This document provides a tutorial on digitizing analogue signals. It discusses the key principles of analogue-to-digital conversion including sampling rate, aliasing, anti-alias filtering, and quantization. The main points are:
1) The sampling rate must be greater than twice the maximum frequency present in the signal to avoid aliasing. Aliasing occurs when the sampling rate is too low and different frequencies become ambiguous.
2) An anti-alias filter should be applied before sampling to limit the signal to below half the sampling rate and prevent aliasing.
3) The digitized signal results in discrete amplitude levels due to quantization from the finite number of levels in the analog-to-digital converter. Higher
This document discusses several digital signal processing applications:
1) A two-band digital crossover system that splits an audio signal into low and high frequencies to be played through different speakers.
2) An ECG system that uses notch filters to remove 60Hz interference from power lines and allow detection of heart rate.
3) Speech noise reduction and coding systems that compress speech signals for transmission.
4) The compact disc recording and playback system which uses anti-aliasing filters, sampling, quantization, encoding, and laser etching to store digital audio that is then reconstructed through decoding, interpolation, DAC, and filtering.
The document reports on a digital signal processing project to design and compare different types of FIR and IIR filters. FIR filters designed include a least squares filter and an equiripple filter. IIR filters designed include a Butterworth filter and a Chebyshev type I filter. The filters were designed to meet bandpass specifications and their magnitude and phase responses were analyzed. The filters were also compared in terms of their bit error rate (BER) performance under different signal conditions such as clean, interference, noise, and both interference and noise. The Chebyshev type I filter was found to have a higher BER under noise conditions compared to the other filters.
This document summarizes a digital signal processing project that involves resampling audio signals and modeling signals using autoregressive (AR) processes.
The resampling part involves downsampling two audio signals with correct and incorrect sampling rate conversions. Graphs and analysis show the resampled signals have lower quality and more distortion compared to the originals.
The AR modeling part estimates AR model coefficients from one of the signals using the Yule-Walker equations. A filter is designed to "whiten" the signal, removing noise. Graphs and audio comparison show the filtered signal has less noise but also some quality loss.
This presentation covers noise performance of Continuous wave modulation systems; It explains modelling of white noise , noise figure of DSB-SC, SSB, AM, FM system
This document compares analog and digital filters for use in data acquisition systems. It discusses how analog filters can remove noise from signals before analog-to-digital conversion, while digital filters act after conversion. The document then defines key parameters for designing analog low-pass filters, including cutoff frequency, stop band frequency, maximum gain, and filter order. It explains how these parameters shape the frequency response curve.
Mitigation of Noise in OFDM Based Plc System Using Filter Kernel DesignIJERA Editor
Power line communication is a technology that transforms power line in to pathway for conveyance of
broadband data. It is cost less than other communication approach and for better bandwidth efficiency OFDM
based PLC system is used. In real PLC environment some electrical appliances will produce noise. To mitigate
this noise filter kernel design is used, so periodic impulsive noise and Gaussian noises are removed from PLC
communication system by using this filter kernel design. MATLAB is used for the simulation and the result
shows that filter kernel is simple and effective noise mitigation technique. Further in future, interference due to
obstacles also wants to be mitigated for the better data transmission without noise.
ECG Signal Denoising using Digital Filter and Adaptive FilterIRJET Journal
1. The document discusses methods for denoising electrocardiogram (ECG) signals, including digital filters and adaptive filters.
2. It evaluates the performance of Savitzky-Golay filters, band pass filters, and adaptive noise cancellation techniques for removing noise from ECG signals and improving the signal-to-noise ratio.
3. The key filters discussed are Savitzky-Golay filters, Tompkins filters, Butterworth band pass filters, and least mean square adaptive filters, analyzing their ability to reduce noise like powerline interference, baseline drift, and motion artifacts from ECG data.
1) Noise exists in all communication systems and degrades signal quality. It is caused by random movement of electrons and can be internal or external.
2) Thermal noise, also known as Johnson noise, is generated by thermal agitation of electrons in conductors. It is proportional to temperature and bandwidth.
3) Noise figure and noise temperature are used to measure the degradation of signal to noise ratio caused by components in a communication system. Lower noise figure and temperature indicate less degradation.
This document discusses several digital signal processing applications:
1) A two-band digital crossover system that splits an audio signal into low and high frequencies to be played through different speakers.
2) An ECG system that uses notch filters to remove 60Hz interference from power lines and allow detection of heart rate.
3) Speech noise reduction and coding systems that compress speech signals for transmission.
4) The compact disc recording and playback system which uses anti-aliasing filters, sampling, quantization, encoding, and laser etching to store digital audio that is then reconstructed through decoding, interpolation, DAC, and filtering.
The document reports on a digital signal processing project to design and compare different types of FIR and IIR filters. FIR filters designed include a least squares filter and an equiripple filter. IIR filters designed include a Butterworth filter and a Chebyshev type I filter. The filters were designed to meet bandpass specifications and their magnitude and phase responses were analyzed. The filters were also compared in terms of their bit error rate (BER) performance under different signal conditions such as clean, interference, noise, and both interference and noise. The Chebyshev type I filter was found to have a higher BER under noise conditions compared to the other filters.
This document summarizes a digital signal processing project that involves resampling audio signals and modeling signals using autoregressive (AR) processes.
The resampling part involves downsampling two audio signals with correct and incorrect sampling rate conversions. Graphs and analysis show the resampled signals have lower quality and more distortion compared to the originals.
The AR modeling part estimates AR model coefficients from one of the signals using the Yule-Walker equations. A filter is designed to "whiten" the signal, removing noise. Graphs and audio comparison show the filtered signal has less noise but also some quality loss.
This presentation covers noise performance of Continuous wave modulation systems; It explains modelling of white noise , noise figure of DSB-SC, SSB, AM, FM system
This document compares analog and digital filters for use in data acquisition systems. It discusses how analog filters can remove noise from signals before analog-to-digital conversion, while digital filters act after conversion. The document then defines key parameters for designing analog low-pass filters, including cutoff frequency, stop band frequency, maximum gain, and filter order. It explains how these parameters shape the frequency response curve.
Mitigation of Noise in OFDM Based Plc System Using Filter Kernel DesignIJERA Editor
Power line communication is a technology that transforms power line in to pathway for conveyance of
broadband data. It is cost less than other communication approach and for better bandwidth efficiency OFDM
based PLC system is used. In real PLC environment some electrical appliances will produce noise. To mitigate
this noise filter kernel design is used, so periodic impulsive noise and Gaussian noises are removed from PLC
communication system by using this filter kernel design. MATLAB is used for the simulation and the result
shows that filter kernel is simple and effective noise mitigation technique. Further in future, interference due to
obstacles also wants to be mitigated for the better data transmission without noise.
ECG Signal Denoising using Digital Filter and Adaptive FilterIRJET Journal
1. The document discusses methods for denoising electrocardiogram (ECG) signals, including digital filters and adaptive filters.
2. It evaluates the performance of Savitzky-Golay filters, band pass filters, and adaptive noise cancellation techniques for removing noise from ECG signals and improving the signal-to-noise ratio.
3. The key filters discussed are Savitzky-Golay filters, Tompkins filters, Butterworth band pass filters, and least mean square adaptive filters, analyzing their ability to reduce noise like powerline interference, baseline drift, and motion artifacts from ECG data.
1) Noise exists in all communication systems and degrades signal quality. It is caused by random movement of electrons and can be internal or external.
2) Thermal noise, also known as Johnson noise, is generated by thermal agitation of electrons in conductors. It is proportional to temperature and bandwidth.
3) Noise figure and noise temperature are used to measure the degradation of signal to noise ratio caused by components in a communication system. Lower noise figure and temperature indicate less degradation.
Application of adaptive linear equalizerSayahnarahul
This document discusses various applications of adaptive linear equalizers including: system identification, linear prediction, inverse modeling, jammer suppression, adaptive notch filtering, noise cancellation, echo cancellation in voice/data communications, fetal monitoring, ocular artifact removal from EEGs, and noise cancellation in AC electrical measurements. Adaptive linear equalizers are used across many domains including telecommunications, radar, sonar, video/audio processing, and noise cancellation to adapt filter coefficients over time to compensate for changes in systems and optimize signal recovery/interference rejection.
