1 Measuring DSA decisions' rippling. This metric can measure how long a theater‐based DSA decision can last before another decision has to be made. A good centralized spectrum arbitrator should make DSA decisions that last.
2 Adapting to traffic demand. If spectrum resources are allocated more where traffic demand is high, this will result in higher throughput efficiency of the heterogeneous networks and this higher throughput efficiency can be manifested in established network performance metrics such as packet or message completion ratio, packet delay, packet delay variation (jitter), and packet loss.
3 Avoiding hidden nodes. A bird's eye view capability should overcome the hidden node phenomena and reduce the likelihood of mistakenly using a primary user's spectrum. This metric can be the measuring of the number of complaints from a primary user.example 2
By the end of this chapter, the reader should realize that Figure 4.1 oversimplified the trade space areas in order to illustrate the most important aspects of this trade space. Other factors such as the used antennas' technologies, the use of licensed spectrum, unlicensed spectrum or a mix of both, and the interactions between the cognitive DSA process and other cognitive processes the system uses add more dimensions to this trade space.
4.2 Decision Fusion Cases
Let us consider the case of sensing the presence of a signal where the local decision fusion engine was able to collect a knowledge repository that can be expressed as depicted as in Figure 4.2. This repository is illustrated in Figure 4.2 with three different scenarios of the RSSI collected samples. Scenario 1 represents the case of low noise variance, scenario 2 represents the case of medium noise variance, and scenario 3 represents the case of high noise variance. The local knowledge repository and its associated cognitive engine were able to select a decision threshold based on the detected energy levels where the gray dots to the left of the decision threshold represent the energy detected in the absence of the sensed signal and the black dots to the right of the decision threshold represent the energy detected in the presence of the signal.
Figure 4.2 Local decision fusion based on single‐dimensional knowledge base.5
Notice that to obtain such knowledge repository, the sensor has to have some sort of pre‐knowledge of the sensed signal characteristics. As explained earlier in this book, this can be achieved in both the case of same‐channel in‐band sensing and the case of an augmented sensor sensing the presence of a primary user signal with known cyclostationary characteristics. In the case of same‐channel in‐band sensing, signal marks allow for the differentiation between energy samples representing noise or noise plus interfering signal (left side of the decision threshold in Figure 4.2) and energy samples representing signal plus noise or signal plus noise plus interfering signal (right side of the decision threshold in Figure 4.2).3 In the case of an augmented sensor probing a frequency band for potential use, the signal cyclostationary characteristics will allow the sensor to collect energy samples representing noise (left side of the decision threshold in Figure 4.2) and energy samples representing signal plus noise (right side of the decision threshold in Figure 4.2).
As scenarios 2 and 3 in Figure 4.2 show, when the noise power increases, the RSSI samples will tend to spread wider. Noise power can be due to pure AWGN or another secondary user overlaying its signal that has unknown characteristics. As scenario 3 in Figure 4.2 shows, a higher noise power will lead to the right side points and the left side points to encroach on each other. Note that higher noise power increases the standard deviation of the detected noise energy samples and the detected signal plus noise energy samples. With this case, the local decision fusion engine is able to hypothesize the presence or absence of a communications signal but clearly noise power increase can lead to either a higher probability of false alarm or a higher probability of misdetection depending on where the decision threshold is chosen.
Now let us consider the case when the local decision fusion engine is able to create a more accurate energy detection fusion, as illustrated in Figure 4.3. With this case the knowledge repository can plot a two‐dimensional curve of RSSI versus SNIR. The inclusion of SNIR in the knowledge repository adds a better depiction and more accurate hypothesizing as the increase in noise energy can cause spreading of the plotted points in two dimensions instead of one dimension. This will result in false alarm and misdetection hypotheses only in extreme cases such as the presence of very high noise power.
Figure 4.3 Decision fusion based on two‐dimensional knowledge base.
This example is used to illustrate the importance of coordinating between decision fusion hierarchies. If the local decision fusion follows the approach depicted in Figure 4.2 to reduce computational complexity, the distributed or centralized decision fusion may need to create knowledge repositories equivalent to Figure 4.3 to reduce false alarm and misdetection probabilities. On the other hand, if the local decision fusion engine was able to hypothesize based on Figure 4.3, distributed and centralized decision fusion engines can focus on other DSA aspects, such as spatial location of interference and the creation of a more accurate spectrum utilization map.
The trade space in Figure 4.1 illustrates some important aspects. In reality, there are more aspects in this trade space. For example, having a more detailed knowledge repository at the local node may not be achievable because of processing and power limitations.4,20 At a centralized arbitrator, processing power may not be a limiting factor. On the other hand, sending more detailed spectrum sensing information to a centralized arbitrator can have its own drawbacks to include the use of more bandwidth for DSA control traffic.
Relying on the centralized arbitrator for more decision fusion can result in the loss of information when sensing information is fused locally. For example, local fusion of SNIR values may produce information about SNIR that includes the mean, average, and standard deviation out of a large sample of SNIR values. This processing can result in some information loss and may result in the centralized arbitrator failing to achieve the desired lower probability of false alarm or lower probability of misdetection. Local fusion using raw RSSI and SNIR values can have a more accurate cutoff threshold in