Signal acquisition
A/D conversion
Sampling rate Nyquist-Shannon sampling theorem: If bandlimited signal x(f) holds in [-B;B], then if f s = 1 / T 2B then x k x(t), meaning: the original signal can be transformed from the digitised signal without loss of information
Quantization Ensures the adequate resolution of the digitised signal E.g.: 32bit ADC 2 32 = 4.3*10 9 level coupled to a ±5V range this means a 0.23 nV precision! (16bit levels 0.15mV precision) ( 8bit 256 levels 39mV precision)
Noise 1 – Quantization errors Quantization = rounding measurement error Error limit: | |< / 2 (in case of linear q.) Amplitude constantly between two q. levels: granular noise Signal elevation faster than q. levels could follow: overload noise
Noise 2 – 50/60Hz Electric field perpendicular Magnetic field
Noise 3 – Different ranges
Noise reduction - averaging Very efficient on time-locked signals (EP epochs, spikes, oscillations,…)
Noise reduction - averaging Cutting a signal to epochs locked to a phenomenon (peaks, stimuli, …) Threshold specification manual semi- and fully-automatic (by mean & SD) Peak detection threshold crossing (simple&fast but misleading) detection of local maxima (minima)
Noise reduction - averaging
Noise reduction via spectral analysis
Fast Fourier transform: efficient algorithm to compute the discrete Fourier transform and its inverse: TIME DOMAIN FREQ. DOMAIN Signal: FFT i-FFT
Noise reduction via spectral analysis Noise reduction by filtering in the freq. domain: SIGNALFFT Noisy : Filtered :
Additional analyses in the frequency domain Periodogram Spectrogram Original signal
Seeking correlation between two datasets Peri-Event Time Histogram ECG Blood flow