Covering live news events via traditional SNG technologies has been an expensive and logistically difficult business. However, with the advent of cellular bonding technology, broadcasters and online video professionals can easily and cost-effectively report live from any location using a field-proven method.
Bonding combines multiple cellular and wireless networks — such as 2.5G, 3G, 4G LTE, Wi-Fi and WiMAX — to obtain a reliable, robust video uplink capable of transmitting HD video.
The bonding challenge
A high-quality video experience relies on smooth and uninterrupted video delivery, but cellular links are inherently unstable and fluctuate continuously. Transmitting video over such a link may result in black screens, video breaks, pixelization, jitters, audio problems, lost lip syncs etc., even from a stationary location over 4G. Parameters that affect the experience, such as bandwidth availability, latency, loss rate or all of them together, can change in a millisecond.
Therefore, there's an inherent gap between the desired experience of high-quality video and cellular technology. Cellular bonding has proven to bridge this gap.
Bonding architecture
Cellular bonding takes compressed video (H.264) and transmits some of the packets over each of the multiple cellular modems. Instead of relying on a single unreliable link with a single point of failure, bonding several cellular links together minimizes the inherent risks while achieving the desired or greater performance.
Bonding systems differ according to how low they can reduce the risks and to what degree they can increase the performance. To achieve this, the system has to continuously, and in real time, monitor all available links and understand how best it can use each of them — currently and in the immediate future. It also has to dynamically adapt the video encoder according to the momentary total available bandwidth in all the links; compensate for, and recover from, any losses; and interact smartly with the operator to best manage its needs. On the receive side, which can be anywhere in the world, software installed on any Internet-connected PC receives those multiple packet streams and reconstructs the video.
Building blocks
- The monitoring of all links in millisecond resolution (traffic analyzer): Parameters per channel include: current bandwidth, latency, loss and jittery behavior. Monitoring is best done transparently without using nonpayload packets.
- Modems/links usage: Selection at any moment of which modems to use, at what loading and what type of packets to send on each modem.
- Priority scheduling for optimal bandwidth use: Assessing which information has higher momentary priority so it can be transmitted first or on better-suited modems. Prioritized packet types may include: I-Frame, P/B frames, audio frames, FEC packets or management packets.
- Modem prediction: The ability to anticipate the modems' behavior and proactively change transmission parameters so as to minimize fluctuations and potential link failures.
- Multiple modems at any one time: Keeping as many modems in operation as possible, so that if a problem arises with a few modems, it can be mitigated by using the others.
- R (reliable) UDP or at least UDP-based transport: As cellular networks may be too slow, or drop to being too slow for TCP traffic, effective bonding systems use (proprietary) RUDP. This ensures that the packets arrive while avoiding TCP mechanisms that are not built for cellular networks.
- Buffers: Buffers are used to reorder packets with different latencies, enable packets traveling on longer delayed links to be used and provide some time margins for sudden link fluctuations. The user can usually decide whether to aim for interview mode or best quality mode. Some systems achieve satellite-like subsecond latency with bonding, using several lowest-delay modems rather than a single low-delay modem, which might experience a sudden increase in latency.
- Error/loss handling mechanisms: Since cellular networks have mechanisms to “fix” erroneous bytes, it's more accurate to refer to “loss” rather than “error” packets. Continuous loss rates range from 2 percent to 40 percent. Application-level mechanisms are required to compensate for such losses, e.g. “retry” for resending packets, dynamic FEC for sending extra packets allowing the receiver or video decoder “error concealment” functionality.
To optimize network use, the software should automatically apply the best mechanism with the right parameters under dynamically changing link conditions, such as adapting the FEC payload/extra packet ratio, retry timing, etc.
H.264 video encoder
Another building block is the H.264 encoder, which may or may not be integrated and coupled with the bonding application. There are clear advantages when the encoder is coupled with the application: The software has full control over the encoder, enabling an immediate response. For example, the encoder will respond in real time to quick and slow changes of the overall bandwidth.
More advanced encoders support H.264 High Profile. Although scalable video codec (SVC) has made its debut, it's still in its initial stages, and its maturity level is a concern. Furthermore, a bandwidth overhead is introduced.
There are two types of encoders: VBR and CBR. VBR encoders are intended to self-detect and react to the changes in total bandwidth. However, detection time and reaction time may be too long, resulting in lower than optimal performance (e.g. black screens, pixelization etc.). In addition, the software, which manages the transmission, can only react to the unforeseen encoder bandwidth changes and therefore cannot correlate optimization of other parameters such as scheduling or FEC.
CBR encoders should output a target bandwidth. This allows the software to control both the precise bandwidth and the other transmission parameters. In real time, CBR encoders can proactively be set to lower than the currently measured bandwidth to be prepared for any anticipated bandwidth drops. However, some CBR encoders are either too slow to react and/or not very accurate or consistent with their output bandwidth, usually outputting more than configured, and should therefore be checked.
Encoder configuration changes should have a minimal response time (milliseconds) and shouldn't require an encoder reset, enabling the software to respond in real time to dynamic fluctuations. This also involves the right GOP configuration considering other implications.
Other considerations
- Compliance with radiation standards: A safe design and relevant standard certifications ensure that a system easily meets the U.S. and EU cellular health-related standards (i.e. SAR - Specific Absorption Rate).
- Antennas: Antennas must be designed to improve performance and reduce radiation, rather than merely serving a decorative purpose. Otherwise, each RF splitter/combiner, connector and cable can actually decrease performance.
Impact of 4G LTE networks
Although 4G LTE brings the promise of higher peak bandwidth and shorter delays to bonded cellular video transmission, single-modem video delivery devices are still inherently unstable and may well experience performance fluctuations, loss of transmission and the inability to go live. This is usually due to network overload, as operators start to promote networks more intensely, advertised vs. actual performance and slower 4G roll-out outside city centers.
Summary
The goal of cellular bonding is to find an optimal temporary equilibrium point over multiple channels, changing from one point to another as smoothly as possible, to deliver sustained high-quality video performance over inherently unstable heterogeneous links.
It's always prudent to test the system in live situations to check if it meets current and future needs — for example, underground (basements/parking lots), on the move (walking in the city, in moving vehicles, through tunnels), in the air (helicopters, planes) or on special location (crowded stadiums or courthouses).
Cellular bonding changes the landscape of the broadcast contribution market. By having such a crosslayer technology from the physical layer up to the application layer, it is now possible to transmit full HD video with a very low delay, even while on the move.
Dr. Rony Ohayon is CTO of LiveU, and Baruch Altman is a director of LiveU.
