Complete Communications Engineering

Congestion control, or rate control, is a crucial factor in the success of networks. By intelligently limiting the transmission rate at each video source, the overall throughput of a network can be increased. Though transmission control protocol (TCP) is a very well known solution to this problem and is used in many applications, it is not suitable for wireless multimedia sensor networks (WMSNs) for two important reasons. In this paper we specifically look at WMSNs which are self contained and not connected to the general TCP/IP internet.First, the transmission window in TCP is constantly changing due to the additive increase/ multiplicative decrease (AIMD) nature of TCP. Though this is important to take advantage of available bandwidth, the constant fluctuation of quality in a variable rate video transmission will cause the perceived quality (by the end viewer) to actually be lower than if the window would have just stayed constant at the mean value. This can be solved by implementing an equation based rate controller (such as ARC or TFRC) which solves an algebraic equation modeling the average rate of TCP. Since we are looking at self contained networks, we do not have to take fairness to TCP into account and can instead design an equation which will maximize the overall video quality (i.e. sum PSNR) over multiple video transmissions. This solution could be extended to general networks and work alongside TCP by restricting the equation based rate to what TCP would give for a given delay and packet loss rate, thereby enforcing TCP fairness.

Second, TCP is a protocol which attempts to control the transmission rate across each node. In terms of fairness, video quality is not equivalent to video rate across different videos. For example, take the well known videos Foreman and Akiyo. For roughly the same quality (around 45 dB PSNR), Foreman is 42% larger than Akiyo. This means that a rate controller that uses the transmission rate of these two videos as the only indication of fairness will result in a user watching Akiyo seeing a better quality video than a user watching Foreman.

The solution to this is to use the estimated distortion at the receiver along with the anticipated (or measured) delay in the rate control decision directly. A simple solution to this problem is to use the (MPEG) encoded video compression rate as a weight in the rate control function. For example, assume that there is only a single video transmission in a network. Since there is no congestion, that video will converge to the highest quality which can be supported by the network topology. If a second user joins the network, the first video will decrease its quality (and therefore its bit rate) while the new user increases its quality.

The amount of rate increase or decrease is directly proportional to the current quality of the video. Therefore, given a similar increase in availability, a video transmitting at a very poor quality will increase faster than a video transmitting at a high quality. This forces fairness in terms of video quality.