Virtual Physiological Instruments  

Over-sampling ADC



When the Physiological signals are digitized it is common practice to firstly remove the high Electrode Offset Potential from the signals by means of high-pass filtering, followed by an amplifier with a gain 1000 for large signals such as EMG and ECG to 5000 for EEG signals. The following medium resolution Analog to Digital Converter now only has to comply with the maximum signal range of the specific physiological signal. If each electrode signal has to be sampled individually, the analog high-pass filters exhibit too low frequency accuracy to allow for a high common mode rejection. This is caused by the inaccuracy of the input capacitor used in the high pass filter. Also the anti-aliasing filter in the individual data path is a source of mismatch between channels leading to a low common mode rejection in subsequent stages of signal processing.


For best performance no analog filters are allowed in the individual signal path. Using the Reference Amplifier with a gain of 20 and  to allow for some headroom in the signal, the ADC now needs a resolution of a factor of 200 to 1000 higher than the one used with high pass filters. To achieve this high resolution either a Sigma-Delta ADC or an over-sampling ADC may be used. A rule of thumb is that the cut off frequency of the anti-aliasing filter should lay at least a factor of 100 above the main frequency in the common mode component, in Europe with a line frequency of 50 Hz about 5 kHz. In this case an over-sampling frequency of 20 to 25 kHz is sufficient. Digital decimation filtering is used to increase resolution of the digitized data. Starting with a 16 bits ADC the resolution may be increased by 5 bits at a resulting sub-sample frequency of 2 kHz. A pre-requisite is that the 16 bits data contains at least 1.5 bits random noise before decimation as shown in the figure.

Traces left top to bottom: sine wave sampled at low resolution with 0.5, 1 and 1.5 bits random noise. Traces at the rigth: left traces filtered by means of a simple 9 tap decimation filter. It is clearly seen that adding only 1.5 bits of random noise already gives a good interpolation of the bit steps and hence a higher resolution


              Contact us at :