HOME     CONTACT US      PRESS ROOM     CAREERS

TRAINABLE DIGITAL LOGIC     TRAINABLE DETECTORS      ABOUT NEURAL SYSTEMS CORPORATION     FAQs

Frequently Asked Questions

What is the Shannon channel capacity (sometimes called the Shannon limit)?

How is the Shannon channel capacity achieved?

At high SNR what gain in bit rate could be expected if Shannon channel capacity is realized?

What is the advantage of NSC's trainable detectors and error correction?

What is meant by "trainable detector"?

What is a neural network?

What does trainable digital logic do and what are its advantages?

Please give examples of trainable digital logic network applications?

What is the Shannon channel capacity (sometimes called the Shannon limit)? The Shannon channel capacity is the maximum number of bits per sec. that can be transmitted without error for a given signal-to-noise ratio (SNR) and bandwidth (W). For a Gaussian channel it is: C = W log2 (1 + SNR)

How is the Shannon channel capacity achieved? Shannon proved that it was possible to get arbitrarily close to the capacity but did not show how. Turbo codes were the first method used to achieve near Shannon capacity at lower SNR's. The NSC algorithms can realize near Shannon capacity at SNR's typical of cable channels.

At high SNR what gain in bit rate could be expected if Shannon channel capacity is realized? For example, at a signal-to-noise ratio of 30db approximately a 40% gain could be expected when compared to 256 QAM.

What is the advantage of NSC's trainable detectors and error correction? NSC detectors have been demonstrated to achieve Shannon channel capacity at signal to noise ratios typical of cable channels, thus increasing the bit rate through a channel.

What is meant by "trainable detector"? Trainable detectors are optimized for a particular application by the use of known signals. The NSC detectors are adjusted by algorithms applying known signals in noise to minimize the bit error rate. A "training set" of known signals, where all the bit values and the resulting data patterns are known, can be used during production, or adjustments can be made in the field using headers ahead of data sequences.

What is a neural network? A classical neural network is constructed of a set of interconnected neuron analogs that can be adjusted or "trained" to recognize an object. For example, the network may be taught to recognize printed letters and to indicate which letter is present. The most common neural network is the back-propagation network. Here, the algorithm may use a sequence of letters and adjust the neuron analogs to give better and better recognition accuracy.

What does trainable digital logic do and what are its advantages? Almost always classical neural networks are implemented on a programmable digital computer where recognition requires non-linear functions and large amounts of multiplication and addition. In contrast, trainable digital logic networks implement the same logic in the form of complex switching functions with no arithmetic or non-linear functions. A trainable digital logic network is trained using the same information that is used to train the back-propagation network. It typically has a gate count that is several orders of magnitude smaller and is several orders of magnitude faster. The smaller gate count and much greater speed can overcome some recognized disadvantages of classical networks and allow widespread use in, for example, consumer electronics.

Please give examples of trainable digital logic network applications? The following are application examples: Digital cable channel bit rate increases Larger capacities for hard disk drives Heart arrhythmia warning Printed character recognition Communication modulation type recognition Vehicle sound recognition Sonar sound classification Radar type classification

HOME     CONTACT US      LEGAL INFORMATION      ABOUT NEURAL SYSTEMS CORPORATION     SITEMAP

Neural Systems Corporation© 2005 - 2009 All Rights Reserved