Noises report

^^… Noises …^^

    What is noise figure?

Noise figure is a figure-of-merit that describes the amount of excess noise present in a system. Minimizing noise figure reduces system impairments that result from noise. In our personal lives, noise degrades the image quality

of TV pictures, and adversely impacts the voice quality of cell phone calls. In military applications like radar, receiver noise limits the effective range of the system. With digital communications, noise increases the bit-error rate. System designers always try to optimize the overall signal-to-noise ratio (SNR) of the system. This can be done by increasing the signal, or by reducing noise. In a

transmit/receive system like a radar system, one possibility is to increase the radar’s transmitted power by using bigger, more powerful amplifiers, and/or by using larger antennas. Decreasing the path loss between the transmitter and

receiver also helps increase SNR, but path loss is often defined by the operating

environment and cannot be controlled by the system designer. SNR can also be

increased by decreasing receiver-contributed noise, which is usually determined by the quality of the low-noise amplifier (LNA) at the front end of the receiver. In genera l, it is easier and less expensive to decrease receiver noise (and achieve a better noise figure) than to increase transmitter power.

The definition of noise figure is simple and intuitive. The noise factor (F) of a network is defined as the input SNR divided by the output SNR:

 F = (Si/Ni)/(So/No), where

   Si= input signal power

   So = output signal power

   Ni = input noise power

   No = output noise power

Noise figure (NF) is simply the noise factor expressed in decibels: NF = 10*log (F)

This definition is true for any electrical network, including those that shift the frequency of the input signal to a different output frequency, such as an up or down converter.

To better understand the concept of noise figure, consider an amplifier where the output signal is equal to the input signal multiplied by the gain of the amplifier. If the amplifier is perfect, the output noise is also equal to the input noise multiplied by the amplifier’s gain, resulting in the same SNR at both the input and output of the amplifier. For any real-world amplifier however, the output noise is larger than the input noise multiplied by the gain, so the SNR at the output is smaller than that at the input, resulting in F being greater than one, or NF being greater than 0 dB.

It is important to note that when measuring and comparing noise figures, the test system is assumed to provide perfect 50-ohm terminations at the input and output of the device-under-test (DUT). In real-world scenarios however, this is never the case. Later, we will discuss the accuracy implications if our test system is not exactly 50 ohms, and we will show how calibration and measurement methods can overcome the errors produced from an imperfect 50-ohm source

match.

* Simulating Network Noise;

We propose a simple dependency-based simulation scheme to assess the influence of network noise to complex application communication patterns .Communication Patterns usually consis to f point-to-point messages and dependencies between them. For example a small-message

Broadcast operation is often implemented with a binomial tree. Such an algorithm usually consists of multiple rounds where each non-root and non-leaf node has to receive data and pass this data onto its children. This scheme allows the construction of a global dependence graph in which the send operations of all nodes but the root process depend on a previous receive (from a previous communication round).

Our simulation considers an application communicator of size x and a perturbation communicator of size y . The communication in the application communicator is modeled as a collective operation that consists of multiple rounds.

The perturbation communicator is modeled with random

Communication (each process picks a random peer to send data to so tha teach process is receiving from exactly one peer) for each round of the application communicator .The simulation performs the routing of all messages (application and perturbation) through the network and counts the congestion of each physical link. Each edge in the dependence graph is then annotated with the maximum congestion along the corresponding logical link in the network simulation.

Then, a breadth first search is started at every root node and the longest path in the dependence graph is reported as the time for this collective operation (we assume that the finishing time of the last process of the collective operation is significant).

* Importance of noise :

 accuracy

   One of the goals of this application note is to give the reader a better understanding of accuracy issues related to noise figure measurements.

   Measurement accuracy is important in both R&D and manufacturing environments.

In R&D, better noise figure accuracy means that there will be a better correlation between simulations and measurements, helping designers refine circuit models faster. But higher accuracy also means that a system designer can better optimize transmit/receive systems like those used in radar applications.

   When assigning performance values to all of the individual components of the

system, the system designer must add a guard band based on measurement accuracy, since a component designer will measure their device to verify its performance. For noise figure, improved measurement accuracy and smaller guard bands mean the LNA can have better specifications, which in turn means that lower-power transmit amplifiers can be used for the same overall system

SNR. This translates to smaller, lighter, and cheaper transmitters, all of which is very important for airborne and spaceborne applications.

   In manufacturing, improved measurement accuracy also allows use of smaller guard bands, which provides better correlation among multiple test stations.

   This means fewer products must be reworked, resulting in higher yields and improved throughput, and lower test costs. Smaller guard bands also allow better device specifications, yielding more competitive products that command higher prices or attain higher market share.

* Conclusions and Future Work;

We showed that network noise can have a significant impact to the performance of large-scale parallel applications.