Introduction (779811), страница 2
Текст из файла (страница 2)
Figure 1.2 shows a broad categorisation of some DSPapplications. This section provides a review of several key applications ofdigital signal processing methods.1.3.1 Adaptive Noise Cancellation and Noise ReductionIn speech communication from a noisy acoustic environment such as amoving car or train, or over a noisy telephone channel, the speech signal isobserved in an additive random noise.
In signal measurement systems theinformation-bearing signal is often contaminated by noise from itssurrounding environment. The noisy observation y(m) can be modelled asy(m) = x(m) + n(m)(1.2)6IntroductionDSP ApplicationsInformation Transmission/Storage/RetrievalSource/Channel CodingInformation extractionParameter EstimationChannel EqualisationSpeech coding, image coding,data compression, communicationover noisy channelsSignal and datacommunication onadverse channelsSpectral analysis, radarand sonar signal processing,signal enhancement,geophysics explorationSignal ClassificationSpeech recognition, imageand character recognition,signal detectionFigure 1.2 A classification of the applications of digital signal processing.where x (m) and n( m) are the signal and the noise, and m is the discretetime index.
In some situations, for example when using a mobile telephonein a moving car, or when using a radio communication device in an aircraftcockpit, it may be possible to measure and estimate the instantaneousamplitude of the ambient noise using a directional microphone. The signalx(m) may then be recovered by subtraction of an estimate of the noise fromthe noisy signal.Figure 1.3 shows a two-input adaptive noise cancellation system forenhancement of noisy speech. In this system a directional microphone takesNoisy signaly(m) = x(m) +n(m)Noiseα n(m+ τ )–1w0...–1zzw1w2z–1Signal^x(m)wP-1Adaptationalgorithm^Noise estimate n(m)Noise Estimation FilterFigure 1.3 Configuration of a two-microphone adaptive noise canceller.7Applications of Digital Signal Processingas input the noisy signal x(m) + n(m) , and a second directional microphone,positioned some distance away, measures the noise α n(m + τ ) .
Theattenuation factor α and the time delay τ provide a rather over-simplifiedmodel of the effects of propagation of the noise to different positions in thespace where the microphones are placed. The noise from the secondmicrophone is processed by an adaptive digital filter to make it equal to thenoise contaminating the speech signal, and then subtracted from the noisysignal to cancel out the noise. The adaptive noise canceller is more effectivein cancelling out the low-frequency part of the noise, but generally suffersfrom the non-stationary character of the signals, and from the oversimplified assumption that a linear filter can model the diffusion andpropagation of the noise sound in the space.In many applications, for example at the receiver of atelecommunication system, there is no access to the instantaneous value ofthe contaminating noise, and only the noisy signal is available.
In such casesthe noise cannot be cancelled out, but it may be reduced, in an averagesense, using the statistics of the signal and the noise process. Figure 1.4shows a bank of Wiener filters for reducing additive noise when only theNoisy signaly(m)=x(m)+n(m)Y(0)y(0)y(1)y(2)...Discrete Fourier TransformW0y(N-1)Signal and noisepower spectraY(1)Y(2)...W1W2...Y(N-1)WN -1^X(0)^X(1)^X(2)^X(N-1)Inverse Discrete Fourier TransformRestored signal^x(0)^x(1)^x(2)...^x(N-1)Wiener filterestimatorFigure 1.4 A frequency−domain Wiener filter for reducing additive noise.8Introductionnoisy signal is available. The filter bank coefficients attenuate each noisysignal frequency in inverse proportion to the signal–to–noise ratio at thatfrequency.
The Wiener filter bank coefficients, derived in Chapter 6, arecalculated from estimates of the power spectra of the signal and the noiseprocesses.1.3.2 Blind Channel EqualisationChannel equalisation is the recovery of a signal distorted in transmissionthrough a communication channel with a non-flat magnitude or a non-linearphase response. When the channel response is unknown the process ofsignal recovery is called blind equalisation.
Blind equalisation has a widerange of applications, for example in digital telecommunications forremoval of inter-symbol interference due to non-ideal channel and multipath propagation, in speech recognition for removal of the effects of themicrophones and the communication channels, in correction of distortedimages, analysis of seismic data, de-reverberation of acoustic gramophonerecordings etc.In practice, blind equalisation is feasible only if some useful statistics ofthe channel input are available. The success of a blind equalisation methoddepends on how much is known about the characteristics of the input signaland how useful this knowledge can be in the channel identification andequalisation process. Figure 1.5 illustrates the configuration of a decisiondirected equaliser.
