences. Temporal and spatial variances of texture and depth sequences are observed and compared to understand the reasons why, in some regions,the use the texture motionfield to encode the depth is not appropriate.Through this approach,a better knowledge of the depth sequence characteristics is obtained. The second novelty is the algorithm performance target. The other low complexity methods presented in literature have payed the complexity reduction with RD performance losses. The proposed method is able to detect when the reuse of texture modes and MVs is not adequate because it could cause quality drops.If the texture motionfield reuse is not reliable,other procedures are applied in order to approximate the best RD performance.So the assurance of reaching the best quality achievable by a traditional H.264/A VC-based encoder is guaranteed,reducing the encoding process complexity. Summarizing the performance in terms of distortion,as the target of the algorithm is the best quality obtained by a traditional H.264/A VC encoder,is always better than any other low complexity depth encoding algorithm presented in literature.The complexity is reduced only when it does not produce losses.III.V+D Features AnalysisAs said above,thefirst goal of this paper is to state simi-larities and differences between texture and depth sequences. Such characteristics are analyzed to know which features can be useful for the design of a low complexity depth encoding algorithm.The paper has been made focusing on those statistical features that affect the MD and the ME stages as,for example, the signal variance.772IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY,VOL.23,NO.5,MAY2013Fig.2.Intra modes.(a)Intra4×4.(b)Intra16×16.A.Sequence Features Involved in MD and MEIn video coding,compression performance is closely related to the sequence variations:little changes are easier to predict and easier to manage than bigger ones.Video signal variations have a different nature and they depend on various factors as changes between the values of adjacent pixels in the same frame or changes between different frames pixels values.The image values variations can be analyzed by using probability distribution descriptors as the variance,which shows how far a set of values is spread out from each other.According to the RD theory[23],[24],the distortion[eval-uated through the mean square error(MSE)]of a Gaussian source X,with varianceσ2x,can be written as a function of its variance and of the bit-rate R needed for its representationD x(R)=2−2Rσ2x.(1) Therefore less variance means less distortion of the com-pressed signal for afixed rate,or less rate needed for the signal representation with the same distortion.The MD and ME stages aim to reduce as much as possible the signal variation in order to reduce the distortion or the rate,so signal variances are the features that mostly affect these stages.Depending on which one of the two stages is observed,different variances have to be considered.In this section,the various variances involved in the MD and the ME are observed and highlighted. The MD stage is divided into two main parts:intramode decision and intermode decision.Intramodes are used to spatially compress the macro block (MB).In this case,the predicted blocks are generated by tech-niques based on interpolation,extrapolation,or means of the pixels values of the same frame located around the area to be encoded(Fig.2)[25].When the intracompression is applied, the spatial variance is considered in the process and,when it is reduced,the spatial-based compression is able to represent the signal with a reduced number of bits and lowdistortion.Fig.3.Inter motion estimation.Intermodes are used to compress the MB by dividing it into subsections of different sizes and making the prediction of every subsection through pixels of other frames[25].The region of any other frame that best approximates the area to compress is found in the ME stage(Fig.3)and it is chosen in order to minimize the residual to encode.As this prediction is made through pixels of the other frames,the temporal variance is the main feature affecting this stage.When the temporal variance is reduced,it is possible to obtain a good compression of the frame through the interprediction.As temporal and spatial variance are involved in the MD and ME stages,the analysis is assessed over these features.MD and ME are evaluated independently for every MB of every frame;hence,we consider a MB-wise variance evaluated in different ways depending on the cases.For the spatial variance,pixel value changes,with respect to the adjacent ones,are considered.In this case,the variance is evaluated on the luminance values of a16×16MB as shown inσ2B=E[(B−μB)2](2) where B represents the luminance values of one MB andμB is the average luminance value.CERNIGLIARO et al.:LOW COMPLEXITY MODE DECISION AND MOTION ESTIMATION773Fig.4.(a)Frame#2of thefifth view of Beergarden with its(b)associated depth map.For the temporal variance evaluation,the scene motion has to be considered,so the pixel variations among different frames have to be evaluated.The differences between pixels of two consecutive frames without the application of any motion compensation algorithm are evaluated as inδn=|B n−B n−1|(3) where B n is one MB of the frame n and B n−1is the MB located in the same position in the frame n−1.The temporal variance is evaluated as variance of the differencesδn asσ2δn=E[(δn−μδn)2].(4) In the next subsection the spatial and the temporal variances of a set of V+D sequences,evaluated as stated above,are observed and analyzed in order to obtain a model to follow in the algorithm design.B.Variance of V+D SequencesThe goal of the proposed algorithm is to take advantage of the similarities between texture and depth for the complexity reduction of the depth encoding.The purpose of this paper is to know when a direct copy of the texture modes and MVs provides good performance or when it is preferable to use other strategies.The signal variance affects the video encoder performance;therefore,it is necessary to analyze the texture and depth variances focusing on differences and similarities that are useful for the design of the method.The variances analysis is discussed in the following two subsections:the first one shows a visual analysis of the variance differences between the texture and the depth of one sequence frame;in the second,to make consistent the considerations made in the first one,the analysis is done over a large number