| • | Data Acquisition Unit |
PSG Data Acquisition Hardware and Specifications
The purpose here is to discuss a digital acquisition unit (DAU) for the digital acquisition of a single channel of PSG data, to recommend specifications, and finally to describe a simple method for evaluating the performance of such a unit. A DAU as described here is necessary for the acquisition of most channels of PSG data. The DAU includes the hardware and software necessary to amplify, filter, and digitize a single channel of PSG data. The DAU takes the analog-input signal, amplifies and filters the signal, and then digitizes it and sends it to the computer.
At the input to the device is a differential amplifier (DA) which amplifies the difference in the signal between two input leads. The differential feature (in contrast to amplifying a single input signal) is a very important aspect of this amplifier, as input PSG data typically have signal components common to both input leads that are much larger than the differential signal of interest. The most important function of the DA is to attenuate this common-mode signal to a level much smaller than the differential signal of interest. A DA has both an input resistance between the two input leads and a resistance from each input lead to ground. An input resistance to ground on each input of 1 to 10 megohms is adequate (1-megohm resistance will provide less than 1% signal loss); higher input resistances are readily obtained, but this may restrict the implementation to a less (perhaps size, input power, and/or noise) desirable fabrication technology. Almost all DAs have a sufficiently fast frequency response (bandwidth) that the frequency response of the DA should not be a factor in determining the frequency response of the DAU.
Bandwidth: The bandwidth is determined primarily by the frequency response of the analog filters in the DAU. The overall processing includes both digital filters and analog filters. The digital filters are easier to design and implement than the analog filters, but the analog filters are a must. Practically, the analog filter has two tasks: first, it must block any dc component of the input signal, and, second, it must include the anti-aliasing filter. Input signals often contain a large dc component which must be blocked. This component can arise from the electro-chemical interface created when a metal electrode is connected to biological tissue. The anti-aliasing filter is needed to reduce the amplitude of the higher- frequency signals present in the input data (as well as noise created in the analog section of the DAU). The DAU should not contain 50 or 60 Hz filters. There are no standards for the analog filters. One possibility is to use the same type of filtering used with the polygraphs. The analog low-frequency filter was defined by the time constant. This filter attenuates signals below the cutoff frequency such that for every factor of two reduction in frequency below the cutoff frequency results in a factor of two reduction in the gain. The high end of the bandpass filter is normally determined by a second or third order filter, which means that beyond the higher cutoff frequency there is a factor of four reduction in gain for every doubling of the signal frequency. Note that the filter passes input signals above the cutoff frequency. They are only attenuated as described here. The characteristics of this filter are critical in describing the characteristics of the DAU.
Digital Conversion: The sampling rate, together with the anti-aliasing filter, is the most important factor in determining the fidelity of the digitized signal. In brief, the faster the sampling rate, the less aliasing present in the conversion and hence the more accurate the digital representation of the analog signal. Any input signal has higher frequency components which cause frequency aliasing. Once the signal is digitized, frequency aliasing that occurs in the sampling process cannot be recognized or removed. In fact, frequency aliasing may make the digitized version appear less noisy when in actuality it can be a poor representation of the analog signal. I have previously published examples of the effects of frequency aliasing. [see e.g. Smith, J.R. “Transferring EEG Polysomnography to the Home Environment,” chapter in Medical Monitoring in the Home and Work Environment, L. Miles and R. Broughton (eds), Raven Press, New York, 1990: 219-229.] The Nyquist sampling criteria states that the sampled data will be an exact replica of the input signal if the sampling rate is at least twice that of the fastest input signal. It is important to note that the frequency of the signal to be digitized is not the DAU bandwidth or even ten times the frequency of the input bandwidth. Once the data is digitized the sampling rate can be reduced with the incorporation of digital anti-aliasing filters.
The number of bits needed depends on what is to be done with the data. Only about six bits are needed to represent the data on a multi-channel display on a computer monitor. The use of more bits will improve the dynamic range of the digitized data. First the minimum signal level and signal resolution must be described; a specification of the number of bits is not by itself adequate without specifying the minimum detected signal.
DAU Evaluation
Specifications are helpful, but the user can judge the performance without understanding the specifications by applying a square-wave calibration signal at the input to the DAU. The calibration signal must come from a separate isolated unit and be applied at the DAU input. Almost all commercial systems do not apply the calibration signal at the input, and thus the overall performance of the amplifier cannot be evaluated. This form of calibration is similar to what was used with analog polygraphs. The advantage of a square-wave signal rather than a sinewave for the calibration signal is that the frequency response of the DAU can be obtained with the square wave signal. The leading edge of the response to the square-wave provides the necessary information for the DAU’s high frequency response, (the shorter the rise time the wider the bandwidth), and the decay time of the response can be used to determine the DAU’s low frequency response.