Okay, here are the steps to find the velocity of the rocket at t = 10s:
1) Take the derivative of the position function x(t) to get the velocity function v(t):
v(t) = 4 + 14t + 15t^2 - 1.4t^3
2) Plug t = 10s into the velocity function:
v(10) = 4 + 140 + 1500 - 140 = 1404 m/s
So the velocity of the rocket at t = 10s is 1404 m/s.
Performance Evaluation of Different Thresholding Method for De-Noising of Vib...IJERA Editor
De-noising of the raw vibration signal is essential requirement to improve the accuracy and efficiency of any fault diagnosis of method. In many cases the noise signal is even stronger than the actual signal, so it is important to have such system in which noise elimination can be done effectively, there are many time domain and frequency domain methods are already available, where use of wavelet as time-frequency domain method in the field of de-noising the vibration signal is relatively new, it gives multi resolution analysis in both is time-frequency domain. In this paper various conventional thresholding methods based on discrete wavelet transform are compared with adaptive thresholding method and Penalized thresholding method for the de-noising of vibration signal of rotating machine. Signal to noise ratio (SNR), root mean square error (RMSE) in between de-noised signal with original signal are used as an indicator for selecting the effective thesholding method.
Acoustic echo cancellation using nlms adaptive algorithm ranbeerRanbeer Tyagi
The document discusses acoustic echo cancellation using the NLMS adaptive algorithm. It introduces the acoustic echo problem in hands-free communication systems and how echo cancellation works by using an adaptive filter to generate an echo replica that is subtracted from the echo signal. It then describes the NLMS adaptive algorithm and how it offers improved convergence over LMS with low computational complexity. Simulation results show NLMS effectively cancels echo. Future work topics are enhancing performance in noisy and double-talk conditions.
Echo Cancellation Algorithms using Adaptive Filters: A Comparative Studyidescitation
An adaptive filter is a filter that self-adjusts its transfer function according to an
optimization algorithm driven by an error signal. Adaptive filter finds its essence in
applications such as echo cancellation, noise cancellation, system identification and many
others. This paper briefly discusses LMS, NLMS and RLS adaptive filter algorithms for
echo cancellation. For the analysis, an acoustic echo canceller is built using LMS, NLMS
and RLS algorithms and the echo cancelled samples are studied using Spectrogram. The
analysis is further extended with its cross-correlation and ERLE (Echo Return Loss
Enhancement) results. Finally, this paper concludes with a better adaptive filter algorithm
for Echo cancellation. The implementation and analysis is done using MATLAB®,
SIMULINK® and SPECTROGRAM V5.0®.
The document discusses active noise cancellation and noise reduction techniques. It describes how active noise cancellation works by generating a sound wave with equal amplitude but opposite phase to the original noise, cancelling it out. Adaptive filters are used, with algorithms like LMS and RLS, to analyze input sounds and adjust filter coefficients to minimize noise. Applications include headphones, vehicles, aircraft, and noise-cancelling devices that can reduce ambient sounds.
Lock-in amplifiers use phase-sensitive detection to isolate signals at a specific reference frequency, even when obscured by noise much larger than the signal. They multiply the input signal with an internal reference signal that is phase-locked to an external reference, extracting the component that matches the reference frequency as a DC output. This allows accurate measurement of nanovolt-level signals. Digital lock-ins implement phase-sensitive detection through digital multiplication of digitized input and reference signals, avoiding issues like harmonic detection that can occur in analog implementations.
echo types, how to cancel echo in each type, which is more complex, echo cancellation implementation in matlab
prepared by : OLA MASHAQI ,, SUHAD MALAYSHE
The document discusses different types of noise that affect measurement accuracy, including thermal noise, shot noise, flicker noise, and interference. It explains how noise arises from random fluctuations in instruments and how it is characterized by its peak-to-peak amplitude and root-mean-square value. The key point is that the ratio of the signal to noise determines measurement ability, and factors like cooling, filtering, and modulation can be used to reduce noise and improve this ratio.
Research: Applying Various DSP-Related Techniques for Robust Recognition of A...Roman Atachiants
This paper approaches speaker recognition in a new way. A speaker recognition system has been realized that works on adult and child speakers, both male and female. Furthermore, the system employs text-dependent and text-independent algorithms, which makes robust speaker recognition possible in many applications. Single-speaker classication is achieved by age/sex pre-classication and is implemented using classic text-dependent techniques, as well as a novel technology for text-independent recognition. This new research uses Evolutionary Stable Strategies to model human speech and allows speaker recognition by analyzing just one vowel.
The document discusses noise in analog communication systems. It defines noise as an unwanted signal that affects the wanted signal. There are two main categories of noise: interference from human sources and naturally occurring random noise. Noise is generated internally in communication equipment and externally from environmental changes. Noise corrupts signals and degrades the signal-to-noise ratio, leading to bit errors in digital communication. The bit error rate and bit error probability are used to characterize the noise performance of a system. Additive noise is the combination of all noises added to the signal as it travels through the communication channel.
Chebyshev filters are analog or digital filters having a steeper roll-off and more passband ripple (type I) or stop
band ripple (type II) than Butterworth filters. Chebyshev filters have the property that they minimize the error
between the idealized and the actual filter characteristic over the range of the filter,[citation needed] but with
ripples in the pass band. This type of filter is named after Pafnuty Chebyshev because its mathematical
characteristics are derived from Chebyshev polynomials.
Noise cancellation and suppression techniques use signal processing to extract useful information from a mixture of signals. Noise cancellation works by passing a corrupted signal through a filter to suppress noise while preserving the original signal. Active noise cancellation systems emit an inverted phase signal to destructively interfere with and cancel out the original noise signal. Noise can be modeled as white noise, colored noise, impulsive noise, or acoustic echo. White noise suppression techniques use quantization and filter banks. Impulsive noise compensation uses non-linear limiting. Acoustic echo cancellation uses subband analysis and synthesis with adaptation to model and subtract the echo from microphone signals.
Acoustic fMRI noise reduction: a perceived loudness approachDimitri Vrehen
This document discusses a study that measured the subjective loudness of acoustic noise from fMRI scanners. The study recorded noise from three echo planar imaging sequences on a 3 Tesla MRI scanner. In a psychophysical experiment with 9 subjects, the perceived loudness of the fMRI noise did not increase linearly with sound pressure level. Noises with lower damping factors and frequencies in the 2.5-6kHz range of ear sensitivity were perceived as louder. EPI sequences with suppressed frequencies in the ear's most sensitive range and a highly impulsive nature distributed over longer times should reduce perceived loudness of fMRI acoustic noise.
Design and Implementation of Polyphase based Subband Adaptive Structure for N...Pratik Ghotkar
With the tremendous growth in the Digital Signal processing technology, there are many techniques available to remove noise from the speech signals which is used in the speech processing. Widely used LMS algorithm is modified with much advancement but still there are many limitations are introducing. This paper consist of a new approach i.e. subband adaptive processing for noise cancelation in the speech signals. Subband processing employs the multirate signal processing. The polyphase based subband adaptive implementation finds better results in term of MMSE , PSNR and processing time; also the synthesis filter bank is works on the lower data rate which reduces the computational Burdon as compare to the direct implementation of Subband adaptive filter. The normalized least mean squares (NLMS) algorithm is a class of adaptive filter used.
This document discusses moving average filters and their properties. It begins by defining the moving average filter equation and explaining that it operates by averaging neighboring points in the input signal. While simple, the moving average filter is optimal for reducing random noise while maintaining a sharp step response. It has poor performance in the frequency domain, however, with a slow roll-off and inability to separate frequencies. Relatives like multiple-pass moving average filters have slightly better frequency response at the cost of increased computation. The document provides examples and equations to illustrate the properties of moving average filters.
