Is BioSemi going to produce systems with more than 16-bit resolution ?
 

No, we are not planning to do this in the near future.

 

A first question is: how many bits are needed ? In order to answer this question, we first have to determine the optimal bitstep (the LSB value). In the gain question, it was described that 1uV per bit is quite optimal for typical biopotential systems. It should be noted that this value is valid for a Successive Approximation Register type ADC (SAR type) with not more than LSB/3 inherent noise (only very few precision types meet this demand). For an oversampling Sigma-Delta type converter, the number of bits is a quite meaningless figure. Since the output words from such a converter are calculated from the average of a high-number of one-bit conversion, the word length can be made of any arbitrary number. For this type of converter, we have to look at the dynamic range at a certain output word-rate. Given the 1 uV bitstep, the number of bits of the ADC determines the dynamic range of the system, thus the largest signal that can be handled without distortion (clipping).

 
The plot at the left gives an impression of the signal amplitudes that can be expected. Using an ADC with a high dynamic range (many bits), and with a high sample rate, is attractive because it makes the system applicable for a broad range of biopotential measurements. However, increasing the number of bits and sample range quickly increases the power consumption and costs of a converter. Again, designing a good system is a question of making a sensible tradeoff.

 

  
There are two routes to go:
Suppress the large offset potential with a reduced gain at low frequencies.
Use an ADC with enough dynamic range to handle the offset potentials any high-pass filtering
 
It is obvious from the plot above that large signals have large amplitudes, whereas small signals generally have small amplitudes. Therefore, a frequency response of the amplifier that is tailored to the signal content uses the ADC most effectively (see also the filter question). For example: a 12bit 2kHz converter is fully sufficient for a system specially designed for surface EEG recordings. Going to a 16 bit, 4kHz converter makes the system suitable for all surface recordings (EEG, ECG and EMG). If the system can be configured for 8 or even 16kHz sample rate, all biopotential measurements are possible. Reducing the gain at low frequencies with a factor of 16 can provides a DC input range of +500 mV to -500 mV, allowing all common electrodes to be used without saturation on electrode offset voltages. This is exactly what we have done with the ActiveOne system operating in AC mode.
 
Route two seems attractive because of two reasons. The first reason is that DC measurement are possible without any software equalizing. The second advantage is that there is zero recovery time after an overload condition has been removed (no filter delays). Now, how many bits, or dynamic range would you ideally want. This depends of course on the electrode offset potentials. From our experience and measurements we have come to the following guideline:
 
With good Ag-AgCl electrodes, de offset voltages are surprisingly low, maximally in the order of 10 mV.
With electrodes other than Ag-AgCl such as tin, plain silver and stainless steel, the offset voltages are in the order of maximally a few hundred mV, provided all electrodes are of the same material, and the electrodes are not contaminated with metal particles of different metal.
If electrodes of different materials are mixed, offset voltages up to several volts are possible, according to the difference in half-cell potential. For example a worst case situation would be to combine a gold or platinum electrode and an aluminum electrode. This which would produce a potential difference of approx. 3.4 V

 

This gives us some specifications for the needed resolution of the ADC. So long as we use Ag-AgCl electrodes, a 16 bit converter is sufficient: 2^16 x 1 uV = 66mV, so we have a +33mV to -33mV input range. In order to cover other than Ag-AgCl electrodes, we need more bits. In our experience, an input range in the order of +250mV to - 250mV is sufficient for most applications, while +500mV to -500 mV gives a little more headroom in situations with difficult electrodes: for example stainless steel needles, contaminated electrodes, and worn down Ag-AgCl electrodes (combination of AgAgCl electrodes with pure silver): something like 20 bits would thus be required (+524mV to - 524mV input range) With an even higher number of bits we would be able to cover all combinations of electrode materials. However, this is not a common demand, and we will not go into this further. So the ideal biopotential measurement system would have a 1uV bitstep, and a 20 bit ADC with a sample rate of up to 16kHz. To allow for a large number of channels operating on a reasonable sized battery (or low capacitance DC-DC converter) the power consumption of the ADC should remain limited to something like 10mW at the most.

Unfortunately, at the moment none of the commercially available AD's comes near the described ideal. There are no indications that one of the leading manufactures will be able to develop something like the "perfect biopotential ADC" in the near future. So again the process of designing a good measurement system is a trade off between opposing demands: We want high dynamic range, high sample-rate but at the same time a low power consumption.

And what about all these recent 22-24 bit Sigma-Delta ADCs then, you might ask. First, you should note that the bit numbers derived in the above apply to low-noise SAR type ADCs. Like we said earlier, the number of bits in a Sigma-Delta type converter is a quite meaningless figure, the dynamic range is the figure of merit here. If we look at the specs of dynamic range, in combination with sample-rate and power consumption, the strong image of the Sigma-Delta begins to faint somewhat. The demanded 20-bit successive SAR performance can be compared with approx. 114 dB dynamic range of a Sigma-Delta (a 20 bit SAR would have a pk-pk noise level of approx. 2 LSB, the dynamic range is 2^19 = 114 dB). Going over the various datasheets again will make you realize that no current sigma delta comes close to the 114 dB in combination with a sample rate in the several kHz order and a less than 10 mW power consumption. Consider for example the popular AD7716 SD converter from Analog Devices. Several manufacturers of biopotential instrumentation use this converter in their products and advertise them as 22-bit systems. However, looking a bit closer at the data sheet reveals that this is a highly exaggerated claim: the converter provides just 99 dB dynamic range at 2 kHz sample rate. Therefore, the performance lies somewhere between an 17 bit and 18 bit SAR ADC (2^16 = 96 dB, 2^17 = 102 dB). Consequently, the DC range with this Sigma-Delta converter is limited to something like +100 to -100 mV (a 17 bit successive-approx. offers 66 mV, an 18-bit offer 132 mV). This limits the system to good Ag-AgCl electrodes, an application where the 16 bit SAR type ADC does already suffice. However, the typical 16 bit SAR ADC has a much lower power consumption, and can be operated without quality loss to much higher sample rates (up to 16 kHz). These are the reasons why we have chosen a 16-bit SAR ADC with switchable AC/DC amplifier in our digital systems.