Milen Nikolov

Communicating data is among the most energy expensive routines across different types of wireless networks. For instance, receiving and transmitting data in wireless sensor networks (WSN) consisting of Mica2 nodes running TinyDB applications constitutes about 59% of the total energy consumption. Reducing the amount of data transmitted and/or energy consumed per transmission could lead to longer network lifetimes and significantly impact network functions.

Most fauna organisms have encountered a similar problem (albeit not in the context of wireless communications). Animals sense different environment qualia via large number of sensors, reporting to cortexes organized withing some form of a brain. In many cases, the number of sensors allow for the so-called population coding. Combination of subsets of sensors encode only a particular message. This is highly efficient since it requires only small subsets of neurons to be active and consume energy at a time.

Following a similar population code principle, we introduce encoded sensing (ES) that substantially reduces the energy required for transmission of data. As shown on the Fig. 1, each subset of a group of sensors (analogous to neurons in a bio system) is responsible for a particular stimulus and transmits (analogous to neurons firing) only when that stimulus is detected. I.e. each sensor has a specific receptive field in the parlance of neuroscience. Based on that principle, we achieve at least double the efficiency of traditional transmission schemes. Inspired by the population code efficiency, we design a sparse direct sequence spread spectrum (DSSS) receiver. The circuitry of the receiver contains Klog(N) vs. N elements of a traditional DSSS receiver. The sparse DSSS receiver is about 60% cheaper to produce (it requires 60% less of the cost-heavy components). More details can be found in our article

Fig. 3: The cortical homunculus; large number of sensors concentrated in certain areas (and representing different fractions of cortical domain) encode selected frames of the environment via population coding.

Fig. 2: Nodes (white) placed uniformly at random. 4 responsible nodes per measurement. 21 in a group. 5985 possible measurements. 36 groups in a network. We require 4 signals vs the 13 needed traditionally.

Fig. 1: The encoding sensing scheme: subsets of sensors assigned to different portions of the measuring spectrum. It is hypothesized that different sensory system such as olfaction could work on a similar principle.