High-Sensitivity HydrophoneBased on Fiber Grating Laser And Acorrugated Diaph...IJRESJOURNAL
ABSTRACT: In this work, we present afiber optic hydrophones based on dual-frequency fiber grating lasers and a corrugated diaphragm. The laser is employed as sensing element and an elastic corrugated diaphragm is used to translate acoustic pressure P intolateral point loadNon the laser cavity. Experimental result shows the fiber laser hydrophone has a working bandwidth over 1 kHz with sub100 μPa/Hz1/2minimum detectable pressure at 1 kHz
PERFORMANCE EVALUATION OF COMPUTED TOMOGRAPHY (CT) SCANNERSBhuvi palaniswamy
This document discusses performance evaluation tests for computed tomography (CT) scanners. It can be broadly classified into electromechanical tests, x-ray generator (electrical) tests, image quality tests, radiation dose tests, and general tests related to CT number. Electromechanical tests evaluate the scan localization laser lights, table movement accuracy, and gantry tilt. X-ray generator tests check the accelerating voltage, milliampere linearity, and radiation output reproducibility. Image quality tests evaluate low-contrast resolution, high-contrast resolution, and noise. Radiation dose tests measure the computed tomography dose index and multiple scan average dose using phantoms.
This document discusses band pass filters that are constructed using a combination of low pass and high pass filters integrated with operational amplifiers. It provides equations to calculate key filter parameters like resonance frequency, bandwidth, and cutoff frequencies based on resistor and capacitor values. Simulation results show that as the near infrared operating wavelength increases, the filter capacitance increases, resistance decreases, resonance frequency and cutoff frequencies decrease, and gain increases. The document examines these filter characteristics over a wide range of parameters.
This project simulated sending an image through a communication channel using bipolar pulse amplitude modulation. Two pulse shaping functions - half sine and root raised cosine - were used to modulate the data. The modulated signal then passed through a channel with echoes, adding inter-symbol interference. Noise was added and the signal passed through a matched filter and equalization filters. The minimum mean square error filter performed better than the zero forcing filter, recovering images up to 2dB lower signal-to-noise ratio. Testing with additional indoor and outdoor channels showed expected performance of the filters.
This document discusses signal conversion systems for capturing and digitizing biomedical signals. It covers sampling theory, including the Nyquist sampling theorem. A typical analog-to-digital conversion system is described, involving sensors, amplifiers, filters, a sample-and-hold circuit and ADC. Requirements for converting biomedical signals include high accuracy, appropriate sampling rate, gain, speed, low power and small size. Common circuit elements for analog-digital and digital-analog conversion are also outlined.
This document discusses analog to digital conversion techniques. It explains that analog signals are continuous while digital signals are discrete. It also describes three main techniques in A/D conversion: sampling, quantization, and encoding. Sampling converts a continuous signal to discrete samples. Quantization maps samples to a smaller set of values, introducing quantization error. Encoding converts quantized values into a binary format.
Application of adaptive linear equalizerSayahnarahul
This document discusses various applications of adaptive linear equalizers including: system identification, linear prediction, inverse modeling, jammer suppression, adaptive notch filtering, noise cancellation, echo cancellation in voice/data communications, fetal monitoring, ocular artifact removal from EEGs, and noise cancellation in AC electrical measurements. Adaptive linear equalizers are used across many domains including telecommunications, radar, sonar, video/audio processing, and noise cancellation to adapt filter coefficients over time to compensate for changes in systems and optimize signal recovery/interference rejection.
Okay, here are the steps to find the velocity of the rocket at t = 10s:
1) Take the derivative of the position function x(t) to get the velocity function v(t):
v(t) = 4 + 14t + 15t^2 - 1.4t^3
2) Plug t = 10s into the velocity function:
v(10) = 4 + 140 + 1500 - 140 = 1404 m/s
So the velocity of the rocket at t = 10s is 1404 m/s.
Performance Evaluation of Different Thresholding Method for De-Noising of Vib...IJERA Editor
De-noising of the raw vibration signal is essential requirement to improve the accuracy and efficiency of any fault diagnosis of method. In many cases the noise signal is even stronger than the actual signal, so it is important to have such system in which noise elimination can be done effectively, there are many time domain and frequency domain methods are already available, where use of wavelet as time-frequency domain method in the field of de-noising the vibration signal is relatively new, it gives multi resolution analysis in both is time-frequency domain. In this paper various conventional thresholding methods based on discrete wavelet transform are compared with adaptive thresholding method and Penalized thresholding method for the de-noising of vibration signal of rotating machine. Signal to noise ratio (SNR), root mean square error (RMSE) in between de-noised signal with original signal are used as an indicator for selecting the effective thesholding method.
Acoustic echo cancellation using nlms adaptive algorithm ranbeerRanbeer Tyagi
The document discusses acoustic echo cancellation using the NLMS adaptive algorithm. It introduces the acoustic echo problem in hands-free communication systems and how echo cancellation works by using an adaptive filter to generate an echo replica that is subtracted from the echo signal. It then describes the NLMS adaptive algorithm and how it offers improved convergence over LMS with low computational complexity. Simulation results show NLMS effectively cancels echo. Future work topics are enhancing performance in noisy and double-talk conditions.
Echo Cancellation Algorithms using Adaptive Filters: A Comparative Studyidescitation
An adaptive filter is a filter that self-adjusts its transfer function according to an
optimization algorithm driven by an error signal. Adaptive filter finds its essence in
applications such as echo cancellation, noise cancellation, system identification and many
others. This paper briefly discusses LMS, NLMS and RLS adaptive filter algorithms for
echo cancellation. For the analysis, an acoustic echo canceller is built using LMS, NLMS
and RLS algorithms and the echo cancelled samples are studied using Spectrogram. The
analysis is further extended with its cross-correlation and ERLE (Echo Return Loss
Enhancement) results. Finally, this paper concludes with a better adaptive filter algorithm
for Echo cancellation. The implementation and analysis is done using MATLAB®,
SIMULINK® and SPECTROGRAM V5.0®.
The document discusses active noise cancellation and noise reduction techniques. It describes how active noise cancellation works by generating a sound wave with equal amplitude but opposite phase to the original noise, cancelling it out. Adaptive filters are used, with algorithms like LMS and RLS, to analyze input sounds and adjust filter coefficients to minimize noise. Applications include headphones, vehicles, aircraft, and noise-cancelling devices that can reduce ambient sounds.
Lock-in amplifiers use phase-sensitive detection to isolate signals at a specific reference frequency, even when obscured by noise much larger than the signal. They multiply the input signal with an internal reference signal that is phase-locked to an external reference, extracting the component that matches the reference frequency as a DC output. This allows accurate measurement of nanovolt-level signals. Digital lock-ins implement phase-sensitive detection through digital multiplication of digitized input and reference signals, avoiding issues like harmonic detection that can occur in analog implementations.
echo types, how to cancel echo in each type, which is more complex, echo cancellation implementation in matlab
prepared by : OLA MASHAQI ,, SUHAD MALAYSHE
The document discusses different types of noise that affect measurement accuracy, including thermal noise, shot noise, flicker noise, and interference. It explains how noise arises from random fluctuations in instruments and how it is characterized by its peak-to-peak amplitude and root-mean-square value. The key point is that the ratio of the signal to noise determines measurement ability, and factors like cooling, filtering, and modulation can be used to reduce noise and improve this ratio.
Research: Applying Various DSP-Related Techniques for Robust Recognition of A...Roman Atachiants
This paper approaches speaker recognition in a new way. A speaker recognition system has been realized that works on adult and child speakers, both male and female. Furthermore, the system employs text-dependent and text-independent algorithms, which makes robust speaker recognition possible in many applications. Single-speaker classication is achieved by age/sex pre-classication and is implemented using classic text-dependent techniques, as well as a novel technology for text-independent recognition. This new research uses Evolutionary Stable Strategies to model human speech and allows speaker recognition by analyzing just one vowel.
The document discusses noise in analog communication systems. It defines noise as an unwanted signal that affects the wanted signal. There are two main categories of noise: interference from human sources and naturally occurring random noise. Noise is generated internally in communication equipment and externally from environmental changes. Noise corrupts signals and degrades the signal-to-noise ratio, leading to bit errors in digital communication. The bit error rate and bit error probability are used to characterize the noise performance of a system. Additive noise is the combination of all noises added to the signal as it travels through the communication channel.
