Communications – electrical: acoustic wave systems and devices – Receiver circuitry
Reexamination Certificate
2000-01-28
2001-07-17
Pihulic, Daniel T. (Department: 3662)
Communications, electrical: acoustic wave systems and devices
Receiver circuitry
C367S901000
Reexamination Certificate
active
06262943
ABSTRACT:
This invention relates to a method of localising noise effects and to a signal processing system for applying the method. The invention also relates to applications of the signal processing system in signal detection and tracking system associated with analysis of multicomponent signals.
Regardless of the source of a signal, its detection will be imperfect and detected real signals will contain noise and often a systematic error of comparable magnitude. In particular, tracking a time-varying signal may automatically generate a systematic error between one time period and the next because of dynamic change. There is a perceived need to be able to track changes in a signal with time. Such changes may arise for numerous reasons and in many different types of signal. Of particular interest are periodic phenomena which commonly occur in many aspects of life. For example communication signals, the sound of rotating machinery, the heartbeat and speech can all be broken down into component periodic parts.
Many modelling schemes have been used in the analysis of periodic phenomena. Such analysis admits of numerous applications: for example, subtracting periodic signal components from the acoustic spectrum of a rotating engine makes it possible to detect relatively low intensity non-periodic sounds such as those from a gearbox. Similarly if each periodic component can be isolated and ascribed to a particular single source, the operational sound of that source can be monitored. This enables early diagnosis of a failing component, indication of which would otherwise be lost amongst a cacophony of sounds. Similarly the periodic components of speech may be isolated from other sounds and spoken communication may be made in a noisy environment. Medical applications also exist: heartbeat can be analysed and harmonic interference removed from biomedical signals to enable them to be monitored. Furthermore such a representation of a complex signal requires far less bandwidth than the signal itself, and can be passed through a lower bandwidth telecommunications channel.
The prior art contains various techniques for estimating periodicities in periodic systems. Most are based on Fourier analysis which does not cope well with resolving multiple signals present in an overall received signal nor does it deal effectively with tracking time-varying periodicities or amplitudes. Fourier analysis requires an initial evaluation of a measure of likelihood for each of many possible test values of the periodicity. The range of potential periodicities is continuous and in constructing a discrete set of test values a sufficiently fine scale must be used. Generally the analysis leading to an appropriate set is performed either in the Fourier domain or indirectly by using comb filters. The use of fixed frequency Fourier components as the basis of analysis precludes ready adaptation of either technique to tracking non-stationary periodicities. Furthermore the Fourier calculation is computationally expensive making real-time analysis more difficult and any approximations result in models with a lower signal to noise ratio (SNR). The comb filter offers an improvement in terms of SNR, but still retains the basic disadvantages of a Fourier technique.
In the Fourier calculation, harmonically related Fourier components must be recombined. In general the coefficients are complex valued and so direct linear addition cannot be used to assemble the harmonic set to form the overall modelled signal. If non-coherent second order summation of the coefficients is used, it follows that this measurement is not optimised in terms of its SNR at any test value of period.
Alternatively a bank of comb filters, one filter per test value matched to a given test period, is used to reconstruct trial estimates of periodic components. Directly applied in the time domain, a comb filter may be calculated by pre-multiplying the Inverse Fourier and Fourier transforms using only those harmonic components which are relevant to a particular test value of period. The Inverse Fourier transform provides a more optimal method of recombining the harmonically related Fourier coefficients. Again a likelihood spectrum is obtained as a function of test period; in this method it is by evaluating the mean power output of each comb filter. The Fourier components are not therefore calculated explicitly and the SNR is improved when compared to the non-coherent Fourier method of the previous paragraph.
It is an object of this invention to provide a method applicable to the detection and tracking of multiple non-stationary components of a signal.
In this patent specification the term “cyclet” is used to refer to any basis component of a signal which is replicated at regular time intervals. Theoretically, the amplitude profile and repetition frequency of a cyclet are not constrained in any way. A cyclet “track” describes the variation of cyclet periodicity and amplitude profile over time.
The present invention provides a method of localising noise effects in a first signal by means of a representation thereof: a model signal comprising at least one modelled component representable by a parameter set comprising at least one model parameter characterised in that the method comprises the steps of:
(a) subtracting the model signal from the first signal to derive a residual signal;
(b) scaling the residual signal by a multiplicative pseudo-integrating factor &rgr; having a value between zero and unity;
(c) adding the scaled residual signal obtained in Step (b) to at least a proportion of the model signal to give at least one pseudo-integrated signal, wherein perturbing noise effects in the scaled residual signal are localised.
This invention enables noise to be distinguished from systematic factors which affect signal clarity and magnitude. Potential applications of the technique used in this invention, termed pseudo-integration, are numerous. Often noise is of similar magnitude to and masks more important, systematic, signal effects. This invention provides a means of localising the effects of noise on parameter estimates and enabling information regarding systematic signal effects to be extracted.
Prior to Step (c), the model signal is preferably multiplied by (1−&rgr;) to obtain a proportion thereof comprising at least one scaled model component for addition to the scaled residual signal. The pseudo-integrating factor &rgr; must be less than or equal to unity, and in most applications of this technique it is of the order 0.01. The real signal is considered in this invention to be equivalent to the model signal perturbed by systematic effects and those of random noise. By scaling the residual signal, containing random noise and systematic effects, by a small factor &rgr; and the model signal by (1−&rgr;), peaks in the modelled signal are ~100 times the scaled level of the noise which perturbs them. Within the pseudo-integrated data the perturbing effects of noise are therefore localised to be in the vicinity of each modelled component or parameter. In general peaks within a signal which are masked by noise are not localised. Integration is required to enable such peaks to be distinguished above the effects of random noise.
The model parameter set representative of each modelled component may be corrected for systematic errors by the following additional steps:
(a) modelling the pseudo-integrated signal of each particular modelled component to obtain a corresponding set of locally perturbed model parameters;
(b) subtracting each model parameter from its corresponding locally perturbed model parameter to obtain a corresponding set of at least one localised perturbation error;
(c) resealing each localised perturbation error back to its original level by division with the pseudo-integrating factor &rgr;; and
(d) adding each rescaled localised perturbation error to its corresponding model parameter thereby obtaining a set of respective pseudo-integrated model parameters.
This extension of pseudo-integration enables systematic errors or effects to be corrected.
Nixon & Vanderhye P.C.
Pihulic Daniel T.
The Secretary of State for Defence in Her Britannic Majesty&apos
LandOfFree
Signal processing system for sensing a periodic signal in noise does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Signal processing system for sensing a periodic signal in noise, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Signal processing system for sensing a periodic signal in noise will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2513413