Neuronal phase-locked loops

Details Neuronal phase-locked loops Neuronal phase-locked loops

1998-10-09

2003-06-17

Starks, Jr., Wilbert L. (Department: 2121)

Data processing: artificial intelligence

Neural network

C711S200000

Reexamination Certificate

active

06581046

ABSTRACT:

FIELD OF THE INVENTION
This invention is directed to a novel device, called a “neuronal phase-locked loop” (NPLL), that can decode temporally-encoded information and convert it to a rate code.
A phase locked loop (“PLL”) is a circuit for synchronizing a local variable or rate controlled oscillator with the phase of a transmitted signal.. Electronic embodiments of PLL's are well known and widely used, for example, in communications technology.
The NPLL is based on an algorithm similar to that of the electronic PLL, but utilizes a totally different implementation. It is a stochastic device, implemented by neural networks (real or simulated). Nueral systems in general, while relatively recently developed, are well known to those skilled in the art. See for example, U.S. Pat. No. 4,937,872 to Hopfield et al.
The simplest form of the NPLL includes a phase detector (that is, a neuronal-plausible version of an ideal coincidence detector) and a controllable local oscillator that are connected in a negative feedback loop. The phase detector compares the firing times of the local oscillator and the input and provides an output whose firing rate is monotonically related to the time difference. The output rate is fed back to the local oscillator and forces it to phase-lock to the input. Every temporal interval at the input is associated with a specific pair of output rate and time difference values; the higher the output rate the further the local oscillator is driven from its intrinsic frequency. Sequences of input intervals, which by definition encode input information, are thus represented by sequences of firing rates at the PLL's output.
Unlike the electronic PLL, the NPLL is an adaptive device which can deal with signals whose exact characteristics are not known in advance and can adapt to changing conditions. It is ideal for low frequency applications (1-200 Hz) . NPLLs can be applied to artificial, perceptual-like, operations such as auditory speech recognition, visual and tactile object localization, texture identification and motion detection.
BACKGROUND OF THE INVENTION
The distinction between rate and temporal coding is not always clear. For example, temporal coding is sometimes regarded as rate coding with a fine time resolution. Herein “temporal coding” will refer to coding in which the exact time of every spike (pulse) is informative. Whereas, rate coding will be associated here with a temporal window (the “rate-bin”), within which the information is carried by the average firing rate over the entire temporal window, and the exact temporal information is not informative. The rate-bin is usually determined by the integration times of the ‘readout’ mechanisms. A “rate-encoded” signal can thus be described by a series of numbers each of which represents the average firing rate in a single rate-bin (Appendix A.1). Fluctuations in the average firing rate of a neuron over different rate-bins are considered here as fluctuations of rate-encoded information, and not as temporal coding, as has been considered previously.
A “temporally-encoded” signal is described by a series of numbers each of which represents either the timing of a single spike or a single inter-spike-interval (hereinafter referred to as “ISI”; see Appendix A.1) . The information contained in the spiking times can be presented in different ways, two of which are depicted in FIG.
1
: M(n) describes the deviations of the actual train from an imaginary, ideally periodic, “carrier” train, and I(n) describes the ISI's.
FIG. 1
also demonstrates the distinction between temporal and rate coding; the spike train in this example carries a significant amount of information if a temporal coding is assumed (
FIG. 1
a
), but almost no information if a rate coding, with a particular rate-bin, is assumed (
FIG. 1
b
). Practically, this distinction is important for reading out the information of the spike train. A readout mechanism based on rate will lose more and more information as its integration time increases. To readout temporally-encoded information, on the other hand, a rate-based mechanism needs to employ integration times shorter than half of the input temporal resolution, an implementation that is both non-efficient and, with fine input resolution, not practical for neuronal application. The other alternative is to utilize pre-IQ processing by time-sensitive mechanisms, i.e., mechanisms that produce populations of spikes, where the number of spikes in a population directly represents the ISI at the input and the exact times of these output spikes is not important.
In mammals, sensory information is encoded by both rate and temporal coding. Whereas spatial static information is usually encoded by rate, dynamic information, that is generated during movements of either the stimulus or the sensory organ, is encoded also by temporal cues (see, for example, encoding of spatial intervals by ISIs of tactile and visual neurons). In contrast, motor control is assumed to predominately utilize rate-coding already at the early stage of motor planning. Thus, information carried by the sensory temporal components is likely translated, by neuronal circuits in the brain, to rate-encoded signals that are “readable” by the motor system. If such a translation occurs early in a sensory pathway, the translation would also facilitate integration of temporally-encoded information with other, rate-encoded, sensory information. This necessity for translation was elegantly demonstrated by Mountcastle and his colleagues, over the last few decades.
A mechanism that utilizes neuronal delay-lines to transform temporal coding to rate coding has been suggested. Such delay-lines exist in the electric sensory system of electric fishes and in the subcortical auditory systems of birds and mammals. These delay-lines are probably utilized to decode temporal disparities, which in the submillisecond and low millisecond ranges can determine interaural time differences and echo delays, respectively. As the delay increases above a few hundred microseconds, implementations of delay-lines require multiple neuronal elements and the accuracy decreases. A mechanism that Aid utilizes synaptic time-constants appears more suitable to decode temporally-encoded information in the millisecond range. Both these mechanisms describe “passive”, open-loop decoding schemes that are based on classification of different ISI's according to their interaction with fixed neuronal temporal features.
The present inention, on the other hand, provides an “active”, closed-loop decoding mechanism, which dynamically adapts its working parameters to match the incoming signal. This model, the neuronal phase-locked loop (NPLL) model, is based on a local oscillator “measuring” the instantaneous temporal period of the input by comparing it to its own period. During decoding, the local oscillator updates its period according to the result of the comparison, such that it remains tuned to the changing input.
The PLL is a well-known mechanism in electrical engineering and is often used for the decoding of phase-modulated signals. The present invention is based on continuous-time electronic PLLs, modified to fit discrete-time NPLLS. The implementation is totally different than electronic implementations and is based on neuronal, or neuronal-like, elements organized in small networks.
Other objects, advantages and novel features of the present invention will become apparent from the following detailed description of the invention when considered in conjunction with the accompanying drawings.


REFERENCES:
patent: 5495512 (1996-02-01), Kovacs et al.
patent: 5602884 (1997-02-01), Wieczorkiewicz et al.
patent: 5610705 (1997-03-01), Brosnan et al.
patent: 5625506 (1997-04-01), Dovek et al.
patent: 5677802 (1997-10-01), Saiki et al.
patent: 5682112 (1997-10-01), Fukushima
patent: 5705956 (1998-01-01), Neely
Hoppensteadt, F.C.; Izhikevich, E.M., Pattern recognition via synchronization in phase-locked loop neutral networks, Neural Networks, IEEE Transactions on, vol.: 11 Issue: 3, Center fo

Affiliated with

Also associated with

Rating

Neuronal phase-locked loops does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Neuronal phase-locked loops, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Neuronal phase-locked loops will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-3147265