Chebyshev filters are analog or digital filters having a steeper roll-off and more passband ripple (type I) or stop
band ripple (type II) than Butterworth filters. Chebyshev filters have the property that they minimize the error
between the idealized and the actual filter characteristic over the range of the filter,[citation needed] but with
ripples in the pass band. This type of filter is named after Pafnuty Chebyshev because its mathematical
characteristics are derived from Chebyshev polynomials.
Noise cancellation and suppression techniques use signal processing to extract useful information from a mixture of signals. Noise cancellation works by passing a corrupted signal through a filter to suppress noise while preserving the original signal. Active noise cancellation systems emit an inverted phase signal to destructively interfere with and cancel out the original noise signal. Noise can be modeled as white noise, colored noise, impulsive noise, or acoustic echo. White noise suppression techniques use quantization and filter banks. Impulsive noise compensation uses non-linear limiting. Acoustic echo cancellation uses subband analysis and synthesis with adaptation to model and subtract the echo from microphone signals.
Acoustic fMRI noise reduction: a perceived loudness approachDimitri Vrehen
This document discusses a study that measured the subjective loudness of acoustic noise from fMRI scanners. The study recorded noise from three echo planar imaging sequences on a 3 Tesla MRI scanner. In a psychophysical experiment with 9 subjects, the perceived loudness of the fMRI noise did not increase linearly with sound pressure level. Noises with lower damping factors and frequencies in the 2.5-6kHz range of ear sensitivity were perceived as louder. EPI sequences with suppressed frequencies in the ear's most sensitive range and a highly impulsive nature distributed over longer times should reduce perceived loudness of fMRI acoustic noise.
Design and Implementation of Polyphase based Subband Adaptive Structure for N...Pratik Ghotkar
With the tremendous growth in the Digital Signal processing technology, there are many techniques available to remove noise from the speech signals which is used in the speech processing. Widely used LMS algorithm is modified with much advancement but still there are many limitations are introducing. This paper consist of a new approach i.e. subband adaptive processing for noise cancelation in the speech signals. Subband processing employs the multirate signal processing. The polyphase based subband adaptive implementation finds better results in term of MMSE , PSNR and processing time; also the synthesis filter bank is works on the lower data rate which reduces the computational Burdon as compare to the direct implementation of Subband adaptive filter. The normalized least mean squares (NLMS) algorithm is a class of adaptive filter used.
This document discusses moving average filters and their properties. It begins by defining the moving average filter equation and explaining that it operates by averaging neighboring points in the input signal. While simple, the moving average filter is optimal for reducing random noise while maintaining a sharp step response. It has poor performance in the frequency domain, however, with a slow roll-off and inability to separate frequencies. Relatives like multiple-pass moving average filters have slightly better frequency response at the cost of increased computation. The document provides examples and equations to illustrate the properties of moving average filters.
High-Sensitivity HydrophoneBased on Fiber Grating Laser And Acorrugated Diaph...IJRESJOURNAL
ABSTRACT: In this work, we present afiber optic hydrophones based on dual-frequency fiber grating lasers and a corrugated diaphragm. The laser is employed as sensing element and an elastic corrugated diaphragm is used to translate acoustic pressure P intolateral point loadNon the laser cavity. Experimental result shows the fiber laser hydrophone has a working bandwidth over 1 kHz with sub100 μPa/Hz1/2minimum detectable pressure at 1 kHz
PERFORMANCE EVALUATION OF COMPUTED TOMOGRAPHY (CT) SCANNERSBhuvi palaniswamy
This document discusses performance evaluation tests for computed tomography (CT) scanners. It can be broadly classified into electromechanical tests, x-ray generator (electrical) tests, image quality tests, radiation dose tests, and general tests related to CT number. Electromechanical tests evaluate the scan localization laser lights, table movement accuracy, and gantry tilt. X-ray generator tests check the accelerating voltage, milliampere linearity, and radiation output reproducibility. Image quality tests evaluate low-contrast resolution, high-contrast resolution, and noise. Radiation dose tests measure the computed tomography dose index and multiple scan average dose using phantoms.
This document discusses band pass filters that are constructed using a combination of low pass and high pass filters integrated with operational amplifiers. It provides equations to calculate key filter parameters like resonance frequency, bandwidth, and cutoff frequencies based on resistor and capacitor values. Simulation results show that as the near infrared operating wavelength increases, the filter capacitance increases, resistance decreases, resonance frequency and cutoff frequencies decrease, and gain increases. The document examines these filter characteristics over a wide range of parameters.
This project simulated sending an image through a communication channel using bipolar pulse amplitude modulation. Two pulse shaping functions - half sine and root raised cosine - were used to modulate the data. The modulated signal then passed through a channel with echoes, adding inter-symbol interference. Noise was added and the signal passed through a matched filter and equalization filters. The minimum mean square error filter performed better than the zero forcing filter, recovering images up to 2dB lower signal-to-noise ratio. Testing with additional indoor and outdoor channels showed expected performance of the filters.
This document discusses signal conversion systems for capturing and digitizing biomedical signals. It covers sampling theory, including the Nyquist sampling theorem. A typical analog-to-digital conversion system is described, involving sensors, amplifiers, filters, a sample-and-hold circuit and ADC. Requirements for converting biomedical signals include high accuracy, appropriate sampling rate, gain, speed, low power and small size. Common circuit elements for analog-digital and digital-analog conversion are also outlined.
This document discusses analog to digital conversion techniques. It explains that analog signals are continuous while digital signals are discrete. It also describes three main techniques in A/D conversion: sampling, quantization, and encoding. Sampling converts a continuous signal to discrete samples. Quantization maps samples to a smaller set of values, introducing quantization error. Encoding converts quantized values into a binary format.
This document discusses digital-to-analog conversion (DAC) and analog-to-digital conversion (ADC). It covers key topics such as signal quantization, sampling theory, aliasing, and reconstruction of signals from sampled data. Proper sampling requires a sampling frequency of at least twice the highest frequency component of the signal to avoid aliasing. DACs reconstruct the analog signal from its digital representation, but introduce a zeroth-order hold effect that can be corrected through filtering. Understanding ADC and DAC, including their limitations and filter requirements, is important for digital signal processing applications.
This digital method is built using chirp z-transform(CZT) and provides 100% alias-free bandwidth such as using ideal LPF. This noble method is efficient for economic and practical considerations.
The document discusses digital communication systems and their advantages over analog communication. It describes how analog signals are digitized using sampling and quantization. Sampling must occur at least twice the maximum frequency of the signal to avoid aliasing, as stated by the Nyquist sampling theorem. Quantization converts continuous amplitudes to discrete levels, causing quantization error and noise. Digital communication provides benefits like lower distortion and easier signal processing. Companding is discussed as a technique used in pulse code modulation to improve signal-to-noise ratio by compressing higher amplitudes before transmission.
The document discusses several medical applications of digital signal processing (DSP) including hearing aids, electroencephalograms (EEGs), and acquiring blood pressure signals. DSP techniques such as sampling, filtering, frequency analysis, and spectral estimation are used to process analog signals from the body, like brain waves or sound, into digital signals. This allows signals to be filtered and analyzed to extract clinically useful information for diagnosing conditions and monitoring patients.
This section discusses two techniques for analog to digital conversion: pulse code modulation (PCM) and delta modulation. PCM involves sampling an analog signal, quantizing the sample amplitudes into discrete levels, and encoding the levels into binary digits. The sampling rate must be at least twice the highest signal frequency according to the Nyquist theorem. Delta modulation encodes the difference between samples rather than their absolute values, using one bit per sample. It has a simpler design than PCM but can result in larger errors for signals with large amplitude changes between samples.
1) The document discusses various pulse modulation techniques including pulse amplitude modulation (PAM), pulse width modulation (PWM), and pulse position modulation (PPM).
2) It provides details on the generation and detection of PAM and PWM signals, explaining the use of sampling, comparators, sawtooth waves, and filters.
3) The document compares different sampling techniques for PAM including natural sampling, flat top sampling, and discusses the need for analog to digital conversion in communication systems.