This blind channel equaliser is composed of two distinctsections: an adaptive equaliser that removes a large part of the channeldistortion, followed by a non-linear decision device for an improvedestimate of the channel input. The output of the decision device is the finalChannel noisen(m)x(m)y(m)Channel distortionH(f)+Decision deviceEqualiser^x(m)Hinv(f)ffAdaptationalgorithm-++Error signalBlind decision-directed equaliserFigure 1.5 Configuration of a decision-directed blind channel equaliser.9Applications of Digital Signal Processingestimate of the channel input, and it is used as the desired signal to directthe equaliser adaptation process.
Blind equalisation is covered in detail inChapter 15.1.3.3 Signal Classification and Pattern RecognitionSignal classification is used in detection, pattern recognition and decisionmaking systems. For example, a simple binary-state classifier can act as thedetector of the presence, or the absence, of a known waveform in noise. Insignal classification, the aim is to design a minimum-error system forlabelling a signal with one of a number of likely classes of signal.To design a classifier; a set of models are trained for the classes ofsignals that are of interest in the application. The simplest form that themodels can assume is a bank, or code book, of waveforms, eachrepresenting the prototype for one class of signals. A more complete modelfor each class of signals takes the form of a probability distribution function.In the classification phase, a signal is labelled with the nearest or the mostlikely class.
For example, in communication of a binary bit stream over aband-pass channel, the binary phase–shift keying (BPSK) scheme signalsthe bit “1” using the waveform Ac sin ω ct and the bit “0” using − Ac sin ω c t .At the receiver, the decoder has the task of classifying and labelling thereceived noisy signal as a “1” or a “0”. Figure 1.6 illustrates a correlationreceiver for a BPSK signalling scheme. The receiver has two correlators,each programmed with one of the two symbols representing the binaryCorel(1)Received noisy symbolCorel(0)"1" if Corel(1) ≥ Corel(0)"0 " if Corel(1) < Corel(0)Correlator for symbol "1"Decisiondevice"1"Correlator for symbol "0"Figure 1.6 A block diagram illustration of the classifier in a binary phase-shift keyingdemodulation.10IntroductionM1likelihoodof M 1fY|M (Y| M 1 )Word modelSpeechsignalFeatureextractorFeaturesequenceYM2likelihoodof M 2fY|M (Y| M 2 )...Word modelMVlikelihoodof M vfY|M (Y| MV )Most likely word selectorWord modelM MLSilence model MsilfY|M (Y| Msil )likelihoodof M silFigure 1.7 Configuration of speech recognition system, f(Y|Mi) is the likelihood ofthe model Mi given an observation sequence Y.states for the bit “1” and the bit “0”.
The decoder correlates the unlabelledinput signal with each of the two candidate symbols and selects thecandidate that has a higher correlation with the input.Figure 1.7 illustrates the use of a classifier in a limited–vocabulary,isolated-word speech recognition system. Assume there are V words in thevocabulary. For each word a model is trained, on many different examplesof the spoken word, to capture the average characteristics and the statisticalvariations of the word. The classifier has access to a bank of V+1 models,one for each word in the vocabulary and an additional model for the silenceperiods.
In the speech recognition phase, the task is to decode and label an11Applications of Digital Signal Processingacoustic speech feature sequence, representing an unlabelled spoken word,as one of the V likely words or silence. For each candidate word theclassifier calculates a probability score and selects the word with the highestscore.1.3.4 Linear Prediction Modelling of SpeechLinear predictive models are widely used in speech processing applicationssuch as low–bit–rate speech coding in cellular telephony, speechenhancement and speech recognition. Speech is generated by inhaling airinto the lungs, and then exhaling it through the vibrating glottis cords andthe vocal tract.
The random, noise-like, air flow from the lungs is spectrallyshaped and amplified by the vibrations of the glottal cords and the resonanceof the vocal tract. The effect of the vibrations of the glottal cords and thevocal tract is to introduce a measure of correlation and predictability on therandom variations of the air from the lungs. Figure 1.8 illustrates a modelfor speech production. The source models the lung and emits a randomexcitation signal which is filtered, first by a pitch filter model of the glottalcords and then by a model of the vocal tract.The main source of correlation in speech is the vocal tract modelled by alinear predictor.
![](https://s.studizba.com/z.php?f=/templates/si/i/noavatar.png&w=100&h=100&t=1"){kind=avatar}
