Brain map temporary address : https://twilight-fanyi.gitee.io/mind-map/
Article directory
1 Introduction
In recent years, with the continuous development of electronic devices, everyone may realize that there are more intelligent electronic devices around them that can be connected to the Internet. Smart agriculture, smart transportation, etc. are also constantly developing and applied to corresponding scenarios; Network research is also increasing, and more and more people are gradually realizing the infinite applicability of wireless sensor networks. For example, sensor networks can be used to collect data in situations such as environmental monitoring, habitat detection, structure detection, equipment diagnosis, disaster management, and emergency response.
However, wireless sensor networks (WSNs) have some resource constraints when applied: limited power supply , communication bandwidth , processing speed and memory space . One possible way to maximize these resources is to compress the sensor data.
Usually, processing data consumes much less energy than transmitting data in wireless medium, so using data compression before transmitting data can effectively reduce the total power consumption of sensor nodes. However, most of the existing compression algorithms are too large for sensor nodes with very weak processing capabilities, and each sensor node is limited by resources such as electricity. Therefore, how to design our compression algorithm in the case of very large resource constraints such as sensor nodes is the most important problem .
2. Analysis of energy consumption in wireless medium
In terms of power consumption, the operation of wireless sensor nodes can be divided into three parts: sensing, processing, and transmission. Among the three operations, the task with the most energy consumption is known to be data transmission, and about 80% of the power consumption of each sensor node is used for data transmission.
Therefore, if we can minimize the size of the data through data compression , it will reduce the transmission power. On the other hand, with data compression, more processing power will be required to execute the compression algorithm. In order to reduce the total power consumption, the total power consumption of transmission and processing must be reduced. Power consumed to compress a data string of "a" bits into a data string of "b" bits, where a>b.
Experiment ① Energy consumption experiment for sending data
This lab collects power consumption data by sending 1 bit by executing a simple 32-bit addition instruction.
The results show that sending a bit of data consumes about 0.4µJ of energy, and executing an addition instruction consumes only 0.86nJ of energy. The power consumption of transmitting one bit over the radio medium is at least 480 times that of executing one additional instruction.
So if more than one bit (equivalent to 480 addition instructions) is removed from the original data bit string by a compression operation, the total power consumption of the sensor node will be reduced.
Experiment ② The total power consumption of various lossless data compression for text and web data
The compression algorithms tested in this experiment are bzip2(BWT algorithm), compress(LZE algorithm), LZO(LZ77), PPMd(PPM) and zlib(LZ77).
The above table shows the number of instructions required to delete a bit for each algorithm. The number of instructions for each data compression algorithm is much less than 480 instructions, so when these algorithms are applied, it is expected to consume less total energy than simply transferring a string of data.
Experimental results show that for most compression algorithms, compressing data before transmitting it can reduce the total power consumption. In some cases, however, applying data compression can increase overall power consumption due to memory accesses during compression execution. Accessing memory is expensive in terms of energy consumption.
in conclusion
Using data compression before transmitting data in wireless media is an effective way to reduce energy consumption. However, it is critical to choose a data compression algorithm that requires fewer memory accesses during execution.
3. Data compression technology
① Sort code
As part of the data funnel routing, an sequentially encoded data compression scheme was introduced. The compression scheme is as follows:
① Pass the data from the sensor nodes in the interest region (Interested region) to the collector nodes, as shown in Figure 1. In data funnel routing, some sensor nodes work as data sink nodes.
For example: Node A, Node B, Node D are data sink nodes. On the sink node, the sensor data collected by other nodes is combined, and the aggregated data is sent to its parent node. At node D in Figure 3, the data collected by node E is combined with the data collected by node D itself.
② Then, transmit the aggregated data to Node B.
In this algorithm, when data is merged on an aggregation node, some data is removed. In order to include information about missing data in the aggregated data, the order of the data packets is exploited.
For example, four nodes N1, N2, N3, and N4 send data to the aggregation node (Na). The data value of each node is any integer between 0 and 5. If we decide to discard N4's data and represent N4's data by ordering packets from the other 3 nodes (N1, N2, and N3), then there are 3 nodes != 6 possible orderings. Therefore, by using a three-packet arrangement, it is possible to include N4's data values in one aggregate packet without actually including N4's packets. The possible combinations of permutations and data values are shown in the figure below.
in conclusion
- Advantages
The data compression method has a low compression ratio and a simple algorithm, and may be applied to wireless sensor networks. - Disadvantages
One difficulty of using this scheme is that it requires a mapping table since there is no efficient algorithm for mapping permutations to data values. The size of the table grows exponentially as the number of aggregated sensor nodes increases.