of frames for a set of sequences.1)Visual Analysis:The analysis starts by observing one sequence frame and its corresponding depth map.Fig.4shows an example of a V+D frame:a texture frame, with the corresponding depth map,of the sequence Beergar-den.Observing the frames,some analogous characteristics can be perceived:silhouettes and boundaries belong to the same objects,so similar motions should be noticed along the time.Hence it is apparently reasonable to use the modes and MVs of the texture for the depth encoding.However,the low complexity methods based on the sharing of the motion information between texture and depth show how this strategy often provides quality losses.In order to identify in which situations the re-use of the texture information deteriorates the depth compression quality,the statistics features of the sequences have been observed.Fig.5shows the variance of every texture and depth MB for the frame number2of the5th view of the sequence Beergarden.Fig.5(a)and(c)shows the temporal variances of the texture frame and the depth map,respectively.Within the texture,high temporal variance is present in almost the whole area of the moving objects.In the depth,due to the more homogeneous signal,the high temporal variance is perceived on the boundaries.Considering that the objects in the texture and in the depth are the same,the motion information used for the texture compression should well represent the depth motion,except in a situation:depth boundaries that are in motion have a very high temporal variance that is often higher than the texture one, hence the motion prediction made through the texture motion information,in this case,does not ensure good performance. Summarizing,the texture motion information should provide good performance when used for the depth encoding except on the depth boundaries with high temporal variance(depth motion boundaries).Fig.5(b)and(d)shows,respectively,the MB-wise spatial variances of the texture frame and of its depth.The images indicate the main difference between the two types of video representation:in the depth map all the patterns are lost and the spatial variance results are considerably reduced and very different to the texture one.Therefore,the sharing,from texture to depth,of spatial-based prediction modes could make worse the compression performance.Summarizing,according to thisfirst visual analysis on the MB-wise variances,there are two specific situations where the reuse of the texture information does not ensure good compression performances:1)texture compression based on a spatial prediction(toodifferent spatial variances);2)depth boundary in motion(depth temporal variancehigher than the texture one).2)Statistical Analysis:To check if the considerations made analyzing the variance of one sequence frame are conclusive,the variance of a set offive V+D sequences has been evaluated.Both temporal and spatial variances of texture and depth of every sequence have been considered, evaluated as mean of the MB-wise variances of each frame. The comparison has been represented in Figs.6and7. Fig.6shows the spatial variances of texture and depth, confirming that the texture has higher spatial variance due to the presence of patterns.A compression based on the spatial variance,in H.264/A VC,is performed by the intramodes.Due774IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY,VOL.23,NO.5,MAY2013Fig.5.MB-wise spatial and temporal variances of the texture and of the depth frame#2of thefifth view of the sequence Beergarden.(a)Texture temporal variance.(b)Texture spatial variance.(c)Depth map temporal variance.(d)Depth map spatial variance.to mentioned differences,when a spatial based compressionis used in the texture encoding,the performance of using thetexture mode for the depth encoding is unpredictable.Fig.7shows a comparison between the temporal variancesof texture and depth.Also in this case,the variances aredifferent but the curves show similar trends,although thevalues are not the same.This tendency indicates that themotions of the two sequences have similarities that can beexploited.The variance values can be analyzed using the Pearson cor-relation coefficients evaluated between the depth and texturevariancesρX,Y=E[(X−μX)(Y−μY)]σXσY(5)where the variables X and Y represent the two signals on which the coefficients are evaluated.For the evaluation of the the spatial variance correlation coefficients,the following is used:ρVS,T ,V S,D=E[(V S,T−μV S,T)(V S,D−μV S,D)]σVS,TσVS,D(6)where V S,T and V S,D represent the spatial variances of texture and depth,respectively.For the evaluation of the temporal variances,correlation coefficients are usedρVT,T ,V T,D=E[(V T,T−μV T,T)(V T,D−μV T,D)]σVT,TσVT,D(7)where V T,T and V T,D represent the temporal variances of texture and depth,respectively.The results of the coefficient evaluation are shown in Table I.Almost all the correlation coefficients evaluated on the temporal variance are high enough to affirm that the trends are similar.On the contrary,the coefficients evaluated on the spatial variances are lower,confirming the poor spatialTABLE IPearson Correlation Coefficients Evaluated on the Temporal and Spatial Variances Between Texture and DepthSequenceρVS,T,V S,DρVT,T,V T,DBeergarden0.10.85Book-Arrival0.670.91Cafe0.70.36Kendo-0.210.86Newspaper0.240.94TABLE IIExample Pearson Correlation Coefficients Evaluated on the Temporal and Spatial Variances Between Texture and Depth for Every Second of the Cafe SequenceCafe sequence secondρVarS,T,Var S,DρVarT,T,Var T,D1st0.950.942nd0.190.83th0.70.934th0.480.02variance similarities between texture and depth.The corre-lation coefficients evaluated on the sequence Cafe do not follow the trend of the other ones.In this case,the spatial variance correlation is higher than the temporal one.To check if the temporal correlation is actually so poor,the correlation coefficients have been evaluated dividing the sequence into four segments corresponding to every second(Table II). The results show how in thefirst three seconds the corre-lation of the temporal variances is high and how,in the last one,the coefficient is almost zero.Some depth artifacts can be noticed in this segment of the sequence and they have been considered as the cause of the low correlation. Concluding,the similar trends noticed for the temporal variance curves show how the re-use of the texture motion information could provide good results in the depth encoding. The differences between the curves are due to what noticed。