Audio codecs are typically transform-domain based and efficiently code stationary audio signals, but they struggle with speech and signals containing dense transient events such as applause. Specifically, with these two classes of signals as examples, we demonstrate a technique for restoring audio from coding noise based on generative adversarial networks (GAN). A primary advantage of the proposed GAN-based coded audio enhancer is that the method operates end-to-end directly on decoded audio samples, eliminating the need to design any manually-crafted frontend. Furthermore, the enhancement approach described in this paper can improve the sound quality of low-bit rate coded audio without any modifications to the existent standard-compliant encoders. Subjective tests illustrate that the proposed enhancer improves the quality of speech and difficult to code applause excerpts significantly.

The online property of fountain codes enables the encoder to efficiently find the optimal encoding strategy that minimizes the encoding overhead based on the instantaneous decoding state. Therefore, the receiver is able to optimally recover data from losses that differ significantly from the initial expectation. In this paper, we propose a framework to analyze the relationship between overhead and the number of recovered source symbols for online fountain codes based on random graph theory. Motivated by the analysis, we propose improved online fountain codes (IOFCs) by introducing a designated selection of source symbols. Theoretical analysis shows that IOFC has lower overhead compared with the conventional online fountain codes. We verify the proposed analysis via simulation results and demonstrate the tradeoff between full recovery and intermediate performance in comparison to other online fountain codes.

In legacy long term evolution (LTE) networks, multiple transmission modes are defined to cater to diverse wireless environment and improve the spectrum utilization. However, constrained by user equipment (UE) processing capability on blind detection of downlink control information (DCI), two transmission modes are allowed to be configured to UE simultaneously. In recent 5G standardization, the polar codes have supplanted the tail biting convolution codes (TBCC), becoming the channel coding scheme for downlink control information (DCI). Motivated by its successive decoding property, a novel design of DCI encoding and decoding is proposed in this paper. The proposed scheme could support dynamic configuration of transmission modes with decreasing the complexity of blind detection. Evaluation results from link level simulations show that the performance loss compared to conventional encoding/decoding scheme is generally negligible and the proposed scheme can comply with the false alarm rate (FAR) target of 5G standardization.

In this paper, we investigate a rateless code design for a two-time slot two-way relay channel (TWRC), where network coding operation is employed at the relay. Rateless codes are used at both users and the relay nodes as they can cope with various channel conditions. We consider a general TWRC framework, where the two users may use different rateless code generator matrices, thus, the relay cannot directly recover the physical-layer network coded (PNC) message of the two users. We propose a hybrid estimate and forward (EF)/decode and forward (DF) protocol with rateless codes for the TWRC. In the proposed scheme, the relay uses a joint belief propagation decoding process to decode the messages of two users. Depending on whether the relay recovers both users' messages, only one user's message or no users' messages, the relay will generate three different types of signals, namely, NC, soft NC, and soft PNC, respectively. We design the degree distributions of the rateless codes for the three nodes of the TWRC, to make them suitable not only for decoding at two users at high signal-to-noise ratio (SNR), but also for decoding at the relay at low SNR. We show that the proposed hybrid EF/DF protocol with the designed rateless codes outperforms other schemes significantly in terms of bit error rate performance.