The notion of noise is relative to the problem you are trying to solve. Think of the algorithm and thresholds, then consider time domain and frequency domain before freezing. 

A good rule though is to always think of Nyquist, sampling and avoiding aliasing (low pass filter before sampling!) and be completely aware of your signal behaviour... 

Actually I get the sampled graph from the BIH database and also I found that I can generate a filter in the filter design toolbox and export an array of filter coeff and multiply it with my original signal but that isn't working right.  Im not getting a filtered output rather getting the curve of the coeffs itself.

Also, do know how to pass the signal through a low pass filter using functions ?

There are two ways to implement a filter, if the data is in the time domain.

1.  In the time domain - add together samples in a moving window according to the coefficients given for the filter, e.g. add 10 samples together and average (rectangular filter), or add 1/4 first sample, half second sample and 1/4 third sample.

2.  Transform data into the Frequency domain and apply the filter shape to the whole data set, e.g. for the rectanguar filter apply a sinc envelope factor for the appropriate width of rectangle, then perform inverse transform.

In 1. the filter only applies to some of the data at a time, in a moving window, and the window is moved along the data to get new data and generate new results.

In 2. the shape of the filter is as wide as the whole frequency band - the Nyquist bandwidth.  The filter and data stay still and each data point is multiplied by the filter value at that frequency.

Digital filters (FIR) are usually given in terms of the factors to multiply successive samples by, before adding them together.  Then move the data on one point and repeat.  e.g for simple filter with data a b c d e f g h  i  j, result is a/4, a/2+b/4, a/4+b/2+c/4, b/4+c/2+d/4, c/4+d/2+e/4, d/4+e/2+f/4, etc.   The first few results are wrong because the filter isn't full (or are right if the data really was zero before the first data point, and not merely unknown).

Manish,

welcome,

The colleagues above have pointed out some techniques to filter you ECG signal.

You have to determine at first how your signal is contaminated by noise by comparing your real signal with the desired one. Knowing the difference, one can device the specifications of the filters one can use to clean the signal.

One practical way is to make FFT of your signal and discard the extreme frequency components  at lowest frequency and that at the highest frequency successively and reconstruct the signal again until you get satisfactory waveform of the ECG signal. This is some iterative method of low pass and high filtering.

Best wishes

See the documentation on MATLAB's filtfilt() command, it already contains an ECG example. I'll explain briefly here, first you must design a filter (can be any types of lowpass, highpass, bandpass, etc,...) using the designfilt() function or the filter design toolbox that you already found, the toolbox GUI can export MATLAB code (customized functions) for redesigning that digital filter which can be used in your scripts. then use the filtfilt() function to get an undelayed filtered data set, since it is especially designed to remove delay from digital filters by filtering the data back and forth. however, if you can't find this function in your installed version of MATLAB then it means that you don't have the DSP toolbox, and you'll have to use filter() command instead, but i would advise that you use a FIR filter on this command since IIR filters introduce delays on the dataset.

the syntax for these filter functions are both:

>> y = filtfilt(d,x) and

>> y = filter(d,x) 

where x is the input "raw" signal, d is the digital filter that you design and store, and y is the resulting output "filtered" signal. run the example code below to see both filter commands in action (credit goes to Mathworks) :

wform = ecg(500);

x = wform' + 0.25*randn(500,1);

d = designfilt('lowpassfir', ...

'PassbandFrequency',0.15,'StopbandFrequency',0.2, ...

'PassbandRipple',1,'StopbandAttenuation',60, ...

'DesignMethod','equiripple');

y = filtfilt(d,x);

y1 = filter(d,x);

subplot(2,1,1)

plot([y y1])

title('Filtered Waveforms')

legend('Zero-phase Filtering','Conventional Filtering')

subplot(2,1,2)

plot(wform)

title('Original Waveform')

The following article shows exactly what you need and working on. If you want further information or data or even matlab code please let me know.

best wishes

Article Low Complexity Adaptive Noise Canceller for Mobile Phones Ba...

is anyone have the code to remove noise from ECG signals??

please share