As video calls and personal broadcasting become popular, the demand for mobile live streaming over cellular uplink channels is growing fast. However, current live streaming solutions are known to suffer from frequent uplink throughput fluctuations causing unnecessary video stalls and quality drops. As a remedy to this problem, we propose PERCEIVE, a deep learning-based uplink throughput prediction framework. PERCEIVE exploits a 2-stage LSTM (Long Short Term Memory) design and makes throughput predictions for the next 100ms. Our extensive evaluations show that PERCEIVE, trained with LTE network traces from three major operators in the U.S., achieves high accuracy in the uplink throughput prediction with only 7.67% mean absolute error and outperforms existing prediction techniques. We integrate PERCEIVE with WebRTC, a popular video streaming platform from Google, as a rate adaptation module. Our implementation on the Android phone demonstrates that it can improve PSNR by up to 6dB (4x) over the default WebRTC while providing less streaming latency.

A majority of American households with broadband today subscribe to a cable broadband service. The technology underpinning this cable broadband service is DOCSIS (Data-Over-Cable Service Interface Specifications). The latest version of this technology, DOCSIS 3.1, enables higher modem speeds compared to the previous DOCSIS 3.0 technology. DOCSIS 3.1 provides faster speeds by increasing the modulation order, and by using Orthogonal Frequency-Division Multiplexing (OFDM) to enable the use of such higher modulation orders. DOCSIS 3.1 allows different modems to use different modulations on different parts of the spectrum. Due to the complexity of selecting and configuring these varying modulation levels across the spectrum, most operators use the same modulation level across the entire spectrum. They configure multiple such profiles on a channel, so that different modems can use the most appropriate profile for them. This enable modems on a cleaner part of the plant to run higher modulations. However, the problem with such an approach is that noise in one part of the spectrum might reduce the entire channel's modulation to limit bit errors. A more sophisticated approach is to use profiles that dynamically vary the modulation on different sub-carriers to best match the noise conditions experienced in the plant. We leverage a Machine Learning based dynamic profile management approach and tested it on data from a top 10 US Cable Company and we observe more than 4% broadband bandwidth gains on average, with some consumers in noisy plant conditions seeing over 30% increase in their bandwidth speeds. And the gains are far more dramatic when compared with using lower modulation profiles (similar to DOCSIS 3.0), with average gain itself being over 37%.

This paper investigates how video-quality information can be exploited by HTTP-based adaptive streaming clients in their rate adaptation schemes. We also seek to answer the question: How much network coordination is required to achieve quality fairness across multiple competing clients? To that end, we envision a loosely coupled architecture where a lightweight centralized coordinator complements multiple federated quality-aware clients. Instead of actively programming network bandwidth allocation to each client, in our proposed architecture, the central coordinator simply collects quality and buffer level information from individual clients and publishes the aggregate quality statistics as reference. Such global reference statistics help to guide each federated client to self-tune its aggressiveness in its quality-aware adaptation scheme. Testbed-based evaluations show that such a lightweight approach is effective in reaping most of the performance gains previously achieved by a centralized joint optimization scheme. Compared to both state-of-the-art rate-based clients and independent quality-aware clients, our proposed federated quality-aware scheme can save up to 50% of total bandwidth while maintaining comparable aggregate video quality across all clients.

This paper explores how video quality information can be exploited by HTTP-based adaptive streaming (HAS) clients in their rate adaptation schemes. We first present an idealized combinatorial optimization formulation of the quality-aware rate adaptation problem and its corresponding solution. A few design principles are then introduced to arrive at a practical and robust rate adaptation scheme in the presence of real-world uncertainties (e.g., inaccuracy in bandwidth estimation, unforeseen changes in future video contents). Results from testbed-based experiments verify that by exploiting video quality information, our proposed quality-aware rate adaptation scheme significantly outperforms conventional rate-based clients in various contention-free and multi-client contention scenarios.

Rapidly increasing volume of video streaming traffic creates a bandwidth crunch for service providers and network operators. Increased user expectations for higher quality video does not imply their willingness to pay higher monthly fees. Hence, more efficient bandwidth management schemes are needed to bridge the gap between growing demand from video traffic and existing network infrastructure. In this work, we present a novel bandwidth management solution for optimizing overall quality of experience (QoE) of multiple video streaming sessions. Instead of allocating bandwidth equally among competing flows, we propose to tailor the bandwidth allocation to both content complexity of requested video and playout buffer status of individual clients. We formulate the multi client bandwidth allocation problem within the convex optimization framework, which is flexible enough to accommodate a wide variety of video quality metrics. Further, we present a practical architecture based on software defined networking (SDN) with two components: video quality monitoring and video quality optimization. Testbed-based experiments confirm that with quality-optimized allocation the network can support up to 75% more users at the same level of quality-of-experience (QoE) than conventional equal-rate allocations.

Many image compression techniques require the quantization of multiple vector sources with significantly different distributions. With vector quantization (VQ), these sources are optimally quantized using separate codebooks, which may collectively require an enormous memory space. Since storage is limited in most applications, a convenient way to gracefully trade between performance and storage is needed. Earlier work addressed this problem by clustering the multiple sources into a small number of source groups, where each group shares a codebook. We propose a new solution based on a size-limited universal codebook that can be viewed as the union of overlapping source codebooks. This framework allows each source codebook to consist of any desired subset of the universal code vectors and provides greater design flexibility which improves the storage-constrained performance. A key feature of this approach is that no two sources need be encoded at the same rate. An additional advantage of the proposed method is its close relation to universal, adaptive, finite-state and classified quantization. Necessary conditions for optimality of the universal codebook and the extracted source codebooks are derived. An iterative design algorithm is introduced to obtain a solution satisfying these conditions. Possible applications of the proposed technique are enumerated, and its effectiveness is illustrated for coding of images using finite-state vector quantization, multistage vector quantization, and tree-structured vector quantization.

Many compression applications consist of compressing multiple sources with significantly different distributions. In the context of vector quantization (VQ) these sources are typically quantized using separate codebooks. Since memory is limited in most applications, a convenient way to gracefully trade between performance and storage is needed. Earlier work addressed this problem by clustering the multiple sources into a small number of source groups, where each group shares a codebook. As a natural generalization, we propose the design of a size-limited universal codebook consisting of the union of overlapping source codebooks. This framework allows each source codebook to consist of any desired subset of the universal codevectors and provides greater design flexibility which improves the storage-constrained performance. Further advantages of the proposed approach include the fact that no two sources need be encoded at the same rate, and the close relation to universal, adaptive, and classified quantization. Necessary conditions for optimality of the universal codebook and the extracted source codebooks are derived. An iterative descent algorithm is introduced to impose these conditions on the resulting quantizer. Possible applications of the proposed technique are enumerated and its effectiveness is illustrated for coding of images using finite-state vector quantization.