The document discusses the sampling theorem, which states that a signal must be sampled at a rate at least twice the highest frequency present in the signal (fs > 2fa(max)) in order to perfectly reconstruct the original signal from the samples. If the sampling rate is lower than this Nyquist rate, aliasing distortion will occur as frequency components fold over each other. Digital transmission is preferable to analog as the signal is more robust to noise and can be easily recovered, corrected, and amplified. There are three main sampling methods: ideal sampling uses an impulse, natural sampling uses a short pulse, and flat top sampling "samples and holds" a single amplitude value.
The document discusses sampling a signal using an impulse train. It introduces the impulse train as a theoretical concept consisting of a series of narrow spikes that match the original signal at sampling instants. This allows making an "apples-to-apples" comparison between the original analog signal and the sampled signal. The Fourier transform of the impulse train is a train of Dirac delta functions. Sampling a signal is equivalent to multiplying it with the impulse train. The Fourier transform of the sampled signal is equal to the original Fourier transform multiplied by the Fourier transform of the impulse train.
Ch4 1 Data communication and networking by neha g. kuraleNeha Kurale
This document discusses analog-to-digital conversion techniques, specifically pulse code modulation (PCM) and delta modulation. PCM consists of sampling an analog signal, quantizing the sample amplitudes into discrete levels, and encoding the levels into binary codes. The sampling rate must be at least twice the highest frequency in the signal according to the Nyquist theorem. Quantization introduces error but more levels reduce the error. Delta modulation encodes changes in signal amplitude rather than absolute levels. Serial transmission can be asynchronous, synchronous, or isochronous depending on whether start/stop bits are used and if gaps between frames are of fixed duration.
The document discusses the basics of signal processing in polysomnography (PSG) systems from the physiological signals being recorded to their digital representation. It covers topics like: where EEG signals originate from neurons, the analog components like electrodes, amplifiers and filters used to process analog signals, and the digital components like sampling rate and resolution that determine the digital waveform display. The key stages of signal processing from the patient to the PSG tracing are also outlined.
4. Analog to digital conversation (1).ppttest22333
This document discusses digital transmission techniques, including analog-to-digital conversion methods like pulse code modulation (PCM) and delta modulation. It explains the key steps in PCM - sampling, quantization, and binary encoding. It also covers important concepts like the Nyquist sampling theorem and quantization error. Delta modulation is introduced as an alternative that transmits only the differences between signal pulses. The document concludes by describing parallel, asynchronous, synchronous, and isochronous transmission modes.
This document discusses digital transmission techniques, including analog-to-digital conversion methods like pulse code modulation (PCM) and delta modulation. It explains the key steps in PCM - sampling, quantization, and binary encoding. It also covers important concepts like the Nyquist sampling theorem and quantization error. Delta modulation is introduced as an alternative that transmits only the differences between signal pulses. The document concludes by describing parallel, asynchronous, synchronous, and isochronous transmission modes for sending digital data across a link.
- Digital signal processing systems convert analog signals to digital signals for processing. They consist of anti-aliasing filters, analog-to-digital converters (ADCs), digital signal processors, digital-to-analog converters (DACs), and reconstruction filters.
- ADCs sample analog signals and convert them to discrete digital values. Sampling must occur at least twice the maximum frequency of the analog signal, as per the Nyquist-Shannon sampling theorem, to avoid aliasing.
- Aliasing occurs when the sampling rate is too low, causing high frequency signals to appear as lower frequencies. Anti-aliasing filters are used before sampling to remove frequencies above half the sampling rate.
The document discusses aliasing in signal processing and sampling. It defines aliasing as distortion that occurs when the sampling rate is below the minimum required rate. Aliasing causes high frequency signals to appear as lower frequencies. To prevent aliasing, signals must be low-pass filtered before sampling to remove frequencies above half the sampling rate. The sampling rate must be at least twice the highest frequency component in the original signal as per the Nyquist sampling theorem. Oversampling can help reduce demands on anti-aliasing filters.
This document discusses digital transmission through analog-to-digital conversion techniques like pulse code modulation (PCM) and delta modulation. It describes the key steps of PCM: sampling the analog signal at regular intervals, quantizing the sampled signal amplitudes into discrete levels, and encoding the quantized levels into binary code. The sampling rate must be at least twice the highest signal frequency to avoid aliasing, as per the Nyquist theorem. Quantization introduces approximation errors that can be reduced by using more levels, though this increases the required bit rate. PCM allows digital transmission of signals in a noise-robust way, though it requires more bandwidth than direct analog transmission.
DSP_2018_FOEHU - Lec 02 - Sampling of Continuous Time SignalsAmr E. Mohamed
The document discusses sampling of continuous-time signals. It defines different types of signals and sampling methods. Ideal sampling involves multiplying the signal by a train of impulse functions to select sample values at regular intervals. For practical sampling, a train of rectangular pulses is used to approximate ideal sampling. Flat-top sampling is achieved by convolving the ideally sampled signal with a rectangular pulse, resulting in samples held at a constant height for the sample period. The Nyquist sampling theorem states that a signal must be sampled at least twice its maximum frequency to avoid aliasing when reconstructing the original signal from samples. An anti-aliasing filter can be used before sampling to prevent aliasing from high frequencies above half the sampling rate.
Noise reduction is the process of removing noise from a signal. In this project, two audio files are given: (1) speech.au and (2) noisy_speech.au. The first file contains the original speech signal and the second one contains the noisy version of the first signal. The objective of this project is to reduce the noise from the noisy file
Echo and reverberation effects are used extensively in the music industry. Here we will design a digital filter that will create the echo and reverb effect on audio signals.
This document discusses sampling and quantization in digital communication. It introduces sampling as the process of converting a continuous-time analog signal into a discrete-time signal by taking samples at regular intervals. The sampling theorem states that if a signal is sampled at least twice the maximum frequency of the signal, it can be reconstructed without distortion. Quantization is the process of converting the discrete-time continuous amplitude samples into discrete amplitude values. The document covers topics such as the Nyquist rate, aliasing, ideal sampling, and methods of sampling like impulse sampling.
1. SCOPE • Volume 13 • Issue 4 • DECEMBER 2004 9
Introduction
Digitizing a signal is the first step in digital signal processing
– get it wrong and all subsequent work may be wasted. We
will discuss the principles underpinning this procedure, and
some of the practical problems, and potential pitfalls along
the way. The ECG (electrocardiogram) signal will be used to
provide practical illustrations, and further examples from
other biomedical signals are given at the end of this tutorial.
In analogue-to-digital conversion, or digitizing, an analogue
signal is converted into digital format. Analogue signal have
amplitudes that are known at every
instant in time, and can take on any
value (usually limited by a minimum
and maximum determined by the
type of signal and the equipment
used to acquire it). Clearly the ECG
signal, as measured on the patient’s
chest (fig. 1, top plot) and then
amplified by the ECG system, falls
within this definition. Digital signals,
on the other hand, are given by a
sequence of numbers that
represent samples of the signals, as
illustrated in fig. 1 (bottom plot).
Digital signals can then be
processed in ‘real-time’ (or on-line,
i.e. simultaneously with acquisition),
or stored for later ‘off-line’
processing and analysis by a
computer. Essentially, any signal
processing operation that can be
performed on analogue signals can
also be carried out on the digitized
version, and in many cases more
easily, cheaply, and flexibly. There
are also a number of operations that are readily applied on
digital signals, that could not easily be implemented in
analogue form (e.g. operations that require future, as well as
past samples). Digital processing thus holds many practical
advantages, and is progressively replacing analogue
techniques.
The process of analogue-to-digital conversion involves a
number of different stages, as illustrated in fig. 2. Each of
these will now be considered, with particular attention to if
and when information is lost by sampling.
Digitizing Signals – a Short
Tutorial Guide
David M. Simpson, Antonio De Stefano
Institute of Sound and Vibration Research, University of Southampton SO17 1BJ
Tel. 023 8059 3221 e-mail: ds@isvr.soton.ac.uk
Abstract
Converting the analogue signal, as captured from a patient, into digital format is known as digitizing, or analogue to
digital conversion. This is a vital first step in for digital signal processing. The acquisition of high-quality data requires
appropriate choices of system and parameters (sampling rate, anti-alias filter, amplification, number of ‘bits’). Thus
tutorial aims to provide a practical guide to making these choices, and explains the underlying principles (rather than the
mathematical theory and proofs) and potential pitfalls. Illustrative examples from different physiological signals are
provided.
