Cui Huijuan

Adaptive quantization scheme for very low bit rate video coding

This paper presents an efficient two-layer adaptive quantization scheme, which uses Q-N (Quantization step-Number of output bits) mapping tables and fuzzy perceptual classifiers. With this adaptive quantization scheme, the consistent coding quality and good visual effect can be kept under constant bit rate.

Voice Activity Detection in Non-stationary Noise

Existing voice activity detection algorithms degraded severely in low SNR or in non-stationary noise environment. This paper proposes a new fusion method in such environment. The method is based on the fusion of the SNR for selected sub-bands of the input speech and this fusion is implemented through a specific function called SAF (sum of activation function). The results show that the algorithm could give reliable voice activity detection result in low SNR and even in the presence of non-stationary noise

Switch-Type Hybrid Hard Decision Decoding Algorithms for Regular Low-Density Parity-Check Codes

This correspondence presents the switch strategies for switch-type hybrid hard decision decoding algorithms for regular low-density parity-check (LDPC) codes. After the piecewise analysis of extrinsic information transfer functions for Gallager decoding algorithm B (GB), the normalized switch scheme (NSS), for which the majority-based algorithms and GB are two examples, is proposed. Then, several other examples of NSSs are presented and their convergence properties are analyzed based on the extrinsic information transfer (EXIT) functions. In simulations, the proposed NSSs show meaningful performance improvements and less sensitivity to channel parameter underestimations compared with GB for small and moderate block length codes.

Voiced/Unvoiced classification recovery in the speech decoder based on GMM

Voiced/Unvoiced (V/U) classification is an important parameter in low bit-rate speech coding algorithms. An algorithm that recovers the V/U classification from the linear prediction coding (LPC) coefficients and the gain in the speech decoder is proposed. Two Gaussian mixture models (GMM) are employed to model the joint probability of these parameters and to perform the V/U estimation. Experiments show the performance improvements of the proposed algorithm over the V/U classifier used in mixed excitation LPC vocoder (MELP). The proposed algorithm operates only at the receiving end and saves all the bits originally used for V/U quantization.

Robust Video Transmission Scheme Using Dynamic Rate Selection LDPC and RS codes

Compressed video is very vulnerable to bit errors or erasure errors introduced during transmission. Thus FEC is necessary to achieve robust video communication. In this paper, we propose a hybrid dynamic rate selection FEC scheme to utilize low density parity check (LDPC) codes and Reed-Solomon (RS) code to achieve robust video communication. We use the powerful LDPC error correcting code to combat bit-errors and RS code to combat erasure errors. The code rates of the LDPC code and RS code are dynamically adjusted according to the channel bit-error rate and erasure error rate. But the overall code rate in the transmission is fixed. We compare the considered schemes in terms of video quality using the new H.264/AVC video standard. Simulation results show that with this hybrid FEC scheme, robust video transmission can be achieved and the video quality is greatly improved compared with the constant code rate FEC scheme

Superior step-size theorem and its application

With independence assumption, this paper proposes and proves the superior step-size theorem on least mean square (LMS) algorithm, from the view of minimizing mean squared error (MSE). Following the theorem we construct a parallel variable step-size LMS filters algorithm. The theoretical model of the proposed algorithm is analyzed in detail. Simulations show the proposed theoretical model is quite close to the optimal variable step-size LMS (OVS-LMS) model. The experimental learning curves of the proposed algorithm also show the fastest convergence and fine tracking performance. The proposed algorithm is therefore a good realization of the OVS-LMS model.

Implement of system control part in a multimedia communication terminal

This paper presents how to realize the system control part of a multimedia communication terminal on a LAN. The system control part is the core of the terminal. The design partly refers to ITU-T H.323. System control is composed of call control and H.245 control. Call control signals are used for call establishment, disconnect and other call control functions. H.225.0 defines the syntax of call-signal messages. We use C language to achieve encode and decode of those messages. TCP/IP network protocols serve as a friendly interface to the LAN. We establish a TCP channel (at 1720 port) to transfer those messages to finish the call control process. Through call control negotiation, we establish another TCP channel for H.245 control. Master/slave determination, capability exchange, logical channel signalling and close logical channel procedures are finished at the H.245 control channel through a message exchange method. Abstract syntax notation one (ASN.1) serves as a structured data specification language. It is widely used in H.225.0 and H.245 protocols. It is very important to finish the ASN.1 language to C language translation. A compiler translates the ASN.1 notations into C language sources, which function as the encoder and decoder for the ASN.1 notations according to X.691 (packed encoding rules or PER). Other parts of the terminal are H.263 video codec, G.723.1 audio codec and RTP (real-time protocol). The terminal can run perfectly point-point multimedia communication on the LAN. It can also interoperate with Microsoft NetMeeting for video and audio communication.

Chessboard-interpolation-based multiple description video coding

To enhance the robustness of video transmission over noisy channels, this paper presents a multiple description video coding algorithm based on chessboard-interpolation. In the algorithm, the input image is decomposed according to the chessboard pattern, and then interpolated to produce two approximate images with the same resolution. Consequently, the state-of-the-art DCT+MC (Discrete Cosine Transform + Motion Compensation) video codec is independently applied to the two approximate images to generate two descriptions of the original image. In this framework, a fairely good reconstructed image quality is obtained when two descriptions are received simultaneously, while an acceptable reconstructed image quality could be yielded if only one description is available. Moreover, the mismatch between the encoder and the decoder could be effectively controlled through partial coding of the difference signal between two descriptions. In bidirectional video communications, a drift control scheme is further proposed, in which the error drift could be eliminated after the encoder imitating the error concealment actions of the decoder. Since the inherent correlation among adjacent blocks of DCT+MC video coding is efficiently exploited, this algorithm has a better redundancy-rate-distortion (RRD) performance than other multiple description algorithms. Simulation results show that the proposed algorithm is fairly robust while preserves a high compression rate. A more constant reconstructed image quality is achieved over extremely noisy channels, compared with traditional single description coding. In addition, it is observed that the mismatch and the error drift are effectively controlled.

Optimal variable step-size LMS model and algorithm with independence assumption

To solve the contradiction between convergence rate and steady-state error in least mean square (LMS) algorithm, basing on independence assumption, this paper proposes and proves the optimal step-size theorem from the view of minimizing mean squared error (MSE). The theorem reveals the one-to-one mapping between the optimal step-size and MSE. Following the theorem, optimal variable step-size LMS (OVS-LMS) model, describing the theoretical bound of the convergence rate of LMS algorithm, is constructed. Then we discuss the selection of initial optimal step-size and updating of optimal step-size at the time of unknown system changing. At last an optimal step-size LMS algorithm is proposed and tested in various environments. Simulation results show the proposed algorithm is very close to the theoretical bound.

Normalized Switch Schemes for Low Density Parity Check Codes

This paper addresses switch schemes for switch type hybrid hard decision iterative algorithms decoding finite length low density parity check (LDPC) codes. After piecewise analysis for extrinsic transfer information (EXIT) chart family and exploration of fixed points, one general switch scheme named normalized switch scheme, for which majority based (MB) algorithms and Gallagers decoding algorithm B (GB) are extreme cases, is proposed. Several specific normalized switch scheme, named expanded optimal switch scheme and tail-cut (TC) optimal switch scheme, are given and their convergence properties are analyzed. Simulation shows that compared with GB that holds optimality for infinity length codes, proposed switch schemes show performance improvements while enjoying less sensitivity to channel parameter underestimations for finite length codes