220 likes | 407 Views
Measurement Techniques in Biomechanics Lecture 2. August 31 th Numerical Methods 1. Signal processing. Extract cleaner data Average waveforms Correlate to find similarities and differences Transform into other domains for information. Data Smoothing.
E N D
Measurement Techniques in BiomechanicsLecture 2 August 31th Numerical Methods 1
Signal processing • Extract cleaner data • Average waveforms • Correlate to find similarities and differences • Transform into other domains for information
Data Smoothing • An attempt to obtain the simplest representation of data that adequately describes an underlying process, while eliminating from consideration data scatter arising from experimental error Wood, 1982
A Little BackgroundLinear systems • A system is any process that generates an output signal in response to an input • A system is considered linear if it has 2 mathematical properties • Homogeniety • Additivity • A third property, Shift Invariance, is mandatory for DSP techniques
A Little Background Linear Systems have special properties • Commutative • You can add them and order does not matter • Can multiply by a constant • Can multiply by another signal • Synthesis/decomposition * Remember complex signals are a superposition (sum) of simpler signals
A Little Background First and Second Order Linear Systems • Order of a system relates to the differential equation needed to define it • First Order (first derivative = velocity) • Second Order (second derivative = acceleration) • Third Order (third derivative = jerk)
Some definitions • Derivative • Instantaneous rate of change (slope) • Differentiation • Process of finding the derivative • Integration • Used to measure totality (e.g. volume, area)
Finite Difference Techniques • Difference techniques are basically slopes of curves -derivatives • First order systems • Forward finite difference • Two point central difference • Tedious and error prone • Rarely used in processing today due to advent of computers REF Wood pg 315
Polynomials • Second order systems of finite difference (acceleration) • Typically parabolic – not linear • Used in a moving average manner (see Fig 4 Wood pg 323) • Are used to approximate underlying biological behaviour • Restricts to lower frequency as it does not go through all points (see figure) • Limits on use on data (see figure)
Splines • A number of low order polynomials • Strung together by “knots” • Used for interpolation of data • Kinematic data-missing markers • Rules of thumb • As few knots as possible • No more than one Max/Min (inflection point) • Extreme points should be in center of spline range • No more than a 10-15 point gap*
When to use derivatives • Finite differences, polynomials and splines are typically used to smooth clean data • They do not do well with a lot of noise • Can use these for missing data points and for some simple analyses
Fourier Smoothing • Convert signal into frequency domain • Choose signal content desired • Reconstruct using only those frequency components (harmonics) • This is a better way to remove noise from signal or to simplify a signal • A better (more common) way however is to filter
Digital Filters • Used for 2 general purposes • Signal separation • Needed when a signal has been contaminated (e.g. interference, noise) • Signal restoration • Needed when a signal is distorted (e.g. use of poor equipment)
How do digital filters work conceptually • A digital filter allows certain frequencies to pass while others are stopped • Which frequencies are passed depends on what type of filter you are using and the cut off frequency you have set • High • Low • Band • Notch
You must know the frequency content of your signal to choose what type of filter to use • You must then determine the Fc • Winter advocates using residual analysis to determine the appropriate Fc • However this only gives you the “bulk” of the signal and therefore f filtered at that frequency you may lose data • Another method to consider is the -3dB mark • A decibal is a logarithmic expression of amplitude • 3dB is half the power of the signal** = .707 of amplitude
Some other considerations when choosing a filter • How does the filter performance? • 3 main regions to consider for a filter • Pass band • Stop band • Transition band
Frequency Response • Ideally you want a filter: • Fast roll off • No pass band ripple • Good stop band attenuation • Every filter has a • Step response • Frequency response • Impulse response
When digital filtering • 3 considerations • Order (higher the order the better frequency magnitude response) • Fc • Type Reference DSP Ch 14 & 21
Other numerical methods • Cross correlation • A pearson correlation of two signal that shows you what the signals have in common • This is used a lot in EMG and we will talk more about it then • Spike triggered averaging • Counting the number of spikes in a given time interval • This is used a lot in neurophysiology for motor neuron activity
Next Lecture • Transducers and Force measurement • Review of analog to digital conversion (lab 2) • Quantization
Group Work • Answer the following question: If a low pass filter is employed at exactly half the sampling rate ( e.g. low pass filter data acquired at 50 Hz Fc set at 25Hz) does this introduce errors into the signal?
Group Work • Find examples of cross correlation and spike triggered averaging on the internet • Be able to draw and describe what each process is to the other group • Open for discussion