Figure 1. An ECG signal (analogue signal in the top plot), and the sampled version (digital signal,
bottom plot). This signal was sampled at a sampling rate of 100 Hz.
2. SCOPE • Volume 13 • Issue 4 • DECEMBER 200410
The sampling theorem and aliasing
The ‘sampling rate’ is defined as the number of samples
acquired, per unit time, and is usually given in samples/sec
or Hz. It is intuitively clear that at a higher sampling rate,
the digital signal provides a better approximation to the
analogue signal. It may furthermore be shown from theory
that if the sampling rate is sufficiently high, the analogue
signal can be reconstructed exactly from the samples, i.e.
there is no loss of information in the process of sampling.
The sampling theorem states that this recovery of the
analogue signal from its sampled version is possible, when
the sampling rate is greater than twice the maximum
frequency present in the signal. This provides the main
criterion for selecting the sampling rate.
THE SAMPLING RATE MUST BE GREATER THAN TWICE
THE MAXIMUM FREQUENCY PRESENT IN THE SIGNAL.
Thus, for example in fig. 1, the ECG signal had a maximum
frequency of about 40 Hz (determined by the filter settings
during acquisition, and confirmed by observing the
spectrum), and was sampled at 100 Hz (i.e. > 2*40). The
analogue signal (solid line) could therefore be reconstructed
perfectly from the samples (•)1
. It would thus also be possible
to calculate the samples corresponding to any other arbitrary
sampling rate, from the digital signal acquired at 100 Hz.
When the sampling rate is lower than required, aliasing
occurs, as illustrated in fig. 3. Consider a sine-wave of 1.5
Hz (fig. 3a), sampled at 8.5 Hz. According to the sampling
theorem, this sampling rate is adequate, and the original
sine-wave can be reconstructed from the samples. Now
consider a 10 Hz sine-wave (fig. 3b), again sampled at
fs=8.5 Hz, i.e. much below the minimum of 20 Hz that is
required according to the sampling theorem. It may be
noted, that the samples obtained are identical to those in fig.
3a, i.e. it would appear that the signal was made up of a
sine-wave of 1.5 Hz, rather than one of 10 Hz. This change
of frequency (from 10 to 1.5 Hz in this example) is known as
aliasing. Sine-waves at 18.5 and 27 Hz also give identical
digital signals (fig. 3c and d), and from these samples it
would be impossible to determine which frequency was
present in the original analogue signal. Only when the
sampling theorem is obeyed, is there no ambiguity as to the
frequency content of the signal.
Figure 2. The main steps in digital signal acquisition. The plots illustrate
the manner in which each step modifies the input.
Figure 3. Sine waves at 1.5 Hz, 10 Hz, 18.5 Hz, and 27 Hz, all give
exactly the same sample-values (marked as o), when sampled at 8.5
Hz. Thus, the sampled sine-waves of 10, 18.5 and 27 Hz appear to
have been acquired from a 1.5 Hz sine-wave, i.e. they have been
'aliased' down to 1.5 Hz. In order to avoid this ambiguity as to the
frequency in the analogue signal, the sampling rate must always be
greater than twice the frequency present in the data.
1
This reconstruction can be achieved by applying an analogue low-pass filter to the sampled signal, when each sample is
represented by an impulse. This reconstruction process is clearly quite different from the simple linear interpolation between
samples usually employed, when signals are plotted on a computer screen.
a
b
c
d
amplitude
amplitude
time (s)
time (s)
time (s)
time (s)
amplitudeamplitude
3. SCOPE • Volume 13 • Issue 4 • DECEMBER 2004 11
If a signal has been sampled with a sampling rate that is too
low and aliasing has arisen, it is (normally) impossible to
restore the original data. The error cannot be undone and
the signal should be discarded. Practical signals, such as
the ECG, can be considered as made up of a sum of sine
(and cosine) waves of different frequencies, in accordance
with Fourier analysis [1]
. In order to adequately digitize such
signals, the maximum frequency present in the signal must
be considered, and a sampling rate that is more than twice
this value must be chosen. In order to ensure a known upper
limit to the frequency content of a signal, a low-pass filter
(the anti-alias filter – see fig. 2) should be applied prior to
sampling. For the example of the ECG signal, typically all the
important information is contained in the band up to about
100 Hz. In the specific example below (fig. 4a shows a small
segment), the signal is was found to be contained in the
band below about 40 Hz (fig. 4c), and a sampling rate
above 80 Hz would thus seem adequate for the ECG signal.
However, there is also noise present at higher frequencies,
indicated by the sharp spikes at the mains frequency of 50
Hz, and its odd harmonics (150, 250, 350 and 450 Hz). Thus
a much higher sampling rate (above 900 Hz) would be
required. However, if we suppress this noise prior to
sampling, a lower sampling rate would be adequate. In this
example, we apply a low-pass filter (fig. 4b and d) that
retains frequencies below 45 Hz (ECG), and attenuates the
higher components (noise). We then sample at 256 Hz (i.e.
well above the theoretical minimum of 2*45 Hz). With the
anti-alias filter (fig. 4b, d and f), aliasing is avoided, but the
ECG signal itself is preserved. Without the anti-alias filter
(fig. 4a, c and e), aliasing occurs, and the noise above 256/2
Hz is aliased: 150 Hz appears as 106 Hz, 250 as 6 Hz, 350
as 94 Hz, and 450 Hz as 62 Hz . Of particular concern would
be the harmonic that has been aliased to 6 Hz2
. This is in the
middle of the frequency band containing the ECG and
contaminates the ECG signal. Clearly, after digitizing no
filtering can remove that harmonic without also affecting the
ECG signal. Furthermore, it would be impossible to tell
whether the activity at that particular frequency arose as a
result of aliasing, or if it was present in the original analogue
ECG signal (or how much of it was present). The aliasing
that has occurred cannot be undone, and the digitized
signal should be discarded.
Anti-alias filters are not perfect and cannot eliminate all
noise above their ‘cut-off’ frequency (45 Hz in the example).
This is evident in fig. 4, where the 50 Hz noise is attenuated,
but not completely cancelled. As a result of this limitation of
the filters, the sampling rate should normally be set to some
three to five times above the cut-off frequency of the anti-
alias filter, not at the theoretical lower limit of twice. In the
above example with the anti-alias filter cutting off at 45 Hz,
we used a sampling rate of fs=256 Hz. Excessively high
sampling rates should also be avoided, since they result in
more data (i.e. larger data files), require more computer
memory and computing time when processing, and may
require faster and more expensive A/D converter hardware.
The example shown underlines the importance of selecting
the sampling rate based on the maximum frequency present
in the signal, and not simply the maximum frequency that we
may be interested in. Thus, in order to select a suitable
sampling rate, first the maximum frequency of interest in the
signal should be determined (fmax). The cut-off frequency (fc)
of the anti-alias filter (i.e. the maximum frequency that the
filter passes) should then be chosen as a little above fmax, so
as not to attenuate the band of interest. The sampling
frequency is then chosen as fs > 3*fc. It may be noted that
signal acquisition systems often include a low-pass filter as
part of the in-built analogue signal processing, in order to
suppress noise (for ECG systems the cut-off frequency of
this filter might typically be set at 100 Hz). This filter can be
taken as the anti-alias filter, and the output signal digitized at
a sampling rate some 3 – 5 times above the filter cut-off
frequency, without the need for a further specific anti-alias
filter. It should also be emphasized again that the anti-alias
filter has to be applied prior to digitizing, i.e. it must be an
analogue filter. A digital filter cannot perform this task, since
it would operate on a signal that has already suffered
aliasing.
