Paper Review: Understanding Packet Delivery Performance In Dense Wireless Sensor Networks

J. Zhao and R. Govindan, November 2003

This paper studies the volatility and indeterminacy in packet delivery of wireless sensor network. They measured the link quality in three different environments: indoor office, habitat and open parking lot with a medium-scale development of 60 Mica motes.

Their key observation is the existence of "gray area", which "in some environments, is almost a third of the communication range". In their indoor experiment, half of the links experienced more than 10% packet loss, and a third more than 30%. To make thing more complicate, they observed that the gray area changes significantly over time.

Their second observation is that, there's no easy way to measure or estimate poor quality links. At MAC layer lever, they investigate on five factors that may affect packet transmission, they asked the following questions: 1) how distance affects packet reception, 2) whether signal strength could be used to estimate link quality, 3) whether sophisticated coding schemes will help to mask gray area 4) is there a correlation between adjacent node 5) how does packet delivery vary with time. Their answer to these questions, however, does not give optimistic result to these questions.

The last observation in their study is the asymmetric communication between nodes. At the medium access layer, they found that more than 10% of the links exhibits asymmetric packet loss in their indoor experiment. Another disappointing finding in their experiment is that, 50% to 80% of the communication energy is wasted on repairing lost transmissions.

Maybe not too bad

In my opinion, the negative impact of gray area is more depends on the application of a particular system rather than a critical problem of wireless sensor network in general. For example, gray area problem is more complicated in mobile networks than in static networks. According to different usage and implementation, we could minimize the impact of gray area by properly deploy the sensors to its suitable location or use a router algorithm that will snoop on reliable communication paths.

Snooping on surrounding traffic probably is the best solution from a protocol designer's aspect, but maybe simpler solution will also help to solve the problem. We have a similar problem when installing Wi-Fi access points inside a building. The solution for reducing interference and get best coverage is to carefully selection the installation location. I think we can use the same idea when deploying sensor nodes. Instead of placing nodes equally after a certain distance, we might want to place them inside the range where they can get a good signal.

Except for data-intensive wireless network, we can afford packet loss to some point. For many applications such as forest fire sensing and agriculture monitoring, there's not much data needs to be transmitted and doesn't need to have a persistent connection of the entire network. Particularly, given that we could nicely synchronize each node, I guess we could use some kind of smart scheduling method to reduce interference between different nodes.

What could affect the packet delivery?

This paper gives a good description about the relationship between different factors and packet delivery performance in their experiment environment. But it doesn't provide an explanation of what caused those phenomenons. The paper gives some hypothesis such as multi-path, frequency spectrum under use and difference between devices. But they did not verify them, neither provide a systematic investigate strategy.

I think we could start with investigating the following aspects:

1. What is the pattern of signal attenuation according to distance? We have learned from this paper that it is not linearly decreased, but is there other mathematic model that could simulate this decrease?
   
2. What could we do about the asymmetry links? One possible solution I read from Sigcomm poster suggests that use pairwise-reachabilities to infer the entire network graph at individual nodes. By reachability they mean the end-to-end, multipath probability of a packet to reach its destination. From my understanding, they use an intermediate way to find a link that is most probably reliable, thus achieve better availability.
   
3. What affects the radio signal strength in different directions? How does the shape and angle of antenna affect the signal strength? Maybe we could find a way to make the radio to achieve circular signal coverage, or at least less irregular.
   
4. How does multi-path and attenuation affect packet delivery, and on what extend? It seems that many people blame the multi-path effect, but we do not have a quantitative measurement of it.
   
5. How does the gray area changes over time? What is the cause of it? It seems that packet losses are highly correlated over short time periods, but are independent over longer periods.
   

Obviously many other questions can be added to this list. Wireless sensor network is greatly affected by the environment, and it is hard to evaluate all the factors, but by investigating on each of the above questions, we could get a better understanding of what problems we are facing at.

Think the opposite way

    What if high packet lost is inevitable? For example, the disposable router project for firefighters. That project is designed for harsh environment – in a fire scene, we are expecting high packet lost and possible node lost, so another research direction would be how to achieve reliable connection with all these obstacles for communication. The challenge part of such application would be how to find the tradeoff between reliability and throughput.

    To sum up, this paper provides many insights for investigating packet lose in wireless sensor network. They also left many open issues regards to a systematic technique study about packet delivery performance. The topology control mechanism mentioned in their paper is an interesting research direction. Meanwhile, design a system that could work in terrible radio environment would also be an interesting topic. A slight digression of this topic, maybe one day we could use Quantum Teleportation to solve the interference problem, and live happily ever after.