② Pipeline network compression
Pipelined network compression schemes are discussed here. The basic idea is to trade high data transmission delay for low transmission energy consumption . The collected sensor data is stored in the buffer of the aggregation node for a period of time. During this time, the data packets are merged into one data packet, the redundancy in the data packets is eliminated, and the data transmission is minimized.
in conclusion
- Advantages :
One advantage of this simple compression scheme is that a shared prefix system can be used for node ids and timestamps. By doing this, more data compression can be achieved.
The efficiency of data compression depends on the length of the shared prefix. If we can set a long shared prefix, and the measurements have commonality, the compression ratio increases. - Disadvantage :
However, there is no similarity in the measured sensor values. Even setting a long shared prefix will reduce the efficiency of pipelined compression within the network.
Also, if we are consolidating a large number of packets, then a large data buffer is required to temporarily store these packets. Due to the limited memory space of the sensor node, not enough buffer space is available.
③ Low-complexity video compression
A low-complexity video compression scheme is introduced here. Since most current video coding techniques are designed with motion estimation and compensation, they require high computing power, which is usually not available in sensor nodes. Therefore, the method is based on the block change detection algorithm and JPEG data compression.
The figure above gives a block diagram of the image data processing flow. The algorithm is specially designed for wireless video surveillance system. This method divides each video frame into small blocks, each block contains eight 8 (64) pixels. To reduce computational complexity, only a subset of blocks (in this case all white blocks) are considered in each frame. In addition, within each block, a subset of pixels (number of allocated pixels) will be checked for changes, as shown in FIG. 7 . The number assigned to the pixel indicates the importance of the pixel (1=most important, 3=least important).
in conclusion
The experimental results show that the image quality processed by the algorithm is comparable to the image quality processed by MPEG-2, and a certain energy saving is achieved at the same time.
④ Distributed compression
The basic idea behind distributed compression schemes is to use a side information to encode a source information.
For example, there are two sources (X and Y) as shown in Figure 4. They are independent and identically distributed for the associated discrete letters. Since in a sensor network, sensor nodes are densely distributed in a sensor field, this association condition is easy to satisfy. X can then be compressed at the theoretical rate H(X|Y) of its conditional entropy without encoder 1 accessing Y. Conditional entropy can be expressed as
The decoder then selects a codevector in the coset that is closest in value to the codevector value sent by Y.
Below is a simple example of distributed compression. There are two (X and Y) 3-bit datasets. The Hamming distance between X and Y is no more than 1 bit. If both encoder 2 and decoder know Y, X can be compressed to 2 bits. So what happens to the compression ratio of X if Y is only known to the decoder?
In a distributed compression scheme, it is not efficient to distinguish between X=111 and X=000 since the decoder knows that Y and X are only 1 Hamming distance away from Y. Similarly, X=001 and 110, X=010 and 101, X=011 and 110 do not need to be distinguished. The two sets of X values are grouped into 4 costs and assigned 4 different binary index numbers:
If X=010, Y=110, the Y=110 received by the decoder is the side information from Y, and X=10 is the partial information from X. Then at the decoder, since the Hamming distance between 110 and 110 is 2, X=010 is chosen in coset 2. Whether X knows Y or not, X can still compress 3 bits of information into 2 bits.
in conclusion
Distributed compression schemes can be applied not only to discrete sources as in the above example, but also to continuous sources. Also, it can be used for lossless and lossy compression schemes.
For example, an 8-level quantizer can form 4 companion sets. It is worth noting that the two encoding vectors in each coset are grouped so that they can have the largest possible distance from each other.
4. Summary
In recent years, the application field of wireless sensor networks has been widely discussed. In the future, with the development of technology, the application field of wireless sensor network will be more extensive than now. People will get them more easily than they are now. However, as these days come, there are still many hurdles to overcome for the practical application of sensor networks. One of the obstacles is the limited resources of wireless nodes.
This paper has dealt with five different types of data compression schemes: sequential encoding, pipeline network compression, JPEG200, low-complexity video compression, distributed compression . Although these compression schemes are still in the development stage, the experimental results show that their compression ratio and power reduction are quite impressive. They are a feasible approach to resource constraints of wireless sensor nodes.