In the example of the ECG signal, the maximum frequency
of interest is known to be usually around 100 Hz, and thus
provides guidance as to the selection of the anti-alias filter
and the sampling rate for this signal. However, in an
unknown signal, such guidance may not be available. In that
case, you may make an initial assumption (chose a value as
high as possible), and apply an anti-alias filter and A/D
converter accordingly. You may then apply digital low-pass
filters to the digitized signal, using progressively lower cut-
off frequencies, until you are confident that you have
reached a limit, beyond which the important parts of the
signal are distorted. That will provide you with an indication
of the maximum frequency of interest. Plotting the amplitude
(or power) spectrum of the signal may be of assistance in
this process. Further acquisitions of this signal may then be
carried out using this ‘experimental’ maximum frequency,
and a corresponding anti-alias filter and sampling rate.
Quantization
The output of the A/D converter is actually a series of integer
numbers, each representing the signals amplitude at a
sample. This is illustrated in fig. 5, where the relationship
between the A/D converter's input (analogue) and output
(integer values) are shown, and figs. 6 and 7, where
analogue and the corresponding digital signals are
2
The aliased frequencies fa can be calculated as fa =|f –ifs|, where f is the original frequency, fs is the sampling rate, and i is
an integer such that fa≤
fs
⁄2. Thus, f =150 Hz becomes fa =|150–256|= 106 Hz, and f=450 Hz becomes
fa =|450–2*256|= 62 Hz, etc.
4. SCOPE • Volume 13 • Issue 4 • DECEMBER 200412
illustrated. In fig. 5, on the horizontal axis the analogue value
of any sample is shown, and the vertical axis gives the
corresponding integer-valued output of the A/D converter.
The integer output is obtained by rounding (to the nearest
integer) or truncation (to the next lower integer). Thus the
amplitude of the digitized signal varies in discrete steps; this
is known as ‘quantization’ of the amplitude.
The example in fig. 5 illustrates an A/D converter that has a
total of 32 distinct levels (steps), covering the range ±10
mV. If the input signal exceeds this range, the A/D converter
'saturates', i.e. gives the minimum or maximum value (-16
and +15, respectively). In fig. 6a, this A/D converter is
applied to a segment of ECG signal, giving integer values
between 1 and 13, corresponding to the ECG signal
amplitudes between about 1 and 9 mV. The quantization of
Figure 4. A segment of an ECG signal without (left column) and with (right column) anti-alias filtering. The samples of the signal (•) obtained at 256 Hz,
before (a) and after (b) anti-alias filtering. The anti-alias filter has clearly reduced the mains-noise, and smoothed the signal. c) The amplitude spectrum of the
analogue signal, with ECG activity below about 45 Hz, and mains-noise at 50, 150, 250, 350 and 450 Hz (fundamental and odd harmonics). d) After anti-alias
filtering (cut-off at 45 Hz), the higher harmonics are suppressed, and noise at 50 Hz is attenuated. Below 45 Hz, the filter has little effect. e) The amplitude
spectrum of the sampled signal, if no anti-alias filter is applied. The higher harmonics of the mains noise are aliased and appear at 106, 6 (buried in the ECG
signal), 94, and 62 Hz, respectively. f) The amplitude spectrum with the anti-alias filter and sampling at 256 Hz. No aliasing is evident, and the spectrum of the
ECG signal itself is maintained. Note that c and d show frequencies up to 500 Hz, and e and f only up to 128 Hz.
a b
time (s) time (s)
frequency (Hz) frequency (Hz)
frequency (Hz) frequency (Hz)
ECG
ECG
amplitude
amplitude
amplitude
amplitude
c d
e f
5. SCOPE • Volume 13 • Issue 4 • DECEMBER 2004 13
the signal is clearly evident. In fig. 6b, a higher-resolution
A/D converter is applied, and the quantization is almost
imperceptible, and would be quite adequate for most signal
processing tasks. Thus careful selection of the ‘amplitude
resolution’ of the A/D converter is vital in order to ensure
high quality digital signals. Any signal detail lost due to
quantization cannot (normally) be recovered; very poorly
quantized data may have to be discarded.
The integer output values of the A/D can be converter back
to the desired units (mV in fig. 6), by applying the
appropriate calibration. The final calibrated signal however
still consists of a series of discreet amplitude levels.
The resolution of the A/D converter is determined by the
‘number of bits’, and the number of different amplitude
levels = 2bits
. Thus the converter in fig. 5 and 6a used 5 bits
giving 32 distinct levels, and the 8-bit converter in fig. 6b,
256 levels. A higher number of bits (typically 10, 12 or 16 are
used) would lead to even smaller quantization errors.
Increasing the number of bits is not the only means of
reducing the approximation (or error) due to quantization.
The step-size of the A/D converter (the amplitude or vertical
resolution) depends both on the number of bits, and the
range of the converter: resolution=range/(levels-1). For the
example above, with a range of 20 mV (± 10 mV) and 5 bits,
the resolution=20/(25
-1)=0.645mV, and with 8 bits,
0.078mV. The resolution could be improved by reducing the
range. Since the signal in fig. 6 only occupies the range of
approximately 1 – 9 mV (known as the ‘dynamic range’ of the
signal), we could use a range of say 0 – 10 mV for the A/D
converter. The results are shown in fig. 7a, again using the 5
bit converter, where clear improvement compared to fig. 6a
can be observed. Note that now the integer output of the
converter nearly covers the full range from –16 to 15. The
equivalent can be noted for the 8-bit converter (fig. 7b). If the
range of the converter is chosen too small (e.g. 0 – 5 mV in
fig. 7c), the signal is clipped, as the A/D converter cannot
exceed its maximum (127 for the 8-bit converter). If a very
large range is chosen (-100 to 100 mV in fig. 7d), resolution
again becomes very poor, and even the 8-bit A/D converter
is not adequate. A good match between the dynamic
ranges of the signal and the A/D converter provides for the
smallest possible quantization error.
In practice, the range of the A/D converter is often fixed (or
can only be adjusted according to very limited options, e.g.
±1, ±5, ± 10, 0 – 1, 0 – 5, 0 – 10 V). The gain and offset of
a signal pre-amplifier is then used to adjust the signal to the
range of the A/D converter (rather than adjusting the range
of the converter to the signal).
How important a given quantization error is depends on the
signal: a quantization error of 0.078mV is quite important in
a signal that has a range of say ±1 mV, but much less so in
a signal covering a range of ±5V. Furthermore, reducing the
quantization error provides little benefit, if the signal is
already very noisy due to other sources (e.g. mains
interference or poor quality electronics). Since the
quantization error is relatively easy to control (e.g. by
Figure 6. An analogue (solid line – left-hand vertical scale) and digitized (• – right hand vertical scale) ECG signal, sampled at 200 Hz using a) 5 and b) 8
bits, and a range of ± 10 mV (see fig. 5). Note that the A/D converter output is integer-valued, and that the scales are different in a and b. Due to the low
number of quantization levels in a), the digital signal shows very large steps of about 0.6 mV. With a 8 bits (b), the amplitude resolution of the A/D converter
is improved (step-size of 0.08 mV), and most detail is preserved.
Figure 5. Illustration of quantization: the continuous-amplitude (analogue)
samples (x-axis) are converted to discrete values (y-axis), by rounding, or
truncation. Above and below the ‘range’ of the A/D converter (±10 mV),
the output is saturated. In this example, the converter cannot give a value
smaller than –16 or larger than +15 (integer output).
Output(integer)
Input (mV)
time (s) time (s)
ECG(mV)
ECG(integer)
a
ECG(mV)
ECG(integer)
b
6. SCOPE • Volume 13 • Issue 4 • DECEMBER 200414
employing a better A/D converter with finer amplitude
resolution), quantization errors should normally be small
compared to other (more difficult to control) sources of
noise.
An A/D converter with a high number of bits will provide
lower ‘quantization noise’, and may provide acceptable
results even if the dynamic range of the signal and A/D
converter have not been well matched. However, these
converters are more expensive, and a higher number of bits
may also increase the size of files in which the data is
stored: an 8-bit A/D converter requires only one byte per
sample, but 2 bytes are required for a 16-bit converter3
.
Summary and Discussion
In the above we have considered the most important steps
and choices in converting signals from analogue to digital
form. In fig. 8 we show further examples to aid discussion
and review.
