Basic information

Title: Prosuming Live Media Content in 5G-enabled Smart Cities
Acronym: Prosumer Framework
Company: TNO
Type: Non profit research organization

Where: Bristol
When (time-plan): November 2018 to December 2020

Description

TNO’s Prosumer framework aims to leverage the FLAME infrastructure to create a localized crowd-sourced multimedia production and distribution workflow to enable today’s mobile producers of multimedia content to serve their content in real-time to other users, who may be both located nearby or remotely, while at the same time drastically reducing the bandwidth utilization, both in uplink and downlink, as compared to traditional crowd-sourced multimedia production and distribution services. This will enable regular users in a city to share their experiences in real-time with others while not burdening the internet backend.

Figure 1 – The Prosumer App
Figure 1 – The Prosumer App

Our framework’s front-end, the Prosumer App, is an Android application where end users can enjoy a series of functionalities such as stream discovery, live stream upload and live stream playback. Our framework’s back-end responds to fundamental needs of content providers and network providers. Specifically, for content providers it is essential to provide end users with the ability of finding relevant and interesting content, especially with thousands of streams uploaded constantly to large media platforms. With this in mind, we have included image processing components to help tag content and limit sub-par content. For network providers, it is important to limit bandwidth utilisation in the Internet / network backbone. To address this need, our framework implements a “local breakout”, where content requested locally to its production is directly served by the nearby edge nodes without the need for it to travel back and forth to / from the broadcaster backend.

Scope/objective of the trial

The Prosumer framework is designed to exploit the edge processing capabilities of the FLAME platform, and is based on the idea that the actuality of the content is more important for consumers who are near the action or a place of interest compared to consumers who are further away and not in view of it. In other words, the closer the consumers are to the producers, the lower the latency needs to be.

Figure 2 - High level architecture for the Prosumer framework
Figure 2 – High level architecture for the Prosumer framework

Figure 2 shows the ingest and delivery flows in the Prosumer framework in a scenario with 4 users. Client 1 is producing, and sends an RTMP stream to the edge node located in close proximity of the access point that the client is connected to (Edge 2). The Edge 2 node forwards the RTMP stream to another node, Edge 1, which has been selected as central processing node for our Prosumer framework. In Edge node 1, the RTMP stream is converted into a DASH stream, which is better suited for efficient video delivery because DASH segments are cacheable in the other edge nodes, thereby reducing bandwidth utilization and reducing the distance between the client and the content. A client wishing to watch the stream produced by client 1 will receive it in RTMP format if it is connected to the same access point as client 1, otherwise it will receive it in DASH format. Since the RTMP stream does not need processing, a client being delivered the RTMP stream will experience lower delay (around 3 seconds) than a client being delivered a DASH stream (around 10 seconds), which has gone through a segmentation process.

In each node an image detection module is running, which filters bad quality (e.g. blurred) content and tags content in realtime. The tags help users navigate and browse through the content while also facilitating the job of the producer (who no longer needs to manually tag their content). At the same time, filtering helps reduce the bandwidth utilisation even more, since unsuitable or bad quality content is not further distributed thorough the network.

Results

The experiment approach consisted of two steps:

  • Technical trial 28th/29th August: performed in Bristol with no participants. This trial was targeted at verifying all the technical features of the platform and the workflow, such as scale in terms of supported users and performance in terms of bandwidth and latency
  • User trial 5th December: performed in Bristol during the Festive Market event in December, with 14 participants divided into 2 trial periods. This trial was targeted at verifying whether the users perceive value from using our framework deployed on the FLAME platform
Figure 3 - Left: technical trial; Right: user trial
Figure 3 – Left: technical trial; Right: user trial

The user experience of the Prosumer framework was rated positively by the participants: 79% of the participants declared that they enjoyed using the app very much. The user evaluation further indicated that a sufficient amount of users (45%) enjoyed producing content, an important prerequisite for prosuming apps to thrive.

A total of 81 streams were recorded, with a mean duration of 102.6 seconds. During the first run, participants recorded more videos compared to the second run (65 vs 16), but of a shorter duration (64.9s vs 255.7s). From the total of 81 streams, three streams attracted a high number of distant views (14, 17, and 12 views) which were also of long duration (525, 427, and 995 seconds). This viewing pattern, combined with the edge capabilities of the FLAME platform and the smart local ingest and distribution of the Prosumer framework, made it possible to realise a downlink bandwidth saving of 93%, as compared to a traditional Prosumer system with cloud-based ingest and distribution.


To know more about Prosumer Framework experiment, read also their final blog post.