EEC: A Brilliant Approach

In a recent SIGCOMM publication on improving wireless communication, someone had a brilliant idea. Usually, when a packet is received by a wireless device, the packet is checked for errors. If the packet has errors that can't be recovered, it is discarded. The whole packet. The paper gives the following graph showing the fraction of bits that are typically corrupted in a wireless mesh network setting:



This shows that 50% of packets have some corruption within 0 to 2%. They claim that corrupted packets have some value, if the fraction of corrupted packets can be estimated. For example, error-tolerant applications, like video streaming could leverage corrupted packets to increase throughput if the error rate is below a threshold. Other applications could measure the error rate to determine setting selection (such as route selection or WiFi rate adaptation).

As a result, they propose a concept known as error estimation codes (EEC). The idea is to introduce a small amount of extra data to a packet used for error estimation. This is similar to error correction codes, which also introduce redundancy, but use the redundancy for error correction. The problem with error correction is that it requires a lot more redundancy to correct a "more corrupted" packet. EEC's additional redundancy guarantees estimation precision, and is the same regardless of the error rate.