The respiratory flow signal (fig. 8a) was found (by inspecting
the spectrum) to have a maximal frequency content of about
20 Hz, and was sampled at 200 Hz. This sampling rate is
rather higher than strictly required, leading to rather larger
files than necessary. One consideration here, however, was
that a number of signals were acquired simultaneously, and
it is then easiest to sample all channels at the same rate,
corresponding to the maximum sampling rate required.
The digitized EMG (electromyogram) signal (fig. 8b) shows
amplitudes that are hard-limited at ±40. This is clearly not
‘physiological’, and is characteristic of clipping which arises
when the signal exceeds the range of the A/D converter (or
the amplifier). In other applications, clipping may only be
observed at either the peaks or the troughs of the signal, or
it may be asymmetrical, occurring at different absolute
values for minima and maxima.
The blood-flow signal (fig. 8c) varies in discrete steps in
amplitude, showing plateaus between steps. This is typical
for signals that have been acquired with an inadequate
amplitude resolution. It should be noted that these signals
have been plotted, as is usual, by drawing straight lines
between their samples, which enhances the stepped
appearance of the data. In order to obtain a better quality
Figure 7. The analogue signal (solid line – left axis) and the sampled version (• – right axis) sampled at 200 Hz. a) 5-bit A/D converter with an amplitude
range of 0-10 mV; b) 8 bit A/D converter, 0 – 10 mV; c) 8 bit A/D converter, 0 – 5 mV; d) 8-bit A/D converter, ±100 mV. Note the saturation at 5mV (integer
output of 127) in c, and the poor amplitude resolution in a and d.
3
Usually 10 or 12 bit converters also use 2 bytes of file-storage for every sample.
time (s)
ECG(mV)
ECG(integer)
a
time (s)
ECG(mV)
ECG(integer)
b
time (s)
ECG(mV)
ECG(integer)
c
time (s)
ECG(mV)
ECG(integer)
d
7. SCOPE • Volume 13 • Issue 4 • DECEMBER 2004 15
signal, either an A/D converter with a higher number of bits
should be used, or the gain of the pre-amplifier should be
reduced.
The ECG signal (fig. 8d) has been sampled at 60 Hz. This is
rather low (and probably too low according to the sampling
theorem – though we do not know the characteristics of the
anti-alias filter used in acquisition, and thus cannot be
certain). Due to the low sampling rate, the samples do not
always coincide with the peaks of the ECG, and thus the
peak-values appear to vary greatly between beats – much
more than expected in a normal ECG. If the ECG signal had
been adequately sampled, the analogue ECG signal (or a
digital version sampled at an arbitrarily high frequency)
could be reconstructed from the samples, to show the more
constant amplitudes of the original ECG. If aliasing has
occurred, then accurate calculation of intermediate sample
values would of course not be possible.
Figure 8. Examples of digitized signals. a) A respiratory flow signal, sampled adequately at 200 Hz. b) An EMG signal, showing clear evidence of clipping
(saturation) at about ±40 (arbitrary units). c) A blood flow signal (from transcranial Doppler ultrasound), that shows well-defined amplitude steps, due to
poor quantization. d) An ECG signal sampled at 60 Hz. Due to the low sampling rate, the peak-values appear to fluctuate strongly between beats. e) Blood
pressure sampled at 40 Hz, showing an ectopic beat. f) The heart-rate, derived from the blood-pressure signal. The beat-interval varies in discrete steps
corresponding to the interval between samples (0.025 ms). Note that the signals displayed here are not calibrated in terms of their amplitude, and that
straight lines have been drawn between samples.
time (s) time (s)
time (s) time (s)
time (s) time (s)
RespiratoryflowBloodflowBloodpressure
EMGECGBeat-intervals(s)
a b
c d
e f
8. SCOPE • Volume 13 • Issue 4 • DECEMBER 200416
I consider myself an engineer – and have a certificate to
prove it. I am proud to be an engineer. I do not see myself as
a scientist. For me, engineering is not just an applied science
in the same way that medicine isn’t either. I will try to explain.
When I was only 6, with my grandmother as baby sitter,
and to my parents’ great horror when they returned,
I successfully repaired a household mains fuse – the type
with real wire you had to poke through a hole and fasten
each end. I went on to make models of all sorts, pull to bits
and sometimes repair all nature of mechanisms, electrical
and mechanical and in my early teens made radio sets –
with valves, high voltages etc. receiving a few nasty shocks
and burns in the process. Heaven was a transistor radio
kit – just two transistors, each costing 15 shillings (75p), a
fortune in those days (1958) with pocket money of two
shilling per week. I seemed to have a fascination about and
an instinct for how things were made and worked. Naturally
for those days I wanted to be an engine driver and my father
occasionally arranged for me to ride on the footplate of
real steam locomotives, then still in regular service – what
excitement.
Both my grandfathers were engineers. One a chemical
engineer and ‘soft drinks’ entrepreneur, the other an ex-army
engineer who built and raced his own cars (I still have his
gold medals and cups). Is engineering genetic, much like
art and music? I don’t know but having taught electronics for
a long time, it is clear that some students ‘get the hang of it’
and some don’t. I am saddened though by the seeming
decline of interest in ‘engineering’ hobbies amongst young
people today. I have not been able to pass it on to my
children. Being a whiz kid at computer games is not quite
the same thing.
But to get back to ‘engineering’. Wind the clock back a few
thousand years. There was no shortage of engineering
achievements, many are still with us today. The techniques
What is Engineering?
Mike Bolton
Chair - IPEM Engineering Group Board
The blood pressure signal (fig. 8e), was low-pass filtered at
12 Hz, and sampled at 40 Hz, and thus adequately
according to the sampling theorem. The maximum increase
in blood pressure was then used as a marker to measure
the time interval between beats, in order to determine the
heart rate. The results are shown in fig. 8f. Since the beat-
interval was determined from the number of samples
between each increase, and the time between beats varies
in discrete steps of 0.025 s (considering the sampling rate
of 40 Hz), the beat-interval varies in rather large discrete
steps. Clearly the beat-interval signal is of poor quality, and
a higher sampling rate for the blood pressure signal would
have lead to better results, showing smoother variations.
While the sampled signal contains all the information in the
original analogue data (since the sampling theorem was
obeyed), precise measurement of heart-rate should be
possible. However, our simple detection algorithm did not
exploit the data adequately. This illustrates the important
point that strict adherence to the minimum requirements of
the sampling theorem does not guarantee good results.
However, adequately sampled signals do allow data to be
reconstructed at an arbitrary (higher) sampling rate, by
using the appropriate interpolation methods. If the sampling
theorem was not obeyed, such reconstruction is not
possible4
.
In many applications, particularly in biomedicine, you may
only get one chance to collect the signals. It is therefore vital
that signal acquisition is adequately prepared and planned.
We hope this tutorial will have clarified the underlying
principles, and allow well-informed choices to be made5
.
References
[1] D. M. Simpson and A. De Stefano; ‘A Tutorial Review of the
Fourier Transform’; Scope, Vol.13, Nr.1, pp. 28-33, March 2004
[2] M. Merri, D.C. Farden, J.G. Mottley, E.L. Titlebaum; ‘Sampling
frequency of the electrocardiogram for spectral analysis of the
heart rate variability’; IEEE Trans Biomed Eng. Vol. 37(1) pp. 99-
106. January 1990;
[3] A. V. Oppenheim and R. W. Schafer; ‘Discrete Time Signal
Processing’; Prentice Hall, 1989
[4] J. Proakis and D. Manolakis; ‘Digital Signal Processing’;
Macmillan, New York, 1988
[5] P.S.R. Diniz, E.A.B. da Silva, S.L. Netto; ‘Digital Signal
Processing System Analysis and Design’; Press Syndicate of the
University of Cambridge 2002
[6] J. Jerri; ‘The Shannon sampling theorem—its various
extensions and applications: A tutorial review’; Proceedings IEEE,
Vol. 11, pp. 1565–1596, 1977
4
Detailed discussion and analysis of sampling requirements for determining heart-rate from the ECG may be found in [2]
.
5
Further details on the theory underlying sampling may be found in standard text books, such as [3-5]
. Extensions of the sampling
theorem, including considerations on unequally spaced samples, can be found in [6]
.